Remote Sensing of Wetlands - Applications and Advances-Crc Press (2014)

Edited by
Ralph W. Tiner
Megan W. Lang
Victor V. Klemas
Remote Sensing
of Wetlands
Applications and Advances
Remote Sensing
of Wetlands
Remote Sensing
of Wetlands
Edited by
Ralph W. Tiner
Megan W. Lang
Victor V. Klemas
Boca Raton London New York
CRC Press is an imprint of the

Taylor & Francis Group, an informa business
Cover: June 14, 2014 Landsat 8 TM image of southern Nova Scotia and the Bay of Fundy (15 meter resolution; RGB – bands
6, 2, and 1; image enhanced by Mike O’Brien, MDA Information Systems LLC, Gaithersburg, Maryland). Lake Kejimkujik
is the most prominent feature in the inset. The yellow arrow points to Holdrights Lake that is shown in the ground-level
photo. Marsh, aquatic bed, and black spruce bog are the wetlands present within the lake basin. (Photo by Ralph Tiner)
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2015 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works

Version Date: 20150129
International Standard Book Number-13: 978-1-4822-3738-2 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti-
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy-
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For
organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Contents
Preface...............................................................................................................................................ix
Editors................................................................................................................................................xi
Contributors.................................................................................................................................... xiii
Acronyms........................................................................................................................................xvii
Section I
Introduction to the Use of Remote Sensing for
Wetland Mapping
Chapter 1 Wetlands: An Overview................................................................................................ 3

Ralph W. Tiner
Chapter 2 Classification of Wetland Types for Mapping and Large-Scale Inventories............... 19

Ralph W. Tiner
Chapter 3 Introduction to Wetland Mapping and Its Challenges................................................ 43

Ralph W. Tiner
Chapter 4 Early Applications of Remote Sensing for Mapping Wetlands................................... 67

Ralph W. Tiner
Chapter 5 Advances in Remotely Sensed Data and Techniques for Wetland Mapping
and Monitoring............................................................................................................ 79
Megan W. Lang, Laura L. Bourgeau-Chavez, Ralph W. Tiner,
and Victor V. Klemas
Section II
Summaries of Remote Sensing Technologies
and Their Application for Mapping Wetlands
Chapter 6 Mapping and Monitoring Surface Water and Wetlands with Synthetic
Aperture Radar.......................................................................................................... 119
Brian Brisco
Chapter 7 Wetland InSAR: A Review of the Technique and Applications............................... 137

Shimon Wdowinski and Sang-Hoon Hong
v
vi Contents
Chapter 8 Radar and Optical Image Fusion and Mapping of Wetland Resources.................... 155
Elijah Ramsey III and Amina Rangoonwala
Chapter 9 Theory and Applications of Object-Based Image Analysis and Emerging

Methods in Wetland Mapping................................................................................... 175
Joseph F. Knight, Jennifer M. Corcoran, Lian P. Rampi, and Keith C. Pelletier
Chapter 10 Unmanned Aerial Systems and Structure from Motion Revolutionize

Wetlands Mapping.................................................................................................... 195
Marguerite Madden, Thomas Jordan, Sergio Bernardes, David L. Cotten,
Nancy O’Hare, and Alessandro Pasqua
Section III
Applications of Remote Sensing for Mapping
Specific Wetland Habitats
Chapter 11 Remote Sensing of Submerged Aquatic Vegetation and Coral Reefs....................... 223
Sam Purkis and Chris Roelfsema
Chapter 12 Remote Sensing of Mangroves.................................................................................. 243

Victor V. Klemas
Chapter 13 Tidal Marsh Classification Approaches and Future Marsh Migration Mapping
Methods for Long Island Sound, Connecticut, and New York................................. 265
Mark Hoover and Adam Walton Whelchel
Chapter 14 Using Moderate-Resolution Satellite Sensors for Monitoring the Biophysical

Parameters and Phenology of Tidal Marshes............................................................ 283
Deepak R. Mishra and Shuvankar Ghosh
Chapter 15 Great Lakes Coastal Wetland Mapping.................................................................... 315

Laura L. Bourgeau-Chavez, Zachary M. Laubach, Anthony J. Landon,
Elizabeth C. Banda, Michael J. Battaglia, Sarah L. Endres, Mary Ellen Miller,
Robb D. Macleod, and Colin N. Brooks
Chapter 16 Mapping Wetlands and Surface Water in the Prairie Pothole Region of
North America........................................................................................................... 347
Jennifer Rover and David M. Mushet
Chapter 17 Mapping the State and Dynamics of Boreal Wetlands Using Synthetic
Aperture Radar.......................................................................................................... 369
Daniel Clewley, Jane Whitcomb, Mahta Moghaddam, and Kyle McDonald
Contents vii
Chapter 18 Fusion of Multispectral Imagery and LiDAR Digital Terrain Derivatives

for Ecosystem Mapping and Morphological Characterization of a Northern
Peatland Complex...................................................................................................... 399
Antonio Difebo, Murray Richardson, and Jonathan Price
Chapter 19 Airborne LiDAR-Based Wetland and Permafrost-Feature Mapping on an

Arctic Coastal Plain, North Slope, Alaska................................................................ 413
Jeffrey G. Paine, John R. Andrews, Kutalmis Saylam, and Thomas A. Tremblay
Chapter 20 Hybrid Mapping of Pantropical Wetlands from Optical Satellite Images,

Hydrology, and Geomorphology............................................................................... 435
Thomas Gumbricht
Chapter 21 Capturing the Dynamics of Amazonian Wetlands Using Synthetic Aperture

Radar: Lessons Learned and Future Directions....................................................... 455
Thiago Sanna Freire Silva, John Melack, Annia Susin Streher,
Jefferson Ferreira-Ferreira, and Luiz Felipe de Almeida Furtado
Chapter 22 Mapping China’s Wetlands and Recent Changes with Remotely Sensed Data........ 473
Zhenguo Niu
Chapter 23 Mapping Invasive Wetland Plants............................................................................. 491

Carol A. Johnston
Chapter 24 Multisatellite Remote Sensing of Global Wetland Extent and Dynamics................ 511
Catherine Prigent and Fabrice Papa
Section IV Promising Developments and Future Challenges
Chapter 25 Promising Developments and Future Challenges for Remote

Sensing of Wetlands.................................................................................................533
Megan W. Lang, Sam Purkis, Victor V. Klemas, and Ralph W. Tiner
Preface
The need for wetland mapping has never been greater as the world’s population continues to
increase. Human demands on wetlands for agriculture, aquaculture, forestry, and development are
accelerating especially in underdeveloped countries. Climate change impacts will further exacer-
bate wetland alteration and loss, but cumulative effects are difficult to predict. Given the environ-
mental services that natural wetlands freely provide, it is in society’s best interest to sustainably
manage wetlands at both local and global scales. In order to effectively manage any environmental
resource, its distribution and extent must be known along with its diversity and how and why it is
changing. Wetland maps, inventories, and monitoring programs provide vital information to aid in
natural resource planning and wise use of wetlands. While wetland maps do not guarantee the best
use of wetland resources, they do provide governments and developers with information that can
be used to do so.
The science of wetland mapping has developed rapidly over the past few decades. Wetland
mapping with remotely sensed data began with the manual interpretation of aerial photography.
Interpreters would trace wetland boundaries on overlays of analog aerial photographs, and a car-
tographer would later convert those drawings into a map using traditional cartographic techniques.
This technique produced relatively accurate wetland maps at the fine spatial scale necessary to
support most environmental management decisions, but this method had drawbacks. It was slow,
was resource intensive, and encountered challenges with some land covers, especially forests and
drier-end wetlands. With the advent of satellites, advanced sensors, and geographic information
systems (GIS), new tools became available for wetland mapping. Digital imagery and GIS changed
traditional mapping to combine the interpretation and cartographic operations of wetland mapping
into a single step. The interpreter now became both the interpreter and cartographer—the image
analyst. Scientists have helped advance remote sensing technology to improve the spatial, spec-
tral, radiometric, and temporal resolution of the sensors and the analytical processes necessary to
interpret the data. We now appear to be at a crossroad where digital imagery and more automated
techniques can be used to produce high-quality wetland maps on a more frequent basis than can
be accomplished by manually based aerial photo analysis. The automated approach is quicker and
usually costs less when data and interpreter time are accounted for. This book reviews many of the
datasets and techniques that are responsible for this advance.
The book is arranged in four sections: Section I is designed as an introduction to wetlands and the
evolving use of remote sensing for wetland mapping; Section II presents summaries of some of the
newer technologies and their use in wetland mapping; Section III highlights applications of the new
technologies for mapping different types of wetlands or various analyses; and Section IV, the final
chapter, summarizes the most impactful advancements and identifies areas for further development.
While the book is not an instruction manual for remote sensing of wetlands, it is our hope that readers
will gain an appreciation of the challenges that wetlands pose and a broad understanding of the latest
remote sensing technologies and their strengths and limitations for inventorying wetlands. The book
also provides some insights into future directions. Information on the fundamentals of remote sens-
ing and its varied applications in other environments can be found in other books, including
Remote Sensing of Coastal Environments (Y. Wang, editor; 2010; CRC Press)
Remote Sensing of Forest Environments: Concepts and Case Studies (M.A. Wulder and
S.E. Franklin, editors; 2002; Kluwer Academic Publishers)
Remote Sensing and Global Environmental Change (S.J. Purkis and V.V. Klemas; 2011;
Wiley-Blackwell)
ix
x Preface
Remote Sensing: Models and Methods of Image Processing (R.A. Schowengerdt, editor;
2007; Academic Press)
Remote Sensing and Image Interpretation (T. Lillesand, R.W. Kiefer, and J.W. Chipman;
2008; John Wiley & Sons, Inc.)
Introduction to Remote Sensing (J.B. Campbell and R.H. Wynne; 2011; The Guilford Press)
Remote Sensing: Principles and Interpretation (F.F. Sabins; 2007; Waveland Press, Inc.)
Remote Sensing of the Environment: An Earth Resource Perspective (J.R. Jensen; 2006;
Prentice Hall)
Remote Sensing of Vegetation: Principles, Techniques and Applications (H.G. Jones and
R.A. Vaughan; 2010; Oxford University Press)
Another book—Wetland and Environmental Applications of GIS (J.G. Lyon and J. McCarthy, edi-
tors; 1995; Lewis Publishers)—addresses GIS applications of remote sensing, which include some
examples of interest to wetland specialists. Remote Sensing of Coastal Environments (listed earlier)
also contains some wetland examples.
While aerial photointerpretation is still used as a valuable tool for wetland mapping, our book
focuses on the application of more advanced remote sensing techniques for wetland detection. We
are thankful for the researchers who dedicated their time and expertise to producing the unique
collection of well-informed chapters presented in this book. The authors have technical and envi-
ronmental knowledge in difference areas. The cumulative wetland mapping knowledge presented
in this text is unparalleled. We are also grateful to the publisher CRC Press—for making this
information available in a single book.
Ralph W. Tiner
Megan W. Lang
Victor V. Klemas
Editors
Collectively, the editors have nearly 100 years of experience in wetland mapping, having mapped
wetlands from the coasts to mountaintops and nearly everywhere in between. Their work has
ranged from interpreting aerial photography to analyzing digital imagery and using geographic
information system (GIS) techniques to study wetland processes. They have made major con-
tributions to mapping wetlands on the ground and through remote sensing by developing new
techniques using hyperspectral, synthetic aperture radar, and LiDAR data. The editors have also
applied remote sensing methods to monitor changes in wetlands and to predict their function at
the landscape scale.
Ralph W. Tiner recently retired from the U.S. Fish and Wildlife Service (FWS) where he served
as regional wetland coordinator for the Northeast Region, Hadley, Massachusetts. He has been
mapping wetlands since 1970, first as a graduate student at the University of Connecticut, then with
the State of South Carolina, and for the past 37 years for the FWS’s National Wetlands Inventory
Program. He is a wetland ecologist and the author of more than 200 publications including sev-
eral books—a few field guides to wetland plants, Wetland Indicators (Lewis Publishers), and most
recently Tidal Wetlands Primer (UMass Press). Through his participation on three national com-
mittees, he has helped standardize on-the-ground wetland delineation practices for implementing
federal wetland regulations in the United States. He is editor of Wetland Science and Practice
(a journal of the Society of Wetland Scientists) and was recently chosen as treasurer for the U.S.
National Ramsar Committee, working to designate Wetlands of International Importance and fur-
ther wetland conservation and education in the United States.
Megan W. Lang is a research associate professor at the Department of Geographical Sciences,

University of Maryland, College Park, Maryland. Her career in wetland sciences began in 1996, as
an undergraduate student researching succession in abandoned rice fields while at the College of
Charleston, South Carolina. Megan leads the U.S. Department of Agriculture Mid-Atlantic Wetland
Conservation Effects Assessment Project. She is also a lead scientist on multiple grants, serves as
an associate editor for the journal Wetlands, and has published over 40 scientific articles and book
chapters. Megan is cofounder of the U.S. Association of State Wetland Managers Wetland Mapping
Consortium and has helped to develop wetland monitoring strategies for the state of Maryland and
the United States.
Victor V. Klemas is professor emeritus at the School of Marine Science and Policy, University of
Delaware, Newark, Delaware. Since 1976, he has directed the university’s Center for Remote Sensing,
where he pioneered the application of a wide range of remote sensing techniques in the study of
wetland and estuarine ecosystems in the United States and overseas. One result of his research has
been the development of satellite remote sensing techniques for studying wetland stress and biomass
changes. Dr. Klemas has published his research results in 120 scientific journal articles and coau-
thored several books, including the most recent one, Remote Sensing and Global Environmental
Change (Wiley-Blackwell). He has served on six scientific committees of the National Research
Council (National Academy of Science) and various government advisory panels.
xi
Contributors
John R. Andrews Jennifer M. Corcoran
Bureau of Economic Geology Division of Forestry
Jackson School of Geosciences Minnesota Department of Natural Resources
The University of Texas at Austin Saint Paul, Minnesota
Austin, Texas
David L. Cotten
Elizabeth C. Banda Department of Geography
Michigan Tech Research Institute Center for Geospatial Research
Michigan Technological University University of Georgia
Ann Arbor, Michigan Athens, Georgia
Michael J. Battaglia Antonio Difebo

Michigan Tech Research Institute Department of Geography and Environmental
Michigan Technological University Management
Ann Arbor, Michigan University of Waterloo
Waterloo, Ontario, Canada
Sergio Bernardes
Department of Geography Sarah L. Endres
Center for Geospatial Research Michigan Tech Research Institute
University of Georgia Michigan Technological University
Athens, Georgia Ann Arbor, Michigan
Laura L. Bourgeau-Chavez Jefferson Ferreira-Ferreira

Michigan Tech Research Institute Coordination of Information Technology and
Michigan Technological University Communication
Ann Arbor, Michigan Mamirauá Sustainable Development Institute
Tefé, Brazil
Brian Brisco
Canada Centre for Remote Sensing Luiz Felipe de Almeida Furtado
Natural Resources Canada Remote Sensing Division
Ottawa, Ontario, Canada Coordination for Earth Observation
National Institute for Space Research
Colin N. Brooks São José dos Campos, Brazil
Michigan Tech Research Institute
Michigan Technological University Shuvankar Ghosh
Ann Arbor, Michigan Department of Geography
University of Georgia
Daniel Clewley Athens, Georgia
Ming Hsieh Department of Electrical
Engineering Thomas Gumbricht
University of Southern California Birger Jarlsgatan 102b
Los Angeles, California Stockholm, Sweden
xiii
xiv Contributors
Sang-Hoon Hong Zachary M. Laubach

Division of Marine Geology and Geophysics Michigan Tech Research Institute
Rosenstiel School of Marine and Atmospheric Michigan Technological University
Science Ann Arbor, Michigan
University of Miami
Miami, Florida
Robb D. Macleod
and Ducks Unlimited, Inc.
Satellite Information Research Institute Ann Arbor, Michigan
Korea Aerospace Research Institute
Daejeon, South Korea
Marguerite Madden
Department of Geography
Mark Hoover Center for Geospatial Research
University of Connecticut University of Georgia
Storrs, Connecticut Athens, Georgia
Carol A. Johnston Kyle McDonald

Department of Natural Resource Management Department of Earth and Atmospheric Sciences
South Dakota State University City College of New York
Brookings, South Dakota City University of New York
New York, New York
Thomas Jordan
Department of Geography John Melack
Center for Geospatial Research Bren School of Environmental Science and
University of Georgia Management
Athens, Georgia University of California, Santa Barbara
Santa Barbara, California
Victor V. Klemas
School of Marine Science and Policy Mary Ellen Miller
University of Delaware Michigan Tech Research Institute
Newark, Delaware Michigan Technological University
Ann Arbor, Michigan
Joseph F. Knight
Remote Sensing and Geospatial Analysis Deepak R. Mishra
Laboratory Department of Geography
Department of Forest Resources University of Georgia
University of Minnesota Athens, Georgia
Saint Paul, Minnesota
Mahta Moghaddam
Anthony J. Landon Ming Hsieh Department of Electrical
Michigan Tech Research Institute Engineering
Michigan Technological University University of Southern California
Ann Arbor, Michigan Los Angeles, California
Megan W. Lang David M. Mushet

Department of Geographical Sciences Northern Prairie Wildlife Research Center
University of Maryland U.S. Geological Survey
College Park, Maryland Jamestown, North Dakota
Contributors xv
Zhenguo Niu Sam Purkis

Institute of Remote Sensing and Digital Earth National Coral Reef Institute
Chinese Academy of Sciences Nova Southeastern University
Haidian, Beijing, People’s Republic of China Dania Beach, Florida
Nancy O’Hare Lian P. Rampi

Department of Geography Remote Sensing and Geospatial Analysis
Center for Geospatial Research Laboratory
University of Georgia Department of Forest Resources
Athens, Georgia University of Minnesota
Saint Paul, Minnesota
Jeffrey G. Paine
Bureau of Economic Geology Elijah Ramsey III
Jackson School of Geosciences National Wetlands Research Center
The University of Texas at Austin U.S. Geological Survey
Austin, Texas Lafayette, Louisiana
Fabrice Papa
Amina Rangoonwala
Laboratory for Space Geophysics and
National Wetlands Research Center
Oceanography
U.S. Geological Survey
Institut for Research and Development
Lafayette, Louisiana
Toulouse, France
and Murray Richardson
Indo-French Cell for Water Sciences Department of Geography and Environmental
IRD-IISc Joint International Laboratory Studies
Indian Institute of Science Carleton University
Bangalore, India Ottawa, Ontario, Canada
Alessandro Pasqua Chris Roelfsema

Department of Geography School of Geography, Planning and
Center for Geospatial Research Environmental Management
University of Georgia The University of Queensland
Athens, Georgia Saint Lucia, Queensland, Australia
Keith C. Pelletier
Remote Sensing and Geospatial Analysis Jennifer Rover
Laboratory Earth Resources Observation and Science
Department of Forest Resources Center
University of Minnesota U.S. Geological Survey
Saint Paul, Minnesota Sioux Falls, South Dakota
Jonathan Price Kutalmis Saylam

Department of Geography and Environmental Bureau of Economic Geology
Management Jackson School of Geosciences
University of Waterloo The University of Texas at Austin
Waterloo, Ontario, Canada Austin, Texas
Catherine Prigent Thiago Sanna Freire Silva

National Center for Scientific Research Institute for Geosciences and Exact Sciences
Paris Observatory São Paulo State University
Paris, France Rio Claro, Brazil
xvi Contributors
Annia Susin Streher Shimon Wdowinski

Coordination of Information Technology and Division of Marine Geology and Geophysics
Communication Rosenstiel School of Marine and Atmospheric
Mamirauá Sustainable Development Institute Science
Tefé, Brazil University of Miami
Miami, Florida
Ralph W. Tiner
National Wetlands Inventory (retired)
U.S. Fish and Wildlife Service
Hadley, Massachusetts Adam Walton Whelchel
and The Nature Conservancy
Institute for Wetland & Environmental New Haven, Connecticut
Education & Research
Leverett, Massachusetts
Thomas A. Tremblay Jane Whitcomb

Bureau of Economic Geology Ming Hsieh Department of Electrical
Jackson School of Geosciences Engineering
The University of Texas at Austin University of Southern California
Austin, Texas Los Angeles, California
Acronyms
ALOS Advanced Land Observing Satellite
ASTER Advanced spaceborne thermal emission and reflection radiometer
AVHRR Advanced very high resolution radiometer
AVIRIS Advanced visible/infrared imaging spectrometer
CIR Color infrared
DEM Digital elevation model
EM Electromagnetic
ETM+ Enhanced Thermal Mapper Plus
GDEM Global digital elevation model
GIEMS Global Inundation Extent from Multi-Satellite
GIS Geographic information system
GLDW Global lakes and wetlands dataset
GPS Global positioning system
HGM Hydrogeomorphic method
HH Horizontal transmitted and horizontal received (polarization mode)
HH+HV Horizontal transmitted and two types of receptions, horizontal and vertical
InSAR Interferometric synthetic aperture radar
LAI Leaf area index
LiDAR Light detection and ranging
MIR Mid-infrared
MODIS Moderate resolution imaging spectroradiometer
MSS Multispectral scanner
NASA National Aeronautics and Space Administration
NDVI Normalized difference vegetation index
NDWI Normalized difference water index
NIR Near infrared
NLCD National Land Cover Data
NWI National Wetlands Inventory
OBIA Object-based image analysis
PALSAR Phased Array L-Band Synthetic Aperture Radar
PolSAR Polarimetric synthetic aperture radar
SAR Synthetic aperture radar
SfM Structure from motion
SPOT HRV Systeme Pour l’Obervation de la Terre, high resolution visible
SRTM Shuttle Radar Topography Mission
TM Thematic Mapper
UAS Unmanned aerial system
VV Vertical transmitted and vertical received (polarization mode)
VV+VH Vertical transmitted and two types of receptions, vertical and horizontal
xvii
Section I
Introduction to the Use of Remote
Sensing for Wetland Mapping
1 An OverviewWetlands
Ralph W. Tiner
CONTENTS
Introduction......................................................................................................................................... 3
Definitions and Concept...................................................................................................................... 3
Hydrology........................................................................................................................................... 6
Factors Affecting Wetland Formation................................................................................................. 7
General Wetland Types..................................................................................................................... 10
Global Distribution of Wetlands....................................................................................................... 11
Threats to the World’s Wetlands....................................................................................................... 13
Conclusion........................................................................................................................................ 14
References......................................................................................................................................... 16
Further Readings............................................................................................................................... 18
INTRODUCTION
While many indigenous people and rural populations still depend on local wetlands and waters for
their livelihood (e.g., Schuyt 2005), most if not all nations now recognize wetlands as one of the
world’s most valuable natural resources. Besides their intrinsic value to specially adapted plants and
animals, wetlands produce a wealth of environmental services that directly benefit human societies
(Table 1.1). Given this recognition, there has been increasing attention given to wetland conserva-
tion around the globe (e.g., Millennium Ecosystem Assessment 2005). Wetlands have been acquired
as nature preserves, sanctuaries, and wildlife management areas or identified as globally important
for water birds or vital for storing carbon to help ameliorate the effect of human-induced greenhouse
gases on atmospheric temperature. Wetland restoration has become a subject for scientific investiga-
tion as well as for commercial enterprise. Some governments have enacted laws and/or policies to
protect wetlands or regulate conversion of these valued resources and initiated wetland inventories
to aid wetland management and conservation. Since this book focuses on advances in remote sens-
ing technology to perform such inventories, a brief overview of wetlands seems warranted. This
chapter is intended to provide that introduction and includes a list of references for readers to gain a
more complete understanding of wetlands.
DEFINITIONS AND CONCEPT

Wetlands are lands that are at least periodically wet during the growing season in most years or
in arid regions during the wet phase of the hydrologic cycle (e.g., episodic or ephemeral wetlands).
They range from permanently flooded areas (i.e., open water) to lands that are flooded or satu-
rated for extended periods usually with some frequency (Figure 1.1). Technical definitions vary to
some degree but have much in common (Table 1.2). The presence of water may be permanent or
temporary, but if the latter, it must be present long and often enough to affect plant establishment,
soil development, animal life, and/or the variety of functions attributed to these natural resources.
3
4 Remote Sensing of Wetlands: Applications and Advances
TABLE 1.1
Wetland Functions and Values
Function Values
Water storage Flood- and storm-damage protection, water source during dry season, peat deposits, fish and
shellfish habitat, waterfowl and water bird habitat, recreational boating, fishing, shellfishing,
waterfowl hunting, fur-bearer trapping, nature photography, aesthetic appreciation
Nutrient retention and Water-quality improvement, increases in plant productivity, aquatic productivity, decreases in
transformation eutrophication, pollutant abatement, global cycling of nitrogen, phosphorus, and sulfur
Carbon sequestration Global cycling of methane and carbon dioxide, peat deposits, greenhouse gas reduction
Sediment and particulate Water-quality renovation, reduced sedimentation of waterways, pollution abatement
retention (contaminant retention)
Streamflow maintenance Support of aquatic life in rivers, streams, and lakes; fish and aquatic invertebrate habitat;
waterfowl and water bird habitat; transportation routes; water supply
Provision of substrate for Shoreline stabilization, reduction of flood crests and water’s erosive potential, plant-biomass
plant colonization productivity, edible plants, organic export, aquatic productivity, fish nursery grounds, bird
breeding habitat, other fish and wildlife habitat, trapping, hunting, fishing, nature
observation, production of timber (mangroves and tidal swamps), production of salt hay,
reed and other natural products, scientific study, environmental education, nature
photography, aesthetic appreciation
Notes: It is important to recognize that (1) all wetlands do not provide all functions; (2) functions are not necessarily per-
formed continuously, but are often performed on a frequent and recurring basis; (3) wetlands typically have a limited
capacity to perform functions; and (4) the maximum value of wetlands is realized in their collective capacity to
perform functions at landscape scales (e.g., watershed).
Wetlands include the shallow water zone of deepwater habitats, permanently saturated lands (e.g.,
bogs and fens), temporarily inundated floodplains and depressions, and other lands with seasonal
high water tables (e.g., seasonally saturated forests and meadows). While most wetlands are colo-
nized by vascular plants, others may only support microscopic algae and other aquatic-dependent
organisms. Some wetland definitions include lakes and rivers as wetlands, while others do not. In
the United States, freshwater aquatic systems or the parts of those systems that are greater than
2.5 m deep are classified as “deepwater habitats” and are not wetlands unless they support per-
sistent self-supporting vegetation (Wetlands Subcommittee 2013). In contrast, the wetland defini-
tion adopted at the Ramsar Convention on Wetlands of International Importance Especially as
Waterfowl Habitat includes deepwater lakes and rivers as freshwater wetlands as well as the shallow
water zone of marine waters. Wetlands represent the collection of the world’s wet habitats that are
typically referred to by a host of terms including marshes, swamps, bogs, fens, and aquatic beds in
North America and by these and other names globally.
Wetlands are often described as ecotones—transitional habitats situated between dry land
(upland) and a water body (e.g., estuary, lake, river, or stream) (e.g., Mitsch and Gosselink 2007).
If wetlands were always associated with permanent open water, they would be more easily identi-
fied. While many wetlands occur as fringing habitats on the floodplains of water bodies, many
others are completely surrounded by dry land with no direct surface connection to open water.
The latter types have been referred to as “geographically isolated wetlands” (Tiner 2003). Yet
even they may be connected to other waters via intermittent surface flowage or groundwater inter-
actions. Although some wetlands are ecotonal habitats sharing species with neighboring habitats,
wetlands possess many species that are unique to these ecosystems and are therefore considered
separate ecosystems (Tiner 1999). Wetlands may be modified for agricultural or silvicultural uses
or created for these and other purposes (e.g., rice paddies, commercial cranberry bogs, and ponds
Wetlands 5
(a) (b)
(c) (d)
(e) (f)
FIGURE 1.1 Wetlands vary in their source and degree of wetness. (a) Permanently flooded aquatic bed with
saturated forested bog in background (Nova Scotia, Canada). (b) Semipermanently flooded marsh (Florida,
United States). (c) Seasonally flooded southern swamp (South Carolina, United States). (d) Temporarily
flooded floodplain forest in the spring (note the silt coating on trees from recent flood) (Massachusetts, United
States). (e) Seasonally saturated flatwood in the spring (Virginia, United States). (f) Continuously saturated
shrub bog (Maine, United States). (Photos courtesy of Ralph Tiner.)
constructed for storm water and wastewater treatment, aquaculture [shrimp and fish], and live-
stock watering). They may also be created by beaver or be the result of unintentional activities
(e.g., undersized culverts). Wetlands may form in other altered landscapes, such as mined lands
(e.g., abandoned sand pits). These variable conditions produce a diversity of wet habitats and plant
communities around the globe, some of which are particularly challenging to identify both on the
ground and through remote sensing (see Chapter 3).
TABLE 1.2
Examples of Wetland Definitions
Geographic Applicability
Wetland Definition (Source)
“Areas of marsh, fen, peatland, or water, whether natural or artificial, permanent or International (Ramsar
temporary, with water that is static or flowing, fresh, brackish, or salt, including areas of Bureau 1998)
marine water the depth of which at low tide does not exceed 6 m … may incorporate
riparian and coastal zone adjacent to wetlands, and islands or bodies of marine water
deeper than 6 m at low tide lying within the wetlands”
“Areas of seasonally, intermittently, or permanently waterlogged soils or inundated land, Australia (Semeniuk and
whether natural or otherwise, fresh or saline” Semeniuk 1995)
“Land that is saturated with water long enough to promote wetland or aquatic processes as Canada (Warner and
indicated by poorly drained soils, hydrophytic vegetation, and various kinds of biological Rubec 1997)
activity which are adapted to a wet environment”
“Includes permanently or intermittently wet areas, shallow water, or land water margins New Zealand (Johnson and
that support a natural ecosystem of plants and animals that are adapted to wet conditions” Gerbeaux 2004)
“Lands transitional between terrestrial and aquatic systems where the water table is United States (Cowardin
usually at or near the surface or the land is covered by shallow water” et al. 1979)
“Land which is transitional between terrestrial and aquatic systems where the water table South Africa (S.A. National
is at or near the surface, or the land is periodically covered with shallow water, and which Water Act; No. 36 of
land in normal circumstances supports or would support vegetation typically adapted to 1998; Ollis et al. 2013)
life in saturated soils”
HYDROLOGY
Wetlands derive water from many sources, including groundwater, river overflow, surface water
runoff, precipitation, snowmelt, tides, melting permafrost, and seepage from impoundments or
irrigation projects or are artificially flooded by pumps and siphons or other means (Figure 1.2).
The frequent occurrence of extended wetness (i.e., soil saturation or inundation) typically creates
anaerobic soil conditions that affect plants and animals alike as well as soil development (hydric
soil morphology). Variations in hydrology coupled with other factors create many different wetland
types (Figures 1.1 and 1.3).
Tidal wetlands are often characterized by two zones: the regularly flooded zone and the irregu-
larly flooded zone. The former is typically flooded daily while the latter is inundated less frequently
(i.e., every few days, a few times a month, or less often). Local weather conditions (e.g., storms,
rainfall, and wind direction) play an important role in tidal flooding. Figure 1.4 shows examples of
tidal marsh and mangrove swamp, the two most common vegetated tidal wetlands.
Given the diversity of inland or nontidal wetlands around the globe, the frequency and duration
of their wetness vary with local groundwater conditions, geomorphology, climate (e.g., regional
precipitation patterns), and other factors. Worldwide, some wetlands, such as marshes, bogs, fens,
and organic-rich swamps, are typically permanently wet or nearly so; in higher latitudes and eleva-
tions, they may be snow covered or frozen for much of the year. Other wetlands are only season-
ally wet, and for some, the degree of wetness may change markedly from year to year (e.g., arid
regions). In humid temperate midlatitudes, wetlands tend to have high water tables from late winter
to early summer. Seasonal wetlands in Mediterranean climates may be wettest during the winter and
extremely dry in summer, whereas in other regions experiencing frequent droughts, monsoons bring
heavy rains thereby reinvigorating wetlands. Interannual differences in precipitation cause variable
inundation in wetlands, often accompanied by shifts in vegetation patterns (e.g., U.S. Great Lakes
and prairie pothole wetlands). While hydrology may be generally predicted based on past trends, the
hydrology of all wetlands is by its nature dynamic, subject to daily, seasonal, and annual variations.
Wetlands 7
SW
I ET
P
High water
table
S
Low water
table
GWI SWO
G
W
O
FIGURE 1.2 Wetlands form where water inputs exceed outputs for prolonged periods; thus, water is stored
at, near, or above the soil surface. Inputs: precipitation (P), surface water inflow (SWI), and groundwater
inflow (GWI = discharge). Outputs: evapotranspiration (ET), surface water outflow (SWO), and groundwa-
ter outflow (GWO = recharge). ΔS represents change in storage. The diagram is for a nontidal wetland as
tidal effects are not shown. (From Carter, V., Wetland hydrology, water quality, and associated functions, in:
Fretwell, J.D., Williams, J.S., and Redman, P.J. [compilers], National Water Summary on Wetland Resources,
U.S. Geological Survey, Reston, VA, 1996, pp. 35–48.)
In the United States, the minimum wetness for a federally regulated wetland has been defined
by saturation within 30 cm of the surface for two weeks or more during the growing season in most
years (i.e., every other year on average; National Research Council 1995). While this definition may
be appropriate for temperate regions, it does not address the creation of what may be called “ephem-
eral” wetlands in arid regions where episodic rainfall during the wet phase of the hydrologic cycle
forms wetlands on broad flats or depressions that later dry out and are exposed, dry substrates for
years (e.g., wetlands associated with episodic lakes in Australia).
FACTORS AFFECTING WETLAND FORMATION

Many factors interact to produce areas that are wet enough for wetland establishment and to affect
the types of plants that can colonize such sites. Wetlands typically form in certain positions on the
landscape where surface or tidal water collects and/or where groundwater is discharged (e.g., hot,
warm, or cold springs or seeps). They are associated with floodplains (overflow lands), depres-
sions, broad flats (e.g., coastal or glaciolacustrine plains), or slopes in areas of active groundwater
discharge. Wetlands may form by a successional process called “terrestrialization” where an open
water body (e.g., lake or pond) becomes shallower as sediments and organic matter from the remains
of algae and aquatic bed plants settle to the bottom creating an organic substrate (gyttja), while
emergent hydrophytes (helophytes) expand outward from the shore and eventually cover the entire
surface of what once was the water body (Dansereau and Segadas-Vianna 1952; Marek 1992). As
noted previously, precipitation patterns characteristic of regional climates strongly influence hydrol-
ogy and in some places can cause wetlands to form in landscapes where they would not typically
form. For example, the wetland-forming process of “paludification” occurs in boreal and arctic cli-
mates with low evapotranspiration rates that allow for peat mosses (Sphagnum spp.) to grow upslope
+90
+60
Marsh
+30
0 Ground level
+30
0 Ground level
–30 Wet meadow
–60
–90
+30
Bog Ground level
0
–30
+60
0 Ground level
Hardwood swamp
–60
–120
Ground level
0
Flatwood
–60
–120
+120
+60
0 Ground level
Floodplain forest
–60
–120
Jan Feb Mar Apr May June July Aug Sep Oct Nov Dec
FIGURE 1.3 Simplified hydrographs showing relative changes in the water table through an average year for
some common wetland types in eastern North America. Units are centimeters; a plus (+) indicates the pres-
ence of surface water, whereas a minus (−) indicates the position of the water table below the ground surface.
(From Tiner, R.W., In Search of Swampland: A Wetland Sourcebook and Field Guide, Rutgers University
Press, New Brunswick, NJ, 2005.)
creating a wetland landscape of hills and valleys (e.g., blanket bogs; Figure 1.5). The world’s peat-
lands (organic soils) are mostly found in these regions, with the tropical swamp forests of Southeast
Asia (Malaysia) also having extensive peat formations. Along the world’s tidal shores, sea-level rise
is causing a swamping of lowland forests, allowing tidal wetlands to “migrate” landward. Episodic
heavy precipitation events during the wet phase of the hydrologic cycle in arid regions create condi-
tions for wetland formation in areas that have been dry for many years. Climate also affects water
Wetlands 9
(a) (b)
(c)
FIGURE 1.4 In temperate and boreal regions, salt marshes are the most common vegetated tidal wetland
(a, Massachusetts; b, South Carolina), (a and b: Photos courtesy of Ralph Tiner.) while in the tropics, mangrove
swamps prevail (c, Cambodia; Photo courtesy of Leon Petrosyan).
FIGURE 1.5 Alaskan blanket bog, a product of paludification. (Photo courtesy of David Cooper.)
U.S. Dept. of Agriculture

Natural Resources Conservation Service
Soil Survey Division
World Soil Resources Distribution of wetlands
Upland
Lowland
Organic
Salt affected Miller projection
Scale 1:100,000,000
Permafrost affected 0 500 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000
Inland water bodies km
No wetlands (or too small to display)
Country boundaries are not authoritative. Washington D.C. 1996
FIGURE 1.6 Generalized global distribution of wetland types. (From USDA Natural Resources
Conservation Service.)
and soil salinity. In arid climates, high evapotranspiration rates coupled with low precipitation have
produced an abundance of saline wetlands where salt stress and hydrology operate to affect plant
communities or the lack thereof (i.e., salt barrens; Figure 1.6). The vegetation of wetlands subject to
flooding by ocean tides is also strongly influenced by salt water. The chemistry of groundwater dis-
charged into wetlands is another important factor affecting plant composition. In limestone regions,
nutrient-rich groundwater produces more alkaline conditions (hydrogen ion concentration, pH > 7.4)
promoting colonization by calciphiles (calcium-loving plants). In other region, soils of wetlands
not subject to such enrichment are acidic (pH < 5.5) or circumneutral (pH = 5.5–7.4). Soil type also
affects wetland establishment. Clayey soils, impervious rock, or other restrictions near the surface
may create a perched water table that promotes wetland formation.
GENERAL WETLAND TYPES

Wetlands may be classified in various ways depending on one’s interest (see Chapter 2). From a general
hydrologic standpoint, they can be separated into tidal and nontidal wetlands or by their soils as peat-
lands (organic soils) or nonpeatlands (mineral soils). From the water chemistry perspective, they can
be saline or nonsaline with the latter further divided into acid, neutral, and alkaline types. Wetlands
can also be categorized by a combination of their vegetation and other factors (Table 1.3). They can
also be further separated by other factors summarized in Chapter 2 including landscape position (e.g.,
marine, estuarine, lotic, lentic, or terrene), landform (e.g., basin, flat, floodplain, interfluve, peatland,
slope, highland/hill, swale, and delta), their connection to surface waters (e.g., throughflow, outflow,
inflow, or isolated), and their hydrology (e.g., permanently flooded, semipermanently flooded, season-
ally flooded, temporarily flooded, seasonally saturated, continuously saturated, or tidally flooded).
Common in arid and semiarid regions, “floodout” wetlands are associated with inland or closed
drainage systems where seasonal flows inundate a broad floodplain (e.g., alluvial fan plain or inland
delta) with Botswana’s Okavango Delta being an excellent example (Figure 1.7a; see Chapter 20).
Wetlands 11
TABLE 1.3
Major Types of Wetlands
Type Brief Description
Tidal flat Nonvegetated shore flooded daily by the tides.
Salt flat Nonvegetated wetland where significant accumulations of salts restrict the growth of most vascular
plants; they occur in estuaries where they are infrequently flooded by tides or form inland in
semiarid and arid regions (also called “salinas”).
Mangrove swamp Wetland dominated by shrubs and/or trees in several families (e.g., Rhizophoraceae, Verbenaceae,
Combretaceae, Pellicieraceae, Meliaceae, Acanthaceae, and Sonneratiaceae) typically found along
tropical and subtropical coasts.
Tidal salt marsh Herb-dominated wetland in estuaries and saltwater lagoons subject to frequent or infrequent tidal
flooding, often dominated by halophytic (salt-tolerant) plants.
Freshwater marsh Herb-dominated wetland in shallow water or in areas flooded for long periods during the year, often
found along margins of freshwater lakes, ponds, rivers and streams.
Inland saline Similar to freshwater marsh, but occurring in arid and semiarid regions where surface water is high in
marsh salts creating inland saline soils.
Wet meadow Herb-dominated freshwater wetland in seasonally saturated soils, may be briefly inundated and
duration of saturation may be temporary or prolonged.
Inland saline Similar to wet meadow, but occurring in arid and semiarid regions where evapotranspiration exceeds
meadow precipitation creating saline soil conditions.
Fen A peat-forming, nearly continuously saturated wetland where groundwater discharge often enriches
the soil chemistry to create more alkaline (calcareous) conditions; surface water may be present;
plant community is often characterized by calciphiles; plant community ranges from herb-
dominated to forested.
Bog A peat-forming, nearly continuously saturated wetland formed under nutrient-poor (acidic)
conditions; surface water if present is limited to depressions; may occur in isolated depressions or
along streams but hydrologically decoupled from them with most of its nutrients obtained from rain,
snow, fog, or dust; plant community ranges from shrub-dominated bogs with ericaceous evergreen
shrubs to black spruce and other evergreen forests, with peat mosses forming the base substrate;
carnivorous plants (e.g., Drosera and Sarracenia) may be common in these low-nutrient habitats.
Shrub swamp Freshwater wetland dominated by low-growing woody plants (e.g., less than 6 m tall) occurring
under a wide range of hydrologic conditions from permanently flooded to seasonally saturated; soils
mineral or organic depending on duration of saturation or inundation.
Forested swamp Freshwater wetland dominated by mature trees (e.g., greater than 6 m) occurring under a wide range
of hydrologic conditions from permanently flooded to seasonally saturated; soils mineral or organic
depending on the duration of saturation or inundation.
Floodplain forest Freshwater forested wetland along a river where it is subject to periodic inundation during high river
discharge periods; soils tend to mineral derived from alluvium.
Aquatic bed A wetland in mostly shallow water characterized by submerged, floating-leaved, or free-floating
plants (e.g., Najas, Vallisneria, Zostera, Potamogeton, Nuphar, Nymphaea, Victoria, Eichhornia,
Ceratophyllum, Lemna, and Azolla).
Note: These general types may be referred to by different names around the globe.
Note that the term “swamp” has sometimes been applied to herb-dominated wetlands (i.e., marshes).
Some examples of wetland types are shown in Figures 1.1, 1.4, and 1.7.
GLOBAL DISTRIBUTION OF WETLANDS

Comprehensive wetland inventories are lacking for most countries for three main reasons: (1) the
expense of conducting a nationwide inventory; (2) technical challenges to actually locate, map,
(a) (b)
(c) (d)
(e) (f)
FIGURE 1.7 Examples of some of diversity of wetlands around the globe. (a) Wetlands in Botswana’s
Okavango Delta during floodout. (Photo courtesy of Thomas Gumbricht.) (b) Uganda marsh in hot springs
area. (c) Brazilian palm swamp. (b and c: Photos courtesy of Aram Calhoun.) (d) Peruvian fen. (e) Labrador
patterned peatland (note: “strings” indicate low ridges and “flarks” water bodies aligned perpendicular to water
flow; Photo courtesy of Richard LaPaix). (f) Slovakian rich fen. (d and f: Photos courtesy of David Cooper.)
and quantify all the various types of wetlands through remote sensing techniques (see Chapter
3 for details); and (3) other priorities for funding (Finlayson et al. 2001). Furthermore, the inven-
tories that have been conducted are not consistent in their methodology, wetland definition, and
wetland classification making the results of comparative analysis tentative. North America and
parts of Western Europe are probably the only regions with adequate inventories of their wetlands,
Wetlands 13
TABLE 1.4
Estimated Area of Wetlands in Various Regions
Region Area (ha) Source
Africa 121,321,683–124,686,189 Stevenson and Frazier (1999a)
Asia 211,501,790–224,117,790 Watkins and Parish (1999)
Oceania 4,792,700+ Watkins (1999)
Europe
Eastern 225,821,930 Stevenson and Frazier (1999b) (errata)
Western 28,821,979 Stevenson and Frazier (1999c)
Middle East 7,434,790 Frazier and Stevenson (1999)
North America
Canada 127,199,000–150,000,000 Davidson et al. (1999b)
United States 114,544,800 Tiner (1999)
Neotropics 414,996,613 Davidson et al. (1999a)
Total 1,295,216,791
recognizing that even those inventories are not complete and have significant limitations. Data for
tropical, subtropical, and southern temperate regions are generalized or nonexistent. Remote sens-
ing technology has greatly assisted with generating improved information on the status of wetlands
around the globe and advances in remote sensing technology are well positioned to improve our
ability to inventory wetlands.
Based on existing data, about 1.3 billion hectares of wetlands occur worldwide (Table 1.4).
Wetlands cover about 1%–2% of the Earth’s surface or somewhere between 3% and 8% of its land
surface depending on the estimate used. Peatlands may represent one-third of the world’s wetlands,
occupying more than 400 million hectares. Nearly half of the world’s wetlands may occur between
50°N and 70°N latitude in boreal and arctic regions where bogs and fens are abundant (Figure 1.6).
These two types are the two most abundant wetland types in the world (Table 1.5). More than
one-third of the Earth’s wetlands may lie between 20°N and 30°S latitude where forested wetlands
and marshes prevail including mangrove swamps, tropical rainforests, and floodplain wetlands.
The remaining 20% may be found in temperate regions. Ten countries have over 2 million hectares
of peatlands alone, with Canada leading at nearly 130 million hectares (about 18% of the country)
followed by the former U.S.S.R. at 83 million hectares. About a third of Finland is covered by
peatlands (10 million hectares). The Pantanal of South America, referred to by some as the largest
wetland in the world, reportedly covers 140,000–210,000 km2 (or 14–21 million hectares) (Swarts
2000). Current estimates report 15.2 million hectares of mangroves distributed in 123 countries and
territories (Spalding et al. 2010).
THREATS TO THE WORLD’S WETLANDS

Despite recognition of their many uses by people, their environmental services to humankind
(e.g., flood storage, shoreline stabilization, carbon sequestration, and water-quality renovation),
and their ecological significance, natural wetlands continue to be converted to nonwetlands and
subjected to pollution (Table 1.6). In some countries, environmental laws or policies that pro-
mote wetland conservation have significantly reduced the rate of wetland loss, yet in many other
countries, wetlands are still being destroyed at alarming and increasing rates. As much as 50%
of the world’s wetlands may have been lost (Russi et al. 2013). According to the Organisation for
Economic Co-operation and Development, drainage for agricultural production has been the prin-
cipal cause of global wetland loss (OECD Development Assistance Committee 1996). By 1985,
TABLE 1.5
Estimated Worldwide Area of Certain Wetland Types Plus Lakes and Coral Reefs
Wetland Type Global Area (ha)
Bogs 190,000,000–297,500,000
Fens 150,000,000
Swamps 110,000,000
Floodplains 80,000,000
Marshes 27,000,000
Lakes 12,000,000
Rice paddies 130,000,000
Mangroves 18,000,000
Coral reefs 30,000,000–60,000,000
Total 747,000,000–884,500,000
Source: Finlayson, C.M. et al. (eds.), Wetland inventory, assessment and monitoring: Practical techniques and
identification of major issues, in: Proceedings of Workshop 4, Second International Conference on
Wetlands and Development, November 8–14, 1998, Dakar, Senegal, Environment Australia, Supervising
Scientist, Darwin, Northern Territory, Australia, Supervising Scientist Report No. 161, 2001.
Note: Data include Africa, Asia, Europe, Neotropics (Caribbean, South America, and Central America),
Oceania, and North America.
it was estimated that 56%–65% of European and North American wetlands had been drained for
intensive agriculture, 27% for Asia, 6% for South America, and 2% for Africa, for a worldwide
loss of 26% due to agriculture. Tropical wetlands may be in greatest jeopardy for future devel-
opment as mangroves are harvested for timber or fuel; converted to aquaculture impoundments
(e.g., shrimp or fish ponds), salt farms, and agricultural land (e.g., for rice, cassava, palm, coconut,
or rubber); filled in urban areas for real estate and port expansion; and degraded by pollution or
altered hydrology (e.g., irrigation and freshwater diversions). It was estimated that by 2001 about
35% of the world’s mangroves had disappeared (Valiela et al. 2001). Sixteen of the world’s 70
known species of mangroves are now on the verge of extinction (Polidoro et al. 2010). Inland
tropical wetlands are being converted to agricultural land and adversely affected by reservoir
construction, water diversions, urban development, and pollution from multiple sources (e.g., in
Brazil; Diegues 1994). A recent study using remote sensing techniques found that the world’s wet-
lands have declined by 6% between 1993 and 2007, largely due to human development (Prigent
et al. 2012). Besides human actions, climate change is having a significant impact on wetlands in
most regions (Erwin 2009). Arctic lakes and thermokarst ponds are disappearing in the region of
discontinuous permafrost, while open water bodies are increasing in the continuous permafrost
zone due to thawing of permafrost (Smith et al. 2005; Smol and Douglas 2007). Coastal wetlands
are exposed to rising sea levels that are permanently inundating low-lying tidal flats, marshes,
and mangroves and causing landward shifts in vegetation patterns where more favorable condi-
tions exist (Gilman et al. 2008; Tiner 2013). Changing precipitation patterns will cause prolonged
droughts in some regions with significant adverse impacts on wetlands, while other areas will
experience wetter conditions.
CONCLUSION
As you can tell from this broad overview, wetlands are extremely diverse and are quite abundant in
some regions while uncommon in others. They are also undergoing change due to natural processes
Wetlands 15
TABLE 1.6
Some Impacts to Wetlands from Natural Processes or Activities and
Human-Induced Disturbances
Possible Impacts
Natural Processes or Activities
Sea-level rise and coastal plain Submergence (loss) of wetland, change from vegetated wetland to tidal flat/open
subsidence water, change from tidal flat to open water, change in shorelines, landward
migration of marshes if suitable space is available, increase in salinity with
changes in plant communities
Increased temperature and changes in Drying wetlands in some regions while increasing wetness and expanding
precipitation patterns due to climate wetlands in other regions; altered hydrology with effect on salinity and wetland
change vegetation; increasing Arctic ponds in continuous permafrost and decreasing
ponds in discontinuous permafrost; changes in species composition—shift
northward in Northern Hemisphere and southward in Southern Hemisphere
(e.g., mangroves); shifts in wetland usage by migratory species
Coastal processes (wave action, Erosion, sedimentation, wetland loss or gain, smothering of vegetation (e.g.,
currents, and tides) from deposition of tidal wrack or overwash sediments), and changing shorelines
Hurricanes and other storms Sedimentation (including overwash deposits), erosion, vegetation impacts,
wetland loss or gain, saltwater intrusion, and changing shorelines
Ice scour Erosion, vegetation removal, creation of open water in marsh
Grazing by animals (e.g., waterfowl, Loss of vegetation (denuded areas) and vegetation changes
furbearers, and invertebrates)
Insect infestations Loss of vegetation and vegetation changes
Disease outbreaks Loss of vegetation and vegetation changes
Beaver dams Reduce tidal flowage to tidal freshwater wetlands and vegetation changes
Fire from lightning Vegetation impacts
Droughts Brown marsh syndrome, vegetation changes
Filling for development (industrial, Loss of wetland
commercial, residential, resort, etc.)
Disposal of dredged material or Loss of wetland or change in plant community depending on amount of spoil
garbage
Construction of jetties and groins Loss of wetland, changes in wetland type, altered hydrology, shoreline changes,
and changes in wildlife use
Armoring shorelines (e.g., bulkheads Loss of wetland and stops landward migration of tidal wetland along the coast
and riprap)
Dredging for marinas and residential Loss of wetland and degraded water quality
development (canal development)
Drainage for mosquito control Altered hydrology, increased salinity, vegetation changes, and local subsidence
Diking (for many purposes) Altered hydrology, vegetation changes, localized subsidence, loss or diminished
estuarine exchange, and loss of wetland
Road and railroad crossings Loss of wetland, altered hydrology, and vegetation changes
Installation of water-control structures Altered hydrology, salinity changes (ocean coast and arid regions), and changes
(e.g., tide gates) in vegetation and aquatic life
Runoff from farms, lawns, or other Eutrophication vegetation changes and changes in aquatic life
developed areas (non-point-source
pollution)
Discharge of industrial or municipal Water pollution, fish kills, degraded water quality, increased algal blooms,
wastewater hypoxia, and changes in vegetation and aquatic life
(Continued )
TABLE 1.6 (Continued)

Some Impacts to Wetlands from Natural Processes or Activities and
Possible Impacts
Oil spills Vegetation die-off, changes in aquatic life, substrate contamination, death for
oiled wildlife, fish kills, and water pollution
Deepening channels and dredging Increase saltwater intrusion, altered hydrology, shoreline erosion, vegetation
canals for navigation changes, and change in aquatic life
Damming of rivers for power Reduction in sediment load, loss of wetland, altered hydrology, altered salinity
generation, water supply, or other regimes, vegetation changes, and changes in aquatic life (e.g., fish migration)
purposes
Water withdrawals Altered hydrology, possible subsidence, altered salinity regimes, and
corresponding changes in vegetation and aquatic life
Diversion of river flows Altered hydrology, increased salinity (at least seasonally), and corresponding
changes in aquatic life
Groundwater withdrawals Subsidence and changes in vegetation
Oil and gas withdrawals Possible subsidence, corresponding changes in vegetation and aquatic life, and
pollution from spills and leaks (see “oil spills”)
Prescribed burning for wildlife Loss of soil organic matter and vegetation changes management
Timber harvest Vegetation changes, may facilitate conversion to other uses
Log storage Topographic changes, vegetation changes, habitat degradation
Aquaculture Change in current patterns, disturbance to water birds, possible loss of wetland
(via conversion to open water)
Plant introductions Invasive species replacement of native flora and altered hydrology
Animal introductions Replacement of native fauna
Grazing by livestock Soil compaction and changes in vegetation
All-terrain vehicle traffic Vegetation impacts, disturbance of nesting birds, destruction of nests, and
incidental kill of young birds
Plant collection Loss or reduction of local populations (overharvest, e.g., sea lavender)
Nitrogen inputs from runoff Change in vegetation, plant productivity, and aquatic life
Spraying of pesticides Aquatic organism kills
Source: Tiner, R.W., Tidal Wetlands Primer: An Introduction to Their Ecology, Natural History, Status, and Conservation,
University of Massachusetts Press, Amherst, MA, 2013.
Note: With human impacts, some changes are one directional (e.g., losses due to agriculture and filling for roads and other
development). Also where a significant change in vegetation occurs, a change in wildlife use will likely occur.
and human actions. Given their value to humankind as well as to fish and wildlife, many countries
now realize the importance of these lands in their natural condition and are actively seeking to
improve wetland conservation. An important step in conserving wetlands is knowing where they
are, so that appropriate measures can be taken to better manage them. This is not a simple exercise
for many reasons (see Chapter 3). The rest of this book begins with a review of wetland classification
and then focuses on remote sensing techniques that may allow for more accurate and timely wetland
inventories in the future.
REFERENCES
Carter, V. 1996. Wetland hydrology, water quality, and associated functions. In: J.D. Fretwell, J.S. Williams, and
P.J. Redman (compilers). National Water Summary on Wetland Resources, pp. 35–48. U.S. Geological
Survey, Reston, VA, Water-Supply Paper 2425.
Wetlands 17
Cowardin, L.M., V. Carter, F.C. Golet, and E.T. LaRoe. 1979. Classification of Wetlands and Deepwater
Habitats of the United States. U.S. Fish and Wildlife Service, Washington, DC, FWS/OBS-79/31.
Dansereau, P. and F. Segadas-Vianna. 1952. Ecological study of the peat bogs of eastern North America.
Canadian Journal of Botany 30: 490–520.
Davidson, I., R. Vanderkam, and M. Padilla. 1999a. Review of wetland inventory information in the Neotropics.
In: C.M. Finlayson and A.G. Spiers (eds.). Global Review of Wetland Resources and Priorities for
Wetland Inventory. Supervising Scientist Report 144, Canberra, Australian Capital Territory, Australia.
Davidson, I., R. Vanderkam, and M. Padilla. 1999b. Review of wetland inventory information in North America.
In: C.M. Finlayson and A.G. Spiers (eds.). Global Review of Wetland Resources and Priorities for
Wetland Inventory. Supervising Scientist Report 144. Canberra, Australian Capital Territory, Australia.
Diegues, A.C.S. (ed.). 1994. An Inventory of Brazilian Wetlands. International Union for Conservation of
Nature and Natural Resources, Gland, Switzerland.
Erwin, K.L. 2009. Wetlands and global climate change: The role of wetland restoration in a changing world.
Wetlands Ecology and Management 17: 71–84.
Finlayson, C.M., N.C. Davidson, and N.J. Stevenson (eds.). 2001. Wetland Inventory, Assessment and
Monitoring: Practical Techniques and Identification of Major Issues. In: Proceedings of Workshop 4,
Second International Conference on Wetlands and Development, November 8–14, 1998, Dakar, Senegal.
Environment Australia, Supervising Scientist, Darwin, Northern Territory, Australia. Supervising
Scientist Report No. 161.
Frazier, S. and N. Stevenson. 1999. Review of wetland inventory information in the Middle East. In: C.M.
Finlayson and A.G. Spiers (eds.). Global Review of Wetland Resources and Priorities for Wetland
Inventory. Supervising Scientist Report 144, Canberra, Australian Capital Territory, Australia.
Gilman, E., J. Ellison, N.C. Duke, and C. Field. 2008. Threats to mangroves from climate change and adapta-
tion options: A review. Aquatic Botany 89(2): 237–250.
Johnson, P. and P. Gerbeaux. 2004. New Zealand Wetland Types. Department of Conservation, Te Papa Atawhai,
Wellington, New Zealand.
Marek, S. 1992. Transformation of lakes into mires. Acta Societatis Botanicorum Poloniae 61(1): 103–113.
Millennium Ecosystem Assessment. 2005. Ecosystem Services and Human Well-Being: Wetlands and Water
Synthesis. World Resources Institute, Washington, DC.
Mitsch, W.J. and J.G. Gosselink. 2007. Wetlands. John Wiley & Sons, Hoboken, NJ.
National Research Council. 1995. Wetlands: Characteristics and Boundaries. The National Academies Press,
Washington, DC.
OECD, Development Assistance Committee. 1996. DAC Guidelines on Aid and Environment—Guidelines for
Aid Agencies for Improved Conservation and Sustainable Use of Tropical and Sub-Tropical Wetlands.
Organisation for Economic Co-operation and Development, Paris, France, http://www.oecd.org/dac/
environment-development/1887748.pdf. Accessed November 4, 2014.
Ollis, D.J., C.D. Snaddon, N.M. Job, and N. Mbona. 2013. Classification System for Wetlands and Other
Aquatic Ecosystems in South Africa. User Manual: Inland Systems. South African National Biodiversity
Institute, Pretoria, South Africa. SANBI Biodiversity Series 22.
Polidoro, B.A., K.E. Carpenter, L. Collins, N.C. Duke, A.M. Ellison, J.C. Ellison, E.J. Farnsworth et al. 2010.
The loss of species: Mangrove extinction risk and geographic areas of global concern. PLoS ONE 5(4):
e10095.
Prigent, C., F. Papa, F. Aires, C. Jiménez, W.B. Rossow, and E. Matthews. 2012. Changes in land surface water
dynamics since the 1990s and relation to population pressure. Geophysical Research Letters 39: L08403.
doi:10.1029/2012GL051276.
Ramsar Convention Bureau. 1998. What are Wetlands? Ramsar Information Paper No. 1. Gland, Switzerland.
Russi, D., P. ten Brink, A. Farmer, T. Badura, D. Coates, J. Förster, R. Kumar, and N. Davidson. 2013. The
Economics of Ecosystems and Biodiversity for Water and Wetlands. IEEP, London, U.K. Ramsar
Secretariat, Gland, Switzerland.
Schuyt, K.D. 2005. Economic consequences of wetland degradation for local populations in Africa. Ecological
Economics 53: 177–190.
Semeniuk, C.A. and V. Semeniuk, 1995. A geomorphic approach to global classification for inland wetlands.
Vegetatio 118: 103–124.
Smith, L.C., Y. Sheng, G.M. MacDonald, and L.D. Hinzman. 2005. Disappearing Arctic lakes. Science
308(5727): 1429.
Smol, J.P. and M.S.V. Douglas. 2007. Crossing the final ecological threshold in high Arctic ponds. Proceedings
National Academy of Science of the United States of America 104(30): 12395–12397.
Spalding, M., M. Kainuma, and L. Collins. 2010. World Atlas of Mangroves. Earthscan, Taylor & Francis
Group, New York.
Stevenson, N. and S. Frazier. 1999a. Review of wetland inventory information in Africa. In: C.M. Finlayson
and A.G. Spiers (eds.). Global Review of Wetland Resources and Priorities for Wetland Inventory.
Supervising Scientist Report 144, Canberra, Australian Capital Territory, Australia.
Stevenson, N. and S. Frazier. 1999b. Review of wetland inventory information in Eastern Europe. In: C.M.
Stevenson, N. and S. Frazier. 1999c. Review of wetland inventory information in Western Europe. In: C.M.
Swarts, F.A. 2000. The Pantanal in the twenty-first century: For the planet’s largest wetland, an uncertain future.
In: F.A. Swarts (ed.). The Pantanal of Brazil, Bolivia, and Paraguay. Hudson MacArthur Publishers,
Waterland Research Institute, Gouldsboro, PA. (2nd edn.—The Pantanal: Understanding and Preserving
the World’s Largest Wetland; Paragon House, St. Paul, MN.)
Tiner, R.W. 1999. Wetland Indicators: A Guide to Wetland Identification, Delineation, Classification, and
Mapping. Lewis Publishers, CRC Press, Boca Raton, FL.
Tiner, R.W. 2003. Geographically isolated wetlands of the United States. Wetlands 23(3): 494–516.
Tiner, R.W. 2005. In Search of Swampland: A Wetland Sourcebook and Field Guide. Rutgers University Press,
New Brunswick, NJ.
Tiner, R.W. 2013. Tidal Wetlands Primer: An Introduction to Their Ecology, Natural History, Status, and
Conservation. University of Massachusetts Press, Amherst, MA.
Valiela, I., J.L. Bowen, and J.K. York. 2001. Mangrove forests: One of the world’s most threatened major tropi-
cal environments. BioScience 51: 807–815.
Warner, B.G. and C.D.A. Rubec (eds.). 1997. The Canadian Wetland Classification System. 2nd edn. Research
Centre, University of Waterloo Waterloo, Ontario, Canada.
Watkins, D. 1999. Review of wetland inventory information in Oceania. In: C.M. Finlayson and A.G. Spiers
(eds.). Global Review of Wetland Resources and Priorities for Wetland Inventory. Supervising Scientist
Report 144. Canberra, Australian Capital Territory, Australia.
Watkins, D. and F. Parish. 1999. Review of wetland inventory information in Asia. In: C.M. Finlayson and A.G.
Spiers (eds.). Global Review of Wetland Resources and Priorities for Wetland Inventory. Supervising
Scientist Report 144. Canberra, Australian Capital Territory, Australia.
Wetlands Subcommittee. 2013. Classification of Wetlands and Deepwater Habitats of the United States. 2nd
edn. Federal Geographic Data Committee, Reston, VA, August 2013. FGDC-STD-004-2013.
FURTHER READINGS
Batzer, D.P. and A. Baldwin (eds.). 2010. Wetland Habitats of North America: Ecology and Conservation
Concerns. University of California Press, Berkeley, CA.
Fraser, L.H. and P.A. Keddy (eds.). 2005. The World’s Largest Wetlands: Ecology and Conservation. Cambridge
University Press, Cambridge, UK.
Gore, A.J.P. (ed.). 1983. Mires: Swamp, Bog, Fen and Moor, Regional Studies. Ecosystems of the World 4B.
Elsevier Scientific Publishing Co., Amsterdam, the Netherlands.
Hogarth, P. 2007. The Biology of Mangroves and Seagrasses. Oxford University Press, Oxford, U.K.
International Water Management Institute. 2014. Wetlands and People. Colombo, Sri Lanka, http://www.iwmi.
cgiar.org/Publications/Books/PDF/wetlands-and-people.pdf. Accessed November 4, 2014.
Junk, W.J. (ed.). 1997. The Central Amazon Floodplain. Ecology of a Pulsing System. Ecological Studies
No. 126. Springer-Verlag, Berlin, Germany.
Keddy, P.A. 2012. Wetland Ecology: Principles and Conservation. Cambridge University Press, Cambridge, UK.
Mitsch, W.J. and J.G. Gosselink. 2007. Wetlands. John Wiley & Sons, Hoboken, NJ.
Spalding, M., M. Kainuma, and L. Collins. 2010. World Atlas of Mangroves. Earthscan, Taylor & Francis
Group, New York.
Conservation. The University of Massachusetts Press, Amherst, MA.
Tomlinson, P.B. 1996. The Biology of Mangroves. Cambridge University Press, Cambridge, U.K.
Whigham, D., D. Dykyjova, and S. Hejny (eds.). 1993. Wetlands of the World I: Inventory, Ecology and
Management. Kluwer Academic Publishers, Dordrecht, the Netherlands (reprinted in 2012).
2 Classification of Wetland
Types for Mapping and
Large-Scale Inventories
Ralph W. Tiner
CONTENTS
Introduction....................................................................................................................................... 19
Characteristics Used for Wetland Classification...............................................................................20
Types of Wetland Classification Systems.........................................................................................20
Major Wetland Classification Systems............................................................................................. 22
Ramsar International Wetland Classification System.................................................................. 22
Cowardin et al. Classification System..........................................................................................24
Brinson’s Hydrogeomorphic Classification System....................................................................26
Other Preexisting Classification Systems......................................................................................... 27
Recent Developments in Wetland Classification..............................................................................28
Modification of the Cowardin et al. System.................................................................................28
Updating the Cowardin et al. System by the 2013 Federal Geographic Data Committee......28
Adapting Cowardin et al. for Mediterranean Wetlands........................................................... 30
Adapting the HGM Concept for Wetland Mapping and Landscape-Level
Functional Assessment................................................................................................................. 30
Other New Approaches to Wetland Classification....................................................................... 37
Conclusion........................................................................................................................................ 39
References......................................................................................................................................... 39
INTRODUCTION
Recognizing the diversity of wetlands that exist around the globe, scientists have developed numer-
ous classification schemes to group wetlands into categories based on similar characteristics, envi-
ronmental influences, functions, and/or uses. Various systems have been created for many purposes
including (1) to group wetlands into categories for scientific studies and comparison; (2) to pro-
vide common terminology to describe similarities and differences among wetlands; (3) to provide
the foundation for conducting wetland inventories that report on the status, trends, functions, and
condition of wetlands in specific geographic areas (e.g., watersheds, counties, provinces, states,
countries, and the world); and (4) for conservation, use, and management purposes. Wetlands may
be classified biologically, physically, chemically, hydrogeomorphically, and in other ways depend-
ing on the discipline and interests of the classifier (e.g., scientist, organization, or government
agency). The term “classification” has also been used to refer to various methods to identify wet-
lands using remotely sensed data—image classification—but this type of classification is not the
subject of this chapter (see Chapters 5, 6, 8, 9, and others in Section III of this book for discussion
of image classification).
19
This chapter provides an overview of classification systems used for characterizing wetlands for
mapping and large-scale inventories. It is important to note that it is not a comprehensive review
of all wetland classification systems as there are other systems that are part of national vegetation
classification systems to assess biodiversity (e.g., Alvo and Ponomarenko 2003; Comer et al. 2003;
Grossman et al. 1998; Josse et al. 2007; Vegetation Subcommittee 2008) or have been designed for
research, mitigation, restoration, or other management purposes (e.g., Albert et al. 2005; Brooks
et al. 2011; Morrice et al. 2011; Simenstad et al. 2011; Stewart and Kantrud 1971). Readers inter-
ested in a more in-depth review of classification systems used for inventories in the 1990s should
consult Volume 118 of the journal Vegetatio (1995, also published as Finlayson and van der Valk
1995). Another publication worth consulting is Mires: Swamp, Bog, Fen, and Moor (Gore 1983),
which includes various peatland classification systems for describing these wetlands around the
globe. Also remember that there is not a single universally accepted definition of wetland and that
some definitions may include lakes, rivers, and other open waterbodies, while others include only
the shallow water zone, in addition to the vegetated marshes, swamps, bogs, and fens (see Chapter 1
for brief discussion).
CHARACTERISTICS USED FOR WETLAND CLASSIFICATION

Wetlands have formed in a variety of climates, landscapes, and hydrologic settings producing a
wide diversity of types. No two wetlands are exactly the same. They differ in size, shape, biological
characteristics, environmental conditions, topography, and human influences. Despite these differ-
ences, wetlands can be categorized into a few or many types based on common properties and the
level of separation required. Formal classification systems have been developed by scientists with
a particular expertise (e.g., ecologists, botanists, foresters, wildlife biologists, soil scientists, and
hydrologists) with the distinguishing features based largely on their area of expertise, but such clas-
sifications have limited applications. Wetland classification systems developed for natural resource
conservation purposes often require analysis of multiple properties such as vegetation, hydrology,
water chemistry, soil type, landscape position, and landform. Table 2.1 lists features used to classify
wetlands.
TYPES OF WETLAND CLASSIFICATION SYSTEMS

There are two basic approaches to classification: horizontal and vertical. A horizontal classification
system is a simple listing of a finite number of types based on a few major characteristics. Horizontal
classifications tend to be highly generalized out of necessity to limit the number of wetland types for
mapping or other purposes (e.g., Martin et al. 1953). These classifications have produced terms like
marsh, swamp, and bog that are familiar to the public (see Chapter 1 for examples of common types;
Tiner 2005). Unfortunately, many of these common terms lack universally accepted definitions and
have been inconsistently used. This issue is highlighted when maps from multiple years or sources
with different definitions are compared, making it difficult to attribute area and type differences
to either physical change or classification differences. Many wetland mapping efforts that rely on
automated classification for processing imagery have used horizontal classifications.
A vertical system employs a hierarchical approach that begins with general characteristics and
then further subdivides groups into smaller and smaller units (e.g., Cowardin et al. 1979; Tiner 2014).
These approaches to wetland classification consider more variables and often have better definition
of terms that will likely produce more consistent mapping and more detailed characterization than
possible with a horizontal classification. Higher levels share more generalized characteristics, such
as landscape position and water source, while lower levels are based on more detailed and spe-
cific characters like vegetation life-form (including dominant species), substrate characteristics, and
water level fluctuations. In providing more descriptive characterizations of wetlands, a hierarchical
Classification of Wetland Types for Mapping and Large-Scale Inventories 21
TABLE 2.1
Properties Used to Classify Wetlands
Property Distinguishing Feature(s) Examples of Terminology
Vegetation Presence or absence of vegetation Vegetated or nonvegetated
Woody versus herbaceous plants, Marsh, wet meadow, aquatic bed, shrub swamp, forested wetland
height
Leaf type Needle leaved, broad leaved, narrow leaved
Leaf persistence Evergreen or deciduous, persistent or nonpersistent
(herbaceous plants)
Structure Treeland, shrubland, tussockland, grassland, mossfield
Plant physiognomy Lawn, carpet, hummock, mud-bottom communities
Dominant species in the Red maple swamp, cattail marsh, leatherleaf bog, Picea
community mariana–Chamaedaphne calyculata, Spartina patens–Distichlis
spicata, Nymphaea odorata–Nuphar luteum
Hydrology Tidal flooding or not Tidal wetlands, nontidal wetlands
Frequency and duration of Permanently flooded, semipermanently flooded, seasonally
wetness flooded, temporarily flooded, seasonally saturated, continuously
saturated (nontidal); regularly flooded, irregularly flooded
(tidal); episodic; ephemeral
Depth of flooding Deep, shallow
Directional flow of water source Outflow, inflow, throughflow, vertical flow, sink, pass-through,
unidirectional, bidirectional
Groundwater interaction Discharge, recharge, flow-through
Soil drainage class Poorly drained, very poorly drained
Source of water Topogenous (ponded or saturated), soligenous (spring fed or
seepage fed), limnogenous (flooding from rivers, streams, or
lakes), groundwater, rainfall, runoff, precipitation, river, estuary,
tributary dominated, lake dominated
Origin of water Telluric (groundwater), meteoric (atmosphere), marine (ocean)
Water chemistry Hydrogen ion concentration (low Acid, circumneutral, or alkaline
to high)
Salt concentration (high to low) Euhaline, mesohaline, oligohaline (coastal salts, mostly sodium
chloride); eusaline, mesosaline, oligosaline (inland salts,
calcium, magnesium, sodium, potassium, chloride, bicarbonates,
and sulfates); fresh
Nutrient status (poor to rich) Ombrotrophic, mesotrophic, eutrophic
Soil or substrate Substrate Rocky shore, unconsolidated shore, sandy beach, mud flat, gravel
bar, sand flat, organic flat
Amount of organic matter Histosols (organic soils), mineral soils
Amount of peat Peatland, nonpeatland
Amount of decomposition of Muckland, peatland
organics
Other soil characteristics Alluvial, fluvaquents, gleysols, spodosols, planosols
Peatland Presence of series of open water When present, eccentric or patterned; when absent, concentric
pools (“flarks”) and ridges raised (domed), plateau, blanket, aapa mire, palsa
(“strings”)
Landscape Association with a waterbody Marine, estuarine, lotic (rivers and streams), lentic (lakes),
position or not terrene, geographically isolated
Landform Physical shape of wetland Basin (depression), flat, slope, floodplain, interfluve, fringe, hill
slope, highland/hill, swale, delta, shore, channel; plateau,
cushion, domed, string, and blanket (for bogs)
(Continued)
TABLE 2.1 (Continued )

Properties Used to Classify Wetlands
Property Distinguishing Feature(s) Examples of Terminology
Impacts Human caused or natural Diked (impounded, leveed), partly drained, farmed, excavated,
ditched, beaver
Wetland origin Natural or artificial Natural, human created, beaver modified
Function Significant for a particular Surface water detention, streamflow maintenance, waterfowl and
function waterbird habitat, sediment and other particulate retention,
nutrient transformation, carbon sequestration
Note: Wetland types are typically described by a combination of factors, most commonly by vegetation and hydrology.
classification system can be used to better assess the potential functions of individual wetlands than
a horizontal classification offers. This type of classification has commonly been applied through
manual interpretation techniques or a combination of automated and manual interpretation routines.
More advanced remotely sensed data and techniques are beginning to support the increased use of
automated or semiautomated techniques to create vertical classification-based maps.
MAJOR WETLAND CLASSIFICATION SYSTEMS

Many wetland classification systems have been developed by various countries and by wetland sci-
entists for a variety of purposes. Two systems have received more widespread usage, and in other
cases, hybrid systems have been developed from them: the Ramsar international wetland classifica-
tion and the Cowardin et al. classification system. A third system, the hydrogeomorphic (HGM)
classification, has also received much attention in the United States by government agencies, and
the concept has been incorporated in some new classification systems.
Ramsar International Wetland Classification System

In 1971 at the Convention on Wetlands of International Importance Especially as Waterfowl Habitat
(Ramsar, Iran), representatives from 18 nations signed an international treaty to promote conser-
vation and wise use of wetlands recognizing their economic, cultural, scientific, and recreational
value, especially as habitat for migratory waterfowl. To date, nearly 160 contracting partners or
member states have joined the movement. One outcome of the Convention was a multinational wet-
land classification published by the Ramsar Convention Bureau. Its purpose was to provide a consis-
tent framework for wetland classification and for describing wetlands of international importance.
The classification is essentially a horizontal classification system, despite the fact that wet-
lands are first separated into three broad groups: marine/coastal wetlands, inland wetlands, and
human-made wetlands, since these categories are then divided into a number of other types
(Table 2.2). The specific types are only generally defined or even lack definition, whereas addi-
tional guidance has been given for defining peatlands, wet grasslands, mangroves, coral reefs,
temporary pools, and karst and other subterranean hydrological systems and for identifying spe-
cific types as wetlands of international importance. The system includes many permanent open
waterbodies as wetland including freshwater rivers, lakes, and streams plus reservoirs, large
impoundments, coastal lagoons, and the shallow water zone of marine waters. The classifica-
tion is being used by various countries as a foundation to collect information on their wetlands,
especially for their wetland inventories (e.g., Denmark, Finland, Germany, Norway, Sweden, the
Netherlands, and the United Kingdom; Schultink and van Vliet 1997) (see Wetlands Inventory:
TABLE 2.2
Wetland Types Based on Multinational Wetland Classification from Ramsar Convention
Marine or coastal wetlands
Permanent shallow marine waters (in most cases <6 m deep)
Marine subtidal aquatic beds (kelp beds, sea grass beds, and tropical marine meadows)
Coral reefs
Rocky marine shores (including rocky offshore islands and sea cliffs)
Sand, shingle, or pebble shores (including sand bars, spits, sandy inlets, dune systems, and humid dune slacks)
Estuarine waters (permanent waters of estuaries and estuarine systems of deltas)
Intertidal mud, sand, or salt flats
Intertidal marshes (salt marshes, salt meadows, saltings, raised salt marshes, and tidal brackish
and freshwater marshes)
Intertidal forests (mangrove swamps, nipah swamps, and tidal freshwater swamp forests)
Coastal brackish/saline lagoons (with at least one relatively narrow connection to the sea)
Coastal freshwater lagoons (includes freshwater delta lagoons)
Karst and other subterranean hydrological systems (marine/coastal)
Inland wetlands
Permanent inland deltas
Permanent rivers/streams/creeks (including waterfalls)
Seasonal, intermittent, or irregular rivers/streams/creeks
Permanent freshwater lakes (over 8 ha, including large oxbow lakes)
Seasonal or intermittent freshwater lakes (over 8 ha, including floodplain lakes)
Permanent saline/brackish/alkaline lakes
Seasonal or intermittent saline/brackish/alkaline lakes and flats
Permanent saline/brackish/alkaline marshes and pools
Seasonal or intermittent saline/brackish/alkaline marshes and pools
Permanent freshwater marshes and pools (ponds below 8 ha, marshes and swamps on inorganic soils, with emergent
vegetation waterlogged for at least most of the growing season)
Seasonal or intermittent freshwater marshes and pools on inorganic soils (including sloughs, potholes, seasonally flooded
meadows, sedge marshes)
Nonforested peatlands (including shrub or open bogs, swamps, fens)
Alpine wetlands (including alpine meadows and temporary waters from snowmelt)
Tundra wetlands (includes tundra pools and temporary waters from snowmelt)
Shrub-dominated wetlands (shrub swamps, shrub-dominated freshwater marshes, shrub carr, alder
thicket on inorganic soils)
Freshwater tree-dominated wetlands (freshwater swamp forest, seasonally flooded forests, and wooded swamps on
inorganic soils)
Forested peatlands (peat swamp forests)
Freshwater springs and oases
Geothermal wetlands
Karst and other subterranean hydrological system (inland)
Human-made wetlands
Aquaculture ponds (e.g., fish/shrimp)
Ponds (including farm ponds, stock ponds, small tanks; generally below 8 ha)
Irrigated land (including irrigation channels and rice fields)
Seasonally flooded agricultural land (including intensively managed or grazed wet meadow or pasture)
Salt exploitation sites (including salt pans and salines)
Water storage areas (including reservoirs, barrages, dams, and impoundments; generally over 8 ha)
(Continued)

Wetland Types Based on Multinational Wetland Classification from Ramsar Convention
Excavations (gravel/brick/clay pits, borrow pits, and mining pools)
Wastewater treatment areas (e.g., sewage farms, settling ponds, and oxidation basins)
Canals, drainage channels, and ditches
Karst and other subterranean hydrological system (human made)
Source: The Ramsar Convention Bureau, Strategic Framework and Guidelines for the Future Development of the List of
Wetlands of International Importance of the Convention on Wetlands (Ramsar, Iran, 1971), Gland, Switzerland,
http://www.ramsar.org/key_guide_list2006_e.htm#B (updated on the web 3/12/2012).
A Ramsar Framework for Wetland Inventory and Ecological Character Description [Ramsar
Convention Secretariat 2010]). This system has been modified to address regional conditions and
needs in some countries (e.g., China [Zuo et al. 2013]) (see Chapter 22).
Cowardin et al. Classification System

The U.S. Fish and Wildlife Service (FWS) has a long history of wetland conservation and manage-
ment due to the dependency of many fish and wildlife species on wetlands. Since the 1950s, the
FWS has published reports on wetland status and vulnerability. In the 1970s, the agency decided to
undertake a more comprehensive wetlands inventory by producing a set of maps (National Wetlands
Inventory [NWI] maps) to assist their biologists and others in reviewing environmental impacts
of proposed development and for promoting wetland conservation. The first step in conducting an
inventory involves establishing a wetland definition and classification system for identifying dif-
ferent types of interest. In 1975, the FWS held a national conference to bring together the nation’s
leading wetland experts, federal and state biologists, and others mapping wetlands to help develop
a classification for conducting this national survey (Sather 1976). One of the findings of this meet-
ing was that common terms such as marsh, swamp, and bog had been used differently across the
country, and experts felt that it was best to create new terminology and avoid the use of these famil-
iar terms. After going through a couple of iterations, in 1979, the FWS published a hierarchical
classification system (Cowardin et al. 1979) for its NWI program. The purpose of this new system
was threefold: (1) to group ecologically similar habitats so that value judgments can be made, (2)
to furnish habitat units for inventory and mapping, and (3) to provide uniformity in concepts and
terminology throughout the entire United States. The classification has been used for wetland map-
ping in the United States for nearly 40 years and has been adapted for use in a few other countries
or regions.
This classification system is a vertical or hierarchical system beginning with wetlands defined
by five ecological systems with more detailed classification to follow based on specific properties
such as vegetation, substrate, hydrology, water chemistry, and impacts (Table 2.3). The five systems
are (1) marine (open ocean and its shoreline), (2) estuarine (estuary and associated salt and brackish
wetlands and tidal flats), (3) riverine (within banks of rivers and streams), (4) lacustrine (deep and
shallow waters of lakes), and (5) palustrine (the majority of inland wetlands including ponds). Each
system, with the exception of the palustrine, is further subdivided into subsystems. For the marine,
estuarine, and lacustrine systems, subsystems separate wetlands from deepwater habitats. The
marine and estuarine systems have two subsystems defined by tidal water levels: (1) subtidal (con-
tinuously submerged areas = deepwater habitats) and (2) intertidal (areas alternately flooded by tides
and exposed to air = wetlands), whereas the lacustrine system is separated into two systems based
on water depth: littoral (e.g., from the lake shore to a depth of 6.6 ft or 2 m below low water = wet-
lands) and limnetic (at depths below 6.6 ft at low water = deepwater habitats). The riverine system
TABLE 2.3
Wetland Types in the United States (Excluding Modifiers) Based on the U.S. Fish and
Wildlife Service’s Wetland Classification System
System Subsystem Classes Subclasses
Marine intertidal Unconsolidated shore Cobble gravel, sand, mud, organic
Rocky shore Bedrock, rubble
Aquatic bed Algal, rooted vascular
Reef Coral, worm
Estuarine intertidal Unconsolidated shore Cobble gravel, sand, mud, organic
Rocky shore Bedrock, rubble
Aquatic bed Algal, rooted vascular, floating vascular
Reef Mollusk, worm
Emergent Persistent, nonpersistent
Scrub–shrub and forested Broad-leaved deciduous, needle-leaved deciduous,
broad-leaved evergreen, needle-leaved evergreen, dead
Palustrine (none) Emergent Persistent, nonpersistent
Scrub–shrub and forested Broad-leaved deciduous, needle-leaved deciduous,
broad-leaved evergreen, needle-leaved evergreen, dead
Moss–lichen Moss, lichen
Aquatic bed Algal, aquatic moss, rooted vascular, floating vascular
Unconsolidated shore and bottom Cobble gravel, sand, mud, organic, vegetated (pioneer)
Rocky bottom Bedrock, rubble
Lacustrine littoral Rocky bottom and rocky shore Bedrock, rubble
Unconsolidated shore and bottom Cobble gravel, sand, mud, organic
Aquatic bed Algal, aquatic moss, rooted vascular, floating vascular
Emergent Nonpersistent
Riverine tidal Unconsolidated shore and streambed Cobble gravel, sand, mud, organic, vegetated (pioneer)
Lower perennial Rocky shore Bedrock, rubble
Upper perennial Aquatic bed Algal, aquatic moss, rooted vascular, floating vascular
Intermittent Emergent Nonpersistent
Source: Cowardin, L.M. et al., Classification of wetlands and deepwater habitats of the United States, U.S. Fish and Wildlife
Service, Washington, DC, FWS/OBS-79/31, 1979, http://www.fws.gov/nwi/Pubs_Reports/Class_Manual/class_
titlepg.htm.
Modifiers: water regime (see Table 2.4), soils (organic or mineral), water chemistry (pH and salinity/halinity; Table 2.5),
special (excavated, impounded [to obstruct outflow of water]; diked [to obstruct inflow of water]; partly drained, farmed,
artificial [materials deposited to create or modify a wetland or deepwater habitat]; and beaver).
is further defined by four subsystems that represent different reaches of a flowing or lotic system:
(1) tidal (freshwater), (2) lower perennial (permanent, flowing waters with a well-developed flood-
plain), (3) upper perennial (permanent, flowing water with very little or no floodplain development),
and (4) intermittent (channel containing nontidal flowing water for only part of the year). The class
level describes the general appearance of the wetland or deepwater habitat in terms of the dominant
vegetative life-form or the nature and composition of the substrate where vegetative cover is <30%.
Each class is further divided into subclasses to better define the type of substrate in nonvegetated
areas or the type of dominant vegetation. Below the subclass level, dominance type can be applied
to specify the predominant plant or animal in the wetland community. The fifth level of the clas-
sification is represented by a set of descriptors that can be used to better describe hydrologic, chemi-
cal, and soil characteristics and human impacts: (1) water regime (hydrology [Table 2.4]), (2) water
chemistry (pH—acid <5.5, circumneutral 5.5–7.4, and alkaline >7.4; salinity/halinity [Table 2.5]),
TABLE 2.4
Water Regimes (Hydrology) Based on the U.S. Fish and Wildlife Service’s Wetland
Classification System
Water Regime Modifiers General Definition
Tidal regimes
Subtidal Continuously covered by tidal waters.
Irregularly exposed Exposed less than daily by the tides.
Regularly flooded Flooded daily by the tides.
Irregularly flooded Flooded less than daily by the tides.
Nontidal regimes
Permanently flooded Inundated continuously, year-round in all years.
Intermittently exposed Inundated year-round in most years, but exposed during extreme droughts.
Semipermanently flooded Inundated throughout the growing season in most years.
Seasonally flooded Inundated for extended periods during the growing season, but usually not flooded later in
the growing season; water table may be near the surface for much of the time when not
flooded (seasonally flooded/saturated) or may be well below the surface.
Temporarily flooded Inundated for brief periods during the growing season (usually a couple of weeks or less
early in the growing season), with the water table typically well below the surface for
extended periods thereafter.
Intermittently flooded Inundated for variable periods with no detectable seasonally; area is usually exposed.
Continuously saturated Substrate is saturated at or near the surface throughout the year in all or most years;
widespread surface inundation is rare, but water may be present in shallow depressions that
intersect the groundwater table, particularly on a floating peat mat.
Seasonally saturated Substrate is saturated at or near the surface for extended periods during the growing season,
but unsaturated conditions prevail by the end of the season in most years; surface water is
typically absent, but may occur for a few days after heavy rain and upland runoff, especially
during the nongrowing season.
Artificially flooded The frequency and duration of inundation are controlled by humans; in the strictest sense, the
control is purposeful by means of pumps or siphons, but more generally, wetlands flooded
by any artificial means qualify, including irrigation.
Sources: Cowardin, L.M. et al., Classification of wetlands and deepwater habitats of the United States, U.S. Fish and Wildlife
Service, Washington, DC, FWS/OBS-79/31, 1979, http://www.fws.gov/nwi/Pubs_Reports/Class_Manual/class_
titlepg.htm; Wetlands Subcommittee, Classification of wetlands and deepwater habitats of the United States, 2nd
edn., Federal Geographic Data Committee, August 2013, FGDC-STD-004-2013, https://www.fgdc.gov/standards/
projects/FGDC-standards-projects/wetlands/nvcs-2013.
(3) soil (organic or mineral), and (4) special (human and beaver impacts: diked/impounded, exca-
vated, partly drained, farmed, artificial, and beaver).
Brinson’s Hydrogeomorphic Classification System

Wetland ecologist Mark Brinson (1993) developed an HGM-based wetland classification to aid in
performing functional assessments of wetlands. He realized the shortcomings of the Cowardin et al.
system, in that it did not address abiotic features, namely, hydrogeomorphology, which is directly
linked to many wetland functions. The HGM wetland classification is actually more of an approach
to provide a framework for wetland evaluation, rather than a classification system for mapping wet-
lands. Consequently, some of the types are not mutually exclusive (e.g., fringe wetlands associated
with rivers and lakes versus riverine and lacustrine wetlands, depressional wetlands [oxbows and
sloughs] within the riverine wetland type).
TABLE 2.5
Salinity Modifiers for Coastal and Inland Areas
Approximate Specific
Coastal Modifiersa Inland Modifiersb Salinity (ppt) Conductance (Mhos at 25°C)
Hyperhaline Hypersaline >40 >60,000
Euhaline Eusaline 30–40 45,000–60,000
Mixohaline (brackish) Mixosalinec 0.5–30 800–45,000
Polyhaline Polysaline 18–30 30,000–45,000
Mesohaline Mesosaline 5–18 8,000–30,000
Oligohaline Oligosaline 0.5–5 800–8,000
Fresh Fresh <0.5 <800
Source: Cowardin, L.M. et al., Classification of wetlands and deepwater habitats of the United
States, U.S. Fish and Wildlife Service, Washington, DC, FWS/OBS-79/31, 1979, http://
www.fws.gov/nwi/Pubs_Reports/Class_Manual/class_titlepg.htm.
a Coastal modifiers are employed in the marine and estuarine systems.
b Inland modifiers are employed in the riverine, lacustrine, and palustrine systems.
c The term “brackish” should not be used for inland wetlands or deepwater habitats.
The HGM system identified seven basic hydromorphic classes: (1) depressional wetlands (within
topographic depressions, e.g., kettles, potholes, vernal pools, playas, and Carolina bays), (2) organic
soil flats (extensive peatlands, e.g., bogs and pocosins), (3) mineral soil flats (broad nearly level
wetlands with inorganic soils, e.g., flatwoods), (4) riverine wetlands (along rivers and streams, e.g.,
floodplains and riparian areas), (5) slope wetlands, (6) lacustrine fringe (lakeshore wetlands), and
(7) estuarine fringe (tidal wetlands) (Smith et al. 1995). Riverine wetlands are further divided into
three gradients that reflect stream flow and fluvial processes: high gradient (flow likely continu-
ous, but mostly flashy; wetland on coarse substrate maintained by upslope groundwater source;
unstable substrate colonized by pioneer species; streamside vegetation that contributes to allochtho-
nous organic supply), middle gradient (flow likely continuous; channel processes establish variable
topography/hydroperiod/habitat interspersion on floodplain; alluvium is renewed by surface accre-
tion and point bar deposition; interspersion of plant communities contributes to beta diversity), and
low gradient (flow continuous with cool season flooding, high suspended sediments, flood storage,
conserves groundwater discharge, major fish and wildlife habitat and biodiversity, strong biogeo-
chemical activity and nutrient retention).
The development of HGM was rapid in the 1990s and early 2000s, and a number of guidebooks
have been prepared by the U.S. Army Engineer Research and Development Center (search “HGM”
at http://acwc.sdp.sirsi.net/client/default). Modification of the original classification has occurred
through usage (e.g., Adamus 2001; Brooks et al. 2011; Table 2.6). Researchers in Oklahoma have
used GIS to reclassify NWI wetlands according to a regional HGM scheme for inventory purposes
(Dvorett et al. 2012).
OTHER PREEXISTING CLASSIFICATION SYSTEMS

Classifications from some other countries are presented in a series of tables to give readers a broader
perspective on other approaches to wetland classification for regional and national inventories
(Tables 2.7 through 2.11). For definition of terms, consult the applicable reference. Some other clas-
sifications of note include Gopal and Sah (1995) and Pakarinen (1995), while Whigham et al. (1993),
Semeniuk and Semeniuk (1997), and Tiner (1999) offer additional information on global wetland
types and wetland classification.
TABLE 2.6
Example of an Application of the HGM Concept to Classify Arkansas Wetlands
Class Subclass Community Types
Depression Headwater depression Headwater swamp
Isolated depression Mountaintop depression
Sinkhole
Sand pond
Valley train pond
Unconnected alluvial depression
Connected depression Connected floodplain depression
Flat Alkali flat Alkali wet prairie
Alkali post oak flat
Nonalkali flat Wet tallgrass prairie
Pine flat
Hardwood flat
Post oak flat
Fringe Reservoir fringe Reservoir shore
Connected lacustrine fringe Connected lake margin
Isolated lacustrine fringe Unconnected lake margin
Slope Calcareous slope Calcareous perennial seep
Noncalcareous slope Noncalcareous perennial seep
Bayhead
Wet weather seep
Sandstone glade
Riverine Spring run Spring run
High-gradient riverine High-gradient riparian zone
Mid-gradient riverine Mid-gradient floodplain
Mid-gradient backwater
Low-gradient riverine Low-gradient overbank
Low-gradient backwater
Sand prairie
Riverine impounded Beaver complex
Wildlife management impoundment
Source: Acquired from Arkansas Watershed Resource Information Management System, 2014, HGM classifica-
tion & characteristics of Arkansas wetlands, http://anrc.cast.uark.edu/home/mawpt/statewide-planning-
initiatives/hgm-characterization/hgm-classification-and-characteristics-of-arkansas-wetlands.html.
RECENT DEVELOPMENTS IN WETLAND CLASSIFICATION

A few new approaches to wetland classification have been created during the past decade. Some
have been used for mapping while others have been used in research studies. The former types are
briefly summarized, whereas the latter are beyond the scope of this chapter.
Modification of the Cowardin et al. System

Updating the Cowardin et al. System by the 2013 Federal Geographic Data Committee
In 1996, the Federal Geographic Data Committee (FGDC) first adopted the Cowardin et al. system
as the national wetland classification system. Since then, advances in the science and 30 years of
TABLE 2.7
Wetland Types according to the Canadian Wetland Classification System
Class Forms
Bog Atlantic plateau bog, basin bog, blanket bog, collapse scar bog, domed bog, flat bog, floating
bog, lowland polygon bog, mound bog, northern plateau bog, palsa bog, peat mound bog,
peat plateau bog, polygonal peat plateau bog, shore bog, string bog, and veneer bog
Fen Atlantic ribbed fen, basin fen, channel fen, collapse scar fen, feather fen, floating fen,
horizontal fen, ladder fen, lowland polygon fen, net fen, northern ribbed fen, palsa fen, shore
fen, slope fen, snowpatch fen, spring fen, and stream fen
Marsh Active delta marsh, channel marsh, coastal high marsh, coastal low marsh, estuarine high
marsh, estuarine low marsh, floodplain marsh, inactive delta marsh, kettle marsh, seepage
track marsh, shallow basin marsh, shore marsh, stream marsh, terminal basin marsh, and
tidal freshwater marsh
Swamp Basin swamp, flat swamp, floodplain swamp, peat margin swamp, shore swamp, spring
swamp, and stream swamp
Shallow water Channel water, delta water, estuarine water, kettle water, nontidal water, oxbow water, shallow
basin water, shore water, stream water, terminal basin water, thermokarst water, tidal water,
and tundra pool water
Source: Warner, B.G. and Rubec, C.D.A. (eds.), The Canadian Wetland Classification System, 2nd edn., National
Wetlands Working Group, Research Centre, University of Waterloo, Waterloo, Ontario, Canada, 1997,
http://www.gret-perg.ulaval.ca/fileadmin/fichiers/fichiersGRET/pdf/Doc_generale/Wetlands.pdf.
applying the system for the NWI have warranted some minor changes. In August 2013, the FGDC
published a revision of the Cowardin et al. (1979) wetland classification system for use in mapping
and monitoring wetland changes by federal agencies (Wetlands Subcommittee 2013). The most
significant of the changes involved the “saturated” water regime that was split into two regimes:
“continuous saturated” to describe the hydrology of most bogs and fens, for example, and “season-
ally saturated” to address the hydrology of most drier-end wetlands (e.g., flatwoods—coastal plain
forested wetlands on poorly drained soils). “Continuously saturated” is defined as “The substrate
is saturated at or near the surface throughout the year in all or most years. Widespread surface
inundation is rare, but water may be present in shallow depressions that intersect the groundwater
table, particularly on a floating peat mat.” In contrast, the definition of “seasonally saturated” is
“The substrate is saturated at or near the surface for extended periods during the growing season,
but unsaturated conditions prevail by the end of the season in most years. Surface water is typi-
cally absent, but may occur for a few days after heavy rain and upland runoff.” The other notable
changes were (1) more precise definitions of freshwater tidal water regimes (including adding the
word “fresh” to the pre-existing terms, e.g., “seasonally flooded–tidal fresh” and changing “regu-
larly flooded” to “regularly flooded–tidal fresh”), (2) adding an appendix describing the differ-
ence between organic and mineral soils, (3) changing the maximum depth of the shallow water
zone of ponds and lakes (littoral zone) from 2.0 to 2.5 m based on the revised U.S. Department
of Agriculture definition of soil, and (4) a revised definition of “growing season” to account for
green-up of native species in the spring as opposed to an agriculture-centric definition based on the
frost-free season. The latter is consistent with the concept developed for identifying wetlands for
regulatory purposes at the federal level.
The FGDC wetland classification system is intended to replace the Cowardin et al. (1979) system
for mapping wetlands by agencies of the U.S. government. Its use by states and other organizations
is also encouraged. With the decline of federal funding for the FWS’s NWI program, it is uncertain
as to how much NWI data will be updated with this system.
TABLE 2.8
New Zealand Wetland Classification Is Somewhat of a Hybrid of the American and
Canadian Systems with New Categories Reflective of the Country’s Uniqueness
Structural Classes
Hydrosystems Classes Forms Vegetated Nonvegetated
Riverine Bog Plateau mire Forest Rockland
Lacustrine Fen Cushion mire Treeland Boulderfield
Palustrine Swamp Domed mire Scrub Stonefield
Inland Saline Marsh String mire Shrubland Gravelfield
Plutonic Seepage Blanket mire Flaxland Sandfield
Geothermal Shallow water Floating Tussockland Siltfield
Nival Ephemeral wetland Shore Fernland Clayfield
Pakihi and gumland Riparian Reedland Mudfield
Channel Rushland Loamfield
Floodplain Sedgeland Peatfield
Delta Grassland Driftwoodfield
Basin Cushionfield Bacteriafield
Swale Herbfield Shellbeds
Flat Turf Salt crust
Mossfield
Lichenfield
Algalfield
Heathland
Marine and estuarine wetlands are not represented in this table (Johnson and Gerbeaux 2004).
Note: Swamp includes all types of vegetated wetlands (not only woody plants) on substrates comprising both peat
and mineral materials with a water table usually permanently above the surface at least in some places,
while marsh is limited to herbaceous-dominated wetlands on mineral soils with better drainage, lower
water tables, and a higher pH.
Adapting Cowardin et al. for Mediterranean Wetlands

Scientists responsible for inventorying wetlands in the Mediterranean have adapted the Cowardin
et al. system (Farinha et al. 2005). While maintaining most of the elements, they added a few fea-
tures. An “underground” subsystem was added to the lacustrine system. For the riverine system,
“underground” and “ephemeral” subsystems were added, plus the “intermittent” subsystem was
changed to “upper nonperennial” and “lower nonperennial” subsystems. The HGM concept was
adopted to add six subsystems to the palustrine system—slope, pan, basin, flat, floodplain, and
fringe. Other changes include establishing a combined “rocky/unconsolidated substrate” for non-
vegetated wetlands flooded for half or more of the growing season and adding “naked soil” class for
nonvegetated wetlands flooded for a shorter duration, a “salt crust” subclass, and “hyperhaline” and
“hypersaline” salinity modifiers. A “riparian” system with lentic and lotic subsystems was added
to the classification.
Adapting the HGM Concept for Wetland Mapping and

Landscape-Level Functional Assessment
Recognizing the merits of Brinson’s HGM system, the FWS Northeast Region applied the concept
of HGM to its updated NWI mapping to increase the functionality of the NWI database. Since the
HGM classification used some of the same terminology as the Cowardin et al. system but defined
TABLE 2.9
Geomorphic Wetland Classification for Inland Wetlands in Southeast Australia
Landform
Water Longevity Basin Channel Flat Slope Highland/Hill
Permanently inundated Lake River — — —
Seasonally inundated Sumpland Creek Floodplain — —
Intermittently inundated Playa Wadi Barlkarra — —
Seasonally waterlogging Dampland Trough Palusplain Paluslope Palusmont
Source: Semeniuk, C.A. and Semeniuk, V., Vegetatio, 118, 103, 1995.
This system focused on inland wetlands and was expanded for worldwide application.
Water descriptors: salinity (fresh, brackish—mixosaline, saline, and hyperhaline), consistency of salinity
(poikilohaline, fluctuating; stasohaline, consistent).
Landform descriptors: wetland shapes (linear, elongate, irregular, fan shaped, ovoid, or round), channel
shapes (straight, sinuous, anastomosing, and irregular).
Other descriptors: geomorphic scale (megascale, very large, >100 km2; macroscale, large, 1,000,000 m2 to
100 km2; mesoscale, 250,000–1,000,000 m2; microscale, small, 10,000–250,000 m2; leptoscale, very
small, <10,000 m2), vegetation cover (peripheral, mosaic, or complete), and complexity of vegetation
(homogenous, zoned, or heterogenous).
TABLE 2.10
Australian Wetland Classification for New South Wales
General Type Specific Types
Coastal wetlands Mangrove and salt marsh swamps
Estuarine lakes and lagoons
Dune swamps and lagoons
Coastal floodplain swamps and lagoons
Coastal floodplain forest
Tableland wetlands Upland lakes and lagoons
Upland swamps
Inland wetlands Permanent inland wetlands
Inland floodplain lakes and lagoons
Inland floodplain meadows
Reed swamps
Lignum swamps
Inland floodplain forests and woodlands
Arid wetlands
Source: Green, D.L., Wetland Management Technical Manual: Wetland Classification, Department of Land
and Water Conservation, Ecological Services Unit, Parramatta, New South Wales, Australia, 1997.
them differently (e.g., lacustrine and riverine) and the HGM system was not developed for mapping
wetlands, new terms and a classification hierarchy were created to address these and other issues.
The new descriptors include landscape position, landform, water flow path, and waterbody type,
commonly referred to as “LLWW descriptors” (Tiner 1997, 2003a, 2011a, 2014) (see Table 2.12
for simplified dichotomous keys). When added to NWI geospatial data, the descriptors improved
TABLE 2.11
Wetland Classification for East Africa
Marine wetlands
Subtidal types: sea grass beds; coral reefs
Intertidal types: rocky marine shores and reefs; mud flats, sand flats, and salt flats; intertidal vegetated sediments (salt
marshes and mangroves)
Estuarine wetlands
Subtidal types: estuaries and marine deltas
Intertidal types: mud flats, sand flats, and salt flats; estuarine marshes and salt marshes
Estuarine swamps and mangrove swamps
Sodic and/or saline water

Lacustrine permanent types: sodic lakes and salt lakes
Lacustrine temporary types: seasonally/occasionally inundated depressions and salt pans
Palustrine permanent types: sodic and salt marshes and swamps; springs, soaks, and resultant pools
Freshwater wetlands
Riverine permanent types: edges of perennial rivers, streams, and waterfalls; inland deltas (including deltas in lakes)
Riverine temporary types: seasonal/occasional rivers, streams, and waterfalls; riverine floodplains, river flats, deltaic plains,
riverine grasslands, and mbugas
Lacustrine permanent types: freshwater lakes (>10 ha) including shores subject to seasonal or irregular inundation
(drawdown floodplains); freshwater ponds and pools (<10 ha)
Lacustrine temporary types: seasonal lakes (>10 ha); seasonal ponds and pools (<10 ha)
Palustrine herbaceous types: permanent swamps, marshes, and dambosa; seasonal/occasional swamps, marshes, and
dambos; peatlands and fens; montane wetlands (including bogs); springs and soaks
Palustrine woody types: shrub swamps and thicket wetlands; swamp forests
Man-made wetlands
Aquaculture/mariculture types: fish ponds and prawn ponds
Agriculture types: farm ponds and dams; irrigated lands, rice paddy, channels, canals, and ditches; seasonally flooded
arable lands
Salt production types: salt evaporation pans
Urban/industrial types: borrow pits, brick pits, mining pools, road impoundments, and quarries; wastewater treatment
facilities
Water storage types: ponds, dams, and reservoirs
Source: Food and Agriculture Organization of the United Nations, Wetland characterization and classification for sustain-
able agricultural development, Sub-Regional Office for East and Southern Africa, Harare, Zimbabwe, 1998, http://
www.fao.org/docrep/003/X6611E/X6611E00.HTM.
Note: This system is a modification of the Ramsar classification.
a Dambos—seasonally or permanently wet grassy valley, depression, or seepage zone on slopes.
wetland characterization, creating what is now called “NWI+ data.” The ultimate objective of this
effort was to use the NWI+ data to predict wetland functions at the landscape level (Tiner 1995,
2003b, 2010, 2011b).
Landscape position defines the relationship between a wetland and an adjacent waterbody, if
present (Figure 2.1). Five wetland landscape positions were identified: (1) marine (on the shores
of the open ocean and its embayments), (2) estuarine (associated with tidal brackish waters
[estuaries]), (3) lotic (along freshwater rivers and streams and periodically flooded frequently
at least during high discharge periods [including freshwater tidal reaches of coastal rivers]),
(4) lentic (in lakes, reservoirs, and their basins where water levels are significantly affected by
TABLE 2.12
Simplified Keys for Classifying Wetlands by Landscape Position (A), Landform (B), and
Water Flow Path (C)
Key A. Landscape position
1. Wetland borders a river, stream, lake, reservoir, in-stream pond, estuary, or ocean. 2
1. Wetland does not border one of these waterbodies; it is surrounded by upland or borders a Terrene
pond that is surrounded by upland (i.e., an isolated pond).
2. Wetland lies along an ocean shore and is subject to tidal flooding. Marine
2. Wetland does not lie along an ocean shore, or if oceanside, it is not subject to tidal flooding. 3
3. Wetland lies along an estuary (salt–brackish waters) and is subject to tidal flooding. Estuarine
3. Wetland does not lie along an estuary, or if along the estuary, it is not subject to tidal 4
flooding.
4. Wetland lies along a lake or reservoir or within its basin (i.e., the relatively flat plain Lentic
contiguous to the lake or reservoir).
4. Wetland lies along a river or stream, or in-stream pond, or borders a marine or estuarine 5
wetland or associated waters but is not flooded by tides (except episodically).
5. Wetland is associated with a river or stream. 6
5. Wetland is not associated with a river or stream; it is a freshwater nontidal wetland Terrene
bordering a marine or estuarine wetland or associated waters.
6. Wetland is the source of a river or stream, and this watercourse does not flow completely Terrene
through the wetland.
6. A river or stream flows through or alongside the wetland. 7
7. Wetland is periodically flooded by river or stream. Lotica
7. Wetland is not periodically flooded by the river or stream. Terrene
Key B. Landform
1. Wetland is formed by the accumulation of peat forming thick organic deposits. Peatland
1. Wetland is not a peat deposit. 2
2. Wetland occurs on a slope >2%. Slope
2. Wetland does not occur on a slope >2%. 3
3. Wetland forms an island completely surrounded by water (not from ditching). Island
3. Wetland is not an island. 4
4. Wetland occurs in the shallow water zone of a permanent nontidal waterbody, the intertidal Fringe
zone of an estuary with unrestricted tidal flow, or the regularly flooded (daily tidal
inundation) zone of freshwater tidal wetlands.
4. Wetland does not occur in these waters or in estuarine intertidal zones with unrestricted 5
tidal flow.
5. Wetland occurs in a portion of an estuary with restricted tidal flow due to tide gates, Basin
undersized culverts, dikes, or similar obstructions.
5. Wetland does not occur in such location. 6
6. Wetland forms a nonvegetated bank or is within the banks of a river or stream. Fringe
6. Wetland is a vegetated river or stream bank or not within the banks. 7
7. Wetland occurs on an active alluvial plain of a river (a polygonal watercourse). Floodplainb
7. Wetland does not occur on an active floodplain. 8
8. Wetland occurs in a distinct depression. Basin
8. Wetland occurs on a nearly level landform. Flat
Key C. Water flow path

1. Wetland is formed by paludification processes where in areas of low evapotranspiration and Paludified
high rainfall, peat moss moves uphill creating wetlands on hillslopes (i.e., wetland
develops upslope of primary water source).
1. Wetland is not formed by paludification processes. 2
(Continued)

Simplified Keys for Classifying Wetlands by Landscape Position (A), Landform (B), and
Water Flow Path (C)
2. Wetland is typically surrounded by upland (nonhydric soil) and receives precipitation and Vertical flowc
runoff from adjacent areas with no apparent surface water inflow or outflow.
2. Wetland is not geographically isolated. 3
3. Water flow is mainly bidirectional from tides or lake/reservoir fluctuations. 4
3. Water flow is essential one directional (downstream). 5
4. Wetland is subjected to tidal flooding. Bidirectional–tidal
(apply tidal range:
megatidal, macrotidal,
mesotidal, microtidal,
or nanotidal)
4. Wetland is located along a lake or reservoir and not along a river or stream entering this Bidirectional–nontidald
waterbody; water levels are mainly affected by the rise and fall of lake or reservoir water
levels.
5. Wetland is a sink, receiving water from a river, stream, or other surface water source and Inflow
lacking surface water outflow.
5. Wetland is not a sink; surface water flows through or out of the wetland. 6
6. Water flows out of the wetland but does not flow into this wetland from another source. Outflow
6. Water flows through the wetland, often coming from upstream or uphill sources (typically Throughflow
wetlands along rivers and streams).
Sources: Adapted from Tiner, R.W., Dichotomous Keys and Mapping Codes for Wetland Landscape Position, Landform,
Water Flow Path, and Waterbody Type Descriptors, Versions 2.0 and 3.0, U.S. Fish and Wildlife Service, National
Wetlands Inventory Program, Northeast Region, Hadley, MA, 2011a, 2014, respectively.
a Lotic wetlands are separated into river and stream sections based on watercourse width—polygon (lotic river) versus linear
(lotic stream)—at a scale of 1:24,000.

b Basin and flat sublandforms can be identified within these landforms when desirable.
c Wetland is geographically isolated; hydrological relationship to other wetlands and watercourses may be more complex
than can be determined by simple visual assessment of surface water conditions. If groundwater relationships are known,
one can apply other water flow paths as appropriate, but add “groundwater” to the term (e.g., vertical flow/
outflow-groundwater).
d Bidirectional–nontidal flow is typically expanded to also reference the water flow path of the associated waterbody:
bidirectional–nontidal/throughflow, bidirectional–nontidal/inflow, bidirectional–nontidal/outflow, and bidirectional–

nontidal/isolated.
the presence of these waterbodies), and (5) terrene (geographically isolated wetlands, headwa-
ter wetlands [sources of streams], fragments of former isolated or headwater wetlands that are
now connected to downslope wetlands via drainage ditches, and wetlands on broad, flat ter-
rain cut through by stream but where overbank flooding does not occur [e.g., hydrologically
decoupled from streams]). In the early versions of this system, lotic wetlands are further sepa-
rated by river and stream sections (based on watercourse width, polygon [river] versus linear
[stream], at a scale of 1:24,000) and then divided into one of six gradients: (1) high (e.g., shal-
low mountain streams on steep slopes—not present in the study areas), (2) middle (e.g., streams
with moderate slopes—not present in the study areas), (3) low (e.g., main stem rivers with
considerable floodplain development and slow-moving streams), (4) intermittent (i.e., periodic
flows), (5) dammed reach (i.e., dammed rivers and streams where water periodically over-
flows the dam), and (6) tidal (i.e., under the influence of the tides). Today, these gradients have
been replaced with terms that address water permanence: (1) tidal, (2) dammed, (3) perennial,
(4) intermittent, and (5) ephemeral (Tiner 2014).
Terrene
Stream
Terrene
Lotic
Lotic
Lentic
Lake
Lentic Terrene
Stream
River
Terrene
Lotic river
Lotic stream
Lotic river Head of tide
Limit of salt water
Estuarine
Estuarine
Estuary
Ocean
Marine
FIGURE 2.1 Classification of wetlands by landscape position. (From Tiner, R.W., Dichotomous Keys
and Mapping Codes for Wetland Landscape Position, Landform, Water Flow Path, and Waterbody Type
Descriptors, Version 3.0, U.S. Fish and Wildlife Service, National Wetlands Inventory Program, Northeast
Region, Hadley, MA, 2014.)
Landform is the physical form of a wetland or the predominant land mass upon which it occurs
(e.g., floodplain). Seven types are recognized: basin, flat, floodplain, peatland, fringe, island, and
slope (Table 2.13). Basin and flat can be applied as sublandforms for floodplain.
Water flow path descriptors indicate the directional flow of water affecting wetlands:
bidirectional–tidal, bidirectional–nontidal, throughflow, throughflow–intermittent, inflow, outflow,
or vertical flow (formerly called “isolated”). Surface water connections are emphasized because
they are more readily identified than groundwater linkages (Figure 2.2). Bidirectional flow is a
two-way flow related to either tidal influence (bidirectional–tidal) or water level fluctuations in
lakes and impoundments (bidirectional–nontidal). Since estuarine and marine wetlands are tidal
by their nature and adding bidirectional–tidal to their classification is redundant, the tidal range of
these wetlands is now emphasized: megatidal (≥8 m), macrotidal (4 to <8 m), mesotidal (2 to <4 m),
microtidal (>0.3 to <2 m), and nanotidal (≤0.3 m) (Tiner 2014). Throughflow wetlands have either
a watercourse or another type of wetland above and below them, so water flows through these wet-
lands. Most lotic wetlands are throughflow types. Inflow wetlands are sinks where no surface water
outlets exist, yet water is entering via a stream or river (often intermittent) or an upslope wetland.
Outflow wetlands have water leaving them and moving downstream via a watercourse or a slope
wetland; they are often sources of streams. Wetlands lacking an apparent surface water inlet or
outlet are classified as having vertical flow—water flow moves up and down in response to direct
TABLE 2.13
Definitions and Examples of Landform Types
Landform Type General Definition Examples
Basina A depressional (concave) landform Lakefill bogs; wetlands in the saddle between two
hills; wetlands in closed or open depressions,
including narrow stream valleys
Slope A landform extending uphill (on a slope) Seepage wetlands on hillside; wetlands along
drainageways or mountain streams on slopes
Flata A relatively level landform, often on broad-level Wetlands on flat areas with high seasonal
landscapes groundwater levels; wetlands on terraces along
rivers and streams; wetlands on hillside benches;
wetlands at toes of slopes
Floodplain A broad, generally flat landform occurring on a Wetlands on alluvium; bottomland swamps
landscape shaped by fluvial or riverine processes
Fringe A landform occurring along a flowing or standing Buttonbush swamps; aquatic beds;
waterbody (lake, river, stream) and often subject semipermanently flooded marshes; wetlands in
to permanent, semipermanent, or tidal flooding river channels; salt and brackish marshes with
unrestricted tidal flow; river gravel bars
Peatland A landform created by the accumulation of Bogs; fens
organic matter forming thick organic deposits
Island A landform completely surrounded by water Deltaic and insular wetlands; floating bog islands
Sources: Tiner, 2011a; Tiner, R.W., Dichotomous Keys and Mapping Codes for Wetland Landscape Position, Landform,
Water Flow Path, and Waterbody Type Descriptors, Version 3.0, U.S. Fish and Wildlife Service, National Wetlands
Inventory Program, Northeast Region, Hadley, MA, 2014.
a May be applied as sublandforms within the floodplain landform.
precipitation, localized surface water runoff, and/or groundwater connections. These wetlands have
been commonly referred to as “geographically isolated wetlands” (e.g., wetlands completely sur-
rounded by upland) (Tiner 2003c) or “topographically isolated” (Brinson 1993). Despite their lack
of an apparent surface water connection, it must be recognized that these wetlands may be hydro-
logically linked to other wetlands and waterbodies via groundwater or subsurface flows or through
intermittent spillovers during extremely wet periods. (Note: When mapping such wetlands through
remote sensing, the interpretation scale and the season and quality of imagery affect detection of
surface water connections such as small streams, ditches, and narrow drainageways; consequently,
wetlands mapped with vertical flow may actually be connected by such features that were not visible
on the imagery at the scale of observation.)
Depending on project objectives (user needs), numerous other descriptors may be applied to
mapped wetlands including headwater, drainage divide, tidally restricted, and partly drained. For
open water habitats, descriptors for water flow path are applied as well as the so-called “waterbody
type” descriptors that better address the wide variety of pond types that occur on the landscape.
This classification is used in combination with the Cowardin et al. (1979) system to create a
“NWI+ database” that is used to better characterize wetlands and to predict wetland functions at
the watershed or landscape level. To date, 11 functions are predicted: surface water detention (for
nontidal wetlands), coastal storm surge detention, streamflow maintenance, sediment and other par-
ticulate retention, carbon sequestration, bank and shoreline stabilization, nutrient transformation,
provision of habitat for fish and aquatic invertebrates, provision of waterfowl and waterbird habitat,
provision of habitat for other wildlife, and provision of habitat for unique, uncommon, or highly
diverse plant communities. Wetlands predicted to be significant for these functions are highlighted
Outflow
Outflow
Throughflow
Throughflow
Throughflow
Bidirectional-
throughflow Vertical
flow
Vertical
flow Throughflow
Throughflow
Microtidal Head of tide
Limit of salt water
Microtidal
Microtidial
Microtidal
FIGURE 2.2 Classification of wetlands by water flow path. (From Tiner, R.W., Dichotomous Keys and
Mapping Codes for Wetland Landscape Position, Landform, Water Flow Path, and Waterbody Type
Descriptors, Version 3.0, U.S. Fish and Wildlife Service, National Wetlands Inventory Program, Northeast
Region, Hadley, MA, 2014.)
on online maps. These data and analyses have been conducted by the FWS for several states in the
United States including Massachusetts, Rhode Island, Connecticut, New Jersey, and Delaware and
for several large geographic areas across the United States. A few states have also applied LLWW
descriptors to enhance recent or ongoing wetland inventories (e.g., Delaware, Georgia, Michigan,
Montana, New Mexico, Oregon, and Wisconsin). Online NWI+ maps and reports are available
for FWS project areas and some states at the “wetlands one-stop mapping” website: http://aswm.
org/wetland-science/wetlands-one-stop-mapping. In its 2009 wetland mapping standard, the FGDC
recommended the use of this classification to increase the utility of the wetland geospatial data
produced with federal funds (Wetlands Subcommittee 2009).
Other New Approaches to Wetland Classification

Other classification systems have been created for a variety of purposes since the 1990s. Two of
particular note that have been used to assess the extent of wetlands in large geographic areas are
briefly described later. Since wetlands represent an estimated 30% of the Amazon Basin, scientists
have created a new system for categorizing lowland wetlands (Junk et al. 2011). The new clas-
sification applies Brinson’s concept recognizing the importance of physical properties to classify
wetlands into various categories. This classification considers numerous variables: climate, hydrol-
ogy, water and soil chemistry, and vegetation to create a hierarchical classification scheme for
Amazon lowland wetlands. Wetlands are first separated into two major groups based on water lev-
els (stable versus fluctuating) and then divided into other categories by vegetation and region for the
stable group and for wetlands with “oscillating” water levels by predictability and duration of the
fluctuations and then by other factors including river type (whitewater, blackwater, or clearwater),
flood height, and fertility (Table 2.14). The authors used this classification and data from a number
of sources to describe the extent and distribution of 14 types of Amazonian wetlands.
Another classification system including elements of the HGM approach coupled with vegetation
and hydrology was created to map wetlands in the Republic of Korea (Kim et al. 2006). Wetlands
were classified in six levels—with three broad level 1 types, inland wetlands, estuarine wetlands,
and coastal wetlands—followed by ecoregion (level 2, e.g., mountain, plain, river, delta, and lake/
reservoir), HGM feature (level 3, e.g., fringe, depression, flat, riverine), general wetland type
(level 4, e.g., tidal mudflat shore, salt flat marsh, tidal marsh margin, alkali fen, acid bog, acid wet
grass, rice paddy depression, low-gradient floodplain, steep gradient wet weather seep, and farm
pond), vegetation type (level 5, e.g., swamp or marsh), and dominant species in plant community
(level 6, e.g., willow community, reed community, or alder community). The first four levels can be
interpreted from imagery and maps, whereas levels 5 and 6 are classified through field inspections.
Considering the first 4 levels of this system, 125 types of wetlands may occur in the Republic of
Korea: 74 inland types, 36 estuarine types, and 15 coastal types.
TABLE 2.14
Classification of Amazonian Lowland Wetlands
Group 1. Wetlands with relatively stable water levels
1.1 Herbaceous and forested swamps (campos unidos, veredas, and buritizais) in the savanna belt
1.2 Forested swamps in the rainforest (palm swamps such as buritizais and mixed forests)
1.3 Open, waterlogged vegetation on the table mountains of the Guiana Shield (tepuis)
Group 2. Wetlands with oscillating water levels

2.1 Wetlands subjected to predictable, long-lasting, monomodal flood pulses
2.1.1 River floodplains with high flood amplitudes (large river floodplains of greater than fifth order along the Amazon
and its large tributaries)
2.1.1.1 Floodplains of high fertility (white-water river floodplains; várzeas)
2.1.1.2 Floodplains of intermediate fertility type A (clearwater river floodplains; igapós)
2.1.1.3 Floodplains of intermediate fertility type B (blackwater floodplains on paleowhite-water substrates)
2.1.1.4 Floodplains of low fertility (blackwater river floodplains; igapós)
2.1.2 Wetlands with low flood amplitudes (large interfluvial wetlands that are inundated mostly by rainwater)
2.1.2.1 Wetlands of low fertility and long flood periods (large areas in the upper Negro River basin)
2.1.2.2 Hydromorphic edaphic savannas of low fertility and of short-to-intermediate flood periods (some
Amazonian campinas, banas, and varillales)
2.1.2.3 Hydromorphic climatic savannas on variable soil types and variable flood length (e.g., Humaitá
savannas, Bananal/Araguaia savannas, Roraima savannas, and Beni savannas)
2.2 Wetlands subjected to short, predictable, polymodal flood pulses
2.2.1 Marine and brackish water tidal wetlands (Amazon estuary)
2.2.2 Freshwater wetlands indirectly affected by the tide (Amazon estuary)
2.3 Wetlands subjected to short, unpredictable, polymodal flood pulses
2.3.1 Wetlands associated with small rivers and streams (first to fifth order)
2.3.2 Wetlands in depressions fed by rainwater
Source: Junk, W.J. et al., Wetlands, 31, 622, 2011.

CONCLUSION
To date, most wetland mapping projects covering large geographic areas have used a horizontal
approach to wetland classification focusing on a few discrete types. Where more detailed assess-
ments are desired, hierarchical classifications like the Cowardin et al. system or this system in
combination with some form of HGM classification have been employed. Both approaches have
advantages and limitations but are tailored to meet the needs of the mapping project. While not a
classification issue per se, the biggest concern is the lack of a universally accepted wetland defini-
tion. This poses problems for researchers trying to estimate the extent of wetlands globally and to
use such data in predicting the effects of climate change on carbon sequestration and methane emis-
sion, for example. At some point, international scientists are going to have to come to grips with this
issue. In the last chapter of this book, we offer a definition that may serve as a starting point for such
dialogue. In terms of classification, researchers have a suite of variables that they can consider based
on their project objectives. Finally, the HGM concept should be incorporated into classification for
improved use of wetland inventory data for predicting wetland functions and using the data for a
wide variety of applications (e.g., ecological modeling, resource planning, and education).
REFERENCES
Adamus, P.R. 2001. Guidebook for Hydrogeomorphic (HGM)-Based Assessment of Oregon Wetland and
Riparian Sites: Statewide Classification and Profiles. Oregon Division of State Lands, Salem, OR.
Albert, D.A., D.A. Wilcox, J.W. Ingram, and T.A. Thompson. 2005. Hydrogeomorphic classification for Great
Lakes coastal wetlands. Journal of Great Lakes Research 31(Suppl. 1):129–146.
Alvo, R. and S. Ponomarenko. 2003. Vegetation classification standard for Canada workshop: May 31–June 2
2000 (Hull, Quebec). Canadian Field Naturalist 117(1):125–139.
Arkansas Watershed Resource Information Management System, 2014, HGM classification & characteris-
tics of Arkansas wetlands, http://anrc.cast.uark.edu/home/mawpt/statewide-planning-initiatives/hgm-
characterization/hgm-classificationand-characteristics-of-arkansas-wetlands.html. Accessed November 7,
2014.
Brinson, M.M. 1993. A hydrogeomorphic classification for wetlands. U.S. Army Engineers Waterways
Experiment Station, Vicksburg, MS. Wetland Research Program Technical Report WRP-DE-4. http://
el.erdc.usace.army.mil/elpubs/pdf/wrpde4.pdf. Accessed November 7, 2014.
Brooks, R.P., M.M. Brinson, K.J. Havens, C.S. Hershner, R.D. Rheinhardt, D.H. Wardrop, D.F. Whigham,
A.D. Jacobs, and J.M. Rubbo. 2011. Proposed hydrogeomorphic classification for wetlands of the mid-
Atlantic region, USA. Wetlands 31(2):207–219.
Comer, P., D. Faber-Langendoen, R. Evans, S. Gawler, C. Josse, G. Kittel, S. Menard et al. 2003. Ecological
Systems of the United States: A Working Classification of U.S. Terrestrial Systems. NatureServe,
Arlington, VA.
Cowardin, L.M., V. Carter, F.C. Golet, and E.T. LaRoe. 1979. Classification of wetlands and deepwater habitats
of the United States. U.S. Fish and Wildlife Service, Washington, DC. FWS/OBS-79/31. http://nctc.fws.
gov/resources/knowledge-resources/FWS-OBS/79_31.pdf. Accessed November 7, 2014.
Dvorett, D., J. Bidwell, C. Davis, and C. DuBois. 2012. Developing a hydrogeomorphic wetlands inventory:
Reclassifying National Wetlands Inventory polygons in geographic information systems. Wetlands
32:83–93.
Farinha, J.C., P.R. Araújo, E.P. Silva, S. Carvalho, E. Fonseca, and C. Lavinas. 2005. MedWet habitat
description system. MedWet publication. http://www.medwet.org/codde/OtherResources/Habitat.pdf.
Accessed November 7, 2014.
Finlayson, C.M. and A.G. van der Valk (eds.). 1995. Classification and Inventory of the World’s Wetlands.
Kluwer Academic Publishers, Dordrecht, the Netherlands.
Food and Agriculture Organization of the United Nations. 1998. Wetland characterization and classifica-
tion for sustainable agricultural development. Sub-Regional Office for East and Southern Africa,
Harare, Zimbabwe. http://www.fao.org/docrep/003/X6611E/X6611E00.HTM. Accessed November
7, 2014.
Gopal, B. and M. Sah. 1995. Inventory and classification of wetlands in India. Vegetatio 118:39–48.
Gore, A.J.P. (ed.). 1983. Mires: Swamp, Bog, Fen, and Moor. Regional Studies. Elsevier Scientific Publishing
Company, Amsterdam, the Netherlands.
Green, D.L. 1997. Wetland Management Technical Manual: Wetland Classification. Department of Land and
Water Conservation, Ecological Services Unit, Parramatta, New South Wales, Australia.
Grossman, D.H., D. Faber-Langendoen, A.S. Weakley, M. Anderson, P. Bourgeron, R. Crawford, K. Goodin
et al. 1998. International Classification of Ecological Communities: Terrestrial Vegetation of the United
States, Vol. I, The National Vegetation Classification System: Development, Status, and Applications.
The Nature Conservancy, Arlington, VA.
Johnson, P. and P. Gerbeaux. 2004. New Zealand wetland types. Department of Conservation, Te Papa Atawhai,
Nelson, New Zealand. http://www.doc.govt.nz/publications/science-and-technical/products/reports-and-
books/. Accessed November 7, 2014.
Josse, C., G. Navarro, F. Encarnación, A. Tovar, P. Comer, W. Ferreira, F. Rodríguez et al. 2007. Ecological
Systems of the Amazon Basin of Peru and Bolivia. Classification and Mapping. NatureServe,
Arlington, VA.
Junk, W.J., M.T. Fernandez Piedade, J. Schongart, M. Cohn-Haft, J.M. Adency, and F. Wittmann. 2011. A clas-
sification of major naturally-occurring Amazonian lowland wetlands. Wetlands 31:622–640.
Kim, K.-G., M.-Y. Park, and H.-S. Choi. 2006. Developing a wetland-type classification system in the Republic
of Korea. Landscape and Ecological Engineering 2:93–110.
Martin, A.C., N. Hotchkiss, F.M. Uhler, and W.S. Bourn. 1953. Classification of wetlands of the United States.
U.S. Fish and Wildlife Service, Washington, DC. Special Scientific Report, Wildlife, No. 20.
Morrice, J.A., A.S. Trebitz, J.R. Kelly, M.E. Sierszen, A.M. Cotter, and T. Hollenhurst. 2011. Determining
sources of water to Great Lakes coastal wetlands: A classification approach. Wetlands 31:1199–1213.
Pakarinen, P. 1995. Classification of boreal mires in Finland and Scandinavia—A review. Vegetatio 118:29–38.
Ramsar Convention Secretariat. 2010. Wetlands Inventory: A Ramsar Framework for Wetland Inventory
and Ecological Character Description. Ramsar Handbooks for the Wise Use of Wetlands, 4th edn.,
Vol. 15. Gland, Switzerland. http://www.doe.ir/portal/theme/talab/0DB/2-BS/INV/SO/bs-inv-so-bk-
gud-V15-2010.pdf. Accessed November 7, 2014
Sather, J.H. (ed.). 1976. Proceedings of the National Wetland Classification and Inventory Workshop, July
20–23, 1975, University of Maryland, College Park, MD. U.S. Fish and Wildlife Service Report,
Washington, DC.
Schultink, G. and R. van Vliet. 1997. Wetland identification and protection: North American and European
perspectives. Michigan State University, Michigan Agricultural Experiment Station. Research
Report 554.
Semeniuk, C.A. and V. Semeniuk. 1995. A geomorphic approach to global classification for inland wetlands.
Vegetatio 118:103–124.
Semeniuk, V. and C.A. Semeniuk. 1997. A geomorphic approach to global classification for natural inland wet-
lands and rationalization of the system used by the Ramsar Convention—A discussion. Wetlands Ecology
and Management 5:145–158.
Simenstad, C.A., J.L. Burke, J.E. O’Connor, C. Cannon, D.W. Heatwole, M.F. Ramirez, I.R. Waite, T.D.
Counihan, and K.L. Jones. 2011. Columbia River Estuary ecosystem classification—Concept and appli-
cations. U.S. Geological Survey Open-File Report 2011-1228.
Smith, R.D., A. Ammann, C. Bartoldus, and M.M. Brinson. 1995. An approach for assessing wetland functions
using hydrogeomorphic classification, reference wetlands, and functional indices. U.S. Army Corps of
Engineers, Waterways Experiment Station, Vicksburg, MS. Wetland Research Program Technical Report
WRP-DE-9. http://el.erdc.usace.army.mil/elpubs/pdf/wrpde9.pdf. Accessed November 7, 2014.
Stewart, R.E. and H.A. Kantrud. 1971. Classification of Natural Ponds and Lakes in the Glaciated Prairie
Region. Bureau of Sport Fisheries and Wildlife, U.S. Fish and Wildlife Service, Washington, DC.
Resource Publication 92.
The Ramsar Convention Bureau. 2006. Strategic Framework and Guidelines for the Future Development of
the List of Wetlands of International Importance of the Convention on Wetlands (Ramsar, Iran, 1971).
The Ramsar Convention Secretariat, Gland, Switzerland. http://www.ramsar.org/sites/default/files/
documents/pdf/guide/guide-list2009-e.pdf. Accessed November 8, 2014.
Tiner, R.W. 1995. Piloting a more descriptive NWI. National Wetlands Newsletter 19:14–16.
Tiner, R.W. 1997. Keys to landscape position and landform descriptors for U.S. wetlands. Operational
draft. U.S. Fish and Wildlife Service, National Wetlands Inventory Program, Northeast Region,
Hadley, MA.
Mapping. Lewis Publishers, CRC Press, Boca Raton, FL.
Tiner, R.W. 2003a. Dichotomous keys and mapping codes for wetland landscape position, landform,
water flow path, and waterbody type descriptors. U.S. Fish and Wildlife Service, Northeast Region,
Hadley, MA. http://www.fws.gov/northeast/EcologicalServices/pdf/wetlands/dichotomouskeys0903.
pdf. Accessed November 7, 2014
Tiner, R.W. 2003b. Correlating enhanced national wetlands inventory data with wetland functions for water-
shed assessments: A rationale for northeastern U.S. wetlands. U.S. Fish and Wildlife Service, National
Wetlands Inventory Program, Northeast Region, Hadley, MA.
Tiner, R.W. 2003c. Geographically isolated wetlands of the United States. Wetlands 23(3):494–516. http://
digitalmedia.fws.gov/cdm/ref/collection/document/id/1334. Accessed November 7, 2014
Tiner, R.W. 2005. In Search of Swampland: A Wetland Sourcebook and Field Guide, 2nd edn., Revised and
Expanded. Rutgers University Press, New Brunswick, NJ.
Tiner, R.W. 2010. NWIPlus: Geospatial database for watershed-level functional assessment. National Wetlands
Newsletter 32(3):4–7, 23.
Tiner, R.W. 2011a. Dichotomous Keys and Mapping Codes for Wetland Landscape Position, Landform, Water
Flow Path, and Waterbody Type Descriptors, Version 2.0. U.S. Fish and Wildlife Service, National
Tiner, R.W. 2011b. Predicting Wetland Functions at the Landscape Level for Coastal Georgia using NWIPlus
Data. U.S. Fish and Wildlife Service, Region 5, Hadley, MA. In cooperation with the Georgia Department
of Natural Resources, Coastal Resources Division, Brunswick, GA and Atkins North America,
Raleigh, NC.
Tiner, R.W. 2014. Dichotomous Keys and Mapping Codes for Wetland Landscape Position, Landform, Water
Flow Path, and Waterbody Type Descriptors, Version 3.0. U.S. Fish and Wildlife Service, National
Vegetation Subcommittee. 2008. National vegetation classification standard, Version 2. Federal Geographic
Data Committee. February 2008. FGDC-STD-005-2008. http://usnvc.org/wp-content/uploads/2011/02/
NVCS_V2_FINAL_2008-02.pdf. Accessed November 7, 2014.
Warner, B.G. and C.D.A. Rubec (eds.). 1997. The Canadian Wetland Classification System, 2nd edn., National
Wetlands Working Group. Research Centre, University of Waterloo, Waterloo, Ontario, Canada.
http://www.gret-perg.ulaval.ca/fileadmin/fichiers/fichiersGRET/pdf/Doc_generale/Wetlands.pdf.
Wetlands Subcommittee. 2009. Wetlands mapping standard. Federal Geographic Data Committee. FGDC-
STD-015-2009. http://www.fgdc.gov/standards/projects/FGDC-standards-projects/wetlands-mapping/
2009-08%20FGDC%20Wetlands%20Mapping%20Standard_final.pdf. Accessed November 7, 2014.
Wetlands Subcommittee. 2013. Classification of wetlands and deepwater habitats of the United States, 2nd
edn. Federal Geographic Data Committee. August 2013. FGDC-STD-004-2013. https://www.fgdc.gov/
standards/projects/FGDC-standards-projects/wetlands/nvcs-2013. Accessed November 7, 2014.
Whigham, D., D. Dykyjová, and S. Henjný (eds.). 1993. Wetlands of the World: Inventory, Ecology and
Management, Vol. I. Kluwer Academic Publishers, Dordrecht, the Netherlands.
Zuo, P., Y. Li, C.-A. Lin, S.-H. Zhao, and D.-M. Guan. 2013. Coastal wetlands of China: Changes from the
1970s to 2007 based on a new wetland classification system. Estuaries and Coasts 36:390–400.
3 Introduction to Wetland
Mapping and Its Challenges
Ralph W. Tiner
CONTENTS
Introduction....................................................................................................................................... 43
The Need for Wetland Maps............................................................................................................. 43
Challenges for Mapping Wetlands.................................................................................................... 55
Nature of Wetlands....................................................................................................................... 55
Environmental Factors................................................................................................................. 59
Technical Challenges...................................................................................................................60
Acknowledgments............................................................................................................................. 62
References......................................................................................................................................... 62
INTRODUCTION
Mapmaking has its origins in ancient times most likely when civilizations formed to build settle-
ments supporting large masses of people. Since then, expansion of human populations has produced
an interest and need for identifying features important for exploration, transportation, naviga-
tion, municipal planning, and strategic military planning and for depicting the geographic limits
of territories or nation states and property boundaries (public versus private lands) on paper (see
Harley and Woodard 1987 for the early history of cartography). Maps are an attempt to illustrate a
spatial concept of a feature deemed important enough by the cartographer or sponsoring agency/
organization to be depicted graphically. The feature is something seen on land, in water, or in the
sky or interpreted from various sources. The phrase “a picture is worth a thousand words” is true for
maps, for the visible representation conveys a lot of information that can be interpreted by people of
varied backgrounds. According to one source, the first maps prepared to represent vegetation types
were drafted in the mid-nineteenth century (Spalding et al. 1997).
Mapmaking has been part science and part art, yet with advances in remote sensing, the over-
whelming emphasis is on science. Consequently, all maps have limitations based on the mappability
of the feature, the effort exerted to delineate the feature, the objective and skill of the mapper, the
method or technology used to prepare the map, and, of course, scale. In this chapter, after a brief
introduction to wetland mapping with emphasis on the United States, the challenges faced by agen-
cies, organizations, and individuals interested in mapping these valuable natural resources will be
discussed. Since remote sensing applications are the focus of this book, other techniques for map-
ping such as ground surveys or combining existing geospatial data layers (e.g., derivative maps,
such as the peatland map of Ireland by Connelly et al. 2007 or the Mauritius wetland inventory by
Mamoun et al. 2013) will not be examined.
THE NEED FOR WETLAND MAPS

Throughout history, the values human societies place on wetlands have changed with their evolv-
ing cultures and needs (e.g., Vileisis 1997). Societies more directly dependent on natural resources
43
(a)
(b)
FIGURE 3.1 Early survey maps showing wetlands. (a) 1748 survey map for planning the town of Alexandria,
Virginia, produced by George Washington. He was a land surveyor prior to achieving his later accomplish-
ments in the America’s Revolutionary War and as the first president of the United States. (From Library of
Congress; g3884a ct000368r http://hdl.loc.gov/loc.gmd/g3884a.ct000368r.) (b) Portion of a 1880 map showing
a location of private land parcels for Prince Edward Island, Canada. The rectangular shape of the lots was
designed to include some wetland (shown by symbols) for raising livestock. (Provided by Matthew Hatvany.)
Introduction to Wetland Mapping and Its Challenges 45
may have mapped wetlands because of their value for harvest of natural products. For example,
when North America was first colonized by Europeans, salt marshes were a vital resource for a
farm-based economy (Tiner 2013). The earliest villages were established in close proximity to these
resources. Land surveyors identified wetlands and waterways on their field-based maps that were
then converted to plat maps (Figure 3.1a). Such maps often used symbols to represent the more con-
spicuous wetlands. When Prince Edward Island was first settled in the 1600s, the island was divided
up into rectangular parcels so that each landowner would have access to marsh hay to feed livestock
(Hatvany 2001; Figure 3.1b).
Wetlands have been depicted on maps of various kinds designed for other purposes. During
wars, maps were produced showing enemy positions and fortifications as well as waterways and
wetlands that were important for military logistics (Figure 3.2). In 1807, U.S. President Thomas
Jefferson established the Survey of the Coast to publish nautical charts that, in addition to map-
ping water depths and hazards to navigation (e.g., shoals), showed the location of tidal wetlands
(Figure 3.3a). Since 1884, the U.S. government has produced topographic maps that show general
locations of marshes and swamps among other features (Usery et al. 2009; Figure 3.3b).
While early settlements depended on salt marshes for raising livestock, as the American popu-
lation expanded, the industrial revolution commenced, and eastern forests were cleared for crop-
land and pasture, these wetlands became less vital to the economy as food for livestock could be
secured from upland pastures. Salt marshes and other wetlands became viewed as “wastelands”
or public nuisances (e.g., mosquito-breeding areas) that were unhealthy and should be filled or
drained and used for “productive” purposes (e.g., real estate or agricultural land) (Tiner 2013).
Reclamation—the conversion of wetlands to “dry land”—became a major goal of the U.S. govern-
ment (e.g., passage of Swamp Land Acts in mid-1800s). The development of mechanized equipment
for dredging, ditch-digging, and land clearing and the invention of tile drains were instrumental
in promoting wetland conversion to agricultural land and marketable real estate (see Vileisis 1997
for a detailed history of U.S. wetlands). In 1906 and 1922, the U.S. Department of Agriculture
(USDA) conducted what may be considered the first two national wetland inventories for the inter-
est in reclamation for agriculture (Shaw and Fredine 1956). Surveys of peat deposits were also
FIGURE 3.2 Section of a 1775 map showing Boston, Massachusetts, prepared to show British fortifications.
Also shown are wetlands along surrounding waters. (From Library of Congress; g3764b ct000070 http://hdl.
loc.gov/loc.gmd/g3764b.ct000070.)
(a)
(b)
FIGURE 3.3 Early government maps used symbols to display wetlands; here are two examples showing
wetlands along the Connecticut shore of Long Island Sound. (a) 1838 coastal geodetic survey map. (From
NOAA.) (b) 1893 topographic map. (From U.S. Geological Survey archives.)
conducted due to their economic importance. For example, peat and muck were some of the soil
types mapped by the USDA on their earliest soil survey maps in the early 1900s. Only after more
than a century of reclamation did wetlands begin to regain their importance to American society
as natural resources.
In the 1950s, government wildlife biologists became quite concerned about the escalating
losses of wetlands and their effect on wildlife, especially waterfowl. The post–World War II build-
ing boom was exacting a heavy toll on wetlands, especially coastal marshes. The U.S. Fish and
Wildlife Service (FWS) in cooperation with state wildlife departments conducted its first national
wetlands inventory in 1953 and 1954 to identify wetlands in regions of most states that contained
90% of the wetlands recognized as important for waterfowl. The findings were summarized in a
national report commonly known as Circular 39 (Shaw and Fredine 1956). A few years later, they
expanded this effort to include all wetlands greater than 16 ha (40 acres) in size and rated wetlands
in terms of their vulnerability to development. A series of state reports documented the status of
wetlands and later ones emphasized losses of coastal wetlands (e.g., U.S. Fish and Wildlife Service
1959, 1965). Small-scale “state wetland maps” were included in the former reports showing the
distribution of major wetland complexes (Figure 3.4). At the same time, marine scientists study-
ing estuaries documented the role of salt marshes as nursery grounds for commercially important
fishes and shellfish. Faced with accelerating losses of these and other wetlands, state legislatures
began to take action to restrict or limit these losses. In 1963, the state of Massachusetts passed the
first law to protect salt marshes from filling and other activities. Two years later, Massachusetts
passed a similar law to protect inland wetlands recognizing their importance to wildlife and for
providing a multitude of environmental services (e.g., flood water storage, shoreline stabiliza-
tion, and water quality renovation). Other northeastern states later passed similar laws and the
U.S. federal government also began regulating uses of wetlands more strictly under the Rivers
and Harbors Act and the Clean Water Act (CWA). Since the 1970s, wetlands in the United States
have been considered natural resources worthy of special protection by “environmental” laws.
The environmental movement underscored the need to map the location of these valuable natural
resources in more detail.
FIGURE 3.4 In the 1950s, the U.S. FWS produced generalized state wetland maps showing wetlands rated
by their importance to waterfowl. This figure shows a portion of the Maryland map. Red, high value; blue,
moderate; yellow, low; and orange, negligible. The maps were enclosed within state wetland reports. (From
U.S. Fish and Wildlife Service.)
While government agencies depicted wetlands on maps for municipal master plans and land use
planning documents or for acquisition of wildlife management areas beforehand, it was not until the
1970s that “wetland inventories” and the production of wetland maps were initiated by some states.
The first of these surveys were often designed to identify tidal marshes protected by state tidal wet-
land laws. Few states mapped inland wetlands probably due to the extent, challenge, and expense of
doing so (see Martel Laboratories Inc. 1976 for a review of state and local wetland surveys between
1965 and 1975). The federal regulatory agencies (Corps of Engineers and Environmental Protection
Agency) did not produce maps, but required on-the-ground delineations for wetland identification.
Comprehensive wetland mapping by the U.S. government was initiated by the U.S. FWS in
the 1970s. Prior to that time, U.S. Geological Survey (USGS) topographic maps were the only
maps showing wetlands across the country (http://nationalmap.gov/ ) (Figure 3.5). In 1974, the FWS
established the National Wetlands Inventory Project (NWI) to produce maps showing the general
boundaries of wetlands to promote wetland conservation. At the beginning of the NWI, coarse-
spatial-scale maps (small-scale; 1:100,000) were the primary product, but it was soon realized that
these maps did not provide enough detail to aid FWS biologists and others in their review of projects
requiring federal permits or expenditures. Finer-spatial-scale maps (1:24,000 for the conterminous
United States and 1:63,360 for Alaska) became the standard product until the mid-1990s when
they were replaced by digital geospatial data and online maps (Tiner 2009). The production of the
first generation of NWI maps was generally a two-step process involving photointerpretation and
cartography, with various levels of quality assurance and field review. In the first step, pen and ink
overlays of aerial photos were produced through viewing images on a stereoscope, while the sec-
ond step transferred these data to a USGS base map. Map overlays were first produced but proved
too cumbersome for field use, so hard-copy maps were produced by photographing the wetland
overlay on top of a topographic map (Figure 3.6). Later with the advent of geographic information
system (GIS) technology, NWI data were produced by on-screen interpretation and digitization.
FIGURE 3.5 A portion of a modern-era topographic map for Connecticut. Note the use of symbols to indi-
cate marsh, swamp, and aquatic beds. The more obvious wetlands are usually depicted, while drier-end wet-
land areas typically lack these symbols. (From U.S. Geological Survey.)
FIGURE 3.6 A portion of a 1980s NWI map for Connecticut. Wetlands and deepwater habitats are delin-
eated on a USGS topographic map and labeled with an alphanumeric code (e.g., PFO1E—palustrine forested
wetland, broad-leaved deciduous, seasonally flooded/saturated). (From U.S. Fish and Wildlife Service.)
Today, NWI data are viewed via an online mapper and users can produce custom maps for their
area of interest (Figure 3.7; http://www.fws.gov/wetlands/Data/Mapper.html). Many state agencies
and some local governments (e.g., City of New York) have provided funds or contributed in other
ways to the NWI to produce wetland inventories for their areas of interest. Since the NWI process
remains labor intensive, it has not been completed for the country, although priority areas (e.g., areas
of heavy development) have been updated one or more times to provide more current information.
In May 2014, the FWS completed the NWI for the lower 48 states, Hawaii, U.S. Trust Territories,
and 35% of Alaska (Bergeson 2014). Most of the NWI data are from the 1980s and of limited utility
in many if not most areas for a variety of reasons. Significant areas that have been recently mapped
have been done so with “scalable” maps derived from varied sources due to declining budgets. Many
states have been or are major contributors to the NWI. A relatively recent initiative by the NWI in
cooperation largely with state agencies has produced maps showing wetlands classified by land-
scape position, landform, water flow path, and their predicted significance for providing different
functions, potential wetland restoration sites, and areas that may support wetlands based on hydric
soil mapping (Figure 3.8). These products called NWI+ data have been produced for a limited
number of areas due to priorities and funding limitations. The results are presented via an online
mapper—NWI+ web mapper—that is sponsored by the Association of State Wetland Managers
through their “Wetlands One-Stop Mapping” site (http://aswm.org/wetland-science/wetlands-one-
stop-mapping). The standard NWI data and the NWI+ data can be displayed on a variety of base
maps (e.g., topographic map or imagery) using the online tools. NWI geospatial data are a primary
source of collateral data used by remote sensing analysts. For nearly four decades, NWI maps have
helped regulators, landowners, and would-be property owners locate wetlands that may be subject
to federal and other regulations when planning development or land acquisition. The maps have also
played a vital role in assisting natural resource agencies/organizations in formulating wetland con-
servation strategies and for predicting the effect of sea-level rise on coastal communities.
The National Oceanic and Atmospheric Administration has also produced land use/land cover
maps that show wetlands as part of their Coastal Change Analysis Program (C-CAP; http://www.
csc.noaa.gov/digitalcoast/tools/lca/; see Chapter 5 for further discussion). They utilize remote sens-
ing techniques to map land cover and land use changes in coastal watersheds, with wetlands desig-
nated as one type of land cover. To date, analyses have been conducted for 1996, 2001, and 2006.
(a)
(b)
FIGURE 3.7 Examples of displays of wetlands from the Wetlands Mapper for Ellendale, Delaware.
(a) Wetlands on aerial image (all wetlands are forested or scrub–shrub). (b) Wetlands on a topographic map
(darker green areas are wetlands; other green from base map). (From U.S. Fish and Wildlife Service, http://
www.fws.gov/wetlands/Data/Data-Download.html.)
(a)
(b)
(c)
FIGURE 3.8 Examples of displays from the NWI+ mapper for Ellendale, Delaware. (a) Wetlands classified
by water flow path (green, outflow types; blue, throughflow; and red, vertical flow). (b) Wetlands predicted to
be important for streamflow maintenance (red, high potential; orange, moderate potential). (c) Potential wet-
land restoration sites displayed on imagery (darker green, drained wetlands now cropland; lighter green, partly
drained wetlands; blue, excavated wetlands). (From Association of State Wetland Managers, http://nwiplus.
cmi.vt.edu/nwiplusmapper/.)
(a) (b)
(c) (d)
FIGURE 3.9 Examples from NOAA’s C-CAPS. (a) Land cover/use map showing tidal wetlands (purple
areas) around Charleston, South Carolina. (b) Change image showing wetlands and uplands lost in one loca-
tion in this area from 1990 to 2011 (red areas). (c) 1990 image of area. (d) 2011 image. (From NOAA Digital
Coast; supplied by Nate Herold.)
The data are viewable via an online mapper—C-CAP Land Cover Atlas (http://www.csc.noaa.gov/
ccapatlas/#). From the wetland perspective, the atlas shows a few types of wetlands and changes
from one time period to the next (Figure 3.9). With increasing concern about rising sea levels,
NOAA’s Sea Level Rise and Coastal Flooding Impacts Viewer should be invaluable to coastal com-
munities. The viewer produces online maps showing the predicted effect of different sea-level rise
scenarios on the coast (http://www.csc.noaa.gov/digitalcoast/tools/slrviewer).
Since the 1890s, the USDA has been producing soil survey maps for use in agricultural and land
use planning and soil conservation. While not strictly wetland maps, they have been used to predict
the location of wetlands or potential wetlands. The maps now show soils classified by soil series
or associations and were originally published in hard-copy form in often county-based soil survey
reports. Early soil maps identified certain wetlands as land types, such as tidal marsh, swamp,
and muck. Today, the maps like the NWI maps are displayed via an online mapper—Web Soil
Survey (http://websoilsurvey.sc.egov.usda.gov/App/WebSoilSurvey.aspx). In addition to display-
ing soils by their soil series, the mapper can highlight soils designated as “hydric soils” thereby
showing areas that are wetlands, former wetlands, and sites that have potential to contain wetlands
(Figure 3.10). These data have been invaluable in locating wetlands and areas that likely support
wetlands and are frequently used to aid in wetland interpretation through remote sensing as well as
for field investigations.
While the U.S. Environmental Protection Agency and the U.S. Army Corps of Engineers do
not have their own large-scale wetland mapping efforts underway, they do produce wetland maps
for site-specific areas that involve unauthorized alterations of wetlands under federal juris-
diction of the U.S. CWA and use remote sensing for implementing environmental policies
WoeB
FapA
FrkB
FapA
Water
eC HocA
pA FapA HocB
HocB
EveC
A HocB
FapA HocA
FapA
HumAt
ShrA
SacB SacC
WogA FapA
(a)
WoeB HocA
WoeB
FapA
FrkB
FapA
Water
HocA
FapA FapA HocB
HocB
EveC
gA HocB
FapA FapA HocA
HumAt
ShrA
SacB SacC
WogA FapA
FapA WoeB FapA
EkaAr
(b)
FIGURE 3.10 Section of display of soil mapping from USDA’s Web Soil Survey for Assunpink, New Jersey.
(a) Showing soil map units—hydric ones are EkaAr (Elkton loam), HumAt (Humaquepts, frequently flooded),
and FapA (Fallsington loam). (b) Showing hydric soil ratings of all soils: red, 100% hydric; orange, pre-
dominantly hydric (66%–99%). (From USDA Natural Resources Conservation Service, http://websoilsurvey.
sc.egov.usda.gov/App/WebSoilSurvey.aspx.)
(Mayer and Lopez 2011). They use remote sensing techniques for environmental compliance and
to determine pre- and post-violation condition of wetlands for court cases. These agencies have
also provided financial support for wetland mapping by the NWI, state, and local governments
or conducted their own mapping for identifying wetlands of federal concern under the CWA or
on federal installations.
Mapping of potential wetland sites and stream locations has also been conducted in the Canadian
Maritimes using GIS techniques (derivative mapping) (Murphy et al. 2007). The initiative called
“Provincial Wet Areas Mapping” (WAM) is being conducted by the Nexfor/Bowater Forest
Watershed Research Center at the University of New Brunswick and has been completed for New
Brunswick, Nova Scotia, and large areas in Alberta and Maine among other locations (Ogilvie et al.
2013). The WAM model produces a soil wetness index that can be used to predict the flow of water
across the landscape and where it will accumulate. The analysis is based on digital elevation model
(DEM) data and the known location of surface water bodies and wetlands identified through pho-
tointerpretation and more recently by analysis of Light Detection and Ranging (LiDAR) data. The
mapping has been invaluable for planning timber harvest (e.g., road alignments), wildlife habitat
management, and planning development in general. (See Chapter 6 for additional discussion on
Canadian wetland mapping efforts.)
Globally, mangroves, sea grasses, coral reefs, and peatlands have been recognized as valuable natu-
ral resources whose distribution and trends are of the utmost importance. Consequently, researchers
from around the globe have been involved in mapping these features (see Chapters 11, 12, and 17–19)
(Krankina et al. 2008). The results were initially presented on paper maps but more recent efforts
portray mangrove extent on digital products. Compilation of maps into atlases have been prepared
for mangroves (Spalding et al. 1997, 2010), sea grasses (Green and Short 2003), and corals (Spalding
et al. 2001). Maps of peatlands may be available for some countries because of their economic impor-
tance (e.g., Canada) (Tarnocai et al. 2002). Also given recent interest in climate change and estimating
global carbon sequestration, many researchers have used remote sensing to produce maps or data that
predict carbon storage and methane production (e.g., Sitch et al. 2007; see Chapter 24). The Joint
Research Centre of the European Commission has produced soil atlases for Europe, the Northern
Circumpolar region, and Africa, with another one in production for Latin America (http://eusoils.jrc.
ec.europa.eu/library/maps/maps.html). Certain soil types can be used to visualize possible locations
of only the larger wetlands across these regions since the maps are produced at a coarse spatial scale.
In the United States, the National Aeronautics and Space Administration (NASA) has initiated the
MEaSUREs Program to use remotely sensed data to help better understand the Earth as a water–land–
atmospheric system. One of the projects supported by this program is mapping the world’s “inun-
dated” wetlands and changes from 1992 to 2009 (NASA 2014). The data will be used by researchers
who study (1) global cycling of methane, carbon dioxide, and water; (2) interactions among climate,
greenhouse-gas emissions, and water exchange; (3) climate change impacts and feedbacks; (4) eco-
logical health and biodiversity; and (5) water resource management at regional and global levels.
While mapmaking has migrated from handmade paper maps to computer-generated online prod-
ucts, wetland maps or “geospatial data” continue to serve as valuable resources for identifying the
distribution and status of wetlands, monitoring trends, portraying the connectivity between wetlands
and water bodies and among wetlands, evaluating wetland functions and values, and modeling eco-
logical processes and the impacts of climate change. As such, they are widely used by government
agencies, organizations, and individuals for land use and natural resource planning and by scientists
in designing and conducting research. Wetland maps are vital for persons seeking to purchase land
as wetlands may have legal restrictions on them in some countries, and even if not, those lands have
certain physical limitations for development due to their soil properties and hydrology. Simply by
producing wetland maps, government agencies have shown the public that wetlands are valuable
resources—important enough for significant expenditures of time, energy, and funds. Furthermore,
given the widespread recognition of wetlands as important natural resources, remote sensing is
being used to map their distribution and condition and to track changes in wetlands over time (e.g.,
see many examples presented in the other chapters of this book plus Jensen et al. 1987; Choung and
Ulliman 1992; Munyati 2000; Paula and William 2000; Ozesmi and Bauer 2002; McCarthy et al.
2003; Sugumaran et al. 2004; Kashaigili et al. 2006; Baker et al. 2007; Davidson and Finlayson
2007; McHugh et al. 2007; Islam et al. 2008; Lee and Yeh 2009; Hoekman 2009; Reif et al. 2009;
Li et al. 2010; Zhao et al. 2010; Gao and Xing 2011; Klemas 2011; Mahmud et al. 2011; Melack and
Hess 2011; Ghobadi et al. 2012; Mwita et al. 2012; Evans and Costa 2013; Huang et al. 2014; and
Martínez-López et al. 2014).
CHALLENGES FOR MAPPING WETLANDS

Nature of Wetlands
The diversity of wetlands around the globe makes wetland identification and conducting wetland
inventories a challenge for both the field scientist and the remote sensing analyst. Wetlands include
a range of habitats from permanently flooded sites to areas that are never flooded but are season-
ally wet for variable durations (Chapter 1). In general terms, the wetter the wetland, the easier it is
to identify both on the ground and through remote sensing. This can be restated another way—the
more difficult the wetland is to identify on the ground, the less likely it will be able to be detected
by remote sensing. Consequently, drier-end wetlands offer the greatest test to both the field delinea-
tor and the remote sensing analyst. Local expertise may be required to identify these wetlands on
the ground.
The near permanence of water creates conditions that favor colonization by obligate hydro-
phytes forming vegetation that is markedly different than that growing on dry land. Distinctive
plant communities and the presence of water combine to make for easy identification (Figure 3.11).
Consequently, aquatic beds, marshes, the wetter wet meadows, mangrove swamps, and many bogs
and fens are virtually unmistakable types and can be mapped with consistency and high reliability.
While many of these types can be detected on imagery at any season (except when snow covered),
aquatic beds and marshes dominated by nonpersistent vegetation require that imagery be acquired
when plants are leafed out and, for the former, also during calm water conditions (e.g., minimal
wind or wave action and no turbidity).
Tidal salt and brackish marshes and mangrove swamps are often large expanses of vegetation
that is distinctly different from neighboring uplands (Figure 3.12). Also the drainage pattern of tidal
creeks and/or mosquito ditches offers further clues to their occurrence. The extent of most tidal
wetlands (including high marsh salt flats—“salinas”) should be identifiable year-round since most
FIGURE 3.11 Color infrared aerial photo showing clear distinction between deciduous forested wetlands
(darker gray areas) and upland evergreen forests (red). Notice the evergreen trees (red) surrounding a small
pond at the top middle of the photo; this is a black spruce (P. mariana) bog. (From State of Massachusetts.)
FIGURE 3.12 Satellite image of salt marshes behind Georgia’s barrier islands and tidal wetlands along the
Altamaha River. (Copyright and used with permission from Geospatial Division, MDA Information Systems
Inc., Gaithersburg, MD.)
of the vegetation is persistent. Freshwater tidal marshes, however, may be dominated by nonper-
sistent species (dieback in winter) or by annual plants, so the extent of their vegetation is limited to
the growing season; they appear as mudflats or open water at other times. Large stands dominated
by a single species such as smooth cordgrass (Spartina alterniflora) and black needlerush (Juncus
roemerianus), and even different growth forms of these species (short versus tall) should be read-
ily detected, whereas mixed communities may make it difficult to distinguish species composition.
Different densities of mangroves have been successfully identified and separation of some spe-
cies may be possible especially with more advanced technologies (Heumann 2011; Elmahdy and
Mohamed 2013). Identification of tidal flats requires acquisition of low-tide imagery for best results.
Freshwater marshes, fens, and bogs are typically dominated by species that do not occur in
abundance on uplands, with some exceptions (e.g., black spruce Picea mariana in boreal regions).
Consequently, they should be readily mapped at least in more temperate regions (Figure 3.13).
Association of these wetlands with a water body further facilitates their recognition. Beaver-
influenced wetlands should be recognized by the presence of a beaver pond, dam, and perhaps the
beaver lodge itself. Similarly muskrat and nutria marshes may be detected by the presence of musk-
rat houses for the former and eat-outs (open areas cleared by these furbearers) for both species. In
boreal regions, separating wet black spruce stands from dry-site stands is challenging and requires
consideration of tree density (more dense for wetlands) and plant height (stunted for bogs) and per-
haps multiseason imagery to capture dry conditions for upland sites. High-resolution topographic
data (e.g., LiDAR) may be required to identify wetlands from uplands in areas of low relief like the
Arctic Coastal Plain (Paine et al. 2013).
Other wetlands are colonized and dominated by mesic (moist soil) vegetation—plants that have
been considered facultative-type hydrophytes (Tiner 1991). These plants are not unique to wetlands
since they are capable of growing in both wetlands and uplands. This led early Dutch plant ecologist
Eugenius Warming (1909) to conclude that in some places “it is impossible to establish a sharp dis-
tinction between swamp-forests and forests on dry land” (p. 192). Early American plant ecologists
Weaver and Clements (1929) remarked “amphibious plants have a wide range of adjustment and
may grow for a time as mesophytes or partially submerged” and that they are the “least specialized
FIGURE 3.13 Color infrared aerial photo showing a variety of wetlands along ponds and streams. Ericaceous
shrub bog (orange-brown area along large pond), marshes (white areas), deciduous shrub thickets (medium
grayish areas), and mixed deciduous/evergreen swamps (dark gray with red clusters, bottom middle of image).
(From State of Massachusetts.)
of aquatic plants” (p. 343). The fact that these plants can grow well in both wetlands and moist
uplands complicates wetland recognition. Many of the wetlands dominated by these species are
difficult to identify by vegetation alone. Examination of soils and verification of hydric soil proper-
ties is often required for positive identification (Tiner 1999). Landscape position, landform, and the
presence of a neighboring water body may help one recognize these plant communities as wetlands.
Temporarily or seasonally flooded types can be readily identified if imagery is captured when the
sites are flooded or when soils are saturated to the surface, especially when deciduous woody plants
are in their leaf-off condition. Yet low-lying uplands when subject to rare flooding can be mistaken
for wetlands, especially on the upper terraces of floodplains. After all, floodplains include both wet-
lands and uplands. When woody plants are in full leaf (including evergreen species), their identifica-
tion on imagery is not straightforward. Interpretation by association—a low-lying landform along a
water body—might be used to map them. The driest of wetlands—seasonally saturated types—are
the most challenging even for identification in the field and are likely missed or underrepresented
by most inventories. Their identification typically requires examination of soil properties to docu-
ment the occurrence of “hydric soils” (e.g., histosols, mineral soils with histic pipedons, and gleyed
soils). Although the distinction between hydric soils and nonhydric soils are quite obvious in many
cases, in some situations (e.g., low-gradient topography on coastal and glaciolacustrine plains and
the permafrost region), such differences may not be apparent. Wet season imagery offers, of course,
the best chances for differentiation.
Regions experiencing marked annual changes in precipitation also create challenges for wetland
mapping. For example, interannual differences in precipitation and runoff cause significant water
level fluctuations in North America’s Great Lakes producing changes in the vegetation of lakeshore
wetlands (see Chapter 15). Typically, vegetation shifts from aquatic beds/marshes in wet years to
marshes/wet meadows in dry years. Vegetation changes also occur elsewhere in regions subject
to pronounced fluctuations in annual precipitation amounts, distinct wet–dry cycles (including
Mediterranean climates), and long-term droughts. In arid regions, ephemeral wetlands may appear
on the landscape during extremely wet periods and be dry for years thereafter.
Tropical wetlands, with the exception of mangroves, are relatively poorly studied. By some
accounts, they are largely permanently or periodically flooded evergreen forested wetlands and
grasslands (Bwangoy et al. 2010). Collecting imagery in the equatorial zone has been a problem
due to persistent cloud cover. Central Africa’s Congo River Basin experiences two wet seasons and
two dry seasons annually. Its northern basin has a rainy season from March to November, whereas
the southern basin receives precipitation from September to early June (Bwangoy et al. 2010). This
is quite different from the Amazon where wetlands are exposed to prolonged drought and can be
recognized by their greenness during droughts (Saleska et al. 2007). This observation suggests
that acquisition of imagery during the dry season may be best for identifying tropical wetlands
where they experience a pronounced drawdown of the water table. Wetland detection could also
be complicated by the heterogeneity of the diverse tropical forests—their canopies and understo-
ries (Saalovara et al. 2005). Access to high-quality topographic data may be required to improve
mapping accuracy of tropical wetlands based on the Congo work (Bwangoy et al. 2010) and that of
Gumbricht (Chapter 20). Field studies in many tropical regions are hampered by the lack of access
in remote areas as well as the difficulties coping with the heat and natural environmental conditions
that wetlands present (e.g., wet soils make travel difficult). The latter is true for other regions where
landscapes are dominated by wetlands, such as northern circumpolar regions.
Arctic wetlands pose some unique challenges due to the presence of permafrost and low spe-
cies diversity in general. Some problems that encountered mapping wetlands in Alaska based on
NWI’s interpretation of aerial photography include (1) defining wetland hydrology (i.e., water
regime) in regions of continuous and discontinuous permafrost due to surface water drying during
the summer period of nearly 24 h of sunlight and also its effect on melting of permafrost through
summer; (2) determining hydrology of streamside shrub thickets (e.g., flooded or saturated or not
upland); (3) uplands can be flooded by heavy precipitation in areas of shallow discontinuous perma-
frost; (4) mapping slope wetlands on paludified slopes and on north-facing slopes since the spectral
signature of black spruce (P. mariana) is similar to that of white spruce (Picea glauca); (5) separat-
ing wet from dry tundra; (6) distinguishing deciduous trees from tall willows and short willows
(i.e., requires stereoscopic interpretation); (7) confusing shrubs with herbaceous cover in flooded
areas; (8) separating deciduous forested wetlands from upland forests on floodplain terraces; (9)
determining the break between estuarine marshes and freshwater tidal marshes in areas of low tidal
range (0.5 m); (10) vegetation and hydrologic complexity in polygonal wetlands and freeze–thaw
lakes makes it difficult to separate individual types; (11) determining the limits of tidal action in
coastal rivers and associated wetlands (Julie Michaelson and John Anderson, personal communica-
tion 2013). The lack of sufficient sunlight during a large portion of the year also presents problems
for capturing aerial photographs (Megan Lang, personal communication 2014). In general, northern
peat-dominated wetlands tend to be underrepresented on global wetland maps that focus on inun-
dated wetlands, yet peatland mapping can be improved by applying a continuous field mapping
approach, using multitemporal data and other sensors (e.g., MODIS), and placing more attention on
spectral differences especially that of peat mosses (Sphagnum spp.) on open lands and on canopy
architecture in forested situations (Krankina et al. 2008).
Some researchers have been experimenting with creating spectral libraries of dominant wet-
land species and mixed vegetation for use in generating plant community maps of wetlands from
hyperspectral imagery (Zomer et al. 2009). Spectrometers document in-field spectra of vegetation
that can then be applied to interpret remotely sensed data. Clearly, there are challenges for doing
this on a regional, national, or global scale in that species subject to varied environmental condi-
tions and growing in different densities may produce varied spectral responses that will require
more advanced image analysis and processing techniques as well as significant expenditures. The
more homogeneous the plant community, the easier it should be to detect through remote sensing
(e.g., large monotypic stands should be the easiest), recognizing that differences in other features
(e.g., bare earth and water) may produce different spectral signatures for the same species due to
changes in plant density and vigor. Depending on imagery scale and the size of the patches, group-
ings of the same species that are subject to different environmental conditions may be classified as
separate units or may have to be combined into mixed groups. More heterogeneous communities or
randomly distributed patchy stands are more challenging and classifications may have to be limited
to general categories or mixed types with low-spectral-resolution or low-spatial-resolution imagery
or else analyzed with higher-resolution imagery (Rosso et al. 2005).
Environmental Factors
Environmental factors influencing the quality and utility of imagery for wetland mapping include
plant phenology (e.g., leaf-off condition for deciduous trees or flowering of herbs), shadows cast by
tall objects (e.g., trees and buildings), variable hydrologic conditions (e.g., flooding, soil saturation,
drought, and tide levels), cloud cover (tropics, e.g., Amazon, Congo, and Southeast Asia), snow and
ice in higher latitudes and elevations (montane), and growing season (short growing season in arctic
versus year-round in subtropics and tropics). Timing of imagery is critical to facilitating interpre-
tation. For deciduous forested wetlands, imagery captured during leaf-off, wet season conditions
(e.g., spring in most temperate regions) is best. Wet season imagery is also required to identify
temporarily inundated prairie pothole wetlands before they are planted (Kevin Bon, personal com-
munication 2013) as well as for the most accurate identification of temporarily flooded wetlands
(e.g., floodplain wetlands), seasonally saturated wetlands (e.g., flatwoods), and farmed wetlands
elsewhere (Figure 3.14). For similar wetlands in Mediterranean climates, imagery captured dur-
ing the rainy season (winter) provides the best results (e.g., De Roeck et al. 2008). Identification of
wetland hydrology based on one-time imagery is challenging, especially if the imagery is captured
during the dry season or drought.
Human activities have had an enormous impact on wetland detection. Without human interven-
tion, wetlands exhibited patterns that followed natural drainages, certain landforms (e.g., depres-
sions), or landscapes that were largely wet regions (e.g., arctic coastal plain and paludified landscapes
of the boreal forest and tundra). Land clearing, drainage, filling, and construction of roads, railroads,
buildings, and other infrastructure have fragmented the natural landscape to varying degrees around
the globe. This fragmentation has divided large wetlands into smaller and smaller pieces, leaving
only remnants today in many places, especially agricultural districts and urban/suburban locales.
FIGURE 3.14 Color infrared aerial photo showing coastal plain wetlands on the Delmarva Peninsula,
United States. Wet season, leaf-off imagery provided the best detection of wet areas. The interspersion of wet
and dry areas can be seen left half of the photograph. (From U.S. Geological Survey.)
In some cases, clearing of forests may make it easier to detect wetlands from uplands, especially
in areas of low topography where wet soils may be observed. In other cases where the wet forests
were readily distinguished from dry forests on remotely sensed imagery, timber harvest and for-
est regeneration may make it more difficult to delineate wetlands, particularly when saturated or
flooded soils are not apparent. Human activities have also altered natural drainage patterns that
either effectively drain wetlands or in other situations create wetlands where they formerly did not
exist. Drainage and channelization projects, both large and small, have complicated wetland iden-
tification and delineation for both on-the-ground and remote sensing. For many areas, field inspec-
tions and hydrologic monitoring are required to definitely establish whether the site is effectively
drained or still maintains the minimum wetness required to be classified as wetlands for regulatory
purposes. Irrigated lands pose similar problems as some areas may represent “artificial wetlands”
while other areas are not wet long enough to be considered wetland. Impounding streams and dam-
ming rivers have altered natural flooding regimes, reduced sedimentation, and diverted water for
various uses, thereby disrupting downstream wetland and other aquatic ecosystems. This compli-
cates the interpretation of floodplain forests—determining whether cottonwoods represent forested
wetlands on an active floodplain or nonwetland forests on a former floodplain along a river with
altered hydrology (Kevin Bon, personal communication 2013). These actions may have also caused
permanent flooding of former wetlands for some distance upstream behind the impoundment. In
urban areas, connectivity between wetlands and watercourses is often obscured by impervious sur-
faces. Creation of forest plantations in wetlands may produce confusing spectral signatures between
evergreen shrub and forested wetlands (e.g., pine plantations in Florida; John Anderson, personal
communication 2013). Finally, even smoke from fires (man-made or natural) creates turbidity in the
air that can complicate or obscure spectral signatures.
Technical Challenges
Remote sensing has been used for wetland inventories (e.g., Rebelo et al. 2009; Tiner 2009; Ramsar
Convention Secretariat 2010) as well as for research investigations (e.g., Akumu et al. 2010; Klemas
2013; see Lampman 1993 for an extensive bibliography of the latter uses). Remote sensing includes
aerial photointerpretation and analysis of satellite imagery or other geospatial data collected by a
variety of airborne or satellite-borne sensors. Aerial photography (black and white, true color, or
color infrared) has been used for decades to map wetlands in detail (see Chapter 4). Satellite imag-
ery collected by a few sensors were initially used to produce more generalized wetland surveys (e.g.,
regional studies), but technological advances have now improved the ability of satellites to produce
higher-resolution data for investigators (see Chapter 5 and other chapters in this book). Collection
of remotely sensed data through airborne sensors like LiDAR (Light Detection and Ranging) have
helped bridge the gap between large-scale aerial photography and satellite imagery and are being
used to produce high-resolution geospatial data (Jensen 2007; Wang 2010; Klemas 2011).
As mentioned previously, for several decades, wetland mapping was conducted through inter-
pretation of aerial photographs (Tiner 1997). Most of this mapping was done by stereoscopic
interpretation—simultaneously viewing pairs of aerial photos that produced a 3D image of the
landscape. This process facilitated identification of topography so that depressions, slopes, flats,
bottomlands, and floodplain terraces could be readily detected. With the advent of the computer
age, GIS, and digital imagery, stereoscopic interpretation has been largely replaced with on-screen
monoscopic viewing and heads-up digitizing. On-screen interpretation does have an advantage over
traditional photointerpretation by being able to zoom in and view areas at varying scales versus a
fixed scale for hard-copy aerial photographs. Stereoscopic on-screen interpretation is possible but
requires costly hardware, specialized optics, and digital stereo imagery that are beyond the budget
of most agencies. The State of Massachusetts is one exception as it uses these techniques to produce
highly detailed and accurate wetland maps for their Wetlands Conservancy Program. The quality of
aerial imagery has always been a concern. For example, it is not uncommon for the colors of aerial
photographs and corresponding wetland signatures to vary from one set of images to another due to
differences in image processing.
While interpretation of aerial photography has many limitations, many of those limitations and
others are also present when analyzing satellite images. For example, one challenge with aerial
photography is that spectral signatures of neighboring plant species may have similar reflectance
values. Even when using hyperspectral images, it may be impossible to separate some distinct veg-
etation types, such as distinguishing low salt marsh (S. alterniflora) from short form of Phragmites
australis and S. patens in the New Jersey Meadowlands in June (Yang and Artigas 2010). Using
SPOT imagery (20 m resolution) to map upland bogs in New Zealand, Arbuckle et al. (1998) experi-
enced difficulty separating bogs and some tussock grasslands due to similar spectral signatures. The
problem was resolved by consulting aerial imagery and reclassifying features as necessary. Scale is
an issue with any mapping project as it determines the targeted map unit and the degree of mixing
of classes (e.g., Tiner 1999, 1997; Muster et al. 2013).
Timing of imagery is also important for wetland detection on satellite images. For example,
varying water conditions (e.g., tide stage) in the coastal wetland ecosystem generate complex spec-
tral signatures of vegetation types that make wetland classification challenging for both photoin-
terpretation and satellite image analysis, requiring an innovative approach to improve classification
(Zhang et al. 2011; Zhang and Xie 2012).
Satellite image analysis is a largely computer-intensive procedure with automated classification
routines compared to manual interpretation of aerial photographs (Chapter 5 and other chapters in
this book). Its upfront costs are higher, requiring expensive software, substantial computer process-
ing power, and large amounts of available memory (storage) as well as a higher learning curve for
analysts. Analysts using either method must have a significant knowledge base of wetlands and their
signatures on imagery. In the long run, however, satellite image analysis may offer considerable
time saving over manual interpretation of aerial photographs and can be updated more frequently,
although the level of wetland classification produced may need to be adjusted. The availability
of multitemporal and multispectra imagery help improve classification, yet georeferencing images
from different sensors and dates presents a constant challenge for data fusion efforts.
The north–south path of finer-spatial-resolution satellites may pose problems for synchronized
imagery for geographic features oriented east to west. For full coverage of Long Island, New York,
QuickBird high-resolution imagery required numerous days with varying environmental conditions
to collect data producing difference reflectance values for similar features requiring separate analy-
sis of each image (Wang et al. 2010). While acquisition of aerial photography involves considerable
advanced planning and may also result in capturing images during different environmental condi-
tions (e.g., before and after leaf out), the manual interpretation process can often compensate for
these differences.
All sensors face the challenges posed by wetlands and the environment in general for accurate
detection and classification (see Chapter 5 for more detailed discussions of the technical issues faced
when attempting to map wetlands via remote sensing technologies). For the best results, acquisition
of the remotely sensed data must be timed to capture wetlands in a condition that will facilitate
their identification or collected at multiple times to cover a range of conditions that will also aid
the image analysts. Classification systems may have to be simplified to match the capability of the
sensor to accurately and consistently categorize wetlands. Furthermore, the production of maps that
show wetland likelihood (higher probability of being wetland) rather than maps that are intended to
show the actual (exact) limits of specific wetland types may be a better way to go. For example, the
U.S. FWS has, in some areas, produced NWI+ data that show “P-wet areas”—areas that may sup-
port some wetland based on soil mapping (Figure 3.15). This product however is a derivative map
generated from selecting hydric soil map units from USDA soil surveys and then reviewing aerial
imagery and selecting only those units or portions of units that are undeveloped.
FIGURE 3.15 Example of P-wet areas and mapped wetlands for Ellendale, Delaware, from the NWI+ web
mapper. Red areas, P-wet areas; purple, scrub–shrub wetlands; brown, deciduous forested wetlands; green,
evergreen forested wetlands; and blue, ponds. (From Association of State Wetland Managers; http://nwiplus.
cmi.vt.edu/nwiplusmapper/.)
ACKNOWLEDGMENTS
Special thanks to the following individuals for offering their experiences from mapping wetlands
in the United States for inclusion in this chapter: John Anderson, Kevin Bon, Julie Michaelson,
and Jerry Tande. Nate Herold (NOAA) graciously provided an example of their wetland mapping
for inclusion in this chapter. Thanks also to Greg Koeln and Mike O’Brien (MDA Information
Systems, Inc.) for providing the satellite image of the Georgia coast and to Matthew Hatvany (Laval
University) for the historic map of Prince Edward Island. Review of draft manuscripts by Megan
Lang and Vic Klemas helped improve the quality of this chapter.
REFERENCES
Akumu, C.E., S. Pathirana, S. Baban, and D. Bucher. 2010. Modeling methane emission from wetlands in
North-Eastern New South Wales, Australia using Landsat ETM+. Remote Sensing 2(5):1378–1399.
Arbuckle, C.J., A.D. Huryn, and S.A. Israel. 1998. Applications of remote sensing and GIS to wetland inventory:
Upland bogs. Proceedings of the Spatial Information Research Centre’s 10th Colloquium, University of
Otago, New Zealand, November 16–19, 1998, pp. 15–24.
Baker, C., R. Lawrence, C. Montagne, and D. Patten. 2007. Change detection of wetland ecosystems using
Landsat imagery and change vector analysis. Wetlands 27(3):610–619.
Bergeson, M.T. 2014. U.S. Fish and Wildlife Service completes digital wetland coverage for 48 states, Hawaii,
trust territories and 35% of Alaska. Wetland Science and Practice 31(2):20–21.
Bwangoy, J.-R.B., M.C. Hansen, D.P. Roy, G. De Grandi, and C.O. Justice. 2010. Wetland mapping in the
Congo Basin using optical and radar remotely sensed data and derived topographical indices. Remote
Sensing of Environment 114:73–86.
Choung, S.H. and J.J. Ulliman. 1992. A comparative study of wetland change detection techniques using post-
classification comparison and image differencing on Landsat-5 TM data. Journal of Korean Forestry
Society 81:346–356.
Connolly, J., N.M. Holden, and S.M. Ward. 2007. Mapping peatlands in Ireland using rule-based methodology
and digital data. Soil Science Society of America Journal 71(2):492–499.
Davidson, N.C. and C.M. Finlayson. 2007. Earth observation for wetland inventory, assessment and monitor-
ing. Aquatic Conservation: Marine and Freshwater Ecosystems 17:219–228.
De Roeck, E.R., N.E.C. Verhoest, M.H. Miya, H. Lievens, O. Batelaan, A. Thomas, and L. Brendonck. 2008.
Remote sensing and wetland ecology: A South African case study. Sensors 8:3542–3556.
Elmahdy, S.I. and M.M. Mohamed. 2013. Change detection and mapping of mangrove using multi-temporal
remote sensing data: A case study of Abu Dhabi, UAE. Journal of Geomatics 7(1):41–46.
Evans, T.L. and M. Costa. 2013. Landcover classification of the Lower Nhecolândia subregion of the Brazilian
Pantanal wetlands using ALOS/PALSAR, RADARSAT-2 and ENVISAT/ASAR imagery. Remote
Sensing of Environment 128:118–137.
Gao, Z. and L. Xing. 2011. A study on dynamic change features of wetlands in Dongying City based on RS.
Procedia Environmental Sciences 10:2141–2146.
Ghobadi, Y., B. Pradham, K. Kabiri, S. Pirasteh, H.Z.M. Shafri, and G.A. Sayyad. 2012. Use of multi-temporal
remote sensing data and GIS for wetland change monitoring and degradation. 2012 IEEE Colloquium
on Humanities, Science and Engineering Research (CHUSER 2012), Kota Kinabalu, Sabah, Malaysia,
December 3–4, 2012, pp. 97–102.
Green, E.P. and F.T. Short (eds.). 2003. World Atlas of Seagrasses. University of California Press,
Berkeley, CA.
Harley, J.B. and D. Woodard (eds.). 1987. The History of Cartography. Vol. 1. The University of Chicago Press,
Chicago, IL.
Hatvany, M.G. 2001. Wedded to the marshes: Salt marshes and socio-economic differentiation in early Prince
Edward Island. Acadiensis 30(2):40–55.
Heumann, B.W. 2011. Satellite remote sensing of mangrove forests: Recent advances and future opportunities.
Progress in Physical Geography 35:87–108.
Hoekman, D. 2009. Monitoring tropical peat swamp deforestation and hydrological dynamics by ASAR
and PALSAR. Chapter 13. In: Pei-Gee Peter Ho (ed.). Geoscience and Remote Sensing. InTech.
pp. 257–275. http://edepot.wur.nl/242290. Accessed November 8, 2014.
Huang, C., Y. Peng, M. Lang, I.-Y. Yeo, and G. McCarty. 2014. Wetland inundation mapping and change
monitoring using Landsat and airborne LiDAR data. Remote Sensing of Environment 141:231–242.
Islam, Md.A., P.S. Thenkabail, R.W. Kulawardhana, R. Alankara, S. Gunasinghe, C. Edussriya, and
A. Gunawardana, 2008. Semi-automated methods for mapping wetlands using Landsat ETM+ and
SRTM data. International Journal of Remote Sensing 29:7077–7106.
Jensen, J.R. 2007. Remote Sensing of the Environment: An Earth Resource Perspective. Prentice Hall, Upper
Saddle River, NJ.
Jensen, J.R., E.W. Ramsat, H.E. Mackey, E.J. Christensen, and R.P. Sharitz. 1987. Inland wetland change
detection using aircraft MSS data. Photogrammetric Engineering and Remote Sensing 53:521–529.
Kashaigili, J.J., B.P. Mbilinyi, M. McCartney, and F.L. Mwanuzi. 2006. Dynamics of Usangu plains wetlands:
Use of remote sensing and GIS as management decision tools. Physics and Chemistry of the Earth (Parts
A/B/C) 31(15–16):967–975.
Klemas, V. 2011. Remote sensing of wetlands: Case studies comparing practical techniques. Journal of Coastal
Research 27(3):418–427.
Klemas, V. 2013. Remote sensing of coastal wetland biomass: An overview. Journal of Coastal Research
29(5):1016–1028.
Krankina, O.N., D. Pflugmacher, M. Friedl, W.B. Cohen, P. Nelson, and A. Baccini. 2008. Meeting the chal-
lenge of mapping peatlands with remotely sensed data. Biogeosciences 5:1809–1820.
Lampman, J.L. 1993. Bibliography of Remote Sensing Techniques Used in Wetland Research. U.S. Army Corps
of Engineers, Waterways Experiment Station, Vicksburg, MS. Wetlands Research Program Technical
Report WRP-SM-2.
Lee, T.-M. and H.-C. Yeh. 2009. Applying remote sensing techniques to monitor shifting wetland veg-
etation: A case study of Danshui River estuary mangrove communities. Ecological Engineering
35:487–496.
Li, A., W. Deng, B. Kong, M. Song, W. Feng, X. Lu, G. Lei, and J. Bai. 2010. A comparative analysis on spatial
patterns and processes of three typical wetland ecosystems in 3H area, China. Proceida Environmental
Sciences 2:315–332.
Mahmud, M.S., U. Habiba, F. Haider, and A. Masrur. 2011. Remote sensing and GIS based spatio-temporal
change analysis of wetland in Dhaka City, Bangladesh. Journal of Water Resource and Protection
3:781–787.
Mamoun, C.M., R. Nigel, and S.D.D.V. Rughoopath. 2013. Wetlands inventory, mapping and land cover index
assessment on Mauritius. Wetlands 33:585–595.
Martel Laboratories, Inc. 1976. Existing State and Local Wetlands Surveys (1965–1975). Vol. II. Narrative.
Prepared for the U.S. Fish and Wildlife Service, Office of Biological Services, Washington, DC.
Martínez-López, J., M.F. Carreño, J.A. Palazón-Ferrando, J. Martínez-Fernández, and M.A. Esteve. 2014.
Remote sensing of plant communities as a tool for assessing the condition of semiarid Mediterranean
saline wetlands in agricultural catchments. International Journal of Applied Earth Observation and
Geoinformation 26:193–204.
Mayer, A.L. and R.D. Lopez. 2011. Use of remote sensing to support forest and wetlands policies in the USA.
Remote Sensing 3:121–123.
McCarthy, J.M., T. Gumbricht, T. McCarthy, P. Frost, K. Wessels, and F. Seidel. 2003. Flooding patterns of the
Okavango wetland in Botswana between 1972 and 2000. AMBIO 32(7):453–457.
McHugh, O.V., A.N. McHugh, P.M. Eloundou-Enyegue, and T.S. Steenhuis. 2007. Integrated qualitative
assessment of wetland hydrological and land cover changes in a data scarce dry Ethiopian highland
watershed. Land Degradation and Development 18:643–658.
Melack, J.M. and L.L. Hess. 2011. Remote sensing of the distribution and extent of wetlands in the Amazon
Basin. Chapter 3. In: W.J. Junk, M.T.F. Piedade, F. Wittmann, J. Schöngart, and P. Parolin (eds.).
Amazon Floodplain Forests: Ecophysiology, Biodiversity and Sustainable Management. Springer
Science+Business Media, B.V., Dordrecht, the Netherlands.
Munyati, C. 2000. Wetland change detection on the Kafue Flats, Zambia, by classification of a multi-temporal
remote sensing image dataset. International Journal of Remote Sensing 21:1787–1806.
Murphy, P.N.C., J. Ogilvie, K. Connor, and P.A. Arp. 2007. Mapping wetlands: A comparison of two different
approaches for New Brunswick, Canada. Wetlands 27:846–854.
Muster, S., B. Heim, A. Abnizova, and J. Boike. 2013. Water body distributions across scales: A remote sensing
based comparison of three Arctic tundra wetlands. Remote Sensing 5(4):1498–1523.
Mwita, E., G. Menz, S. Misana, M. Becker, D. Kisanga, and B. Boehme. 2013. Mapping small wetlands in
Kenya and Tanzania using remote sensing techniques. International Journal of Applied Earth Observation
and Geoinformation 21:173–183.
NASA. 2014. An inundated wetlands earth system data record: Global monitoring of wetland extent and
dynamics. Accessed on May 24, 2014, https://earthdata.nasa.gov/our-community/community-data-
system-programs/measures-projects/global-monitoring-wetlands.
Ogilvie, J., C. Mouland, M. Castonguay, G. Moran, and P.A. Arp. 2013. Introduction to wet-areas map-
ping: Creating a new and innovative base layer for rural, municipal and urban planning. Accessed on
December 6, 2013 from https://www.academia.edu/1348010/Introduction_to_wet-areas_mapping_
creating_a_new_and_innovative_base_layer_for_rural_municipal_and_urban_planning.
Ozesmi, S.L. and M.E. Bauer. 2002. Satellite remote sensing of wetlands. Wetlands Ecology and Management
10:381–402.
Paine, J.G., J.R. Andrews, K. Saylam, T.A. Tremblay, A.R. Averett, T.L. Caudle, T. Meyer, and M.H. Young.
2013. Airborne LiDAR on the Alaskan North Slope: Wetlands mapping, lake volumes, and permafrost
features. The Leading Edge (July 2013):798–805.
Paula, F.H. and K.M. William. 2000. Detecting wetland change: A rule-based approach using NWI and
SPOT-XS data. Photogrammetric Engineering and Remote Sensing 66:205–211.
Ramsar Convention Secretariat. 2010. Wetland Inventory: A Ramsar Framework for Wetland Inventory and
Ecological Character Description. Ramsar Handbooks for the Wise Use of Wetlands. Handbook 15.
Ramsar Convention Secretariat, Gland, Switzerland. http://www.ramsar.org/sites/default/files/documents/
pdf/lib/hbk4-15.pdf. Accessed November 8, 2014.
Rebelo, L.-M., C.M. Finlayson, and N. Nagabhatia. 2009. Remote sensing and GIS for wetland inventory,
mapping and change analysis. Journal of Environmental Management 90(7):2144–2153.
Reif, M., R.C. Frohn, C.R. Lane, and B. Autrey. 2009. Mapping isolated wetlands in a karst landscape: GIS and
remote sensing methods. GIScience and Remote Sensing 46:1548−1603.
Rosso, P.H., S.L. Ustin, and A. Hastings. 2005. Mapping marshland vegetation of San Francisco Bay, California,
using hyperspectral data. International Journal of Remote Sensing 26(23):5169–5191.
Saalovara, K., S. Thessler, R.N. Malik, and H. Tuomisto. 2005. Classification of Amazonian primary rain forest
vegetation using Landsat ETM+ satellite imagery. Remote Sensing of Environment 97:39−51.
Saleska, S.R., K. Didan, A.R. Huete, and H.R. da Rocha. 2007. Amazon forests green-up during 2005 drought.
Science 318(5850):318−612.
Shaw, S.P. and C.G. Fredine. 1956. Wetlands of the United States—Their Extent and Their Value to Waterfowl
and Other Wildlife. U.S. Department of the Interior, Washington, DC, circular 39.
Sitch, S., A.D. McGuire, J. Kimball, N. Gedney, J. Gamon, R. Engstrom, A. Wolf, Q. Zhuang, J.S. Clein, and
K.C. McDonald. 2007. Assessing the carbon balance of circumpolar Arctic tundra using remote sensing
and process modeling. Ecological Applications 17:213–234.
Spalding, M.D., F. Blasco, and C.D. Field. 1997. World Mangrove Atlas. International Society for Mangrove
Ecosystems, Okinawa, Japan.
Spalding, M.D., E.P. Green, and C. Ravilious. 2001. World Atlas of Coral Reefs. University of California Press,
Berkeley, CA.
Spalding, M.D., M. Kainuma, and L. Collins. 2010. World Atlas of Mangroves. Earthscan, London, U.K.
Sugumaran, R., J. Harken, and J. Gerjevic. 2004. Using Remote Sensing Data to Study Wetland Dynamics in
Iowa. University of Northern Iowa, Cedar Rapids, IA, Iowa Space Grant (Seed) Technical Report.
Tarnocai, C., I.M. Kettles, and B. Lacelle. 2002. Peatlands of Canada Database. Natural Resources Canada,
Ottawa, Ontario, Canada. Geological Survey of Canada Open File 4002.
Tiner, R.W. 1991. The concept of a hydrophyte for wetland identification. BioScience 41(4):236–247.
Tiner, R.W. 1997. Wetlands. Chapter 13. In: W.R. Philipson (ed.). Manual of Photographic Interpretation. 2nd
edn. American Society for Photogrammetry and Remote Sensing, Bethesda, MD, pp. 475–494.
Mapping. CRC Press, Boca Raton, FL.
Tiner, R.W. (ed.). 2009. Status Report for the National Wetlands Inventory Program: 2009. U.S. Fish and
Wildlife Service, Division of Habitat and Resource Conservation, Branch of Resource and Mapping
Support, Arlington, VA.
Tiner, R.W. 2013. Tidal Wetlands Primer: An Introduction to their Ecology, Natural History, Status, and
U.S. Fish and Wildlife Service. 1959. Wetlands Inventory of Connecticut. Region V, Office of River Basin
Studies, Boston, MA.
U.S. Fish and Wildlife Service. 1965. Supplementary Report on the Coastal Wetlands Inventory of Connecticut.
Region V, Division of River Basin Studies, Boston, MA, June 1965.
Usery, E.L., D. Varanka, and M.P. Finn. 2009. A 125 year history of topographic mapping and GIS in the U.S.
Geological Survey 1884 to 2009, part 1. U.S. Geological Survey, Center of Excellence for Geospatial
Information Science, Rolla, MO. http://nationalmap.gov/ustopo/125history.html. Accessed November 8,
2014.
Vileisis, A. 1997. Discovering the Unknown Landscape: A History of America’s Wetlands. Island Press,
Washington, DC.
Wang, Y. (ed.). 2010. Remote Sensing of Coastal Environments. CRC Press, Boca Raton, FL.
Wang, Y., M. Christiano, and M. Traber. 2010. Mapping salt marshes in Jamaica Bay and terrestrial vegetation
in Fire Island National Seashore using QuickBird satellite data. In: Y. Wang (ed.). Remote Sensing of
Coastal Environments. CRC Press, Boca Raton, FL, pp. 191–208.
Warming, E. 1909. Oecology of Plants: An Introduction to the Study of Plant-Communities. (Updated English
version of a 1896 text.) Clarendon Press, Oxford, U.K.
Weaver, J.E. and F.E. Clements. 1929. Plant Ecology. McGraw-Hill, New York.
Yang, J. and F.J. Artigas. 2010. Mapping salt marsh vegetation by integrating hyperspectral and LiDAR remote
sensing. Chapter 9. In: Y. Wang (ed.). Remote Sensing of Coastal Environments. CRC Press, Boca Raton,
FL, pp. 173–187.
Zhang, C. and Z. Xie. 2012. Combining object-based texture measures with a neural network for vegetation
mapping in the Everglades from hyperspectral imagery. Remote Sensing of Environment 124:310–320.
Zhang, Y., D. Lu, B. Yang, C. Sun, and M. Sun. 2011. Coastal wetland vegetation classification with a Landsat
Thematic Mapper image. International Journal of Remote Sensing 32:545–561.
Zhao, H., B. Cui, H. Zhang, X. Fan, Z. Zhang, and X. Lei. 2010. A landscape approach for wetland change
detection (1979–2009) in the Pearl River Estuary. Procedia Environmental Science 2:1265–1278.
Zomer, R.J., A. Trabucco, and S.L. Ustin. 2009. Building spectral libraries for wetlands land cover classification
and hyperspectral remote sensing. Journal of Environmental Management 90:2170–2177.
4 Early Applications of Remote
Sensing for Mapping Wetlands
Ralph W. Tiner
CONTENTS
Introduction....................................................................................................................................... 67
Aerial Photography........................................................................................................................... 67
Satellite Imagery............................................................................................................................... 72
References......................................................................................................................................... 75
INTRODUCTION
Earth-based remote sensing involves the collection of information regarding land, water, and atmo-
spheric properties from above the Earth’s surface with cameras and other recording instruments.
Airborne and satellite imagery are major data sources that are used to classify the Earth’s land
and water resources. When mapping wetland extent, the purpose of using these data is to identify,
delineate, classify, and display the boundaries of wetlands and calculate their distribution across
the landscape. Aerial photography has its origins in the late 1800s when pictures of Petit Bicêtre,
France, and Boston, Massachusetts, were taken from hot air balloons (Colwell 1997). Images were
first captured by cameras on black-and-white (BW) film, then on true color or color infrared (CIR)
film, and later from satellites or aircraft armed with sensors that detect electromagnetic wavelengths
beyond the visible and infrared spectra (e.g., thermal) or sensors that transmit and detect their own
energy (e.g., radar and LiDAR). Since the rest of the book addresses recent applications of remote
sensing using the latter sensors, this chapter will briefly discuss the early uses of remote sensing for
mapping wetlands generally prior to the 1990s. Although current applications of remote sensing for
wetland mapping in the United States were generally described in Chapter 3 when discussing the
need for wetland mapping, an overview of historic uses of remote sensing for such purposes pro-
vides background for readers that are new to this topic before delving into more advanced applica-
tions of the technology. This chapter therefore presents an introduction to the uses of remote sensing
for wetland detection prior to the 1990s and as such is intentionally brief with numerous references
cited for further examination. Most early wetland mapping was conducted with aerial photos and
less with satellite-derived imagery, hence the focus for this brief review.
AERIAL PHOTOGRAPHY
Three types of aerial photography have been widely used for wetland mapping: (1) BW (panchromatic),
(2) true color, and (3) CIR. Stereoscopic viewing of pairs of aerial photos provides a 3D view that allows
depressions to be readily seen, especially in areas of moderate or high topographic relief (Figures 4.1
through 4.3). BW aerial photos captured by aircraft have been readily available for landscape and
land use mapping since the 1930s (Lee 1926). The U.S. government and state agencies often worked
together to acquire such imagery. For a review of the history of aerial photography and photointer-
pretation, see Colwell (1997). See Paine and Kiser (2012) for general information on the use of aerial
photography and Tiner (1997) for a description of the utility of aerial photography for wetland mapping.
67
FIGURE 4.1 Stereogram of CIR, leaf-off aerial photo showing flooded depressional (basin) wetlands (dark
gray to black) and forested wetlands along stream (dark gray). Water is black. To see area in 3D, use a pocket
stereoscope. (Images from State of Massachusetts.)
FIGURE 4.2 Stereogram of CIR, leaf-off aerial photo showing bog (orange-brown area), marsh (whitish
area), deciduous shrub swamps (light gray), deciduous forested wetland (darker gray), and mixed evergreen/
deciduous forested wetland (dark gray with pinkish purple clusters). (Images from State of Massachusetts.)
Early Applications of Remote Sensing for Mapping Wetlands 69
FIGURE 4.3 Stereogram of CIR, leaf-off aerial photo showing floodplain wetlands (semipermanently
flooded and seasonally flooded types). (Images from State of Massachusetts.)
The acquisition of aerial photographs fundamentally changed the practice of mapping from
one based on land surveys to one relying more on interpretation of aerial photographs. The mili-
tary probably produced the first air photo-derived maps showing wetlands as they posed obstacles
for movements of troops and vehicles during World War I. After the war, aerial photography
became available for other uses and was interpreted for forestry management, soil mapping, wild-
life surveys, and evaluating land use changes (Colwell 1997). The first inventory of U.S. wetlands
important to waterfowl was conducted in the 1950s by the U.S. Fish and Wildlife Service in
cooperation with state wildlife departments. This survey used a combination of interpretation
of the U.S. Department of Agriculture BW aerial photos (Figure 4.4); analysis of topographic,
coastal geodetic, and other maps; and field investigations to identify 90% of the wetlands “used
significantly” by waterfowl across the country (e.g., Spinner 1955; Shaw and Fredine 1956; see
Chapter 3 for sample map).
In the 1960s, researchers began using CIR and true color photos to map wetlands and sub-
merged aquatic vegetation (SAV), respectively (e.g., Olson 1964; Anderson 1968; Stroud and Cooper
1968; Anderson and Wobber 1972; Carter and Anderson 1972; Reimold et al. 1973; Seher and
Tueller 1973; Cowardin and Myers 1974; McEwen et al. 1976; Shima et al. 1976; Carter et al. 1977,
1979; Austin and Adams 1978; Gammon and Carter 1979; Stewart et al. 1980; Mead and Gammon
1981; Hathout and Simpson 1982; see Orth and Moore 1983 for early SAV studies). Roller (1977)
described the early use of remote sensing for mapping wetlands. Lee (1991) documented the use of
remote sensing for wetland detection, while Lampman (1993) provided a detailed review of early
remote sensing applications for wetland research. Readers should consult these sources and the
references in this chapter for more in-depth discussion.
The use of aerial photos for operational wetland mapping has been widespread in the United
States for over 60 years (e.g., Tiner 1997). Several states have used air photointerpretation to map
wetlands. Klemas et al. (1973) produced the first maps of Delaware’s coastal wetlands through
manual interpretation of 1:60,000 CIR photos and automated analysis of 1:120,000 CIR images.
Ground truthing was performed on foot or by boat in the summer of 1972, while other sources of
collateral data included large-scale aerial photos captured from small planes to aid identification
10-22-57 AHU-6T
FIGURE 4.4 Portion of 1957 large-scale BW aerial photograph showing freshwater tidal wetlands, uplands,
and the Mattawoman Creek (Charles County, MD). (Images from U.S. Fish and Wildlife Service.)
of wetland plant communities for inaccessible sites and to verify interpretations for large areas and
1:20,000 BW aerial photos from the Delaware Department of Agriculture for mapping ditched
wetlands. This work also included a large-scale attempt to classify imagery through an automated
routine using the General Electric Multispectral Data Processing System (GEMSDPS; Figure 4.5).
Hardisky and Klemas (1983) produced another set of tidal wetland maps from 1:12,000 CIR photo-
graphs; these maps were adopted by the state as regulatory maps showing the official boundaries of
wetlands. Wetlands in Massachusetts and Rhode Island were mapped using large-scale BW photos
(1:20,000 and 1:12,000) as part of statewide land cover/land use mapping (e.g., MacConnell 1974,
1975). Connecticut produced tidal wetland maps through ground surveys but recorded the boundary
points on large-scale aerial photographs. The State of South Carolina inventoried its extensive tidal
marshes by interpreting large-scale BW photographs supplemented by CIR photos (Tiner 1977),
while New York, New Jersey, and Maryland used 1:12,000 CIR photography to map their tidal wet-
lands (Brown 1978; McCormick and Somes 1982). New York State surveyed its freshwater wetlands
with 1:24,000 BW photos (Cole and Fried 1981). Hawaii used a variety of aerial photos (scales and
emulsion) to map its wetlands (Chime et al. 1978).
FIGURE 4.5 Computer-generated map showing different plant communities in Delaware tidal wetlands.
Water and mud, black; Spartina patens and Distichlis spicata, blue; Spartina alterniflora and Spartina
cynosuroides, yellow. (Map from Klemas, V. et al., Coastal Vegetation of Delaware. The Mapping of
Delaware’s Coastal Marshes, University of Delaware, College of Marine Studies, Newark, DE, 1973, assisted
by University of Michigan.)
Prior to initiating the National Wetlands Inventory (NWI) Program, the U.S. Fish and Wildlife
Service commissioned an exhaustive survey of state and local wetland inventories (Martel
Laboratories 1976). From the mid-1970s to the mid-1990s, the NWI relied on stereoscopic inter-
pretation of a variety of aerial photography (scales and emulsions) to map wetlands (see Tiner 2009
for review of techniques; see Chapter 3 for examples of maps). Today, NWI’s approach employs
monoscopic on-screen interpretation of digital imagery and works at scales of 5,000–10,000 for
operational mapping with the ability to zoom in even further as necessary.
Analysis of historical aerial photos is required to document wetland trends (e.g., Frayer et al.
1983; Hardisky and Klemas 1983; Niedzwiedz and Batie 1984; Figure 4.6) and changes in sea grass
beds (e.g., Orth and Moore 1983; Ball et al. 2009) prior to the 1970s. Such analyses are vital to
documenting violations (i.e., unauthorized wetland alterations) of the U.S. Clean Water Act for court
cases. Today, both aerial photography and satellite imagery are available for historic analyses and to
use multitemporal images to evaluate changes in wetland hydrology.
1952 1971
1991
FIGURE 4.6 Time series of aerial photos showing wetland changes from 1952 to 1971 to 1991. (Note: The
1991 images are a stereogram and can be viewed in 3D with a pocket stereoscope.) (Images from State of
Massachusetts.)
SATELLITE IMAGERY
While Russia launched the first satellite—Sputnik—in 1957, it was not until the 1970s that satel-
lites specifically designed to monitor the Earth’s surface were launched. The launch of the United
States’ first Earth resources satellite (ERTS 1; now called the first Landsat satellite—Landsat 1)
on July 23, 1972, offered a new technology for remote sensing of wetlands. Since then, seven
more Landsat satellites have been launched (e.g., Landsat 8 on February 11, 2013) by NASA,
while other countries have launched satellites providing multitemporal and multispectral data for
remote sensing specialists that can be utilized for wetland mapping (e.g., France, SPOT; India,
IRS; Israel, EROS; Japan, JERS; Russia, Meteor-Priroda and Resurs). Some books that provide
reviews of the use of satellite imagery for land use and vegetation mapping are listed in this
book’s preface (e.g., Lillesand et al. 2008).
The Earth Resources Technology Satellite (ERTS) or Landsat 1 collected the first satellite images
that were widely used for wetland mapping (Figure 4.7). In the early 1970s, researchers were evalu-
ating the utility of ERTS 1 multispectral scanner data to map coastal wetlands in the United States.
(Figure 4.8; Anderson et al. 1973; Carter et al. 1974; see Bartlett and Klemas 1980 for detailed sum-
mary). This sensor had a spatial resolution of 80 m, which limited its utility for wetland mapping
relative to aerial photointerpretation (e.g., Jensen et al. 1980). Increased spatial resolution of later
generations of Landsat helped improved its use for mapping wetlands (Dottavio and Dottavio 1984).
FIGURE 4.7 Landsat 1 image (October 12, 1972) of Charleston, South Carolina and vicinity. This image
was used by Carter et al. (1976) for one of the earliest applications of satellite imagery for mapping of wet-
lands. (Image from U.S. Geological Survey.)
Table 4.1 lists some early applications of satellite imagery for wetland mapping. A number of these
studies compared satellite results with those derived from aerial photointerpretation (e.g., Jensen
et al. 1980; Ernst-Dottavio et al. 1981; Wood 1983; Hardin 1985; May 1986; Rutchey and Vilcheck
1994). Ozesmi and Bauer (2002) offer a must-read review of the use of satellite imagery for wetland
mapping in the first 30 years that satellite data were available.
In the United States, the Wetlands Subcommittee of the Federal Geographic Data Committee
(1992) prepared a review of applications of satellite data for mapping and monitoring wetlands.
Their conclusion was that “Synergistic effects created by combining both satellite data and NWI
digital data have greater value than using either data source alone. Such data sets have the potential
to be synoptic and accurate.” Unfortunately, this type of data fusion has not been integrated into
standard NWI mapping techniques due to the program’s continued focus on highly detailed classi-
fication (Wilen and Bates 1995; Wilen and Smith 1996) and possibly also due to funding reductions
and to the costs of adding this technology to the current mapping procedure.
Since the 1990s much progress has been made on the use of remote sensing for wetland map-
ping around the world. The rest of this book provides an overview of these advances and numerous
examples of applications.
FIGURE 4.8 A computer-generated map showing wetlands in Chincoteague, VA (legend deleted). This
may be the earliest test of using automated techniques to produce a wetland map from satellite data. Water,
blue; organic mud flat, orange; S. alterniflora, green; S. patens, yellow; spoil, purple; upland, red. (Map from
Carter, V., Computer mapping of coastal wetlands, in: Williams, R.S., Jr. and Carter, W.D. (eds.), ERTS-1 A
New Window on Our Planet, U.S. Government Printing Office, Washington, DC, U.S. Geological Survey
Professional Paper 929, 1976, pp. 280–282.)
TABLE 4.1
Examples of the Use of Satellite Imagery for Wetland Identification from 1972 to 2004
Wetland Type Location References
Tidal marshes Delaware Hardin (1985) and Hardisky et al. (1986)
Southeast USA Anderson et al. (1973)
Coastal Louisiana May (1986) and Pickus (1990)
Mangroves Florida Patterson (1986) and Kushwaha et al. (2000)
Forested Florida Kasischke and Bourgeau-Chavez (1997)
Maine Sader et al. (1995)
North Carolina Townsend and Walsh (1998)
Virginia/North Carolina Carter (1976)
Ponds and potholes North Dakota Gilmer et al. (1976, 1980) and Work and Gilmer (1976)
Wet meadows Montana Kinderscher et al. (1998)
Playas Iran Krinsley (1976)
Kelp beds California Jensen et al. (1980)
Everglades Florida Rutchey and Vilcheck (1994) and Jensen et al. (1995)
Inland freshwater Alaska Jacobson et al. (1987)
Australia Johnston and Barson (1993)
Delaware Hardin (1985)
Florida Hodgson et al. (1988)
Georgia/South Carolina Jensen et al. (1986)
Indiana Ernst-Dottavio et al. (1981)
Louisiana May (1986) and Pickus (1990)
Montana Wright (2004)
Ontario Gluck et al. (1996) and Wang et al. (1998)
Saskatchewan Dechka et al. (2002)
Tennessee Jones and Shahrokhi (1977)
Wisconsin Bolstad and Lillesand (1992)
REFERENCES
Anderson, R.R. 1968. Remote sensing of marshlands and estuaries using color infrared photography. NASA
Earth Resources Aircraft Program Status Review III:26-1–26-23.
Anderson, R.R., V. Carter, and J. McGuinness. March 26–28, 1973. Mapping southern Atlantic coastal
marshlands, South Carolina-Georgia, using ERTS-1 imagery. In: Remote Sensing of Earth Resources;
Proceedings of the Second Conference on Earth Resources Observation and Information Analysis System
(Tullahoma, TN). Vol. 2. (A74-25386 10–13). pp. 1021–1028.
Anderson, R.R. and F.J. Wobber. 1972. Wetlands mapping in New Jersey. Photogrammetric Engineering
39:353–358.
Austin, A. and R. Adams. 1978. Aerial color and color infra-red survey of marine plant resources. Photo
grammetric Engineering 44:469–480.
Ball, D., M. Soto-Berelov, P. Young, and A. Coots. 2009. Baywide Seagrass Monitoring Program—Historical
Seagrass Mapping. Department of Primary Industries, Queenscliff, Victoria, Australia. Fisheries Victoria
Technical Report Series No. 70.
Bartlett, D.S. and V. Klemas. 1980. Quantitative assessment of tidal wetlands using remote sensing.
Environmental Management 4:337–345.
Bolstad, P.V. and T.M. Lillesand. 1992. Rule-based classification models: Flexible integration of satellite imag-
ery and thematic spatial data. Photogrammetric Engineering and Remote Sensing 58:965–971.
Brown, W.W. 1978. Wetland mapping in New Jersey and New York. Photogrammetric Engineering and Remote
Sensing 44:303–314.
Carter, V. 1976. The great dismal swamp of Virginia and North Carolina. In: R.S. Williams, Jr. and W.D. Carter
(eds.). ERTS-1 A New Window on Our Planet (U.S. Government Printing Office, Washington, DC). U.S.
Geological Survey Professional Paper 929. pp. 316–320.
Carter, V. and R.R. Anderson. 1972. Interpretation of wetlands imagery based on spectral reflectance character-
istics of selected plant species. Proceedings of the American Society of Photogrammetry. pp. 580–595.
Carter, V., M.K. Garrett, L. Shima, and P. Gammon. 1977. The Great Dismal Swamp: Management of a hydro-
logic resource with the aid of remote sensing. Water Resources Bulletin 13:1–12.
Carter, V., D.L. Malone, and J.H. Burbank. 1979. Wetland classification and mapping in western Tennessee.
Photogrammetric Engineering and Remote Sensing 45:273–284.
Carter, V., J. McGuinness, and R.R. Anderson. March 26–28, 1974. Mapping northern Atlantic coastal marsh-
lands, using ERTS-1 imagery. In: Remote Sensing of Earth Resources; Proceedings of the Second
Conference on Earth Resources Observation and Information Analysis System (Tullahoma, TN). Vol. 2.
(A74-25386 10-13). pp. 1012–1020.
Chime, L.R., G.E. Gnauck, and J. Maragos. March 14–16, 1978. Hawaii wetlands mapping. In: Coastal
Zone’78 (San Francisco, CA). American Society of Civil Engineers, New York. pp. 1223–1241.
Cole, N.B. and E. Fried. 1981. Technical Manual: Freshwater Wetlands Inventory. New York Department of
Environmental Conservation, Division of Fish and Wildlife, Albany, NY.
Colwell, R.N. 1997. History and place of photographic interpretation. Chapter 1. In: W.R. Philipson (ed.).
Manual of Photographic Interpretation (American Society for Photogrammetry and Remote Sensing,
Bethesda, MD). pp. 3–47.
Cowardin, L.M. and V.I. Myers. 1974. Remote sensing for identification and classification of wetland vegeta-
tion. Journal of Wildlife Management 38:308–314.
Dechka, J.A., S.E. Franklin, M.D. Watmough, R.P. Bennett, and D.W. Ingstrup. 2002. Classification of wetland
habitat and vegetation communities using multi-temporal Ikonos imagery in southern Saskatchewan.
Canadian Journal of Remote Sensing 28:679–685.
Dottavio, C.L. and F.D. Dottavio. 1984. Potential benefits of new satellite sensors to wetland mapping.
Ernst-Dottavio, C.L., R.M. Hoffer, and R.P. Mroczynski. 1981. Spectral characteristics of wetland habitats.
Frayer, W.E., T.J. Monahan, D.C. Bowden, and F.A. Graybill. 1983. Status and Trends of Wetlands and
Deepwater Habitats in the Conterminous United States, 1950’s to 1970’s. Colorado State University,
Department of Forest and Wood Sciences, Fort Collins, CO.
Gammon, P.T. and V. Carter. 1979. Vegetation mapping with seasonal color infrared photographs.
Gilmer, D.S., A.T. Klett, and E.A. Work. 1976. Monitoring breeding habitat of migratory waterfowl. In: R.S.
Williams, Jr. and W.D. Carter (eds.). ERTS-1 A New Window on Our Planet (U.S. Government Printing
Office, Washington, DC). U.S. Geological Survey Professional Paper 929. pp. 321–325.
Gilmer, D.S., E.A. Work Jr., J.E. Colwell, and D.L. Rebel. 1980. Enumeration of prairie wetlands with Landsat
and aircraft data. Photogrammetric Engineering and Remote Sensing 46:631–634.
Gluck, M., R. Rempel, and P.W.C. Uhlig. 1996. An Evaluation of Remote Sensing for Regional Wetland
Mapping Applications. Ontario Forest Research Institute, Sault Ste Marie, Ontario, Canada. Forest
Research Report No. 137.
Hardin, D.L. 1985. Remote sensing of wetlands for fish and wildlife habitat management in Delaware—A
comparison of data sources. In: CERMA Proceedings, Integration of Remotely Sensed Data in GIS for
Processing of Global Resource Information (Washington, DC).
Hardisky, M.A., M.F. Gross, and V. Klemas. 1986. Remote sensing of coastal wetlands. BioScience 36:453–460.
Hardisky, M.A. and V. Klemas. 1983. Tidal wetlands natural and human-made changes from 1973 to 1979 in
Delaware: mapping and results. Environmental Management 7(4):339–344.
Hathout, S. and J. Simpson. 1982. A vegetation survey of Netley Marsh using color and color infra-red imagery.
Journal of Environmental Management 15:25–34.
Hodgson, M.E., J.R. Jensen, H.E. Mackey, and M.C. Coulter. 1988. Monitoring wood stork foraging habitat
using remote sensing and GIS. Photogrammetric Engineering and Remote Sensing 54(11):1601–1607.
Jacobson, J.E., R.A. Ritter, and G.T. Koeln. 1987. Accuracy of Thematic Mapper derived wetlands as based on
National Wetlands Inventory data. ASPRS/ACSM/WFPLS. 1987 Fall Convention, October 4–9, 1987 at
Reno, Nevada. American Society of Photogrammetry and Remote Sensing, Falls Church, VA. pp. 109–118.
Jensen, J.R., J.E. Estes, and L. Tinney (1980). Remote sensing techniques for kelp surveys. Photogrammetric
Engineering and Remote Sensing 46(6):743–755.
Jensen, J.R., M.E. Hodgson, E. Christensen, H.E. Mackey, L.R. Tinney, and R. Sharitz. 1986. Remote sens-
ing inland wetlands: A multispectral approach. Photogrammetric Engineering and Remote Sensing
52:87–100.
Jensen, J.R., K. Rutchey, M.S. Koch, and S. Narumalani. 1995. Inland wetland change detection in the Everglades
water conservation area 2A using a time series of normalized remotely sensed data. Photogrammetric
Engineering and Remote Sensing 61:199–209.
Johnston, R.M. and M.M. Barson. 1993. Remote sensing of Australian wetlands: An evaluation of Landsat TM
data for inventory and classification. Australian Journal of Marine Freshwater Resources 44:235–252.
Jones, N.L. and F. Shahrokhi. 1977. Application of Landsat data to wetland study and land use classification in
west Tennessee. Proceedings of the 11th International Symposium on Remote Sensing of Environment.
Vol. I, Environmental Research Institute of Michigan, Ann Arbor, MI. pp. 609–613.
Kasischke, E.S. and L.L. Bourgeau-Chavez. 1997. Monitoring south Florida wetlands using ERS-1 SAR imag-
ery. Photogrammetric Engineering and Remote Sensing 63:281–291.
Kindscher, K., A. Fraser, M.E. Jakubauskas, and D.M. Debinski. 1998. Identifying wetland meadows in
Grand Teton National Park using remote sensing and average wetland values. Wetlands Ecology and
Management 5:265–273.
Klemas, V., F.C. Daiber, D.S. Bartlett, O.W. Crichton, and A.O. Fornes. 1973. Coastal Vegetation of Delaware.
The Mapping of Delaware’s Coastal Marshes. University of Delaware, College of Marine Studies,
Newark, DE. Delaware Sea Grant Report DE-SG-15-73.
Krinsley, D.B. 1976. Monitoring water resources in Qom Playa, West-Central Iran. In: R.S. Williams, Jr. and
W.D. Carter (eds.). ERTS-1 A New Window on Our Planet (U.S. Government Printing Office, Washington,
DC). U.S. Geological Survey Professional Paper 929. pp. 139–149.
Kushwaha, S.P.S., R.S. Dwivedi, and B.R.M. Rao. 2000. Evaluation of various digital image processing tech-
niques for detection of coastal wetlands using ERS-1 SAR data. International Journal of Remote Sensing
21:565–579.
Lampman, J.L. 1993. Bibliography of Remote Sensing Techniques Used in Wetland Research. U.S. Army Corps
of Engineers, Waterways Experiment Station, Wetland Research Program, Vicksburg, MS. Technical
Report WRP-SM-2.
Lee, K.H. 1991. Wetlands Detection Methods Investigation. U.S. Environmental Protection Agency, Systems
Laboratory, Environmental Monitoring, Las Vegas, NV. Report No. 600/4-91/014.
Lee, W.T. 1926. The Face of the Earth as Seen From the Air: A Study in the Application of Airplane Photography
to Geography. American Geographical Society, New York. Special Publication 4.
Lillesand, T.M., R.W. Kiefer, and J.W. Chipman. 2008. Remote Sensing and Image Interpretation, 6th edn.
John Wiley & Sons, Hoboken, NJ.
MacConnell, W.P. 1974. Remote Sensing Land Use and Vegetative Cover in Rhode Island. University of Rhode
Island, Cooperative Extension Service, Kingston, RI. Bulletin No. 200.
MacConnell, W.P. 1975. Remote Sensing 20 Years Change in Massachusetts 1952–1972. Classification Manual:
Land-Use and Vegetative Cover Mapping. Massachusetts Agricultural Experiment Station, University of
Massachusetts, Amherst, MA. Bulletin No. 631.
Martel Laboratories, Inc. 1976. Existing State and Local Wetlands Surveys (1965–1975). Vol. II. Narrative. U.S.
Department of the Interior, Fish and Wildlife Service, Office of Biological Services, Washington, DC.
May, L.N., Jr. 1986. An evaluation of Landsat MSS digital data for updating habitats maps of the Louisiana
coastal zone. Photogrammetric Engineering and Remote Sensing 52:1147–1158.
McCormick, J. and H.A. Somes, Jr. 1982. The Coastal Wetlands of Maryland. Maryland Department of Natural
Resources, Annapolis, MD.
McEwen, R.B., W.J. Kosco, and V. Carter. 1976. Coastal wetland mapping. Photogrammetric Engineering and
Mead, R.A. and P.T. Gammon. 1981. Mapping wetlands using orthophotoquads and 35mm aerial photographs.
Niedzwiedz, W.R. and S.S. Batie. 1984. An assessment of urban development into coastal wetlands using his-
torical aerial photography: A case study. Environmental Management 8:205–214.
Olson, D.P. 1964. The Use of Aerial Photographs in Studies of Marsh Vegetation. Maine Agricultural Experiment
Station, Orono, ME. Bulletin No. 13.
Orth, R.J. and K.A. Moore. 1983. Submersed vascular plants: Techniques for analyzing their distribution and
abundance. Marine Technology Society Journal 17:38–62.
10:381–402.
Paine, D.P. and J.D. Kiser. 2012. Aerial Photography and Image Interpretation, 3rd edn. John Wiley & Sons,
Hoboken, NJ.
Patterson, W.G. 1986. Mangrove Community Boundary Interpretation and Detection of Areal Changes in
Marco Island, Florida: Application of Digital Image Processing and Remote Sensing Techniques. U.S.
Fish and Wildlife Service, Washington, DC. Biological Report 86(10).
Pickus, J. 1990. Pearl River Wetlands Advanced Identification: A Geographical Information Systems
Demonstration Project. U.S. EPA, Environmental Monitoring Systems Laboratory, Las Vegas, NV.
Report No. 215-90C05.
Reimold, R.J., J.L. Gallagher, and D.E. Thompson. 1973. Remote sensing of tidal marsh. Photogrammetric
Engineering 39:157–166.
Roller, N.E.G. 1977. Remote Sensing of Wetlands. Environmental Research Institute of Michigan (ERIM), Ann
Arbor, MI. ERIM Report No. 193400-14-T.
Rutchey, K. and L. Vilcheck. 1994. Development of an Everglades vegetation map using a SPOT image and the
global positioning system. Photogrammetric Engineering and Remote Sensing 60:767–775.
Sader, S.A., D. Ahl, and W.-S. Liou. 1995. Accuracy of Landsat-TM and GIS rule-based methods for forest
wetland classification in Maine. Remote Sensing Environment 53:133–144.
Seher, J.S. and P.T. Tueller. 1973. Aerial photos for marshland. Photogrammetric Engineering 39:489–499.
Shaw, S.P. and C.G. Fredine. 1956. Wetlands of the United States—Their Extent and Their Value to Waterfowl
and Other Wildlife. U.S. Department of the Interior, Washington, DC Circular 39.
Shima, L.J., R.R. Anderson, and V.P. Carter. 1976. The use of aerial color infrared photography in mapping the
vegetation of a freshwater marsh. Chesapeake Science 19:74–85.
Spinner, G.P. March 23–25, 1955. Wetlands of the Northeast. U.S. Department of the Interior, Fish and Wildlife
Service, Office of River Basin Studies, Region V, Boston, MA. Prepared for presentation at the Northeast
Wildlife Conference (Atlantic City, NJ).
Stewart, W.R., V. Carter, and P.D. Brooks. 1980. Inland (non-tidal) wetland mapping. Photogrammetric
Stroud, L.M. and A.W. Cooper. 1968. Color-Infrared Aerial Photographic Interpretation and Net Primary
Productivity of Regularly Flooded North Carolina Salt Marsh. University of North Carolina Resources
Research Institute Publication No. 14., Wilmington, NC.
Tiner, R.W. 1977a. An Inventory of South Carolina’s Coastal Marshes. South Carolina Marine Resources
Center, Charleston, SC. Technical Report No. 23.
Tiner, R.W. 1997b. Wetlands. Chapter 13. In: W.R. Philipson (ed.). Manual of Photographic Interpretation
(American Society for Photogrammetry and Remote Sensing, Bethesda, MD). pp. 475–494.
Tiner, R.W. (ed). 2009. Status Report for the National Wetlands Inventory Program: 2009. U.S. Fish and
Wildlife Service, Division of Habitat and Resource Conservation, Branch of Resource Mapping Support,
Arlington, VA.
Townsend, P.A. and S.J. Walsh. 1998. Modeling floodplain inundation using an integrated GIS with radar and
optical remote sensing. Geomorphology 21:295–312.
Wang, J., J. Shang, B. Brisco, and R.J. Brown. 1998. Evaluation of multidate ERS-1 and multispectral Landsat
imagery for wetland detection in southern Ontario. Canadian Journal of Remote Sensing 24:60–68.
Wetlands Subcommittee. 1992. Application of Satellite Data for Mapping and Monitoring Wetlands. Federal
Geographic Data Committee, Reston, VA. Technical Report No. 1.
Wilen, B.O. and M.K. Bates. 1995. The U.S. Fish and Wildlife Service’s national wetlands inventory project.
Wilen, B.O. and G.S. Smith. 1996. Assessment of remote sensing/GIS technologies to improve National
Wetlands Inventory maps. In: Proceedings of the Sixth Biennial Forest Service Remote Sensing
Applications Conference (April 29–May 3, 1996; Denver, CO). 15pp.
Wood, B.L. 1983. Wetland mapping in Colusa County, California. Proceedings of the Society of American
Foresters International Renewable Resource Inventories for Monitoring Conference; August 15–19,
1983; Corvallis, OR. pp. 345–349.
Work, E.A. and D.S. Gilmer. 1976. Utilization of satellite data for inventorying prairie ponds and potholes.
Wright, C.K. 2004. Remote Sensing of Wetlands in Yellowstone National Park. Montana State University,
Bozeman, MT, PhD dissertation.
5 Advances in Remotely
Sensed Data and Techniques
for Wetland Mapping
and Monitoring
Megan W. Lang, Laura L. Bourgeau-Chavez,
Ralph W. Tiner, and Victor V. Klemas
CONTENTS
Introduction....................................................................................................................................... 79
Potential of Different Sensors for Wetland Mapping........................................................................80
Aerial Photography...................................................................................................................... 82
Multispectral Imagery..................................................................................................................84
Hyperspectral Imagery................................................................................................................. 89
Synthetic Aperture Radar............................................................................................................. 91
LiDAR Data................................................................................................................................. 95
Methods to Support Wetland Map Creation...................................................................................... 98
Image Classification.....................................................................................................................99
Collection of In Situ Data.......................................................................................................... 101
Current Wetland Mapping Programs.............................................................................................. 102
U.S. Fish and Wildlife Service’s National Wetlands Inventory................................................. 102
NOAA Coastal Change Analysis Program................................................................................. 103
Canadian Wetland Inventory...................................................................................................... 104
Future Sensors and Missions Relevant to Wetland Mapping.......................................................... 105
Conclusion...................................................................................................................................... 106
Acknowledgments........................................................................................................................... 107
References....................................................................................................................................... 108
INTRODUCTION
Remotely sensed imagery has the potential to add to our understanding of wetlands within the
wider landscape setting and to help ensure their preservation by providing natural resource man-
agers with more accurate and timely information for decision making. Remotely sensed data can
be used to create not only maps of wetland extent and distribution but also estimates of wetland
function and ecosystem services. Being able to do so at a landscape scale allows scientists and
managers to identify and assess the implications of gains and losses and also better understand
spatial interactions among wetlands and between wetlands and other land cover types. Using
multitemporal imagery to address dynamic conditions helps identify and characterize drivers of
wetland change and their impact on wetland functions and services. The full scope and effect of
these drivers can usually only be detected remotely as wetlands are often difficult to access on
79
the ground, on-site mapping at the landscape scale is usually cost prohibitive, and landscape-scale
patterns simply may not be evident at the field scale (Harvey and Hill 2001; Rundquist et al. 2001;
Baker et al. 2006; Silva et al. 2008).
Significant effort has been made by scientists and managers to provide quality map products, and
recently developed remote sensing technologies have the potential to further improve map accuracy
and information content. The diversity of remotely sensed data and techniques available to process
these data have increased rapidly since the 1950s, when the United States first began to systemically
map the nation’s wetland resources using aerial photography. When satellite data initially became
available in the 1970s, it was used to produce more generalized wetland maps. Since then, techno-
logical advances have led to the creation of more accurate, finer-spatial-resolution satellite-based
wetland maps. Still, the dynamic nature of wetlands (e.g., ephemeral wetness), their diversity (e.g.,
variations in plant structure and phenology), the multitude of landscape positions they occupy, and
the variable proportion of the landscape that they represent continue to challenge the capabilities of
existing sensors, whether they be airborne or satellite based (see Chapter 25). Recent advances in
the quality and availability of light detection and ranging (LiDAR), synthetic aperture radar (SAR),
multispectral and other types of data, and the introduction of new data dissemination and processing
methods hold great potential for improving wetland mapping and monitoring efforts. These differ-
ent geospatial datasets provide complementary information about wetland presence and character.
Many of these datasets and the techniques necessary to apply them are uniquely suited for specific
landscapes, wetland types, and mapping goals. The use of different wetland classification systems
(e.g., Cowardin et al. 1979) also influences the suitability of datasets and techniques (Powers et al.
2012). This chapter reviews the advantages and disadvantages of using different types of remotely
sensed data, including aerial photographs, multispectral, hyperspectral, SAR, and LiDAR to map
wetlands. The primary techniques used to produce wetland maps are discussed, and then a short
review of three national wetland mapping programs, which demonstrate how these different tools
and techniques are applied, is provided.
POTENTIAL OF DIFFERENT SENSORS FOR WETLAND MAPPING

Aerial photographs have traditionally been used to map wetlands, but more recently initiated wet-
land mapping efforts have integrated satellite data, including both multispectral and SAR data,
into their wetland mapping protocols. Indeed, a considerable amount of literature supports the
use of other types of data, in addition to aerial photography, for wetland mapping. This section
investigates the advantages and disadvantages of using aerial photography, multispectral, hyper-
spectral, SAR, and LiDAR imagery for wetland mapping. This is not an exhaustive review of all
types of remotely sensed imagery. Instead, it addresses the types of imagery that are most com-
mon and that current evidence suggests have the greatest potential to improve wetland mapping.
Both spaceborne and airborne imaging systems are discussed. It is important to not only differen-
tiate between different types of imagery but also between the different platforms used to collect
those images since there are advantages and disadvantages inherent to these vantage points. The
advantages of using satellite imagery include spatial coverage, timeliness, repeatability, and often
lower costs (Federal Geographic Data Committee 1992; Dobson et al. 1995; Papa et al. 2006).
However, the use of satellite data also implies certain limitations. These limitations include the
often greater interference of weather and atmosphere with data collection and, in some cases, the
reduced ability to collect data during key time periods (e.g., when wetlands are inundated). In
the past, the use of satellite data has meant coarser spatial resolutions, but this limitation is less
of an issue as more satellite-based sensors are now collecting finer-spatial-resolution data. The
collection of submeter satellite data is now possible, with data resolutions of a few meters being
fairly common and sub-half-meter data being available as well (e.g., GeoEye-1; 0.41 m spatial
resolution; see Tables 5.1 and 5.2).
TABLE 5.1
General Specifications of Some Commonly Used, Relatively Recent Sensors That Have or Could Be Utilized to
Map Wetlands and Monitor Changes, with the Exception of MODIS, VIIRS, ASTER, SPOT-6, and Landsat 8
Sensor Resolution (m) Bands Width (km) Repeat Life Span Information
OrbView-3 4 (1 P) P, B, G, R, NIR 8 <3 days 2003–2007 digitalglobe.com
SPOT 1, 2, and 3 20 (10 P) P, G, R, NIR 60 var 1986–2009 astrium-geo.com
Landsat TM 30 (120 TIR) B, G, R, NIR, SWIRx2, TH 185 16 days 1982–2011 eros.usgs.gov
SPOT 4 20 (10 mono) M, G, R, NIR, SWIR 60 var 1998–2013 astrium-geo.com
AVHRR/3 1100 G, NIR, SWIR, MIR, TH, FIR 3000 2× daily 1998–Pres. noaasis.noaa.gov
Landsat ETM+ 30 (60 TIR; 15 P) P, B, G, R, NIR, SWIRx2, TH 185 16 days 1999–Pres. eros.usgs.gov
IKONOS 3.2 (0.82 P) P, B, G, R, NIR 11 var 1999–Pres. digitalglobe.com
QuickBird 2.4 (0.61) P, B, G, R, NIR 18 var 2001–Pres. digitalglobe.com
SPOT 5 10 (20 MIR; 2.5–5 P) P, G, R, NIR, SWIR 60 var 2002–Pres. astrium-geo.com
MERIS 300–1000 Bx3, Gx2, Rx4, NIRx6 1150 3 days 2002–2012 earth.esa.int
IRS LISS-III 24 G, R, NIR, SWIR 141 24 days 2003–Pres. nrsa.gov.in
IRS LISS-IV 5.8 P, G, R, NIR 24–70 5 days 2003–Pres. nrsa.gov.in
AWiFS 56–70 G, R, NIR, SWIR 370–740 5 days 2003–Pres. nrsa.gov.in
WorldView-1 0.55 P 17.6 var 2007–Pres. digitalglobe.com
GeoEye-1 1.65 (0.41) P, B, G, R, NIR 15.2 <3 days 2008–Pres. digitalglobe.com
RapidEye 5 B, G, R, NIRx2 77 1–5 days 2008–Pres. blackbridge.com
WorldView-2 1.84 (0.46) P, Bx2, G, Y, Rx2, NIRx2 16.4 1–3 days 2009–Pres. digitalglobe.com
Pleiades 2 (0.50) P, B, G, R, NIR 20 1 day 2011–Pres. astrium-geo.com
Notes: Sensor name (Sensor), spatial resolution in meters (Resolution), the portion of the electromagnetic spectrum sampled by each band (Bands),
swath width in kilometers (Width), temporal resolution or number of days between overpasses (Repeat), years the sensor has operated (Life Span;
present abbreviated as Pres.), and websites where more information about the sensors can be found are provided in the table. Bands are named to
indicate the portion of the electromagnetic spectrum that they sample (P, panchromatic; B, blue; G, green; Y, yellow; R, red; NIR, near-infrared;
SWIR, shortwave infrared; MIR, midwave infrared; TIR, thermal infrared; and FIR, far infrared; see http://database.eohandbook.com for wave-
length ranges). Specialty bands (e.g., coastal and red edge) are included in the previously mentioned classes. For example, coastal bands may be
listed under blue (e.g., WorldView-2). Multiple bands in one portion of the electromagnetic spectrum are denoted with an “x” and then the number
Advances in Remotely Sensed Data and Techniques for Wetland Mapping and Monitoring
of bands (e.g., x2). Note that some columns represent more than one sensor (constellation) and that specifications may vary slightly with time.
81
TABLE 5.2
General Specifications of Other Relatively Recent, Commonly Used Sensors That Have
Been Utilized to Map Wetlands or Have the Potential to Do So
Resolution Incident Width
Sensor Plat. (m) Bands (°) Polarization (km) Life Span Information
SIR-C sh 10–200 X, C, L 20–50 Polarimetric 15–90 1994 jpl.nasa.gov
X-SAR sh 10–200 X 20–50 VV 15–40 1994 jpl.nasa.gov
JERS sa 18 L 39 HH 75 1992–1998 global.jaxa.jp/
ERS-1 sa 30 C 23 VV 100 1991–2000 earth.esa.int
ERS-2 sa 30 C 23 VV 100 1995–2011 earth.esa.int
RADARSAT-1 sa 8–100 C 10–59 HH 50–500 1995–2013 space.gc.ca
ASAR sa 30 m to C 15–45 HH, VV, 56–400 2002–2012 earth.esa.int
1 km VH, HV
PALSAR sa 10–100 L 8–60 Polarimetric 70–250 2006–2011 global.jaxa.jp/
TerraSAR-X sa 1–40 X 15–60 Polarimetric 10–100 2007–Pres. dlr.de/en
RADARSAT-2 sa 3–300 C 70–60 Polarimetric 10–500 2007–Pres. radarsat2.info
COSMO- sa 1–100 X 20–59 Polarimetric 10–200 2007–Pres. cosmo-skymed.it
SkyMed
Notes: Specifications are approximate and not all specifications are available at the same time. Sensor name (Sensor), plat-
form type (Plat.), spatial resolution in meters (Resolution), wavelength band (Bands), incident angle (Incident),
polarization, swath width in kilometers (Width), the years the sensor has operated (Life Span; present abbreviated as
Pres.), and websites where more information about the sensors can be found are provided in the table.
Aerial Photography
Aerial photography has been collected since the 1800s and is still utilized for a variety of opera-
tional mapping tasks today, including the mapping of wetlands and water bodies where spatial
detail is desired (see Chapter 4). It is also used to validate or provide classification training data
when using coarser-resolution satellite data to map wetlands. The first aerial photograph was col-
lected in 1858 by Gaspard Felix Tournachon aboard a hot air balloon (Jensen 2000). Since then,
aerial photography has matured through advancements in both platform and sensor. Airplanes and
even unmanned aerial vehicles are now used to collect aerial photographs, and black-and-white
film has largely been replaced by color film and digital images. Although similar to other types
of optical data, aerial photography cannot be used to quantitatively measure soil moisture and has
other limitations including the following: (1) detection of inundation in vegetated ecosystems can be
compromised by the presence of a vegetative canopy, (2) small water bodies can be misinterpreted
as shadows, and (3) turbid water can be misidentified as soil. However, aerial photography provides
a consistent long-term historic record of surface conditions and changes at a relatively fine spatial
scale. Furthermore, the aerial platform provides flexibility with the timing of data collection that the
satellite platform lacks. For example, airplanes can collect images at certain times of the day that
correspond with optimal tidal levels, whereas the overpass time of satellites is predetermined by the
orbit. It is not uncommon to have a historic record of aerial photographs dating back to the 1950s
or earlier (Morgan et al. 2010). Some of the most common sources of aerial photography in the
United States are the National Aerial Photography Program (NAPP; 1987–2007), the National High
Altitude Aerial Photography (NHAP; 1980–1987) program, and the National Agriculture Imagery
Program (NAIP; 2001–Present). Aerial photographs are also commonly collected by states and
regional/local entities.
Although aerial photographs can be analyzed using automated techniques (i.e., an approach
that has become more popular since digital aerial photography was made available in the 1990s;
Advances in Remotely Sensed Data and Techniques for Wetland Mapping and Monitoring 83
Morgan et al. 2010), aerial photographs are still commonly analyzed using photointerpretation
techniques. Photointerpretation relies on the photointerpreter to make qualitative decisions based on
the image’s tone, color, spatial patterns, texture, height/topography, associations, and other charac-
teristics (Olson 1960). Although these criteria are commonly applied to many different ecosystems,
they can be adapted to map wetlands through expert knowledge. For example, a dark tone is often
indicative of higher soil moisture or higher soil organic carbon levels caused by anoxic soil condi-
tions. The accuracy of aerial photointerpretation relies heavily on the experience and judgment of
the photointerpreter (see Morgan et al. 2010 and Jensen 2000 for reviews of aerial photography his-
tory, fundamentals, and interpretation).
When aerial photographs are viewed stereoscopically, additional information can be gained
regarding topography and the height of features within the image. This process facilitates the iden-
tification of topography so that depressions, slopes, flats, bottomlands, and floodplain terraces can
be readily detected. With recent advances in computer-assisted cartography (e.g., geographic infor-
mation systems [GISs]) and digital imagery, stereoscopic interpretation has been largely replaced
with on-screen monoscopic viewing. On-screen interpretation does have advantages over traditional
photointerpretation. For example, the interpreter is able to zoom in and view areas at varying scales
versus the fixed scale of hard copy aerial photographs. Stereoscopic on-screen interpretation is pos-
sible but requires costly hardware, specialized optics, and digital stereo imagery, which are beyond
the budget of most agencies. The State of Massachusetts is one exception as it uses these techniques
to produce highly detailed and accurate wetland maps for the state’s wetland conservancy pro-
gram. Digital elevation models (DEMs) originally produced using stereo photointerpretation are
commonly available throughout the United States. However, the majority of aerial photographs
collected today are not collected as part of stereopairs. DEMs derived from LiDAR are rapidly
replacing aerial photograph-based DEMs.
Studies have compared the use of aerial photographs and aerial photointerpretation to the use
of multispectral data for wetland mapping in different environments (Shapiro 1995; Rutchey and
Vilchek 1999; Harvey and Hill 2001). One of the major disadvantages of aerial photographs as
compared to multispectral data is the relatively lower spectral resolution of the aerial photographs,
but the superior spatial resolution of aerial photographs may compensate for this disadvantage.
A study conducted in the Florida Everglades found that Systeme Pour l’Observation de la Terre
(SPOT) data overestimated the amount of cattail present in an impounded wetland due to confound-
ing factors, such as fire, hydrology, periphyton species composition, and macrophyte morphology.
Aerial photography was more effective for mapping cattail in these areas due to the experience
and reasoning of the photointerpreter (Rutchey and Vilchek 1999). Harvey and Hill (2001) com-
pared the ability of multispectral data (SPOT and Landsat Thematic Mapper [TM]) to that of aerial
photographs and manual interpretation techniques for mapping an Australian freshwater swamp.
Although 14 land cover types could be delineated from 1:15,000 aerial photographs with 89%
accuracy, only 3 broad land cover types could be distinguished using the multispectral data while
meeting minimum accuracy requirements (≤80%). The difference in accuracy between the aerial
photographs and multispectral data was, in part, attributed to the contextual and textural informa-
tion used by the photointerpreter (Harvey and Hill 2001).
Although manual interpretation of aerial photography is the most commonly used method for
detailed wetland mapping and this method is well developed within the operational framework, it is
subjective (Finlayson and van der Valk 1995; Shapiro 1995; Augusteijn and Warrender 1998; Baker
et al. 2006), time consuming and therefore cannot be quickly updated (Phinn et al. 1999; Harvey
and Hill 2001; Wright and Gallant 2007), and relatively expensive (Dobson et al. 1995; Lunetta and
Balogh 1999; Mumby et al. 1999; Phinn et al. 1999; Harvey and Hill 2001; Hirano et al. 2003; Klemas
2013). There are limitations to the use of human vision to classify a map (Rutchey and Vilchek 1999),
and these decisions are inherently subjective so resultant maps may vary according to the person who
creates them (Finlayson and van der Valk 1995; Augusteijn and Warrender 1998; Baker et al. 2006).
Mumby et al. (1999) estimate that the creation of maps for a relatively small area (~150 km2) may
take six times as long using aerial photointerpretation rather than automated interpretation of digital
airborne imagery and that the cost of manual interpretation of aerial photographs will increase at a
faster rate than the cost of digital interpretation. The inability of the U.S. Fish and Wildlife Service
(FWS) to complete mapping the country’s wetlands in 35+ years using aerial photointerpretation
methods perhaps illustrates the limitations of aerial photointerpretation for mapping large geographic
areas, although mapping priorities and detail were the major factors affecting completion.
The type of aerial photograph used to create a wetland map has the potential to affect map
accuracy. Although photointerpretation has traditionally been a manual process, which utilized
analog photographs rather than digital images, analog aerial photographs can be converted to
digital. Still the challenge of scanning in large numbers of photographs and normalizing photo-
graphs for variations in illumination, angle of acquisition, and sometimes even phenology and
atmospheric conditions remains a challenge (Phinn et al. 1999). Digital cameras are an improve-
ment over analog aerial photographs and can be processed using methods similar to those used
for processing multispectral images (Phinn et al. 1999), although the lack of spectral information
is still a limitation (Becker et al. 2007; Silva et al. 2008). Aerial photographs are not calibrated to
one another so images collected by different sensors may look quite different from one another.
Film processing can produce versions of the same image with different characteristics (e.g., domi-
nance of blue or red). If possible, it is usually advantageous to use color or color-infrared (CIR)
aerial photographs to map wetlands, rather than black-and-white images. CIR film is usually
preferred for wetlands mapping because it includes more spectral information than true color
film and allows for greater contrast between different plant communities (Federal Geographic
Data Committee 1992; Tiner 1999). Dale et al. (1996) found that CIR aerial photography was a
valuable tool in the identification of wetlands. A major exception to this statement occurs when
mapping submerged aquatic vegetation (SAV). The relatively low spectral resolution of aerial
photographs significantly limits their use for the detection of SAV in general (Silva et al. 2008).
However, the use of true color film for SAV detection is preferable to the use of CIR film due to
the superior ability of shorter-wavelength energy (the visible bands that compose true color imag-
ery) to penetrate the water column (Tiner 1999).
Multispectral Imagery
Similar to aerial photographs, multispectral images are collected by passive sensors that gather
data in the visible and near-infrared portions of the electromagnetic spectrum. However, multispec-
tral sensors also collect information on the reflectance or emittance of longer-wavelength energy,
including mid-infrared and thermal, and typically sample several discrete portions of the electro-
magnetic spectrum (Table 5.1). Studies have found that the increased spectral resolution of multi-
spectral data as compared to aerial photography enhances wetland mapping (Federal Geographic
Data Committee 1992; Phinn et al. 1999; Harvey and Hill 2001; Töyrä et al. 2002) and, in some
cases, helps compensate for reduced spatial resolution (Harvey and Hill 2001). However, some
satellite-based multispectral sensors are now available with a spatial resolution similar to that of
aerial photographs (i.e., <1 m; GeoEye and WorldView). In addition, these images are usually less
expensive than aerial photographs (Mumby et al. 1999), and they can be collected regularly, as dic-
tated by satellite orbits (Rundquist et al. 2001; Li and Chen 2005). Therefore, these sensors provide
not only increased spectral but also better temporal resolution as compared to aerial photographs,
both of which are important for distinguishing different wetland classes. The comparatively fine
temporal resolution of these multispectral sensors is complemented by the relatively long historic
record of moderate-spatial-resolution multispectral data (e.g., Landsat MSS first launched in 1972).
Finally, their digital format and robust calibration allow for automated, repeatable classification of
wetlands and other land cover types (Houhoulis and Mitchner 2000; Ausseil et al. 2007), which is
not the case for aerial photographs. Their ability to capture a larger areal footprint also simplifies
image interpretation (Rundquist et al. 2001; Li and Chen 2005).
(a) (b)
FIGURE 5.1 Series of images from Landsat 8 showing preflood (a; May 17, 2013) and flooding (b; October
24, 2013) conditions after a typhoon and heavy rains in Cambodia. (From U.S. Geological Survey.)
Multispectral satellite data have been broadly available since 1972, when the Earth Resource
and Technology Satellite (ERTS) was launched. The name of the ERTS program was subse-
quently changed to Landsat, and several follow-on missions were launched including Landsat 2
(Multispectral Scanner) through 7 (Enhanced Thematic Mapper + [ETM+]; presently operating).
The resulting dataset is the only long-term civilian archive of satellite imagery at scales of human
influence, resolving individual farm fields, deforestation patterns, urban expansion, and other types
of land use and land cover (LULC) change. The Landsat Data Continuity Mission (LDCM) or
Landsat 8 was launched in February 2013 (http://science.nasa.gov/missions/ldcm/) and will help
ensure the provision of Landsat data into the future (Figure 5.1). LDCM consists of two sensors: the
Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS). OLI collects data in nine
visible, near-infrared, and shortwave infrared bands at a spatial resolution of 30 m for most bands
and 15 m for a panchromatic band. TIRS collects data in two thermal bands with a spatial resolu-
tion of 100 m. SPOT is a group of fine-to-moderate resolution satellite-based multispectral sensors
(SPOT 1–6). SPOT data have been available since 1986 and are still being collected today. SPOT
1, 2, and 3 had a 10 m panchromatic band and three 20 m multispectral bands (green, red, and near-
infrared), and SPOT 6, which was launched in 2012, has a 1.5 m panchromatic band and four 8 m
multispectral bands from blue (0.45) to near-infrared (0.89 µm). SPOT 7, which is similar to SPOT
6, is scheduled to be launched in 2014 (Airbus Defense & Space 2014). The Advanced Spaceborne
Thermal Emission and Reflection Radiometer (ASTER) collects data in 15 bands spanning the vis-
ible, near-infrared, shortwave infrared, and thermal portions of the electromagnetic spectrum with
spatial resolutions between 15 and 90 m. ASTER initiated data collection in 1999 and is currently
collecting data, with the exclusion of bands 5–9 (1.6–2.37 µm; Abrams et al., accessed 2012). The
Moderate-Resolution Imaging Spectroradiometer (MODIS) collects imagery from 0.41 to 14.39 µm
with a repeat time of 1–2 days and a variable spatial resolution (250 m, bands 1–2; 500 m, bands
3–7; 1000 m, bands 8–36; http://modis.gsfc.nasa.gov). MODIS includes bands that were designed
to monitor a number of biophysical parameters, including vegetation greenness, terrestrial surface
temperatures, and atmospheric conditions. Visible Infrared Imaging Radiometer Suite (VIIRS)
was designed to improve and extend the MODIS and Advanced Very High Resolution Radiometer
(AVHRR) historic data record, collecting data in 22 spectral bands spanning the visible as well
as the near, middle, and thermal infrared (TIR) wavelength regions with a spatial resolution of
750 m (http://npp.gsfc.nasa.gov/viirs.html). Other commonly used multispectral datasets include
the Indian Remote Sensing Satellites such as the Linear Imaging Self Scanner (LISS), QuickBird,
IKONOS, OrbView, GeoEye, and WorldView (Table 5.1).
The infrared portion of the electromagnetic spectrum is considered to be particularly well suited
for wetland mapping applications (Federal Geographic Data Committee 1992; Phinn et al. 1999;
Munyati 2000; Davranche et al. 2013; Huang et al. 2014). This is due to the high reflectance of veg-
etation in the near-infrared coupled with the strong absorption of water in the same region, which
makes for a sharp contrast between water and vegetation (Lyon and McCarthy 1995) and water
absorption features (portions of the electromagnetic spectrum with distinctly lower reflectance
when water is present) in the mid-infrared. Thermal images have been used to map inundation (e.g.,
Leblanc et al. 2011) and even soil moisture, but this portion of the spectrum is infrequently used to
map wetlands due to its relatively coarse spatial resolution and other limitations. The use of CIR
aerial photos can also improve wetland detection via information on the near-infrared portion of the
electromagnetic spectrum. However, since CIR aerial photographs only collect information on one
portion of the infrared, use of multispectral images may be more advantageous for mapping wet-
lands. Multispectral sensors can sample multiple portions of the infrared spectrum, and these cali-
brated bands can be quantitatively assessed using a greater variety of image processing procedures.
Adequate spatial resolution is necessary to map wetlands since wetland patches are often small
and their boundaries can be complex and challenging to delineate even on the ground (Chapter 3).
Multispectral sensors collect data with a wide range of spatial resolutions, but sensors with coarse
resolutions are not typically used to map wetlands. However, they can be used to augment finer-
resolution sensors when improved temporal resolution is needed to depict phenologic or hydrologic
dynamics. For example, AVHRR with an approximate spatial resolution of 1 km is usually not con-
sidered a viable option for wetland mapping (Munyati 2000; Ringrose et al. 2003), but it collects data
approximately every 12 h. Due to their limited utility for wetland mapping, coarse-resolution mul-
tispectral sensors are not discussed in detail below. Even the spatial resolution of well-established
multispectral sensors, like Landsat TM and ETM+ (30 m) and SPOT (8–20 m), is often insufficient
for mapping smaller or more complex wetlands (Ramsey and Laine 1997; Welch et al. 1999; Ausseil
et al. 2007; Wright and Gallant 2007). Phinn et al. (1999) suggest that some multispectral sensors
(Landsat TM, SPOT XS, and IRS-1C) are limited to mapping wetlands with a minimum mapping
unit (MMU) greater than 9 ha. The MMU represents the smallest wetland that appears on the map.
An MMU is different from a target mapping unit (TMU), which estimates the smallest wetland
that is consistently mapped. To map wetlands at a certain MMU requires an appropriate scale of
imagery. A study evaluating remote sensing for mapping South African wetlands found that at least
88% of wetlands identified via ground surveys were not detected by Landsat ETM+ (30 m spatial
resolution) as most of the wetlands were less than 1 ha in size (De Roeck et al. 2008). However, other
studies estimate that Landsat TM is capable of mapping wetlands at much finer MMU (0.8–1.0 ha;
Federal Geographic Data Committee 1992; Lunetta and Balogh 1999; Wright and Gallant 2007). In
comparison, when mapping wetlands in Maryland using CIR aerial photography, Tiner and Smith
(1992) recommended imagery with a spatial scale of 1:58,000 to produce maps with an MMU of
0.4 ha and 1:24,000 imagery to produce maps with an MMU of 0.1–0.2 ha. A TMU of 0.2 ha is now
the recommended standard for federal wetland mapping in the United States (Federal Geographic
Data Committee 2007).
A number of studies have indicated that the spatial resolution of commonly available multispec-
tral sensors has limited their use for wetland mapping. Landsat MSS data (80 m spatial resolution)
have been used to map wetlands with varying levels of success (Moore and North 1974; Bennet
1987). Severs et al. (1974) were able to use Landsat MSS bands 5 (red) and 7 (near-infrared) to clas-
sify wetland patches of ~4 ha or more into four broad categories (i.e., marsh, seasonally flooded
depressions, meadow, and open water). However, in another case, researchers had difficulties using
Landsat MSS to map large, homogeneous coastal wetlands, which are among the easiest wetland
types to map (Ramsey and Laine 1997). The use of Landsat TM and other moderate-spatial-
resolution multispectral sensors (e.g., Indian Remote Sensing Satellite LISS-III) for inventorying
wetlands is also limited due to the lack of spatial detail (Wilen and Tiner 1989; Ramsey and Laine
1997; Shanmugam et al. 2006). Mixed pixels are often blamed for this difficulty. In using Landsat
for locating wetlands, Zomer et al. (2009) found that a majority of pixels within a Landsat image
were mixtures of several plant species or vegetation types in various proportions, especially along
the wetland–upland border or between communities due to the short distance or sharp boundary,
respectively. Civco et al. (2006) had difficulty using Landsat ETM to separate the upper salt marsh
from upland forest (i.e., a simple task for manual interpretation of aerial photos) due to the spatial
resolution of that sensor that produced many mixed pixels (part forest, part marsh). Instead of rely-
ing on Landsat ETM, they used elevation data to help make the break. The use of LiDAR also made
it possible to distinguish between cattails (Typha spp.), common reed (Phragmites australis), and
salt hay cordgrass (Spartina patens). Muster et al. (2013) found similar problems with Landsat for
detection of small Arctic ponds. They concluded that the best results were obtained with imagery
having spatial resolutions of 2 m or better. Images with lower spatial resolution not only missed
small water bodies but also produced local overestimates of water surface area in specific locales
when clusters of small water bodies were aggregated into single larger water bodies. The use of
regression trees for subpixel mapping of image classes can help remedy the mixed pixel problem,
especially when finer-resolution data are used for calibration (Huang et al. 2014). The application of
MODIS with its 36-band sensor is even less promising for local to regional wetland mapping due to
its relatively coarse spatial resolution (250–1000 m), although it is well suited for global scale map-
ping and has been used to extrapolate maps created with finer-spatial-resolution data to a broader
extent (e.g., Deepwater Horizon oil spill; Leifer et al. 2012; Petus et al. 2013) and correct for pheno-
logic variability in maps created using finer-resolution data (Zhao et al. 2012).
Finer-spatial-resolution multispectral satellite data are now available (Table 5.1). For example,
IKONOS collects 4 m resolution data over the blue, green, red, and near-infrared areas of the elec-
tromagnetic spectrum, whereas QuickBird collects 2.4 m data in the blue, green, red, and near-
infrared regions. WorldView-2 provides nine data bands with a spatial resolution between 0.5 and
1.8 m. It was predicted that these finer-resolution multispectral sensors would increase wetland
mapping capability (Phinn et al. 1999), and there has been a recent increase in the use of this
type of imagery for wetland mapping (e.g., Belluco et al. 2006; Ouyang et al. 2011; Ashraf et al.
2012; Timm and McGarigal 2012). However, the increased difficulty and expense of acquiring fine-
resolution multispectral imagery relative to moderate-resolution (e.g., 30 m) imagery (Nagendra
et al. 2013) has probably slowed this advance.
As with aerial photography, multispectral images can only be collected during the day, and
their collection is limited by cloud cover and atmospheric conditions (e.g., haze). These limita-
tions may preclude the regular and repetitive collection of data that is required for certain wet-
land studies (Baghdadi et al. 2001; Costa 2004; Costa and Telmer 2007). This is especially true
if combined with infrequent temporal coverage (e.g., satellite repeat frequency). Unfortunately,
the optimal period for data collection may be short. For example, in many forested regions of
the United States, there is a short time period between ice/snow melt and the leafing-out of trees,
after which features below the forest canopy are usually obscured. More than one study has cited
the presence of a vegetative canopy as one of the main deterrents to accurate wetland mapping
using multispectral data (Moore and North 1974; Carter 1982; Alsdorf et al. 2007; Costa and
Telmer 2007; Davranche et al. 2013). To help avoid this issue, Huang et al. (2014) used leaf-off
Landsat data to map inundation below the forest canopy. Unless significant gaps in the canopy are
present, hydrologic conditions on the ground cannot be observed. Unfortunately, in the subtrop-
ics and tropics, optimal time for image collection often corresponds with periods of increased
cloudiness (precipitation events—the rainy season). Since Landsat can only be acquired every
16 days (8 days if both operational Landsat sensors are used), the chance of collecting a relatively
cloud-free image during this optimal period is reduced (Federal Geographic Data Committee
1992). Temporal resolution was found to be an important variable when using Landsat TM to map
a diverse group of wetlands on the border between Maryland and Delaware. When the accuracy
of a wetland map produced with single-date leaf-on (June) TM imagery was compared to one that
was produced using multitemporal TM data (leaf-on and leaf-off [April]), the multitemporal map
was found to be more reliable (Lunetta and Balogh 1999). The leaf-on data were used primar-
ily to produce a land cover map, while the leaf-off data were used to detect wetland hydrology.
The accuracy of the multitemporal wetland map was superior to that of the single-date map,
88% accuracy versus 69% accuracy, respectively. Davranche et al. (2010) found that different
SPOT-based indices performed best for detecting different wetland plant classes (i.e., submerged
macrophytes and common reed) during different seasons. Forested wetlands were found to be
especially difficult to map without the use of multidate imagery. On the other hand, open water
areas are relatively easy to identify using Landsat data. Beeri and Phillips (2007) used Landsat
TM and ETM+ to estimate hydroperiod in wetlands with open water with high accuracy (96%
detection of water bodies greater than ~15 m). They suggested the use of SAR data to improve
the detection of water beneath emergent vegetation and fine-resolution elevation data to locate the
areas of likely inundation since the presence of vegetation (Moore and North 1974; Carter 1982)
and sediment (Engman and Gurney 1991) reduces the effectiveness of multispectral data for the
detection of inundation. The ability of some sensors (e.g., SPOT and QuickBird) to collect data
at multiple view angles and the use of microsatellites (Dronova et al. 2011) have the potential to
reduce the time between satellite overpasses and therefore increase the likelihood that images
can be collected during times when conditions on the ground are optimal. However, spectral
resolution is also important because the spectral character of different land cover classes is often
similar and adequate spectral resolution is necessary to distinguish different land cover types
(Harvey and Hill 2001).
Two of the most commonly used series of multispectral sensors are SPOT (Jensen et al. 1991;
Rutchey and Vilcheck 1999; Ringrose et al. 2003; Töyrä and Pietroniro 2005; Davranche et al. 2010,
2013; Poulin et al. 2010) and Landsat (Sader et al. 1995; Townsend and Walsh 1998; Baghdadi et al.
2001; Töyrä et al. 2002; Baker et al. 2006; Wright and Gallant 2007; Bwangoy et al. 2010; Song
et al. 2012; Zhao et al. 2012; Huang et al. 2014). Landsat data have been found to be preferable to
other multispectral satellite data for wetland detection (Hewitt 1990; Bolstad and Lillesand 1992;
Harvey and Hill 2001), and the fact that it is being distributed free of charge makes its use even more
attractive. Although Harvey and Hill (2001) found that both Landsat TM and SPOT XS were ade-
quate for mapping broad categories of wetland types, the coarser-spatial-resolution 7-band Landsat
TM data were preferable to the finer-resolution 4-band SPOT data for mapping wetlands in a fresh-
water Australian swamp. Longer-wavelength bands, especially the mid-infrared Landsat TM band
(band 5; 1.55–1.75 μm), have been found to be especially useful for the detection of water (Federal
Geographic Data Committee 1992; Phinn et al. 1999; Harvey and Hill 2001; Töyrä et al. 2002). The
Federal Geographic Data Committee (1992) identified Landsat TM bands 4 and 5 as most effective
for wetland delineation. Landsat TM bands 2–4 have been found to be helpful for identifying the
presence of understory vegetation (Congalton et al. 1993), which can assist in mapping some for-
ested wetlands (Harvey and Hill 2001). Sader et al. (1995) found that Landsat TM could be used to
distinguish between forested wetlands, other wetlands, forested uplands, and other uplands with an
accuracy of approximately 80%. The relatively recent distribution of no-cost Landsat data has the
potential to significantly improve not only change detection (Pattanaik and Prasad 2011; Song et al.
2012; Huang et al. 2014) but also wetland boundary detection. Temporal variations in phenology,
as well as other dynamic environmental factors, have been found to improve wetland identification
(Gao and Zhang 2006). SPOT data have also been useful for wetland mapping, especially when
differentiating medium and small wetland patches, which are spectrally distinct from other land
cover types. Töyrä et al. (2002) found that SPOT was capable of detecting flooded wetlands with
between 66% and 80% accuracy in freshwater wetlands with a combination of forests and herba-
ceous vegetation. Davranche et al. (2010) used multiseason SPOT-5 data to map common reed and
submerged macrophytes at relatively high levels of accuracy (98%–99% for common reed and 86%
for submerged macrophytes). Similarly, Poulin et al. (2010) used SPOT-5 to map reed habitats, using
multiseasonal data to create a number of spectral indices that were successfully used to characterize
reed structure.
Acquiring multispectral imagery at multiple view angles may enhance the discrimination of
different land cover types, including wetlands (Vanderbilt et al. 2002; Dupigny-Giroux 2007). This
enhanced discrimination is due to the varying appearance of wetland vegetation and water signa-
tures when viewed from different angles and is sometimes called a directional signature (Dupigny-
Giroux 2007). The Polarization and Directionality of the Earth’s Reflectance (POLDER) data were
found to reliably discriminate three cover types (open water, emergent vegetation above inunda-
tion, and non-inundated cover types) when images were collected at angles that maximize spec-
ular reflectance of the water surface (i.e., maximize the interception of sun glint by the sensor;
Vanderbilt et al. 2002). Pinty et al. (2002) found that multidirectional signatures could be used to
distinguish between inundation and dark soils, which often appear similar on multispectral images.
Another study found that ratios of AirMISR data collected at different angles (fore and aft of nadir:
26.1°, 45.6°, and 60.0°) were capable of distinguishing not only different wetland types in a forested
area of Maine but also moisture gradients in emergent wetlands, species type and vigor, the relative
proportion of water and vegetation, and areas of vegetation undergoing moisture stress (Dupigny-
Giroux 2007). Dupigny-Giroux (2007) postulates that a multiangle approach may help map ever-
green wetlands. Another study found that a combination data collected to maximize sun glint and
spectral mixture analysis could be used to detect the areas of wetland inundation, with accuracy of
within pixel estimates of inundation increasing with increasing pixel size (Vanderbilt et al. 2007).
Wetlands that are spectrally distinct from surrounding land cover types will be mapped more
accurately with multispectral images than those that are not (McCarthy et al. 2005). Finer-spatial-
resolution sensors are better adapted to mapping smaller wetlands, but coarser-resolution data may
be perfectly suited for mapping larger wetlands. Thus, no one existing multispectral sensor is opti-
mal for wetland mapping in all situations, and the success of wetland mapping efforts often depends
just as much on the type of wetland being mapped and/or the wetland classification system being
used as the type of imagery being used (Becker et al. 2007).
Hyperspectral Imagery
Hyperspectral data are characterized by numerous, narrow, contiguous spectral bands collected
in the visible, near-infrared, mid-infrared, and sometimes thermal portions of the electromagnetic
spectrum. This is in contrast to multispectral bands that are typically wider, often not contiguous,
and not as abundant. Numerous, finely segregated spectral bands of hyperspectral data allow image
analysts to identify different materials based on their “spectral signature” or diagnostic patterns
in absorption and reflection, usually associated with the molecular and/or cellular properties of
the material (Kokaly et al. 2003; Schmidt and Skidmore 2003). For example, the U.S. National
Aeronautics and Space Administration’s (NASA’s) Airborne Visible/Infrared Imaging Spectrometer
(AVIRIS) has 224 narrow, contiguous spectral bands from 0.4 to 2.5 µm (10 nm wide bands) and
a pixel size of 20 m (Vane et al. 1993) although spatial resolution varies based on altitude (Kokaly
et al. 2013). NASA’s satellite-borne Hyperion sensor provides imagery with 220 spectral bands
(10 nm spectral resolution from 0.4 to 2.5 µm) at a spatial resolution of 30 m and swath width of
7.5 km (Figure 5.2; Pearlman et al. 2001). The Hyperspectral Imager for the Coastal Ocean (HICO)
has been flown aboard the international space station since 2009, collecting 128 bands at a spatial
resolution ~95 m (0.38–0.96 µm sampled at 5.7 nm; Oregon State University 2014).
Hyperspectral data are not typically used to map wetland boundaries, probably due to a com-
bination of relatively poor data availability, high spectral resolution (which results in image pro-
cessing challenges that are not necessary when mapping most wetlands), and often insufficient
spatial resolution. That said, it should be noted that hyperspectral data are particularly well suited
for the mapping of SAV, since the water column reduces reflectance from the vegetation itself
often making detailed examination of spectral characteristics necessary to identify SAV (Silva
et al. 2008; Klemas 2011). Generally, hyperspectral data are used to better characterize wetlands.
Current wetland-related applications include the identification of plant species (e.g., invasive plants),
(a) (b)
FIGURE 5.2 August 17, 2007, images derived from E01 Hyperion enhanced for a portion of the eastern
shore of the Chesapeake Bay: (a) False color image (bands 40:31:13) and (b) green color display (bands
31:40:13). Tidal marshes (brownish purple on b) along coastline are easily recognized from inland agriculture
(light green on b) and forests (darker green on b). (Imagery enhanced by MDA Information Systems LLC,
Gaithersburg, MD.)
(Hirano et al. 2003; Schmidt and Skidmore 2003; Klemas 2011; Zhang and Xie 2012), plant mois-
ture stress (Anderson and Perry 1996), plant biochemical properties (e.g., nutrient and chlorophyll
content), (Schmidt and Skidmore 2003; Judd et al. 2007), water quality (Klemas 2011), and the map-
ping of oiled marsh (Leifer et al. 2012; Kokaly et al. 2013). Schmidt and Skidmore (2003) found that
the spectral signatures of different salt marsh vegetation types were significantly different, and they
proposed that these differences should allow identification of individual species using hyperspectral
imagery. Other studies have found that the mapping of various marsh vegetation species is possible
(Judd et al. 2007). Hirano et al. (2003) found higher accuracies when mapping spike rush (100%) as
compared to red mangroves (40%). A review of several research projects involving the use of hyper-
spectral data for mapping aquatic vegetation found that accuracies ranged from 70% to 96% (Silva
et al. 2008). However, it is not always possible to map wetland plant species using hyperspectral
data. In some instances, spectral signatures of neighboring plant species may have similar reflec-
tance values on hyperspectral images making it impossible to separate some distinct vegetation such
as low salt marsh (Spartina alterniflora) from short form of P. australis and S. patens in the New
Jersey Meadowlands in June (Yang and Artigas 2009). Although hyperspectral data, like all optical
data, are limited by the tree canopy during much of the year, hyperspectral data have been used to
detect forested wetlands during the leaf-on period. This can be accomplished by detecting water
stress, because some tree species when exposed to flooding have elevated reflectance in the green
(~550 nm) and near-infrared (~770 nm) portions of the electromagnetic spectrum. This approach has
been shown to be effective in monospecific stands of facultative wetland trees (Anderson and Perry
1996). However, monospecific forests are not commonly found in natural settings.
Similar to the analysis of multispectral data, spectral unmixing methods can be used with hyper-
spectral data to estimate the fraction of endmembers included in each pixel (Roberts et al. 1993;
Rosso et al. 2005; Judd et al. 2007). This type of analysis has the potential to be more precise when
using the numerous bands available with hyperspectral images, although the selection of truly pure
(unmixed) endmembers can be a challenge (Filippi and Jensen 2006; Judd et al. 2007). By estimat-
ing the fraction of plant species per pixel, spectral unmixing methods can be used to produce a
more accurate wetland vegetation map (McCarthy et al. 2005). Rosso et al. (2005) applied spectral
mixture analysis and multiple endmember spectral mixture analysis to AVIRIS data in order to map
marsh vegetation. Both approaches were found to be suitable for mapping marshes, although the
multiple endmember spectral mixture analysis had the advantage of incorporating more than one
endmember per class. Other classification approaches, such as neural networks, have also proved
beneficial. Filippi and Jensen (2006) found that neural networks performed better than endmember-
based approaches. The neural network approach has the additional advantage of not requiring the
initial input of spectral endmembers.
Hyperspectral data provide more spectral information than other sensors and are therefore able
to resolve details that other types of imagery cannot. However, this data characteristic also imparts
certain restrictions. In addition to the restrictions of all optical data, such as an inability to penetrate
vegetation cover, sensitivity to clouds and other weather events, and restricted use to daylight hours,
users of hyperspectral imagery face unique challenges. For example, spectral signatures can vary
temporally with phenology and environmental conditions (Judd et al. 2007; Silva et al. 2008) mak-
ing generalizations difficult and therefore mapping through time and space challenging (Schmidt
and Skidmore 2003). Additional restrictions include large data volume (Phinn et al. 1999; Hirano
et al. 2003; Laba et al. 2005; Belluco et al. 2006; Becker et al. 2007; Klemas 2011), less developed/
more complex image processing techniques (Phinn et al. 1999; Klemas 2013; Hirano et al. 2003;
Laba et al. 2005), and the relatively poor availability of hyperspectral data in general and high price
of commercial data (Klemas 2011). These drawbacks should diminish as techniques are refined and
data availability increases (Phinn et al. 1999; Govender et al. 2007). However, it should be noted
that increased spectral resolution is not always necessary. Belluco et al. (2006) found that the spec-
tral content of hyperspectral data was largely redundant when mapping salt marsh classes and that
comparable maps could be produced using multispectral data.
A number of additional hyperspectral sensors have recently started to collect data or will soon be
in service. These include NASA’s Hyperspectral Thermal Emission Spectrometer, Next-Generation
Airborne Visible/Infrared Imaging Spectrometer, and Portable Remote Imaging Spectrometer.
Although the previously mentioned sensors represent advancements in technology, they are airborne
sensors primarily designed for research and will probably not significantly increase the availability
of hyperspectral data. The successful launch of the Hyperspectral Infrared Imager (HyspIRI), on
the other hand, could significantly increase the availability of hyperspectral data. There are also a
number of commercial hyperspectral airborne sensors including the AisaFENIX (380–2500 nm
with 620 bands; SPECIM 2014) and CASI 1500 (programmable between 365 and 1050 nm with up
to 288 bands; ITRES 2014).
Synthetic Aperture Radar

SAR, a common imaging radar technology, provides the increased spatial resolution that is often
necessary for wetland mapping. SAR data are collected by active sensors that sample the electromag-
netic spectrum at much longer wavelengths than optical sensors (e.g., aerial photography, multispec-
tral, and hyperspectral). For this reason, SARs provide information that is fundamentally different
from sensors that operate in the visible and infrared portions of the electromagnetic spectrum. While
optical sensors respond to variations at the cellular and molecular levels, SARs are sensitive to dif-
ferences in water content, size/roughness, and relatively broad scale structural differences (e.g., tree
branching pattern or buildings). The capabilities of SARs are also unique. Unlike optical sensors,
radar sensors can collect data regardless of solar illumination, cloud cover, and most rain events.
Until recently, only moderate-resolution (30 m) satellite-borne SAR data were available, but more
recent sensors (e.g., RADARSAT-2) are now providing finer-spatial-resolution data (e.g., 3 m).
SAR sensors have different operating parameters, including not only multiple wavelengths but
also multiple polarizations and incident angles (Table 5.2). Microwave wavelengths commonly
used for remote sensing include X-band (2.4–3.8 cm), C-band (3.9–7.5 cm), L-band (15–30 cm),
and P-band (30–100 cm; Jensen 2000). Electromagnetic energy transmitted from the SAR sensor
toward the surface of the Earth is composed of an electric and a magnetic component. These two
components travel, at the speed of light (≈3 × 108 m s−1), orthogonal to one another. The orientation
of the electric component of electromagnetic energy (perpendicular to the direction of travel) deter-
mines the polarization of that energy. Older sensors (e.g., the SAR flown aboard the Japanese Earth
Resource Satellite-1, the SAR flown aboard the European Remote Sensing Satellite-1 [ERS-1], and
Canada’s RADARSAT-1) transmitted and received energy polarized either horizontally (H) or ver-
tically (V), relative to the surface of the Earth. However, newer sensors including RADARSAT-2,
Phased Array type L-band Synthetic Aperture Radar (PALSAR), and COSMO-SkyMed can collect
data in multiple polarizations and may also have fully polarimetric modes, which will be discussed
further later in this chapter. SAR bands are often described by their wavelength (e.g., X, C, L, or P)
and polarization (e.g., HH = horizontally transmitted and received and VV = vertically transmitted
and received). The energy from SAR sensors is transmitted and received at different angles relative
to the Earth’s surface (i.e., incident angles).
It is important to appreciate differences in these instrument specifications because the selection
of optimal specifications must be tailored to different applications to help ensure project success. It is
primarily the wavelength, polarization, and incident angle of the microwave energy in combination
with certain key characteristics of Earth’s surface (dielectric property, size/roughness, and structure)
that determine the amount of energy reflected in the direction of the sensor and therefore received
by the sensor. For example, the scattering exhibited from image elements (e.g., leaves, branches,
trunks, and soil) is a function of wavelength such that landscape components with physical dimen-
sions similar to the incoming wavelength generate the most backscatter (Raney 1998). Different
send and receive polarization combinations are sensitive to different image features. HH-polarized
data have been found to be most useful for the detection of inundation beneath a plant canopy (Hess
et al. 1995), while cross polarized data (i.e., HV or VH) are more sensitive to differences in biomass
(Bourgeau-Chavez et al. 2009). Water content determines the dielectric property of most natural
materials. Typically, the higher the water content, the higher the dielectric constant (a measure of
the aptitude of a substance to conduct electrical energy) of the material and therefore the greater the
amount of incident energy returned from the material (Jensen 2000). Optimal specifications vary
based on the goals of the mapping project and the environment in which it is being carried out.
When mapping and monitoring wetland ecosystems, imaging radars have many advantages
over sensors that operate in the visible and infrared portions of the electromagnetic spectrum.
Microwave energy is sensitive to variations in soil moisture and inundation and is only partially
attenuated by vegetation canopies in areas of lower biomass (Kasischke and Bourgeau-Chavez
1997; Kasischke et al. 1997a,b; Townsend and Walsh 1998; Baghdadi et al. 2001; Townsend 2001,
2002; Rosenqvist et al. 2007; Lang and Kasischke 2008). The longer the wavelength, the greater the
penetration of a canopy (Hess et al. 1990, 1995; Martinez and Le Toan 2007). Although microwave
sensors can currently only be used to estimate soil moisture within a relatively narrow layer of soil
(e.g., 10 cm) at the surface, new models are being developed to help extend this estimate (Li et al.
1998). The sensitivity of microwave energy to water and its ability to penetrate vegetative canopies
make SAR ideal for the detection of hydrologic features below vegetation (Figure 5.3), (Hall 1996;
Kasischke and Bourgeau-Chavez 1997; Kasischke et al. 1997; Phinn et al. 1999; Rao et al. 1999;
FIGURE 5.3 Radar image of Napo River and environs in Ecuador and Peru (March 17, 2013) captured by
NASA’s Uninhabited Aerial Vehicle SAR. The image colors represent the likelihood of inundation (flooding)
beneath the forest canopy, which is difficult to determine using traditional optical sensors. Red and yellow
shades indicate a high likelihood of standing water with emergent vegetation, while blue and green shades are
areas less likely to be inundated, and black is open water. (From NASA/JPL-CalTech.)
Wilson and Rashid 2005; Lang and Kasischke 2008; Lang et al. 2008b). SAR data can even be used
to detect freeze/thaw events (Bartsch et al. 2007) and salinity (Guo et al. 2013), estimate biomass
and other plant structural parameters (Le Toan et al. 1992; Kasischke et al. 1997b; Kellndorfer
et al. 1998; Mougin et al. 1999; Martinez and Le Toan 2007), and detect oil slick thickness (Leifer
et al. 2012).
The presence of standing water interacts with the radar signal differently depending on the domi-
nant vegetation type/structure (Hess et al. 1995) as well as the biomass and condition of vegeta-
tion (Töyrä et al. 2002; Costa and Telmer 2007). When exposed to open water without vegetation,
specular reflection occurs and a dark signal (weak or no return) is observed (Dwivedi et al. 1999),
making the detection of open water relatively simple. The radar signal is often reduced in wetlands
dominated by lower biomass herbaceous vegetation when high water, relative to the height of the
vegetation, is present due largely to specular reflectance (Kasischke et al. 1997a). However, when
the water level is low relative to the vegetation, the backscatter can be enhanced through double-
bounce scattering (Bourgeau-Chavez et al. 2005). In fact, the relative water level can be monitored
remotely through the backscatter signal (Tanis et al. 1994; Bourgeau-Chavez et al. 2005; Kasischke
et al. 2009). Similarly, the radar signal is often increased in forested wetlands when standing water
is present due to the double-bounce effect (Harris and Digby-Arbus 1986; Dwivedi et al. 1999; Lang
et al. 2008b). This occurs in flooded forests when energy that is sent out by the sensor is reflected
strongly by the water surface away from the sensor (specular reflectance) but is then redirected back
toward the sensor by a second reflection from a nearby tree trunk. The use of small incident angles
(closer to nadir) is thought to enhance the ability to map hydrology beneath the forest canopy via
increased penetration of the forest canopy (Hess et al. 1990; Bourgeau-Chavez et al. 2001; Toyra
et al. 2001). However, Lang et al. (2008a) found that the optimum incident angle for inundation
detection varies between forest types and tree phenology (i.e., leaf-on versus leaf-off). Furthermore,
although a general decreasing trend in inundation detection capability was found between incident
angles of 27.5° and 47° when all forest types were considered together, the smallest incident angle
(23.5°) demonstrated detection capabilities at or below the largest angle considered in the study
(47°). Therefore, smaller incident angles may not always be best suited for detecting inundation.
The temporal resolution of available radar data is relatively high for a variety of reasons. Radar
sensors can collect data regardless of solar illumination, cloud cover, and most rain events (shorter-
wavelength radar such as X-band can be sensitive to intense rainstorms). For example, the use of
radar data is particularly helpful in areas of northern Alaska where daylight can be limited and
cloud cover prevalent (Evans et al. 1995; Baghdadi et al. 2001) and the tropics where cloud cover
is frequent (Evans and Costa 2013). The temporal resolution of SAR data is also increased by the
ability of certain satellite-borne sensors (e.g., European Space Agency’s Envisat Advanced SAR
[ASAR], RADARSAT-1, and RADARSAT-2) to collect data at multiple incident angles, thus reduc-
ing the time between satellite overpasses. The availability of multiple well-calibrated satellite-
borne sensors further increases the temporal resolution of SAR data because those data can be
used in concert to gather information about the same area of interest (Martinez and Le Toan 2007;
Evans and Costa 2013). These characteristics allow not only for the detection of wetlands during
key time periods but also for the monitoring of these ecosystems and important functional drivers
through time.
Seasat, launched in 1978, was one of the first imaging radars to be used to study wetlands and
other ecosystems (Place 1985; Pope et al. 1997). Researchers found that the L-HH microwave energy
transmitted by Seasat was particularly sensitive to flooding, even below forest canopies due to the
increase in backscatter caused by double-bounce scattering between tree trunks and the flooded
surface (Krohn et al. 1983; Place 1985; Hess et al. 1990; Pope et al. 1997). In addition to Seasat,
the ability of L-HH SAR to map inundation in forested wetlands has been well documented with a
variety of other sensors (Table 5.2) including AIRSAR (an airborne sensor), Spaceborne Imaging
Radar (SIR)-C, and Japanese Earth Resources Satellite (JERS)-1 (available from 1992 to 1998;
Ormsby et al. 1985; Hess et al. 1995; Townsend and Walsh 1998; Bourgeau-Chavez et al. 2001, 2008;
Martinez and Le Toan 2007).
After the successful launch of ERS-1 (1991–2000), ERS-2 (1995–2011), and Canada’s
RADARSAT-1 (1995–2013), C-band data became increasingly available (Table 5.2) and wetland
studies using these shorter-wavelength data were initiated. Researchers found that although C-HH
band radar data were not as well suited for forested wetland studies as those from L-HH SARs, they
could be used to monitor inundation patterns, especially in areas of lower biomass (Townsend and
Walsh 1998; Townsend 2000; Costa 2004; Lang and Kasischke 2008; Lang et al. 2008b). C-VV data
from the ERS systems have primarily been used to study herbaceous vegetation (e.g., Kasischke
et al. 2003; Grings et al. 2006) but have also been successful in detecting inundation under forest
canopies during the leaf-off period (Kasischke et al. 1997b; Townsend 2002; Lang and Kasischke
2008) and in monitoring soil moisture in low-biomass ecosystems (Bourgeau-Chavez et al. 2007,
2013; Lang and Kasischke 2008).
Although SIR-C/X-SAR (deployed onboard NASA’s Space Shuttle) only collected data dur-
ing the spring and fall of 1994, it provided new details regarding the advantages and limitations
of spaceborne multiwavelength (X-, C-, and L-bands), multipolarization (capable of sending and
receiving multiple polarizations at the same time) SAR (Hall 1996). This advancement led to the
ability to detect finer details of vegetation structure and therefore improved the accuracy of wetland
classifications from space (Pope et al. 1994; Hess et al. 1995; Hall 1996; Smith 1997; Bourgeau-
Chavez et al. 2001). For example, Hess et al. (1995) and Pope et al. (1994) used SIR-C data to identify
various wetland vegetation types including aquatic macrophytes (Hess), marsh (Pope), and swamps
(both) and found that the discrimination of these classes was best when using multiple wavelengths
and polarizations. More recently, SAR sensors have been launched, which offer multiple polariza-
tions (ASAR launched in 2002) and even polarimetric data (PALSAR in 2006, RADARSAT-2 in
2007, and COSMO-SkyMed in 2007). Polarimetric SAR is a quadrature-polarized radar system
consisting of any two orthogonal polarizations in the transmit signal and those same two orthogonal
polarizations (typically H and V) in the receive signal, as long as the amplitude and phase informa-
tion are preserved. SAR polarimetry has potential to provide additional information on the struc-
tural and dielectric (moisture) variation of vegetation and surface roughness across a landscape and
the effects of those parameters on scattering by allowing the scattered signal to be broken down into
polarization state. Polarization state is determined by three parameter amplitudes: HH, HV, and VV
and two phase (φ) differences between them: Φlike = φHH − φVV and φcross = φHH − φHV. This
provides five independent measures that completely define the polarimetric behavior of the image
elements (scatterers). The additional information offered by these sensors has further improved
the utility of SAR data for wetland mapping (Baghdadi et al. 2001; Horritt et al. 2003; Sokol et al.
2004; Touzi et al. 2007; Brisco et al. 2013; Evans and Costa 2013). Other SAR sensors are currently
being planned (see Future Sensors and Missions Relevant to Wetland Mapping later in this chapter).
These extra satellites will not only increase the amount of data available for analysis, they will also
increase collection frequency and the variety of polarizations and frequencies available.
Additional techniques, such as interferometry, can also be used to better characterize the wetland
environment. Interferometric techniques can be used to create extremely accurate DEMs using SAR
data. Differences (i.e., phase shifts) between two or more SAR images collected from slightly dif-
ferent vantage points make the determination of subtle differences in elevation possible. After cali-
bration with elevation from in situ measurements or ancillary data, interferometric SAR (InSAR)
can be used to create DEMs over large areas relatively inexpensively or can be used to monitor
surface deformation related to a number of processes, including groundwater extraction (Meijerink
et al. 2007) that can impact wetland hydroperiod. InSAR can also be used to monitor water levels
of relatively large wetlands and the directional flow of those waters (5 cm accuracy and 1–2 cm
precision; Wdowinski et al. 2008) and wetland vegetation height and biomass (Simard et al. 2006).
Applications of InSAR data have expanded over the past decade due to wider availability, improve-
ment in data and processing methods, and the unique benefits provided by the data.
Although knowledge of radar data characteristics and processing methods have developed con-
siderably, further research is required to utilize the full potential of SAR data for wetland mapping
(Horritt et al. 2003; Costa 2004). The interpretation of SAR data is less intuitive than that of optical
imagery (Silva et al. 2008), and the methods and software used to process SAR data are less devel-
oped than those for optical data. However, Canada’s inclusion of SAR into the operational wetland
mapping process (Milton et al. 2003; Li and Chen 2005; see Canadian Wetland Inventory [CWI]
later in this chapter) indicates that these restrictions are now being overcome within an operational
framework. The Canadian method uses a combination of SAR and multispectral data. Methods that
use a combination of different bands and polarizations or synergistic approaches that utilize imag-
ery from multiple radar instruments as well as optical data often provide superior results because
they bring different pieces of information to the process (Sahagian and Melack 1996; Smith 1997;
Augusteijn and Warrender 1998; Töyrä et al. 2002; Li and Chen 2005; Töyrä and Pietroniro 2005;
Bwangoy et al. 2010; Gala and Melesse 2012; Evans and Costa 2013; Marti-Cardona et al. 2013).
LiDAR Data
Similar to SAR, LiDAR sensors are also active systems, sending and receiving energy produced by
the sensor. LiDAR sensors use energy with much shorter wavelengths (visible and near-infrared)
than radars (microwave). There are multiple types of LiDAR sensors including waveform and more
readily available discrete point return LiDAR. Waveform LiDARs sample the entire laser pulse
return, whereas discrete point LiDAR sensors only record a certain number of returns (i.e., 2–6).
This discussion will focus on discrete point return data that are often available from federal, state,
or local governments and are readily acquired from commercial mapping companies, although it
should be noted that waveform LiDAR data are also commercially available. The vast majority of
LiDAR data used to inform wetland mapping have been discrete point return LiDAR data. LiDAR
sensors emit short pulses of energy, typically from the infrared portion of the electromagnetic spec-
trum, although bathymetric LiDAR sensors use energy from the blue-green portion (e.g., see Collin
et al. 2010; Collin and Planes 2012). Most terrestrial LiDAR sensors operate in the 900–1550 nm
range (Lemmens 2007).
LiDAR data can be used to calculate highly accurate x, y, z locations through the use of onboard
Global Positioning and Inertial Navigation Systems and by calculating the distance to an object
by recording the amount of time it takes for an emitted pulse, or a portion of that pulse, to return
to the sensor (Vierling et al. 2008). This information is often used to create DEMs. DEMs created
using other types of remotely sensed data (i.e., stereo interpretation of aerial photography [InSAR])
are also available, but airborne LiDAR-based DEMs have a finer spatial resolution and greater
vertical accuracy. The spatial resolution and accuracy of non-LiDAR-based DEMs are often insuf-
ficient to map wetlands in areas with relatively little topographic relief (e.g., Gulf–Atlantic Coastal
Plain of the eastern United States), especially when small changes in elevation can change wetland
vulnerability (e.g., sea-level rise Chmura 2013). In general, non-LiDAR-derived DEMs have much
coarser vertical accuracies (1–10 m) than those derived from LiDAR (~10–15 cm; Coren et al.
2006). LiDAR-derived DEMs also have relatively fine horizontal resolution (~100–200 cm; Coren
and Sterzai 2006).
Techniques for incorporating LiDAR-based DEMs into the wetland mapping process range from
the simple use of DEMs to support hand interpretation of wetland boundaries to the use of these
DEMs to create topographic metrics, which are then automatically incorporated into the wetland
mapping process (Figure 5.4). DEMs can be used to derive primary topographic metrics, such as
slope and aspect, and secondary metrics, which are based on the relationship between multiple
primary metrics. One example of a secondary topographic metric is the topographic wetness index
that has been used to determine where surface water is likely to accumulate based on slope and
upslope contributing area (Beven and Kirkby 1979), but multiple topographic metrics are available
and have been used to map wetlands (Murphy et al. 2008; Lang et al. 2012; Leonard et al. 2012).
These other metrics include height above an inferred groundwater table based on the elevation of
water bodies that are assumed to be surface water expressions of groundwater (Murphy et al. 2008)
and relief relative to local topographic maximums within relatively flat landscapes with fairly high
groundwater tables (e.g., mid-Atlantic Coastal Plain), (Lang et al. 2012). Because these metrics
provide information on where wetlands should be located based on topographic position and this
can be modified by artificial drainage and other means, care should be taken when interpreting
these data in altered landscapes. In landscapes where hydrology has not been severely altered,
FIGURE 5.4 Image produced from U.S. Geological Survey LiDAR data showing Carolina Bays (oval-
shaped features) and drainage system for coastal North Carolina. (From N.C. Department of Transportation
data; public domain, not copyrightable.)
LiDAR-derived topographic metrics can provide independent confirmation of wetland location that
is extremely useful for improving detection in difficult to map landscapes, such as forests. LiDAR-
derived topographic information can also be used to map surface hydrologic flow pathways, which
influence the ability of wetlands to provide ecosystem services (e.g., mitigation of nutrient and sedi-
ment pollution).
Although LiDAR DEMs alone can be used to create wetland maps, they are often used in con-
junction with optical images (Lichvar et al. 2006; Vierling et al. 2008), SAR data (Marti-Cardona
et al. 2013), or both (Li and Chen 2005; Töyrä and Pietroniro 2005; Bwangoy et al. 2010; Gala and
Melesse 2012). In this way, landscape position (e.g., slope, depression, or peak) can be made part of
the wetland mapping process. This additional information can aid in the detection of wetlands that
are normally difficult to identify, such as vernal pools (Figure 5.5), (Lichvar et al. 2006; Leonard
et al. 2012), or can be used to help compensate for the lack of spatial detail when using moderate- or
coarse-spatial-resolution images. For example, as mentioned earlier when Civco et al. (2006) had
difficulty using Landsat ETM to separate the upper salt marsh from upland forest due to mixed
pixels, they used elevation data to help distinguish the two classes. It should be noted that LiDAR
data provide information not only on elevation but also on vegetation characteristics. Although
optical data can be used to detect and characterize vegetation, LiDAR data can be used to enhance
this characterization through increased information on vegetation height, biomass, and structure
(Vierling et al. 2008).
While return time provides information on elevation, LiDAR intensity (i.e., the strength of the
returned LiDAR signal relative to the amount of energy transmitted by the sensor per laser pulse;
Chust et al. 2008) provides information regarding the identity of materials reflecting the LiDAR
signal. LiDAR intensity data are particularly well suited for the detection of inundation due to the
strong absorption of the most commonly used LiDAR laser energy (near-infrared) by water relative
to most dry areas. A considerable strength of LiDAR intensity data over other sensors that detect
near-infrared energy is the ability to use the elevation of the LiDAR return to limit analysis to
Probability
Vernal pools 0 125 250 500
Meters High Medium Low
Seasonal swales
(a) (b)
FIGURE 5.5 Maps derived from preliminary field survey (a) and LiDAR (b) showing vernal pools in
north-central California. LiDAR data identified more potential vernal pools than were located during an
initial ground survey and will aid in refining wetland delineations in the future. (From Lichvar, R.W. et al.,
Delineating and evaluating vegetation conditions of vernal pools using spaceborne and airborne remote sens-
ing techniques, U.S. Army Corps of Engineers Engineer Research and Development Center, ERDC\CRREL
TM-06-3, Beale Air Force Base, CA.)
returns from the Earth’s surface instead of the vegetative canopy. Doing so dramatically reduces
mixing of the signal of interest with reflectance from confounding influences (e.g., tree branches).
LiDAR intensity was found to improve the accuracy of inundation mapping within a deciduous for-
est by approximately 30% relative to aerial photography (Lang and McCarty 2009). LiDAR inten-
sity data have also been used to improve the remote identification of different salt marsh vegetation
classes (Collin et al. 2010; Collin and Planes 2012). However, it should be noted that LiDAR inten-
sity data are not currently standardized and cannot be compared between different acquisitions that
utilize varying specifications (Lang and McCarty 2009; Newcomb and Lang 2012). Furthermore,
use of these data is likely to be most advantageous in relatively flat landscapes with deciduous
woody vegetation.
Although airborne LiDAR data are currently only available for about 1/3 of the conterminous
United States, the majority of airborne LiDAR data have been collected for the eastern region,
particularly in coastal areas (Snyder and Lang 2012). This coverage of available LiDAR data is
advantageous for mapping of wetlands because the majority of these ecosystems are found in the
eastern United States due to higher average precipitation to evapotranspiration ratio as compared
to most of the western half of the country. The U.S. Geological Survey (USGS) recently conducted
the National Enhanced Elevation Assessment (NEEA) to (1) determine national requirements for
enhanced elevation data, (2) estimate the costs and benefits of meeting the documented require-
ments, and (3) assess different national elevation program implementation scenarios. As a result of
the NEEA, the USGS has endorsed an implementation scenario focused on the collection of InSAR
data in Alaska and LiDAR data with a horizontal point spacing of 0.7 m and a vertical accuracy of
9.25 cm throughout the rest of the United States (Snyder and Lang 2012). The USGS is currently
working with other federal agencies to develop a funding strategy and governance model to best
assure the collection of the endorsed dataset. The collection and processing of LiDAR data to dif-
ferent specifications have recently been given more attention, and documents now exist to guide the
collection and processing of LiDAR data to appropriate standards (e.g., USGS 2014).
LiDAR data have the potential to greatly improve wetland mapping, both in the research and
operational realms, but like other types of remotely sensed data, certain considerations should be
kept in mind. Similar to SAR, the processing methods for LiDAR vary significantly from those
developed for more traditional optical sensors (e.g., multispectral). This is partly because LiDAR
data may be stored as point clouds, which must then be interpolated to create DEM grids. For this
reason, software has been developed to take full advantage of these point returns, although more
commonly available image processing software can also ingest LiDAR point return data and per-
form more standard operations. Image analysts should be aware that LiDAR data can be collected
using different specifications and that when possible these specifications should be based on their
intended uses. Data collected for one application may not be suitable for another. For example,
higher point densities may be needed for mapping in forested areas of low relief. Point density can
be improved by increasing the number of pulses, flying the platform at a lower altitude, decreas-
ing the beam divergence angle, or simply collecting data over the same area twice (Goodwin et al.
2006). The utility of LiDAR data is also limited in certain vegetation types. For example, Rosso
et al. (2006) found that LiDAR data could not be used to accurately map ground elevation under
dense herbaceous vegetation (i.e., a S. alterniflora–Spartina foliosa hybrid). Hladik et al. (2013)
used hyperspectral data to map S. alterniflora height and other classes and applied elevation correc-
tion factors to improve DEM accuracy.
METHODS TO SUPPORT WETLAND MAP CREATION

Even remotely sensed data that are best suited to create a wetland map are only as good as the
methods used to transform that image or group of images into a map. The methods used to create
a map or categorize pixels into different groups (e.g., swamp versus marsh) are broadly referred
to as image classification. Image classification is one phase of image processing and is generally
conducted after image preprocessing, which includes image radiometric calibration and georegis-
tration. Classification training often relies on the collection of field data, as does the determination
of map accuracy (i.e., validation). Manual interpretation of aerial photographs is briefly discussed
earlier. The following discussions will focus on image classification using digital data and auto-
mated or semiautomated processing and the collection of field data to support map training and
validation.
Image Classification
Supervised classification of digital imagery is one of the two fundamental approaches to semiau-
tomated mapping of land use land cover types, along with unsupervised classification (Piwowar
2005). Supervised classification uses image analyst–defined land cover classes that are extracted
from remote sensing data and used as training data. Ideally, these training sites would be verified in
the field but can also be generated using expert knowledge. The most common supervised classifi-
ers are minimum distance, parallelepiped, and maximum likelihood. Minimum distance classifiers
assign pixels to a particular class based on a pixel value’s proximity to a simple average of each class
type. With parallelepiped, pixel assignment is constrained between the minimum and maximum
ranges of each class types, a range defined by training data. The minimum–maximum space can
be n-dimensional with more than three image bands. Maximum likelihood classifiers compare sig-
nature statistics for each class type, and condition probabilities define membership within a class.
Although slow and more complex than other supervised classifiers, maximum likelihood is gener-
ally considered to produce more accurate classification results and is therefore one of the more pop-
ular methods of land cover classification using remote sensing data (Moreno and De Larriva 2012).
In unsupervised classification, pixels are automatically grouped into statistically similar cat-
egories. Similar statistical groupings are clustered and stratified by class types according to image
analyst interpretation or mathematical correlations. This method is particularly valuable where
preexisting field data are limited or nonexistent (Jensen et al. 1986). “Mixed” classes, where the
statistical distributions of pixel values are overlapping, can impede classification accuracy. When
multiple class types share nonexclusive variation, confusion can arise. Reduction of variation
into discrete and finite map classification schemes provides inherent challenges. For example, a
30 × 30 m Landsat pixel containing half water and half pine forest would not resemble a pure water
or pure pine forest pixel. Using both supervised and unsupervised classifiers in concert is a com-
mon practice.
In complex wetland environments, more traditional statistical classifications commonly used
with multispectral imagery (e.g., supervised and unsupervised classifications) may have limited util-
ity (McCarthy et al. 2005). In these cases and when greater accuracy is required in general, the use
of different classification techniques can improve the accuracy of resultant maps. For example, deci-
sion trees have proven helpful for wetland mapping (Wright and Gallant 2007), partially because
they can be used to incorporate data from a variety of sources (Wright and Gallant 2007; Hladik
et al. 2013), such as SAR and optical data (Li and Chen 2005; Bwangoy et al. 2010) or even different
types of SAR (Baghdadi et al. 2001). Decision trees are a type of machine learning algorithm that
can be used for regression (continuous variable) or classification (categorical variable). A number
of different approaches exist as implemented in different software (e.g., See5, Cubist, and Random
Forests), but all require user-supplied training data consisting of known classes and their attributes.
Decision trees are used in a variety of fields beyond remote sensing, and data inputs can be diverse.
When mapping wetlands, decision tree inputs could consist of passive optical and thermal imagery;
spectral indices (e.g., normalized difference vegetation index [NDVI]); tasseled cap transformation
bands; active remote sensing measurements from SAR and LiDAR; ancillary vector information,
such as soil maps; and even results from supervised or unsupervised classifications. A regression
tree quantifies the relationship between a predictor and response variables, and terminal nodes rep-
resent values. A classification tree is a series of binary decisions that serve to identify the class of
an object (e.g., pixel). The classifier uses training data to develop a series of binary decisions (i.e., a
model) using user-supplied input data. The root node of a decision tree contains all the training data
and all the possible classes; at each node, the best split is sought in order to separate the data into the
correct classes. The data are split until ideally terminal nodes contain only one class. Postprocessing
steps, like pruning or the reduction of classification tree nodes, are commonly used to improve
model results. Once this process is complete, the model can then be applied to the larger dataset to
map the remaining area. The advantages of decision trees include their ability to work with a wide
variety of input data including combinations of binary and continuous data, data that are not nor-
mally distributed, data with different ranges or resolution, and data with missing values or outliers.
Furthermore, decision splits can be presented as outputs for analysis, thus allowing for interpreta-
tion and logic testing. However, the models can be overfit with training data causing poor results
when applied to the full dataset. Models can be improved through various postprocessing steps.
Random Forests (Breiman 2001) is a machine learning algorithm that takes advantage of an
ensemble of decision trees. Other ensemble tree classifiers include boosting where trees are suc-
cessively weighted in order to improve weaker classifiers (Schapire et al. 1998) and bagging where
sets of trees are created from an independent “bootstrap” sample of training data (Breiman 1996;
Bwangoy et al. 2010). When using Landsat ETM+ images to identify wetlands, a Stochastic Gradient
Boosting decision tree technique was found to improve classification accuracy (Baker et al. 2006).
Random Forests adds an additional layer of randomness to bagging by using a random selection
of attributes for creating splits. Once the forest is created, a pixel’s classification is determined by
which class receives the most “votes” from each decision tree. The performance of individual bands
or attributes in forest creation is determined when individual trees are created. The training data not
selected for use in the creation of an individual tree are permuted one band at a time; the larger the
misclassification error is due to the permutation, the more important the band. Attribute importance
is also ranked based on the band’s ability to increase pixel purity after a spilt (Liaw and Wiener
2002). The strengths of the Random Forests algorithm include handling datasets with a small num-
ber of observations and a large number of attributes; it is more powerful than a single classifier, well
suited to parallel processing, and insensitive to nonpredictive inputs. Additionally, the algorithm can
easily handle missing attributes (e.g., cloud cover) as a subset of the decision trees that were built
without the use of the missing attributes can be used to classify the compromised data. The main
drawback to the algorithm is it is not intuitive, and the model is very sensitive to inaccurate training
data (Liaw and Wiener 2002). The Random Forests algorithm is a popular option for mapping in
a variety of wetland environments (e.g., coastal dune and salt marsh; Timm and McGarigal 2012).
A number of relatively simple processing steps can be used to produce inputs for decision trees
or create wetland maps directly. These procedures include thresholding using one input band and
the creation of ratios using multiple input bands. Thresholding is a simple yet powerful technique
that works best when classes occupy distinct ranges of image values. For example, land and water
classes can easily be separated using a single split in an infrared band (Lillesand et al. 2004), since
water strongly absorbs infrared energy and thus typically has much lower infrared reflectance than
land. Ratios are created by dividing one spectral band with another. This technique allows for
spectral variation between classes to remain distinct even when illumination varies due to shadow-
ing or variations in sun angle. However, a simple ratio of bands can only compensate for illumina-
tion effects that influence both bands equally. One of the most commonly used ratios is the NDVI
(Tucker 1979). NDVI values can range from −1 to 1; the higher values indicate higher concentra-
tions of green vegetation. The use of indices, such as the NDVI or transformations, for example,
the tasseled cap transformation, may improve the discrimination of different wetland types (Sader
et al. 1995; McCarthy et al. 2005; Wright and Gallant 2007; Lee and Yeh 2009; Poulin et al. 2010;
Davranche et al. 2013; Petus et al. 2013).
The traditional methods of classifying remote sensing images (see text on the aforementioned
supervised and unsupervised classifications) are based upon statistical classification of single
pixels in a single digital image (Lillesand et al. 2004). Recent research indicates that pixel-based
classification methods may be less than optimal for some wetland mapping applications since they
do not consider the spatial relationships of landscape features (Dronova et al. 2011; Ouyang et al.
2011). Object-based classification has been identified as a method of incorporating spatial context
into the classification process. This approach complements a principle of landscape ecology that it
is preferable to work with a meaningful object representing spatial patterns rather than a single pixel
(Blaschke and Strobl 2001). It also better mimics human interpretation by identifying objects with
similar spatial and spectral characteristics (Hay and Castilla 2006). Object-based approaches incor-
porate two steps: segmentation and classification. In the segmentation phase, homogeneous image
objects are derived from both spectral and spatial information. In the classification phase, image
objects are classified using established classification algorithms, knowledge-based approaches, or a
combination of classification methods (e.g., see the aforementioned decision trees; Civco et al. 2002;
Powers et al. 2012; Zhang and Xie 2012). Approaches that incorporate object-based classification
have been found to enhance wetland mapping (Dronova et al. 2011; Ouyang et al. 2011; Zhang
and Xie 2012). Trimble eCognition (formerly Definiens Developer) is a popular software package
that supports object-based classification. In Trimble eCognition, image objects are created using
user-defined scale and homogeneity parameters. The scale parameter determines the size of image
objects and how different objects are from one another. The homogeneity criteria are defined by the
spectral value of the image and shape of image objects. Shape is further defined by the smoothness
of object borders and compactness of the resulting image objects. For more information on object-
based image analysis, please see Chapter 9.
Collection of In Situ Data

The accuracy of remote sensing products varies with respect to the classification methodology,
inherent sensor limitations, and the desired product utility. In addition to understanding the interac-
tion between the sensor and the target, an accurate characterization of the target is important and
often best achieved through direct field sampling. Field data, ranging from unique observations or
specimen collections to periodic sampling, provide an invaluable calibration and validation stan-
dard. Biological surveys of species occurrence, distribution, density, and diversity all contribute to
community description and can be applied to map classifications. Knowing what types of plants
exist in a particular area allows for a correlation to be made between spectral signatures and specific
land cover types. Additionally, spectral measurements made on the ground with a spectrometer may
provide a site-specific correlation with reflectance. However, issues of scale may obscure the spec-
tral reflectance recorded by an airborne sensor in comparison to a handheld sensor. For example, the
collective “green” of a monotypic stand of a wetland plant may differ from the spectral reflectance
or “green” of a single leaf or multiple randomly selected leaves from that same patch. Yet, observ-
ing differences in chlorophyll content by species in combination with structural variation between
species is an integral aspect of many wetland remote sensing applications.
Another useful suite of field measurements includes localized hydrology. Variations in soil mois-
ture or levels of standing water are readily detected by radar, and these parameters can be verified
on the ground with the use of soil moisture probes or visual observations of inundation depth and
extent through time. Furthermore, differing amounts of available water have considerable impact on
the spectral signatures of plant species, related to cellular pigmentation and patterns of emergence
and senescence. In combination, seasonal changes in wetland hydrology result in changes to plants
and soils that are detectable by both active and passive sensors. Quantification of soil moisture and
standing water depth should be considered as a part of a thorough field sampling protocol, espe-
cially since hydroperiod is a primary driver of wetland functions.
Finally, when collecting field data, it is important for measurements to be scaled appropriately
to the resolution of the sensor and the MMU. Whether point locations or 2D polygons are measured
in the field, the spatial location data and associated biophysical characteristics need to distinguish
variation in wetland types, both gradual and discrete. All field data should ascribe biological and
geophysical attributes of a representative area corresponding to the MMU. Any variation or change
in wetland phenomena that occurs in areas smaller than the MMU will likely go undetected from
a remote sensing standpoint, because the MMU is determined by sensor resolution. While it is still
important that fine-scale features and site-specific abnormalities be well documented in field notes,
in general, field data should describe biological and physical features of interest at a meaningful
spatial scale relative to sensor limitations. There are benefits and limitations associated with the col-
lection of field data, but such data both provide a means to train remote sensing classification models
and most importantly provide a reference to gauge classification accuracy and error.
CURRENT WETLAND MAPPING PROGRAMS

Both government and private groups have produced wetland maps, but most wetland maps are
produced by federal or state governments (see Chapter 3 and this section) due to the significant
amount of resources and expertise necessary to do so. Within the United States, cartographers
have been using a combination of remotely sensed imagery and field data to produce wetland
maps for natural resource management since the 1950s (see Chapter 4). Two of the world’s most
comprehensive wetland mapping efforts the U.S. FWS’s National Wetlands Inventory (NWI)
and the National Oceanic and Atmospheric Administration’s Coastal Change Analysis Program
(C-CAP) are ongoing within the United States, but wetland mapping is also being conducted by
other countries, including Canada. The NWI, C-CAP, and the CWI employ unique datasets and
methods to achieve program goals. The following text highlights their efforts in an attempt to
illustrate not only the different strengths of applied methods but also the influence of program
goals on selected methods.
U.S. Fish and Wildlife Service’s National Wetlands Inventory

One of the earliest and most commonly relied upon wetland mapping efforts was the NWI. This
wetland mapping and monitoring effort was initiated in 1974. NWI maps are created to provide nat-
ural resource managers with information on distribution and types of wetlands necessary to make
well-informed decisions regarding the conservation of wetlands. NWI maps are used for a variety
of applications but are often used at the field scale to help make decisions regarding specific parcels
of land and have been used by the Army Corps of Engineers and some states to support regulatory
decision making (Tiner 1999). NWI maps have relatively detailed wetland classes, usually based
on vegetation, hydrologic regime, and salinity and deepwater type based on water regime, ecologic
system, and the shape and location of wetlands (see Chapters 2 and 3). NWI maps are primarily
produced using aerial photographs, photointerpretation techniques, field verification, and some col-
lateral data sources (Federal Geographic Data Committee 1994). Interpretation of images was origi-
nally conducted by viewing stereopairs of aerial photographs and delineating wetland boundaries
and classification on acetate overlays by hand with pen and ink. More recent methods involve digi-
tizing wetland boundaries using GIS software. The TMU, or smallest wetland that is consistently
mapped, by NWI varies based on the types of imagery used (e.g., scale of aerial photography) and
the type of wetland being mapped. However, NWI TMUs are known to vary between 0.2 and 2 ha
(see Tiner 2009 for additional information on NWI methods; US FWS 2014).
Although great care has been taken in the production of these maps, challenges to the carto-
graphic process, and product delivery remain. Producing relatively detailed wetland maps using
manual or semiautomated techniques is time consuming and resources intensive. For this reason,
NWI maps are not currently available for the entire United States although they have been updated
in some areas of rapid change. Although NWI maps are generally considered to be the most accu-
rate wetland maps available in the United States, some wetland types are mapped better than others.
Mapping is most accurate when there is a distinct change between vegetation, hydrology, and soil
at the wetland boundary (National Research Council 1995). Other wetland types, such as evergreen
forested wetlands, temporarily flooded wetlands, and seasonally saturated wetlands, are mapped
more conservatively (Tiner 1997, 2003; Kudray and Gale 2000).
Although the U.S. FWS continues to primarily use aerial photography to create and update
NWI, it is also exploring the use of different types of remotely sensed data and more automated
techniques. The U.S. FWS has adopted some technological advances since the inception of NWI,
including the use of finer-resolution imagery, digital instead of analog imagery, computer-based
digitization instead of pen-and-ink drawing on acetate overlays, refinements to the Cowardin clas-
sification system (Cowardin et al. 1979), and, in a few cases, feature recognition software to semi-
automate boundary delineation and the use of LiDAR as a supplementary base map. The NWI has
also enhanced its classification of wetlands by adding hydrogeomorphic descriptors to the geospa-
tial databases for a number of special project areas across the country (see Chapter 2 for details).
The expanded database—NWI+ data—is used to predict wetland functions at the landscape level
with the results displayed via an online mapper at the “Wetlands One-Stop Mapping” Association
of State Wetland Managers 2014 (e.g., see Chapter 3). The U.S. FWS is presently investigating the
use of additional tools and techniques, including the incorporation of LiDAR-derived topographic
metrics and SAR into their mapping process (e.g., Chapter 15).
Additional data and advanced remote sensing techniques could further refine the NWI meth-
odology in the future. The spatial detail needed to meet program goals demands the use of fine-
spatial-resolution data, such as aerial photography, and these optical data also meet the need for
relatively detailed classes based on plant type (e.g., evergreen versus deciduous and woody versus
herbaceous). However, fine-resolution multispectral data, data that can be used to distinguish vary-
ing levels of actual or potential wetness (e.g., SAR and LiDAR), and the use of enhanced classifica-
tion algorithms (e.g., machine learning algorithms) could retain the spatial resolution of NWI data
while increasing accuracy and decreasing production time. Whether or not these new approaches
will be adopted depends on a number of variables including the availability of funding, technical
expertise, and data and the acceptability of this new product.
NOAA Coastal Change Analysis Program

The NOAA C-CAP is a nationally standardized database of land cover and habitat change produced
for the U.S. coastal region. This regularly updated database was established in the 1990s to assist
scientists, managers, and regulators in better understanding the coastal environment, the interac-
tions between major land cover types (e.g., wetlands and uplands), and the impact of those interac-
tions on marine organisms (Dobson et al. 1995). C-CAP products from different years are compared
to determine the amount, rate, and types of changes to land cover, which have occurred between
map dates. By doing so, C-CAP can be used to assess cumulative impacts of land cover change on
ecosystem health (e.g., water quality and habitat quality).
C-CAP maps include the same upland land cover classes that are mapped by the National Land
Cover Dataset (NLCD; http://www.mrlc.gov) in the interior United States but also display more
wetland classes: palustrine forested, palustrine scrub/shrub, palustrine emergent, estuarine forested,
estuarine scrub/shrub, and estuarine emergent wetlands, as well as palustrine aquatic beds and estu-
arine aquatic beds. In this way, C-CAP contains more detailed wetland classes than NLCD, but it is
less detailed than NWI, providing less information about hydrologic regime and other parameters.
C-CAP maps are created in a digital environment using image processing software and
advanced machine learning classification algorithms. The mapping process primarily relies on
digital satellite imagery, field data, and other ancillary data, including NWI. C-CAP primarily
analyzes Landsat 5 and 7 images, although additional satellite imagery is used in some regions,
including SPOT, IKONOS, and QuickBird (Dobson et al. 1995). C-CAP is expected to use
Landsat 8 in the future. The finer-resolution imagery, such as IKONOS and QuickBird, is used
to gather more detailed information for training or in areas of enhanced interest (e.g., areas of
rapid change), (Nate Herold, personal communication 2008). C-CAP maps have been produced
for multiple years, including 1985, 1992, 1996, 2001, 2006, and 2001 (availability varies by loca-
tion), at a 30 m spatial resolution.
To conserve resources and encourage map consistency between dates, C-CAP mapping is
focused on dynamic areas where change is to be expected. NOAA uses various change products
to determine areas where maps need to be updated, and once these areas are identified, map-
ping is focused within these areas, while unchanged areas remain the same. One of the methods
used to identify change areas is cross correlation analysis (CCA), a technique developed to detect
changes in land cover using multispectral satellite data (Koeln and Bissonnette 1999). Houhoulis
and Michener (2000) and Nielsen et al. (2008) have developed similar processes that can also be
used to update wetland maps although these particular methods are not currently used by C-CAP.
An important advantage to these methods is that coarser-resolution digital data can be used to
rapidly screen for changes in areas that were originally mapped with more costly, finer-resolution
imagery. Data summaries, reports, and digital maps are available online (e.g., see Chapter 3)
(NOAA 2014).
Owing to their varying input datasets and production methods, C-CAP and NWI maps are unique
in terms of spatial resolution, mapped classes, and update interval. As suggested by the divergent
objectives of these two programs, they are also better suited for different applications. While NWI
maps offer more spatial and map class detail and are generally considered to be more accurate at
the time of mapping, C-CAP provides a more temporally consistent, regularly updated product.
However, these datasets are not independent of one another since NWI data are often used as an
ancillary dataset to support the C-CAP mapping process. For this reason, it is not surprising that
these two maps exhibited similar accuracy levels during a 1991 field verification campaign. Random
locations were examined and NWI was found to be 88%–100% accurate, while C-CAP was found
to be 63%–97% accurate (Burgess 1995). Overall, NWI tends to be more conservative and has more
omission errors, while C-CAP has more commission errors (Shapiro 1995).
Canadian Wetland Inventory

The objective of the CWI is to provide a national wetland map that can be used for natural
resource management to conserve and protect critical wetlands and provide societal benefits.
The CWI is currently under development and represents the Canadian counterpart to the U.S.
NWI (see Chapter 6). While Canada has roughly a quarter of the Earth’s wetlands (an estimated
1.5 million km2), a nationwide inventory has not been previously developed (Reimer 2009).
Methodologies are being developed to create a consistent map for all areas of Canada using pri-
marily satellite imagery with an MMU of 1 ha (Fournier et al. 2007).
Advanced remote sensing techniques have been developed that take advantage of multiple data
sources and object-based image processing (Grenier et al. 2007). Since certain regions of Canada,
such as the Prairie ecozone, the Maritime provinces, and portions of British Columbia’s Lower
Mainland and Interior have many small wetlands in high densities, mapping in these areas needs
to be based on fine-spatial-resolution imagery (Reimer 2009). This applies to approximately 7%
of Canada where aerial photography is being used to accurately identify wetlands down to 0.2 ha.
This level of accuracy is important because over 80% of the wetlands in the Prairie ecozone are less
than 1 ha in size (Reimer 2009). An estimated 93% of Canada will be mapped at medium resolu-
tion, similar to C-CAP. Medium-resolution mapping relies on satellite imagery such as Landsat
(optical) and RADARSAT (SAR) and will have an MMU of 1 ha. The combination of medium-
and fine-resolution methods is the most cost-effective approach to mapping the entire nation. This
approach will produce a standardized CWI that will provide consistent wetland information to
support conservation, management, and decision-making needs. At the time this chapter was writ-
ten, a significant portion of Canada’s wetlands had been mapped or were in the process of being
mapped. However, these data were not yet available to the public.
FUTURE SENSORS AND MISSIONS RELEVANT TO WETLAND MAPPING

The future holds promise for an expanded wetland mapping toolbox with greater availability of
currently used remotely sensed data, as well as new datasets. The deployment of additional, often
enhanced, satellite and airborne sensors by the U.S. government, international governments, and
commercial entities will increase wetland mapping capabilities. The deployment of a range of sen-
sor types is being planned, including new multispectral, hyperspectral, and SAR instruments. The
spatial and temporal resolution of these new sensors is generally finer, while other technical capa-
bilities are improving or remaining unchanged (Tables 5.1 and 5.2).
Irrespective of other sensor improvements, enhancement of spatial and temporal resolution
should significantly influence the quality and detail of future wetland maps. The simultaneous
advancement of spatial and temporal resolution is a laudable achievement. In the past, the spatial
or temporal resolution of many sensors improved independently. It is only recently that both have
become possible through advanced technologies, such as the deployment of multiple microsatellites,
multiple view angle technology, or simply the deployment of more fine-resolution sensors. These
advancements are important due to the dynamic nature of wetland hydroperiod and the fine spatial
scale of some wetlands.
Although aerial photography has long provided relatively fine-spatial-resolution imagery, more
recent multispectral and even SAR and LiDAR data are now available with a similar spatial
resolution (Tables 5.1 and 5.2). Additional enhanced fine-resolution multispectral sensors are
currently being developed. For example, GeoEye-2 will provide 0.34–1.36 m resolution multi-
spectral data (http://www.satimagingcorp.com/satellite-sensors/geoeye-2/). DigitalGlobe is devel-
oping WorldView-3 that will collect 29-band data with spatial resolutions between 0.31 and 30 m
and a one-day repeat time. At the time this chapter was written, WorldView-3 was scheduled
for launch in mid-2014 (http://www.satimagingcorp.com/satellite-sensors/worldview-3/). Various
SAR sensors are currently planned including RADARSAT Constellation that will collect 3–100 m
resolution C-band data with multiple polarizations and a repeat cycle of 12 days (Canadian Space
Agency 2014). The PALSAR-2 sensor that will be aboard the Japan Aerospace Exploration
Agency’s Advanced Land Observing Satellite-2 (ALOS-2 planned for launch in 2014) will be
an advanced L-band SAR capable of collecting data with a spatial resolution as fine as 1 m
with a revisit time of 14 days (Japan Aerospace Exploration Agency 2014). Tandem-L is a mis-
sion concept currently supported by the Deutsches Zentrum für Luft- und Raumfahrt (DLR),
Germany’s national aeronautics and space research center. Tandem-L would consist of two L-band
SAR sensors to provide fine-spatial-resolution interferometric data for the monitoring of the soil
moisture and plant structure, among other things (http://www.dlr.de/hr/en/desktopdefault.aspx/
tabid-8113/14171_read-35837/).
In the United States, NASA is working toward the launch of multiple SAR sensors, including
the Deformation, Ecosystem Structure, and Dynamics of Ice Radar only (DESDynI-R) and Surface
Water and Ocean Topography (SWOT). SWOT is a collaborative effort between NASA and the
French Space Agency (CNES), which is, in part, designed to inventory freshwater storage and change
in storage in water bodies greater than 250 m2, including wetlands. The launch of SWOT is planned
for 2020 and it is expected to contain a ka-band radar interferometer and a nadir-looking altim-
eter (https://swot.jpl.nasa.gov/files/swot/SWOT_science_reqs_release2_v1.14.pdf). The DESDynI-R
sensor is proposed to contain a multiple polarization, L-band SAR capable of collecting ~10 m
resolution data with a 12–16-day repeat time, for repeat pass interferometry. Originally planned
as a LiDAR and Radar mission, in 2012, NASA reformulated the DESDynI mission as an L-Band
SAR only to meet reduced budget guidelines. The DESDynI-R mission will use radar to measure
the deformation of the Earth, study change occurring in the polar ice sheets, and characterize
global ecosystems and is scheduled for launch in ~2021 (Paul Siqueira 2013). These details and
more regarding the sensors discussed earlier can be found in the U.S. National Research Council
(2007) publication entitled “Earth Science and Applications from Space: National Imperatives for
the Next Decade and Beyond.” In addition, a feasibility study is underway for the joint development
of a satellite equipped with InSARs working in two frequency bands (L-band and S-band) by NASA
and ISRO (NI-SAR) for launch in 2020. The S-band radar will be made by ISRO and the L-band by
Jet Propulsion Laboratory (JPL) (http://eospso.gsfc.nasa.gov/missions/nasa-isro-synthetic-aperture-
radar). The HyspIRI has also been recommended for development by NASA (National Research
Council 2007) and is currently proposed for launch in the 2020s. The HyspIRI mission includes two
sensors: a visible shortwave infrared (VSWIR) imaging spectrometer operating between 0.38 and
2.5 µm in 10 nm contiguous bands with a swath width of 145 km and a TIR multispectral scanner
operating between 4 and 12 µm with a swath width of 600 km. Both sensors have a proposed spa-
tial resolution of 60 m. The larger spatial extent of the TIR instrument will allow for a revisit time
of 5 days, while the smaller spatial extent of the VSWIR instrument will provide a revisit time of
19 days. Among other applications, HyspIRI will be used to identify plant communities and plant
stress (http://hyspiri.jpl.nasa.gov/).
CONCLUSION
Newer technologies (e.g., hyperspectral, SAR, and LiDAR) have and will continue to significantly
advance wetland mapping capabilities. Hyperspectral data can be used to identify wetland patches
that are spectrally indiscernible using multispectral data and are often better at identifying indi-
vidual plant species (e.g., invasive plants) than multispectral data. SAR data can be used to reveal
subtle patterns in inundation and soil moisture, which indicate the presence of wetlands that are
normally difficult to identify (e.g., forested wetlands). The ability of SARs to monitor hydroperiod
is important due to the strong influence of hydroperiod on biota (e.g., provision of habitat to rare
or endangered species), biogeochemical processes (e.g., denitrification), and other ecosystem func-
tions. LiDAR data can be used to locate low-lying areas that often support wetlands, especially
when they have a large upland contributing area. They can also be used to map surface hydrologic
flow pathways, which influence the ability of wetlands to provide valuable ecosystem services (e.g.,
water quality).
There is not a single source of remotely sensed data that is best for mapping wetlands under
all circumstances. Ideally, image analysts should be well versed in the advantages and disadvan-
tages of applying not only different types of imagery (e.g., aerial photography, multispectral, radar,
and LiDAR) but also imagery collected with varying specifications, including wavelengths, spa-
tial resolution, angles of incidence, and repeat times. This is because different types of remotely
sensed data are uniquely suited for detecting specific components of the landscape (e.g., inundation,
soil moisture, elevation, plant biomass, structure, and cellular characteristics) during different time
periods, and the importance of these landscape components varies with wetland type, project goal,
landscape, and more. If more than one type of imagery is needed to detect landscape components,
these datasets can be synthesized during the map production process to produce a superior wetland
map. This data fusion is aimed at reducing classification error by incorporating more information,
whether it be spatial, spectral, temporal, or otherwise. For example, since radar and optical data
are sensitive to very different landscape characteristics, the combination of radar and optical data
can significantly improve wetland mapping (Lyon and McCarthy 1995; Sahagian and Melack 1996;
Kushwaha et al. 2000; Töyrä et al. 2002; Silva et al. 2008; Bwangoy et al. 2010; Chapter 15 of this
book). When using one type of imagery, data from multiple seasons allow improved discrimination
of wetland types via enhanced information on seasonal variations due to phenology and hydro-
logic variations. Even the fusion of finer- and coarser-spatial-resolution imagery of the same type
collected at the same time can improve mapping results. This is often desirable when the coarser-
resolution dataset is more robust spectrally.
In the past, researchers have debated as to whether spatial or spectral resolution is more impor-
tant to the mapping of wetlands. It now appears from available research that a combination of spatial
and spectral resolutions is needed to map wetlands with remotely sensed data since varying wetland
environments necessitate the use of different types of data. For example, fine-spatial-resolution data
are particularly important when mapping salt marshes (Belluco et al. 2006; Ouyang et al. 2011;
Ashraf et al. 2012) or oil-affected shorelines (Kokaly et al. 2013). As the spatial resolution of imag-
ery used to map wetlands becomes finer, wetland boundaries become more distinct and refined and
smaller wetlands may be mapped. Of course, the required spatial resolution depends on the size of
the wetland of interest. As a general rule, it takes an area three by three pixels wide to identify an
object on the ground. The more similar a cover type is to adjacent cover types, the larger the area
that will be needed to distinguish that object (Federal Geographic Data Committee 1992). Although
a minimum spatial resolution is needed to map wetlands, increased spatial resolution is only useful
to a certain point (e.g., Powers et al. 2012), and spectral information is also vital, especially when
mapping wetlands with indistinct spectral signatures (e.g., some types of forested wetlands). It is the
optimal combination of spatial and spectral resolutions that is necessary to best map wetlands, and
this combination varies between wetland types, time of year, and project goals.
Temporal resolution is also vital to the advancement of wetland mapping. Multitemporal imag-
ery can improve wetland mapping capabilities (Töyrä et al. 2001; Li and Chen 2005), especially
when images are collected during different seasons and moisture regimes (Kushwaha et al. 2000).
Wetlands are inherently dynamic systems and multitemporal data are often needed to detect these
changes. Changes in plant phenology (Silva et al. 2008; Chapter 15 of this book) and biomass and
changes in soil moisture and flooding patterns are seen throughout the year (Lyon and McCarthy
1995). Multitemporal SAR data can be used to map hydroperiod as it changes in response to sea-
sonal variations in weather and phenology and periods of extreme weather (i.e., floods and droughts)
(Lang et al. 2008b). These hydroperiod maps cannot only be used to infer wetland function (e.g.,
denitrification); they can also be used to update wetland boundaries as they shift in response to
climate and land cover change and to identify lands that are transitional between wetlands and
uplands. It should also be noted that the collection of imagery during drought, flood, or normal
years can have a large impact on wetland mapping outcomes (e.g., extent and number of wetlands
mapped) and ideal time of collection depends on the ultimate goal of the mapping project.
Selection of the appropriate type(s) of imagery is vital to enhancing wetland mapping, but the
addition of ancillary data (Sader et al. 1995; Li and Chen 2005) and the use of GIS and hydro-
logic models can also greatly benefit the mapping process. Ancillary datasets that are beneficial
to the wetland mapping process include (but are not limited to) information on soils, water bodies,
topography, current and prior land use, tides, and weather. (A note of caution: any inherent error
in the ancillary data may be propagated in the new map products.) Decision support systems (e.g.,
decision trees, GIS, and hydrologic models) can then be used to extract valuable information from
these ancillary datasets in combination with remotely sensed data. Doren et al. (1999) hypothesized
that in the future, a combination of aerial photography, rapidly advancing satellite imagery (e.g.,
hyperspectral data and fine-spatial-resolution multispectral data), and GIS will allow the detailed
mapping of wetland vegetation comparable to those created with detailed ground surveys. Others
offer a more cautious predication. Cowardin and Golet (1995) argued that although remotely sensed
data in conjunction with GIS and modeling hold promise for improved wetland mapping, wetland
maps will never be perfect as they are an attempt to put artificial boundaries on natural gradations.
Rocchini et al. (2013) add that this forcing of continuous data into discrete classes also leads to the
loss of information and increased uncertainty. This ecological reality can never be fully reconciled,
but wetland scientists can develop improved techniques for dealing with this reality and improving
map accuracy.
ACKNOWLEDGMENTS
We acknowledge all those who have provided contributions to this chapter including Colin Brooks,
Michael Battaglia, Mary Ellen Miller, Anthony Landon, Richard Powell, and Greg McCarty.
Thanks also to Lindsey Lefebvre and Robert Lichvar (U.S. Army Corps of Engineers) and Michael
O’Brien (MDA Information Systems LLC) for providing figures for use in this book. Support
for the production of this chapter was provided by the U.S. Department of Agriculture Natural
Resources Conservation Service Wetlands Component of the Conservation Effects Assessment
Project National Assessment.
REFERENCES
Abrams, M., S. Hook, and B. Ramachandran. 2012. ASTER User Handbook Version 2. Jet Propulsion
Laboratory.
Airbus Defense & Space. http://www.astrium-geo.com/en/143-spot-satellite-imagery (Accessed May 12,
2014).
Alsdorf, E., E. Rodríguez, and D. Lettenmaier. 2007. Measuring surface water from space. Reviews of
Geophysics 45(2):RG2002, doi:10.1029/2006RG000197.
Anderson, J. and J. Perry. 1996. Characterization of wetland plant stress using leaf spectral reflectance:
Implications for wetland remote sensing. Wetlands 16:477–487.
Ashraf, S., L. Brabyn, and B.J. Hicks. 2012. Image data fusion for the remote sensing of freshwater environ-
ments. Applied Geography 32(2):619–628.
Association of State Wetland Managers. http://aswm.org/wetland-science/wetlands-one-stop-mapping
(Accessed May 12, 2014).
Augusteijn, M.F. and C.E. Warrender. 1998. Wetland classification using optical and radar data and neural
network classification. International Journal of Remote Sensing 19:1545–1560.
Ausseil, A., J.R. Dymond, and J.D. Shepherd. 2007. Rapid mapping and prioritization of wetland sites in the
Manawatu-Wanganui region, New Zealand. Environmental Management 39:316–325.
Baghdadi, N., M. Bernier, R. Gauthier, and I. Neeson. 2001. Evaluation of C-band SAR data for wetlands map-
ping. International Journal of Remote Sensing 22:71–88.
Baker, C., R. Lawrence, C. Montagne, and D. Patten. 2006. Mapping wetlands and riparian areas using Landsat
ETM+ imagery and decision-tree-based models. Wetlands 26:465–474.
Bartsch, A., R.A. Kidd, C. Pathe, K. Scipal, and W. Wagner. 2007. Satellite radar imagery for monitoring
inland wetlands in boreal and sub-arctic environments. Aquatic Conservation: Marine and Freshwater
Ecosystems 17:305–317.
Becker, B.L., D.P. Lusch, and J. Qi. 2007. A classification-based assessment of the optimal spectral and
spatial resolutions for Great Lakes coastal wetland imagery. Remote Sensing of the Environment
108:111–120.
Beeri, O. and R.I. Phillips. 2007. Tracking palustrine water seasonal and annual variability in agricultural wet-
land landscapes using Landsat from 1997 to 2005. Global Change Biology 13:897–912.
Belluco, E., M. Camuffo, S. Ferrari, L. Modenese, S. Silvestri, A. Marani, and M. Marani. 2006. Mapping salt-
marsh vegetation by multispectral and hyperspectral remote sensing. Remote Sensing of the Environment
105(1):54–67.
Bennett, M.W. 1987. Rapid monitoring of wetland water status using density slicing. Proceedings of the Fourth
Australasian Remote Sensing Conference, Adelaide, South Australia, Australia, Vol. 14–18, pp. 682–691.
Beven, K.J. and M.J. Kirkby. 1979. A physically based, variable contributing area model of basin hydrology.
Hydrological Sciences Journal 24:43–69.
Blaschke, T. and J. Strobl. 2001. What’s wrong with pixels? Some recent developments interfacing remote
sensing and GIS. Zeitschrift für Geoinformationssysteme 6:12–17.
Bolstad, P.V. and T.M. Lillesand. 1992. Improved classification of forest vegetation in northern Wisconsin
through a rule-based combination of soils, terrain, and Landsat Thematic Mapper Data. Forest Science
38:5–20.
Bourgeau-Chavez, L.L., E.S. Kasischke, S.M. Brunzell, J.P. Mudd, K.B. Smith, and A.L. Frick. 2001. Analysis
of space-borne SAR data for wetland mapping in Virginia riparian ecosystems. International Journal of
Bourgeau-Chavez, L.L., E.S. Kasischke, K. Riordan, S.M. Brunzell, E. Hyer, M. Nolan, M. Medvecz, and
S. Ames. 2007. Remote monitoring of spatial and temporal surface soil moisture in fire disturbed boreal
forest ecosystems with ERS SAR imagery. International Journal of Remote Sensing 28(10):2133–2162.
Bourgeau-Chavez, L.L., B. Leblon, F. Charbonneau, and J.R. Buckley. 2013. Evaluation of polarimetric radar-
sat-2 SAR data for development of soil moisture retrieval algorithms over a chronosequence of black
spruce boreal forests. Remote Sensing of the Environment 132:71–85.
Bourgeau-Chavez, L.L., R.D. Lopez, A. Trebitz, T. Hollenhorst, G.E. Host, B. Huberty, R.L. Gauthier, and
J. Hummer. 2008. Landscape-based indicators. In: Great Lakes Coastal Wetlands Monitoring Plan. Great
Lakes Coastal Wetlands Consortium. Project of the Great Lakes Commission, Vol. 25, pp. 143–171.
Bourgeau-Chavez, L.L., K. Riordan, R.B. Powell, N. Miller, and M. Nowels. 2009. Improving wetland
characterization with multi-sensor, multi-temporal SAR and optical/infrared data fusion. Advances in
Geoscience and Remote Sensing 33:679–708.
Bourgeau-Chavez, L.L., K.B. Smith, E.S. Kasischke, S.M. Brunzell, E.A. Romanowicz, and C.J. Richardson.
2005. Remote monitoring regional scale inundation patterns and hydroperiod in the greater Everglades
ecosystem. Wetlands 25(1):176–191.
Breiman, L. 1996. Bagging predictors. Machine Learning 24(2):123–140.
Breiman, L. 2001. Random forests. Machine Learning 45:5–32.
Brisco, B., K. Li, B. Tedford, F. Charbonneau, S. Yun, and K. Murnaghan. 2013. Compact polarimetry assess-
ment for rice and wetland mapping. International Journal of Remote Sensing 34:1949–1964.
Burgess, W. 1995. Regulatory Wetlands Map Workshop: A Report of Activities and Findings. Maryland
Department of Natural Resources, Annapolis, MD.
Bwangoy, J.R.B., M.C. Hansen, D.P. Roy, G. De Grandi, and C.O. Justice. 2010. Wetland mapping in the
Sensing of the Environment 114(1):73–86.
Canadian Space Agency. January 17, 2009. http://www.asc-csa.gc.ca/eng/satellites/radarsat/description.asp
(Accessed May 12, 2014).
Carter, V. 1982. Applications of remote sensing to wetlands. In: C.J. Johannsen and J.L. Sanders (eds.). Remote
Sensing of Resource Management. Soil Conservation Society of America, Ankeny, IA. pp. 284–300.
Chmura, G.L. 2013. What do we need to assess the sustainability of the tidal salt marsh carbon sink? Ocean &
Coastal Management 83:25–31.
Chust, G., I. Galparsoro, A. Borja, J. Franco, and A. Uriarte. 2008. Coastal and estuarine habitat mapping,
using LIDAR height and intensity and multi-spectral imagery. Estuarine Coastal and Shelf Science
78(4):633–643.
Civco, D., J. Hurd, S. Prisloe, and M. Gilmore (July 2006). Characterization of coastal wetland systems using
multiple remote sensing data types and analytical techniques. In: IEEE International Conference of
Geoscience Remote Sensing Symposium (IGARSS 2006), pp. 3442–3446.
Civco, D.L., J.D. Hurd, E.H. Wilson, M. Song, and Z. Zhang. 2002. A comparison of land use and land cover
change detection methods. ASPRS Annual Convention, Washington, DC.
Collin, A., B. Long, and P. Archambault. 2010. Salt-marsh characterization, zonation assessment and mapping
through a dual-wavelength LiDAR. Remote Sensing of the Environment 114(3):520–530.
Collin, A. and S. Planes. 2012. Enhancing coral health detection using spectral diversity indices from
WorldView-2 imagery and machine learners. Remote Sensing 4(10):3244–3264.
Congalton, R.G., K. Green, and J. Teply. 1993. Mapping old growth forests on national forest and park lands
in the Pacific-Northwest from remotely sensed data. Photogrammetric Engineering and Remote Sensing
59:529–535.
Coren, F. and P. Sterzai. 2006. Radiometric correction in laser scanning. International Journal of Remote
Sensing 27(15):3097–3104.
Costa, M. 2004. Use of SAR satellites for mapping zonation of vegetation communities in the Amazon flood-
plain. International Journal of Remote Sensing 25:1817–1835.
Costa, M. and K.H. Telmer. 2007. Mapping and monitoring lakes in the Brazilian Pantanal wetland using syn-
thetic aperture radar imagery. Marine and Freshwater Ecosystems 17:277–288.
Cowardin, L., V. Carter, F. Golet, and E. LaRoe. 1979. Classification of Wetlands and Deepwater Habitats of
the United States. U.S. Fish and Wildlife Service, Washington, DC.
Cowardin, L. and F. Golet. 1995. U.S. Fish and Wildlife Service 1979 wetland classification: A review. Vegetatio
118:139–152.
Dale, P., A. Chandica, and M. Evans. 1996. Using image subtraction and classification to evaluate change in
sub-tropical intertidal wetlands. International Journal of Remote Sensing 17:703–719.
Davranche, A., G. Lefebvre, and B. Poulin. 2010. Wetland monitoring using classification trees and SPOT-5
seasonal time series. Remote Sensing of the Environment 114(3):552–562.
Davranche, A., B. Poulin, and G. Lefebvre. 2013. Mapping flooding regimes in Camargue wetlands using sea-
sonal multispectral data. Remote Sensing of the Environment 138:165–171.
De Roeck, E.R., N.E.C. Verhoest, M.H. Miya, H. Lievens, O. Batelaan, A. Thomas, and L. Brendonck. 2008.
Remote sensing and wetland ecology: A South African case study. Sensors 8(5):3542–3556.
Dobson, J.E., E.A. Bright, R.L. Ferguson, D.W. Field, L.L. Wood, K.D. Haddad, H. Iredale III et al. 1995.
NOAA Coastal Change Analysis Program (C-CAP) Guidance for Regional Implementation. National
Oceanic and Atmospheric Administration, Seattle, WA. NOAA Technical Report NMFS 123.
Doren, R.F., K. Rutchey, and R. Welch. 1999. The Everglades: A perspective on the requirements and appli-
cations for vegetation map and database products. Photogrammetric Engineering and Remote Sensing
65(2):155–161.
Dronova, I., P. Gong, and L. Wang. 2011. Object-based analysis and change detection of major wetland cover
types and their classification uncertainty during the low water period at Poyang Lake, China. Remote
Dupigny-Giroux, L.A.L. 2007. Using AirMISR data to explore moisture-driven land use–land cover variations
at the Howland Forest, Maine—A case study. Remote Sensing of the Environment 107(1):376–384.
Dwivedi, R., B. Rao, and S. Bhattacharya. 1999. Mapping wetlands of the Sundarban Delta and its environs
using ERS-1 SAR data. International Journal of Remote Sensing 20(11):2235–2247.
Engman, E.T. and R.J. Gurney. 1991. Remote Sensing in Hydrology. Chapman & Hall, London, U.K.
Evans, D.L., J. Apel, R. Arvidson, R. Bindschadler, and F. Carsey. 1995. Spaceborne synthetic aperture radar:
Current status and future directions. A Report to the Committee on Earth Sciences, Space Studies Board,
National Research Council, April 1995, NASA Technical Memorandum 4679.
Evans, T.L. and M. Costa. 2013. Landcover classification of the Lower Nhecolandia subregion of the Brazilian
Pantanal Wetlands using ALOS/PALSAR, RADARSAT-2 and ENVISAT/ASAR imagery. Remote
Sensing of the Environment 128:118–137.
Federal Geographic Data Committee. 1992. Application of satellite data for mapping and monitoring
wetlands—Facts finding report: Technical Report 1, Wetlands Subcommittee, Federal Geographic Data
Committee, Washington, DC.
Federal Geographic Data Committee, Wetlands Subcommittee. 1994. Strategic Interagency Approach to
Developing a National Digital Wetlands Database (second approximation). Federal Geographic Data
Committee, Washington, DC.
Federal Geographic Data Committee, Wetlands Subcommittee. 2007. FGDC Working Draft Wetland
Mapping Standard. Federal Geographic Data Committee, Washington, DC.
Filippi, A.M. and J.R. Jensen. 2006. Fuzzy learning vector quantization for hyperspectral coastal vegetation
classification. Remote Sensing of the Environment 100(4):512–530.
Finlayson, C. and A. Van der Valk. 1995. Wetland classification and inventory: A summary. Vegetatio
118(1–2):185–192.
Fournier, R.A. et al. 2007. Towards a strategy to implement the Canadian Wetland Inventory using satellite
remote sensing. Canadian Journal of Remote Sensing 33:S1–S16.
Gala, T.S. and A.M. Melesse. 2012. Monitoring prairie wet area with an integrated LANDSAT ETM+,
RADARSAT-1 SAR and ancillary data from LiDAR. Catena 95:12–23.
Gao, Z.G. and L.Q. Zhang. 2006. Multi-seasonal spectral characteristics analysis of coastal salt marsh vegeta-
tion in Shanghai, China. Estuarine Coastal and Shelf Science 69(1–2):217–224.
Goodwin, N.R., N.C. Coops, and D.S. Culvenor. 2006. Assessment of forest structure with airborne LiDAR and
the effects of platform altitude. Remote Sensing of the Environment 103(2):140–152.
Govender, M., K. Chetty, and H. Bulcock. 2007. A review of hyperspectral remote sensing and its application
in vegetation and water resource studies. Water SA 33(2):145–151.
Grenier, M., A.M. Demers, S. Labrecque, M. Benoit, R.A. Fournier, and B. Drolet. 2007. An object-based
method to map wetland using RADARSAT-1 and Landsat ETM images: Test case on two sites in Quebec,
Canada. Canadian Journal of Remote Sensing 33(1):28–45.
Grings, F.M., P. Ferrazzoli, J.C. Jacobo-Berlles, H. Karszenbaum, J. Tiffenberg, P. Pratolongo, and P. Kandus.
2006. Monitoring flood condition in marshes using EM models and Envisat ASAR observations. IEEE
Transactions on Geoscience and Remote Sensing 44(4):936–942.
Guo, Y., Z. Shi, H.Y. Li, and J. Triantafilis. 2013. Application of digital soil mapping methods for identify-
ing salinity management classes based on a study on coastal central China. Soil Use and Management
29(3):445–456.
Hall, D.K. 1996. Remote sensing applications to hydrology: Imaging radar. Hydrological Sciences Journal
41(4):609–624.
Harris, J. and S. Digby-Argus. 1987. The detection of wetlands on radar imagery. Paper presented at the 10th
Canadian Symposium on Remote Sensing, Edmonton, Alberta, Canada.
Harvey, K. and G. Hill. 2001. Vegetation mapping of a tropical freshwater swamp in the Northern Territory,
Australia: A comparison of aerial photography, Landsat TM and SPOT satellite imagery. International
Journal of Remote Sensing 22(15):2911–2925.
Hay, G. J., and G. Castilla. 2006. Object-based image analysis: Strengths, weaknesses, opportunities and threats
(SWOT). In: Proceedings of First International Conference on Object-Based Image Analysis, pp. 4–5.
Hess, L.L., J.M. Melack, S. Filoso, and Y. Wang. 1995. Delineation of inundated area and vegetation along
the Amazon floodplain with the SIR-C synthetic aperture radar. IEEE Transactions on Geoscience and
Remote Sensing 33(4):896–904.
Hess, L.L., J.M. Melack, and D.S. Simonett. 1990. Radar detection of flooding beneath the forest canopy:
A review. International Journal of Remote Sensing 11(7):1313–1325.
Hewitt, M.J. 1990. Synoptic inventory of riparian ecosystems: The utility of Landsat Thematic Mapper data.
Forest Ecology and Management 33:605–620.
Hirano, A., M. Madden, and R. Welch. 2003. Hyperspectral image data for mapping wetland vegetation.
Wetlands 23(2):436–448.
Hladik, C., J. Schalles, and M. Alber. 2013. Salt marsh elevation and habitat mapping using hyperspectral and
LIDAR data. Remote Sensing of the Environment 139:318–330.
Horritt, M., D. Mason, D. Cobby, I. Davenport, and P. Bates. 2003. Waterline mapping in flooded vegetation
from airborne SAR imagery. Remote Sensing of the Environment 85(3):271–281.
Houhoulis, P.F. and W.K. Michener. 2000. Detecting wetland change: A rule-based approach using NWI and
SPOT-XS data. Photogrammetric Engineering and Remote Sensing 66(2):205–211.
Huang, C., Y. Peng, M. Lang, I.-Y. Yeo, and G. McCarty. 2014. Wetland inundation mapping and change
monitoring using Landsat and airborne LiDAR data. Remote Sensing of the Environment 141:231–242.
ITRES. http://www.itres.com (Accessed May 12, 2014).
Japan Aerospace Exploration Agency. November 25, 2014, Replace with: http://global.jaxa.jp/projects/sat/
alos2/; http://www.jaxa.jp/pr/brochure/pdf/04/sat29.pdf (Accessed May 12, 2014).
Jensen, J.R. 2000. Remote Sensing of the Environment: An Earth Resource Perspective. Prentice-Hall, Inc.,
Upper Saddle River, NJ.
Jensen, J.R., M.E. Hodgson, E. Christensen, H.E. Mackey, L.R. Tinney, and R. Sharitz. 1986. Remote-sensing
inland wetlands—A multispectral approach. Photogrammetric Engineering and Remote Sensing
52(1):87–100.
Jensen, J.R., H. Lin, X. Yang, E. Ramsey, B.A. Davis, and C.W. Thoemke. 1991. The measurement of man-
grove characteristics in southwest Florida using SPOT multispectral data. Geocarto International
6(2):13–21.
Judd, C., S. Steinberg, F. Shaughnessy, and G. Crawford. 2007. Mapping salt marsh vegetation using aerial
hyperspectral imagery and linear unmixing in Humboldt Bay, California. Wetlands 27(4):1144–1152.
Kasischke, E.S. and L.L. Bourgeau-Chavez. 1997. Monitoring South Florida wetlands using ERS-1 SAR imag-
ery. Photogrammetric Engineering and Remote Sensing 63(3):281–291.
Kasischke, E.S., L.L. Bourgeau-Chavez, A.R. Rober, K.H. Wyatt, J.M. Waddington, and M.R. Turetsky. 2009.
Effects of soil moisture and water depth on ERS SAR backscatter measurements from an Alaskan wet-
land complex. Remote Sensing of the Environment 113:1868–1873.
Kasischke, E.S., L.L. Bourgeau-Chavez, K. Smith, E. Romanowicz, and C. Richardson. 1997a. Monitoring
hydropatterns in South Florida ecosystems using ERS SAR data. Third ERS Symposium on Space at the
Service of our Environment, Florence, Italy, pp. 71–76.
Kasischke, E.S., J.M. Melack, and M.C. Dobson. 1997b. The use of imaging radars for ecological applications—
A review. Remote Sensing of the Environment 59(2):141–156.
Kasischke, E.S., K.B. Smith, L.L. Bourgeau-Chavez, E.A. Romanowicz, S. Brunzell, and C.J. Richardson.
2003. Effects of seasonal hydrologic patterns in south Florida wetlands on radar backscatter measured
from ERS-2 SAR imagery. Remote Sensing of the Environment 88(4):423–441.
Kellndorfer, J.M., L.E. Pierce, M.C. Dobson, and F.T. Ulaby. 1998. Toward consistent regional-to-global-scale
vegetation characterization using orbital SAR systems. IEEE Transactions on Geoscience and Remote
Sensing 36(5):1396–1411.
Klemas, V. 2011. Remote sensing techniques for studying coastal ecosystems: An overview. Journal of Coastal
Research 27(1):2–17.
Klemas, V. 2013. Remote sensing of emergent and submerged wetlands: An overview. International Journal of
Koeln, G. and J. Bissonnette. 2000. Cross-correlation analysis: Mapping landcover change with a historic
landcover database and a recent, single-date multispectral image. Proceedings of the ASPRS Annual
Convention, Washington, DC.
Kokaly, R.F., B.R. Couvillion, J.M. Holloway, D.A. Roberts, S.L. Ustin, S.H. Peterson, S. Khanna, and
S.C. Piazza. 2013. Spectroscopic remote sensing of the distribution and persistence of oil from the
Deepwater Horizon spill in Barataria Bay marshes. Remote Sensing of the Environment 129:210–230.
Kokaly, R.F., D.G. Despain, R.N. Clark, and K.E. Livo. 2003. Mapping vegetation in Yellowstone National
Park using spectral feature analysis of AVIRIS data. Remote Sensing of the Environment 84(3):437–456.
Krohn, M.D., N. Milton, and D.B. Segal. 1983. Seasat synthetic aperture radar (SAR) response to lowland veg-
etation types in eastern Maryland and Virginia. Journal of Geophysical Research: Oceans (1978–2012)
88(C3):1937–1952.
Kudray, G.M. and M.R. Gale. 2000. Evaluation of National Wetland Inventory maps in a heavily forested
region in the upper Great Lakes. Wetlands 20(4):581–587.
Kushwaha, S., R. Dwivedi, and B. Rao. 2000. Evaluation of various digital image processing techniques
for detection of coastal wetlands using ERS-1 SAR data. International Journal of Remote Sensing
21(3):565–579.
Laba, M., F. Tsai, D. Ogurcak, S. Smith, and M.E. Richmond. 2005. Field determination of optimal dates for
the discrimination of invasive wetland plant species using derivative spectral analysis. Photogrammetric
Lang, M., E. Kasischke, S. Prince, and K. Pittman. 2008a. Assessment of C-band synthetic aperture radar
data for mapping Coastal Plain forested wetlands in the Mid-Atlantic Region. Remote Sensing of the
Environment 112:4120–4130.
Lang, M. and G. McCarty. 2009. Lidar intensity for improved detection of inundation below the forest canopy.
Wetlands 29(4):1166–1178.
Lang, M., G. McCarty, R. Oesterling, and I.-Y. Yeo. 2012. Topographic metrics for improved mapping of for-
ested wetlands. Wetlands 33:141–155.
Lang, M.W. and E.S. Kasischke. 2008. Using C-band synthetic aperture radar data to monitor forested wetland
hydrology in Maryland’s Coastal Plain, USA. IEEE Transactions on Geoscience and Remote Sensing
46(2):535–546.
Lang, M.W., P.A. Townsend, and E.S. Kasischke. 2008b. Influence of incidence angle on detecting flooded
forests using C-HH synthetic aperture radar data. Remote Sensing of the Environment 112(10):
3898–3907.
Leblanc, M., J. Lemoalle, J.C. Bader, S. Tweed, and L. Mofor. 2011. Thermal remote sensing of water under
flooded vegetation: New observations of inundation patterns for the ‘Small’ Lake Chad. Journal of
Hydrology 404(1–2):87–98.
Lee, T. and H. Yeh. 2009. Applying remote sensing techniques to monitor shifting wetland vegetation: A case
study of Danshui River estuary mangrove communities, Taiwan. Ecological Engineering 35(4):487–496.
Leifer, I., W.J. Lehr, E. Bradley, R. Clark, P. Dennison, Y. Hu, S. Matheson et al. 2012. State of the art satellite
and airborne marine oil spill remote sensing: Application to the BP Deepwater Horizon oil spill. Remote
Lemmens, M. 2007. Airborne lidar sensors. GIM International 21(2):24–27.
Leonard, P.B., R.F. Baldwin, J.A. Homyack, and T.B. Wigley. 2012. Remote detection of small wetlands in the
Atlantic Coastal Plain of North America: Local relief models, ground validation, and high-throughput
computing. Forest Ecology and Management 284:107–115.
Le Toan, T., A. Beaudoin, J. Riom, and D. Guyon. 1992. Relating forest biomass to SAR data. IEEE Transactions
on Geoscience and Remote Sensing 30(2):403–411.
Li, J. and W. Chen. 2005. A rule-based method for mapping Canada’s wetlands using optical, radar and DEM
data. International Journal of Remote Sensing 26(22):5051–5069.
Li, K., R. De Jong, and J. Boisvert. 1998. Towards estimating soil moisture in the root zone using remotely
sensed surface data. Canadian Journal of Remote Sensing 24(3):255–263.
Liaw, A. and M. Wiener. 2002. Classification and regression by random forest. R News 2(3):18–22.
Lichvar, R.W., D.C. Finnegan, S. Newman, and W. Ochs. 2006. Delineating and evaluating vegetation con-
ditions of vernal pools using spaceborne and airborne remote sensing techniques. Beale Air Force
Base, CA. U.S. Army Corps of Engineers Engineer Research and Development Center, ERDC\CRREL
TM-06–3.
Lillesand, T.M., R.W. Kiefer, and J.W. Chipman. 2004. Remote Sensing and Image Interpretation. John Wiley &
Sons Ltd., New York.
Lunetta, R.S. and M.E. Balogh. 1999. Application of multi-temporal Landsat 5 TM imagery for wetland iden-
tification. Photogrammetric Engineering and Remote Sensing 65(11):1303–1310.
Lyon, J. and J. McCarthy. 1995. Wetland and Environmental Applications of GIS. Lewis Publishers, New York.
Marti-Cardona, B., J. Dolz-Ripolles, and C. Lopez-Martinez. 2013. Wetland inundation monitoring by the syn-
ergistic use of ENVISAT/ASAR imagery and ancillary spatial data. Remote Sensing of the Environment
139:171–184.
Martinez, J.M. and T. Le Toan. 2007. Mapping of flood dynamics and spatial distribution of vegeta-
tion in the Amazon floodplain using multitemporal SAR data. Remote Sensing of the Environment
108(3):209–223.
McCarthy, J., T. Gumbricht, and T. McCarthy. 2005. Ecoregion classification in the Okavango Delta, Botswana
from multitemporal remote sensing. International Journal of Remote Sensing 26(19):4339–4357.
Meijerink, A., D. Bannert, O. Batelaan, M.W. Lubczynski, and T. Pointet. 2007. Remote Sensing Applications
to Groundwater. IHP-VI Series on Groundwater No. 16. United Nations Educational, Scientific and
Cultural Organization.
Milton, G.R., L. Belanger, Y. Crevier, R. Helie, J. Hurley, and B.H. Kazmerik. 2003. Development of a remote-
sensing wetland inventory and classification system for Canada. Backscatter Winter 14(1):3.
Moore, G.K. and G.W. North. 1974. Flood inundation in the southeastern United States from aircraft and satel-
lite imagery. Journal of the American Water Resources Association (JAWRA) 10(5):1082–1096.
Moreno, J.P.M. and J.E.M. De Larriva. 2012. Comparison between new digital image classification methods
and traditional methods for land-cover mapping. In: C.P. Giri (ed.). Remote Sensing of Land Use and
Land Cover: Principles and Applications. CRC Press, Boca Raton, FL.
Morgan, J.L., S.E. Gergel, and N.C. Coops. 2010. Aerial photography: A rapidly evolving tool for ecological
management. Bioscience 60(1):47–59.
Mougin, E., C. Proisy, G. Marty, F. Fromard, H. Puig, J. Betoulle, and J.P. Rudant. 1999. Multifrequency and
multipolarization radar backscattering from mangrove forests. IEEE Transactions on Geoscience and
Mumby, P., E. Green, A. Edwards, and C. Clark. 1999. The cost-effectiveness of remote sensing for tropical
coastal resources assessment and management. Journal of Environmental Management 55(3):157–166.
Munyati, C. 2000. Wetland change detection on the Kafue Flats, Zambia, by classification of a multitemporal
remote sensing image dataset. International Journal of Remote Sensing 21(9):1787–1806.
Murphy, P.N.C., J. Ogilvie, M. Castonguay, C. Zhang, F. Meng, and P.A. Arp. 2008. Improving forest opera-
tions planning through high-resolution flow-channel and wet-areas mapping. The Forestry Chronicle
84(4):568–574.
Muster, S., B. Heim, A. Abnizova, and J. Boike. 2013. Water body distributions across scales: A remote sensing
based comparison of three arctic tundra wetlands. Remote Sensing 5(4):1498–1523.
Nagendra, H., R. Lucas, J.P. Honrado, R.H.G. Jongman, C. Tarantino, M. Adamo, and P. Mairota. October 2013.
Remote sensing for conservation monitoring: Assessing protected areas, habitat extent, habitat condition,
species diversity, and threats. Original Research Article Ecological Indicators, Vol. 33, pp. 45–59. http://
www.sciencedirect.com/science/article/pii/S1470160X12003317.
National Research Council, Committee of Earth Sciences and Applications from Space. 2007. Earth Science
and Applications from Space: National Imperatives for the Next Decade and Beyond. National Academy
Press, Washington, DC.
Newcomb, D. and M. Lang. 2012. Potential of LiDAR intensity data for improved operational mapping of
forested wetlands. National Wetlands Newsletter 34(5):11–15.
Nielsen, E.M., S.D. Prince, and G.T. Koeln. 2008. Wetland change mapping for the US mid-Atlantic region
using an outlier detection technique. Remote Sensing of the Environment 112(11):4061–4074.
NOAA. http://coast.noaa.gov www.csc.noaa.gov/landcover (Accessed May 12, 2014).
Olson Jr., C.E. 1960. Elements of photographic interpretation common to several sensors. Photogrammetric
Engineering 26(4):651–656.
Oregon State University. April 12, 2014. http://hico.coas.oregonstate.edu/ (Accessed May 12, 2014).
Ormsby, J.P., B.J. Blanchard, and A.J. Blanchard. 1985. Detection of lowland flooding using active microwave
systems. Photogrammetric Engineering and Remote Sensing 51(3):317–329.
Ouyang, Z.T., M.Q. Zhang, X. Xie, Q. Shen, H.Q. Guo, and B. Zhao. 2011. A comparison of pixel-based
and object-oriented approaches to VHR imagery for mapping saltmarsh plants. Ecological Informatics
6(2):136–146.
Papa, F., C. Prigent, F. Durand, and W.B. Rossow. 2006. Wetland dynamics using a suite of satellite observa-
tions: A case study of application and evaluation for the Indian Subcontinent. Geophysical Research
Letters 33(8): L08401.
Pattanaik, C. and S.N. Prasad. 2011. Assessment of aquaculture impact on mangroves of Mahanadi Delta (Orissa),
East Coast of India using remote sensing and GIS. Ocean & Coastal Management 54(11):789–795.
Pearlman, S., S. Carman, C. Segal, P. Jarecke, and P. Barry. 2001. Overview of the Hyperion imaging spectrom-
eter for the NASA EO-1 mission. International Geoscience and Remote Sensing Symposium Proceedings,
Sydney, New South Wales, Australia.
Petus, C., M. Lewis, and D. White. 2013. Monitoring temporal dynamics of Great Artesian Basin wetland
vegetation, Australia, using MODIS NDVI. Ecological Indicators 34:41–52.
Phinn, S., L. Hess, and C.M. Finlayson. 1999. An assessment of the usefulness of remote sensing for wetland
inventory and monitoring in Australia. In: C.M. Finlayson and A.G. Spiers (eds.). Techniques for
Enhanced Wetland Inventory and Monitoring. Supervising Scientist Report 147, Canberra, Australian
Capital Territory, Australia.
Pinty, B., J.L. Widlowski, N. Gobron, M.M. Verstraete, and D.J. Diner. 2002. Uniqueness of multiangular
measurements. I. An indicator of subpixel surface heterogeneity from MISR. IEEE Transactions on
Geoscience and Remote Sensing 40(7):1560–1573.
Piwowar, J. 2005. Digital image analysis. In: S. Arnoff (ed.). Remote Sensing for GIS Managers. ESRI Press,
Redlands, CA.
Place, J.L. 1985. Mapping of forested wetland: Use of Seasat RADAR images to complement conventional
sources. The Professional Geographer 37(4):463–469.
Pope, K.O., E. Rejmankova, J.F. Paris, and R. Woodruff. 1997. Detecting seasonal flooding cycles in marshes
of the Yucatan Peninsula with SIR-C polarimetric radar imagery. Remote Sensing of the Environment
59(2):157–166.
Pope, K.O., J.M. Rey-Benayas, and J.F. Paris. 1994. Radar remote sensing of forest and wetland ecosystems in
the Central American tropics. Remote Sensing of the Environment 48(2):205–219.
Poulin, B., A. Davranche, and G. Lefebvre. 2010. Ecological assessment of Phragmites australis wetlands
using multi-season SPOT-5 scenes. Remote Sensing of the Environment 114(7):1602–1609.
Powers, R.P., G.J. Hay, and G. Chen. 2012. How wetland type and area differ through scale: A GEOBIA case
study in Alberta’s Boreal Plains. Remote Sensing of the Environment 117:135–145.
Ramsey, E. and S.C. Laine. 1997. Comparison of Landsat Thematic Mapper and high resolution photography
to identify change in complex coastal wetlands. Journal of Coastal Research 13(2):281–292.
Raney, K. 1998. Radar fundamentals: Technical perspective. In: F.M. Henderson and A.J. Lewis (eds.).
Principles and Application of Imaging Radar. John Wiley & Sons, Inc., New York.
Rao, B., R. Dwivedi, S. Kushwaha, S. Bhattacharya, J. Anand, and S. Dasgupta. 1999. Monitoring the spa-
tial extent of coastal wetlands using ERS-1 SAR data. International Journal of Remote Sensing
20(13):2509–2517.
Reimer, K. 2009. The need for a Canadian wetland inventory. Conservator 30(1):36–45.
Ringrose, S., C. Vanderpost, and W. Matheson. 2003. Mapping ecological conditions in the Okavango delta,
Botswana using fine and coarse resolution systems including simulated SPOT vegetation imagery.
International Journal of Remote Sensing 24(5):1029–1052.
Roberts, D., M. Smith, and J. Adams. 1993. Green vegetation, non-photosynthetic vegetation, and soils in
AVIRIS data. Remote Sensing of the Environment 44(2):255–269.
Rocchini, D., G.M. Foody, H. Nagendra, C. Ricotta, M. Anand, K.S. He, V. Amici et al. 2013. Uncertainty in
ecosystem mapping by remote sensing. Computers & Geosciences 50:128–135.
Rosenqvist, A., C.M. Finlayson, J. Lowry, and D. Taylor. 2007. The potential of long-wavelength satellite-borne
radar to support implementation of the Ramsar wetlands convention. Aquatic Conservation: Marine and
Freshwater Ecosystems 17(3):229–244.
Rosso, P., S. Ustin, and A. Hastings. 2005. Mapping marshland vegetation of San Francisco Bay, California,
using hyperspectral data. International Journal of Remote Sensing 26(23):5169–5191.
Rosso, P., S. Ustin, and A. Hastings. 2006. Use of lidar to study changes associated with Spartina invasion in
San Francisco Bay marshes. Remote Sensing of the Environment 100(3):295–306.
Rundquist, D.C., S. Narumalani, and R.M. Narayanan. 2001. A review of wetlands remote sensing and defining
new considerations. Remote Sensing Reviews 20:207–226.
Rutchey, K. and L. Vilchek. 1999. Air photointerpretation and satellite imagery analysis techniques for map-
ping cattail coverage in a northern Everglades impoundment. Photogrammetric Engineering and Remote
Sensing 65(2):185–191.
Sader, S.A., D. Ahl, and W.S. Liou. 1995. Accuracy of Landsat-TM and GIS rule-based methods for forest
wetland classification in Maine. Remote Sensing of the Environment 53(3):133–144.
Sahagian, D. and J. Melack. 1996. Global wetland distribution and functional characterization: Trace gases
and the hydrologic cycle. International Geosphere-Biosphere Programme (IGBP), Stockholm, Sweden.
IGBP Report 46.
Schapire, R., Y. Freund, P. Bartlett, and W. Lee. 1998. Boosting the margin: A new explanation for the effective-
ness of voting methods. Annals of Statistics 26(5):1651–1686.
Schmidt, K. and A. Skidmore. 2003. Spectral discrimination of vegetation types in a coastal wetland. Remote
Shanmugam, P., Y.H. Ahn, and S. Sanjeevi. 2006. A comparison of the classification of wetland characteristics
by linear spectral mixture modeling and traditional hard classifiers on multispectral remotely sensed
imagery in southern India. Ecological Modelling 194(4):379–394.
Shapiro, C. 1995. Coordination and Integration of Wetland Data for Status and Trends and Inventory Estimates.
Federal Geographic Data Committee Wetlands Subcommittee, Washington, DC. Technical Report 2.
Silva, T.S., M.P. Costa, J.M. Melack, and E.M. Novo. 2008. Remote sensing of aquatic vegetation: Theory and
applications. Environmental Monitoring and Assessment 140(1–3):131–145.
Simard, M., K. Zhang, V.H. Rivera-Monroy, M.S. Ross, P.L. Ruiz, E. Castañeda-Moya, R.R. Twilley, and E.
Rodriguez. 2006. Mapping height and biomass of mangrove forests in Everglades National Park with
SRTM elevation data. Photogrammetric Engineering and Remote Sensing 72(3):299–311.
Siqueira, P. April 12, 2013. DESDynL-R: Radar-only version of DESDynI, http://www.eorc.jaxa.jp/ALOS/en/
kyoto/apr2013_kc19/pdf/4-5_KC19_DESDYnI-R_Siqueira.pdf (Accessed May 12, 2014).
Smith, L.C. 1997. Satellite remote sensing of river inundation area, stage, and discharge: A review. Hydrological
Processes 11(10):1427–1439.
Snyder, G. and M. Lang. 2012. Significance of a 3D elevation program to wetland mapping. National Wetlands
Newsletter 34(5):11–15.
Sokol, J., H. NcNairn, and T. Pultz. 2004. Case studies demonstrating the hydrological applications of C-band
multipolarized and polarimetric SAR. Canadian Journal of Remote Sensing 30(3):470–483.
Song, K.S., Z.M. Wang, L. Li, L. Tedesco, F. Li, C. Jin, and J. Du. 2012. Wetlands shrinkage, fragmentation
and their links to agriculture in the Muleng-Xingkai Plain, China. Journal of Environmental Management
111:120–132.
SPECIM. http://www.specim.fi (Accessed May 12, 2014).
Tanis, F., L. Bourgeau-Chavez, and M.C. Dobson. 1994. Application of ERS-1 SAR for coastal inundation. IEEE
International Geoscience and Remote Sensing Symposium Proceedings (August 1994). #94CH3378-7,
Pasadena, CA, Vol. 3, pp. 1481–1483.
Timm, B.C. and K. McGarigal. 2012. Fine-scale remotely-sensed cover mapping of coastal dune and salt marsh
ecosystems at Cape Cod National Seashore using random forests. Remote Sensing of the Environment
127:106–117.
Tiner, R.W. 1997. NWI maps: What they tell us. National Wetlands Newsletter 19(2):7–12.
Mapping. Lewis Publishers, Boca Raton, FL.
Tiner, R.W. 2003. Estimated extent of geographically isolated wetlands in selected areas of the United States.
Wetlands 23(3):636–652.
Tiner, R.W. 2009. Status Report for the National Wetlands Inventory Program: 2009. U.S. Fish and Wildlife
Service, Washington, DC.
Tiner, R.W. and G. Smith. 1992. Comparisons of Four Scales of Color Infrared Photography for Wetland
Mapping in Maryland. U.S. Fish and Wildlife Service, Region 5, Newton, MA. National Wetlands
Inventory Report R5-92/03.
Touzi, R., A. Deschamps, and G. Rother. 2007. Wetland characterization using polarimetric RADARSAT-2
capability. Canadian Journal of Remote Sensing 33:S56–S67.
Townsend, P.A. 2000. A quantitative fuzzy approach to assess mapped vegetation classifications for ecological
applications. Remote Sensing of the Environment 72(3):253–267.
Townsend, P.A. 2001. Mapping seasonal flooding in forested wetlands using multi-temporal radarsat SAR.
Photogrammetric Engineering and Remote Sensing 67(7):857–864.
Townsend, P.A. 2002. Relationships between forest structure and the detection of flood inundation in forested
wetlands using C-band SAR. International Journal of Remote Sensing 23(3):443–460.
Townsend, P.A. and S.J. Walsh. 1998. Modeling floodplain inundation using an integrated GIS with radar and
optical remote sensing. Geomorphology 21(3):295–312.
Töyrä, J. and A. Pietroniro. 2005. Towards operational monitoring of a northern wetland using geomatics-based
techniques. Remote Sensing of the Environment 97(2):174–191.
Töyrä, J., A. Pietroniro, and L.W. Martz. 2001. Multisensor hydrologic assessment of a freshwater wetland.
Remote Sensing of the Environment 75(2):162–173.
Töyrä, J., A. Pietroniro, L.W. Martz, and T.D. Prowse. 2002. A multi-sensor approach to wetland flood monitor-
ing. Hydrological Processes 16(8):1569–1581.
Tucker, C.J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote
US FWS. November 25, 2014. http://www.fws.gov/wetlands/index.html (Accessed May 12, 2014).
USGS. November 24, 2014. http://pubs.usgs.gov/tm/11b4/ (Accessed May 12, 2014).
Vanderbilt, V.C., S. Khanna, and S.L. Ustin. 2007. Impact of pixel size on mapping surface water in subsolar
imagery. Remote Sensing of the Environment 109(1):1–9.
Vanderbilt, V.C., G.L. Perry, G.P. Livingston, S.L. Ustin, D. Barrios, F.M. Bréon, M.M. Leroy, J.Y. Balois, L.A.
Morrissey, and S.R. Shewchuk. 2002. Inundation discriminated using sun glint. IEEE Transactions on
Vane, G., R.O. Green, T.G. Chrien, H.T. Enmark, E.G. Hansen, and W.M. Porter. 1993. The airborne visible/
infrared imaging spectrometer (AVIRIS). Remote Sensing of the Environment 44(2):127–143.
Vierling, K.T., L.A. Vierling, W.A. Gould, S. Martinuzzi, and R.M. Clawges. 2008. Lidar: Shedding new light
on habitat characterization and modeling. Frontiers in Ecology and the Environment 6(2):90–98.
Wdowinski, S., S.W. Kim, F. Amelung, T.H. Dixon, F. Miralles-Wilhelm, and R. Sonenshein. 2008. Space-
based detection of wetlands’ surface water level changes from L-band SAR interferometry. Remote
Welch, R., M. Madden, and R.F. Doren. 1999. Mapping the everglades. Photogrammetric Engineering and
Wilen, B. and R. Tiner. September 1–12, 1989. The National Wetlands Inventory—The first ten years. American
Water Resources and Concerns. American Water Resources Association, Middleburg, VA.
Wilson, B.A. and H. Rashid. 2005. Monitoring the 1997 flood in the Red River Valley using hydrologic regimes
and RADARSAT imagery. The Canadian Geographer/Le Géographe Canadien 49(1):100–109.
Wright, C. and A. Gallant. 2007. Improved wetland remote sensing in Yellowstone National Park using
classification trees to combine TM imagery and ancillary environmental data. Remote Sensing of the
Environment 107(4):582–605.
Yang, J. and F.J. Artigas. 2010. Mapping salt marsh vegetation by integrating hyperspectral and LiDAR remote
sensing. In: Q. Weng (Series Editor) Remote Sensing of Coastal Environment. Taylor & Francis Series in
Remote Sensing Applications 2010, pp. 173–190.
mapping in the Everglades from hyperspectral imagery. Remote Sensing of the Environment 124:310–320.
Zhao, D., H. Jiang, T. Yang, Y. Cai, D. Xu, and S. An. 2012. Remote sensing of aquatic vegetation distribution
in Taihu Lake using an improved classification tree with modified thresholds. Journal of Environmental
Management 95(1):98–107.
Zomer, R.J., A. Trabucco, and S.L. Ustin. 2009. Building spectral libraries for wetlands land cover classifica-
tion and hyperspectral remote sensing. Journal of Environmental Management 90(7):2170–2177.
Section II
Summaries of Remote Sensing
Technologies and Their Application
for Mapping Wetlands
6 Mapping and Monitoring
Surface Water and Wetlands
with Synthetic Aperture Radar
Brian Brisco
CONTENTS
Introduction..................................................................................................................................... 119
SAR Background............................................................................................................................ 120
Applications of SAR for Flood and Wetland Mapping................................................................... 121
Mapping and Monitoring Surface Water and Wetlands.................................................................. 122
Background................................................................................................................................ 122
Detection of Open Water............................................................................................................ 122
SAR Image Processing for Flood and Surface Water Mapping................................................. 125
Flood Mapping with SAR.......................................................................................................... 125
Surface Water Monitoring with SAR......................................................................................... 126
Wetland Classification with SAR.................................................................................................... 126
Monitoring Flooded Vegetation with SAR..................................................................................... 129
Conclusion...................................................................................................................................... 132
References....................................................................................................................................... 132
INTRODUCTION
Freshwater resources are becoming increasingly strained throughout the world. This is due to a
number of causes including larger populations; increasing agricultural irrigation that often results
in overuse or salinization; populations increasing in sunny dry climates that have limited water
to begin with; more pollution from cities, agriculture, and industry; increased human and indus-
trial consumption; and climate change. Knowing the condition and quantity of water supplies is
absolutely critical for planning future growth. Monitoring surface water and water level changes
throughout a watershed could improve hydrological models and the information available for envi-
ronmental management and social planning.
Canada, a country blessed with abundant water in most places, is experiencing water problems.
In the Prairies, drought years put strains on water resources for both urban areas (e.g., Regina)
and rural areas (e.g., southern Alberta and Saskatchewan). Industrial use of water is placing heavy
demands on existing water resources. For example, oil extraction from the tar sands requires large
quantities of water from the Athabasca River.
Besides supplying human needs, ample quantities of water are needed to maintain healthy wet-
lands that are critical for cleansing water, flood water retention, carbon storage, provision of fish,
and wildlife habitat and recreation. The increasing frequency of flash floods in urban areas where
impervious surfaces have increased significantly demonstrates the need for more accurate and timely
information on surface water. The 2013 Calgary flood in Alberta is a recent example as well as the
119
annual Red River floods in the United States/Canada border region of North Dakota and Manitoba.
This demonstrates the need for accurate and timely information on surface water conditions.
Five major user communities need hydrometric data and information derived from these data
for water management: (1) structural designers, (2) emergency management personnel, (3) water
resource users, (4) government agencies, and (5) nongovernment organizations. Structural designers
need the information to optimize the design of various water-related industrial structures such as
bridges, culverts, pipeline crossings, and dams, whereas emergency management personnel need
the data for flood prediction and mapping. There is also a large resource-use community who need
hydrometric information on a day-to-day basis including water supply and sewage disposal sectors,
agriculture, forestry, power supply companies, fishing, and ecotourism sectors. The last two groups
include government agencies and water resource and environmental stewardship organizations who
need water quality and quantity information for biodiversity and ecological habitat assessment as
well as other ecological functional assessments.
A broadscale monitoring program that tracks changes in wetlands and water resources would
provide water resource managers and user communities with vital information to help address their
concerns. Due to high variability in both space and time, an effective monitoring system must con-
sider the use of remote sensing to provide timely, cost-effective information.
SAR BACKGROUND
Synthetic aperture radar (SAR) has long been recognized as an important source of data for moni-
toring surface water, especially under inclement weather conditions, and is therefore used opera-
tionally for flood mapping applications (Pultz et al. 1997; Townsend 2002; Brisco et al. 2008, 2009).
Being independent of cloud and smoke cover, able to operate day and night, and not subject to
sunglint, SAR offers a reliable data stream for monitoring water bodies. A SAR is a coherent air-
borne or spaceborne side-looking radar system that utilizes data acquisition along the flight path of
the platform to simulate an extremely large antenna or aperture electronically and generates high-
resolution remotely sensed imagery. Over time, individual transmit/receive cycles are completed
with the data from each cycle being stored electronically. The signal processing uses magnitude and
phase of the received signals over successive pulses from elements of a synthetic aperture. After
a given number of cycles, the stored data are recombined (taking into account the Doppler effects
inherent in the different transmitter to target geometry in each succeeding cycle) to create a high-
resolution image of the terrain being over flown.
Due to space constraints in this chapter, SAR technology will not be reviewed in this chapter, but
readers from an engineering background can refer to the Ulaby et al. three-volume textbook series
by Ulaby et al. (1986) for details on SAR engineering and technology. End users and application sci-
entists are encouraged to review Henderson and Lewis (1998) for background information on a wide
range of SAR applications and Henderson and Lewis (2008) for a detailed review of SAR and wet-
lands. Excellent SAR tutorials are also available online from a number of locations. A few are listed
here for reader convenience (http://www.nrcan.gc.ca/earth-sciences/geography-boundary/remote-
sensing/11810; http://www.radartutorial.eu/20.airborne/ab07.en.html; http://earth.eo.esa.int/download/
eoedu/Earthnet-website-material/to-access-from-Earthnet/2008_Bilko-SAR-Land-Applications-
Tutorial/sar_land_apps_1_theory.pdf; http://elib.dlr.de/82313/1/SAR-Tutorial-March-2013.pdf).
Since the launch of Seasat in 1978, there has been a steady progression of civilian SAR sys-
tems in space. These include SIR-A, SIR-B, SIR-C/X-SAR, ERS-1/2, JERS-1, ENVISAT ASAR,
RADARSAT-1&2, SRTM, COSMO-SkyMed, TerraSAR-X, and Advanced Land Observing
Satellite (ALOS)/PALSAR. Today, there are a number of SAR constellations and satellite platforms
providing a range of SAR data with more planned for the near future insuring a continuing supply
of spaceborne SAR data for wetland and surface water applications (Table 6.1). Note most of these
SAR systems provide a range of beam modes from 1 m spotlight modes (with limited swath cover-
age) to low spatial resolution ScanSAR modes (50–100 m resolution) with large swath coverages.
Mapping and Monitoring Surface Water and Wetlands with Synthetic Aperture Radar 121
TABLE 6.1
List of Current and Future SAR Satellites
SAR System Launch Frequency/Polarization Country
TerraSAR-X/TanDEM-X 2007 X (quad) Germany
RADARSAT-2 2007 C (quad) Canada
COSMO-SkyMed 1–4 2007–2010 X (dual) Italy
RISAT-1 2012 C (quad) India
HJ-1C 2012 S (VV) China
Kompsat-5 2013 X (dual) Korea
PAZ 2013 X (quad) Spain
Sentinel-1 2013 C (quad) ESA (European consortium)
ALOS-2 2013 L (quad) Japan
RADARSAT Constellation Mission 1–3 2017/2018 C (quad) Canada
SAOCOM-1/2 2014/2015 L (quad) Argentina
Each of these SAR systems has a web page or other information sources where detailed information
on the sensor specifics, data processing, and products can be obtained.
APPLICATIONS OF SAR FOR FLOOD AND WETLAND MAPPING

SAR is an attractive sensor for monitoring wetlands due to the timely data acquisition capabili-
ties and the sensitivity to surface water and flooded vegetation. The ability to map and monitor
open water with SAR provides information often unavailable from optical sensors due to clouds
and limited orbits and swath coverage and is of value to a wide range of applications including
hydrology, ecology, meteorology, and flood mapping (Brisco et al. 2008). The ability to map
and monitor flooded vegetation with SAR provides information often unavailable from optical
sensors due to clouds, limited orbits, and swath coverage. Hydrology, ecology, meteorology, and
flood mapping can benefit because SAR data have considerable potential to monitor water under
vegetation canopies, as long as they are not too closed (i.e., dense canopies) for the microwave
energy to penetrate and be received (MacDonald et al. 1980; Hess et al. 1990; Kasischke and
Bourgeau-Chavez 1997; Pope et al. 1997; Townsend 2002). The vegetation canopy penetration
capability of the microwaves allows for mapping of flooded vegetation as a result of enhanced
backscatter from a double-bounce scattering mechanism. This leads to an enhanced HH back-
scatter with little increase in VV. Thus, the HH/VV dual polarization data sets can be used to
identify flooded vegetation.
This also allows for the use of polarimetric decomposition techniques to identify regions of
flooded vegetation (Brisco et al. 2011). A few well-known decomposition techniques, such as the
Cloude–Pottier, the Freeman–Durden, the Pauli, the Van Zyl, the Yamaguchi, and the Touzi decom-
position methods (Van Zyl 1989; Cloude and Pottier 1997; Freeman and Durden 1998; Yamaguchi
et al. 2005; Touzi et al. 2007), are now implemented in commercial software that are readily avail-
able to the scientific community. Touzi et al. (2004) provide a thorough review of information
extraction from polarimetric SAR data.
Previous investigations have shown that, under certain conditions, the SAR response signal
from flooded vegetation may also remain coherent over time. Coherence measures the similar-
ity in scattering between two SAR acquisitions with values between 0 for no coherence and 1 for
perfect coherence. Thus, repeat SAR satellite overpasses can be exploited by interferometric SAR
(InSAR) measurements and data processing procedures for estimating water level changes in asso-
ciated wetlands (Alsdorf et al. 2001; Lu et al. 2005; Lu and Kwoun 2008; Wdowinski et al. 2008;
Kim et al. 2009; Hong et al. 2010). Please see Chapter 7, Wetland InSAR: A Review of the Technique
and Applications by Wdowinski and Hong for a detailed description.
The rest of this chapter provides an overview of the use of SAR for water and wetland
resource applications. The habitats of interest are not described in any detail because the tech-
niques described apply to a wide range of habitats from all over the globe. The Canadian Wetland
Classification System is the Canada-wide classification system that provides a common language to
describe wetlands. It is an ecologically based system that incorporates soil, water (water chemistry,
water table), and vegetation characteristics. There are three hierarchical levels within the CWCS:
(1) class, (2) form, and (3) type. The five wetland “classes” differentiate by development character-
istics and the environment in which they exist. The five classes are bog, fen, marsh, swamp, and
shallow open water. These classes can be further recognized by their “form” based on surface,
water, and underlying mineral soil characteristics. The class/form combinations can be further
described by their “type” based on the vegetation communities associated with them. This wetland
classification nomenclature will be used throughout this chapter. A more detailed description of
the five wetland classes along with their forms and types can be found in the National Wetlands
Working Group (1997).
The following sections describe and review flood mapping and temporal surface water moni-
toring approaches as well as wetland classification using SAR alone and in combination with
optical data. Monitoring the temporal variations of flooded vegetation is also reviewed in detail,
especially the use of polarimetric decomposition techniques. This chapter concludes with a
review of the future research and development trends for using SAR for surface water and wetland
applications.
MAPPING AND MONITORING SURFACE WATER AND WETLANDS

Background
The following discussion will describe the use of SAR for mapping and monitoring surface water
that includes not only the classic rivers, lakes, and impoundments but also the shallow open water
areas associated with wetlands. Open water areas are characterized by having little emergent veg-
etation and may be shallow or deep areas depending on local topography and wetland conditions.
This description also applies to recently flooded areas of upland terrain which occur during flood
events for the purpose of the discussion of using SAR for mapping open water as the approach for
information extraction is the same for both situations.
The types of microwave scattering mechanisms that occur in forest and wetland land cover types
including this open water category are illustrated in Figure 6.1. They vary as a function of the
type and amount of vegetation and the presence of standing water. For the open water category,
the scattering is specular or mirrorlike with little energy returning back to the sensor that results
in the dark image tones on SAR images. This makes it easy to discriminate this open water class
from upland land cover using a gray-level thresholding procedure. Flooded vegetation, on the other
hand, produces double-bounce scattering resulting in bright tones on SAR images. For nonflooded
vegetation, the scattering is often described as volume scattering that is diffuse in nature and thus
not generally as bright as double-bounce scattering.
Detection of Open Water

SAR is very attractive for surface water mapping and monitoring applications because of the all-
weather, day- or nighttime capability for data collection. This is particularly important for flood map-
ping that typically is needed during cloudy weather that hampers the use of optical data (e.g., aerial
photography). SAR can also be used to effectively monitor the temporal and interannual changes in
surface water within a watershed or region of interest. With current multisatellite constellations such
Forest Forest Forest
Water
Surface backscattering Volume backscattering Double-bounce backscattering
Marsh
Marsh Marsh
Water
Volume backscattering Double-bounce backscattering Specular scattering
FIGURE 6.1 Schematic illustrating the different microwave scattering processes for forest and wetland
vegetation.
as TerraSAR-X/TanDEM-X and COSMO-SkyMed, as well as the next-generation RADARSAT

Constellation, daily change mapping is possible.
The effect of wind and/or rain on the surface of the water can cause some rough surface back-
scatter rather than the purely specular scattering typical for flat/smooth water (Figure 6.2). Note that
as the water roughens due to increased wind speed or fetch, the surface scattering pattern broadens
and more energy is backscattered. This lowers the contrast between the land and water targets
Incident wave Scattering wave
Specular
Incident wave
Scattering wave
Rough
Incident wave
Scattering wave
Very rough
FIGURE 6.2 Schematic drawing of specular scattering illustrating the increase in backscatter as a function
of increased surface roughness.
making the mapping of the surface water more problematic for SAR. Nonetheless, the all-weather
capability of the SAR data collection usually produces multiple images that can be prescreened to
select images acquired during low-wind conditions.
HH polarization generally yields the highest contrast between upland and open water and is the
preferred polarization (Brisco et al. 2008, 2009). Another approach is to use the cross-polarization
channel that is not as sensitive to increased surface roughness on the water and can still provide
good contrast between upland and surface water targets (White et al. 2013). Thus, an HH/HV dual
polarization image is often requested to insure accurate delineation of the open water. VV polar-
ization highlights surface roughness effects through the interaction of the capillary waves and the
vertical polarization and is the least favored polarization to use. These approaches are illustrated
in Figures 6.3 and 6.4. Shallow incidence angles (30°–50°) are preferred to maximize the contrast
HH HV
(a) (b)
FIGURE 6.3 HH and HV channels over a water body in Tamarack, MN, showing better land–water dis-
crimination with the HH channel. Image (a) was acquired on April 21, 2009, and image (b) on August 19,
2009. Both images are from beam mode FQ12.
HH HV
(a) (b)
FIGURE 6.4 HH and HV channels over a water body, showing better land–water discrimination with the
cross-polarization channel when windy. Images (a) (May 13, 2009, FQ12) and (b) (July 26, 2009, FQ12) were
acquired over Tamarack, MN.
between land and water. The ability to map surface water is independent of frequency although the
shorter wavelengths are preferred (X- and C-band) as once again the land/water contrast is highest
at these frequencies where volume scattering of land targets generally increases the backscatter well
above the value for open water.
SAR Image Processing for Flood and Surface Water Mapping

One of the most common techniques for extracting objects from a picture is the classic gray-level
thresholding or density slicing (Kapur et al. 1985). When the object is clearly distinguishable and
the gray-level histogram is bimodal, then the threshold can be selected at the bottom of the valley,
and a satisfactory result can be easily obtained. For SAR image processing, both radiometric cor-
rections and geometric adjustments and re-projections to some datum are performed prior to the
selection and application of an image threshold. This process includes speckle reduction using adap-
tive filters to maintain the resolution, geometric correction to transform the data to a desired map
projection from the slant range geometry, and radiometric enhancement to improve contrast. The
reference material provided earlier will allow the reader to pursue SAR image processing in more
detail. The remainder of this section will review various approaches that have been developed for
thresholding SAR images for flood or surface water delineation.
Flood Mapping with SAR

Although the use of SAR is now operational for flood response mapping in the near real time
using this thresholding technique along with post classification filtering and error correction, there
are still issues with this approach that are being addressed by a variety of techniques (Matgen
et al. 2011; Pulvirenti et al. 2011; White et al. 2011). A review of satellite remote sensing of river
inundation found that single-polarization, fixed-frequency SARs cannot adequately map floods
and that multifrequency, multipolarization SARs are required for the optimal discrimination of
flood inundation over the wide variety of habitats that occur on the landscape (Smith 1997).
Despite these shortcomings, SAR is the best choice as clouds, trees, or floating vegetation often
obscure the signal from optical sensors. When available, multifrequency, multipolarization SAR
could map flooded areas in most wetland land cover types. For mapping open surface water,
attributed to flood events, single-frequency, fixed-polarization SAR can do the job, but post clas-
sification filtering and manual editing are still needed to fix errors of commission and omission.
A typical flowchart of SAR image processing for flood mapping using PCI modeler is shown in
Figure 6.5.
One approach to reduce errors of commission or omission is the use of both tone and tex-
ture in the thresholding procedure (Horritt et al. 2001). Using this approach, the main source
of confusion was nonflooded vegetation giving similar radar returns to open water. Integrating
physical and topographic information into the classification scheme can help reduce these types
of errors and improve the flood mapping capability (Pierdicca et al. 2008; Hostache et al. 2009).
The combination of optical data for land cover classification, preflood SAR for initialization,
SAR for real-time water boundary estimation, and a high-resolution digital elevation model
(DEM) to constrain the classification will produce the best results. Any concurrent ground truth
(i.e., in situ verification data) is also valuable for validation purposes. With this type of approach,
a semiautomated flood monitoring system can be envisioned (Matgen et al. 2011) especially
with the rapid revisit time current SAR systems such as the COSMO-SkyMed and the future
RCM provide (Pulvirenti et al. 2011). Dellepiane and Angiati (2012) demonstrate how a cross-
normalization procedure using only multitemporal SAR data from before and after the flood
event can be used effectively for the detection of flooded areas. However, this approach is still
enhanced by the use of land cover and topographic information to reduce errors of commission
and omission.
Visually inspect
R2 image; select
the HH or HV
channel and a
threshold
5*5 FGAMMA Scale from Enter and apply
filter to Select pixels that
linear to decibel threshold are water in result
preserve edges
of FGAMMA and
Import the R2 FAV filter. Also
product xml select pixels that
are water in only
the FGAMMA filter
3*3 FAV filter Scale from Enter and apply result
to reduce noise linear to decibel threshold
Orthorectify 3*3 FMO filter

Export to pix file image using to further
rational function reduce noise
FIGURE 6.5 Visual flow diagram of the surface water extraction methodology using PCI modeler.
Surface Water Monitoring with SAR

In many environments, particularly northern temperate ecosystems, there is a seasonal water cycle
that begins with high water after snowmelt followed by a dry-down through the summer months
and an increasing water level in the fall before freeze-up. In tropical and subtropical environments
with a monsoon-type climate, there is a wet period followed by a dry period. This seasonal water
cycle is controlled by numerous factors and can vary significantly within a year as well as between
years. This water cycle can also be influenced by climate warming and anthropogenic factors due
to man’s development and land-use activities. SAR systems with their timely all-weather data col-
lection capability provide a useful data stream for monitoring this seasonal and annual variability
in surface water.
An example of seasonal progression comes from Old Crow Flats (Brisco et al. 2009). Here,
a gradual drop in lake levels can be seen over the course of the summer of 2000. In the spring
(see Figure 6.6a), extensive snowmelt caused high water levels, and many lakes are amalgam-
ated by flooding. As the summer progresses, the lakes become distinct (see Figure 6.6b), and
toward the end of the summer, the SAR results show the lakes shrinking and refragmenting (see
Figure 6.6c).
Comparisons of specific water bodies over long periods (years) can reveal annual changes and
long-term trends. Figure 6.7 shows the outlines of McClelland Lake, Fort McMurray, extracted
from SAR images from August 1998, 2000, and 2005 (Brisco et al. 2009). Annual fluctuations in
lake levels that are significantly greater than the pixel level accuracy of the geometric correction
accuracy can be seen.
WETLAND CLASSIFICATION WITH SAR

As pointed out by Ozesmi and Bauer (2002), wetland classification is very difficult because of a high
degree of confusion between different wetland types and between some wetland types and upland
cover. This is true in both the optical region and the microwave region of the electromagnetic (EM)
spectrum. Consequently, using multitemporal, multisource data including a DEM is the best way
(a) (b)
(c)
FIGURE 6.6 Old Crow Flats subset of RADARSAT-1 standard mode images with SAR-extracted water
vectors in red and NTDB hydrography in blue. The area shown is approximately 8 × 8 km. (a) May 25, 2000,
after extensive surface melt resulting in high surface water levels. (b) July 12, 2000, the land surface is now
dry and the water bodies are distinct. Lakes “A” and “B” are marked for comparison with 5c. (c) August 5,
2000, the SAR-extracted vectors show that water levels are slightly lower than in July, and the lakes marked
“A” and “B” have shrunk and fragmented.
to produce accurate wetland classifications. The purpose of this section is to provide an overview
of the current state of knowledge about the use of SAR alone for wetland classification. It is not a
review of wetland classification using SAR data in combination with other data sources such as opti-
cal (see Henderson and Lewis 2008 for such a review).
Many wetland classification studies have been conducted comparing single-channel SAR data
sets (Bourgeau-Chavez et al. 2001; Parmuchi et al. 2002), multifrequency data sets (Pope et al. 1997;
Costa et al. 1998; Novo et al. 2002), multipolarization and polarimetric data sets (Baghdadi et al.
2001; Grings et al. 2005, 2006; Touzi et al. 2007; Brisco et al. 2011, 2013), and even multi-incidence
angle data sets (Kandus et al. 2001). As one moves up the polarization hierarchy from single polar-
ization to dual polarization and eventually to multipolarization and both compact and full polari-
metric modes for radar system configurations, the information content generally increases (Ulaby
and Elachi 1990; Raney 2007; Charbonneau et al. 2010; Brisco et al. 2013). For a full polarimetric
FIGURE 6.7 McClelland Lake in RADARSAT-1 image from August 19, 2005. SAR-extracted lake outlines
from August 8, 1998 (yellow), August 24, 2000 (green), and August 19, 2005 (red). The area shown is approxi-
mately 3 × 4 km. Water levels in the summer of 2005 are noticeably higher than in 2000 and 1998.
data set, you have at least five independent measurements (three polarizations and the like and
cross-polarization phase difference) as opposed to only one measurement for a single-polarization
magnitude-only radar. In general, the classification accuracy improves on the order of a few percent
as you move up this polarization hierarchy with multitemporal polarimetric or compact polarimet-
ric data sets often producing better than 90% classification accuracy.
A wide range of classification algorithms have been explored as well, and as pointed out earlier
by Ozesmi and Bauer (2002), the task of wetland classification is difficult, and the results are often
site dependent due to the mix of land covers in any particular region. However, it is clear that mul-
tidimensional SAR data sets provide the data for accurate wetland classifications (Henderson and
Lewis 2008). In general, L-band is preferred to X- and C-band for woody wetland vegetation while
the shorter wavelengths are preferred for herbaceous land cover types. Polarization diversity espe-
cially including the phase term is useful as different polarizations interact with different land cover
structures in a synergistic fashion, and the phase information can help resolve magnitude ambigui-
ties by elucidating the scattering mechanism. In general, steep incidence angles are preferred to
maximize penetration of flooded canopies, but sometimes, shallow angles are complimentary, and
in the case of classifying, open water can be the best choice. The resolution/swath width choice is
most often site dependent, but in general, higher-resolution data provide better discrimination espe-
cially of the gradients between wetland types.
As pointed out earlier, the state of the art for wetland classification is multitemporal, multi-
source data. Numerous investigations have shown the synergism between optical and radar data
(Wang et al. 1998; Toyra et al. 2001; Grenier et al. 2007; Corcoran et al. 2011; Koch et al. 2012).
The use of a DEM and other land cover or topographic information can further improve the clas-
sification results as was pointed out for the surface water mapping application. Object-based or
decision tree classifications of multisource data constrained by a DEM should provide accurate
wetland classifications. Optical data have the advantage of improved spectral characterization
of wetland types, but SAR data provide more information about the vegetation structure and the
underlying moisture conditions of the soil or flooded vegetation. The combination of optical and
SAR data produces the best wetland classification. See Chapter 8 by Ramsey and Rangoonwala
for additional information.
MONITORING FLOODED VEGETATION WITH SAR

It is more difficult to map flooded vegetation with magnitude-only SAR data because a number of
other targets can have the same brightness as the flooded vegetation but the increased backscatter
from these types of targets can help delineate wetland extent (see Figure 6.3). This is difficult to do
with optical data because of the lack of penetration through the canopy. Although the use of multi-
polarization and multifrequency data can improve the results, there are still many errors of omission
and commission when using magnitude-only data. As a result, many approaches are using optical
data and topographic information to improve the delineation of flooded vegetation by identifying
topographic lows and examining the land cover to aid the identification (see Horritt et al. 2001;
Pierdicca et al. 2008; Hostache et al. 2009). This is also true for mapping ephemeral water bodies
for ecological applications or monitoring seasonal and/or annual changes in flooded vegetation due
to climate change.
Fortunately, the use of the phase information from polarimetric or compact polarimetric SAR
systems as mentioned in the classification section can overcome the limitations of mapping flooded
vegetation with magnitude-only data (Touzi et al. 2007; Brisco et al. 2013; Schmitt and Brisco 2013;
White et al. 2014). Remember it is the ability of the SAR decomposition techniques to discrimi-
nate between these different scattering mechanisms due to the phase information and in particu-
lar identify areas of double-bounce scattering that are indicative of flooded vegetation due to the
180-degree phase shift (see Figure 6.1). All of the decomposition techniques will produce output
that can identify this double-bounce scattering. The Freeman–Durden decomposition that produces
volume scattering, double-bounce scattering, and surface scattering outputs is well suited for this
approach (Freeman and Durden 1998). Brisco et al. (2013) demonstrated this in a case study show-
ing the ability to monitor changes in flooded vegetation using change detection of the double-
bounce scattering in Dongting Lake, China, using polarimetric RADARSAT-2 data. Figure 6.8
shows an example of using the Freeman–Durden decomposition to map flooded vegetation changes
Rough
Double bounce
Volume
Surface water
vectors
April 27, 2010 May 21, 2010
0 10
km
FIGURE 6.8 Freeman–Durden decomposition of RADARSAT-2 Fine Quad (FQ16) multitemporal images
over Gagetown, NB. Note the dramatic change in double-bounce scattering (green) and open water (black)
between April and May 2010 over the 24-day time period following the onset of snowmelt in this region.
in the spring after snowmelt in a temperate wetland. The extent of surface water and flooded veg-
etation both reduce as the flush from the snowmelt reduces and water level drops. Figure 6.9 shows
an example of using this approach in a change detection scheme to map the flooded vegetation in
Dongting Lake, China.
New research in the preliminary stages of validation has demonstrated how phase information
from the Touzi decomposition is being used to discriminate between bog and fen by showing where
regions of water flow in the subsurface are occurring (Touzi et al. 2011, 2013). Since bogs are char-
acterized by a lack of subsurface flow, this technique can help separate bogs from fens—a challenge
using other remote sensing techniques including optical data. This technique utilizes multitemporal
polarimetric data to monitor the phase changes of the subsurface water flow in the different wetland
types as the snowmelt flush moves through the system during the growing season. Due to the dif-
ferent hydraulic properties and water depths in the stagnant bogs versus the fens, the monitoring of
the subsurface water flow can help characterize wetland type and conditions. L-band is particularly
well suited for this approach due to the significant penetration of vegetation at this frequency, and
+10 dB
+8 dB
+6 dB
+4 dB
+2 dB
+0 dB
–2 dB
–4 dB
–6 dB
–8 dB
–10 dB
FIGURE 6.9 Change in double-bounce scattering between June 6 and August 17, 2008, RADARSAT-2
polarimetric data for Dongting Lake, China and the surrounding area. This is due to the change in SAR
scattering mechanisms as the water level drops after snowmelt runoff ceases and the vegetation is no longer
flooded.
ALOS data have been used to demonstrate this approach (Figure 6.10). Touzi scattering phase for
May 2007 and November 2006 ALOS acquisitions represents significant variations of the water
flow beneath the peat surface. The reddish/pink colors represent the presence of subsurface water
flow. The results are less promising at X- and C-band due to lower vegetation penetration.
Another interesting example of using polarimetric information for monitoring wetland condi-
tions is the segmentation of the Shannon Entropy to delineate areas of saturated soil using a mul-
titemporal RADARSAT-2 polarimetric data set (Marechal et al. 2012). The Shannon Entropy has
an entropy component that represents scattering diversity and the degree of polarization and an
intensity or total power component representing the strength of the return (Morio et al. 2007). This
technique appears to help delineate the wetland perimeter where saturated soils change to upland
soils, which along with the delineation of the flooded vegetation zone and the surface water region
using other SAR-derived information would produce an accurate and timely demarcation of wet-
land extent. This information would enhance the capability of an effective monitoring system.
Recent research suggests that the double-bounce scattering in the wetlands is not as dominate
as imagined due to the observation of high coherence in the cross-polarization channel and that
additional scattering mechanisms between the water and the wetland canopy (branches as well
as trunks) account for the coherence of wetlands in the microwave region (Hong and Wdowinski
2010). This has led to the development of a new decomposition technique especially designed for
wetland applications (Hong and Wdowinski 2014). This observation was supported by Brisco et al.
(2014) who found that double-bounce scattering in coherent wetlands was not as dominant as was
first thought and that other scattering mechanisms must be accounted for the coherence. Ongoing
π/2
–π/2
May November
FIGURE 6.10 Touzi scattering phase for May 2007 and November 2006 ALOS acquisitions representing
significant variations of the water flow beneath the peat surface. The reddish/pink colors represent the pres-
ence of subsurface water flow.
research is continuing to evaluate this new decomposition technique and better understand the
microwave scattering mechanisms from wetlands and flooded vegetation.
CONCLUSION
SAR can provide useful information about surface water and wetlands and because of the timely
nature of data acquisition should make it a key component of surface water monitoring systems. Most
flood mapping approaches currently rely on SAR data for timely data input. The same approach can
be applied to multitemporal data stacks for monitoring seasonal or annual changes in surface water.
By integrating SAR data with other geophysical data such as land cover and an accurate DEM,
this capability can be extended to map flooded vegetation as well. Likewise, research has shown
that using multifrequency or polarimetric data can help map different types of flooded vegetation.
Results have also demonstrated that using multifrequency polarimetric SAR data provides effective
input for monitoring seasonal or annual changes in wetland environments including the ephemeral
change of surface water and flooded vegetation.
Multisource data and information fusion are the state of the art for most remote sensing applica-
tions and are valid for wetland mapping and monitoring applications as well (please see Chapter 8
by Ramsey for additional details). This includes the use of in situ observations, GIS map layers, and
earth observation data (both optical and SAR data).
The combination of polarimetric and interferometric processing techniques (PoLInSAR) is a
very exciting way to improve information extraction from SAR data. For example, by using polari-
metric decomposition to isolate the scattering mechanisms in suitable wetlands that are reason-
ably coherent and highlighting the double-bounce component will allow more accurate water level
changes to be monitored. Another exciting approach is to use multifrequency decomposition tech-
niques to improve the ability to monitor a wide range of habitat types from herbaceous marshes
to woody swamps. Another promising approach involves using the Kennaugh matrix to integrate
multifrequency multipolarimetric data for this type of analyses (Schmitt et al. 2013).
Finally, scaling issues when using different remotely sensed data sets and end-user product
refinement are other issues that need to be addressed. Using multisource data is attractive, but we
must take care about the different scales involved in the different data sets and develop effective
approaches for both upscaling and downscaling. The information should be delivered to the end
user in a seamless and effective format that facilitates the uptake and use of the product by the
people or systems involved.
ACKNOWLEDGMENTS
Fred Campbell, Brian Huberty, Lori White, Kevin Murnaghan, and Frank Ahern provided scientific
and editorial peer review of the chapter and helpful comments to improve content and readability.
Support for this work came from the Remote Sensing Science Program at Earth Science Sector/
Natural Resources Canada. ESS Contribution number/Numéro de contribution du SST: 20130294.
REFERENCES
Alsdorf, D., Smith, L., and J. Melack. 2001. Amazon floodplain water level changes measured with interfero-
metric SIR-C radar. IEEE Transactions on Geoscience and Remote Sensing 39:423−431.
Baghdadi, N., Bernier, M., Gauthier, R., and I. Neeson. 2001. Evaluation of C-band SAR data for wetlands
mapping. International Journal of Remote Sensing 22:71–88.
Bourgeau-Chavez, L., Kasischke, E., Brunsell, S., Mudd, J., Smith, K., and A. Frick. 2001. Analysis of space-
borne SAR data for wetland mapping in Virginia riparian ecosystems. International Journal of Remote
Sensing 22:3665–3687.
Brisco, B., Ahern, F., Murnaghan, K., White, L., and P. Lancaster. 2014. InSAR for monitoring water level
changes in Canadian wetlands. Submitted to RSE.
Brisco, B., Kapfer, M., Hirose, T., Tedford, B., and J. Liu. 2011. Evaluation of C-band polarization diversity
and polarimetry for wetland mapping. Canadian Journal of Remote Sensing 37:82–92.
Brisco, B., Kun, L., Tedford, B., Charbonneau, F., Shao Y., and K. Murnaghan. 2013. Compact polarimetry
assessment for rice and wetland mapping. International Journal of Remote Sensing 34:1949–1964.
Brisco, B., Short, N., van der Sanden, J.J., Landry, R., and D. Raymond. 2009. A semi-automated tool for sur-
face water mapping with RADARSAT-1. Canadian Journal of Remote Sensing 35:336–344.
Brisco, B., Touzi, R., van der Sanden, J.J., Charbonneau, F., Pultz, T.J., and M. D’Iorio. 2008. Water resource
applications with RADARSAT-2—A preview. International Journal of Digital Earth 1(1):130–147.
Charbonneau, F., Brisco, B., Raney, K., McNairn, H., Chen, L., Vachon, P., Shang, J. et al. 2010. Compact
polarimetry overview and applications assessment. Canadian Journal of Remote Sensing 36:S298–S315.
Cloude, S.R. and E. Pottier. 1997. An entropy based classification scheme for land applications of polarimetric
SAR. IEEE Transactions on Geoscience Remote Sensing 35:68–78.
Corcoran, J., Knight, J., Brisco, B., Kaya, S., Cull, A., and K. Murnaghan. 2011. The integration of optical,
topographic, and radar data for wetland mapping in northern Minnesota. Canadian Journal of Remote
Sensing 37:564–582.
Costa, M., Novo, F., Ahern, E., Mitsuo, E. II, Mantovani, J., Allester, M., and R. Pietsch. 1998. The Amazon
flood plain through radar eyes: Lago Grande de Monte Alegre case study. Canadian Journal of Remote
Sensing 24:339–349.
Dellepiane, S.G. and E. Angiati. 2012. A new method of cross-normalization and multi-temporal visualization
of SAR images for the detection of flooded areas. IEEE Transactions on Geoscience and Remote Sensing
50:2765–2779.
Freeman, A. and S.L. Durden. 1998. A three-component scattering model for polarimetric SAR data. IEEE
Transactions on Geoscience and Remote Sensing 36:963–973.
Grenier, M., Demers, A.M., Labrecque, S., Benoit, M., Fournier, R.A., and B. Drolet. 2007. An object-based
method to map wetlands using RADARSAT-1 and Landsat ETM images: Test case on two sites in
Quebec, Canada. Canadian Journal of Remote Sensing 33:S28–S45.
Grings, F.M., Ferrazzoli, P., Jacobo-Berlles, J.C., Karsenbaum, H., Tiffenberg, J., Pratolongo P., and P. Kandus.
2006. Monitoring flood conditions in marshes using EM models and Envisat ASAR observations. IEEE
Grings, F.M., Ferrazzoli, P., Karsenbaum, H., Tiffenberg, J., Kandus, P., Guerriero, L., and J.C. Jacobo-
Berlles. 2005. Modelling temporal evolution of Junco marshes radar signatures. IEEE Transactions on
Henderson, F.M. and A.J. Lewis. 1998. Manual of Remote Sensing: Principles and Applications of Imaging
Radar. Wiley, New York.
Henderson, F.M. and A.J. Lewis. 2008. Radar detection of wetland ecosystems: A review. International Journal
of Remote Sensing 29:5809–5835. doi: 10.1080/01431160801958405.
Hess, L.L., Melack, J.M., and Simonett, D.S. 1990. Radar detection of flooding beneath the forest canopy:
Hong, S.-H. and S. Wdowinski. 2010. Rotated dihedral and volume scattering behavior in cross-polarimetric
SAR. In: IEEE International Geoscience and Remote Sensing Symposium 2010, Honolulu, HI.
Hong, S.-H. and S. Wdowinski. 2014. Double bounce component in cross-polarimetric SAR from a new scat-
tering target decomposition. IEEE Transactions on Geoscience and Remote Sensing 52(6):3039–3051.
Hong, S.-H., Wdowinski, S., Kim, S.W., and J.S. Won. 2010. Multi-temporal monitoring of wetland water
levels in the Florida Everglades using interferometric synthetic aperture radar (INSAR). Remote Sensing
of the Environment 114:2436–2447.
Horritt, M.S., Mason, D.C., and A.J. Luckman. 2001. Flood boundary delineation from synthetic aper-
ture radar imagery using a statistical active contour model. International Journal of Remote Sensing
22:2489–2507.
Hostache, R., Matgen, P., Schumann, G., Puech, C., Hoffman, L., and L. Pfister. 2009. Water level estimation
and reduction of hydraulic model calibration uncertainties using satellite SAR images of floods. IEEE
Kandus, P., Karsenbaum, H., Pultz, T., Parmuchi, G., and J. Bava. 2001. Influence of flood conditions and
vegetation status on the radar backscatter of wetland ecosystems. Canadian Journal of Remote Sensing
27:651–662.
Kapur, J.N., Sahoo, P.K., and A.K.C. Wong. 1985. A new method for gray-level picture thresholding using the
entropy of the histogram. Computer Vision Graphics and Image Processing 29:273–285.
Kasischke, E.S. and L.L. Bourgeau-Chavez. 1997. Monitoring South Florida wetlands using ERS-1 SAR imag-
Kim, J.W., Lu, Z., Lee, H., Shum, C.K., Swarzenski, C.M., Doyle, T.W., and S.H. Baek. 2009. Integrated analy-
sis of PALSAR/Radarsat-1 InSAR and ENVISAT altimeter data for mapping of absolute water level
changes in Louisiana wetlands. Remote Sensing of the Environment 113:2356–2365.
Koch, M., Schmid, T., Reyes, M., and J. Gumuzzio. 2012. Evaluating full polarimetric C- and L-band data
for mapping wetland conditions in a semi-arid environment in Central Spain. IEEE Journal of Selected
Topics in Applied Earth Observation and Remote Sensing 5:1033–1044.
Lu, Z., Crane, M., Kwoun, O. I., Wells, C., Swarzenski, C., and R. Rykhus. 2005. C-band radar observes water
level change in swamp forests. EOS 86:141−144.
Lu, Z. and O. I. Kwoun. 2008. Radarsat-1 and ERS InSAR analysis over southeastern Coastal Louisiana:
Implications for mapping water-level changes beneath swamp Forests. IEEE Transactions on Geoscience
and Remote Sensing 46(8):2167–2184.
MacDonald, H.C., Waite, W.P., and J.S. Demarcke. 1980. Use of Seasat satellite radar imagery for the detection
of standing water beneath forest vegetation. In: Proceedings of the American Society of Photogrammetry
Annual Technical Meeting, RS-3(B), New York, pp. 1–13.
Marechal, C., Pottier, E., Huberty-Moy, L., and Rapinel, S. 2012. One year wetland survey investigations from
quad-pol RADARSAT-2 time-series SAR images. Canadian Journal of Remote Sensing 38(3):240–252.
Matgen, P., Hostache, R., Schumann, G., Pfister, L., Hoffman, L., and H.H.G. Savenije. 2011. Towards an
automated SAR-based flood monitoring system: Lessons learned from two case studies. Physics and
Chemistry of the Earth 36:241–252.
Morio, J., Refregier, P., Goudail, F., Dubois-Fernandez, P., and X. Dupuis. 2007. Application of information
theory measures to polarimetric and interferometric SAR images. In: Program and System Information
Protocol 2007, Mulhouse, France.
National Wetlands Working Group. 1997. The Canadian Wetland Classification System, 2nd edn., Warner, B.G.
and C.D.A. Rubec (eds.), Wetlands Research Centre, University of Waterloo, Waterloo, Ontario, Canada.
Novo, E.M.L.M., Costa, M.P.F., Mantovani, J.E., and I.B.T. Lima. 2002. Relationship between macrophyte
stand variables and radar backscatter at L and C band, Tucuruí Reservoir, Brazil. International Journal
of Remote Sensing 23:1241–1260.
10:381–402.
Parmuchi, M.G., Karszenbaum, H., and Kandus, P. 2002. Mapping wetlands using multi-temporal radarsat-1
data and a decision-based classifier. Canadian Journal of Remote Sensing 28(2):175–186.
Pierdicca, N., Chini, M., Pulvirenti, L., and F. Macina. 2008. Integrating physical and topographic information
into a fuzzy scheme to map flooded area by SAR. Sensors 8:4151–4164.
Pope, K.O., Rejmankova, E., Paris, J.F., and R. Woodruff. 1997. Detecting seasonal flooding cycles in marches
of the Yucatan Peninsula with SIR-C polarimetric radar imagery. Remote Sensing of the Environment
59:157–166.
Pultz, T.J., Crevier, Y., Brown, R.J., and J. Boisvert. 1997. Monitoring of local environmental conditions with
SIR-C/X-SAR. Remote Sensing of the Environment 59:248–255.
Pulvirenti, L., Pierdicca, N., Chini, M., and L. Guerrierro. 2011. An algorithm for operational flood mapping
from synthetic aperture radar (SAR) data using fuzzy logic. Natural Hazards and Earth Systems Sciences
11:525–540.
Raney, R.K. 2007. Hybrid-polarity SAR architecture. IEEE Transactions on Geoscience and Remote Sensing
45:3397–3404.
Schmitt, A. and B. Brisco. 2013. Wetland monitoring using the Curvelet-based change detection method on
polarimetric SAR imagery. Accepted for publication in special issue “Advances in Remote Sensing of
Flooding.” Water 5:1036–1051. doi:10.3390/w5031036.
Schmitt, A., Roth, A., and B. Brisco. 2013. Multi-frequency and multi-temporal Kennaugh decomposition for
wetland monitoring. In: Ninth ASAR Workshop, October 15–18, Montreal, Quebec, Canada.
Smith, L.C. 1997. Satellite remote sensing of river inundation area, stage, and discharge: A review. Hydrological
Processes 11(10): 1427–1439.
Touzi, R., Boerner, W.M., Lee, J.S., and E. Luenberg. 2004. A review of polarimetry in the context of synthetic
aperture radar: Concepts and information extraction. Canadian Journal of Remote Sensing 30:380–407.
Touzi, R., Deschamps, A., and G. Rother. 2007. Wetland characterization using polarimetric RADARSAT-2
capability. Canadian Journal of Remote Sensing 33:S56–S67.
Touzi, R., Gosselin, G., and R. Brooks. 2013. Peatland subsurface water flow monitoring using polarimetric
L-band. In: International Geoscience and Remote Sensing Symposium, Honolulu, HI.
Touzi, R., Gosselin, G., Li, J., and R. Brook. 2011. Peatland subsurface water flow monitoring using polarimet-
ric L-band ALOS. In: Proceedings on POLINSAR’11, Frascatti, Italy.
Townsend, P.A. 2002. Relationship between forest structure and the detection of flood inundation in forested
wetlands using C-band SAR. International Journal of Remote Sensing 23:443–460.
Toyra, J., Pietroniro, A., and L. Martz. 2001. Multisensor hydrologic assessment of freshwater wetland. Remote
Ulaby, F.T. and C. Elachi. 1990. Radar Polarimetry for Geoscience Applications. Artech House, Norwood, MA.
Ulaby, F.T., Moore, R.K., and A.K. Fung. 1986. Microwave Remote Sensing: Active and Passive, Vols. 1–3.
Artech House, Dedham, MA.
Van Zyl, J.J. 1989. Unsupervised classification of scattering behavior using radar polarimetry data. IEEE
Wang, J., Shang, J., Brisco, B., and R. Brown. 1998. Evaluation of multidate ERS-1 and multispectral Landsat
imagery for wetland detection in Southern Ontario. Canadian Journal of Remote Sensing 24:60–68.
Wdowinski, S., Kim, S.W., Amelung, F., Dixon, T.H., Miralles-Wilhelm, F., and R. Sonenshein. 2008. Space-
based detection of wetlands’ surface water level changes from L band SAR interferometry. Remote
White, L., Brisco, B., Pregitzer, M., Tedford, B., and L. Boychuk. 2013. RADARSAT-2 beam mode selection
for surface water and flood mapping. Submitted to CJRS.
White, L., Brisco, B., Pregitzer, M., Tedford, B., and L. Boychuk. 2014. RADARSAT-2 beam mode selection
for surface water and flooded vegetation mapping. Canadian Journal of Remote Sensing 40(2):135–151.
doi.org/10.1080/07038992.2014.943393.
White, L., Kaya, S., and B. Brisco. December 2011. Surface water information extraction from RADARSAT-2
fine quad data. Canada Centre for Remote Sensing Technical Report, Ottawa, ON, Canada.
Yamaguchi, Y., Moriyama, T., Ishido, M., and H. Yamad. 2005. Four-component scattering model for polarimet-
ric SAR image decomposition. IEEE Transactions on Geoscience and Remote Sensing 43:1699–1706.
7 A Review of the Technique
Wetland InSAR
and Applications
Shimon Wdowinski and Sang-Hoon Hong
CONTENTS
Introduction..................................................................................................................................... 137
Synthetic Aperture Radar Observations.......................................................................................... 138
Wetland InSAR............................................................................................................................... 140
Calibration and Validation.............................................................................................................. 145
Interferometric Coherence of Wetlands.......................................................................................... 147
InSAR Time Series......................................................................................................................... 147
Wetland InSAR Studies of Various Wetland Environments............................................................ 149
Yellow River Delta Wetlands..................................................................................................... 149
Louisiana Wetlands.................................................................................................................... 151
Sian Ka’an Biosphere Reserve................................................................................................... 151
Danube Delta Wetlands.............................................................................................................. 151
Discussion....................................................................................................................................... 151
Conclusion...................................................................................................................................... 152
References....................................................................................................................................... 153
INTRODUCTION
Wetlands are often very productive ecosystems, which can flourish due to water availability, nutrient
cycling, and the sun’s energy. They provide critical habitat for a wide variety of plant and animal spe-
cies, including the larval stages of many ocean fish. Wetlands also deliver a wide range of important
services, including water supply, water purification, carbon sequestration, coastal protection, and out-
door recreation (see Chapter 1). Globally, many such regions are under severe environmental stress,
mainly from agriculture and urban developments, pollution, and rising sea level. However, there is an
increasing recognition of the importance of these habitats, and mitigation and restoration activities
have begun in a few regions. A key element in wetland conservation, management, and restoration
involves hydrological monitoring as the entire ecosystem depends on its water supply. Heretofore,
hydrological monitoring of wetlands is mostly conducted by stage (water level) measurements at
gauging stations, which provides good temporal resolution but suffers from poor spatial resolution,
as stage gauging stations are typically distributed several or even tens of kilometers from one another.
Remote sensing observations, in particular satellite imagery, serve as very useful tools for char-
acterizing spatial phenomena, such as land cover and its changes over time. Optical and radar
imageries have been widely used to detect and monitor wetlands, mainly for classifying vegetation
and estimating biological parameters, like aboveground biomass (Evans and Costa 2013; Simard
et al. 2006). Most of these remote sensing techniques cannot detect water level changes in wet-
lands, which occur beneath the vegetation cover. The one technique that is sensitive to water level
137
(i.e., height) changes in vegetated aquatic environments is wetland InSAR (interferometric syn-
thetic aperture radar). This technique provides detailed maps of water level changes between two
acquisition times and can be used to detect water level changes in various wetland environments
(Wdowinski et al. 2004, 2008). The method has been successfully applied to study wetland hydrol-
ogy in the Everglades (Hong et al. 2010a; Wdowinski et al. 2004, 2008), Louisiana (Kim et al. 2009;
Kwoun and Lu 2009), and other locations (e.g., Gondwe et al. 2010).
This chapter provides a review of a wetland InSAR technique and applications, which are based
on space-based synthetic aperture radar (SAR) observations. The first part of the chapter explains
the differences between SAR and InSAR and provides technical information on available SAR data
and satellites. The main part of the chapter focuses on the wetland InSAR technique, by presenting
a brief physical explanation of the method, quality assessment of the results, and advanced InSAR
time series methods for monitoring water level changes over time. The review ends with several
examples where the method was applied for studying wetland hydrology around the globe and dis-
cusses future usage of the application.
SYNTHETIC APERTURE RADAR OBSERVATIONS

Space-based SAR is a very reliable technique for monitoring changes in both the terrestrial and
aquatic surfaces of the Earth. SAR measures two independent observables, backscattered ampli-
tude and phase, over a wide swath (10–400 km) with a pixel resolution of 1–100 m depending on
the satellite acquisition parameters. Backscattered amplitude, which is often presented as gray-
scale images of the surface (Figure 7.1b), is very sensitive to surface dielectric properties, surface
–120° –110° –100° –90° –80° –70°
W82° W81°
40° 40°
South
Florida
30° 30°
N29°
(a) –120° –110° –100° –90° –80° –70°
N28°
N27°
N26°
N25°
(b) (c)
FIGURE 7.1 (a) Location map of our study area in south Florida. (b) Radarsat-1 ScanSAR image of study
area (Radarsat data © Canadian Space Agency/Agence spatiale canadienne 2002. Processed by CSTARS
and distributed by Radarsat International). (c) Cartoon illustrating the double-bounce radar signal return in
vegetated aquatic environments. The red ray bounces twice and returns to the satellite, whereas the black ray
bounces once and scattered away (specular reflectance).
Wetland InSAR 139
ALOS PALSAR Radarsat-1

16-Jul-06/1-Dec-06 15-Oct-07/8-Nov-07
LO
level incre
er
t
as
Wa
e
0 π
TerraSAR-X
an as 17-May-08/28-May-08
e
R
km ge in cre
(a) (b) (c) (d)
FIGURE 7.2 Interferograms showing phase changes, which were induced by water level changes in the
Everglades wetlands, south Florida. (a) JERS-1 amplitude image of south Florida showing the location of the
three interferograms. (b) L-band (24 cm wavelength) ALOS interferogram of 90 km wide ascending track.
Each color cycle corresponds to 15 cm of water level change. (c) C-band (5.6 cm) Radarsat-1 interferogram of
a 75 km wide descending swath. Each color cycle represents 4 cm of water level change. (d) X-band (3.1 cm)
TSX interferogram of a 30 km wide descending swath. Some of the observed changes reflect changes in
atmospheric moisture between the acquisitions. Each color cycle represents 2 cm of water level changes. LO,
Lake Okeechobee; EAA, Everglades Agricultural Area; WCA, Water Conservation Area; ENP, Everglades
National Park. The white boxes mark the location of a zoomed-in area shown in Figure 7.4.
inclination toward the satellite, and wave direction in oceans (Bragg scattering). Amplitude images
are widely used for studying land cover classification, soil moisture content, ocean waves, oil spill
detection, and many other applications.
The second observable, backscattered phase measures the fraction of the radar wavelength that
returns to the satellite’s antenna. It is mainly sensitive to the range between the surface and the
satellite but also to atmospheric conditions and changes in the surface dielectric properties. Phase
data are mainly used in interferometric calculations (InSAR) for detecting centimeter-level dis-
placements of the surface (Figure 7.2). The method compares pixel-by-pixel SAR phase obser-
vations of the same area acquired at different times from roughly the same location in space to
produce high-spatial-resolution displacement maps. Such maps, termed interferograms (Figure 7.2b
through d), are widely used in studies of earthquake-induced crustal deformation, magmatic activ-
ity (volcanoes), land subsidence due to water extraction, glacier movements, and more (Wdowinski
and Eriksson 2009).
InSAR phase observations reflect distance changes in the line-of-sight (LOS) direction between
the surface and the satellite. Because the SAR measurements are conducted in a slanted range
geometry (Figure 7.1b), the LOS phase observations reflect a combined measure of both horizontal
and vertical surface movements. When horizontal movements are negligible, the LOS phase sig-
nal can be translated to vertical surface movements, based on the radar half-wavelength and the
acquisition geometry (vertical change = half-wavelength/[cos(incidence angle)]). We use the half-
wavelength, because SAR satellites measure time differences between transmitted and received
signals, which reflect two-way distance change. Because LOS measures range change between the
satellite and the surface, an increase in LOS reflects surface subsidence, and, vice versa, a decrease
in LOS reflects surface uplift.
Spaceborne SAR data have been acquired since the late 1970s by several satellites using various
systems and acquisition parameters (Table 7.1). Two of the key parameters that determine the inter-
action of the radar signal with vegetation are the radar wavelength (or frequency) and polarization.
TABLE 7.1
Space-Based SAR Sensors
Satellite Wavelength and Polarization Agency Time
SeaSAT L-band, HH polarization (pol) DoD 1978
ERS-1 C-band, VV pol ESA 1992–1996
ERS-2 (SAR) C-band, VV pol ESA 1996–2011
JERS-1 L-band, HH pol JAXA 1992–1998
Radarsat-1 C-band, HH pol CSA 1995–2013
Space shuttle (SRTM) X-, C-, and L-band, fixed baseline interferometer NASA 2000
Envisat (ASAR) C-band, dual pol ESA 2002–2012
ALOS (PALSAR) L-band, dual pol, quad pol JAXA 2006–2011
Radarsat-2 C-band, quad pol CSA 2007–present
TerraSAR-X/TanDEM-X X-band, dual pol, quad pol DLR 2007–present
COSMO-SkyMed X-band, dual pol ASI 2007–present
Source: Wdowinski, S. and Eriksson, S., EOS Trans. Am. Geophys. Union, 90, 153, 2009.
Agencies: ASI, Italian Space Agency; CSA, Canadian Space Agency; DLR, German Aerospace Center; DoD,
Department of Defense (United States); ESA, European Space Agency; JAXA, Japan Aerospace
Exploration Agency; NASA, National Aeronautics and Space Administration (United States).
Most SAR satellites have operated in three frequencies: X-band (wavelength 3.1 cm), C-band
(5.6 cm), and L-band (24 cm). It is commonly accepted that the shorter wavelength X-band signal
interacts mainly with upper sections of the vegetation, the intermediate C-band signal penetrates
further and can penetrate the entire canopy under some circumstances, and the L-band signal can
penetrate throughout the vegetation and interact with the surface beneath the vegetation. X-band
satellites include the German TerraSAR-X (TSX) and TanDEM-X (TDX) and the Italian COSMO-
SkyMed (CSK) constellation. C-band satellites are the ERS-1, ERS-2, and Envisat operated by the
European Space Agency and the Radarsat-1 (RSAT-1) and Radarsat-2 (RSAT-2) operated by the
Canadian Space Agency. L-band satellites include the Japanese JERS-1 and ALOS-1 satellites,
which ceased operation, and the ALOS-2 satellite in which its launch is scheduled for 2014.
Radar polarization determines the direction at which the SAR signal is transmitted from and
received at the satellite’s antenna. The first generation of SAR satellites operated in a single polar-
ization mode, such as HH (horizontal transmitted and horizontal received) by RSAT-1 or VV (verti-
cal transmitted and vertical received) by ERS-1/2. The second generation of SAR satellites, such as
Envisat and ALOS, already operated with dual polarization mode, such as HH + HV (horizontally
transmitted and two types of receptions, horizontal and vertical) or VV + VH. RSAT-2 is the only
civilian satellite that operates in a full quadruple mode (quad-pol) that acquires data in four inde-
pendent channels (HH, HV, VH, and VV). ALOS and TSX have been operated in experimental
quad-pol modes for short time periods and acquired limited amounts of quad-pol data.
WETLAND InSAR
Wetland InSAR is a unique application of the InSAR technique that detects elevation changes of
aquatic surfaces; all other InSAR applications detect displacements of solid surfaces. The technique
works, because the radar pulse is backscattered twice (“double bounce”) from the water surface and
vegetation (Richards et al. 1987; Figure 7.1c). The method was first used with L-band data, which
works better in vegetated environments. Alsdorf et al. (2000) analyzed space shuttle L-band data
to study water level variations in the Amazon floodplain. Wdowinski et al. (2004, 2008) also used
JERS-1 L-band observations of Everglades wetlands for detecting water flow changes throughout
Wetland InSAR 141
(a) (b)
(c) (d)
FIGURE 7.3 Photographs of (a) freshwater herbaceous vegetation (saw grasses); (b) mixed
vegetation—woody vegetation in the tree island and herbaceous around the islands—(c) freshwater woody
vegetation (cypress); and (d) saltwater woody vegetation (mangroves) that grows along a tidal channel.
the wetlands. Although X- and C-band radar signals are often assumed to interact mainly with
upper sections of the vegetation, several studies show that both data types are also suitable for the
wetland InSAR application (Gondwe et al. 2010; Hong et al. 2010a,b; Lu 2005).
We present the principles, methodology, and results of wetland InSAR using examples from our
main study area, the Everglades in south Florida. The Everglades can be viewed as a large-scale
laboratory for testing remote sensing hydrological techniques, because it extends over a wide area,
consists of a variety of vegetation types (herbaceous, woody, freshwater, and saltwater—Figure 7.3),
and contains both controlled and natural hydrological conditions. Furthermore, a dense network of
stage gauging stations located throughout the Everglades provides excellent ground truthing data
for calibrating and validating the wetland InSAR results. Two sets of interferograms are presented:
(1) the freshwater environment, which is dominated by wide and slow sheet flow, and (2) the coastal
wetland environment, which is dominated by daily tidal flow. The data were processed using the
software package ROI_PAC (Pritchard et al. 2014), which generates interferograms from pairs of
the same sensor data.
The freshwater interferograms cover an extensive wetland area located between Lake Okeechobee
and the Gulf of Mexico—known as the “River of Grass”—prior to its massive development (Douglas
1947). Today, the area includes the Everglades Agricultural Area (EAA), a series of managed wet-
lands called Water Conservation Areas (WCAs), and Everglades National Park (ENP) with remnants
of the natural flowing freshwater system (Figure 7.2b). The freshwater interferograms show high
variability in the pattern of the detected phase changes, often called fringes, depending on the sen-
sor type and hydrological conditions occurring during the two acquisition times (Figure 7.2). The
L-band (24 cm wavelength) ALOS interferograms show that most phase changes occur within a
range of a single fringe cycle. Furthermore, some of the fringes are discontinuous (sharp color
change) across physical structures, which control hydrology (e.g., levees) reflecting different water
level changes across hydrological structures. The C-band interferogram also shows fringe disconti-
nuity across hydrological structures but with a larger number of fringes (Figure 7.2c). The X-band
interferogram shows even larger number of fringes, mostly in the WCAs, but some also in the urban
area located east of the WCAs. The fringes in the urban area and some in the WCAs reflect changes
in the atmospheric precipitable water (moisture) level, which can delay the arrival of the radar signal.
The fringe density and fringe pattern detected in the three interferograms (Figure 7.2b through d)
reflect both the sensor’s wavelength and the change in water levels between the two acquisition times.
As indicated earlier, for a given acquisition geometry (incidence angle), the measured phase change
is proportional to the half-wavelength of radar signal. In the example shown in Figure 7.2, one fringe
cycle (2π) of L-band signal reflects 15 cm of vertical displacement, 4 cm in C-band interferograms,
and 2 cm in X-band. Thus, 12 cm of water level change will be reflected as less than a single L-band
fringe cycle (about 0.8 of a cycle), 3 C-band fringes, and 6 X-band fringes. Since SAR satellites
acquire their data at a speed of ~7 km/s, data acquisition over a 100 km long area takes about ~15 s.
Consequently, the observed phase change reflects the difference between two acquisition snapshots
and not an averaged difference. In some areas, where water level changes occur gradually over long
time periods, observed fringes may reflect an average rate of change. But in a dynamic wetland envi-
ronment, when water levels change frequently due to heavy rain, water management, or ocean tides,
InSAR observations measure the difference between the two acquisition snapshots.
Interferograms are color maps displaying phase changes in the range of 0–2π. In order to under-
stand the significance of these maps and their use for calculating water level changes, we zoom into
WCA-1 (Figure 7.4). Four interferograms of this area are presented, each with a different fringe pat-
tern and fringe density. The ALOS interferogram (Figure 7.4a) shows variations within one fringe
cycle, in which the phase in the interior is higher than along the area’s boundaries (according to
the scale in Figure 7.2, the range increases from yellow to pink). This phase change pattern of high
phase in the interior and lower phase near the boundaries can be easily seen in the phase profile
plot across the A–A′ transect (second row in Figure 7.4a). The first RSAT-1 interferogram shows a
roughly N–S fringe pattern, in which phase increases from west to east (Figure 7.4b). Because the
phase change wrapped in the range of 0–2π, the phase change is discontinuous and has a sawtooth
shape pattern (second row in Figure 7.4b). The second RSAT-1 interferogram shows a radial fringe
pattern, in which phase increases from the area’s boundaries toward the interior (Figure 7.4c). In
this case, the sawtooth pattern changes from a westward slope (phase increase from west to east)
in the western side of the area to an eastward slope in the eastern side (second row in Figure 7.4c).
The TSX interferogram shows a combination of radial and linear patterns, which results in a less
symmetrical sawtooth pattern (second row in Figure 7.4d).
The discontinuous phase change pattern occurs because the InSAR-detected phase change is
wrapped in the range of 0–2π. In order to remove these discontinuities, the phase changes should
be unwrapped and expanded over a larger range. The unwrapping procedure is demonstrated in the
third row of Figure 7.4, where the phase change pattern is continuous and extends over the range of
0–π (L-band), 0–14π (first C-band), 0–7π (second C-band), and 0–5π (X-band). All the unwrapped
profiles show a phase change decrease from left (west) to right (east) and some increase at the eastern
part of the profile. As InSAR calculations correct for a possible change in the SAR satellite positions
during the two acquisitions, the observed phase change reflects surface changes with respect to the
satellite. A phase change increase indicates surface subsidence, whereas a phase change decrease
indicates uplift. The actual amount of subsidence or uplift depends on the sensor’s wavelength and
the acquisition geometry. For the four interferograms presented in Figure 7.4, one fringe cycle (2π)
Wetland InSAR 143
ALOS PALSAR Radarsat-1 Radarsat-1 TerraSAR-X

16-Jul-06/01-Dec-06 02-Sep-06/26-Sep-06 07-Dec-06/31-Dec-06 17-May-08/28-May-08
A A A A
A A A A
(a) (b) (c) (d)
Wrapped phase
2π
Unwrapped phase
14π
Water level change (cm)

0
–30
Distance (km)
FIGURE 7.4 Phase change and inferred water level changes in WCA-1. Upper row presents ALOS (a),
Radarsat-1 (b and c), and TSX (d) interferograms of WCA-1. Second row illustrates phase changes along the
A–A′ transect. The phase is wrapped in the range of 0–2π. The third row illustrates the unwrapped phase,
which extends over a larger phase range. The fourth row illustrates the inferred vertical change in water level,
which depends on the SAR signal’s wavelength and acquisition geometry.
of the L-, C-, and X-bands corresponds to 15, 4, and 2 cm of vertical surface changes, respectively.
The conversion of the phase information to water level changes involves both reversing the fringe
shape pattern and multiplication by the sensor’s conversion factor (fourth row in Figure 7.4). In three
of the four examples presented in Figure 7.4, the water level change pattern is similar indicating a
relative subsidence of the interior part of WCA-1 with respect to the area located near the boundar-
ies. But since InSAR is a relative measurement, the same pattern can also represent a relative uplift
of the boundary area with respect to the wetland interior. The water level change pattern in Figure
7.4b shows an overall westward decrease of water levels throughout the area.
The second set of interferograms presented is of coastal wetlands located in the western
Everglades (Figure 7.5). The interferograms show a less organized fringe pattern than those in
the flow-controlled area (WCAs) (Figure 7.2). The L-band interferogram (Figure 7.5b) shows two
ALOS 8-Aug-10/23-Sep-10
level incre
ter a
se
W0 π
an
se
ge in crea
R
(a) (b)
RSAT-2 23-Sep-08/17-Oct-08 TSX 26-Sep-08/18-Oct-08
(c) (d)
FIGURE 7.5 (a) Google Earth composite satellite image showing the location of the three interferograms used
in the coastal wetland area. (b) ALOS interferogram showing a coherent fringe along the transition between
fresh- and saltwater vegetation and partial fringe along the tidal channels. (c) Radarsat-2 interferogram show-
ing short wavelength fringes surrounding tidal channels in the mangrove forest area. (d) TerraSAR-X inter-
ferogram showing a similar fringe patterns around the tidal channels. The different number of fringes in each
interferogram reflects the sensitivity of each sensor to surface changes. Each ALOS (L-band) fringe reflects
15 cm of water level change, Radarsat-2 (C-band) 4 cm, and TerraSAR-X (X-band) 2 cm.
fringes that follow geographic features and most likely reflect water level changes, which are (1)
a linear roughly N–S oriented fringe located east of the Tarpon Bay and (2) a rounded fringe sur-
rounding the Tarpon Bay. The location and orientation of the linear fringe follow the transition
zone between fresh- and saltwater vegetation reflecting tide-induced water level changes in the salt
marshes. The rounded fringe around the Tarpon Bay has a more pronounced appearance in the
other interferograms and is discussed in detail in the following.
The RSAT-2 and TSX interferograms show a more complex fringe pattern (Figure 7.5c and d)
that can be characterized by the following three zones: low fringe gradient in the northeastern
corner, incoherent phase (fuzzy pattern) in the southwest corner, and wide areas in between with
high fringe gradient. The low fringe gradient in the northeast corner occurs in the freshwater
wetlands. The low gradient reflects slow changes in the freshwater sheet flow and possibly some
atmospheric signals. The incoherent signal in the southwest corner occurs over tall mangrove
Wetland InSAR 145
vegetation, which produces unstable scattering in the C- and X-band SAR signals. The high fringe
gradient in the transition zone occurs over variable vegetation types, including intermediate and
short mangrove forests.
The fringes that surround Tarpon Bay and other bodies of water reflect water level changes that
occur due to the high contrast between fast flow in the channel and slow flow through the saltwater
vegetation (Wdowinski et al. 2013). These high fringe gradients mark the extent of the tidal flushing
zone, which occurs over a 2–3 km wide area on each side of the tidal channel. Unwrapping the phase
changes along the Tarpon Bay in both C- and X-band interferograms indicates a total of 12 cm water
level change, about 25% of average daily tidal variation in Tarpon Bay (40–50 cm). It is interesting
to note that the InSAR-detected slope (12 cm/2.5 km) is very small, ~5 × 10 −5, which is less than a
1/100 of a percent. Such a small slope cannot be detected from the ground but can be easily mea-
sured by InSAR from space.
CALIBRATION AND VALIDATION

Interferograms provide high-spatial-resolution (10–300 m pixel resolution) maps of surface water
level changes over a broad wetland area. In order to utilize these space-based observations for
hydrological applications, additional information is needed because the InSAR measurements are
relative in both time and space. In time, the measurements provide the change in water level
(not the actual water level) that occurred between the data acquisition times. In space, the mea-
surements describe the relative change of water levels in the entire interferogram with respect to
a zero change at an arbitrary reference point, because the actual range between the satellite and
the surface cannot be determined accurately. However, relative changes between pixels can be
determined at the centimeter level. In many other InSAR applications, such as earthquake- or
volcano-induced deformation, the reference zero change point is chosen to be in the far field,
where changes are known to be negligible (Massonnet et al. 1993). However, in wetland InSAR,
the assumption of zero surface change in the far field does not hold, especially in the Everglades,
because flow and water levels can be discontinuous across the various water-control structures or
other flow obstacles.
In order to utilize the high-spatial-resolution InSAR observations, independent observations of
water levels can be used for calibrating and validating the InSAR observations. The simplest cali-
bration method relies on ground-based stage data as introduced by Wdowinski et al. (2004, 2008).
When stage data are not available, calibration can be conducted using space-based altimetry data
that were acquired during the same time span as the InSAR data (Kim et al. 2009). Here, we briefly
describe the stage-based calibration technique using an example of an RSAT-1 interferogram from
south Florida (Figure 7.6a) and stage data that are available at the SFWMD’s DBHYDRO database
(SFWMD 2014; for more details of the calibration method, see Wdowinski et al. 2008).
The stage-based calibration procedure involves the following four steps: phase unwrapping, stage
data retrieval, InSAR–stage data comparison, and InSAR data adjustment. The first step is phase
unwrapping for calculating the InSAR-determined water level changes with respect to an arbitrary
reference point that is set with a value of zero. The phase unwrapping is conducted independently
for each body of water (WCA-1, 2A, 2B, 3A, 3B, ENP), as water level changes can be discontinuous
across barriers (e.g., levees). For each stage station location (Figure 7.6c), the InSAR-derived water
level change is calculated by averaging values from nine pixels surrounding the station location. In
parallel (step two), the stage-derived value is calculated by subtracting the stage record from the
two acquisition dates. The comparison between the InSAR- and stage-derived water level changes
(step three) is conducted separately for each body of water, as shown in Figure 7.6b. In all areas, the
data are fitted using a straight line with a slope of one (y = x + b), in order to determine the y-axis
intercept value, which is used to calibrate the InSAR data. In the final step, we subtract the intercept
value from the InSAR water level change observation, so they best fit the stage-derived levels. In
some areas, such as WCA1, the fit between the two independent datasets is very good, whereas in
Interferogram: Radarsat-1
WCA WCA2A
23-Mar-05/16-Apr-05 20 20
Y = X + 10.5 Y = X + 4.6
2 2
10 R = 0.94 10 R = 0.80
0 0
–10 –10
–20 –20
InSAR-measured water level changes (cm)

–30 –20 –10 0 10 –20 –10 0 10 20
WCA2B WCA3B
20 20
Y = X + 12.3 Y = X+1.0
2 2
10 R = –195.35 10 R = 0.83
0 0
–10 –10
–20 –20
–40 –30 –20 –10 0 –30 –20 –10 0 10
WCA3A/BCNP ENP
30 10
Y = X + 10.9 Y = X – 5.3
2
20 2
R = 0.93 R = 0.74
0
10
0
–10
–10
–20 –20
–30 –20 –10 0 10 20 –20 –10 0 10
(a) (b) Stage-determined water level changes (cm)
15
WCA1
0 WCA2A
–15
–25
cm WCA2B
WCA3A
WCA3B
ENP
(c)
FIGURE 7.6 (a) Radarsat-1 interferogram showing phase changes in response to changes in water levels that
occurred between the two acquisition dates. (b) InSAR–stage calibration plots for each water body. The water
bodies and stage station locations are shown in Figure 7.5c. (c) InSAR-derived map of water level changes that
occurred between March 23, 2005, and April 16, 2005. The black dots mark the location of the stage stations.
Wetland InSAR 147
others (e.g., ENP) is not. The poor fit in the ENP most likely reflects water level decreases below
the surface, which are still measured by the gauging stations, but invisible to InSAR measurements.
The quantitative comparison between InSAR and stage data allows us to determine the accu-
racy of the InSAR technique. We found accuracy of the L-band data in the range of 3–5 cm (Hong
and Wdowinski 2014; Wdowinski et al. 2008), whereas the accuracy of C-band RSAT-1 data was
slightly worse, in the range of 6–7 cm (Hong et al. 2010b). These low accuracy levels most likely
reflect low interferometric coherence (see next section), close proximity of some stage gauging
stations to hydrological structures, and atmospheric noise.
INTERFEROMETRIC COHERENCE OF WETLANDS

Interferometric coherence is a quality measure of an interferogram, which calculates spatial con-
sistency of the calculated phase. High interferometric coherence is reflected in continuous fringe
patterns, whereas low coherence appears as a fuzzy phase pattern (e.g., lower part of Figure 7.5c).
We use coherence maps to evaluate which data type and acquisition parameters are most suitable for
the wetland application of InSAR. As coherence strongly depends on the vegetation type, we used
the diverse Everglades vegetation to evaluate coherence values over the various vegetation types.
We subdivided the study area into the following five wetland vegetation types: freshwater woody
vegetation (cypress), saltwater woody vegetation (mangroves), freshwater herbaceous vegetation
(saw grass), freshwater grassy vegetation (graminiod), and mixed vegetation (mixed shrubs) (Figure
7.7a). Some of these vegetation types can be seen in Figure 7.3. Our study includes the analysis of
C- and L-band SAR data that were acquired by the ERS-1/2, RSAT-1, Envisat, and JERS-1 satellites.
A robust coherence analysis of the various data types and vegetation types was conducted by Kim
et al. (2013). Here, we present a brief summary of the coherence analysis. Our results indicate that
woody wetlands like cypress and mixed shrubs marsh have better coherence than herbaceous wet-
lands like saw grass and cattail in all satellite systems (Figure 7.7c). The L-band JERS InSAR pairs, as
much as 3 years apart, still maintained adequate coherence in wetlands, especially in woody wetlands,
while C-band ERS-1/2 required short temporal baselines (<70-day) to maintain coherence in herba-
ceous wetland. Our study also clearly indicates that HH polarization with high resolution and small
incidence angles is more suitable to wetland InSAR application in terms of decorrelation. Further
details of the coherence analysis and explanations for the results are provided by Kim et al. (2013).
InSAR TIME SERIES

A significant progress in InSAR technology was the development of permanent scatterer InSAR
(PS-InSAR) (Ferretti et al. 2000, 2001) and small baseline subset (SBAS) techniques (Berardino et al.
2002; Lanari et al. 2004), which use a large number of SAR observations to monitor time series of
displacement using successive InSAR observations. These techniques are very useful for monitor-
ing slow and continuous deformation of the Earth’s solid surface mainly in urban areas. However,
these methods do not work in the rapidly changing wetland or floodplain environments, which are
subjected to rapid water level changes. Thus, we developed a new technique—small temporal base-
line subset (STBAS)—which utilizes highly coherent interferometric phases obtained only with
relatively short time difference between two SAR acquisitions of a single satellite track (Hong et al.
2010b). The STBAS technique integrates InSAR and stage observations to transform relative wet-
land InSAR observations to absolute frame and generates both detailed maps of water levels and
water level time series for almost each pixel (40 m resolution). We successfully applied the STBAS
technique to WCA-1 by using RSAT-1 and stage data spanning over a 2-year period (2006–2008)
and obtained a time series of high-spatial-resolution water level maps, almost every 24 days, which
is the RSAT-1 satellite repeat pass period.
In order to increase the observation frequency, we recently expanded our method to incorporate
water level maps calculated from several tracks. Although the observation frequency in each track
(a) (b)
Saw grass
0.6
JERS-1
Radarsat-1
0.5
ERS-1/2
Envisat (HH)
0.4
0.3
0.2
Coherence
0 400 800 0 0.2 0.4 0.6

Mangroves
0.6
0.5
0.4
0.3
0.2
0 400 800 0 0.2 0.4 0.6

(c) Temporal baseline (days) |Bp/Bc|
FIGURE 7.7 Characteristic wetland environments in the Everglades superimposed on ERS multireflectiv-
ity SAR image. (a) Selected five wetland types based on 1999 land cover map distributed from SFWMD
(South Florida Water Management District) and NLCD (National Land Cover Database) 2001 land cover map.
(b) Five typical marshes with low backscattering variation selected for statistical analysis of coherence and
backscatter using ERS-1/2, JERS-1, and Radarsat-1 backscatter variation map. White polygons and red poly-
gons indicate open water surface and urban area, respectively. Backscatter and coherence are used to estimate
background noisy coherence and evaluate radar backscatter calibration accuracy. Black polygon represents
saw grass marsh covered by all of four different Radarsat-1 observations. (c) Comparison between the coher-
ence obtained with the JERS-1, Radarsat-1, ERS, and Envisat SAR data as a function of time interval between
acquisitions and baseline normalized by the critical baseline of each SAR system.
Wetland InSAR 149
T148 T149 T464 T465 Multitrack

5m
20/Feb/11 09/Mar/11 09/Mar/11
10/Feb/11
05/Jan/11 22/Jan/11
20/Nov/10 07/Dec/10 20/Nov/10
05/Oct/10
20/Aug/10 06/Sep/10
05/Jul/10
20/May/10 06/Jun/10
04/Apr/10 21/Apr/10 11/Apr/10
25/Mar/10
17/Feb/10 07/Feb/10
19/Jan/10*
07/Nov/09 24/Nov/09 07/Nov/09
02/Oct/09* 19/Oct/09
22/Mar/09
30/Dec/08 20/Dec/08*
14/Nov/08 04/Nov/08 21/Nov/08* 04/Nov/08
06/Oct/08
14/Aug/08
29/Jun/08
14/May/08 31/May/08
14/Jan/08
02/Nov/07 02/Nov/07
4m
FIGURE 7.8 InSAR-derived high-spatial-resolution water level map time series of WCA-1. The four left
columns present the results of four single-track time series analyses, and the right column presents the com-
bined multitrack results. The time span between maps in the single-track solutions is 46 days and its multiples.
The time span between maps in the multitrack solution varies between 7 and 194 days, depending on data
availability.
depends on the satellite’s repeat pass, the multitrack algorithm can increase the observation fre-
quency to a third or a quarter of the repeat pass orbit, depending on the data availability. Although
we cannot process InSAR data from different tracks, we can combine the results from the different
tracks as all water level maps are calibrated with the same set of stage data and are tied to the stage
stations’ datum. We applied the multitrack algorithm again to WCA-1 using ALOS PALSAR data
acquired along four tracks during 2007–2011 and stage data (Figure 7.8). Our results indicate a
significant increase in the observation frequency from 46 days, which is ALOS’ repeat pass period,
up to 7 days in the best case.
WETLAND InSAR STUDIES OF VARIOUS WETLAND ENVIRONMENTS

So far, we presented wetland InSAR results from our main study area, the south Florida Everglades.
Here, we expand the geographic location and present wetland InSAR observations from other wet-
land areas around the globe (Figure 7.9). The results of four studies are presented—all showing the
successful application of the wetland InSAR technique to monitor remotely water level change and
better understand the hydrological regime of the subject wetlands.
Yellow River Delta Wetlands

The Yellow River Delta is located in eastern China, where the river meets the Bohai Sea. The delta
is a dynamic landform changing locations and patterns due to the large sediment supply. The delta
contains a large area of coastal wetlands, including tidal flats, reed marshes, swamp forests, and
saline marshes (Xiea et al. 2013). Wetland InSAR analysis of L-band ALOS data acquired during
Upland
Lowland
Organic
Salt affected
Permafrost affected
Inland water bodies
No wetlands (or too
small to display)
(a)
N
W E
S
Range change Range change

0 11.8 cm 0 1.5 3 6 9 12
km
0 283 cm
(b) (c)
0 3
km
2π
Relative increase
Black Sea
Phase
[rad]0.00 3.14 6.28
(d) (e)
FIGURE 7.9 (a) Global distribution of wetlands. (From http://www.nrcs.usda.gov/wps/portal/nrcs/detail/

soils/use/worldsoils.) (b) ALOS interferogram of the Yellow River Delta in China. (From Xiea, C. et al., Int.
J. Remote Sens., 34, 2047, 2013.) (c) ALOS interferogram of woody wetlands along the Louisiana Coast in the
United States. (From Kim, J.W. et al., Remote Sens. Environ., 113, 2356, 2009.) (d) Radarsat-1 interferogram
of the Sian Ka’an (Yucatan, Mexico) wetlands. (From Gondwe, B.R.N. et al., Wetlands, 30, 1, 2010.) (e) ALOS
interferogram of the Danube Delta (Romania). (From Poncos, V. et al., J. Hydrol., 482, 79, 2013.)
Wetland InSAR 151
the years 2008–2009 indicates a good interferometric coherence over the reed marsh vegetation and
some of the tidal flats (Figure 7.9b). Combined analysis of the InSAR and ground-based water level
observations suggests that most of the InSAR-observed water level changes occurred due to tidal
variations, especially in the tidal flats (Xiea et al. 2013).
Louisiana Wetlands
Southern Louisiana is rich in both coastal and inland wetlands, as excess water from the Mississippi
River drainage gateway accumulates on the region’s very flat topography. The coastal wetlands,
built on deltaic plains, are shrinking at a fast rate due to the combined effect of (1) sea level rise,
(2) rapid subsidence, and (3) limited sediment supply due to the construction of flood preven-
tion levees along the main rivers. The inland wetlands were also affected by the construction of
these levees but since have stabilized. Both coastal and inland wetlands contain diverse habitats
from forested swamps to marshes and floating mats. Wetland InSAR was successfully applied
to examine water level changes in several wetland environments located in southern Louisiana
(Kim et al. 2009; Lu 2005; Lu and Kwoun 2008). Interferometric analysis of ERS-1/2, RSAT-1,
and ALOS data revealed that best results were obtained when using L-band data with HH polar-
ization (ALOS). C-band data with HH polarization also revealed reasonable results especially
with short time span interferograms. All observation types showed higher coherence over forested
swamps and patchy fringe patterns, reflecting flow discontinuities due to man-made and natural
barriers (Figure 7.9c).
Sian K a’an Biosphere Reserve

The Sian Ka’an wetlands are located along the east coast of the Yucatan Peninsula, Mexico. The
pristine wetlands are fed by groundwater from the karst aquifer of the peninsula. Because most of
the inflow occurs through underground karst structures, it is difficult to observe and understand the
wetland hydrology. Gondwe et al. (2010) used RSAT-1 observations acquired between July 2006
and March 2008 to detect water level changes in the wetlands and infer the dynamic nature of the
system. Their study revealed large variation in the flooded area, in which the main water input
areas are associated with water-filled faults that transport groundwater from the catchment to the
wetlands. The wetland InSAR observations also revealed local-scale water divides and surface
water flow directions within the wetlands (Figure 7.9d).
Danube Delta Wetlands

The Danube Delta is located at the outlet of the Danube River to the Black Sea, mostly in Romania.
The delta’s hydrology is dominated by a network of tributaries that supply water and nutrients to a
variety of aquatic ecosystems, including lakes, ponds, and marshes. Poncos et al. (2013) used ALOS
data from the years 2007–2010 to detect water level changes in the delta (Figure 7.9e). They com-
bined the InSAR observations with stage measurements and hydrological modeling to characterize
the surface flow in the marshes and tributaries. Their study revealed a very dynamic flow regime in
which the flow direction in the marshes varies frequently depending on the water supply from the
tributaries.
DISCUSSION
The unique wetland application of the InSAR technique provides high-spatial-resolution observa-
tions (1–50 m pixel resolution) that cannot be obtained by any terrestrial technique. However, the
main disadvantage of the wetland InSAR observations is their relative nature. The observations
are relative in both space (with respect to a reference point) and time (water level changes between
two acquisition times). Thus, it is important to tie the space-based observations to ground obser-
vations of water levels (stage monitoring). In the Everglades, there is a dense stage network, which
allows us to calibrate and validate the InSAR observations and tie them to an absolute reference
frame. Our calibration studies suggest an accuracy level of 3–8 cm (Hong and Wdowinski 2014;
Wdowinski et al. 2004, 2008). The man-made structures in the Everglades create many flow
discontinuities that require such dense network for accurate and reliable monitoring. However, in
natural flow wetland areas, such as in the ENP in the southern part of the Everglades ecosystem,
the flow is continuous and can be monitored by a less dense network. Thus, sparsely distributed
stage stations in natural wetland areas may be sufficient for calibrating the InSAR observations.
In remote wetland areas, where no stage monitoring exists, one can use altimetry data for InSAR
calibration, as used by Kim et al. (2009). However, altimeter observations are also character-
ized by low temporal resolution and may acquire at a different time than the SAR acquisitions.
Nevertheless, altimeter observations can be useful for calibration, if no other accurate measure-
ments can be obtained.
One very useful and important observation can be derived directly from the raw interferogram,
without the need of stage data for calibration. The high-resolution wetland interferograms provide
direct observations of flow patterns and flow discontinuities, as shown in Figure 7.2. As water level
and water level changes tend to be different across barriers, these differences will be shown in the
interferogram as phase discontinuities. This observation is important for wetland restoration efforts
that aim to restore managed or degraded wetlands, such as the Everglades, to their natural undis-
turbed condition.
Another very important application of wetland InSAR is constraining high-resolution flow
models, which are important tools for wetland management and restoration. This application also
does not require stage calibration, as the model results can be converted into the interferogram
phase domain. The stage data are typically used as boundary conditions of the flow model. We
conducted such preliminary study by comparing the InSAR observations with the TIME (Tides
and Inflows in the Marshes of the Everglades) model, which was developed by U.S. Geological
Survey and University of Miami. Our study indicates that the model predicts longer-wavelength
water levels well, which are constrained by the stage data, but misses many of the shorter-wave-
length features.
CONCLUSION
The unique wetland application of the wetland InSAR technique provides high-spatial-resolution
observations of surface water level changes in aquatic environments with emergent vegetation,
such as wetlands and floodplains. The technique works because of the “double-bounce” effect,
in which the radar pulse is backscattered twice, first from the water surface and then from the
vegetation, or vice versa. Interferometric coherence analysis of various SAR data types indicates
that longer wavelength (L-band), short revisit cycles, HH polarization, high spatial resolution, and
small incidence angle are more suitable to wetland InSAR application in terms of decorrelation.
Best results have been obtained with the L-band (ALOS and JERS-1) data, X-band (TSX) data
with 11-day repeat orbits, and fine beam (7 m resolution) C-band RSAT-1/2 data with 24-day
repeat orbits. The application of the wetland InSAR technique to various wetland environments
around the world shows that in many wetland areas, interferometric coherence can be maintained
and, furthermore, the wetland InSAR observations can provide high-spatial-resolution observa-
tions that cannot be obtained by any terrestrial technique. In addition to high-spatial-resolution
water level monitoring, wetland InSAR applications include detection of flow patterns and flow
discontinuities and constraining high-resolution flow models. The development of the wetland
InSAR time series techniques (Hong and Wdowinski 2014; Hong et al. 2010b) provides improved
high-spatial, multitemporal water level observations that can be helpful for water management and
hydrological modeling.
Wetland InSAR 153
ACKNOWLEDGMENTS
This work was enabled by the TSX science proposal (HYD0029) and the SOAR project from CSA
for access to the TerraSAR-X and the Radarsat-1/2 data. We also thank JAXA and ASF for access
to ALOS data. The research was supported by NASA Cooperative Agreement No. NNX10AQ13A
(WaterSCAPES: Science of Coupled Aquatic Processes in Ecosystems from Space) and by the
National Science Foundation through the Florida Coastal Everglades Long-Term Ecological
Research program under Grant No. DEB-1237517.
REFERENCES
Alsdorf, D.E., J.M. Melack, T. Dunne, L.A.K. Mertes, L.L. Hess, and L.C. Smith. 2000. Interferometric radar
measurements of water level changes on the Amazon flood plain. Nature 404:174–177.
Berardino, P., G. Fornaro, R. Lanari, and E. Sansosti. 2002. A new algorithm for surface deformation moni-
toring based on small baseline differential SAR interferograms. IEEE Transactions on Geoscience and
Douglas, M.S. 1947. The Everglades: River of Grass. Rinehart and Company, New York.
Evans, T.L. and M. Costa. 2013. Landcover classification of the Lower Nhecolandia subregion of the Brazilian
Ferretti, A., C. Prati, and F. Rocca. 2000. Nonlinear subsidence rate estimation using permanent scatterers in
differential SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing 38:2202–2212.
Ferretti, A., C. Prati, and F. Rocca. 2001. Permanent scatterers in SAR interferometry. IEEE Transactions on
Gondwe, B.R.N., S.H. Hong, S. Wdowinski, and P. Bauer-Gottwein. 2010. Hydrologic dynamics of the
ground-water-dependent Sian Ka’an Wetlands, Mexico, derived from InSAR and SAR data. Wetlands
30:1–13.
Hong, S.H. and S. Wdowinski. 2014. Multi-temporal, multi-track monitoring of wetland water levels in the
Florida Everglades using ALOS PALSAR data with interferometric processing. Geoscience and Remote
Sensing Letters 11:1355–1359.
Hong, S.H., S. Wdowinski, and S.W. Kim. 2010a. Evaluation of TerraSAR-X observations for wetland InSAR
application. IEEE Transactions on Geoscience and Remote Sensing 48:864–873.
Hong, S.H., S. Wdowinski, S.W. Kim, and J.S. Won. 2010b. Multi-temporal monitoring of wetland water levels
in the Florida Everglades using interferometric synthetic aperture radar (InSAR). Remote Sensing of the
Kim, J.W., Z. Lu, H. Lee, C.K. Shum, C.M. Swarzenski, T.W. Doyle, and S.H. Baek. 2009. Integrated analy-
sis of PALSAR/Radarsat-1 InSAR and ENVISAT altimeter data for mapping of absolute water level
changes in Louisiana wetlands. Remote Sensing of the Environment 113:2356–2365.
Kim, S.W., S. Wdowinski, A. Amelung, T.H. Dixon, and J.S. Won. 2013. Interferometric coherence analysis of
the Everglades wetlands, South Florida. IEEE Transactions on Geoscience and Remote Sensing 51:1–15.
Kwoun, O.I. and Z. Lu. 2009. Multi-temporal RADARSAT-1 and ERS backscattering signatures of coastal
wetlands in southeastern Louisiana. Photogrammetric Engineering and Remote Sensing 75:607–617.
Lanari, R., O. Mora, M. Manunta, J.J. Mallorqui, P. Berardino, and E. Sansosti. 2004. A small-baseline approach
for investigating deformations on full-resolution differential SAR interferograms. IEEE Transactions on
Lu, Z. 2005. C-band radar observes water level change in swamp forests. EOS, Transactions, American
Geophysical Union 86:141–144.
Lu, Z. and O.I. Kwoun. 2008. Radarsat-1 and ERS InSAR analysis over southeastern coastal Louisiana:
Implications for mapping water-level changes beneath swamp forests. IEEE Transactions on Geoscience
and Remote Sensing 46:2167–2184.
Massonnet, D., M. Rossi, C. Carmona, F. Adragna, G. Peltzer, K. Feigl, and T. Rabaute. 1993. The displace-
ment field of the Landers earthquake mapped by radar interferometry. Nature 364:138–142.
Poncos, V., D. Teleaga, C. Bondar, and G. Oaie. 2013. A new insight on the water level dynamics of the Danube
Delta using a high spatial density of SAR measurements. Journal of Hydrology 482:79–91.
Pritchard, M.E. and Contributors to the ROI_PAC wiki. Version: 0.4. 2014. Open-source software for geodetic
imaging: ROI_PAC for InSAR and pixel tracking. Available online at: http://www.geo.cornell.edu/eas/
PeoplePlaces/Faculty/matt/pub/winsar/InSAR_textbook_for_web_2014.pdf.
Richards, L.A., P.W. Woodgate, and A.K. Skidmore. 1987. An explanation of enhanced radar backscattering
from flooded forests. International Journal of Remote Sensing 8:1093.
SFWMD. 2014. DBHYDRO—Corporate environmental database. Available online at: http://www.sfwmd.gov/
dbhydro. http://www.sfwmd.gov/org/ema/dbhydro/index.html.
Simard, M., K.Q. Zhang, V.H. Rivera-Monroy, M.S. Ross, P.L. Ruiz, E. Castaneda-Moya, R.R. Twilley, and
E. Rodriguez. 2006. Mapping height and biomass of mangrove forests in Everglades National Park with
SRTM elevation data. Photogrammetric Engineering and Remote Sensing 72:299–311.
Wdowinski, S., F. Amelung, F. Miralles-Wilhelm, T.H. Dixon, and R. Carande. 2004. Space-based measure-
ments of sheet-flow characteristics in the Everglades wetland, Florida. Geophysical Research Letters
31:L15503.
Wdowinski, S. and S. Eriksson. 2009. Geodesy in the 21st century. EOS, Transactions American Geophysical
Union 90:153–155.
Wdowinski, S., S.-H. Hong, A. Mulcan, and B. Brisco. 2013. Remote-sensing monitoring of tide propagation
through coastal wetlands. Oceanography 26(3):64–69, DOI 10.5670/oceanog.2013.46.
Wdowinski, S., S.W. Kim, F. Amelung, T.H. Dixon, F. Miralles-Wilhelm, and R. Sonenshein. 2008. Space-
based detection of wetlands’ surface water level changes from L-band SAR interferometry. Remote
Xiea, C., Y. Shaoa, J. Xub, Z. Wanc, and L. Fanga. 2013. Analysis of ALOS PALSAR InSAR data for map-
ping water level changes in Yellow River Delta wetlands. International Journal of Remote Sensing
34:2047–2056.
8 Radar and Optical Image
Fusion and Mapping of
Wetland Resources
Elijah Ramsey III and Amina Rangoonwala
CONTENTS
Introduction..................................................................................................................................... 155
Defining Fusion............................................................................................................................... 157
Alignment of Data Sources............................................................................................................. 157
Types of Fusion Processes.............................................................................................................. 158
Pixel-Level Fusion..................................................................................................................... 158
Feature-Level Fusion.................................................................................................................. 161
Decision-Level Fusion............................................................................................................... 161
Fusion and Land Cover Classifications........................................................................................... 161
Pixel-Level Fusion..................................................................................................................... 161
Concatenation Fusion with Land Cover Classification Examples........................................ 162
Concatenation Fusion Applied in Cloudy Environments...................................................... 163
Feature-Level Fusion.................................................................................................................. 164
Decision-Level Fusion............................................................................................................... 166
Fusion Performance........................................................................................................................ 168
Conclusion...................................................................................................................................... 168
References....................................................................................................................................... 170
INTRODUCTION
Wetlands proportionally exert a higher influence on biogeochemical fluxes among land, atmosphere,
and hydrologic systems than their 1% worldwide occurrence (Sahagian and Melack 1996). Although
their frequency of occurrence is low and their importance high, wetlands continue to face high
detrimental pressures from natural and human-induced forces (Ramsey 1998). For proper resource
management and to provide for mitigation of degradation before irreversible change, a regional,
synoptic, detailed, and timely wetland resource monitoring system is needed. Remote sensing offers
the single best source of timely, synoptic wetland status and trends information at a variety of spatial
and temporal scales (Wickland 1991).
The remote sensing of wetlands does not generally differ in technique or process from remote
sensing–based mapping of other terrestrial features (e.g., Ramsey 2005). Differences exist because
wetlands occupy a unique interface, or ecotone, between aquatic and upland ecosystems (Mitsch
and Gosselink 2000). Lacustrine and riparian wetlands reveal that uniqueness in their response to
seasonal and longer-term cycles of hydrology or changing water inputs into the systems. Coastal
wetlands experience similar complexities (Mitsch and Gosselink 2000), but a higher inundation
variability and severe storm frequency further intensify the landscape heterogeneity over time
155
and space (Ramsey et al. 2012). The high temporal and spatial complexity of forests and marshes
that occupy wetlands, particularly coastal wetlands, worldwide make wetlands a more challenging
remote sensing environment than do terrestrial landscapes (Ramsey and Rangoonwala 2011).
Moderate-spatial-resolution passive optical systems (e.g., Landsat Thematic Mapper [TM]) can
address many coastal wetland resource issues (Jensen et al. 1987; Klemas et al. 1993; Lunetta et al.
1998; Ramsey et al. 2001b). However, even though multiple-source integration and subpixel extrac-
tion have improved the spatial detail extractable from moderate-resolution sensor systems (Ramsey
et al. 1998, 2005, 2006), these moderate-resolution sensors are often insufficient to quantify the
amount and types of change within spatially irregular and heterogeneous coastal landscapes, par-
ticularly those of degrading wetlands (Ramsey and Laine 1997).
Although time series analyses and some applications of sensor combinations have improved
the temporal content of optical regional mapping of specific events or broad changes (e.g., flood-
ing, storms, and drought) (Ramsey et al. 2001a), the overriding constraints for consistent collec-
tion of optical data are clouds and turbid atmospheres, especially in subtropical coastal regions
and during severe weather events (Ramsey 1995). Passive optical imaging is restricted to daylight,
and an active light amplification by stimulated emission of radiation (LASER) and passive optical
imaging (or profiling) are limited to nonturbid by atmospheric clarity. Even when reliance on time-
constrained collections is minimized, the potential to maximize extractable information or capture
time-dependent features is limited using optical systems (Moghaddam and McDonald 2003). The
failure of optical remote sensing systems to move beyond an opportunistic to a consistent data
source with an adequate spatial resolution is a critical issue in coastal management and proper
resource evaluation (Ramsey and Laine 1997).
A combined optical and radio detection and ranging (radar) remote sensing system can pro-
vide the spatial and temporal resolutions needed for monitoring the diverse, dynamic, and complex
coastal wetland environments. As optical satellite and aircraft remote sensing systems, satellite and
aircraft radar systems have been used extensively for mapping natural resources, especially when
timely collection is the dominant concern (Lyon and McCarthy 1981; Kasischke and Bourgeau-
Chavez 1997; Ramsey 1998, 2005; Ramsey et al. 2012, 2013). The day and night and nearly all-
weather mapping capabilities of radar remote sensing systems eliminate most of the weather-related
optical limitations, and the high spatial and temporal resolution modes and coverages available
with radar systems adequately address the temporal and spatial dynamics of coastal wetlands (e.g.,
Ramsey et al. 2011a,b; Ramsey and Rangoonwala 2011).
In order to fully utilize the increasing number of satellite sensors that provide increasingly higher
spatial and temporal resolutions and exploit an increasing range and higher detail of the electromag-
netic spectrum, this chapter introduces the fusion of multispectral (MS) optical and radar remote
sensing for mapping and monitoring landscapes. Although we found only a limited number of
optical–radar fusion examples focused on mapping wetlands, all fusion concepts, developments,
and applications described are directly relevant to needed advancements of optical–radar fusion in
coastal wetland mapping.
Fusion theory, strategies, and techniques are presented in a volume of literature; there are many
references that provide excellent reviews of fusion (Pohl and Van Genderen 1998; Wald 2000;
Henderson et al. 2002; Wang et al. 2005; Boström et al. 2007; Hong 2007; Orsomando et al. 2007;
Gangkofner et al. 2008; Fonseca et al. 2011; Zhang 2012). While those referenced manuscripts and
many others detail fusion complexities and provide a wide range of application examples, we gener-
alize the various fusion strategies and techniques for clarity and brevity. Our objective is to provide
map producers, resource managers, and classification protocol and remote sensing technical devel-
opers a general assessment and some perspectives of the state and usefulness of fusion research and
applications pertinent to remote sensing land cover (LC) mapping and expressly wetland mapping.
To reach that objective, a brief history and objective definition of image data fusion relevant to
remote sensing LC mapping is presented. Fusion, as applied to LC classification of remote sens-
ing image data, is described at three levels: pixel, feature, and decision. Pixel-level (PL) fusion
Radar and Optical Image Fusion and Mapping of Wetland Resources 157
transforms and combines single-resolution data, feature-level (FL) fusion identifies and combines
common objects, and decision-level (DL) fusion combines and optimizes multiple algorithm results.
Dominant methods used to accomplish each of the three fusion levels are broadly explained, perti-
nent applications are referenced, and graphical portrayals depicting straightforward fusion applica-
tions in each level are included where appropriate.
DEFINING FUSION
Starting in the 1950s, methods were sought to enhance detection of objects in an image by com-
positing data from various sensors (Wang et al. 2005). These methods were generally referred to
as “fusion.” Specific to remote sensing applications, the overall justification for fusion was that
improvements in eliminating classification error probability and interpretation robustness could
only be achieved with addition of more data collected by more sensors (Wald 2000).
We adapt the definition of Hall and Llinas (1997) by simply adding the concept of fusion qual-
ity (Wald 2000; Wang et al. 2005; Zhang et al. 2008) as an assessment of whether or not the fused
product is more beneficial or relevant for the user or “customer” than possible without fusion. Our
working definition becomes “Processes that use data from sensors and related spatial information
to create quality, fused products that provide improved accuracy and more specific inferences than
could be achieved without fusion.” In essence, fusion is not specific to an application and may be
tailored to a particular objective to provide the needed quality as related to customer needs that are
not possible without the fused product (Wald 2000).
ALIGNMENT OF DATA SOURCES

Sensor fusion requires spatial and structural alignment of the layers of data or information originat-
ing from the different sources in order to preserve the low-resolution spectral information (Wald
2000; Zhang et al. 2008; Palubinskas and Reinartz 2010). Spatial alignment requires that all data
and information sources are represented on a common coordinate frame (Wald 2000; Wegner et al.
2007; Kahler and Blasch 2009).
Structural alignment refers to structural aspects such as the standardization of units, calibration
of sensors, and atmospheric corrections (Wald 2000; Wegner et al. 2007; Kahler and Blasch 2009).
Apart from spatial reprojection (image warping) and interpolation (resampling), even though repre-
senting similar quantities (commensurate sources—radiance measurements) with a known spatial
dependence, Pan and MS fusion requires structural or, in this case, spectral alignment, for Pan spatial
information to be properly transferred into the MS image (see Zhang et al. 2008). The Pan spectral
range must align with the spectral range of the selected MS sensor image bands, most commonly the
visible and near-infrared (VNIR) image bands. In the earliest public satellite sensor systems, the Pan
range generally overlapped the spectral range of the VNIR MS image bands; however, the wavelet
extension of new sensors may not always provide adequate overlap between the MS (VNIR) and
Pan spectral bands (Zhang 2004). Although occupying mostly the same reference space, as spectral
differences of the commensurate data sources increase (e.g., reflected versus emitted), fusion com-
plexities increase (Wald 2000; Kahler and Blasch 2009); however, those complexities are most often
less challenging compared to fusion of heterogeneous (noncommensurate) sources such as optical
and radar sensors (Wald 2000). Complexity is increased because direct transferability of structural
information between images can be compromised (see Zhang et al. 2008).
Radar and optical remote sensing can be defined by interactions according to absorption, reflec-
tance or backscatter, transmittance, and surface reflection; however, the very dissimilar frequency
ranges of optical and microwave (radar) result in differences in the nature of the electromagnetic
interaction and its scatter or reflectance from the target (e.g., Ramsey 2005; Ramsey et al. 2006,
2009a,b). These differences establish what aspects or components of the target are defined by the
two sensor systems (e.g., Ramsey et al. 1998, 1999, 2002, 2009a,b) and differences in the nature
of the energy recorded at the optical and radar sensor. It is not only disparities in the frequency or
wavelength that lead to differences in the nature of optical and radar representation of the wetland
landscape, it is also the contrast of a system based on a coherent source and recordings and a system
based on incoherent sources and recording without consideration of phase or polarization. In radar,
the sensor records the result of constructive and destructive additions of backscatter from elements
comprising ground element at selected polarizations, whereas in optical, the sensor records the sum
of radiance reflected into the solid angle of the optical aperture. Synthetic aperture radar (SAR)
systems focus these backscatter resultant records over one- or multiple-look directions creating
single- to multiple-look images. Passive optical systems have no comparable construction ability.
Ultimately, these differences affect interpretability as represented on the image; optical is more
directly interpretable, whereas radar can be more difficult (Ramsey et al. 1998; Hong et al. 2009).
In addition, radar systems record the phase and polarization of the send and receive waves pro-
viding unique information that is not directly represented in optical systems. Variation in the ampli-
tude and phase of the send and receive electromagnetic waves can deliver direct measures of a
wetland canopy structure (Ramsey 1998, 2005; Ramsey et al. 1999, 2006, 2009a,b) as compared
to passive optical systems that must rely on application of bidirectional models (Kimes and Sellers
1985) to infer structure variables. Lastly, VNIR sensitivity to leaf structure and pigments (Ramsey
and Rangoonwala 2006, 2010, 2011) versus radar sensitivity to the 3D distribution of water and the
100,000 times difference in wavelengths result in radar imaging representing interactions deeper
into wetland forest or marsh canopies than commonly possible with optical systems (Ramsey 1998).
As illustrated in Figure 8.1, the confusion in interpretability may stem from a lack in understand-
ing of what the optical and radar image depictions of the landscape represent. In this case, remote
sensing mapping with MS imagery (as represented by canopy reflectance) largely depicts the loss
and recovery of the bulk green biomass, whereas radar (as represented by canopy structure) largely
embodies changes in the bulk stem density and orientation. Further demonstrations of MS and radar
imaging differences as represented by marsh recovery from burns are described in Ramsey et al.
(1998, 1999, 2002, 2009b).
TYPES OF FUSION PROCESSES

Fusion can be broadly divided into PL, FL, and DL fusion (Kahler and Blasch 2009). These divi-
sions describe the hierarchy or information level that the fusion of information or data operates
on. PL (or image) fusion combines data from two or more image sources into a single-resolution
image (Zhang 2012), such as a synthesized high-spatial-resolution MS image. PL fusion is the most
complicated and sensitive to registration or alignment errors (Kahler and Blasch 2009). FL fusion is
more complex but results in enhanced fusion-detection performance (Kahler and Blasch 2009). In
FL fusion, features (attributes or properties) of objects created from separate or joint segmentation
of image sources are fused (Kahler and Blasch 2009). DL fusion requires that each sensor modality
has a good object recognition algorithm for classification and additionally that the sensors provid-
ing the fusion input operate in the same energy range, process the measured electromagnetic fluxes
similarly, and provide similar object specificity or detail (Kahler and Blasch 2009). To be more
contemporary with remote sensing classifications and fusions based on optical and radar, we define
DL fusion as imposing external knowledge to constrain or guide the fusion process (following ideas
stated in Wald [2000]). And in remote sensing applications, the external knowledge can be from any
number of sources or can be the results of multiple algorithms operating on two or more indepen-
dent sources that in conjunction provide an optimal solution.
Pixel-Level Fusion
Most PL fusion processing is applied at the PL and focuses on injecting spatial information pri-
marily texture content from radar into a low-spatial-resolution MS or hyperspectral image while
preserving spectral properties of the low-resolution image (Henderson et al. 2002; Amarsaikhan
and Douglas 2004; Mercer et al. 2005; Wang et al. 2005; Zhang et al. 2008, 2010; Yuhendra
et al. 2011; Zhang 2012). Even though increase in spatial resolution and texture of Pan and high-
resolution radar is an important aspect of PL fusion, the preservation of spectral information
during fusion is extremely important for most remote sensing applications based on spectral sig-
natures (Mercer et al. 2005; Zhang et al. 2008; Yuhendra et al. 2011). There is, however, always
a compromise between desired spatial enhancement and spectral consistency, and it is the nature
of this compromise as part of the fusion process that separates the different PL fusion methods.
Current research on PL fusion strives to optimize the injection of high-frequency information
while emphasizing the improvement of fusion quality and the reduction of color distortion (Zhang
2004; Zhang et al. 2008).
PL fusion processes can be divided into four classes: arithmetic (AT), statistical, concatenation,
and substitution. AT fusion includes image addition, multiplication, ratioing, and other similar con-
structs (Henderson et al. 2002; Amarsaikhan and Douglas 2004; Palubinskas and Reinartz 2011).
Statistical fusion is processed as a statistical estimation problem (Yuhendra et al. 2011) in order to
improve retention of MS color and lessen dependency on the dataset composition (Zhang 2004).
Concatenation fusion creates stacks or layers (concatenates) of commensurate and noncommen-
surate data into a common dataset in preparation for an inverse transform application (Henderson
et al. 2002). Substitution fusion includes most PL fusion methods. In this chapter, we further sepa-
rate substitution into two fusion categories based on Jinghui et al. (2010): component substitution
and multiresolution.
Nonburnt and preburnt Burnt and initial regrowth Regrowth
(a1) (a2) (a3)
0.20 0.20 0.20

January March September
Preburnt Juncus (J3) Burnt Juncus (J3) Regrown Juncus (J3)
0.15 Nonburnt Juncus (J4) 0.15 Nonburnt Juncus (J4) 0.15 Nonburnt Juncus (J4)
Canopy reflectance
0.10 0.10 0.10
0.05 0.05 0.05
0.00 0.00 0.00

400 500 600 700 800 900 1000 400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
(b1) Wavelength (nm) (b2) Wavelength (nm) (b3) Wavelength (nm)
FIGURE 8.1 The recovery of a Juncus roemerianus marsh canopy monitored with reflectance as an MS imag-
ing analogue and structure as a radar imaging analogue (Ramsey et al. 2002, 2009a). (a1–a3) The top series
contains actual photographs of before, right after, and 1–2 years after the marsh was burnt. (b1–b3) The next
series of plots directly below the cartoon series depict the recovery of the marsh canopy determined with canopy
reflectance data. The canopy reflectance recovery is estimated to take about 6 months. (Continued)
0%
0%
10
10
120
100
Centimeter
0%
80
10
%
60
%
90
60
40
%
15
5-
%
80
20
40
0
(c1) (c2) (c3)
1.0 1.0 1.0

0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
Light penetration
0.6 0.6 0.6

0.5 0.5 0.5
0.4 0.4 0.4
September (2nd year)
0.3 0.3 0.3 and January (3rd year)
September (1st year) April (2nd year)
0.2 0.2 0.2 Regrown Burn (Sept. J3)
Preburnt Juncus (J3) Burnt Juncus (J3) Regrown Burn (Jan. J3)
0.1 Nonburnt Juncus (J4) 0.1 Nonburnt Juncus (J4) 0.1 Nonburnt (Sept. J4)
0.0 Nonburnt (Jan. J4)
0.0 0.0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
(d1) Canopy height (cm) (d2) Canopy height (cm) (d3) Canopy height (cm)
FIGURE 8.1 (Continued) The recovery of a Juncus roemerianus marsh canopy monitored with reflectance
as an MS imaging analogue and structure as a radar imaging analogue (Ramsey et al. 2002, 2009a). (c1–c3)
The series of cartoons below the reflectance plots broadly illustrate the changes in canopy structure; (1) higher
density and mixed orientation of leaves and stacks comprising the preburn marsh; (2) the short, less dense, and
nearly vertical marsh a short time after being burnt; and (3) the increasing taller, denser, and mixed orientation
marsh canopy 1–2 years after being burnt. (d1–d3) These plots illustrate the change in marsh structure (density
and orientation) from preburn, to just after being burnt, to 1 and 1.5 years after being burnt. Canopy structure
is parameterized as changes in light penetration into the depths of the canopy as illustrated in c1–c3. (Adapted
from Ramsey III, E. et al., Mapping fire scars and marsh recovery with remote sensing image data, in: X.
Yang (ed.), Remote Sensing and GIS for Coastal Ecosystem Assessment and Management, Springer, Berlin,
Germany, Publisher Springer Lecture Notes in Geoinformation and Cartography, 2009a, pp. 415–438.)
Component-substitution fusion can be further divided into fusion processes that are based on
color-related techniques, such as intensity–hue–saturation (IHS), and statistical–mathematical tech-
niques, such as principal component analysis (PCA) (Xie and Keller 2006; Saberioon et al. 2009;
Yuhendra et al. 2011). Multiresolution fusion processes are based on spatial filters, such as high-pass
filter (HPF) and À Trous wavelet (ATF) (Mercer et al. 2005; Xie and Keller 2006; Jinghui et al.
2010; Yuhendra et al. 2011), and the application of multiple linear regressions, such as the synthetic
variable ratio (SVR) (Hong 2007; Zhang et al. 2010).
Although fusions based on noncommensurate datasets, such as optical and radar, are less com-
mon than MS and optical Pan, in most typical applications of radar, the fusion process is altered
simply by substituting radar for the optical Pan image band (Mercer et al. 2005). In both commen-
surate and noncommensurate datasets, the fusion processes most commonly involve IHS and PCA
transforms, and the most common fusion methods apply image band component substitution and
AT techniques (Henderson et al. 2002; Kumaran and Shyamala 2002; Amarsaikhan and Douglas
2004). Lastly, many PL applications mix fusion types within an overall fusion process, such as a
wavelet–IHS or wavelet–PCA fusions (Zhang and Hong 2005; Jinghui et al. 2010).
Feature-Level Fusion
FL fusion is not carried out pixel by pixel but by identifying homogeneous regions, features, or
segments in each image that can be used to fuse one or more images or enhance performance in
interpreting image data collected by another sensor (Orsomando et al. 2007). FL and DL fusions
in remote sensing mapping can be separated based on the dominant application objective associ-
ated with each fusion strategy. FL fusion is based on segmentation of the landscape into uniform
segments, leading directly to LC classifications. The objective of the segmentation is to identify
features based on the input dataset relation to the landscape composition. A classifier is then applied
to the segmented landscape to provide physical meaning to the identified features. The DL fusion
definition offered in this chapter does not require that the sensors used in the image band fusion
provide similar object specificity or detail, only that each sensor provides object recognition useful
in decision classification.
The advantage of FL fusion is that it reduces the stringency of PL fusion and reduces issues
related to data alignment. In FL fusion, the difference and variety of responses recorded by an
optical and radar sensor of a landscape are considered independent and need not be presupposed
(Kahler and Blasch 2009); it is only necessary that the feature be distinguishable in at least
one sensor image or obtainable with auxiliary information. This decreased dependency of input
image data sources lowers the influence of alignment of the two sensor landscape recordings that
may be problematic in PL fusion. In fact, FL fusion better accounts for and takes advantage of
how different sensors record the same features in the image representation. Although highest per-
formance is achieved when images are closely associated in space and time (Kahler and Blasch
2009), FL fusion provides a convenient method to fuse images collected at different times or over
a temporal series.
Decision-Level Fusion
In contrast to FL fusion, DL fusion is not an input into the final mapping classifier; instead, it is
an input into the final objective of the mapping process, normally involving biophysical mapping.
The cases identified as DL fusion contain mixtures of principally FL fusion components and are
applied primarily as additional biophysical information derived from MS data sources that improve
biophysical information extraction with radar data sources. The different and varying responses of
MS and radar to the same LC structural features provide a complementary data source to possibly
constrain the range of biophysical variables derived most often from radar with canopy biophysical
variables derived from MS data.
Most processed MS and radar biophysical information are optimized to ensure the best esti-
mate for a yet undetermined biophysical property. This type of DL fusion based on optimization
was reported to have improved estimation of specific biophysical variables; however, the applica-
tion examples were few and dominantly associated with interferometric mapping of forest canopy
structure. Although limited in scope, the DL fusion shows promise, and comparable approaches for
similar reasons have been suggested in wetland biophysical mapping that includes marshes as well
as forested wetlands (Ramsey et al. 2006, 2011b).
FUSION AND LAND COVER CLASSIFICATIONS

Pixel-Level Fusion
We choose concatenation applications to highlight PL fusion because it is straightforward to apply
within current image processing software (Henderson et al. 2002; Kumaran and Shyamala 2002),
and along with substitution, it is a common method used in LC classifications. Concatenation is used
to stack or layer commensurate and noncommensurate data into a common dataset in preparation
for substitution of inverse transforms, such as AT, IHS, and PCA, before classification or more
commonly input directly into an LC classification procedure. In effect, the concatenation with or
without an interjected transform acts to form the input dataset for the LC classification, with the
classification more or less functioning as the fusing procedure.
Concatenation Fusion with Land Cover Classification Examples

Leckie (1990) provided an early example of a stacked concatenation fusion with good perfor-
mance assessment. As in many of the early optical radar fusion applications, the study objective
was discernment of change in forested landscapes, in this case differentiation of hardwood and
softwood forest stands. A similar early fusion study concatenated MS (TM), color infrared (CIR)
photography, and radar (Earth Resources Satellite [ERS]) image bands into a single concatenated
dataset that was input into an LC classification procedure (Ramsey et al. 1998). The 25 m MS
and radar data as well as 3 m CIR were registered and resampled into a 10 m pixel resolution
common database. As in Leckie (1990), the concatenated dataset was input into a progressive LC
classification of the coastal marsh and adjacent inland forest system. LC classes included hard-
woods and pine forests, scrub shrub, a fresh marsh, four saline classes within a nearly monotypic
black needlerush marsh (high, medium high, medium, and low), shallow water, coastal water,
sand flats, and recent burns. Classifications were performed on individual sensor inputs and com-
binations of MS, CIR, and radar sensor inputs (Figure 8.2). Classification accuracies based on 991
classified ground sites were obtained per LC class and per classification. In all cases, the addi-
tion of radar to the six reflective MS image bands improved the overall classification accuracy
obtained solely on MS image bands, evaluated per class, and increased in individual class accura-
cies from 0% to 10%.
Open water
F
l Emergent herbaceous wetlands
o
r Woody wetlands
St. Marks i
NWR
d
a
Row crops
Deciduous forest
Evergreen forest
Mixed forest
Shrubland
Commercial industrial
High intensity residential
Gulf of Mexico Transitional
Grassland herbaceous
Deciduous forest
Evergreen forest
Shrubland
Fresh marsh
Saline high marsh
Saline medium marsh
Saline low marsh
Shallow water
Coastal water
0 4 8 km Open water
Sand flats
Recent burns
FIGURE 8.2 Classifying coastal resources by fusing optical and radar imagery and CIR photography (e.g.,
Ramsey et al. 1998), in this case coastal Florida (top: Florida LC map from Florida Game and Fish Commission
[FGFC; 1992]). The optical–radar fusion (bottom map) dramatically increased the detail of the single FGFC
emergent herbaceous wetland LC class to four classes: fresh marsh, saline high marsh, saline medium marsh,
and saline low marsh. Reference points of 991 were used in the fused map LC accuracy assessment.
As reflected in Leckie (1990) and Ramsey et al.’s (1998) fusion applications, fusion performance
results are not straightforward but related to the particular LC class and mixture and number of
image bands entered into the LC classification algorithm. Although commonly lacking in detailed
accuracy assessments based on recognized LC classification protocols, concatenation fusion meth-
ods and many related variants are dominant in LC classifications that are specific to wetland eco-
systems. Selected applications can show how concatenation fusion methods are commonly applied
to LC classifications; these examples are based on Leckie (1990) and Ramsey et al. (1998) and help
interpret the performance and relevancy of LC classifications.
In another study, Dehouch et al. (2012) apply concatenation fusion of MS (SPOT XS) and radar
(TerraSAR-X) image data to improve classification of intertidal flats and coastal salt marshes.
Training samples that included dry and wet sands; mud; sandy mud; aquatic beds; and low, inter-
mediate, and upper salt marshes, as well as oyster beds, were extracted from the concatenated MS
and radar dataset and entered into LC classifications. As found in other concatenations and similar
fusion-based LC classifications, visual interpretation revealed that the various concatenation vari-
ants entered into the multiple classifications enhanced classification performance in some classes
and reduced it in others as compared to solely optical-based classifications (Dehouch et al. 2012).
Henderson et al. (2002) also combined MS (TM) and radar (Radarsat-1) image data in an
attempt to improve wetland mapping. Their study not only provided a detailed example of what
we generally describe as concatenation fusion, they as well compared those results to AT and PCA
substitution fusion classification results. Based on variations of input datasets and fusion meth-
ods, results showed that only concatenation fusion increased the overall classification accuracy.
As typical in all fusion performance assessments, however, when the final classification output
is compared to actual ground-based data, the increase in overall accuracy over that classifica-
tion obtained solely with optical data was very modest (<3%). The overall accuracy incompletely
describes the fusion quality when based on the user objective. Within the highest overall perform-
ing optical and radar concatenation fusions, the classification accuracy improvement was unequal
among LC classes. In this study, the palustrine emergent class accuracy fared worse (from 7% to
0%) after fusion than when solely based on MS, and the low MS classification accuracies associ-
ated with estuarine emergent (60%), palustrine forest (31%), and scrub shrub (3.3%) remained low
after fusion.
Concatenation Fusion Applied in Cloudy Environments

One of the primary motivations for fusion of optical and radar is the need to overcome the inability
of optical systems to provide useable data when cloudy or turbid atmospheres exist. Cloud impedi-
ments to optical classifications were the focus of a study by Wang et al. (1998) that compared classi-
fication performances based on concatenation fusion of a monthly sequence of radar (ERS-1) images
to a classification based on MS (TM) images. Overall classification accuracy of phragmites and cat-
tail marshes and swamp LC classes progressively increased with the addition of radar images from
55% to 85%; however, the method used to calculate the overall classification accuracy was unclear.
Results clearly showed that even the classification based on the concatenation of multiple dates (up
to 9 months) of radar images underperformed the single-date MS classification of 92%–97% (two
separate dates classified). In addition, the single-date MS class accuracies associated with the cat-
tail (90%–94%) and phragmites (53%–73%) marshes and a forested wetland (70%–90%) equaled or
outperformed class accuracies based on the nine radar image dates.
In the cloud-prone country of Vietnam, IHS and PCA fusions were applied to an MS (SPOT XS)
and radar (TerraSAR-X) concatenated dataset in order to create fused inputs into an LC classifica-
tion (Hanh and Tuan 2009). The classification produced water, forest, four types of cultivated lands,
transport, and residential classes with an overall accuracy of 94%. In addition, the combination of
concatenation fusion and a maximum likelihood (ML) classification procedure provided adequate
classification of cloud-covered areas on the SPOT image. Fusion performance results could not be
assessed, however, because classifications based solely on MS data were not provided for comparison.
In a mapping study that sought to improve LC classification of a humid and cloud-prone land-
scape, MS (TM) and radar (Radarsat) substitution fusion (PCA and AT) provided >75% accura-
cies associated with rangeland, forest, rice, water, and residential classes but low accuracy (34%)
associated with the agriculture class (Sadidy et al. 2009). As found in other classification studies,
class accuracies varied with classification method. In addition, although clouds prevented LC clas-
sification in parts of the MS image, the fused product produced acceptable classification results
even where clouds obscured the MS image. A similar MS (SPOT XS) and radar (Radarsat) LC clas-
sification also found that fusion, specifically PCA substitution fusion, provided adequate separation
of two forest types even in cloudy regions (Saberioon et al. 2009). Unfortunately, in neither Sadidy
et al. (2009) nor Saberioon et al. (2009), classification results were based solely on MS data provided
for comparison.
Feature-Level Fusion
Feature-extraction methods depend on characteristics of the data sources (Zhang 2012); however,
most common examples of FL fusion are associated with independent or joint segmentation of
optical and radar images (Orsomando et al. 2007). In many cases, segmentation is used to reduce
complexities intrinsic to PL fusion by transforming the fusion approach to the FL; however, the
probability of feature identification is dependent on how many segments are identified; selecting
the wrong features or too many features reduces the FL fusion performance (Kahler and Blasch
2009). Various segmentation methods and FL fusion theories are applied to identify the optimum
mix of features for FL fusion (Orsomando et al. 2007). Many literature examples of segmentation
are accomplished with probabilistic, Bayesian, and other fusion theories that are each applied dif-
ferently to accomplish FL fusion (Nsaibi and Chaabane 2008).
The more common application of segmentation fusion is to identify abrupt change or homogene-
ity on single radar or optical images or image stack (Zhang 2012). There are many different varia-
tions on how to accomplish the segmentation as a basis for FL fusion. One of the clearest formats
for describing the construct of FL fusion based on segmentation is provided by Orsomando et al.
(2007). In that study, the optical and radar joint segmentation fusion was accomplished by first con-
catenation of the optical image bands onto the radar image stack and then application of procedures
involving multivariate probability density function calculations and ML testing for each optical and
radar segment and the combined segments. The ML function was used to define estimates in the
search for abrupt changes with incorporation of both time lapse (series of radar images) and seg-
ment size (Orsomando et al. 2007).
A more a straightforward but still powerful application of FL fusion involved delineating wetland
forest (bottomland hardwood [BLH], cypress-tupelo) LC features identifiable at MS and radar data
of moderate spatial resolutions (25 m) and transfer of that feature information to low-spatial- but
high-temporal-resolution MS data sources (Ramsey et al. 2009a, 2011a). Image data were collected
before, soon after, and 1 month following hurricane wind damage to the cypress and hardwood wet-
land forests. The three dates of optical and radar images were segmented into four feature classes:
cypress forest damage class and severe, moderate, and light damage BLH classes (Figure 8.3a and b).
Comparison of the optical and radar damage-recovery feature classes confirmed that radar could
adequately replace optical damage assessment when persistent cloud cover prevents the collection
of useable optical image data (Ramsey et al. 2009a).
The second study objective was to improve temporal monitoring of LC features that although
evident were not necessarily delineable at the low spatial and spectral resolutions of high revisit and
large coverage satellite sensor systems, such as the twice daily collected 250–1000 m spatial reso-
lution moderate resolution imaging spectrometer (MODIS) onboard the Terra and Aqua satellites
(Figure 8.3c) (Ramsey et al. 2011a). The study purpose was to advance the ability to track subtle,
latent long-term changes of natural resources (e.g., forests) brought about by some causal agent, such
as oil contamination, drought, persistent flooding, or in this case, hurricane impact.
0.9
1
0 1 2 3km
Satellite optical 0.8 0.5
MODIS vegetation index

0.7 0
2004
0.6
0.5
Typical forest phenology
0.4
0.3
Cypress-tupelo
Low impact BLH
0.2 Moderate impact BLH
Severe impact BLH
N 0.1
1 Jan
29 Jul
31 Jan
29 Jun
30 Apr
1 Mar
28 Aug
31 Mar
30 May
27 Sep
27 Oct
26 Nov
26 Dec
0.9
1
0.8 0.5
0.7 0
(a) 0.6 2005
0.5
Hurricane impact
0.4
29 Aug Katrina
0.3
0 1 2 3km
Satellite radar 0.2
0.1
29 Jul
30 Jan
29 Jun
30 Apr
1 Mar
28 Aug
31 Mar
30 May
27 Sep
27 Oct
26 Nov
31 Dec
26 Dec
0 2 4 km 0.9 1
0.8 0.5
0.7 0
Based on 9 October 2005 Based on 25 October 2005 2006
Light-moderate BLH Light damaged BLH
Moderate-servere BLH Moderate damaged BLH 0.6
Cypress Severe damaged BLH
0.5
Latent damage response
25 m impact severity Daily satellite 250 m 0.4
and recovery classes MODIS monitoring 0.3
with NASATSPT 0.2
0.1
29 Jul
30 Jan
29 Jun
30 Apr
1 Mar
28 Aug
31 Mar
30 May
27 Sep
27 Oct
26 Nov
31 Dec
26 Dec
(b) (c) (d) (e)
FIGURE 8.3 Optical and radar cypress and BLH forest impact and recovery features fused to daily VI
monitoring. The forest damage features based on image collections before, right after, and 1–1.5 months after
hurricane impact. (a) MS (25 m), (b) radar (25 m) images, and (c) fused MS and radar damage features (With
kind permission from Springer Science and Business Media: Wetlands, Satellite optical and radar image data
of forested wetland impact on and short-term recovery from Hurricane Katrina in the lower Pearl River flood
plain of Louisiana, USA, 29(1), 2009b, 66–79, Ramsey III, E., Rangoonwala, A., Middleton, B., and Lu, Z.)
(d) A color composite of VI images based on MODIS image data (250 m) collected on the same days as the
MS image data (color rendition as in [a]). (e) The damage features developed on MS and radar image data
were used to segment the daily MODIS-VI image data and produce daily records of VI change from 2004
(prehurricane) through 2006 (1 year after) the hurricane impact (cloud cover percent [top curve], VI vegeta-
tion phenology [bottom curve]). (Adapted from Ramsey III, E. et al., Photogram. Eng. Remote Sens., 77(11),
1133, 2011b.)
The first step in forming the high-temporal-frequency dataset applied the Time Series Product
Tool (TSPT) software produced at NASA Stennis Space Center. TSPT was used to reproject the
multitemporal MODIS image datasets and then compute vegetation indexes (VIs) from the input
red and near-infrared reflectance data. TSPT then removed clouds and restricted the cloud-free data
to sensor zenith angles of less than 47°. The retained quality cloud-free data from Aqua and Terra
MODIS sensors then were fused into composite images that yielded the best available quality data
on a daily basis. The next step was to convert the raster-based forest type and damage features iden-
tified in the 25 m MS and radar image data segmentation into polygons that were overlain on TSPT
generated daily VI image dataset representing 3 years of biophysical data (Ramsey et al. 2011a).
Guided by the MS–radar feature classes, daily VIs representing temporal dynamics of each damage
class were constructed and compared for the year preceding, during, and 1 year after the hurricane
impact (Figure 8.3d). This straightforward fusion process resulted in previously unobserved high-
frequency subtleties of wetland forest regrowth, replacement, and regeneration dynamics that are
used to advance predictions of long-term trends (Ramsey et al. 2011a).
Decision-Level Fusion
Our DL fusion examples illustrate how knowledge gained from external data sources; in particu-
lar, knowledge gained from optical or radar data sources can be used to constrain or guide a fusion
process. Our first examples illustrate such a DL fusion where additional data sources including
MS image data were used to mitigate phase noise in interferometric SAR (NASA Airborne SAR)
LC mapping (Walker et al. 2007). Noise mitigation was achieved by using available presegmented
images to produce image features of sufficient size to provide adequate sample and noise reduc-
tion and to produce homogeneous features (segmented) in terms of topographic slope and vertical
(e.g., height) and horizontal (e.g., canopy density) forest. To provide a segmented landscape based
on the fusion of these variables, a concatenated dataset was formed within a geographic informa-
tion system that included a slope variable layer derived from the National Elevation Dataset, a
canopy density variable layer derived from the National Land Cover Data (NLCD; TM based),
and a vertical canopy height plus ground elevation variable layer derived from InSAR analysis
of the Shuttle Radar Topography Mission SAR data pairs (Walker et al. 2007). FL fusion was
used to extract and combine features in a common decision; however, the final step incorporated
a GIS segmented dataset to constrain the interpretation of the InSAR product. The forest height
estimates were based on a constrained solution set that provided the highest propensity to arrive
at the best decision. Goodenough et al. (2008) proposed a similar concept in fusing MS (NASA
Airborne Visible/Infrared Imaging Spectrometer) and polarimetric radar (JAXA Advanced Land
Observing Satellite Phased Array type L-band SAR); however, processing first decomposed the
polarimetric radar data into estimates of features representing backscatter mechanisms as a first-
order classification.
The final DL fusion example reverses the roles of MS and radar and depicts the straight-
forward enhancement of an MS (TM)-based LC classification with spatial contexture available
from InSAR (ERS-1/2) products (Ramsey et al. 2006). The study site located in the panhandle
region of Florida is centered on a mixed landscape of emergent herbaceous wetlands (marsh),
woody wetlands (bottom hardwood forest), and evergreen forest following the NLCD classifi-
cation based on 1992 MS image data (Wardlow and Egbert 2003) (Figure 8.4). The MS image
data collected on May 1995 includes an evergreen forest parcel that was clear-cut after the 1992
MS data collection (Figure 8.4c and d). The InSAR phase and coherence products were created
from a 6-month suite of coupled ERS-1 and ERS-2 scenes collected 1 day apart in 2005 (Figure
8.4a and b). From a wider regional inspection of the panhandle landscape, phase and coherence
image depictions and histogram results indicated that forest canopy structures were less spatially
uniform (lower coherence) than marshes, and of the forest classes, only BLHs exhibited highly
textured phase and low-coherence patterns in the winter (Ramsey et al. 2006). These results and
the documented flood attenuation of backscatter from these marshes (Ramsey 1995) were applied
to the study site in order to illustrate how InSAR phase and coherence products add informa-
tion relative to optical LC mapping (outside of decreased collection constraints) and what pos-
sible advantages coherence and phase (neglecting vegetation height and deformation) offer with
respect to SAR backscatter intensity.
First, the northwest corner was classified as evergreen forest in the NLCD image (Figure 8.4d);
however, the same area in the 1995 TM image did not exhibit a spectral signature indicative of a for-
est (Figure 8.4c) and the relatively lower backscatter in the radar intensity scenes indicate flooded
marsh (Figure 8.4a and b). Second, the intensity, coherence, and phase distributions indicated that the
NLCD transition class was dominantly marsh transitioning to an older regrowth containing marsh
(highlighted in the blue outlined box). In addition, the clear-cut (highlighted in red outline box)
(a) (b) (c) (d)
0 2π 0 2π
(e) (f) (g) (h)
FIGURE 8.4 Interferometric SAR coherence and phase used to improve LC classifications (From
Ramsey III, E. et al., GISci. Remote Sens., 43(4), 283, 2006). Intensity ERS-1 recorded on (a) October 11,
1995, and (b) December 20, 1995; (c) TM bands 421; and (d) NLCD—emergent herbaceous wetlands, woody
wetlands, evergreen forest, mixed forest, transitional. Coherence recorded on (e) October 11–December 20,
1995; (f) October 11–Decemebr 21, 1995; (g) December 20–21, 1995; and (h) December 20–21, 1995. The
blue box outlines an NLCD transition class that was dominantly marsh or an older regrowth containing
marsh, and the red box outlines an evergreen forest clear-cut between the 1992 NLCD map production and
the TM and ERS radar scene collections in 1995. (Adapted from Ramsey III, E., Lu, Z., Rangoonwala, A.,
and Rykhus, R., Multiple baseline radar interferometry applied to coastal landscape classification and change
analyses, GIScience and Remote Sensing, 43(4), 283–309, 2006, with permission from Taylor & Francis
LTD.). The DL fusion illustration displays how the selective LC classes can be discriminated with InSAR
image products and used to enhance MS LC classifications. The 1995 ERS-1 and ERS-2 scenes captured 1 day
apart were processed to create daily intensity scenes and InSAR coherence and phase products at 2-, 40-day,
and greater time differences. The SAR intensity and InSAR coherence and phase were compared to 1995
Landsat TM image collected within the same time period of the ERS-1/2 scenes and a 1992 NLCD. The USGS
NLCD classified map data and a description of the map construction, classification protocols, and accuracy
assessment can be obtained at http://landcover.usgs.gov/prodescription.php (accessed November 2013).
exhibited high backscatter and coherence indicative of a double-bounce backscatter mechanism con-
ducive for enhancing coherent radar backscatter intensity and increased optical reflectance resulting
from lowered volume scatter attenuation losses (Figure 8.4a through d and f) (Ramsey 1998, 2005).
The accompanying phase change indicated a small change in above surface flood levels between the
two collection dates (Figure 8.4e and f) (Ramsey et al. 2006). Third, outside the clear-cut and tran-
sition areas, the most ubiquitous NLCD class was wetland forest, followed by evergreen, and then
mixed forest (Figure 8.4d); however, the higher-coherence areas depict marshes, and the relatively
lower-coherence areas depict forests (Figure 8.4e). In addition, the low coherences exhibited in winter
were indicative of hardwood not evergreen forests (Figure 8.4g). Taken together, the radar intensity,
MS, coherence, and phase images indicated that extensive portions of the NLCD woody wetlands
class included evergreen and mixed forest classes and that a large portion of the NLCD evergreen and
mixed forest classes included marshes.
FUSION PERFORMANCE
Reported performances of the various fusion processes vary widely. Performance, however, is
somewhat of a subjective term in most published applications and development papers. In all
but a few fusion-based studies referenced, performance was largely a statement on how well the
fused image represented the higher-spatial-resolution component and the spectral resolution of the
moderate-spatial-resolution component. The overriding concern throughout most fusion-based
applications and developments was color distortion exhibited in the final fused image product
(e.g., Amarsaikhan and Douglas 2004; Wang et al. 2005; Gangkofner et al. 2008; Zhang et al.
2010; Yuhendra et al. 2011). In LC mapping, high-quality synthesis of spectral information in the
fusion process is more than a concern of spectral fidelity; it is also necessary for retaining the sen-
sor’s recorded spectral information on which the remote sensing LC classification is based (Wang
et al. 2005).
Visual assessment is often used to evaluate fusion performance. Although seldom applied, a
higher level of fusion assessment, particularly associated with PL fusion, is the statistical analysis of
spectral quality of fused image by comparing input images to fused image correlation coefficients,
mean, standard deviation (STD), etc. (Hong et al. 2009; Yuhendra et al. 2011). In PL fusion, the
objective is to maximize both the MS spectral and Pan (panchromatic or radar) spatial informa-
tion in the fused image; however, there is always a trade-off. For instance, the correlation between
MS and the fused image, and radar and the fused image provides a bulk indicator of information
transfer. If the correlation were high between the radar and fused images, good textural information
transfer is indicated. If, however, the correlation between the MS and fused image is high, good
spectral information transfer is indicated (Yuhendra et al. 2011). Somewhere in between these two
extremes lies the best fusion solution that maximizes the contributions of spectral and spatial (con-
texture) information in the fused image product.
The next higher level of fusion performance assessment is most often associated with LC clas-
sifications, and although mostly constrained to a concatenation subset of PL fusion, it is applied
relatively often throughout FL and DL fusions associated with LC classifications. In FL fusion,
LC classes are largely dependent on the segmentation (i.e., classification) performance and the
landscape composition. Somewhat as in PL concatenation fusion, the DL fusion LC classes are
more likely prespecified than in FL fusion; however, in both PL concatenation fusion and DL clas-
sification fusion, the LC classes tend to be broadly defined and selective relative to the landscape
composition. In both FL and DL, fusion assessment normally involves statistical comparison of
LC class spatial occurrences based on classification of the fused image product to the actual spatial
occurrences of these LC classes. The overriding lack in the assessment of fusion methods is that
LC classes used in the classification accuracy assessments do not follow a standard protocol (e.g.,
Sheoran et al. 2009).
CONCLUSION
Although PL fusion techniques are widely applied, the fusion techniques that include concatena-
tion, mathematical, and IHS and PCA substitution fusions exhibited inconsistent performances.
IHS and PCA transforms are widely acknowledged to have unacceptable spectral distortion and
sensitivity to input variance when applied in fusion processes. Often in PL concatenation–based
fusion examples, even though selected and generalized, LC classes exhibited varied accuracies
when compared to ground-based observations. This inconsistent performance of optical–radar
PL concatenation fusions to advance LC classifications was discussed and fully reported in the
included illustration (Figure 8.2). The illustrated wetland and adjacent upland LC classification
was assessed based on class protocols developed by organizations (e.g., resource managers and
ecologists) not part of the remote sensing classification. That applied study revealed that the fused
concatenation of radar data with MS data could increase the mapping accuracy of some LC classes;
however, that change varied depending on the uniqueness of the MS sensor response relative to the
radar sensor response.
A similar problem occurred when comparing FL fusion performance in increasing mapping
accuracy. After segmentation, the segmented image product was classified into meaningful LC
classes; however, the class definitions were not based on standard protocols but were arbitrarily a
function of the segmentation process. In one sense, the FL fusion might genuinely represent the
highest class detail definable based on the source images and landscape composition. And its direct
association of FL fusion with feature delineation makes this fusion level most similar to standard
image data classifications. Even so, to standardize the FL fusion product and classification assess-
ment, the final classification should be constrained to the classification protocols (classes) by some
means, possibly rules, normally imposed as limits in the process of classifying the fusion-produced
segmented image.
As with other fusion-based classifications, the success of DL in improving biophysical estimation
performance is not fully demonstrated. Although limited in scope, the DL fusion shows promise.
Using similar approaches, Ramsey et al. (2006, 2011a) applied DL fusion in wetland biophysical
mapping that included marshes as well as forested wetlands. Rather than solely improving radar-
based biophysical estimates, these latter studies suggest that the unique and determinable sensitiv-
ity of radar and InSAR imaging to canopy structure could help remove the largely indeterminable
influence of canopy structure from MS wetland mapping. Removal of the unknown canopy influ-
ence would improve the MS mapping of plant leaf pigment and in turn would improve remote sens-
ing detection and monitoring of abnormal plant distress.
The fusion of optical and radar pertinent to LC classification advancement shows promise in
PL, FL, and DL fusions. Of the PL fusion methods, those that incorporate spatial pattern extrac-
tions, such as fusions based on HPF, wavelet transforms, and multiple linear regressions, show
most promise. Both the segmentation FL fusion for LC feature classifications and optimization DL
fusion for LC biophysical classifications show promise. The included FL and DL fusion illustrations
in Figures 8.3 and 8.4 provide straightforward methods to transfer information from high spatial
to low spatial but high temporal scales and from radar and InSAR to MS-based LC classification
without resorting to more complicated optimization approaches.
Although showing promise, fusion developments and applications have resulted in a mixed
record of accomplishments and uncertain direction in the advancement of LC class and biophysical
classifications. In order to improve this mixed record, optical and radar fusion should advance in
two areas; better define the physical basis for the expected improvement and better assess the per-
formance based on standard classification protocols. Concerning the former, most fusion implemen-
tations pay little attention to understanding the physical basis of the values represented on images
collected by different sensor systems. Although there are exceptions, this lack of attention seems
most prevalent in PL fusion and least prevalent in DL fusion. For instance, other than providing
cloud penetration and additional measures of image texture, a physical basis supporting the need for
fusing radar with MS image data is often lacking in PL concatenation fusions.
Requiring better assessment of fusion accuracy should collaterally improve understanding of
the physical basis of the fusion. By defining the quality of the fused product (image) relative to its
performance in accurately representing the LC classes–based standard classification protocols, the
map producer must know the advantages and disadvantages of each sensor representation of the
preset LC classes. With this knowledge and adequate ground-based observations, the choice of
proper fusion type (PL, FL, or DL) and the proper mechanism within each fusion type should lead
to more consistent and extendable classification performance and more direct ability to evaluate and
compare classification assessments.
ACKNOWLEDGMENTS
We thank Dr. Gang Hong with the Earth Observation and Geosolutions Division at the Canada Centre
for Remote Sensing and Professor Valerie Thomas with the Center for Environmental Applications
in Remote Sensing in the Department of Forest Resources and Environmental Conservation at
Virginia Polytechnic Institute and State University for their thoughtful, detailed, and effective
reviews. We are grateful for the technical editing done by the USGS Lafayette Publishing Service
Center. This work was funded entirely by the U.S. Federal Government and performed by federal
employees and employees (under contract no. G11PC00013 with the federal government) of Five
Rivers Services, LLC. Any use of trade, product, or firm names is for descriptive purposes only and
does not imply endorsement by the U.S. government.
REFERENCES
Amarsaikhan, D. and T. Douglas. 2004. Data fusion and multisource image classification. International Journal
of Remote Sensing 25(17):3529–3539.
Boström, H., S.F. Andler, M. Brohede, R. Johansson, A. Karlsson, J. van Laere, L. Niklasson, M. Nilsson,
A. Persson, and T. Ziemke. 2007. On the definition of information fusion as a field of research. Technical
report at the University of Skövde, Skövde, Sweden, HS-IKI-TR-07-006, pp. 1–8.
Dehouch, A., V. Lafon, N. Baghdadi, and V. Marieu. 2012. Use of optical and radar data in synergy for map-
ping intertidal flats and coastal salt-marshes. 2012 IEEE International Geoscience and Remote Sensing
Symposium, IEEE Conference Publications, Arcachon Lagoon, France. pp. 2853–2856.
Florida Game and Fresh Water Fish Commission (FGFC). 1992. Map entitled Florida land cover from landsat
thematic mapper satellite imagery (1985–1989). Tallahassee, FL.
Fonseca, L., L. Namikawa, E. Castejon, L. Carvalho, C. Pinho, and A. Pagamisse. 2011. In: Y. Zheng
(ed.). Image fusion for remote sensing applications, image fusion and its applications, accessed
November 9, 2013. InTech. http://www.intechopen.com/books/image-fusion-andits-applications/
image-fusion-for-remote-sensing-applications.
Gangkofner, U.G., P.S. Pradhan, and D.W. Holcomb. 2008. Optimizing the high-pass filter addition technique
for image fusion. Photogrammetric Engineering and Remote Sensing 74(9):1107–1118.
Goodenough, D.G., H. Chen, A. Dyk, G. Hobart, and A. Richardson. 2008. Data fusion study between polari-
metric SAR, hyperspectral and lidar data for forest information. Proceedings of the IEEE International
Geoscience and Remote Sensing Symposium 2:281–284.
Hall, D.L. and J. Llinas. 1997. An introduction to multisensor fusion. Proceedings of the IEEE 85(1):6–23.
Hanh, T.H. and V.A. Tuan. 2009. Combination of microwave and optical remote sensing in land cover mapping.
Seventh International Federation of Surveyors (FIG) Regional Conference, Hanoi, Vietnam, October
19–22, 2009, pp. 1–10.
Henderson, F.M., R. Chasan, J. Portolese, and T. Hart, Jr. 2002. Evaluation of SAR-Optical imagery syn-
thesis techniques in a complex coastal ecosystem. Photogrammetric Engineering and Remote Sensing
68(8):839–846.
Hong, G. 2007. Image fusion, image registration, and radiometric normalization for high resolution image
processing. Dissertation Thesis, University of New Brunswick, Fredericton, New Brunswick, Canada.
Hong, G., Y. Zhang, and J.B. Mercer. 2009. A wavelet and IHS integration method to fuse high resolution
SAR with moderate resolution multispectral images. Photogrammetric Engineering and Remote Sensing
75(10):1213–1223.
Jensen, J., E. Ramsey, H. Mackey, E. Christensen, and R. Sharitz. 1987. Inland wetland change detection using
aircraft MSS data. Photogrammetric Engineering and Remote Sensing 53:521–528.
Jinghui, Y., Z. Jixian, L. Haitao, S. Yushan, and P. Pengxian. 2010. Pixel level fusion methods for remote sens-
ing images: A current review. In: W. Wagner and B. Székely (eds.). ISPRS TC VII Symposium, Vienna,
Austria, July 5–7, 2010, pp. 680–686. Accessed November 9, 2013. http://www.isprs.org/proceedings/
xxxviii/part7/b/pdf/680_XXXVIII-part7B.pdf.
Kahler, B. and E. Blasch. 2009. Predicted radar/optical feature fusion gains for target identification. Proceedings
of the IEEE 2010 National Aerospace and Electronics Conference (NAECON), Fairborn, Ohio, USA,
pp. 405–412.
Kasischke, E. and L. Bourgeau-Chavez. 1997. Monitoring South Florida wetlands using ERS-1 SAR imagery.
Kimes, D.S. and P.J. Sellers. 1985. Inferring hemispherical reflectance of the Earth’s surface for global energy
budgets from remotely sensed nadir or direction radiance values. Remote Sensing of the Environment
18:205–223.
Klemas, V., J. Dobson, R. Ferguson, and K. Haddad. 1993. A coastal land cover classification system for the
NOAA coastwatch change analysis project. Journal of Coastal Research 9(3):862–872.
Kumaran, T.V. and R. Shyamala. 2002. Land cover mapping: Performance analysis of image-fusion methods.
Geospatial Application Papers 1–3. Accessed November 4, 2014, http://www.geospatialworld.net/Paper/
Application/ArticleView.aspx?aid=399.
Leckie, D.G. 1990. Synergism of synthetic aperture radar and visible/infrared data for forest types discrimina-
tion. Photogrammetric Engineering and Remote Sensing 56(9):1237–1246.
Lunetta, R., J. Lyon, B. Guindon, and C. Elvidge. 1998. North American landscape characterization
dataset development and data fusion issues. Photogrammetric Engineering and Remote Sensing
64(8):821–829.
Lyon, J. and J. McCarthy. 1981. Seasat imagery for detection of coastal wetlands. Proceedings of the 15th
International Symposium on Remote Sensing of Environment, Ann Arbor, MI: ERIM, pp. 1475–1485.
Mercer, J.B., D. Edwards, G. Hong, J. Maduck, and Y. Zhang. 2005. Fusion of InSAR high resolution imagery
and low resolution optical imagery. In: C. Heipke, K. Jacobsen, and M. Gerke (eds.). Proceeding ISPRS
Hannover Workshop 2005: High-Resolution Earth Imaging for Geospatial Information, Hannover,
Germany, May 17–20, 2005.
Mitsch, W. and J. Gosselink. 2000. Wetlands, 3rd edn. John Wiley & Sons, New York. (Note: 4th edition was
published in 2007.)
Moghaddam, M. and K. McDonald. 2003. Mapping wetlands of North American boreal zone from satel-
lite radar imagery. Proceedings, Geoscience and Remote Sensing Symposium, IGARSSS’03, Toulouse,
France, pp. 261–263.
Nsaibi, M. and F. Chaabane. 2008. Image fusion of radar and optical remote sensing data for land cover classi-
fication. Third International Conference on Information and Communication Technologies: From Theory
to Applications, Damascus, Syria, pp. 1–4.
Orsomando, R., P. Lombardo, M. Zavagli, and M. Constantini. 2007. SAR and optical data fusion for change
detection. IEEE Conference Publications, Urban Remote Sensing Joint Event Conference, Paris,
France.
Palubinskas, G. and P. Reinartz. 2010. Fusion of optical and radar remote sensing data: Munich city
example. In: W. Wagner and B. Székely (eds.). Proceedings Thematic Processing, Modeling and
Analysis of Remotely Sensed Data, ISPRS Technical Commission VII Symposium, Vienna, Austria,
July 5–7, 2010.
Palubinskas, G. and P. Reinartz. 2011. Multi-resolution, multi-sensor image fusion: General fusion frame-
work. In: U. Stilla, P. Gamba, C. Juergens, and D. Maktav (eds.). Joint Urban Remote Sensing Event
Conference Publication, Munich, Germany, pp. 313–316.
Pohl, C. and J.L. Van Genderen. 1998. Review article multisensor image fusion in remote sensing: Concepts,
methods and applications. International Journal of Remote Sensing 19(5):823–854.
Ramsey III, E. 1995. Monitoring flooding in coastal wetlands by using radar imagery and ground-based mea-
surements. International Journal of Remote Sensing 16:2495–2502.
Ramsey III, E. 1998. Radar remote sensing of wetlands. In: R. Lunetta and C. Elvidge (eds.). Remote Sensing
Change Detection: Environmental Monitoring Methods and Applications. Ann Arbor Press, Inc., Ann
Arbor, MI, pp. 211–243.
Ramsey III, E. 2005. Remote sensing of coastal environments. In: M.L. Schwartz (ed.). Encyclopedia of
Coastal Science. Kluwer Academic Publishers, Dordrecht, the Netherlands, pp. 797–803.
Ramsey III, E., M. Hodgson, S. Sapkota, and G. Nelson. 2001a. Forest impact estimated with NOAA AVHRR
and Landsat TM data related to an empirical hurricane wind field distribution. Remote Sensing of the
Ramsey III, E. and S. Laine. 1997. Comparison of landsat thematic mapper and high resolution photography to
identify change in complex coastal marshes. Journal of Coastal Research 13(2):281–292.
Ramsey III, E., Z. Lu., A. Rangoonwala, and R. Rykhus. 2006. Multiple baseline radar interferometry
applied to coastal landscape classification and change analyses. GIScience and Remote Sensing
43(4):283–309.
Ramsey III, E., G. Nelson, and S. Sapkota. 1998. Classifying coastal resources by integrating optical and radar
imagery and color infrared photography. Mangroves and Salt Marshes 2(2):109–119.
Ramsey III, E., G. Nelson, and S. Sapkota. 2001b. Coastal change analysis program implemented in Louisiana.
Journal of Coastal Research 17(1):55–71.
Ramsey III, E., G. Nelson, S. Sapkota, S. Laine, J. Verdi, and S. Krasznay. 1999. Using multiple polarization
L band radar to monitor marsh burn recovery. IEEE Transactions on Geoscience and Remote Sensing
37(1):635–639.
Ramsey III, E. and A. Rangoonwala. 2006. Site-specific canopy reflectance related to marsh dieback onset and
progression in Coastal Louisiana. Photogrammetry and Remote Sensing 72(6):641–652.
Ramsey III, E. and A. Rangoonwala. 2010. Mapping the onset and progression of marsh dieback. In:
Y. Wang (ed.). Remote Sensing of Coastal Environments. CRC Press, Boca Raton, FL, Remote Sensing
Applications Series, pp. 123–150.
Ramsey III, E. and A. Rangoonwala. 2011. Remote sensing of wetland vegetation focusing on hyperspectral
mapping. Dr. Prasad (editor-in-chief) and S. Thenkabail, J.G. Lyon, and A. Huete (eds.). Hyperspectral
Remote Sensing of Vegetation. CRC Press, Boca Raton, FL, pp. 487–512.
Ramsey III, E., A. Rangoonwala, and T. Bannister. 2013. Coastal flood inundation monitoring with satellite
C-band and L-band synthetic aperture radar data. Journal of the American Water Resources Association
49:1–22.
Ramsey III, E., A. Rangoonwala, F. Barnes, and R. Spell. 2009a. Mapping fire scars and marsh recovery with
remote sensing image data. In: X. Yang (ed.). Remote Sensing and GIS for Coastal Ecosystem Assessment
and Management, Springer, Berlin, Germany. Publisher Springer Lecture Notes in Geoinformation and
Cartography, pp. 415–438.
Ramsey III, E., A. Rangoonwala, and R. Ehrlich, 2005. Mapping the invasive species, Chinese Tallow with
EO1 satellite Hyperion hyperspectral image data and relating tallow percent occurrences to a classified
Landsat Thematic Mapper landcover map. International Journal of Remote Sensing 26(8):1637–1657.
Ramsey III, E., A. Rangoonwala, B. Middleton, and Z. Lu. 2009b. Satellite optical and radar image data of for-
ested wetland impact on and short-term recovery from Hurricane Katrina in the lower Pearl River flood
plain of Louisiana, USA. Wetlands 29(1):66–79.
Ramsey III, E., A., Rangoonwala, Y. Suzuoki, and C. Jones. 2011a. Oil detection in a coastal marsh with polari-
metric SAR. Remote Sensing 3:2630–2662.
Ramsey III, E., S. Sapkota, F. Barnes, and G. Nelson. 2002. Monitoring the recovery of Juncus roemerianus
marsh burns with the normalized vegetation index and landsat thematic mapper data. Wetlands Ecology
and Management 10(1):85–96.
Ramsey III, E., J. Spruce, A. Rangoonwala, Y. Suzuoki, J. Smoot, J. Gasser, and T. Bannister. 2011b. Daily
MODIS trends of hurricane-induced forest impact and early recovery. Photogrammetric Engineering and
Ramsey III, E., D. Werle, Y. Suzuoki, A. Rangoonwala, and Z. Lu. 2012. Limitations and potential of optical
and radar satellite imagery to monitor environmental response to coastal emergencies in Louisiana, USA.
Journal of Coastal Research 28(2):457–476.
Saberioon, M.M., M. Mardan, L. Nordin, A.M. Sood, and A. Gholizaeh. 2009. Fusion SPOT-5 & Radarsat-1
images for mapping major bee plants in Marang District, Malaysia. European Journal of Scientific
Research 38(3):465–473.
Sadidy, J., P.Z. Firouzabadi, and A. Entezari. 2009. The use of Radarsat and Landsat image fusion algorithms and
different supervised classification methods to improve land use map accuracy—Case study: Sari Plain—
Iran. Accessed November 9, 2013. http://www.cirgeo.unipd.it/cirgeo/convegni/mmt2007/proceedings/
papers/sadidy_javad.pdf.
Sahagian, D. and J. Melack. 1996. Global wetland distribution and functional characterization: Trace gases
and the hydrologic cycle. Report of the Joint GAIM-DIS-BAHC-IGAC-LUCC Workshop, held in Santa
Barbara, CA, May 16–20, 1996.
Sheoran, A., B. Haack, and S. Sawaya. 2009. Land cover/use classification using optical and quad polarization
radar imagery. ASPRS 2009 Annual Conference, Baltimore, MD, March 9–13, 2009.
Wald, L. 2000. A conceptual approach to the fusion of earth observation data. Surveys in Geophysics 21:177–186.
Walker, W.S., J.M. Kellndorfer, E. LaPoint, M. Hoppus, and J. Westfall. 2007. An empirical InSAR-optical
fusion approach to mapping vegetation canopy height. Remote Sensing of the Environment 109:482–499.
Wang, J., J. Shang, B. Brisco, and R.J. Brown. 1998. Evaluation of multidate ERS-1 and multispectral landsat
imagery for wetland detection in southern Ontario. Canadian Journal of Remote Sensing 24(1):60–68.
Wang, Z., D. Ziou, C. Armanakis, D. Li, and Q. Li. 2005. A comparative analysis of image fusion methods.
IEEE Transactions on Geoscience and Remote Sensing 43(6):1391–1402.
Wardlow, B. and S. Egbert. 2003. A state-level comparative analysis of the GAP and NLCD land-cover data
sets. Photogrammetric Engineering and Remote Sensing 69(12):1387–1397.
Wegner, J.D., J. Inglada, and C. Tison. 2007. Image analysis of fused SAR and optical images deploying open
source software library OTB. In: C. Heipke, K. Jacobsen, M. Gerke (eds.). ISPRS Hannover Workshop
2007: High-Resolution Earth Imaging for Geospatial Information, Hannover, Germany, Vol. XXXVI-1/
W51, May 29–June 1, 2007. Accessed November 9, 2013. http://www.isprs.org/proceedings/
XXXVI/1-W51/paper/Wegner_inglada_tison.pdf.
Wickland, D. 1991. Mission to planet earth: The ecological perspective. Ecology 72:1923–1933.
Xie, H. and G. R. Keller. 2006. Fusion of Landsat ETM+ and radar data to enhance the extraction of surface
and near-subsurface information. In: A.K. Sinha (ed.). Geoinformatics: Data to Knowledge. Geological
Society of America Special Paper, Boulder, CO, Vol. 397, pp. 141–151.
Yuhendra, J.S., J.T. Sri Sumantyo, and H. Kuze, 2011. Performance analyzing of high resolution pan-sharpening
techniques: Increasing image quality for classification using supervised kernel support vector machine.
Research Journal of Information Technology 3:12–23.
Zhang, J. 2012. Multi-source remote sensing data fusion: Status and trends. International Journal of Image and
Data Fusion 1(1):5–24.
Zhang, J., C. Qi, and W. Tang. 2008. Multi-sensor image fusion based on transferable parameters. Proceedings
of the IEEE International Geoscience and Remote Sensing Symposium 2:1096–1099.
Zhang, J., J. Yang, Z. Zhao, H. Li, and Y. Zhang. 2010. Block-regression based fusion of optical and SAR imag-
ery for feature enhancement. International Journal of Remote Sensing 31(9):2325–2345.
Zhang, Y. 2004. Understanding image fusion. Photogrammetric Engineering and Remote Sensing
68(8):839–846.
Zhang, Y. and G. Hong. 2005. An IHS and wavelet integrated approach to improve pan-sharpening visual qual-
ity of natural colour IKONOS and QuickBird images. Information Fusion 6:225–234.
9 Theory and Applications of
Object-Based Image Analysis
and Emerging Methods
in Wetland Mapping
Joseph F. Knight, Jennifer M. Corcoran,
Lian P. Rampi, and Keith C. Pelletier
CONTENTS
Introduction..................................................................................................................................... 175
OBIA Background.......................................................................................................................... 175
Decision Tree Classifier Background.............................................................................................. 176
History of OBIA Methods in Wetland Mapping............................................................................. 177
Case Studies.................................................................................................................................... 180
Discussion and Conclusion............................................................................................................. 188
References....................................................................................................................................... 189
INTRODUCTION
This chapter addresses object-based image analysis (OBIA) and other emerging methods for map-
ping wetlands using remotely sensed and ancillary data. This treatment includes an introduction to
OBIA and decision tree techniques, a history of the development of OBIA methods in wetland map-
ping, and four case studies describing recent wetland mapping research conducted in the Remote
Sensing and Geospatial Analysis Laboratory at the University of Minnesota. The chapter ends with
a discussion of lessons learned and possibilities for future applications.
OBIA BACKGROUND
Traditional approaches to mapping land cover are photointerpretation and semiautomated per-pixel
classifiers. Human analysts use the elements of image interpretation (EII) for delineating and label-
ing features such as buildings, transportation infrastructure, agricultural fields, forests, wetlands,
soils, and grasslands (Olson 1960). While these maps can be detailed and accurate, the process is
labor intensive, which limits the frequency of updates (Goetz et al. 2003; Van and Bao 2010). Per-
pixel approaches rely on spectral information to either aggregate statistically similar pixels for class
labels (e.g., unsupervised) or use training data to group similar pixels into classes reflecting the
training data (e.g., supervised).
Unlike per-pixel approaches that focus on classification of individual pixels in an image, object-
based analyses delineate and classify homogeneous groups of pixels or objects using spectra, shape,
size, texture, and spatial context, producing land cover classes that better approximate real-world fea-
tures (Benz et al. 2004; Blaschke 2003, 2010; Blaschke and Strobl 2001; Kettig and Landgrebe 1976).
175
These efforts integrate geospatial data from different sensors with varying resolutions and vector
data (Benz et al. 2004). The ability to incorporate disparate such data (e.g., multiple resolutions and
GIS layers) into object-based classification often results in higher mapping accuracy compared to
traditional pixel-based classification techniques (Baatz and Schape 2000; Benz et al. 2004; Blaschke
2010; Bock et al. 2005; Haralick and Shapiro 1985).
An object-based approach starts with the segmentation process, in which an image is partitioned
into image objects based on color, shape, and size. These image object primitives contain attributes
based on spectra, geometry, texture, and context. These attributes are similar to the information
employed by human interpreters using the EII. Classes are defined by creating fuzzy ranges or
thresholds, which are applied to the image objects. The process is iterative, in which some or all of
the image object primitives are classified and then merged or further segmented to delineate addi-
tional classes. An iterative approach to segmentation and classification reduces oversegmentation
and undersegmentation, resulting in objects that correspond to the boundaries of land cover features
(Witharana et al. 2014). Object-based approaches have been successfully used for mapping different
land cover features such as transportation infrastructure (Nobrega et al. 2008; Repaka and Truax
2004), urban infrastructure and vegetation (Frauman and Wolff 2005; Harayama and Jaquet, 2004;
O’Neil-Dunne et al. 2012; Syed et al. 2005; Zhou and Troy 2008), wildlife habitat (Blaschke 2010;
Bock et al. 2005), land use (Kressler and Steinnocher 2008), buildings (Chen et al. 2006; Meng and
Peng 2009), and forests (Kim et al. 2009).
Context can be used for classifying features based on landscape patterns or distances between
objects. As additional classes are identified, the contextual information increases such that areas can
be defined based on pattern; that is, an area with few buildings and roads suggests a rural landscape,
while an area with numerous buildings and roads suggests an urban landscape (MacFaden et al.
2012). The areas can be classified as urban and not urban, while the objects within these classes are
reclassified based on spectral geometry and additional context. For example, a recently plowed or
harvested field can have similar spectral and geometric properties as impervious surfaces. However,
given the landscape context (e.g., rural), the objects are classified as agricultural fields rather than
large parking areas.
Context can be also used to classify land cover features based on topological relationships
or shared boundaries between objects. The relative border or percentage of common boundary
between classes can be used to separate or aggregate neighboring features. Impervious surfaces
found in urban infrastructure often have similar spectral and geometric characteristics as exposed
soils or barren areas. For example, both features often have high reflectance with long and narrow
shape. The spectral and geometric similarities could result in stream banks or shorelines being
misclassified as impervious surfaces. The spatial context or shared border with objects classified as
water allows for reclassifying these features as exposed soils rather than impervious features.
Wetland classes are challenging to classify given the need to characterize vegetation, hydrol-
ogy, and soils for accurately identifying wetland boundaries and differentiating wetland features
from similar upland features. Traditional wetland maps have been produced using photointerpre-
tation (e.g., Tiner 1990) or pixel-based classifiers (e.g., Ozesmi and Bauer 2002). Object-based
approaches for mapping wetlands have integrated optical data, LiDAR-derived surface layers, and
vector data. Recent efforts using LiDAR or integrating LiDAR and high-resolution optical data
have been shown to improve identification of wetlands (e.g., Corcoran et al. 2013; Kim et al. 2011;
Knight et al. 2013; Lang and McCarty 2009; Lang et al. 2012, 2013; Maxa and Bolstad 2009;
Rampi et al. 2014a).
DECISION TREE CLASSIFIER BACKGROUND

Many researchers report that decision tree classifiers can improve wetland mapping accuracy over
traditional approaches such as maximum likelihood classifier (MLC; Baker et al. 2006; Hogg
and Todd 2007; Rodriguez-Galiano et al. 2012; Rover et al. 2011). A decision tree (e.g., Random
Theory and Applications of Object-Based Image Analysis 177
Forest [RF], Classification And Regression Tree [CART], and See5) is a supervised classifica-
tion algorithm that works by creating decision points, or nodes, based on the diagnostic features
of the input variables. Combining these decision points in a set of sequential nodes results in the
“growing” of a “tree” of nodes that can be used to partition the input data into classes of interest
(leaves). RF extends this approach by creating a “forest” of hundreds of trees, each of which has
a “vote” as to the correct class for a given case (e.g., a pixel). In this ensemble approach, the class
chosen by the majority of the trees in the forest is selected. RF uses out-of-bag (OOB) sampling to
compute a measure of internal cross-validation accuracy, in addition to any independent reference
data–based accuracy assessment. Decision tree classifiers are attractive for land cover mapping,
in general, and wetland mapping, in particular, because they require no assumptions about the
underlying distributions of the input datasets, are able to use both continuous and categorical data,
and can be resistant to overfitting (especially RF). Also, decision trees may help to reduce the
subjectivity involved in defining OBIA parameters for wetland mapping (Breiman 2001; Friedl
and Brodley 1997; Quinlan 1993; Smith 2010).
A prominent example of a decision tree classification approach is that used to create the 2001
version of the National Land Cover Data (NLCD) (Homer et al. 2007). The decision tree used
inputs a wide variety of data types: multidate Landsat imagery, topographic derivatives, impervious
surface maps, wetland maps, and other ancillary datasets. Though decision tree use in land cover
mapping is now common, use in the OBIA context is developing but shows great promise (Corcoran
et al. 2013, 2011; Smith 2010).
HISTORY OF OBIA METHODS IN WETLAND MAPPING

The history of OBIA methods used in wetland and aquatic vegetation mapping is relatively brief.
Though image segmentation based on pixel spectral response was a technique introduced by Kettig
and Landgrebe (1976) in the form of their “extraction and classification of homogeneous objects”
(ECHO) algorithm, computing speed, software availability, and algorithmic sophistication did not
allow for widespread adoption of what we now call OBIA techniques until the early 2000s. The
first peer-reviewed study using OBIA to map wetlands was Costa et al. (2002), which used a spec-
tral value segmentation and subsequent region-based classification of Radarsat and JERS-1 radar
images to map Amazon floodplain communities. The authors reported overall accuracies of 97% for
classification of aquatic vegetation, flooded forest, and unflooded forest. However, the use of image
segmentation was not the focus of the study. The authors concerned themselves more with the per-
formance of radar imagery for vegetation mapping. Their innovative use of image segmentation for
mapping wetland vegetation was not discussed.
In a similar study, Hess et al. (2003) used spectral-based image segmentation of JERS-1 radar
imagery to map wetlands in the Amazon basin. Once derived, the segments were classified using
an unsupervised approach based on Mahalanobis distance followed by manual editing of “outli-
ers.” The authors reported 95% wetland/nonwetland mapping accuracy and suggested that, with a
combination of automated classifiers and interpreter editing, image segmentation is well suited for
mapping wetlands at high spatial resolution.
In a pair of 2004 studies, Wang et al. (2004a,b) provided the first peer-reviewed reports of
wetland mapping using image segmentation applied to high-resolution optical satellite imagery.
The authors used IKONOS and QuickBird imagery to map mangroves on the Caribbean coast of
Panama. The 2004a study describes the use of eCognition for object-based segmentation and classi-
fication using MLC, but since the focus of the research was on comparing IKONOS and QuickBird
for mangrove mapping, the authors did not provide much detail on the OBIA approach or its ben-
efits. In contrast, Wang et al. (2004a) used IKONOS imagery to compare the suitability of three
classifiers for mangrove mapping: pixel-based MLC, object-based nearest neighbor (NN), and a
hybrid pixel and object-based classifier (MLCNN). Segmentation was done using the Bhattacharyya
distance measure to group similar pixels. The hybrid MLCNN classifier outperformed the others
(91% overall accuracy versus 89% for MLC and 80% for NN). The authors report a number of use-
ful conclusions. The scale parameter is an important component of the segmentation process but is
subjective. Using the Bhattacharya distance helped to optimize the selection of the scale parameter
and subsequently confirm that separability of mangroves is improved at the object level over the
pixel level. However, a pixel-based classifier may preserve “rich spectral information” in the image,
while OBIA methods tend to generalize such information at the object level. Thus, when mapping
areas with fine spatial reflectance variability, objects may contain mixed classes. The authors report
that such mixed objects were unavoidable, particularly near object edges and class boundaries. The
better performance of the hybrid classifier indicates that it is important to maintain pixel-based
spectral information when spectral variability is high.
Harken and Sugumaran (2005) is the first peer-reviewed wetland mapping paper using OBIA
with hyperspectral imagery and also the first to explicitly mention using textural measures in addi-
tion to spectral information in the segmentation process. The authors used 0.6 m Compact Airborne
Spectrographic Imager (CASI) airborne hyperspectral imagery in a comparison of two classifi-
cation approaches: the pixel-based spectral angle mapper (SAM) and an OBIA approach using
eCognition v3. The authors describe using a texture image to assist in segmenting the hyperspectral
imagery, though little detail is provided on that process. Nine wetland and upland land cover classes
were mapped, with the OBIA approach achieving 92% overall mapping accuracy versus 65% for the
SAM. In anticipation of new directions in wetland mapping, the authors recommended evaluation of
decision tree classifiers such as CART and multitemporal imagery as future work.
Fournier et al. (2007) and Grenier et al. (2007, 2008) jointly described the approach used in the
Canadian Wetland Inventory (CWI). Fournier et al. laid out the general case for using OBIA meth-
ods. Grenier et al. (2007, 2008) followed with case studies of OBIA methods for mapping CWI wet-
land classes. The 2008 study used Landsat-7 multispectral and panchromatic bands with Radarsat-1
imagery in a multiscale segmentation approach—the first mention of multiscale segmentation in a
wetland mapping paper. Supervised NN and fuzzy membership function (MF) approaches incor-
porating spectral, textural, and shape measures were used. The MF approach is described as being
superior due to its greater flexibility. The authors reported overall accuracies for their seven-class
wetland maps as 68%–76%, depending on study area. Li and Chen (2011) described a more devel-
oped rule-based version of a CWI mapping protocol.
Gilmore et al. (2008) provided the first study of LiDAR data in combination with optical imagery
to map wetlands using OBIA. This study was also the first to attempt to map an invasive wetland spe-
cies (Phragmites) with OBIA. The authors used multidate QuickBird imagery with LiDAR-derived
vegetation height data to map marsh plant communities in Connecticut. The authors described the
segmentation parameters used, but did not provide detail on the subsequent classification approach.
Reported accuracies ranged from 63% to 97%, depending on species, with the higher accuracy
applying to Phragmites. Seasonal variability was described as an important consideration when
mapping marsh plant communities, so the authors recommended multitemporal imagery for such
inventories.
By 2008, the basic elements of effective OBIA wetland mapping methods were in place: (1)
segmentation based not just on spectral attributes but also on size, shape, texture, and context, (2)
classification using object-derived measures, and (3) the use of multidate high-resolution optical
imagery, radar, and LiDAR derivatives. Several subsequent studies have confirmed or extended
these results. Conchedda et al. (2008) used multiresolution segmentation to map mangroves in
Senegal using multidate (1986 and 2006) SPOT XS imagery. The object classification was per-
formed using an NN classifier, with an overall accuracy estimate of 86% and a wetland class user’s
accuracy of 97%. Shen et al. (2008) used multiresolution segmentation and MF classification of
ENVISAT ASAR imagery to map inundated areas in the Poyang Lake, China, area. The authors
compared their OBIA approach to an MLC and reported that accuracy estimates of inundation
increased from 86% to as high as 95%. Dissanska et al. (2009) used multiresolution segmentation
and MF classification of QuickBird panchromatic images to map peatlands in James Bay, Quebec,
Canada. Peatland accuracy estimates were 95% for producer’s accuracy and 88% for users. The
overall classification accuracy of their five classes was 88%.
Frohn et al. (2009) provide several recommendations for OBIA-based wetland mapping. The
authors claimed to be the first to attempt to map isolated wetlands using satellite imagery. They
used Landsat-7 imagery and ancillary data to map isolated wetlands of various sizes in Florida
(United States) with an MF classification approach. Reported accuracy estimates for isolated
wetlands varied depending on their size. For small wetlands (>0.5 acres), the user’s and producer’s
accuracies were 88% and 89%, respectively. For larger wetlands (>2 acres), the user’s and
producer’s accuracies were 97% and 95%, respectively. The authors noted that Landsat imagery is
suitable to meet the U.S. Federal Geographic Data Committee (FGDC) wetland mapping accuracy
standards. They recommended using the wettest Landsat scene of the year, reported that rainfall
may not be a useful input to the classification process and suggested using Landsat Band 5 to
determine moisture, and concluded that image segmentation of Landsat data has great potential for
mapping isolated wetlands.
Several recent studies have compared pixel-based versus OBIA methods or assessed ways to opti-
mize the performance of OBIA approaches. Laba et al. (2010) mapped invasive wetland plants in
the Hudson River National Estuarine Research Reserve using IKONOS imagery. They reported that
OBIA textural methods combined with per-pixel spectral information resulted in slight increases
in accuracy over maximum likelihood alone (78% versus 76%), though they suggested that their
field-based method might be biased in favor of pixel-based techniques. Dronova et al. (2011, 2012)
assessed the effects of segmentation scale, thematic detail, and classification methods on the accuracy
of wetland plant functional type maps in Poyang Lake, China. The authors reported that the highest
classification accuracies were obtained using (1) scales that were substantially coarser than pixels, (2)
classification schemes that match the spatial and spectral hierarchy of the landscape, and (3) classifi-
cation methods that operate at multiple scales appropriate to the plant community patch sizes.
Kamal and Phinn (2011) assessed pixel and OBIA methods for mapping mangroves using CASI-2
hyperspectral imagery. They compared SAM and linear spectral unmixing (LSU) pixel-based
approaches with multiscale segmentation. The mapping results showed that the SAM produced
accurate class polygons with only a few unclassified pixels (overall accuracy of 69%), LSU resulted
in a patchy polygon pattern with many unclassified pixels (overall accuracy of 56%), and the object-
based mapping produced the most accurate results (overall accuracy of 76%). Kamal et al. (2014)
extended their study of mangroves by describing the relationship between mangrove patch size and
pixel size in WorldView-2 imagery. They concluded that semivariograms can be used to optimize
scale parameters for more accurate multiscale mapping of mangroves.
Continuing the exploration of optimal OBIA methods, Kim et al. (2011) provided the first
wetland-focused study of 0.3 m aerial imagery in an OBIA context. The authors examined three
issues: a comparison of single- and multiscale segmentation, the benefits of incorporating the gray-
level co-occurrence matrix (GLCM) textural measure in classification, and the effects of GLCM
quantization level in OBIA. Salt marsh classes mapped included vegetation, channel, and mud. The
single-scale approach was found to be the least accurate (76% overall). The multiscale method using
only spectral information performed better (82%). Adding the GLCM texture measure increased
the accuracy to 89%. GLCM quantization level was found to affect the value of including texture
in the classification. Ouyang et al. (2011) suggested that multiscale segmentation provides greater
accuracy benefits than texture and shape features.
Building on work such as Kim et al. (2011) and the preceding technical exploration studies,
Moffett and Gorelick (2013) performed a sensitivity analysis to determine optimal data and method
choices for wetland vegetation and channel mapping in San Francisco Bay (California, United
States). They concluded that, overall, optimal choices depend on the data used, the wetland features
to be mapped, and “conceptual models of wetland organization.” Specifically, the authors recom-
mended high-resolution CIR or RGB imagery, multiresolution segmentation, and scale parameters
appropriate for the patch size of the landscape and the data used. They noted that high-resolution
LiDAR data were not beneficial in distinguishing channels or other features. Concerns raised
included the following: feature shape affects the quality of segmentation results; guidance in using
OBIA is difficult to generalize from the literature due to inherent customization of the approach
for particular study areas and input data; and even optimal OBIA methods may not be reproducible
because of the human interpretation factor.
In the first OBIA-specific wetland mapping study to evaluate a decision tree classifier, Smith
(2010) used the RF algorithm to determine an optimal scale parameter and then to classify SPOT
4/5 images into wetland classes. The study was conducted in North and South Dakota (United
States). The overall accuracy for the seven mapped classes was estimated at 85%, with wetland veg-
etation producer’s and user’s accuracies of 73% and 81%, respectively. The author suggested that,
given the substantially increased numbers of data-derived variables allowed by OBIA techniques,
using an algorithmic approach such as RF may reduce the subjectivity suggested by Moffett and
Gorelick (2013) and others. Despite challenges in the use of OBIA techniques, Antonellini et al.
(2014) suggested that such methods may be especially appropriate for making land use suitability
maps for assessment and management of freshwater resources.
CASE STUDIES
The following are four case studies based on work completed at the University of Minnesota’s
Remote Sensing and Geospatial Analysis Laboratory and published in peer-reviewed journals
between 2011 and 2014. These case studies are intended to provide an overview of recent research
in the context of broader wetland mapping efforts. The cases progress from data selection, to data
preprocessing, to mapping methods and results. Both OBIA and decision tree classifiers were used.
The first case is an assessment of various data types for their usefulness in mapping wetlands.
The second examines topographic preprocessing options for computing the Compound Topographic
Index (CTI), a commonly used topographic derivative for wetland mapping. Third is a study of
wetland mapping of northern Midwest wetlands using multisource and multitemporal data. The
last case presents a study of integration of high-resolution optical and LiDAR data using an OBIA
approach in different types of Midwest wetlands.
Case Study 1 Data Type Suitability for Mapping Midwest

Wetlands (Based on Knight et al. 2013)
Choosing the most useful input data is one of the most important steps to ensuring a high-
quality classification result—regardless of the method used. Given that wetlands are often more
difficult to accurately map than other land cover classes, the use of optimal input data is neces-
sary (Ozesmi and Bauer 2002). Many studies have been done of wetland mapping approaches,
OBIA and other, examining both appropriate input data and classification algorithms. Most of
those studies have focused on radar or Landsat-like sensors (e.g., Baker et al. 2006; Lunetta and
Balogh 1999; Wright and Gallant, 2007). However, in recent years, the increasing availability
of high-spatial-resolution imagery and LiDAR is allowing for new approaches to mapping wet-
lands. Though several studies address this question (e.g., Bowen et al. 2010; Halabinski et al.
2011; Laba et al. 2008; Lang and McCarty 2009; Maxa and Bolstad 2009), there is a need for
broader study of the available high-resolution data types to determine which potentially provide
the largest increases in mapping accuracy. The objective of this research was to examine the
effects of data type selection on wetland mapping accuracy.
This research question was evaluated in two different physiographic regions in Minnesota
(United States): (1) an urban–suburban area in the Twin Cities Metropolitan Area and (2) a
northern boreal forest. The data used included high- and low-resolution topographic deriva-
tives (digital elevation model [DEM], CTI, curvature), National Agricultural Imagery Program
(NAIP) 1 m optical imagery with infrared, 0.6 m spring leaf-off aerial imagery, SSURGO soil
maps, and C-band Radarsat-2 imagery (forested area only). The CTI (Beven and Kirkby 1979)
uses the local upslope contributing area, also known as specific catchment area (SCA), and local
slope to indicate the relative potential wetness of landscape elements. Experimental trials were
conducted by varying input data to the classifier, for example, all available data, optical but no
topography, radar included or excluded, high- versus low-resolution topographic data, and topo-
graphic derivatives included or excluded. The decision tree classifier See5 was used to evaluate
all combinations of input data. The classifier was trained with image- and ground-based train-
ing data. Independent image- and ground-based reference data were used to evaluate the accu-
racies of the various trials. Classes used were a simple wetland/nonwetland determination and
the Cowardin classification system for wetland types (Cowardin and Myers 1974).
The results of this study reinforced the conclusions of existing research and suggested inter-
esting cost–benefit questions for future projects. First, the importance of topographic data to
accurate wetland classification results cannot be overstated. In both study areas, when topog-
raphy layers were removed from the input data stack, accuracies declined significantly. In par-
ticular, when NAIP imagery only was used, accuracy fell by 15%–31%, depending on whether
the simple wetland/nonwetland or Cowardin wetland-type classification schemes were used.
However, the source of the topography data was not a significant factor. Mapping accuracies
were statistically indistinguishable, no matter whether the topographic variables were derived
from the 10 m National Elevation Dataset (NED) or the 3 m LiDAR data. Thus, the presence
of topographic information was more important than its source. This result suggested that if
cost is a significant issue, and high spatial resolution is not a critical specification of a given
project, relatively low-resolution topographic data, when used with imagery, may provide suf-
ficient wetland mapping accuracy. The comparisons between CTI and curvature also indicated
no significant difference in accuracy estimates, regardless of spatial resolution. Thus, in areas
similar to those used in this study, which are not topographically diverse, the simpler curvature
derivative may be a suitable choice over the more computationally intensive CTI.
Second, the choice of classification thematic detail had a significant effect on mapping accu-
racy estimates. Wetland/nonwetland accuracies ranged from 93% to 79% for the suburban and
forested study areas, respectively. The more specific Cowardin wetland-type mapping accura-
cies were lower in both areas: 84% and 58%, respectively. Accuracy estimates were lower in the
forested area because of canopy occlusion of ground features and the often-subtle reflectance
differences between upland and wetland tree species. Thus, as expected, using the simplest clas-
sification scheme that meets the needs of a particular project is recommended.
Finally, the inclusion of Radarsat-2 or leaf-off imagery did not result in higher mapping accu-
racies. C-band radar is not ideally suited for wetland mapping in forested areas due to relatively
low canopy penetration compared with other radar wavelengths, so the lack of improvement
in accuracy was unsurprising. However, spring leaf-off imagery was expected to provide sub-
stantial benefits, due to an enhanced ability to see ground features and moisture and because
of its suitability for discriminating coniferous and deciduous vegetation. Possible explanations
for these results include the particularly dry conditions existing during the radar and leaf-off
acquisitions and canopy illumination differences in the high-spatial-resolution leaf-off imagery.
Case Study 2 Topographic Data Preparation for Wetland

Mapping (Based on Rampi et al. 2014b)
Topography layers and derivatives are often used as inputs to wetland mapping methods because
of their suitability for showing where water might stand on the landscape and thus where
wetlands might exist. The topographic derivative CTI is based on local slope and the SCA.
Because slope is a simple, intuitive calculation, the accuracy of the CTI depends much more on
the quality of the SCA. The SCA is computed using a flow accumulation layer that is normally
derived from a flow direction layer. A flow direction layer can be computed in several ways
that are broadly separated into single flow direction (SFD) and multiple flow direction (MFD)
algorithms (Erskine et al. 2006; Gruber and Peckham 2008; Wilson et al. 2008). This case
study examines the impacts of flow direction algorithm choice on the CTI and resulting wetland
mapping accuracy.
SFD algorithms route flow from a DEM grid cell to only one adjacent cell. The most promi-
nent examples of SFD algorithms are the Deterministic 8 (D8; O’Callaghan and Mark 1984)
and Rho8 (Fairfield and Leymarie 1991). In contrast, MFD algorithms allow water to flow to
multiple neighboring grid cells. Tested MFDs included DEMON (Costa-Cabral and Burges
1994), Deterministic Infinite (D∞; Tarboton 1997), Mass Flux (MF; Gruber and Peckham
2008), Triangular Multiple Flow (MD∞; Siebert and McGlynn 2007), and Divergent Flow
(FD8; Freeman 1991).
This study was conducted in watersheds in three ecoregions in Minnesota: Northern
Glaciated Plains, Central Hardwood Forest, and Northern Lakes and Forest. The base datasets
for flow direction model and CTI creation were 3 m LiDAR DEMs acquired in 2010 and 2011.
For each area, the LiDAR DEM was processed with each SFD and MFD algorithm. The result-
ing flow direction models were used with slope layers to create CTIs. The CTI layers were then
thresholded to establish two simple classes: upland and wetland. The accuracies of these classes
were assessed with respect to field- and image-based reference data and compared to an existing
National Wetlands Inventory (NWI) map.
The results of this study suggested two conclusions about the importance of flow direction in
wetland mapping. First, the MFD algorithms generally performed better than SFD. The differ-
ences in these classes of algorithm were particularly evident in visual comparison. The MFD-
derived wetland maps appear much smoother, clearer, and more realistic in terms of expected
water flow over the DEM (Figure 9.1). MFD results also had fewer visual artifacts, such as
linear features and speckle. Quantitatively, MDF algorithms were generally better in terms of
wetland mapping accuracy. Though the differences were not large and varied per area, the best-
performing algorithms were D∞, MD∞, and MF.
Second, SFD algorithms may be appropriate when software availability and computation
speed are factors or when visual appearance of the flow direction layer is not a major concern.
The SFD algorithms are implemented in a wider array of commercial and open-source software
packages and generally have lighter computational requirements. MFD algorithms require sig-
nificantly more computing resources and are available in a more limited number of software
packages. Thus, an ease-of-use argument can be made in favor of SFD. If the flow direction
model is intended only as in input to the creation of a derived product such as the CTI, which
is then generalized to a relatively coarse wetland map, then the generally poorer appearance of
SFD layers may not be a limiting factor. However, if a particular use relies on detailed analysis
of water movement across the landscape, or requires the most realistic possible depiction of
flow, then an MFD algorithm is recommended.
Further research is recommended to determine the effects of using CTI layers derived from
various flow direction models in the following areas: (1) in more complex studies, such as those
using OBIA methods, where the visual artifacts resulting from using SFD approaches may
affect segmentation boundaries; (2) when using DEM resolutions other than 3 m, for coarser
or finer DEMs, the flow direction algorithm choice may be more important due to variations in
the flow due to resolution; and (3) in areas where nondepressional wetlands are common, for
example, organic flat wetlands, groundwater discharge-fed wetlands, and some types of fens.
Traditional flow direction models are unlikely to capture the factors leading to wetland forma-
tion in such areas.
(a) (b) (c)
(d) (e) (f )
(g) (h) (i)
FIGURE 9.1 Visual comparison of (a) the NWI polygons, (b) CIR aerial imagery 2011, (c) D8 CTI, (d)
Rho8 CTI, (e) DEMON CTI, (f) FD8 CTI, (g) D∞ CTI, (h) MD∞ CTI, and (i) Mass Flux CTI for the Central
Hardwood Forest study area. Bluer colors represent higher CTI values or more water accumulation. Redder
colors represent lower CTI values or dryer areas. The SFD models contain more pixilation and erroneous
linear features. (With kind permission from Springer Science + Business Media: Rampi et al. 2014a, Figure
9.4; reprinted.)
Case Study 3 Multisource, Multitemporal Mapping of Northern

Midwest Wetlands (Based on Corcoran et al. 2011, 2013)
Selection of input data types and their acquisition times is an important determining factor of the
accuracy of wetland classifications from geospatial data. Capturing diagnostic features in a wet-
land’s hydroperiod, or seasonal changes in water storage capacity (“hydrologic signature”), has
been described as the most important aspect of its biodiversity and phenology (Wissinger 1999).
The hydroperiod of a wetland dictates the optimal acquisition timing of geospatial data for clas-
sification and, thus, indirectly affects data availability. Changes in hydroperiod can change a
wetland’s classification. This case study examined the impacts of input data type and timing on
thematic accuracy of wetland maps derived from high-resolution aerial imagery, optical imag-
ery, SAR imagery, DEM-derived topographic measures, and soil maps.
These studies were conducted in the Northern Lakes and Forest ecoregion in Minnesota. The
input data were incorporated into various RF classification trials to determine which data types
and acquisition times were most influential on the accuracy of wetland maps. Thematic accura-
cies were assessed using both image- and field-based reference data.
Corcoran et al. (2011) reported wetland/nonwetland accuracy of 75% when using a combina-
tion of optical, topographic, and radar data (Figure 9.2). The accuracy decreased to 72% when
Legend
Water
Emergent
wetlands
Forested
wetlands
Scrub/shrub
wetlands
Upland
(a) (b)
Confidence
High
Low
1 N
km
(c) (d)
FIGURE 9.2 Subset area showing the upland/water/wetland determinant classification results from an RF
decision tree classification: (a) using optical and topographical data alone and (b) using optical, topographic,
and radar imagery combined. The original NWI data are shown in panel (c), where land cover classes were
consolidated for ease in comparison. The relative confidence of the decision tree classification using all opti-
cal, topographic, and radar data is shown in panel (d). The red circle indicates an area with substantial dif-
ferences in emergent wetland classification between the datasets. (Reprinted with permission from Corcoran,
J.M. et al., Can. J. Remote Sens., 27(5), 564, 2011.)
radar imagery was excluded. Wetland-type accuracies were significantly lower at 63%. No sig-
nificant accuracy differences due to acquisition timing were identified. Corcoran et al. (2013)
reported Level 1 (upland, water, and wetland) mapping accuracy of 85% when using Landsat-5,
topographic derivatives, PALSAR radar imagery, and soil data. Level 2 (Cowardin wetland
types) accuracy was 69% (Figure 9.3). Wetland class user’s and producer’s accuracies were 93%
and 79%, respectively. Spring optical and radar imagery was consistently more influential in the
RF classification process for mapping Level 1 classes; however, spring imagery was no more
valuable than summer imagery for Level 2 classes.
The results of these studies suggested several conclusions related to input data selection
and timing. First, the most influential variables for Level 1 mapping were the red, near-
infrared, and middle-infrared optical bands, NDVI, elevation, curvature, hydric soils, and
L-band H-V polarization SAR. For Level 2 mapping, additional important variables were
Landsat Band 5, tasseled cap greenness and wetness bands, and L-band HH polarization
SAR. Second, SAR imagery can increase mapping accuracies, but care must be taken to select
imagery that is appropriate for the landscape being mapped. In this study area, PALSAR
L-band imagery resulted in higher accuracies than Radarsat-2 C-band imagery for Level 1
mapping—an expected result for a forested area. However, C-band radar did provide some
benefits in Level 2 mapping. Third, variation in precipitation affected wetland mapping accu-
racy. Imagery acquired in above normal precipitation conditions was more influential than
that acquired after below normal precipitation. Fourth, reducing the number of input variables
92°37'30''W 92°33'0''W 92°28'30''W

Cowardin
classification
TM, aerial, Rsat-2,

PALSAR, Soil
all season, all data
Upland
46°47'30''N
Water
Emergent wetland
Forested wetland
Scrub/shrub wetland
Confidence
Low: 0 High: 1
46°44'0''N
0 1 2 km
46°40'30''N
0 1 2 4 N
km
FIGURE 9.3 Output classification of the most accurate full season RF model for the Level 2 land cover clas-
sification using all available Landsat 5 TM, aerial orthophoto, topographic, RADARSAT-2, PALSAR, and
soil data. (From Corcoran, J.M. et al., Remote Sens., 5(7), 3212, 2013.)
provided to the RF classifier can result in lower computational requirements, with only small
decreases in accuracy. Thus, as long as the most influential data types and timing (e.g., spring
with average or above precipitation) are used, there is no need to include additional input data
in an RF approach.
Future research should examine several aspects of RF-based wetland mapping: (1) the use
of RF confidence maps to select areas for additional training and reference data collection, (2)
including additional LiDAR-derived topographic and vegetation structure metrics (e.g., number
of returns, standard deviation of returns, and vegetation height), (3) including context and tex-
ture in an OBIA approach, and (4) more in-depth study of the benefits of the available bands,
polarizations, and decompositions of radar imagery.
Case Study 4 Mapping Upper Midwest Wetlands

Using OBIA (Based on Rampi et al. 2014a)
Integration of multiple high-resolution data types has the potential to improve the accuracy of
wetland maps. OBIA techniques are well suited for high-resolution mapping efforts because
they incorporate spectral, spatial, textural, and contextual information and human knowledge
to delineate and characterize image segments. This case examines the effectiveness of using
an OBIA approach to integrate two types of high-resolution data, multispectral leaf-off aerial
imagery and LiDAR elevation data, across three Midwest physiographic provinces.
This research was carried out in three study areas located in different ecoregions in
Minnesota: Northern Glaciated Plains, Western Corn Belt Plains, and Northern Lakes and
Forest. Data used in this project included (1) multispectral leaf-off imagery with four spectral
bands (near-infrared, red, green, and blue) with a spatial resolution of 0.5 and acquired in
2009 and 2011; (2) the Green Ratio Vegetation Index (GRVI); (3) LiDAR data collected in
2010 and 2011; and (4) five LiDAR-derived products: 3 m DEM, 1 m digital surface model
(DSM), a 1 m normalized DSM (nDSM), 1 m intensity layer, and 3 m CTI. For each study
area, an OBIA approach was developed through the creation of a customized rule set using
the Cognition Network Language (CNL) in Definiens eCognition Developer version 8.8.0.
eCognition was used to develop the three rule sets with a divide and conquer approach, which
is a multiscale iterative method that partitions image objects based on size, shape, and spec-
tral attributes. Four land cover classes were mapped using the OBIA technique: wetlands,
agriculture, forest, and urban. The accuracies of these classes were evaluated against field-
and image-based reference data. Four standard accuracy estimators were used: overall accu-
racy, producer’s accuracy, user’s accuracy, and the kappa coefficient of agreement (Congalton
and Green 2009). In addition, the wetland class was compared to an existing NWI map in
each study area.
The results of this study suggested several conclusions regarding the importance of integrat-
ing high-resolution imagery and LiDAR data for mapping wetlands using OBIA. First, the com-
bination of high-resolution optical imagery and different LiDAR-derived products including
the CTI significantly improved the accuracy of wetland classification compared to traditional
pixel-based approaches (Corcoran et al. 2011; Knight et al. 2013; Rampi et al. 2014b; Sader
et al. 1995). The overall percent accuracy estimates for the three study areas ranged from 90%
to 93%. The wetland-specific user’s and producer’s accuracies ranged from 91% to 96%. Kappa
coefficients were 0.84–0.90.
Second, while many existing OBIA studies focus primarily on segmentation techniques, this
study focused on developing customized rule sets for multiple areas using a multiscale iterative
OBIA wetland polygons

NWI polygons
FIGURE 9.4 Comparison map of the original NWI polygons and OBIA polygons for a small portion of
the Minnesota River Headwaters with background aerial imagery. The OBIA polygons much more closely
match actual wetland boundaries than does the NWI. (Reprinted with permission from Rampi, L.P. et al.,
Photogram. Eng. Remote Sens., 80(5), 53, 2014a.)
approach that combined LiDAR and imagery layers suitable for discriminating wetlands from
upland classes. This OBIA-based method used information about the input data layers, includ-
ing size, shape, spectral, textural, and contextual information, and further refined the approach
using human knowledge and experience relevant to each area. The customized rule sets devel-
oped in this study may be appropriate for wetlands located in areas with similar characteristics
such as topography, soils, vegetation, and land use.
Finally, the OBIA results were significantly improved over the NWI map (Figure 9.4), espe-
cially in terms of wetland omission errors. This result was expected, given that the NWI is
decades old in parts of the state; however, the methods used in this study would be appropriate
as a base classification for a new version of the NWI. In fact, the state of Minnesota is currently
using such an approach in its ongoing statewide update of the NWI, in which OBIA segmenta-
tion and classification are used to provide a base classification from which image analysts work
to produce a final refined product.
These OBIA results for wetland mapping boundaries are encouraging and useful as an
initial classification of these complex and diverse ecosystems. Further research is needed
to classify the wetlands into types (e.g., Cowardin) using high-resolution data and OBIA
techniques.
DISCUSSION AND CONCLUSION

Wetland classification using geospatial data will likely always be among the more difficult tasks
facing remote sensing mapping professionals. However, significant improvements have been made
as a result of new and emerging techniques such as OBIA and decision tree classifiers. Though
these methods require new ways of thinking about remote sensing–based mapping as opposed to
traditional techniques such as maximum likelihood supervised classification, the extensive research
presented in this chapter provides guidance as to how to work within this new paradigm.
Data selection is of paramount importance to producing a high-quality wetland classification.
The use of high-resolution optical imagery has become much more common as its availability
increases. Though such imagery is undeniably useful, care must be taken to select a spatial reso-
lution that is appropriate for mapping the patch size and shape of the landscape, since these fac-
tors affect the accuracy of image segmentation. In addition, aerial imagery often does not provide
middle-infrared bands that have been shown to be useful for identification of moisture. Thus, the
mapping analyst must balance the desire to use higher-resolution imagery with the disadvantages
of not having middle-infrared and sometimes even near-infrared bands (e.g., many NAIP datasets).
Given the increasing availability of LiDAR datasets, high priority should be assigned to acquiring
such data due to the substantial benefits they offer for characterizing wetlands. LiDAR derivatives
like elevation, curvature, CTI, vegetation height, and intensity provide landscape structural infor-
mation that is difficult or impossible to obtain from optical imagery. If LiDAR data are not avail-
able, coarser elevation-only datasets such as the U.S. NED or Shuttle Radar Topography Mission
may be acceptable substitutes. Elevation derivatives like the CTI provide information about water
flow and storage on the landscape, which is helpful in wetland classification. When using metrics
such as CTI, consideration should be given to the method by which flow is accounted for, with MFD
algorithms providing a more realistic depiction of water movement through the landscape. Like
LiDAR, radar imagery provides structural and also textural information. While radar can improve
the accuracy of wetland classification, especially in inundated areas, the limited availability and
often high cost of radar imagery limit its broad use.
Wetlands are 4D (space and time) processes in 3D landscapes. A wetland’s hydroperiod regulates
not only how it might appear on a single remotely sensed image but also its extent, water depth, veg-
etation composition, biological productivity, and, over time, soil type. Given the significant temporal
variability of many wetlands, data acquisition timing is a critical issue. Ideally many images show-
ing the intra- and interannual variability of a wetland landscape would be used in classification, but
rarely is such a rich dataset available. Thus, the analyst should seek to maximize the extent to which
temporal variability is represented in the data that are available. Diagnostic times, for example, in the
spring when areas are often at their wettest, should be prioritized for data acquisition. A contrasting
dataset, for example, in the summer, may be used to highlight the changes in wetland extent and water
level throughout the year and to identify nonpersistent vegetation (e.g., aquatic beds). It is important
to note the prevailing long-term precipitation regime at the time of data collection, since especially
wet or dry intervals may significantly affect the apparent extent of wetlands on the landscape—a
phenomenon that might not be anticipated based only on relatively short-term weather information.
For classification of wetlands, OBIA and decision trees have emerged as high-performing addi-
tions to the mapping analyst’s toolset. The additional information gained from examining image
segments in addition to individual pixels is powerful. The inclusion of texture, shape, size, and
context variables in wetland classification has been shown to result in significantly higher accura-
cies than without that information. Multiscale segmentation has become the standard in creating
segments for subsequent classification. Viewing a landscape at multiple scales allows for hierarchi-
cal processing of its features. This is especially important given the influence patch size and shape
have on both segmentation and classification. Wetland areas having relatively consistent patch sizes
and shapes are likely to be much easier to classify than those with more complex patch characteris-
tics. A challenge to the broad adoption of OBIA techniques is the subjective nature of segmentation
parameter setting and classification rule set creation. More research is needed to describe optimal
OBIA approaches for different wetland types and input data.
Decision trees have several benefits that make them attractive for wetland classification. They
are able to use continuous and categorical data, do not rely on assumptions about the distributions
of input data, and can be resistant to overfitting. Potentially the biggest advantage of decision trees
in the OBIA wetland mapping context may be that they can help to reduce the subjectivity in both
the segmentation and rule set creation processes. Emerging research suggests that using informa-
tion from OBIA image segments within an RF or other tree-based classifier may be advantageous.
Analyst-driven OBIA segmentation and classification will likely perform better when study area
sizes are conducive to such relatively labor-intensive efforts. However, when study areas are large
or highly variable, making rule set customization onerous, a decision tree classifier used within the
OBIA context may represent a good trade of accuracy for efficiency.
Though the use of OBIA and decision trees has resulted in significant advances in the accuracy
of wetland maps, there remain many issues that require future research. Perhaps most prominent of
these is the lack of guidance, in the relatively nascent literature in this area, for how best to imple-
ment OBIA methods across a wide range of wetland types and using the many different data sources
available. A large experimental study or an integrative analysis of widely varying wetlands and the
data types that are most suitable for mapping them would be a welcome addition to the wetland
mapping literature.
REFERENCES
Antonellini, M., T. Dentinho, A. Khattabi, E. Masson, P.N. Mollema, V. Silva, and P. Silveira. 2014. An inte-
grated methodology to assess future water resources under land use and climate change: An application
to the Tahadart drainage basin (Morocco). Environmental Earth Sciences 71(4):1839–1853.
Baatz, M. and A. Schape. 2000. Multiresolution segmentation : An optimization approach for high quality
multi-scale image segmentation. Journal of Photogrammetry and Remote Sensing 58:12–23.
Baker, C., R. Lawrence, C. Montagne, and D. Patten. 2006. Mapping wetlands and riparian areas using Landsat
ETM+ imagery and decision-tree-based models. Wetlands 26(2):465–474.
Benz, U.C., P. Hofmann, G. Willhauck, I. Lingenfelder, and M. Heynen. 2004. Multi-resolution, object-oriented
fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry and
Remote Sensing 58(3–4):239–258.
Hydrological Sciences Journal 24:43–69.
Blaschke, T. 2003. Object-based contextual image classification built on image segmentation. In Proceedings
of the 2003 IEEE Workshop on Advances in Techniques for Analysis of Remotely Sensed Data, 27–28
October, Washington, DC, CD-ROM.
Blaschke, T. 2010. Object based image analysis for remote sensing. ISPRS Journal of Photogrammetry and
Blaschke, T. and J. Strobl. 2001. What’s wrong with pixels? Some recent developments interfacing remote
sensing and GIS. GIS—Zeitschrift fur Geoinformationssysteme 6:12–17.
Bock, M., P. Xofis, J. Mitchley, G. Rossner, and M. Wissen. 2005. Object-oriented methods for habitat map-
ping at multiple scales—Case studies from Northern Germany and Wye Downs, UK. Journal for Nature
Conservation 13(2–3):75–89.
Bowen, M.W., W.C. Johnson, S.L. Egbert, and S.T. Klopfenstein. 2010. A GIS-based approach to identify and
map playa wetlands on the high plains, Kansas, USA. Wetlands 30:675–684.
Chen, L.-C., T.-A. Teo, C.-H. Hsieh, and J.-Y. Rau. 2006. Reconstruction of building models with curvilinear
boundaries from laser scanner and aerial imagery. In Chang, L.-W., Lie, W.-N., and Chiang, R. (Eds.),
Advances in Image and Video Technology (pp. 24–33). Springer, Berlin, Germany.
Conchedda, G., L. Durieux, and P. Mayaux. 2008. An object-based method for mapping and change analysis in
mangrove ecosystems. ISPRS Journal of Photogrammetry and Remote Sensing 63(5):578–589.
Congalton, R.G. and K. Green. 2009. Assessing the Accuracy of Remotely Sensed Data: Principles and
Practices, 2nd edn. CRC Press/Taylor & Francis, Boca Raton, FL.
Corcoran, J.M., J.F. Knight, B. Brisco, S. Kaya, A. Cull, and K. Murhnaghan. 2011. The integration of optical,
topographic, and radar data for wetland mapping in northern Minnesota. Canadian Journal of Remote
Sensing 27(5):564–582.
Corcoran, J.M., J.F. Knight, and A.L. Gallant. 2013. Influence of multi-source and multi-temporal remotely
sensed and ancillary data on the accuracy of random forest classification of wetlands in northern
Minnesota. Remote Sensing 5(7):3212–3238.
Costa, M.P.F., O. Niemann, E. Novo, and F. Ahern. 2002. Biophysical properties and mapping of aquatic veg-
etation during the hydrological cycle of the Amazon floodplain using JERS-1 and Radarsat. International
Journal of Remote Sensing 23:1401–26.
Costa-Cabral, M. and S.J. Burges. 1994. Digital elevation model networks (DEMON): A model of flow
over hillslopes for computation of contributing and dispersal areas. Water Resources Research
30:1681–1692.
Cowardin, L.M. and V.I. Myers. 1974. Remote sensing for identification and classification of wetland vegeta-
tion. The Journal of Wildlife Management 38(2):308–314.
Dissanska, M., M. Bernier, and S. Payette. 2009. Object-based classification of very high resolution panchro-
matic images for evaluating recent change in the structure of patterned peatlands. Canadian Journal of
Dronova, I., P. Gong, N.E. Clinton, L. Wang, W. Fu, S. Qi, and Y. Liu. 2012. Landscape analysis of wetland
plant functional types: The effects of image segmentation scale, vegetation classes and classification
methods. Remote Sensing of the Environment 127:357–369.
Dronova, I., P. Gong, and L. Wang. 2011. Object-based analysis and change detection of major wetland cover
types and their classification uncertainty during the low water period at Poyang Lake, China. Remote
Sensing of Environment 115(12):3220–3236.
Erskine, R.H., T.R. Green, J.A. Ramirez, and L.H. MacDonald. 2006. Comparison of grid-based algorithms for
computing upslope contributing area. Water Resources Research 42:W09416.
Fairﬁeld, J. and P. Leymarie. 1991. Drainage networks from grid digital elevation models. Water Resources
Research 27:709–717.
Fournier, R.A., M. Grenier, A. Lavoie, and R. Helie. 2007. Towards a strategy to implement the Canadian
Wetland Inventory using satellite remote sensing. Canadian Journal of Remote Sensing 33(S1):S1–S16.
Frauman, E. and E. Wolff. 2005. Segmentation of very high spatial resolution satellite images in urban areas
for segments-based classification. In ISPRS WG VII/1 “Human Settlements and Impact Analysis” Third
International Symposium Remote Sensing and Data Fusion Over Urban Areas (URBAN 2005) and Fifth
International Symposium Remote Sensing of Urban Areas (URS 2005), Tempe, AZ.
Freeman, T.G. 1991. Calculating catchment area with divergent flow based on a regular grid. Computers and
Geosciences 17(3):413–422.
Friedl, M.A. and C.E. Brodley. 1997. Decision tree classification of land cover from remotely sensed data.
Frohn, R.C., M. Reif, C.R. Lane, and B. Autrey. 2009. Satellite remote sensing of isolated wetlands using
object-oriented classification of Landsat 7 data. Wetlands 29(3):931–941.
Gilmore, M.S., E.H. Wilson, N. Barrett, D.L. Civco, S. Pirsloe, J.D. Hurd, and C. Chadwick. 2008. Integrating
multi-temporal spectral and structural information to map wetland vegetation in a lower Connecticut
River tidal marsh. Remote Sensing of the Environment 112(11):4048–4060.
Goetz, S.J., R.K. Wright, A.J. Smith, E. Zinecker, and E. Schaub. 2003. IKONOS imagery for resource manage-
ment: Tree cover, impervious surfaces, and riparian buffer analyses in the mid-Atlantic region. Remote
Sensing of the Environment 88(1–2):195–208.
Grenier, M., A.-M. Demers, S. Labrecque, M. Benoit, R.A. Fournier, and B. Drolet, 2007. An object-based
Canada. Canadian Journal of Remote Sensing 33(S1):S28–S45.
Grenier, M., S. Labrecque, M. Garneau, and A. Tremblay. 2008. Object-based classification of a SPOT-4 image
for mapping wetlands in the context of greenhouse gases emissions: The case of the Eastmain region,
Québec, Canada. Canadian Journal of Remote Sensing 34(S2):S398–S413.
Gruber, S. and S. Peckham. 2008. Land-surface parameters and objects in hydrology. In Hengl, T. and
Reuter, H.I. (Eds.), Geomorphometry: Concepts, Software, Applications (pp. 171–194). Elsevier,
Amsterdam, the Netherlands.
Halabisky, M., L.M. Moskal, and S.A. Hall. 2011. Object-based classification of semi-arid wetlands. Journal
of Applied Remote Sensing 5:053511-1–053511-13.
Haralick, R.M. and Shapiro, L.G. 1985. Image segmentation techniques. Computer Vision, Graphics, and
Image Processing 29(1):100–132.
Harayama, A. and J.M. Jaquet. 2004. Multi-source object-oriented classification of landcover using very high
resolution imagery and digital elevation model. In Enviroinfo, Geneva, Switzerland.
Harken, J. and R. Sugumaran, 2005. Classification of Iowa wetlands using an airborne hyperspectral image: A
comparison of the spectral angle mapper classifier and an object-oriented approach. Canadian Journal
Hess, L.L., J.M. Melack, E.M.L.M. Novo, C.C.F. Barbosa, and M. Gastil. 2003. Dual-season mapping of
wetland inundation and vegetation for the central Amazon basin. Remote Sensing of the Environment
87:404–428.
Hogg, A.R. and K.W. Todd, 2007. Automated discrimination of upland and wetland using terrain derivatives.
Canadian Journal of Remote Sensing, 33(1):68–83.
Homer, C., J. Dewitz, J. Fry, M. Coan, N. Hossain, and C. Larson. 2007. Completion of the 2001 national land
cover database for the conterminous United States, Photogrammetric Engineering and Remote Sensing
73(4):337–341.
Kamal, M. and S. Phinn. 2011. Hyperspectral data for mangrove species mapping: A comparison of pixel-based
and object-based approach. Remote Sensing 3(10):2222–2242.
Kamal, M., S. Phinn, and K. Johansen. 2014. Characterizing the spatial structure of mangrove features for
optimizing image-based mangrove mapping. Remote Sensing 6(2):984–1006.
Kettig, R.L. and D.A. Landgrebe. 1976. Classification of multispectral image data by extraction and classifica-
tion of homogeneous objects. IEEE Transactions on Geoscience Electronics GE-14(1):19–26.
Kim, M., M. Madden, and T.A. Warner. 2009. Forest type mapping using object-specific texture measures from
multispectral Ikonos imagery: Segmentation quality and image classification issues. Photogrammetric
Kim, M., T.A. Warner, M. Madden, and D.S. Atkinson. 2011. Multi-scale GEOBIA with very high spatial reso-
lution digital aerial imagery: Scale, texture and image objects. International Journal of Remote Sensing
32(10):2825–2850.
Knight, J.F., B.P. Tolcser, J.M. Corcoran, and L.P. Rampi. 2013. The effects of data selection and thematic
detail on the accuracy of high spatial resolution wetland classifications. Photogrammetric Engineering
Kressler, F. and K. Steinnocher. 2008. Object-oriented analysis of image and LiDAR data and its potential
for a dasymetric mapping application. In Blaschke, T., Lang, S., and Hay, G. J. (Eds.), Object-Based
Image Analysis—Spatial Concepts for Knowledge-Driven Remote Sensing Applications (pp. 611–624).
Springer, Berlin, Germany.
Laba, M., B. Blair, R. Downs, B. Monger, W. Philpot, S. Smith, P. Sullivan, and P.C. Baveye. 2010. Use of
textural measurements to map invasive wetland plants in the Hudson River National Estuarine Research
Reserve with IKONOS satellite imagery. Remote Sensing of the Environment 114(4):876–886.
Laba, M., R. Downs, S. Smith, S. Welsh, C. Neider, S. White, M. Richmond, W. Philpot, and P. Baveye.
2008. Mapping invasive wetland plants in the Hudson River National Estuarine Research Reserve using
QuickBird satellite imagery. Remote Sensing of the Environment 112:286–300.
Lang, M.W. and G.W. McCarty. 2009. Lidar intensity for improved detection of inundation below the forest
canopy. Wetlands 29(4):1166–1178.
Lang, M.W., G.W. McCarty, R. Oesterling, and I. Yeo. 2013. Topographic metrics for improved mapping of
forested wetlands. Wetlands 33:141–155.
Lang, M.W., O. McDonough, G.W. McCarty, R. Oesterling, and B. Wilen. 2012. Enhanced detection of
wetland-stream connectivity using LiDAR. Wetlands 32:461–473.
Li, J. and W. Chen. 2005. A rule-based method for mapping Canada’s wetlands using optical, radar and DEM
data. International Journal of Remote Sensing 26(22):5051–5069.
Lunetta, R. and M. Balogh. 1999. Application of multi-temporal Landsat-5 TM imagery for wetland identifica-
tion. Photogrammetric Engineering and Remote Sensing 65(12):1303–1310.
MacFaden, S. W., J.P. O’Neil-Dunne, A.R. Royar, J.W. Lu, and A.G. Rundle. 2012. High-resolution tree canopy
mapping for New York City using LIDAR and object-based image analysis. Journal of Applied Remote
Sensing 6(1):063567-1.
Maxa, M. and P. Bolstad. 2009. Mapping northern wetlands with high resolution satellite imagers and lidar.
Wetlands 29(1):248–260.
Meng, Y. and S. Peng. 2009. Object-oriented building extraction from high-resolution imagery based on
fuzzy SVM. In International Conference on Information Engineering and Computer Science. IEEE,
2009, Kyiv, Ukraine.
Moffett, K.B. and S.M. Gorelick. 2013. Distinguishing wetland vegetation and channel features with object-
based image segmentation. International Journal of Remote Sensing 34(4):1332–1354.
Nobrega, R.A.A., C.G. O’Hara, and J. A. Quintanilha. 2008. An object-based approach to detect road features
for informal settlements near Sao Paulo, Brazil. In Blaschke, T., Lang, S., and Hay, G.J. (Eds.), Object-
Based Image Analysis (pp. 589–607). Springer, Berlin, Germany.
O’Callaghan, J.F. and D.M. Mark. 1984. The extraction of drainage networks from digital elevation data.
Computer Vision, Graphic and Image Processing 28:328–344
Olson, C.E. 1960. Elements of photographic interpretation common to several sensors. Photogrammetric
Engineering 26(4):651–656.
O’Neil-Dunne, J.P.M., S.W. MacFaden, A.R. Royar, and K.C. Pelletier. 2012. An object-based system for
LiDAR data fusion and feature extraction. Geocarto International 28(3):1–16.
Ouyang, Z.-T., M.-Q. Zhang, X. Xie, Q. Shen, H.-Q. Guo, and B. Zhao. 2011. A comparison of pixel-based
and object-oriented approaches to VHR imagery for mapping saltmarsh plants. Ecological Informatics
6(2):136–146.
10(5):381–402.
Quinlan, J.R.1993. C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers, San Mateo, CA.
Rampi, L.P., J.F. Knight, and C.F. Lenhart. February 2014a. Comparison of flow direction algorithms in the
application of the CTI for mapping wetlands in Minnesota. Wetlands 34(3):513–525.
Rampi, L.P., J.F. Knight, and K.C. Pelletier. 2014b. Wetland mapping in the Upper Midwest United States: An
object-based approach integrating Lidar and imagery data. Photogrammetric Engineering and Remote
Sensing 80(5):53–62.
Repaka, S.R. and D.D. Truax. 2004. Comparing spectral and object based approaches for classification and
transportation feature extraction from high resolution multispectral imagery. In Proceedings of the
ASPRS 2004 Annual Conference, Denver, CO.
Rodriguez-Galiano, V.F., B. Ghimire, J. Rogan, M. Chica-Olmo, and J.P. Rigol-Sanchez. 2012. An assess-
ment of the effectiveness of a random forest classifier for land-cover classification. ISPRS Journal of
Photogrammetry and Remote Sensing 67:93–104.
Rover, J., C.K. Write, N.H. Euliss, D.M. Mushet, and B.K. Wylie. 2011. Classifying the hydrologic function of
prairie potholes with remote sensing and GIS. Wetlands 31:319–332.
Sader, S.A., D. Ahl, and W.S. Liou. 1995. Accuracy of Landsat- TM and GIS rule-based methods for forest
wetland classification in Maine. Remote Sensing of the Environment 53(3):133–144.
Seibert, J. and B. McGlynn. 2007. A new triangular multiple flow direction algorithm for computing upslope
areas from gridded digital elevation models. Water Resources Research 43:1–8.
Shen, G., H. Guo, and J. Liao. 2008. Object oriented method for detection of inundation extent using multi-
polarized synthetic aperture radar image. Journal of Applied Remote Sensing 2(1):023512.
Smith, A. 2010. Image segmentation scale parameter optimization and land cover classification using the
Random Forest algorithm. Journal of Spatial Science 55(1):69–79.
Syed, S., P. Dare, and J. Simon. 2005. Automatic classification of land cover features with high resolution
imagery and lidar data: An object-oriented approach. In Proceedings of SSC2005 Spatial Intelligence,
Innovation and Praxis: The National Biennial Conference of the Spatial Sciences Institute, pp. 512–522,
Melbourne, Australia.
Tarboton, D.G. 1997. A new method for the determination of flow directions and upslope areas in grid digital
elevation models. Water Resources Research 33:309–319.
Tiner, R.W. 1990. Use of high-altitude aerial photography for inventorying forested wetlands in the United
States. Forest Ecology and Management 33–34:593–604.
Van, T.T. and H.D.X. Bao. 2010. Study of the impact of urban development on surface temperature using
remote sensing in Ho Chi Minh City, Northern Vietnam. Geographical Research 48(1):86–96.
Wang, L., W.P. Sousa, and P. Gong. 2004a. Integration of object-based and pixel-based classification for map-
ping mangroves with IKONOS imagery. International Journal of Remote Sensing 25(24):5655–5668.
Wang, L., W.P. Sousa, P. Gong, and G.S. Biging. 2004b. Comparison of IKONOS and QuickBird images
for mapping mangrove species on the Caribbean coast of Panama. Remote Sensing of the Environment
91(3–4):432–440.
Wilson, J.P., G. Aggett, Y.X Deng, and C.S. Lam. 2008. Water in the landscape: A review of contemporary
flow routing algorithms. In Zhou, Q., Lees, B., and Tang, G. (Eds.), Advances in Digital Terrain Analysis
(pp. 213–236). Springer, Berlin, Germany.
Wissinger, S.A. 1999. Ecology of wetland invertebrates. In Batzer, D.P., Rader, R.B., and Wissinger, S.A.
(Eds.), Invertebrates in Freshwater Wetlands of North America: Ecology and Management (pp. 1043–
1086). Wiley, New York.
Witharana, C., D.L. Civco, and T.H. Meyer. 2014. Evaluation of data fusion and image segmentation in earth
observation based rapid mapping workflows. ISPRS Journal of Photogrammetry and Remote Sensing
87:1–18.
Wright, C. and A. Gallant. 2007. Improved wetland remote sensing in Yellowstone National Park using
classification trees to combine TM imagery and ancillary environmental data. Remote Sensing of the
Zhou, W. and A. Troy. 2008. An object-oriented approach for analysing and characterizing urban landscape at
the parcel level. International Journal of Remote Sensing 29(11):3119–3135.
10 Unmanned Aerial Systems
and Structure from
Motion Revolutionize
Wetlands Mapping
Marguerite Madden, Thomas Jordan, Sergio Bernardes,
David L. Cotten, Nancy O’Hare, and Alessandro Pasqua
CONTENTS
Introduction..................................................................................................................................... 195
Background on Existing Imagery for Wetlands Mapping............................................................... 196
Historical Aerial Photographs Opportunistically Include Wetlands.......................................... 196
U.S. National Programs Provide Aerial Photographs and Digital Imagery for
National Wetlands Inventory and Other Wetlands Mapping...................................................... 197
Medium-Resolution Satellite Images, High-Resolution Commercial Satellite Imagery,
and Airborne Multispectral/Hyperspectral Imagery Are Used for Wetlands Mapping.............. 198
Structural Information from LiDAR: Incorporating LiDAR Technology into
Wetlands Mapping..................................................................................................................... 199
UAS for Flexible Data Acquisition................................................................................................. 201
Capabilities of UAS for Ground Truthing and Bird’s-Eye View of Wetlands............................ 201
UAS Data Acquisition and Management...................................................................................204
UAS Image Processing to Create Orthoimages and 3D Point Clouds............................................206
Structure-from-Motion Photogrammetry and Multiple Image Matching..................................209
Considerations of Federal and State Regulations for Use of UAS in the United States................. 211
Future Directions for UAS and Wetlands Mapping........................................................................ 213
References....................................................................................................................................... 214
INTRODUCTION
Emerging geospatial technologies such as unmanned aerial systems (UAS*) and light detection and
ranging (LiDAR) coupled with structure-from-motion (SfM) photogrammetric processing present
both opportunities and challenges for wetlands mapping. In spite of the plethora of aerial and sat-
ellite imagery that are readily available at a variety of scales and formats for wetlands mapping,
the potential for UAS and other technologies to revolutionize mapping is gaining recognition from
resource managers, scientists, surveyors, engineers, foresters, farmers, private practitioners, and
policy makers (Anderson and Gaston 2013; Colomina and Molina 2014). This chapter explores these
emerging and revolutionizing technologies.
* For the purpose of this chapter, UAS, unmanned aerial vehicle (UAV), and remotely piloted aircraft (RPA) are considered
synonyms for a range of systems of unmanned aircraft, navigation system, ground-control system, sensor, first-person
view capability, and/or ground-based pilot.
195
As presented in the first section, the fundamentals of the various optical (i.e., visible and near-
infrared [NIR] spectral) or structural (e.g., LiDAR) sensors for wetland issues apply regardless
of whether the platform for the sensor is satellite, manned aircraft, or UAS. The second section
introduces the UAS platform for research and monitoring, a tool that has only recently become
widely used, in part due to technological advances. Then, we explore recent progress in creating
point clouds of location, elevation, and color data from a series of overlapping images captured
with off-the-shelf UAS and SfM photogrammetric software. These techniques offer one of the
most promising avenues for low-cost, but high-quality, 3D data acquisition. Finally, in several
countries, commercial and research applications of UAS technology are limited by strict rules on
the use of aircraft and sensors or, conversely, by the lack of well-defined policies and regulations.
We focus on the United States, where federal and state regulations currently are being enacted
affecting widespread application of UAS. Regardless, UAS for wetlands, and other resource man-
agement issues, is poised to explode over the next few years, becoming an estimated U.S. $13.6
billion industry in the United States alone, primarily from agricultural applications (Jenkins
and Vasigh 2013). Fine detail and increased accuracy of the bare earth surface, vegetation struc-
ture, and hydrology afforded by UAS, LiDAR, and SfM will enhance existing wetlands map-
ping methods and open up new possibilities for high-resolution 3D change detection in wetlands
imaged over time.
BACKGROUND ON EXISTING IMAGERY FOR WETLANDS MAPPING

Wetlands are difficult features to map from the ground (see Chapter 3). Their boundaries appear
to change with the seasons and the tides, upland to wetland gradients are often gradual, sizes vary
widely, and the diversity of wetland species is vast. Anyone who has ever attempted to walk around
a wetland to map its perimeter realizes the magnitude of this challenge due to thick and tangled
vegetation, saturated and unstable substrates, deep creeks, and difficulty in identifying the wetland
edge from a ground perspective. Early on, wetland mappers realized the requirement for an aerial
viewpoint, preferably in three dimensions and coupled with field experience to fully understand the
character and spatial extent of a wetland.
Historical Aerial Photographs Opportunistically Include Wetlands

Aerial views of wetlands have been photographically captured since the mid-1800s. In 1862, two
years after Gaspard-Félix Tournachon (Nadar) photographed Paris from a balloon, James Wallace
Black acquired aerial photographs of the city of Boston, including wetlands (Hannavy 2008). It is
possible that soldiers from the U.S. Union Army Balloon Corps hovering over battlefields during
the Civil War photographed wetlands during their reconnaissance missions. The same can be said
about photos snapped from an airplane for the first time in 1908, by a photographer accompanying
Wilber Wright or by small cameras strapped to pigeons and sent behind World War I enemy lines
(Colwell 1997; Lillesand et al. 2008). Although wetlands appeared in early photographs taken by
cameras lifted into the sky by balloons, pigeons, kites, and airplanes, they were merely accidental
components of a landscape being imaged for novel aerial views, meteorological data, and enemy
reconnaissance. The first comprehensive program acquiring aerial photographs for the United
States that could be used to map wetlands was conducted by the U.S. Department of Agriculture’s
Agricultural Stabilization and Conservation Service (USDA-ASCS) beginning in 1937 and con-
tinuing to the early 1940s (Lillesand et al. 2008). Although the 1:20,000 scale, black-and-white
(BW) stereo photographs captured wetlands to some extent, they were largely ignored at the time
because (1) the primary objective of the mission was to provide “a more accurate, inexpensive, and
efficient method” of measuring farmland acreage than laying out surveying chains and drawing
maps of fields by hand and (2) wetlands were historically considered wastelands that bred diseases,
hampered travel, and prevented farming, deserving only to be reclaimed by filling or draining
Unmanned Aerial Systems and Structure from Motion Revolutionize Wetlands Mapping 197
(Dahl and Allord 1996; USDA 2014a). The value of wetlands began to be realized in the 1970s with
the first understanding of the ecological importance of estuarine and freshwater marshes in terms
of biodiversity, productivity, and habitat for important commercial and sport fisheries (Odum 1971).
Coupled with an increased awareness of cumulative wetland losses and threats to water quality,
scientists, resource managers, and policy makers recognized the importance of wetlands mapping
and inventory from the vantage of cameras in the sky (Frayer 1983; Tiner 1984).
For nearly 50 years, individuals tasked with detecting and mapping wetlands in the United
States have relied on aerial imagery acquired from cameras attached to or secured within aircraft
and pointed vertically downward to target coastal marshes, inland freshwater marshes, scrub–
shrub wetlands, and wooded swamps. Early investigations of the utility of aerial photographs for
mapping wetlands evaluated film types and optimal flying heights resulting in best image scales
for detection and delineation of aquatic vegetation by manual interpretation. Color aerial photo-
graphs, for example, were found to be superior to panchromatic photographs for water resource
studies in the Florida Everglades, and color film permitted penetration through the water col-
umn for discrimination of both aquatic and terrestrial vegetation (Schneider 1966, 1968). The
National Wetlands Inventory (NWI), established by the U.S. Fish and Wildlife Service (USFWS)
in 1974, used BW and color infrared (CIR) stereo aerial photographs flown by the USFWS, U.S.
Geological Survey (USGS), and U.S. Forest Service (USFS) to conduct the first national effort to
produce wetlands maps (Tiner 1997). Federal Wetlands Mapping Standards, field verification, and
a common classification system designed to fully describe wetlands in marine, estuarine, river-
ine, lacustrine, and palustrine systems—the National Wetlands Classification Standard—ensured
high-quality, accurate map products suitable for assessing national losses, gains, and distributions
of wetlands for conservation efforts (Cowardin et al. 1979; Dahl et al. 1991; Wilen and Bates 1995;
Dahl 2000; Tiner 2009).
U.S. National Programs Provide Aerial Photographs and Digital Imagery

for National Wetlands Inventory and Other Wetlands Mapping
Several national programs for acquiring aerial photograph have provided imagery for the NWI since
1980. Implemented by the USGS and the USDA, the high-resolution film and paper photographs,
and more recently digital orthoimages, have been the fundamental data source for mapping topog-
raphy, hydrography, infrastructure, and wetlands of the conterminous United States at 1:24,000
scale, based on the North American Datum of 1983 (NAD83) and tiled according to 7.5 min USGS
quadrangles. Limitations for wetlands mapping, however, have resulted from these programs being
multipurpose and, in some cases, contributed state funds dictated specifications of film type, scale,
and season of acquisition for optimal topographic, infrastructure, or forest/agriculture mapping. For
example, aerial photographs were largely acquired during leaf-off conditions by the USGS National
High Altitude Aerial Photography (NHAP) program from 1980 to 1989 (1:80,000 scale BW and
1:58,000 CIR) and the National Aerial Photography Program (NAPP) (1:40,000 scale BW and CIR)
in the1990s, to facilitate clear views of the earth surface for topographic mapping and the creation
of digital elevation models (DEMs) (USGS 1996, 2014c,d,e).
The responsibility for national aerial photographs shifted to the USDA Farm Service Agency
(FSA) Aerial Photography Field Office (APFO) in the early 2000s. This National Agricultural
Imagery Program (NAIP) resulted in first film and then digital aerial photographs acquired during
the growing season in leaf-on conditions for largely agricultural monitoring applications (USDA
2014a). Around the same time, scanned imageries from the NAPP were orthorectified by the USGS
to create 7.5 min digital orthophoto quadrangles (DOQs) and digital orthophoto quarter quadran-
gles (DOQQs), forming a 1 m resolution image base specified to meet National Map Accuracy
Standards (NMAS) for 1:12,000 scale products (USGS 2014a). Greenfeld (2001) found the posi-
tional accuracy of a representative DOQ to be ±7.6 m at the 95% confidence. This level of positional
accuracy is suitable for mapping wetlands directly from DOQs/DOQQs and for use as a control
base for rectifying other digital images. As digital NAIP images are acquired, they have been recti-
fied to a root-mean-squared error (RMSE, 68% confidence level) horizontal accuracy of ±5 m of
reference DOQQs (USDA 2014b). These images are tiled relative to 3.75 by 3.75 min USGS quarter
quadrangles and formatted to the UTM coordinate system based on NAD83. In order to promote
the use of these images for agricultural applications, NAIP products are generally available as
color mosaics subset by county at 1 m pixel resolution. Although wetland vegetation can be identi-
fied in the color imagery, large-scale CIR film on the order of 1:8,000–1:20,000 scales or digital
images with pixel sizes <1 m are often desired to detect inland freshwater wetlands <0.5 ha in size,
distinguish wetland vegetation species, interpret/classify moist substrates, and accurately delineate
upland–wetland boundaries (Bogucki and Gruendling 1978; Gruendling et al. 1980; Welch et al.
1988, 1992, 1995, 1999; Remillard and Welch 1992, 1993; Teng 1997; Tiner 1997; Madden et al.
1999; Hirano et al. 2003; Madden 2004).
Medium-Resolution Satellite Images, High-Resolution Commercial Satellite Imagery,

and Airborne Multispectral/Hyperspectral Imagery Are Used for Wetlands Mapping
The past four decades have been particularly data rich for wetlands mapping in the United States
given the aforementioned national aerial photography programs, combined with (1) synoptic, sys-
tematic, and repetitive satellite coverage of medium-resolution (i.e., 30 m pixel sizes) multispectral
images such as imagery captured by the U.S. Landsat program since 1982 and (2) targeted high-
resolution (on the order of <5 m pixel size) multispectral digital images acquired by commercial
satellites since 1999 (DigitalGlobe 2014b; USGS 2014b). These images have been extensively used
with automated classification techniques to map wetlands and are especially suitable for mapping
expansive coastal marshes and large wetland areas such as the Everglades and Chesapeake Bay
(Klemas et al. 1975; Carter et al. 1979; Butera 1983; Ackleson and Klemas 1987; Jensen et al. 1995;
Baker et al. 2006; Klemas 2011, 2013). Nevertheless, wetland mappers are often constrained by
the scale and/or the timing of these images. Synoptic satellite imaging programs such as Landsat
are designed to have sun-synchronous orbits such that sensors capture images of a given location
at the same time of day on each 16-day orbital revisit, crossing the equator on the north-to-south
portion of each orbit at the same local sun time of 9:45 a.m. for Landsat-4, -5, and -7 and 10:00
a.m. for Landsat-8 (Jensen 2007; Lillesand et al. 2008). The multispectral imagery covers a broad
area of 185 × 185 km per scene at a general spatial resolution of 80 m for Landsat-1, -2, and -3
multispectral sensor (MSS) imagery, 30 m pixel size for optical bands of Landsat-5 Thematic
Mapper (TM) and Landsat-7 Enhanced Thematic Mapper Plus (ETM+), and 30 m pixel size for
Operational Land Imager (OLI) multispectral bands and 15 m resolution OLI panchromatic band
(USGS 2014b). The fixed orbital parameters of Landsat provide a consistent dataset that docu-
ments a 40-year history of global land cover. This entire data archive has been open for free access
since December 2008.
Recognizing the need for higher spatial resolution satellite imagery for applications requir-
ing the detection of small features, following 1999, commercial satellite companies in the United
States, such as ORBIMAGE, GeoEye, Space Imaging, and DigitalGlobe, began acquiring pan-
chromatic images at 0.41–1 m spatial resolution and multispectral images at 1.64–4 m for smaller
footprint areas (image swaths on the order of 18 km) and on a tasked basis for targeted loca-
tions (Petrie and Stoney 2009; DigitalGlobe 2014a). Although offering stereo imagery, high spa-
tial resolution, and the ability to task the satellite to acquire images of a particular location at a
particular time via pointable sensors with 1–2-day revisit times, the spaceborne platform orbits
at altitudes of 450–770 km (i.e., IKONOS at 680 km, QuickBird at 482–450 km, WorldView-1 at
496 km, WorldView-2 at 770 km, and WorldView-3, which was launched on August 13, 2014, at
617 km allowing imagery of 31 cm panchromatic and 1.24 m multispectral spatial resolution to
be acquired) (DigitalGlobe 2014b). Images acquired from space require atmospheric corrections
and often include clouds, haze, and smoke that reduce contrast and/or block features of interest.
High spatial resolution images available from commercial satellites are well suited for wetlands
mapping and especially for updating existing wetlands maps (Dahl et al. 2009; FGDC 2009; Maxa
and Bolstad 2009). Limitations of high-resolution commercial satellite data for wetlands mapping
include the time that is required to order/schedule the data acquisition in cloud-free conditions,
availability of archived data, greater expense for tasked image requests, and need to purchase
imagery under license agreements.
Aerial photographs from digital cameras and imagery from airborne multispectral and hyper-
spectral sensors that are contracted on a project basis from photogrammetric firms offer greater
flexibility in flying heights (i.e., potential higher spatial resolutions and larger scales) and timing for
data acquisition in particular seasons and days or times of the day (Jensen et al. 1984; Christensen
et al. 1988; Remillard and Welch 1992; Welch et al. 1992; Hirano et al. 2003). Drawbacks again are
the expense of customized flights of airborne image data, the need to wait for ideal weather condi-
tions and minimum flying heights for aircraft of approximately 300 m for fixed-wing airplanes and
150 m for helicopters (Welch and Thomas 1985; Welch et al. 1999). There is tremendous potential
for UAS technology to overcome these limitations and provide a relatively low-cost, versatile, and
assessable tool for image acquisition at high spatial and temporal resolutions. Typically equipped
with multispectral video cameras, future payloads will include miniaturized LiDAR sensors capa-
ble of emitting light energy and detecting partial returns of signals as points as they are reflected
from canopies and rooftops (first return), the bare earth surface (last return), and one-to-two points
in between (intermediate returns) (Renslow 2012). Even without UAS-mounted LiDAR, overlap-
ping frames can be extracted from UAS video and input to SfM photogrammetric software to
generate accurate and dense 3D point clouds approaching the quality of point clouds created by
airborne and ground-based LiDAR. The reconstruction of 3D landscapes and vegetation structure
from LiDAR and SfM photogrammetry is expected to greatly enhance traditional wetlands map-
ping techniques by providing detailed visualizations related to hydrologic flow, soil moisture, and
vegetation types/species.
Structural Information from LiDAR: Incorporating

LiDAR Technology into Wetlands Mapping
At this writing, most LiDAR sensors employed to characterize topography, bathymetry, and wet-
lands vegetation are mounted primarily on manned aircraft. It is anticipated, however, that due
to rapid technological changes, LiDAR sensor deployment will become more common on UAS.
Multiple LiDAR sensors are currently available for UAS platforms (Colomina and Molina 2014),
although sensor size, weight, and energy source for the LiDAR scanner continue to present chal-
lenges for mounting these sensors on smaller airframes (Yang et al. 2011). Regardless of platform,
many of the fundamentals of LiDAR for wetland issues apply regardless of whether the platform for
the sensor is a manned or an unmanned aircraft.
The 3D characteristics of the ground surface and overlying vegetation are critical in determining
not only wetland extent but also function and ecosystem services (Mitsch and Gosselink 1986). The
USGS National Elevation Dataset (NED) has a nominal vertical accuracy (RMSE) of 2.44 m and
horizontal accuracy of 10–30 m (Gesch 2007), making elevation from NED sufficient for broad-
scale mapping of large wetlands. However, detection and delineation of wetland extent for a regula-
tory framework and detection of smaller wetland areas or those with subtle elevation differences
from adjacent upland areas requires finer-scale resolution (Hogg and Holland 2008; Mahaney and
Klemens 2008; Guzha and Shukla 2012). Photointerpretation has traditionally been used to map
wetlands, yet LiDAR had higher accuracy (84%) compared to CIR images (76%) (Hogg and Holland
2008). More importantly, the variety of economic benefits derived from improved topographic maps
from LiDAR data exceeds data acquisition costs (Snyder and Lang 2012).
LiDAR has a vertical accuracy of ~15–30 cm and thereby can generate higher resolution DEMs
than NED (Bater and Coops 2009; Coveney 2013). Because of this vertical accuracy, the U.S. Army
Corps of Engineers (USACE) began to acquire airborne LiDAR data in the mid-1990s to pro-
duce high-resolution DEMs for many coastal areas of the United States (Irish and Lillycrop 1999).
Subsequently, several states acquired LiDAR data to produce fine-scale DEMs that would be use-
ful to effectively map water features not captured in NED, NWI, or USGS National Land Cover
Dataset (NLCD) maps, including vernal (seasonal) ponds (Burne and Lathrop 2008; Paul et al.
2012) and smaller streams (Lang et al. 2012; Biron et al. 2013). In 2012, the USGS implemented
the recommendations of the National Elevation Enhancement Assessment (NEEA) (Snyder et al.
2014) as part of the 3D Elevation Program (3DEP) to acquire LiDAR across the United States to
improve topographic maps. As of early 2014, publicly available LiDAR data exist for 28% of the
conterminous United States and Hawaii. However, coverage by state varies widely. Thirteen states
have complete coverage (Figure 10.1). Vertical accuracy also varies, but far exceeds resolution of
previous topographic data.
Vegetation coverage is the greatest source of error in LiDAR-derived DEMs, followed by slope
(Su and Bork 2006; Schmid et al. 2011) and surface granularity (Coveney 2013). Vegetation-
associated error derives both from lower penetration to ground when pulses are intercepted by veg-
etation and incorrect classification of low vegetation as ground surface. The size of the feature also
matters. Small, shallow depressions in LiDAR-derived DEMs are more likely artifacts (Zandbergen
2010), which may stem from interpolation methods (Toyra et al. 2003; Li et al. 2011). Regardless,
Paul et al. (2012) developed a workflow to detect small (~0.37 ha) depressional wetlands using local-
ized changes in concavity with ~15% commission and ~5% omission rates.
Commonly, LiDAR sensors simultaneously collect intensity of return. While intensity values
can be difficult to standardize, LiDAR intensity can improve point classification of ground returns
(Julian et al. 2009; Stevens and Wolfe 2012) and detection of inundated areas compared to false
CIR (>96% accuracy for LiDAR versus 70% for CIR) (Lang and McCarty 2009). Potential errors in
Vertical accuracy (cm)

15.0 or less
15.1–50.0
50.1–100.0
>100
Not provided
Variable N
FIGURE 10.1 LiDAR availability by state as of August 2013. There were 1673 individual projects and accu-
racy of vertical elevation varied widely. (Data collated by the United States Interagency Elevation Inventory,
U.S. Department of Commerce 2014.)
DEMs should be considered, since errors may result in inaccurate hydrological flow models (Doctor
and Young 2013).
As an active sensor, LiDAR has limitations determined by the emittance wavelength (Pack et al.
2012). Most LiDAR sensors for topographic applications emit an NIR wavelength (1064 nm). This
wavelength is completely absorbed by water; inundated areas would, therefore, simply appear as
no data. This is particularly problematic in tidal* wetlands with daily variation in hydrology† or
in freshwater wetlands with high seasonal variability, such as the Okefenokee Swamp of Georgia
(Rose et al. 2013) or the prairie pothole regions of the United States and Canada (Shook et al. 2013).
Inundation at the precise time of data collection can restrict data acquisition if only an NIR sensor
is used. To overcome this limitation, a sensor emitting a blue-green wavelength (532 nm) can be
used, as in the Experimental Advanced Airborne Research LiDAR (EAARL) system. Alternately,
Flener et al. (2013) integrated point cloud data derived from optical images obtained by UAS over a
river channel with terrestrial LiDAR data to generate a seamless topo-bathymetric DEM with level
of change detection of channel depth estimated at ~47 cm.
Aside from topographic information, LiDAR also collects data on vegetation height and density
that can be used to quantify biomass. However, there are specific challenges related to wetland
types. For graminoid wetlands, vegetation density may prohibit pulse penetration to ground sur-
face as well as confound point classification based upon elevation. Moreover, since the accuracy of
the LiDAR elevation/height data is significant percentage of graminoid vegetation height (1–2 m),
small inaccuracies in either ground surface model or vegetation height will have greater impact,
especially on biomass estimates. Schmid et al. (2011) found error rates differed by vegetation spe-
cies within a coastal habitat, while Hladik and Alber (2012) applied species-specific corrections
in a salt marsh along Georgia coast to improve overall DEM error from 0.10 ± 0.12 m (standard
deviation) to −0.01 ± 0.09 m (standard deviation). In forested wetlands, general algorithms relat-
ing tree height to biomass were less accurate compared to regional algorithms (Feng et al. 2012).
Wallace et al. (2011, 2012) investigated LiDAR on a UAS platform for forest inventory. There is
some variation in canopy estimates from repeated measurements, but these were generally <5%
(Wallace et al. 2011), which should give confidence in single-date acquisitions and the ability to
detect significant changes.
LiDAR sensors flown on UAS can operate at lower altitudes and therefore achieve higher point
density than the same sensors flown in piloted aircraft. Higher point density currently achievable
from UAS platforms (up to 62 pts/m2) will facilitate point classification and increase classification
accuracy (Wallace et al. 2012). Moreover, LiDAR sensors are likely to be fused with simultaneous
spectral data to further improve classification.
UAS FOR FLEXIBLE DATA ACQUISITION

Capabilities of UAS for Ground Truthing and Bird’s-Eye View of Wetlands
UAS data acquisition platforms, which have traditionally included large gas-powered aircraft, have
benefited from recent technological advances, including the miniaturization of electronic compo-
nents, advances in battery technology, and improved aircraft navigation and control. We are cur-
rently witnessing a transformative popularization in UAS use resulting from a more flexible data
collection system based on smaller battery-powered electric aircraft, including a variety of fixed-
wing (airplanes) and rotary-wing (helicopters) platforms (Figure 10.2). There are systems on the
market that are ready to use out of the box, therefore expanding remote sensing data collection to
* Tidal bathymetric data tend to be referenced to mean lower low water vertical datum, which is updated every 19 years.
LiDAR acquisition over land tends to reference the North American Vertical Datum of 1988 (NAVD88). The absolute
vertical offset of these two data varies geographically but is generally 0.5–1.5 m.
† Often, specification for LiDAR data acquisition in coastal areas, in which only the NIR (1064 nm) sensor is used, includes
provision of within 2 h of mean lower low tide to maximize extent of exposed ground area.
FIGURE 10.2 Clockwise from the top left: fixed-wing aircraft (eBee, SenseFly), rotary-wing aircraft (S800,
DJI), fixed-wing with catapult launch system (Bramor gEO, C-Astral), and quadcopter (Octane, Volt Aerial
Robotics), used for image acquisition and land cover mapping.
the masses. This section introduces these systems, notes their capabilities, and emphasizes the use
of electric aircraft for data acquisition applied to wetlands mapping.
UAS is a multicomponent system that includes a small aircraft combined with a series of
hardware and software elements supporting navigation and control for remotely sensed data
acquisition. These elements are briefly introduced by the paragraphs later; however, the spe-
cific details surrounding these systems are beyond the scope of this text. For that, the reader is
directed to other sources, including Gupta et al. (2013) and Beard and McLain (2012). Moreover,
UAS development and associated applications are very active areas of research and development.
In this context, search engines may assist in identifying recent developments in the use of these
systems.
Fixed-wing and rotary-wing UAS platforms have been used for image acquisition and mapping
along with achieving significant cost savings over traditional manned flight missions. The charac-
teristics of these systems can vary considerably; thus, both fixed- and rotary-based systems possess
unique features that make them more or less suited for different applications. System selection
and use should consider the specific area/region in question and what data acquisition techniques
need to be employed. For instance, fixed-wing aircraft benefit from their gliding capability, usually
resulting in longer flight times and in their ability to cover relatively larger areas following flight
plans by navigating to waypoints. Conversely, rotary-wing systems include multirotor aircraft (usu-
ally using four or more rotors), which give the aircraft the ability of hovering over a particular area
for continuous data acquisition. Hovering can be particularly relevant when investigating dynamic
processes or targets. However, multirotor systems tend to have relatively higher power consumption
and therefore shorter flight times. Many battery-powered quadcopters, for example, can typically
fly for 15–20 min per battery pack.
The longer range of fixed-wing platforms compared to rotary systems is often hampered by the
requirement of larger areas (and operator skill) needed for takeoff and landing. As presented later,
several strategies have been developed to facilitate aircraft takeoff and landing. Since fixed-wing
systems cannot achieve a vertical takeoff without the assistance of a launching mechanism, larger
instruments are often unable to be attached to the base of these aircraft. As a result, using instru-
ments such as LiDAR or other larger instruments, suspended payloads have proven difficult with
fixed-wing systems, and multiple strategies have been proposed to reduce limitations to what sen-
sors these aircraft can carry. Further, fixed-wing flight benefits from simple, aerodynamic designs,
making these systems ideal for large survey missions such as mapping expansive wetland stretches,
including tidal and nontidal marshes and swamps.
The main advantages of multirotor systems arise from the ability of the UAS pilot to have pre-
cise control of even the smallest movements of the aircraft. This level of control allows the aircraft
to achieve vertical takeoff and landing (VTOL), thus providing rotary-wing UAS with much more
versatility pertaining to where missions can be launched. The VTOL feature is particularly relevant
when dry surfaces are limited. In this context, multirotor systems can be launched and landed on
very small areas such as a dock and a boat or even directly from the operator’s hands. In conjunction
with nearly infinite launching and landing platforms, rotary-wing UAS has the ability to hover and
rotate in one stable position, thus providing a continuous and/or 360° view of an area because of its
advanced position processing abilities.
Navigation and control of UAS benefit from the availability of a series of support components,
including inertial measurement units (IMUs), GPS, barometer, and magnetic compass. Data from
these components, including orientation, velocity, gravitational forces, and location, are used to
keep the aircraft leveled and at a fixed height, often without the need for pilot interaction. Being a
critical component of a UAS, a typical IMU consists of gyroscopes mounted orthogonally to each
other in conjunction with orthogonally mounted accelerometers (Kong 2004). Changes reported
by these gyroscopes and accelerometers are used to monitor and control aircraft attitude, includ-
ing roll, pitch, and yaw. By combining information obtained from the IMU with GPS data, the
platform’s inertial navigation system (INS) can calculate the position, velocity, and orientation of
the craft and provide a feedback mechanism to further calibrate the GPS (Wendel et al. 2006). In
addition, the INS determines flying altitude and attitude to provide data for autopilot control, hov-
ering or to follow GPS waypoints. This information, which may include other aircraft flight data,
can be downlinked by telemetry and displayed to the pilot or to other personnel for monitoring of
the aircraft location, heading, and battery status, among others. Furthermore, IMU on multirotor
systems support adjustments for rotor failure, which in many platform configurations allow the pilot
to maintain control when a rotor is lost and to fly the aircraft back to its home location (e.g., original
takeoff location or any other area defined as home).
The ability to have precise control of an aircraft is one of the key reasons why these systems
are so valuable for wetlands mapping and many other projects and fields of research. Being a key
part of this capability, the INS provides the pilot and the autopilot system improved navigation and
control, allowing for fixed-wing and rotary-wing aircraft to follow preconfigured waypoints. In
addition, the INS can work with other system controls to adequately position the craft or the remote
sensor to guarantee targets that are in the field of view during data acquisition. Further, a skilled
UAS pilot can fly under scenarios that conventional airborne and orbital image acquisition systems
cannot. For instance, rotary-wing systems can be flown under tree canopies or set to acquire data
multiple times in a few hours. Using the combination of the heads-up display, a first-person view
(FPV) of the aircraft, and even a line of sight view from the pilot on the ground, images of specific
locations can be accurately controlled. This near absolute control over data acquisition gives pilots
the ability to collect data of normally hard-to-reach places, with short revisit times, and in nearly
any type of weather. In addition to an IMU and a GPS, other instruments may be incorporated into
a UAS and used to improve aircraft navigation and control. These include a barometer for informa-
tion on aircraft vertical movement (Bristeau et al. 2010), a magnetic compass that can be used as a
fail-safe strategy when readings from other positioning systems are not available, and a sonar, for
accurate estimation of height above ground and automatic takeoff and landing capability (Johnson
and Schrage 2004).
Barring heavy winds, severe weather, or extreme cold, both the fixed-wing and rotary-wing
systems can be considered “all-weather” systems. Since UAS can be deployed quickly, they have
the potential to revisit locations in very short intervals and are less affected by cloud cover than
satellites, thus making them an ideal tool for quantifying wetland habitat dynamics. When planning
UAS missions over wetland areas, system limitations should be considered, including limitations
in flight range, which are influenced by battery life and, if remote controlled, by communications
between the transmitter and the receiver. The range of remote control by the pilot varies from each
model and can vary from under a kilometer to several kilometers. Remote control range is not limit-
ing when autonomous flight systems are utilized. Many systems allow for preflight planning that
employ the use of GPS waypoints for data collection of specific locations. These preflight systems
are completely reliant on the INS of the aircraft to fly to specific locations, gather data, and move to
the next location, independent of in-flight operator intervention. These fully automated flights bring
to light many of the issues facing UAS today.
System variability is reflected also in choices of UAS payload, including sensors. Images
acquired in the visible region of the electromagnetic spectrum by nonmetric cameras may prove suf-
ficient when performing less complex tasks, including the general delineation of wetlands and the
identification of some vegetation species. In this context, point-and-shoot red–green–blue (RGB)
cameras have been successfully used as payload components of fixed and rotary-wing aircraft.
These cameras can be mounted and operated using their factory default options or, alternatively,
can have their software and/or hardware modified to incorporate a nonexisting feature. Many of
these cameras have been modified to acquire images also in the NIR part of the spectrum, allow-
ing for insights into plant vigor and health. Other multispectral systems offer native RG, NIR
capabilities or allow for the selection of the wavelengths to be used during image acquisition (e.g.,
Tetracam Mini-MCA). Developments in miniaturization also have produced hyperspectral sen-
sors that can be mounted on unmanned aircraft. These sensors collect data using tens to hundreds
of narrow wavelength intervals, providing a fine spectral characterization of biophysical prop-
erties and chemical composition of vegetation, water, soil, and other targets (Berni et al. 2009;
Burkart et al. 2014; Duan et al. 2014). Sensors operating within nonoptical regions of the electro-
magnetic spectrum including thermal infrared and microwave also have been used for land cover
characterization and can make significant contributions to wetlands mapping and related projects.
Thermal infrared has been used in the identification of spatial variability in water temperature and
seeps, with implications to monitoring biotic responses to changes in temperature and variability
in salinity levels (Lega and Napoli 2010). Microwave-based systems have been used to monitor
water under canopies and flood levels and to investigate soil moisture (Acevo-Herrera et al. 2010).
Further, vegetation structure have been investigated by using data collected by LiDAR systems
mounted on UAS (Wallace et al. 2012) with great potential for wetland studies. Significant strides
in sensor development have allowed the engineering of smaller and lighter sensor systems that can
be used as payload of small aircraft. Examples of these systems include the Micro-Hyperspec™
hyperspectral sensor by Headwall, the Tamarisk 640 thermal sensor by DRS Technologies, and the
HDL-32E LiDAR by Velodyne.
UAS Data Acquisition and Management

Wetlands mapping has benefited from multiple data types acquired by UAS, including images and
video covering a variety of regions of the electromagnetic spectrum (Acevo-Herrera et al. 2010;
Lechner et al. 2012; Christian Knoth et al. 2013). In addition to these sensors, unmanned aircraft are
often equipped with nonimaging systems, such as LiDAR. Data acquisition campaigns using UAS
may employ remote controllers and involve the uplink of radio signals for aircraft maneuvering and
sensor control. An important feature of the downlink component of multiple unmanned systems
includes real-time streaming video, which gives the pilot or assisting personnel FPV capabilities of
the area being seen by the sensor. Flight conditions affecting sensor aiming and data collection from
specific targets over wetland areas make FPV a very desirable feature when mapping wetlands,
particularly when considering the relatively reduced image footprint of UAS. Additional telemetry
information including system location, velocity, and attitude can also be retrieved and displayed in
real time by, for instance, using on-screen display (OSD). In addition, this information can be stored
during flight and can later be accessed when processing data acquired by the UAS.
As an alternative to remote-controlled systems, aircraft also can be programmed to perform
autonomous flights and to collect data over specific areas of interest. Additional flight flexibility
can be achieved by informing the system flight parameters, including coordinates of waypoints,
flight altitude, forward overlap, and sidelap percentages. A flight planning software component
(Figure 10.3) uses the provided information to compute an ideal flight plan, including multiple flight
lines, which can then be uploaded to the aircraft.
Application requirements define what type of sensor should be used for a particular data col-
lection mission. For instance, the number of spectral bands and the region of the electromag-
netic spectrum covered by these bands are critical parameters that determine the type and quality
of information that can be derived from the collected dataset. Additional data collection choices
may include how the sensor is mounted on the platform and the resulting viewing angle. Although
having the benefit of covering larger areas, oblique imaging may impose challenges involving
in-scene scale variability as well as those associated with image coregistration. Conversely, nadir
images may cover smaller footprints, especially considering the often lower flight heights of UAS,
in comparison to manned systems. In this respect, data acquisition intervals, including shutter
intervals, need to be adequately configured considering flight height and speed, to guarantee suf-
ficient end lap when acquiring images. Continuous image acquisition by video uses multiple frames
per second and does not require intervalometer settings, but these systems may still be affected
by image blur during fast flights. Further, in addition to aircraft positioning, data collection may
involve also the aiming of sensors, usually mounted on vibration dampening structures or gimbals.
Similar to the steps associated with flight control, sensor aiming can be performed by ground
personnel during flight or can be preprogrammed and uploaded to the aircraft during mission
preparation. Data collected by aerial platforms are usually stored on board by employing solid-state
or disk-based media, although data can also be downlinked to the ground station for monitoring
sensor activity and aiming.
FIGURE 10.3 Waypoint programming and flight planning of a UAS mission assisted by mission planning
software.
The operation of UAS for data acquisition over wetlands can be affected by the availability of
dry surfaces necessary for takeoff and landing. Fixed-wing systems employ a variety of strategies
for takeoff, including manned launch or release and catapult-based strategies. For landing, these
systems can use landing gears, can land directly on their bellies, or, when landing space is limited,
can employ parachutes that are deployed when the aircraft flies over the landing area. Rotary-wing
systems have the ability of taking off vertically and therefore are less affected by the unavailabil-
ity of large open areas. In this context, rotary-wing systems used in wetlands mapping have been
launched and retrieved considering a variety of scenarios, including small dry patches, docks, and
boats. It is noteworthy that due to their flight characteristics and their proximity to the target being
remotely sensed, UAS have the potential to promote more disturbances due to noise and wind than
conventional manned aircraft. In addition, issues of privacy, public concern of the use of drones,
safety, and regulations of the UAS regulatory agency (e.g., Federal Aviation Administration [FAA],
in the United States) must always be considered.
Wind can greatly affect flight and image acquisition, particularly when interfering with the
correct aiming and stability of the sensor system. High winds are not uncommon over coastal
wetland areas, and these may impose considerable challenges to flight control and maneuver-
ing. Lighter aircraft and those airframes with larger surfaces tend to be more affected by winds.
Recent developments in platform control systems, including IMUs, have allowed successful data
acquisition campaigns under these challenging scenarios. In addition, the introduction of weath-
erproof systems (e.g., those from Microdrones GmbH) has extended data collection capabilities
under a variety of environmental conditions. Further, increased stability for data acquisition can
be achieved by using three-axis camera stabilization mounts. Data acquisition and the quality
of the data acquired by a UAS also can be affected by other atmospheric conditions, especially
under haze resulting from fog and high aerosol concentration. However, due to the relatively thin
atmosphere between the target and the sensor, data derived from UAS are usually less prone
to be affected by these factors, when compared to airborne systems flying at higher altitudes.
Progressively thicker atmosphere characterizes data acquisition of most manned airborne systems
and orbital systems.
Usually, data acquired by UAS include images, videos, or point sample data, which can be
stored using different formats, depending on the sensor used. In addition to the dataset itself, UAS
can record multiple attributes associated with the data collection mission, including acquisition
coordinates and UAS attitude information, which may differ from the values defined during mis-
sion planning. These ancillary data can be later used for image processing and analysis. Following
acquisition, datasets can be transferred to a final storage device and usually include multiple files,
sometimes thousands of images, depending on intervalometer settings and flight time or one or
multiple videos. UAS system characteristics and flight height restrictions (e.g., under 400 ft) often
result in reduced image footprint. As a result, for the same given area, workflows incorporating
images acquired by UAS need to process a larger number of images, when compared to the number
of images acquired by conventional airborne aerial systems. Storage needs for these datasets can be
in the order of gigabytes to terabytes.
UAS IMAGE PROCESSING TO CREATE ORTHOIMAGES

AND 3D POINT CLOUDS
Like most remotely sensed raw images, UAS images need to be corrected for lens distortion and
topography in order to produce differentially georectified orthoimages that can be integrated into
geodatabases and from which accurate measurements can be made. Image correction is required for
images collected from any platform (Lillesand et al. 2008). Unlike satellite and aerial images that
tend to be orthorectified before they are provided to end users, multiple overlapping UAS images,
corrected for lens distortion, create a series of stereopairs that can be used to create point cloud
data, with each point having associated X, Y, and Z value (as in LiDAR). The UAS point clouds are
often more dense than those of LiDAR and inherently have color (red, green, and blue) information
extracted directly from the photographs. Therefore, each resulting point has associated X, Y, Z, R,
G, and B values.
Processing and analysis of images acquired by UAS involve a variety of workflows, which may
include the use of single images, the generation of mosaics (orthorectified or not), the extraction of
3D point clouds from multiple images, and the reconstruction of the 3D geometry of a given area
based on the acquisition of images from different vantage points (Laliberte et al. 2011). This sec-
tion uses a case study to introduce some of these workflows and incorporates images from a video
acquired over tidal marshes of the Isle of Hope, on the coast of Georgia in the United States. The
UAS images used in this study were acquired by a DJI Phantom 2 Vision quadcopter, equipped
with a 14-megapixel camera able to record video in high definition (1080 p) at 30 frames/s. The
UAS camera is attached to a single-axis stabilization mount and is connected to a servo, which
enables the camera to tilt along the X-axis. During video collection, camera tilt was controlled on
the ground by an operator interacting via Wi-Fi with the DJI mobile application on an Apple iPod or
iPad. Video recording was supported by an FPV system, for aircraft control and navigation as well
as for camera aiming (Figure 10.4).
Image processing included the selection and retrieval of individual frames from the high-
definition video stored on a micro SD card attached to the camera. Frame selection and extraction
were performed using the open-source VLC Media Player application (VideoLAN Organization
2014) to extract sequential individual frames having sufficient (on the order of 80%) overlap of
ground cover. Scene overlap guarantees the representation of analog features by more than one
video frame and is a requirement when conducting feature detection and pairwise image matching
as part of 3D reconstruction of scene geometry (Stefanik et al. 2011).
Imaging sensors used as UAS payloads may incorporate wide viewing angles to increase the
imaged area and to compensate for the relatively lower flight altitude of these systems. A drawback
of wide-field-of-view lenses, however, is the associated geometric distortions, which may impact the
adequate identification and location of objects of interest in the scene (Figure 10.5a). During video
acquisition, our UAS was equipped with a 140° field of view camera, and a series of oblique images
extracted from the resulting video file were processed to correct for lens distortion using Photoshop
CS6 and Adobe Lens Profile for the DJI Phantom 2 Vision (Figure 10.5b). Following lens correc-
tion distortion, we used the Photomerge function in Photoshop to merge and color correct the video
frames to create a single seamless image (Figure 10.5c).
FIGURE 10.4 Authors Cotten and Bernardes flying a DJI Phantom 2.

(a) (b)
(c)
FIGURE 10.5 Marsh area represented by (a) single image before lens distortion correction, (b) single image
after lens distortion correction, and (c) uncontrolled photomosaic.
Wetlands mapping using UAS-derived images may incorporate orthoimages and orthomosa-
ics for increased level of detail and positional accuracy. Due to the small footprint of individual
images, mosaicking and the generation of orthomosaics have been incorporated into image pro-
cessing workflows applied to multiple fields. Orthomosaics at submeter accuracy and 3D models
require the identification of analog features on multiple images acquired by the UAS. In addition,
2D images acquired from multiple vantage points can be used to create a 3D representation of the
landscape. While it is difficult to create orthomosaics directly from oblique aerial photography due
to scale variations within and between images, orthographic representations of the landscape can be
created from the color-coded cloud of points resulting from dense photogrammetric point-matching
methods. These methods are described later.
Uncontrolled changes in attitude of the UAS during data collection may contribute to consider-
able challenges during image processing, particularly those associated with abrupt variations in
scale, viewing angle and illumination that occur during windy conditions or as a result of pilot
actions. In general, lighter platforms and those with larger profiles are more prone to increased vari-
ability in attitude, which may not be fully captured and compensated by hardware (i.e., gyroscopes,
accelerometers, and camera stabilization mount), when these are available. Results from this vari-
ability range from the nonrepresentation of a given object of interest to the reduction of overlapping
areas, which may affect further processing steps involving mosaicking and 3D reconstruction of
the scene.
There are obvious advantages brought about by having a flexible, aerial vantage point from
which to observe and monitor wetlands. With the incorporation of photogrammetric techniques
into mapping procedures for UAS images, the utility of these images for mapping wetland areas
has increased even further. In this context, principles of classic photogrammetry, including the use
of overlapping photographs and parallax, have been expanded, and a variety of software solutions
exist to process these images. In addition, new processing approaches are being researched and
developed, several coming from computer vision, and concepts and algorithms traditionally applied
to metric cameras are now being used with more simple and more available nonmetric camera
systems.
Structure-from-Motion Photogrammetry and Multiple Image Matching

Reconstruction of the 3D geometry of the objects of interest, including the generation of point
clouds, employs an image processing concepts dating back to the 1950s, with relatively recent
implementations to photogrammetry. In traditional stereo photogrammetry, 3D structure can be
resolved from a series of overlapping, offset images, but a priori knowledge of scene geometry,
camera parameters, camera orientation, and ground control point (GCP) targets are required. We
used SfM, which differs from traditional photogrammetry in that it does not require ground control,
reference targets, or a priori knowledge of the camera exposure locations and attitudes. Instead,
the geometry of the camera/scene parameters is solved automatically with very little, if any, user
interaction. The approach originated in the computer vision community and incorporates automatic
feature-matching algorithms (Westoby et al. 2012). By using multiple overlapping images, SfM
incorporates simultaneous, highly redundant, iterative bundle adjustment procedures and extracts a
database of features automatically. As a result, a very accurate point matching between photographs
can be achieved, and an extremely dense RGB-encoded point cloud can be extracted. SfM works
best with sets of highly overlapping images that capture the full 3D structure of a scene viewed
from a wide array of positions. The approach works also with images derived from a moving sensor,
such as frames captured from video. Readers interested in knowing more about SfM are directed to
Westoby et al. (2012) and to Dellaert et al. (2000).
Although internally consistent, models derived from SfM typically lack scale and orientation
provided by GCPs. Consequently, resulting 3D point clouds are generated in a relative image-space
coordinate system. Solutions based on relative coordinates may satisfy multiple applications, and
data can be scaled using a known distance or ruler imaged in a scene. If more rigorous repre-
sentations are required, including meeting analysis and repeatability requirements, data must be
aligned to a real-world, object-space coordinate system. A multidimensional data adjustment can be
achieved by 3D similarity transform using a few GCPs measured after the model is complete. The
corresponding processing workflow would then include the consideration of a known control point
and the definition of direction and dimension. In addition, control point insertion may involve the
integration of 3D points measured on photos into the model solution.
Several software solutions exist to process a series of images and generate a point cloud dataset.
Implementations include cloud based (Autodesk 123D Catch), open source (Visual SfM/CMVS/
Meshlab, Insight3D), and commercial (Agisoft PhotoScan, Eos Systems PhotoModeler, University
of Stuttgart SURE). Figure 10.6 shows different stages of the SfM workflow for 3D microterrain
extraction, vegetation structure representation, and wetland geovisualization.
Once the 3D geometry of an area has been reconstructed, the results can be shared with the
public through web mapping tools, such as ArcGIS Online (ESRI 2014), where users can gener-
ate new maps or edit existing web maps created by others. New content can be added to maps by
employing a variety of approaches, including the use of map notes (e.g., a pushpin), whose pop-up
window can be edited by inserting external web links such as YouTube. In this way, users can
generate dynamic web maps demonstrating 3D models and other results in a variety of formats,
such as photos and videos (see example, Jordan 2014). Web maps created in ArcGIS Online can be
shared as a map or presentation through their specific web links, and they can also be embedded
in websites (Figure 10.7). For example, a web map of Wormsloe Historic Site showing 3D models
(a) (b)
(c) (d)
FIGURE 10.6 Reconstruction of 3D geometry of a marsh wetland based on photographs acquired from
multiple photo stations: (a) single frame captured from the high-definition video, (b) photo stations and coarse
point cloud resulting from the SfM orientation procedure, (c) point cloud of marsh surface, and (d) magnified
section of the marsh point cloud showing the extreme density of 3D points in the surface.
FIGURE 10.7 Customizing pop-up windows in ArcGIS Online.
and images of marshland is hosted by the Center for Geospatial Research (CGR) at the University
of Georgia (Pasqua 2014). Given the increasing number and variety of professionals involved in the
collection, distribution, and use of geospatial data derived from UAS and SfM technologies, it is
key to explore the distribution and the geovisualization of geospatial information by means of web
mapping and the Internet.
CONSIDERATIONS OF FEDERAL AND STATE REGULATIONS

FOR USE OF UAS IN THE UNITED STATES
At the moment, regulations surrounding UASs in the United States are in a state of constant flux
with individual states enacting their own legislation while awaiting a broader ruling by the FAA. As
of April 2014, 15 states have authorized 18 laws focusing on UAS use, while in 2014 alone, 35 states
have considered bills or resolutions referencing the use of UASs and 2 states, Indiana and Utah,
have enacted new laws. The laws passed by Indiana in HB 1009 require warrants and sets excep-
tions for the police use of unmanned aircraft and real-time geo-location tracking devices. The law
also creates the crime of “Unlawful Photography and Surveillance on Private Property,” aimed at
curtailing the unwanted survey of private property without permission. Although the Utah SB 167
law also requires a warrant for police to use or obtain data from UAS, it specifically states the law
is not meant to “prohibit or impede the public and private research, development or manufacture of
unmanned aerial vehicles” (Williams 2014).
The U.S. federal regulations have been slower to develop because of the immense scale of the
National Airspace System (NAS) resulting in the need for a much broader and intensive set of rules
and regulations. In 2013, the FAA released a roadmap for UAS regulations stating:
Ultimately, UAS must be integrated into the NAS without reducing existing capacity, decreasing safety,
negatively impacting current operators, or increasing the risk to airspace users or persons and property
on the ground any more than the integration of comparable new and novel technologies.
FAA (2013)
To achieve this, the FAA currently requires public UAS operators to obtain Certificate of Waiver or
Authorization (COA) to gain access to the NAS, and civil applicants require special airworthiness
certificates, yet these are only temporary solutions until “new or revised operating rules and pro-
cedures are in place and UAS are capable of complying with them.” A COA is defined as an FAA
grant of approval for a specific flight operation. This gives a user authorization to operate a UAS in
the NAS as a public aircraft within areas approved for aviation activities. The airworthiness certi-
fication is a process used by the FAA to ensure that an aircraft design conforms to the appropriate
safety standards in the applicable airworthiness regulations (FAA 2013).
One of the main issues policy makers face in the current era of unmanned aerial vehicles (UAVs)
is the removal of the pilot from the aircraft. Even though FPV is common on many UAS, the field of
view offered in these automated systems pales in comparison to having a human being in the cockpit.
Compared to a traditional manned aircraft, UAS presents multiple complications such as (1) a lack
of environmental and sensory cues that traditional manned aircraft pilots experience, (2) full reli-
ance by a UAS pilot on a data link to control the aircraft, (3) the wide range of sizes and operational
capabilities that present issues for air traffic controllers, and (4) the UAS pilot that cannot comply
with the see-and-avoid (SAA) responsibilities currently required for manned aircraft (FAA 2013).
The FAA UAS roadmap points to two key issues that must be addressed to allow both manned
and unmanned aircraft to inhabit the same airspace. The first issue is SAA capabilities in unmanned
systems, which are used to provide collision protection between aircraft. Manned aircraft have
numerous procedures, guidelines, and proven technologies in place to ensure safe operation; how-
ever, unmanned systems are lacking in this regard. Although some UAS have collision avoidance
systems between similar units, interoperability constraints still need to be developed to avoid
collisions with other UAS, manned flights, and ground-based systems. The other major issue
encompasses control and communications and is mainly being handled by the Radio Technical
Commission for Aeronautics (RTCA), a private, nonprofit corporation that functions as a federal
advisory committee (www.rtca.org).
The RTCA Special Committee 203 (SC-203) was established in 2004 to assure the safety and effi-
ciently of UAS with other aircraft operating within the NAS. The FAA uses the recommendations
from the RTCA to aid in creating or amending policy, regulation and program decisions. They have
highlighted the need for advanced research in data link management to allow for constant control
of the UAS. Furthermore, spectrum analysis with frequency management has been emphasized to
enable safe and efficient operation of UAS through the mitigation of specific frequencies used to
operate UAS (Johnson et al. 2012).
Although these regulations exist, small UASs used by hobbyists are exempt as long as they fol-
low the model airplane guidelines issued by the Department of Transportation in 1981. The aircraft
must stay below 400 ft and avoid populated areas and should not be operated within three miles
of an airport, and they are required to give right of way to full-scale aircraft (FAA 1981). When a
system is not reliant on human intervention, or even when a human is in control, there is a need for
communication among the aircraft, ground operator, other aircraft, and, especially as the number
of UAS increases, air traffic control (ITU-R 2009).
These different state-level regulations and developing federal-level regulations demonstrate how
the landscape for UAS use is constantly changing at a rapid pace. Just recently, Raphael Pirker was
fined by the FAA for using a UAS to shoot a promotional video for a university, but an administra-
tive law judge determined the FAA had no authority over small unmanned aircraft (Levin 2014).
This overturning of an FAA ruling accentuates the lack of guidance on what is legal or illegal in
our current policy environment. The different sets of rules may seem confusing considering that a
child can operate a quadcopter as long as it is a hobby and he or she follows the hobbyist rules, yet a
business cannot monitor crop health for a farmer, nor can a mining company use UAS for mapping
operations. It is now generally understood that UAS are here to stay and can serve a great societal
need. Consequently, the U.S. Congress ordered the FAA to establish laws allowing the safe integra-
tion of civil UAS into the national airspace no later than September 30, 2015.
In the time being, the FAA has set up six public entities to develop and test the evolution of
UAS technologies and policies. The determination of each site was based on geographic location,
climate, ground infrastructure, airspace use, safety, aviation experience, risk, and research need
(Table 10.1). Besides these main objectives, other considerations and objectives were determined
for the test sites. The University of Alaska was chosen because their proposal consisted of test sites
located in seven different climate zones including Hawaii and Oregon. The State of Nevada’s focus
will also involve how air traffic control procedures will evolve with the introduction of civil UAS.
The main objective of New York’s Griffiss International Airport is FAA safety oversight, and thus,
its research will focus on UAS sense and avoid capabilities and how UAS can be integrated into con-
gested airspaces like the Northeast. Furthermore, the North Dakota Department of Commerce will
TABLE 10.1
Location and Main Objectives of the Six UAS Test Sites Established in December 2013 by
the FAA to Integrate UAS into the NAS
Location Main Objectives
University of Alaska Set of standards for unmanned aircraft categories, including state monitoring and
navigation
State of Nevada UAS standards and operations including operator standards and certification requirements
New York’s Griffiss Test and evaluation as well as verification and validation processes under FAA safety
International Airport oversight
North Dakota Department of UAS airworthiness, essential data and validation of high reliability link technology
Commerce
Texas A&M University, System safety requirements for UAS vehicles and operations with a goal of protocols and
Corpus Christi procedures for airworthiness testing
Virginia Tech UAS failure mode testing and identification and evaluation of operational and technical
risks areas
focus on the human factor surrounding UAS. Those wishing to adopt UAS technology for wetlands
mapping must closely follow this development of laws and regulations to avoid strict fines, penalties,
and even injury and property damage.
FUTURE DIRECTIONS FOR UAS AND WETLANDS MAPPING

The main development in UAS use for remotely sensed data acquisition in support of wetlands
mapping comes from electronics and miniaturization. Recent advances contribute to the increased
availability of these systems and the popularization of their use. Increased investments on system
development and the emergence of multiple companies offering a variety of airframes and UAS
functionalities should contribute to the popularization of these systems and tools.
Past studies have used UAS technologies for wetland monitoring, but the potential for more quan-
titative work using these new acquisition techniques is immense. Kuria et al. (2014) used orthopho-
tos derived from photographs taken by a UAS in conjunction with radar data to determine phenology
and land cover change in wetlands located in eastern Africa, while Lechner et al. (2012) investi-
gated the effects underground coal mining can have on local hydrological conditions through UAS
image–based analysis to characterize swamp characteristics. Comparisons have been made between
the low altitude imagery gathered from UAS and ground-based LiDAR or terrestrial laser scan-
ning (TLS) for a beach dune system, demonstrating the accuracy (average difference in the vertical
values of 0.05 m) of these methods (Mancini et al. 2013). This is especially noteworthy to resource
management agencies and universities with limited budgets. While a UAS such as the DJI Phantom
2 Vision quadcopter currently costs approximately U.S. $1,200, the TLS instrumentation may be
50–100 times more expensive. The UAS systems have begun to fill the void left between satellite and
airborne imagery and ground-based measurements (d’Oleire-Oltmanns et al. 2012). Bryson et al.
(2013) used a kite-based system to monitor intertidal areas in multiple wavelengths establishing the
usefulness of such a system in monitoring such dynamic wetland regions (Figure 10.8).
Many topics involving wetland research would greatly benefit from the temporal and spatial
resolutions offered by UAS. Zomer et al. (2009) discussed the need for building spectral libraries
for land cover classification of wetland regions from currently available remotely sensed data. Yet,
if UAS can be utilized, the detailed spatial scales of such libraries would be extremely valuable
for evaluating species diversification and distributions. Additionally, monitoring of wetland distur-
bances has been limited to flights of aerial imagery or cloud-free satellite observations, therefore
providing limited knowledge of how disturbances change on short timescales. The flexibility of
launching a UAS within minutes of arriving on location is a key factor in monitoring time-dependent
phenomena such as tidal stages, flooding, extreme weather events, fire, or pollutant dispersion.
Using UAS to bridge the gap between satellite/aerial imagery and ground-based measurements
can greatly benefit wetland research. More knowledge could be gained regarding the temporal and
spatial patterns of marsh dieback through the use of UAS equipped with multispectral imaging
capabilities. With the high spatial and temporal resolutions offered, the health and dynamics of
oyster beds and other intertidal species can be determined before entire regions are left barren.
Exploiting the ease at which these systems can be launched can be critical in monitoring urban or
agricultural development near sensitive areas. Thus, UAS may serve as a forecasting tool, allowing
action to be taken before irreversible damage is done to critical wetland areas.
This technology will change the way we run our everyday lives just like satellites, cell phones,
and computers did in the years past. Once we have laws in place regarding privacy issues and safety
regulations, an unimaginable number of products will utilize UAS for everything from wetland
monitoring, crop health, animal behavior, to transportation. Everyone, however, must be aware that
along with the advantages of emerging technologies, there are concerns and dangers including inva-
sion of privacy, collision with aircraft, and damage to property. It is imperative that users of UAS be
informed, follow state and federal regulations, and operate equipment with care in order to maxi-
mize the potential use and benefits of this new technology for wetlands mapping.
0.5
NDVI
0
–0.5
–1
(b)
2m
(a) (c)
FIGURE 10.8 Normalized difference vegetation index (NDVI) maps derived from raw color and NIR imag-
ery: (a) NDVI map for entire shoreline, (b) detailed view of NDVI, and (c) corresponding color imagery of
detailed section. (From Bryson, M. et al., PLoS One, 8, e73550, 2013.)
REFERENCES
Acevo-Herrera, R., A. Aguasca, X. Bosch-Lluis, A. Camps, J. Martínez-Fernández, N. Sánchez-Martín, and
C. Pérez-Gutiérrez. 2010. Design and first results of an UAV-borne L-Band radiometer for multiple moni-
toring purposes. Remote Sensing 2(7):1662–1679. doi: 10.3390/rs2071662.
Ackleson, S.G. and V. Klemas. 1987. Remote sensing of submerged aquatic vegetation in Lower Chesapeake
Bay: A comparison of Landsat MSS to TM imagery. Remote Sensing of the Environment 22(2):
235–248.
Anderson, K. and K.J. Gaston. 2013. Lightweight unmanned aerial vehicles will revolutionize spatial ecology.
Frontiers in Ecology and the Environment 11(3):138–146. doi: 10.1890/120150.
Baker, C., R. Lawrence, C. Montagne, and D. Patten. 2006. Mapping wetlands and riparian areas
using Landsat ETM+ imagery and decision-tree-based models. Wetlands 26(2):465–474. doi:
10.1672/0277-5212(2006)26[465:MWARAU]2.0.CO;2.
Bater, C.W. and N.C. Coops. 2009. Evaluating error associated with lidar-derived DEM interpolation.
Computers and Geosciences 35(2):289–300.
Beard, R.W. and T.W. McLain. 2012. Small Unmanned Aircraft: Theory and Practice. Princeton, NJ: Princeton
University Press.
Berni, J.A.J., P.J. Zarco-Tejada, L. Suarez, and E. Fereres. 2009. Thermal and narrowband multispectral remote
sensing for vegetation monitoring from an unmanned aerial vehicle. IEEE Transactions on Geoscience
and Remote Sensing 47(3):722–738. doi: 10.1109/Tgrs.2008.2010457.
Biron, P.M., G. Chone, T. Buffin-Belanger, S. Demers, and T. Olsen. 2013. Improvement of streams hydro-
geomorphological assessment using lidar DEMs. Earth Surface Processes and Landforms 38(15):1808–
1821. doi: 10.1002/esp.3425.
Bogucki, D.J. and G.K. Gruendling. 1978. Remote Sensing to Identify, Assess and Predict Ecological Impact on
Lake Champlain Wetlands. Washington , DC: Office of Water Research and Technology, U.S. Department
of the Interior.
Bristeau, P.J., E. Dorveaux, D. Vissiere, and N. Petit. 2010. Hardware and software architecture for state estima-
tion on an experimental low-cost small-scaled helicopter. Control Engineering Practice 18(7):733–746.
doi: 10.1016/j.conengprac.2010.02.014.
Bryson, M., M. Johnson-Roberson, R.J. Murphy, and D. Bongiorno. 2013. Kite aerial photography for low-
cost, ultra-high spatial resolution multi-spectral mapping of intertidal landscapes. PLoS One 8:e73550.
doi: 10.1371/journal.pone.0073550.
Burkart, A., S. Cogliati, A. Schickling, and U. Rascher. 2014. A novel UAV-based ultra-light weight spectrom-
eter for field spectroscopy. IEEE Sensors Journal 14(1):62–67. doi: 10.1109/Jsen.2013.2279720.
Burne, M.R. and R.G. Lathrop. 2008. Remote and field identification of vernal pools. In: A.J.K. Calhoun and
P.G. DeMaynadier (eds.), Science and Conservation of Vernal Pools in Northeastern North America,
pp. 55–68. Boca Raton, FL: CRC Press.
Butera, K.M. 1983. Remote sensing of wetlands. IEEE Transactions on Geoscience and Remote Sensing
GE-21(3):383–392.
Carter, V., D.L. Malone, and J.H. Burbank. 1979. Wetland classification and mapping in western Tennessee.
Christensen, E.J., J.R. Jensen, E.W. Ramsey, and H.E. Mackey, Jr. 1988. Aircraft MSS data registration
and vegetation classification for wetland change detection. International Journal of Remote Sensing
9(1):23–38.
Christian Knoth, B.K., T. Prinz, and T. Kleinebecker. 2013. Unmanned aerial vehicles as innovative remote
sensing platforms for high-resolution infrared imagery to support restoration monitoring in cut-over
bogs. Applied Vegetation Science 16:509–517.
Colomina, I. and P. Molina. 2014. Unmanned aerial systems for photogrammetry and remote sens-
ing: A review. ISPRS Journal of Photogrammetry and Remote Sensing 92:79–97. doi: 10.1016/
j.isprsjprs.2014.02.013.
Colwell, R.N. 1997. History and place of photographic interpretation. In: W.R. Philipson (ed.), Manual of
Photographic Interpretation, pp. 3–47. Bethesda, MD: American Society for Photogrammetry and
Remote Sensing.
Coveney, S. 2013. Association of elevation error with surface type, vegetation class and data origin in discrete-
returns airborne LiDAR. International Journal of Geographical Information Science 27(3):467–483.
doi: 10.1080/13658816.2012.695794.
Habitats of the United States. Washington, DC: U.S. Fish and Wildlife Service.
Dahl, T.E. 2000. Status and Trends of Wetlands in the Conterminous United States 1986 to 1997. Washington,
DC: U.S. Fish and Wildlife Service.
Dahl, T.E. and G.J. Allord. 1996. History of wetlands in the conterminous United States. In: J.D. Fretwell, J.S.
Williams, and P.J. Redman (eds.), National Water Summary on Wetland Resources. Reston, VA: U.S.
Geological Survey.
Dahl, T.E., J. Dick, J. Swords, and B. Wilen. 2009. Data Collection Requirements and Procedures for Mapping
Wetland, Deepwater, and Related Habitats of the United States. Arlington, VA: U.S. Fish and Wildlife
Service, Technical Procedures Report.
Dahl, T.E., W.E. Frayer, and C.E. Johnson. 1991. Wetlands Status and Trends in the Conterminous United
States mid-1970’s to mid-1980’s. Washington, DC: U.S. Fish and Wildlife Service.
Dellaert, F., S.M. Seitz, C.E. Thorpe, and S. Thrun. 2000. Structure from motion without correspondence. In
IEEE Conference on Computer Vision and Pattern Recognition, pp. 557–564. Hilton Head Island, SC:
IEEE.
DigitalGlobe. 2014a. Basic imagery. Accessed April 27, 2014. http://www.digitalglobe.com/sites/default/files/
Basic%20Imagery%20Datasheet_0.pdf.
DigitalGlobe. 2014b. Details of high resolution commercial earth imaging satellites. Accessed April 27, 2014.
http://www.digitalglobe.com/about-us/content-collection#overview.
Doctor, D.H. and J.A. Young. 2013. An evaluation of automated GIS tools for delineating karst sinkholes and
closed depressions from 1 m LiDAR-derived digital elevation data. In National Cave and Karst Research
Institute Symposium, pp. 449–458. Carlsbad, NM: National Cave and Karst Institute (NCKRI).
d’Oleire-Oltmanns, S., I. Marzolff, K. Peter, and J. Ries. 2012. Unmanned Aerial Vehicle (UAV) for monitoring
soil erosion in Morocco. Remote Sensing 4:3390–3416. doi: 10.3390/rs4113390.
Duan, S.B., Z.L. Li, H. Wu, B.H. Tang, L. Ma, E. Zhao, and C. Li. 2014. Inversion of the PROSAIL model
to estimate leaf area index of maize, potato, and sunflower fields from unmanned aerial vehicle hyper-
spectral data. International Journal of Applied Earth Observation and Geoinformation 26:12–20. doi:
10.1016/j.jag.2013.05.007.
ESRI. 2014. ArcGIS, The mapping platform for your organization, Accessed November 6, 2014. http://www.
arcgis.com/features.
FAA. 1981. Model Aircraft Operating Standards AC 91–57. Washington, DC.
FAA. 2013. Integration of Civil Unmanned Aircraft Systems (UAS) in the National Airspace System (NAS)
Roadmap. Washington, DC.
Feng, Z., G. Qinghua, and K. Maggi. 2012. Allometric equation choice impacts LiDAR-based forest bio-
mass estimates: A case study from the Sierra National Forest, CA. Agricultural and Forest Meteorology
165:64–72. doi: 10.1016/j.agrformet.2012.05.019.
FGDC. 2009. Wetlands Mapping Standard. Reston, VA.
Flener, C., M. Vaaja, A. Jaakkola, A. Krooks, H. Kaartinen, A. Kukko, E. Kasvi, H. Hyyppä, J. Hyyppä, and
P. Alho. 2013. Seamless mapping of river channels at high resolution using mobile LiDAR and UAV-
Photography. Remote Sensing 5(12):6382–6407. doi: 10.3390/rs5126382.
Frayer, W.E. 1983. Status and Trends of Wetlands and Deepwater Habitats in the Conterminous United States,
1950’s to 1970’s. Fort Collins, CO: Department of Forest and Wood Sciences, Colorado State University.
Gesch, D. 2007. The national elevation dataset. In: D.F. Maune (ed.), Digital Elevation Model Technologies
and Applications: The DEM Users Manual, pp. 99–118. Bethesda, MD: American Society for
Photogrammetry and Remote Sensing.
Greenfeld, J. 2001. Evaluating the accuracy of digital orthophoto quadrangles (DOQ) in the context of parcel-
based GIS. Photogrammetric Engineering and Remote Sensing 67(2):199–205.
Gruendling, G.K., D.J. Bogucki, and M. Madden. 1980. Remote sensing to monitor water chestnut growth in
Lake Champlain. Journal of Soil and Water Conservation 35(2):79–81.
Gupta, S.G., M.M. Ghonge, and P.M. Jawandhiya. 2013. Review of unmanned aircraft system
(UAS). International Journal of Advanced Research in Computer Engineering and Technology
2(4):1646–1658.
Guzha, A.C. and S. Shukla. 2012. Effect of topographic data accuracy on water storage environmental service
and associated hydrological attributes in South Florida. Journal of Irrigation and Drainage Engineering
138(7):651–661.
Hannavy, J. 2008. Encyclopedia of Nineteenth-Century Photography. 2 vols. New York: Taylor & Francis Group.
Wetlands 23(2):436–448.
Hladik, C. and M. Alber. 2012. Accuracy assessment and correction of a LiDAR-derived salt marsh
digital elevation model. Remote Sensing of the Environment 121:224–235. doi: DOI 10.1016/
j.rse.2012.01.018.
Hogg, A.R. and J. Holland. 2008. An evaluation of DEMs derived from LiDAR and photogrammetry for wet-
land mapping. Forestry Chronicle 84(6):840–9.
Irish, J.L., and W.J. Lillycrop. 1999. Scanning laser mapping of the coastal zone; the SHOALS system. ISPRS
Journal of Photogrammetry and Remote Sensing 54(2–3):123–129.
ITU-R. 2009. Characteristics of unmanned aircraft systems and spectrum requirements to support their safe
operation in non-segregated airspace. Radiocommunication Sector of International Telecommunication
Union, Geneva, Switzerland. Technical Report.
Jenkins, D. and B. Vasigh. 2013. The Economic Impact of Unmanned Aircraft Systems Integration in the United
States. Arlington, VA: AUVSI Economic Report.
Jensen, J.R. 2007. Remote Sensing of the Environment: An Earth Resource Perspective 2nd edn., Prentice Hall
Series in Geographic Information Science. Upper Saddle River, NJ: Pearson Prentice Hall.
Water Conservation Area 2A using a time series of normalized remotely sensed data. Photogrammetric
Jensen, J.R., R. Sharitz, and E.J. Christensen. 1984. Nontidal wetland mapping in South Carolina using air-
borne multispectral scanner data. Remote Sensing of the Environment 16(1):1–12.
Johnson, C., J. Griner, K. Hayhurst, J. Shively, M. Consiglio, E. Muller, J. Murphy, and S. Kim. 2012.
Unmanned aircraft systems (UAS) integration in the national airspace system (NAS) project subcom-
mittee final. NASA Advisory Council Aeronautics Committee, UAS Subcommittee, Accessed May 13,
2014. https://publicintelligence.net/nasa-uas-nas/.
Johnson, E.N. and D.P. Schrage. 2004. System integration and operation of a research Unmanned Aerial
Vehicle. Journal of Aerospace Computing, Information, and Communication 1(1):5–18.
Jordan, T. 2014. 3D model of the marsh at wormsloe historic site video. YouTube, Accessed November 6, 2014.
http://www.youtube.com/watch?v=tfP_8yk70Fg.
Julian, J.T., J.A. Young, J.W. Jones, C.D. Snyder, and C.W. Wright. 2009. The use of local indicators of spatial
association to improve LiDAR-derived predictions of potential amphibian breeding ponds. Journal of
Geographical Systems 11(1):89–106. doi: 10.1007/s10109-008-0074-4.
Klemas, V. 2013. Remote sensing of wetland biomass: An overview. Journal of Coastal Research 29:1016–1028.
Klemas, V., D. Bartlett, and R. Rogers. 1975. Coastal zone classification from satellite imagery. Photogrammetric
Kong, X.Y. 2004. INS algorithm using quaternion model for low cost IMU. Robotics and Autonomous Systems
46(4):221–246. doi: 10.1016/j.robot.2004.02.001.
Kuria, D., G. Menz, and S. Misana. 2014. Seasonal vegetation changes in the Malinda wetland using bi-temporal,
multi-sensor, very high resolution remote sensing data sets. Advances in Remote Sensing 3:33–48.
Laliberte, A.S., M.A. Goforth, C.M. Steele, and A. Rango. 2011. Multispectral remote sensing from unmanned
aircraft: Image processing workflows and applications for rangeland environments. Remote Sensing
3(11):2529–2551. doi: 10.3390/Rs3112529.
Lang, M. and G. McCarty. 2009. LiDAR intensity for improved detection of inundation below the forest
canopy. Wetlands 29(4):1166–1178. doi: 10.1672/08-197.1.
Lang, M., O. McDonough, G. McCarty, R. Oesterling, and B. Wilen. 2012. Enhanced detection of wetland-
stream connectivity using LiDAR. Wetlands 32(3):461–473.
Lechner, A.M., A. Fletchera, K. Johansen, and P. Erskine. 2012. Characterising upland swamps using object-
based classification methods and hyper-spatial resolution imagery derived from an unmanned aerial
vehicle. Paper presented at the ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial
Information Sciences, Melbourne, Victoria, Australia, August 25–September 01, 2012.
Lega, M. and R.M.A. Napoli. 2010. Aerial infrared thermography in the surface waters contamination monitor-
ing. Desalination and Water Treatment 23:141–151.
Levin, A. March 7, 2014. Commercial drone pilots cheer judge finding against FAA. Bloomberg, Accessed
April 22, 2014. http://www.bloomberg.com/news/2014-03-06/drone-pilot-s-fine-dropped-by-judge-
finding-against-faa.html.
Li, S., R.A. MacMillan, D.A. Lobb, B.G. McConkey, A. Moulin, and W.R. Fraser. 2011. LiDAR DEM error
analyses and topographic depression identification in a hummocky landscape in the prairie region of
Canada. Geomorphology 129(3/4):263–275.
Lillesand, T.M., J.W. Chipman, and R.W. Kiefer. 2008. Remote Sensing and Image Interpretation. 6th edn.
Hoboken, NJ: John Wiley & Sons.
Madden, M. 2004. Remote sensing and geographic information system operations for vegetation mapping of
invasive exotics. Weed Technology 18:1457–1463.
Madden, M., D. Jones, and L. Vilchek. 1999. Photointerpretation key for the Everglades vegetation classifica-
tion system. Photogrammetric Engineering and Remote Sensing 65(2):171–177.
Mahaney, W.S. and M.W. Klemens. 2008. Vernal pool conservation policy: The federal, state, and local con-
text. In: A.J.K. Calhoun and P.G. DeMaynadier (eds.), Science and Conservation of Vernal Pools in
Northeastern North America, pp. 193–212. Boca Raton, FL: CRC Press.
Mancini, F., M. Dubbini, M. Gattelli, F. Stecchi, S. Fabbri, and G. Gabbianelli. 2013. Using Unmanned Aerial
Vehicles (UAV) for high-resolution reconstruction of topography: The structure from motion approach
on coastal environments. Remote Sensing 5(12):6880–6898. doi: 10.3390/rs5126880.
Maxa, M. and P. Bolstad. 2009. Mapping northern wetlands with high resolution satellite images and LiDAR.
Wetlands 29(1):248–260.
Mitsch, W.J. and J.G. Gosselink. 1986. Wetlands. New York: Van Nostrand Reinhold Co.
Odum, E.P. 1971. Fundamentals of Ecology. 3rd edn. Philadelphia, PA: Saunders.
Pack, R.T., V. Brooks, J. Young, N. Vilaca, S. Vatslid, P. Rindle, S. Kurz, C.E. Parrish, R. Craig, and P.W. Smith.
2012. An overview of ALS technology. In: M. Renslow (ed.), Manual of Airborne Topographic LiDAR.
Bethesda, MD: American Society for Photogrammetry and Remote Sensing.
Pasqua, A. 2014. Advanced geospatial techniques and archaeological methods to investigate historic rice culti-
vation at wormsloe historic site. Accessed November 6, 2014. http://www.cgr.uga.edu/wordpress/?page_
id=2689.
Paul, B.L., F.B. Robert, A.H. Jessica, and T.B. Wigley. 2012. Remote detection of small wetlands in the Atlantic
coastal plain of North America: Local relief models, ground validation, and high-throughput computing.
Forest Ecology and Management 284:107–115. doi: 10.1016/j.foreco.2012.07.034.
Petrie, G. and W.E. Stoney. 2009. The current status and future direction of spaceborne remote sensing platforms
and imaging systems. In: M.W. Jackson (ed.), Earth Observing Platforms and Sensors, pp. 387–422.
Bethesda, MD: American Society for Photogrammetry and Remote Sensing.
Remillard, M.M. and R.A. Welch. 1992. GIS technologies for aquatic macrophyte studies: I. Database develop-
ment and changes in the aquatic environment. Landscape Ecology 7(3):151–162.
Remillard, M.M. and R.A. Welch. 1993. GIS technologies for aquatic macrophyte studies: Modeling applica-
tions. Landscape Ecology 8(3):163–175.
Renslow, M.S. 2012. Manual of Airborne Topographic LiDAR. Bethesda, MD: American Society for
Photogrammetry Remote Sensing.
Rose, L.S., J.C. Seong, J. Ogle, E. Beute, J. Indridason, J.D. Hall, S. Nelson, T. Jones, and J. Humphrey. 2013.
Challenges and lessons from a wetland LiDAR project: A case study of the Okefenokee Swamp, Georgia,
USA. Geocarto International 28(3):210–226.
Schmid, K.A., B.C. Hadley, and N. Wijekoon. 2011. Vertical accuracy and use of topographic LiDAR data
in coastal marshes. Journal of Coastal Research 27(6):116–132. doi: 10.2112/jcoastres-d-10-00188.1.
Schneider, W.J. 1966. Water resources in the Everglades. Photogrammetric Engineering and Remote Sensing
32(6):958–65.
Schneider, W.J. 1968. Color photographs for water resources studies. Photogrammetric Engineering and
Shook, K., J.W. Pomeroy, C. Spence, and L. Boychuk. 2013. Storage dynamics simulations in prairie wetland
hydrology models; evaluation and parameterization. Hydrological Processes 27(13):1875–1889. doi:
10.1002/hyp.9867.
Snyder, G. and M. Lang. 2012. Significance of a 3D elevation program to wetland mapping. National Wetlands
Newsletter 34(5):11–5.
Snyder, G.I., L.J. Sugarbaker, A.L. Jason, and D.F. Maune. 2014. National requirements for improved elevation
data. U.S. Geological Survey Open-File Report 2013–1237.371. doi: 10.3113/ofr20131237.
Stefanik, K.V., J.C. Gassaway, K. Kochersberger, and A.L. Abbott. 2011. UAV-based stereo vision for rapid
aerial terrain mapping. Giscience and Remote Sensing 48(1):24–49. doi: 10.2747/1548-1603.48.1.24.
Stevens, C.W. and S.A. Wolfe. 2012. High-resolution mapping of wet terrain within discontinuous permafrost
using LiDAR intensity. Permafrost and Periglacial Processes 23(4):334–341. doi: 10.1002/ppp.1752.
Su, J. and E. Bork. 2006. Influence of vegetation, slope, and LiDAR sampling angle on DEM accuracy.
Teng, W.L. 1997. Fundamentals of photographic interpretation. In: W.R. Philipson (ed.), Manual of
Photographic Interpretation, pp. 49–113. Bethesda, MD: American Society for Photogrammetry and
Remote Sensing.
Tiner, R. 1997. NWI maps: Basic information on the nation’s wetlands. Bioscience 47(5):269. doi:
10.2307/1313186.
Tiner, R.W. 1984. Wetlands of the United States: Current Status and Recent Trends. Washington, DC: U.S. Fish
and Wildlife Service. Technical Report.
Tiner, R.W. 2009. Field Guide to Tidal Wetland Plants of the Northeastern United States and Neighboring
Canada: Vegetation of Beaches, Tidal Flats, Rocky Shores, Marshes, Swamps, and Coastal Ponds.
Amherst, MA: University of Massachusetts Press.
Toyra, J., A. Pietroniro, C. Hopkinson, and W. Kalbfleisch. 2003. Assessment of airborne scanning laser altim-
etry (LiDAR) in a deltaic wetland environment. Canadian Journal of Remote Sensing 29(6):718–728.
U.S. Department of Commerce. 2014. United states Interagency Elevation Inventory, National Oceanic and
Atmospheric Administration (NOAA) Coastal Services Center, Accessed November 6, 2014. http://
coast.noaa.gov/inventory.
USDA. 2014a. History of aerial photography field office (APFO). Accessed April 27, 2014. http://www.fsa.
usda.gov/FSA/apfoapp?area=about&subject=landing&topic=his.
USDA. 2014b. National Agricultural Imagery Program (NAIP) digital orthorectified images (DOQ), Minnesota,
2010. Accessed April 27, 2014. http://www.mngeo.state.mn.us/chouse/metadata/naip10.html.
USGS. 1996. NAPP and NHAP Photographic Enlargements. Reston, VA: United States Geological Survey.
Factsheet.
USGS. 2014a. Digital orthophoto quarter-quadrangles—Metadata. USGS, Accessed April 27, 2014. http://
gisdata.usgs.gov/metadata/doqq.htm.
USGS. 2014b. Landsat project description. United States Geological Survey, Accessed April 27, 2014. http://
landsat.usgs.gov/about_project_descriptions.php.
USGS. 2014c. National Aerial Photography Program (NAPP). USGS, Accessed April 27, 2014. https://lta.
cr.usgs.gov/NAPP.
USGS. 2014d. National High Altitude Photography (NHAP). USGS, Accessed April 27, 2014. https://lta.
cr.usgs.gov/NHAP.
USGS. 2014e. National High Altitude Photography and National Aerial Photography Program. United States
Geological Survey, Accessed April 27, 2014. https://lta.cr.usgs.gov/Guides/napp.html.
VideoLAN Organization. 2014. VLC media player, Accessed November 6, 2014. http://www.videolan.org.
Wallace, L., A. Lucieer, D. Turner, and C. Watson. 2011. Error assessment and mitigation for hyper-temporal
UAV-borne LiDAR surveys of forest inventory. In Proceedings of SilviLaser 2011, 11th International
Conference on LiDAR Applications for Assessing Forest Ecosystems, University of Tasmania, Hobart,
Tasmania, Australia, October 16–20, 2011. Hobart, Tasmania, Australia: Conference Secretariat.
Wallace, L., A. Lucieer, C. Watson, and D. Turner. 2012. Development of a UAV-LiDAR system with applica-
tion to forest inventory. Remote Sensing 4(12):1519–1543. doi: 10.3390/rs4061519.
Welch, R., J. Alberts, and M. Remillard. 1992. Integration of GPS, remote sensing, and GIS techniques for
coastal resource management. Photogrammetric Engineering and Remote Sensing 58(11):1571–1578.
Welch, R., R.F. Doren, and M. Remillard. 1995. GIS database development for South Florida’s National Parks
and Preserves. Photogrammetric Engineering and Remote Sensing 61(11):1371–1381.
Welch, R., M. Madden, and R.F. Doren. 1999. Mapping the Everglades. Photogrammetric Engineering and
Welch, R., R.B. Slack, and M.M. Remillard. 1988. Remote sensing and geographic information system tech-
niques for aquatic resource evaluation. Photogrammetric Engineering and Remote Sensing 54(2):177–185.
Welch, R. and A.W. Thomas. 1985. Aerial photogrammetry: A potential technology for monitoring soil erosion.
Imaging Technology Research and Development 28(5):4–7.
Wendel, J., O. Meister, C. Schlaile, and G.F. Trommer. 2006. An integrated GPS/MEMS-IMU navigation sys-
tem for an autonomous helicopter. Aerospace Science and Technology 10(6):527–533. doi: 10.1016/
j.ast.2006.04.002.
Westoby, M.J., J. Brasington, N.F. Glasser, M.J. Hambrey, and J.M. Reynolds. 2012. “Structure-from-Motion”
photogrammetry: A low-cost, effective tool for geoscience applications. Geomorphology 179:300–314.
doi: 10.1016/j.geomorph.2012.08.021.
Wilen, B.O. and M.K. Bates. 1995. The U.S. fish and wildlife service’s national wetlands inventory project.
Vegetatio 118(1–2):153–169. doi: 10.2307/20046601.
Williams, R. 2014. Current unmanned aircraft state law landscape. Accessed April 9, 2014. http://www.ncsl.
org/research/civil-and-criminal-justice/current-uas-state-law-landscape.aspx.
Yang, J., G. Zhou, X. Yu, and W. Zhu. 2011. Design and implementation of power supply of high-power
diode laser of LiDAR onboard UAV. Paper presented at the International Symposium on Image and Data
Fusion (ISIDF), Tengchong, Yunnan, China.
Zandbergen, P.A. 2010. Accuracy considerations in the analysis of depressions in medium-resolution LiDAR
DEMs. GIScience and Remote Sensing 47(2):187–207.
Zomer, R.J., A. Trabucco, and S.L. Ustin. 2009. Building spectral libraries for wetlands land cover classification
and hyperspectral remote sensing. Journal of Environmental Management 90:2170–2177. doi: 10.1016/
j.jenvman.2007.06.028.
Section III
Applications of Remote Sensing for
Mapping Specific Wetland Habitats
11 Remote Sensing of
Submerged Aquatic
Vegetation and Coral Reefs
Sam Purkis and Chris Roelfsema
CONTENTS
Introduction..................................................................................................................................... 223
Coral Reefs and SAV......................................................................................................................224
Fundamentals of Remote Sensing Coral Reefs and SAV...............................................................224
Light and the Water Column......................................................................................................224
Spectral Reflectance Characteristics of Bottom Features.......................................................... 226
Sensor Type................................................................................................................................ 226
Image Preprocessing.................................................................................................................. 228
Information Extraction Approaches and Validation................................................................... 229
Integration of Field and Image Data.......................................................................................... 230
Examples of Information Extraction from Remotely Sensed Imagery........................................... 231
Benthic Characteristics of Coral Reefs...................................................................................... 231
Benthic Characteristics of SAV: A Case Study—Moreton Bay, Australia................................. 235
Image-Derived Bathymetry........................................................................................................ 235
Cost Benefit................................................................................................................................ 237
Conclusion...................................................................................................................................... 237
References....................................................................................................................................... 238
INTRODUCTION
Submerged aquatic vegetation (SAV) refers to all underwater flowering plants, of which seagrass is
the most important from a marine ecosystem perspective. Being submerged underwater, the remote
sensing of SAV, as well as coral reefs, is considerably more challenging than for terrestrial targets.
The air–water interface and overlying water column form a dynamic medium that strongly influ-
ences the transfer of electromagnetic radiation (e.g., visible sunlight) that is used in a remote sensing
situation to communicate information about submerged targets. As a result, the spectral differentia-
tion of SAV and coral habitats using optical remote sensing demands specialized strategies, even
if the depth of submergence is only a few meters. When submerged features are visible within
remote sensing imagery, a situation termed “optically shallow water,” SAV, and coral habitats can
be mapped. Conversely, if the lake or seabed is invisible due to excessive turbidity or water depth,
termed “optically deep water,” benthic features cannot be mapped using satellite or airborne optical
remote sensing. Therefore, optically deep and optically shallow waters require different application
of technologies and/or also integration of field and image-based datasets. This chapter will provide
insight into the use of remote sensing to map submerged features found in coral reef and SAV
environments.
223
CORAL REEFS AND SAV

During a span of more than 600 million years, reef ecosystems have enjoyed episodes of incredible
production and diversification, building edifices of carbonate rock that exist today in the form of
great mountain ranges such as the Dolomite Alps or the exquisitely preserved Devonian patch reefs
of Australia’s Canning Basin (Purkis and Klemas 2011). Reefs have also had to persevere through
the bad times, dodging the eye of extinction on numerous occasions. As discussed by Brasier (1975),
few paleoecologists would doubt that marine vegetation has played a significant role in ecosystems
of the geologic past. Unfortunately though, this role must remain largely enigmatic because of the
general lack of noncalcified plant material in the fossil record.
As is arguably also the case for seagrass meadows, the coral reefs we observe today are a shadow
of their former selves. Many reefs, the majority perhaps, are dead or dying, with remote and healthy
examples gravely threatened (Glynn 1994; Hughes et al. 2003; Hoegh-Guldberg et al. 2007). Coral
reefs and seagrasses react quickly to new stressors because they thrive in a narrow range of envi-
ronmental conditions and are very sensitive to small changes in temperature, light, water quality,
and hydrodynamic setting. For corals, even if these stress factors are sublethal, the colony has little
spare energy reproduction. Population sizes fall as does growth rate. This decline under stress is
exacerbated by poor survival rates of young colonies. Numerous demographic studies show a gen-
eral decline in mortality rates for older, larger coral colonies (Connell 1973; Tanner et al. 1996).
Recent work has even established a strong link between the prevalence of coral disease and ocean
warming (Bruckner and Bruckner 2006; Bruno et al. 2007; Muller et al. 2008; Rosenberg et al.
2009). A comprehensive global assessment by Waycott et al. (2009), which considered 215 literature
studies, found that seagrasses have been disappearing at a rate of 110 km2 year−1 since 1980 and
that 29% of the known areal extent has disappeared since seagrass areas were initially recorded in
1879. Furthermore, rates of decline have accelerated from a median of 0.9% year−1 before 1940 to
7% year−1 since 1990 (Waycott et al. 2009). These rates of seagrass loss are comparable to those
reported for coral reefs, placing seagrass, like reefs, as one of Earth’s most threatened ecosystems.
While so-called “deepwater” or “cold-water” corals can grow in the absence of sunlight at abys-
sal depths in waters as cold as 4°C (e.g., Correa et al. 2012), the reef-building “hermatypic” corals
are confined to the photic zone where sufficient sunlight penetrates to facilitate photosynthesis.
This zone typically occurs between the surf zone and 50 m water depth. Coral polyps do not pho-
tosynthesize but have a symbiotic relationship with zooxanthellae that live within the tissues of
polyps and provide organic nutrients to nourish them. Although corals exist in both temperate and
tropical waters, shallow-water reefs form only in a zone extending from 30°N to 30°S of the equator
where winter temperatures remain above 20°C. Seagrasses are widely distributed along temperate
and tropical coastlines of the world, and like corals, many species can have ranges that extend for
thousands of kilometers of coastline (Short et al. 2007). Also as for corals, seagrasses rely on photo-
synthesis for nourishment and are confined to waters shallower than 50 m. The maximum tolerable
water temperature for seagrass and corals is similar and lies in the range of 36°C–40°C (Purkis
and Riegl 2005; Campbell et al. 2006), but seagrasses extend into the cold waters of the temperate
Atlantic and North Pacific, the Mediterranean, and the Southern Ocean, although species diversity
is greatest in the tropics (Short et al. 2007).
FUNDAMENTALS OF REMOTE SENSING CORAL REEFS AND SAV

Light and the Water Column
Earth’s atmosphere offers high transmittance of electromagnetic energy in the visible and near-
infrared spectrum, and land targets are well illuminated by sunlight in these wavelengths. The
wavelengths at which submerged targets are illuminated, by contrast, are more limited because the
transmittance of light through water is more restricted than through air (Figure 11.1).
Remote Sensing of Submerged Aquatic Vegetation and Coral Reefs 225
Full light range Partly attenuated light Scattering
Attenuation Highly attenuated light Refraction
FIGURE 11.1 Environmental features and processes in coral reefs affecting the radiative transfer character-
istics that are recorded by passive optical remote sensing instruments, including aerial cameras and multispec-
tral and hyperspectral imaging systems. This diagram identifies measurable features, along with factors that
reduce the ability to use images of coral reef and SAV environments. (From Remote Sensing Toolkit, www.
gpem.uq.edu.au/brg-rstoolkit.)
Upon reaching the water’s surface, shortwave radiation is attenuated and the intensity of light
decreases almost exponentially with increasing water depth, as described by the Beer–Lambert law.
Long-wavelength radiation, for example, infrared light, is absorbed within the first few centime-
ters of a water body, and therefore, these wavelengths are unavailable for remote sensing. Lacking
the infrared wavelengths, tools that are so effective on land, such as the Normalized Difference
Vegetation Index (NDVI), cannot be applied for the identification of submerged targets. Further, the
availability of visible wavelength energy becomes more limited with increasing water depth, creat-
ing a situation whereby only very few spectral bands of a remote sensing instrument contain any
signal that pertains to the character of even a shallow lake bottom or seabed.
The effect of temperature on the rate of absorption of visible light by water is negligible. This
is also the case with salinity, and therefore, the challenges of remote sensing submerged targets
are the same for lakes, as for seas and oceans. Water constituents, such as the concentration of
phytoplankton, suspended matter, and gelbstoff, have a pronounced effect on the attenuation of
electromagnetic energy. In clear open waters, the absorption of light by water proceeds with the
longest wavelengths first. Therefore, in the visible spectrum, red, yellow, and green wavelengths are
absorbed at shallower water depths than blue and violet, which persist to penetrate deeper into the
water column. It is because of this differential absorption of light by wavelength that clear ocean
and lake waters appear blue to the eye. When water contains phytoplankton, which is often the case
in coastal waters and productive lakes, chlorophyll-a pigments absorb electromagnetic energy most
strongly in the shortest wavelengths (blue and violet), such that green penetrates deepest, and the
water appears green or green-blue in color.
Turbid Clear to turbid Clear
You can (almost) You can (almost)

You can see the bottom
never see the bottom, always see the bottom,
through most of the area
even in knee-deep water even in deeper water
FIGURE 11.2 A guide to water column characteristics and the degree to which the bed for coral reefs and
SAV environments can be observed.
As a result of the strong attenuation of sunlight by water, there are limitations for the use of optical
remote sensing to distinguish SAV and coral reef environments. As a general rule, if the lake bottom
or seabed is visible within remote sensing imagery, substrate features can be mapped. By contrast,
if the bed is invisible, even after the imagery has been enhanced, mapping is impossible. There is no
defined and consistent line between these two situations because the water column is vertically and
laterally dynamic in terms of clarity, however. For this reason, it is a common situation, for example,
that a seagrass meadow might be mappable in some areas but not in others (Figure 11.2).
Spectral Reflectance Characteristics of Bottom Features

Each feature on the Earth’s surface, including those that are submerged, has a unique spectral
reflectance signature that characterizes the amount of light reflected from the feature at specific
wavelengths of the electromagnetic spectrum. For submerged environments, spectral reflectance
signatures of bottom features only carry information in the visible light spectrum as longer wave-
lengths are completely attenuated by the overlying water column (Figure 11.3).
As the water column transmits, absorbs, and scatters light along its path, the spectral reflectance
of submerged features changes with both water depth and water type, rendering it a challenge to
distinguish between environments in both deep and turbid waters (Figure 11.4).
Sensor Type
Remote sensing sensors can be divided into passive (e.g., optical) and active (e.g., acoustic) tech-
nologies. Passive sensors measure electromagnetic radiation (e.g., visible light) reflected from the
Earth’s surface, while active sensors emit a signal and interpret its return. Passive and active remote
sensors used for mapping and monitoring the Earth’s surface vary in spatial, spectral, and temporal
characteristics (Figure 11.5).
Spatial characteristics are defined by the pixel size and the extent covered by an image scene.
Pixel size is commonly divided on the basis of pixel width into very high (<0.5 m), high (0.5–10 m),
moderate (10–50 m), and low (50 m to km) spatial resolution. Depending on the acquisition plat-
form, image scenes vary in size between several km2 and tens of thousands of km2. Spectral char-
acteristics of remote sensing data are defined by the number of bands of electromagnetic energy
Halophila spinulosa
0.20
Syringodium spp.
0.18
0.16 Zostera marina
0.14 White sand

Reflectance
0.12 Light brown mud

0.10
Brown mud
0.08
Dark brown mud
0.06
0.04 Hydroclathrus spp.
0.02 Rhizoclonium spp.

0.00 Lyngbya majuscula
500 550 600 650 700
Wavelength (nm)
FIGURE 11.3 Examples of visible-spectrum reflectance signatures for various SAV species and the sub-
strata that they inhabit.
0.15 1 m depth 0.15 2 m depth

0.13 0.13
Reflectance
0.11 0.11
0.09 0.09
0.07 0.07
0.05 0.05
450 550 650 750 450 550 650 750
0.15 Lyngbya spp.

0.15 4 m depth 5 m depth Syringodium spp.
0.13 0.13 Halophila spp.
Reflectance
0.11 0.11 Zostera spp.

Rhizoclonium spp.
0.09 0.09 Hydroclathrus spp.
0.07 0.07
0.05 0.05
450 550 650 750 450 550 650 750
Wavelength (nm) Wavelength (nm)
FIGURE 11.4 Modeled example of how spectral reflectance signatures change with increasing water depth
within a single water type. While these species of SAV are spectrally separable when submerged under 1 m
of water, the separability decreases rapidly with increasing depth. Spectra generated using Hydrolight v4, a
radiative transfer numerical model that computes radiance distributions and related quantities (irradiances,
reflectances, diffuse attenuation functions, etc.) in any water body.
monitored by the instrument. For instance, multispectral imagery is generally considered to be

comprised ten or fewer bands, whereas hyperspectral is characterized by more than ten and up to
hundreds of narrow bands. Temporal characteristics define the minimum revisit time of the remote
sensing instrument to any given point on the Earth’s surface and vary from hours for airborne sen-
sors to days, weeks, and years for satellites.
QuickBird Landsat ETM+ Aqua MODIS
500 m 500 m 500 m

Resolution
Increasing resolution
2.4 m pixel (b) 30 m pixel (c) 250 m pixel
(a)
Extent
Increasing extent
(d) 10 km (e) 180 km (f ) 1000 km
FIGURE 11.5 The different spatial dimensions of remote sensing data for an image of a coral environment,
Eastern Banks, Moreton Bay, Australia. Images (a–c) show the effects of progressively larger pixel sizes
for a 3 km long section of the Eastern Banks area. Images show different image extents, starting at Eastern
Banks (d), Moreton Bay (e), and moving to the entire East Queensland (f). The red box indicates the same area
as shown in images (a–c). (From Phinn, S.R. et al., Visible and infrared overview, in: J. Goodman, S. Purkis,
and S.R. Phinn [eds.], Coral Reef Remote Sensing: A Guide for Multi-Level Sensing Mapping and Assessment,
Springer, Dordrecht, the Netherlands, pp. 328, 2013.)
Both optical and acoustic sensors can be used to discriminate coral reef and SAV environments
(Figure 11.6). Because of their reliance on reflected sunlight, optical sensors are limited to work in
shallow and well-lit areas but hold the advantage over acoustic techniques in that the technology
covers more area at considerably lower cost. Furthermore, optical instruments can capture bot-
tom features at a higher level of detail than acoustic sensors, which mostly discriminate based on
structural properties of the seabed or lake bed. The latter sensors also need to be mounted beneath
a vessel and therefore only capture relatively narrow strips of data. The spatial and spectral charac-
teristics of passive sensors used for mapping coral reefs and SAV are chosen to provide the required
level of ecological detail. While radar sensors are used routinely to monitor terrestrial landscapes
and the surface of water bodies, they are compromised in the detection of submerged features as
microwaves do not penetrate water.
Image Preprocessing
Preprocessing is required to extract anything beyond visual information from remote sensing imag-
ery. For optical remote sensing of submerged targets, radiometric, atmospheric, and geometric cor-
rections are applied, just as is the case for terrestrial landscapes. For remote sensing of coral reefs
AUV
Satellite Aerial Diving
Ocean
Satellite multispectra Airborne
cruises Underwater
hyperspectra hyperspectra acoustics
Airborne Snorkeling
LiDAR Boat-based
Satellite
radar visual ROV
FIGURE 11.6 Summary of remote sensors used to monitor coral reef and SAV environments.
and SAV, however, additional corrections are often required, such as for sunglint (Hedley et al.
2005), air–water surface interface correction (Andréfouët et al. 2003b), and water column correc-
tion (Brando and Dekker 2003). Geometric correction of satellite remote sensing imagery can be
especially challenging in aquatic environments, as compared to terrestrial settings, because the
boundaries of coral reefs and seagrass habitats can be indistinct and often not in the vicinity of
land targets that are commonly used to constrain geographic position. As a result, many more “tie
points” may be needed for a submerged image scene to deliver accurate geometric correction than
required for a terrestrial one.
Information Extraction Approaches and Validation

The techniques used to extract information from remote sensing imagery of SAV and coral reef envi-
ronments are commonly divided into one of two strategies: “empirical” and “physics-based” models.
Empirical models utilize information and knowledge acquired on the ground to produce a clas-
sification. The commonly used empirical classification approaches are “pixel based” and use unsu-
pervised, supervised, or object-based classification and spectral unmixing. Object-based image
analysis proceeds by initially segmenting the image into “objects” using polygons where the objects
represent a group of pixels with specific spectral and textural properties. Next, objects are assigned
to classes on the basis of membership rules (Blaschke 2010).
Physics-based models call on radiative transfer theory to replicate the path of sunlight travel-
ing through the atmosphere and water column, being reflected by the seabed or lake bed, before
making the upward return journey to be intercepted by an overflying remote sensing instrument.
Despite being more computationally demanding, the physics-based approach has the advantage over
empirical models in that it delivers information on both the composition and abundance of the coral
reef and SAV habitat, but also water depth and water quality parameters, such as total suspended
solids and chlorophyll concentration (Kutser et al. 2006). To run a radiative transfer model, precise
information on the optical characteristics of the atmosphere and water column within the image
scene is needed, along with a library of spectral reflectance signatures that correspond to the coral
reef and SAV features predicted to occur in the area. A radiative transfer numerical model, such
as Hydrolight, is often used to compute radiance distributions and related quantities (irradiances,
reflectances, diffuse attenuation functions, etc.) in any water body by varying water constituents
such as chlorophyll or total suspended sediments, water depth, or sea surface state (Mobley 1995).
A validation or accuracy assessment step follows image classification and is used to assess the
quality of information extracted from remotely sensed imagery. In this step, reference data collected
in the field are statistically compared to the remotely sensed map. For maximum statistical power,
the reference data should be independent from the imagery. The most commonly used measures of
map accuracy are the overall accuracy and individual map category accuracies (partitioned between
“users” and “producers”), which are derived from an error matrix that numerically compares refer-
ence and map data (Congalton and Green 2008; Foody 2011).
Multitemporal analysis and change detection approaches are frequently applied to optical remote
sensing data in order to evaluate the dynamics of coral and SAV environments through time. These
approaches are becoming ever more feasible as satellite remote sensing imagery, such as that from
the Landsat program, becomes freely accessible on the Internet (Wulder et al. 2012). For temporal
analysis of coral reefs and SAV, there are several additional considerations as compared to ter-
restrial assessment. For instance, the water column can vary in composition in time and space, as
well as in thickness due to tidal fluctuations. While it is not always possible and as is the case for
terrestrial image time series, temporally separated marine image sequences should consist of data
acquired under similar conditions (e.g., tidal height and water clarity) or else the multiple scenes
should be processed to yield comparable imagery.
Integration of Field and Image Data

Through the process of ground validation, information is collected in the field in order to aid the
process of relating image features to the real world. Such information is typically straightforward
to collect on land because of the ease of accessibility. Submerged areas, by contrast, pose special
challenges:
1. Submerged environments are hidden from casual view, often remote, and their investiga-
tion demands arduous logistics and effort.
2. While the collection of optical measurements to support image processing is routine
on land, collection of these data underwater is problematic and requires specialized
instrumentation.
3. While inclement weather is inconvenient on land, it poses real dangers for work on large
lakes, seas, and oceans, such that areas of interest become inaccessible for large parts of
the year.
While the listed factors make the collection of ground validation for submerged areas rather gruel-
ing, they do not detract from the need for such data to support the remote sensing endeavor.
The ground validation of submerged areas takes several forms. Bottom surveys audit the char-
acter of the seabed or lake bed, whereas bottom shape surveys are used to describe the shape of the
bottom, including benthic features. By contrast, optical water column property measurements and
underwater spectroradiometry are used to characterize the spectral reflectance of benthic features,
without consideration of their size and shape.
“Bottom surveys” aim to describe the abundance, composition, and/or biomass of the benthos
(coral, seagrass, and algae) and substrate (silt, sediment, rubble, and rock/reef matrix), as well as
other details such as morphology and species diversity. Bottom information is typically gathered
through boat-based surveys using drop cameras, towed video, and/or automated underwater vehi-
cles (Davie et al. 2008). Alternatively, in-water surveys can be conducted via snorkeling and scuba
that rely on spot checks, transects, or manta tow surveys (English et al. 1997; Roelfsema et al. 2004;
Roelfsema and Phinn 2010).
“Bottom shape surveys” are used to audit the 3D topographic shape of sea or lake bottoms at
various scales. Rugosity audits provide the highest level of detail but typically require in-water
techniques to access length scales finer than 1 m. Rugosity describes the topography of individual
features, such as corals, and is directly measured using chains or indirectly via stereo photogram-
metric approaches (Friedman et al. 2012). For a coarser resolution assessment of rugosity over large
areas, bathymetric postings acquired acoustically from a vessel or LiDAR are favored (Purkis and
Kohler 2008; Purkis and Brock 2013).
“Optical water property measurements” are used in a remote sensing context to describe how
light is attenuated and scattered through the water column. The measurements are often used in
combination with physics-based coral and seagrass mapping approaches. Such measurements can
additionally be used to audit the composition of the water with regard to organic (chlorophyll) and
inorganic matter (suspended sediments). Optical measurements are usually conducted from a boat
using a variety of instruments and water filtering devices (Dekker et al. 2001).
Collection of “underwater spectra” allows the optical characteristics of benthic features to be
parameterized. Typical targets are various forms of coral, algae, and seagrass species and mor-
phologies, substrate types, or even mixed benthic communities. The measurements combine to
deliver a reflectance library describing the spectral form and variation of benthic targets that can
be used in physics-based classification routines. Spectral measurements have the advantage over
other means of ground validation in that they can support computational modeling experiments
designed to determine the degree to which different benthic targets will be separable in a remote
sensing case (Hochberg and Atkinson 2000; Kutser et al. 2003; Purkis 2005; Leiper et al. 2012).
Further, archived benthic spectra are flexible as can be resampled to the spectral resolution of any
multi- or hyperspectral instrument and therefore can be used to guide remote sensing campaigns
through time, without the need to collect new data. Underwater spectra are either collected by boat,
with optic fibers lowered from above to the seabed, a process that typically must be guided by scuba
divers or uses a terrestrial spectrometer within a waterproof housing placed directly on the seabed
(Daniel et al. 2010; Dekker et al. 2010). In the latter configuration, the spectrometer may be diver
operated or mounted on an autonomous or remotely operated underwater vehicle.
As with terrestrial remote sensing, it is crucial that the spectrometer is rigorously calibrated and
that all the measurements collected in the field are accurately positioned by GPS. Since the GPS
signal does not penetrate water, however, alternative methods need to be employed to translate the
position on the seabed to the surface. Typically employed techniques include a tethered floating GPS
that can be triggered by a diver from beneath the water surface, acoustic underwater positioning sys-
tems, or dead reckoning, as often employed on autonomous underwater vehicles and submersibles.
EXAMPLES OF INFORMATION EXTRACTION

FROM REMOTELY SENSED IMAGERY
Benthic Characteristics of Coral Reefs
Benthic characteristics of coral reefs have been mapped at various spatial scales using remote sens-
ing: reef/non-reef (1,000s–10,000s m), reef type (100s–1,000s m), “geomorphic zone” (10s–100s m),
dominant components of benthic assemblages (1–10s m), and patch reef scale (<1 m) (Figure 11.7). At
a global scale, Andréfouët et al. (2006) employed manual delineation of Landsat imagery to define
WorldView-2 Seagrass
RGB imagery Sand
Coral
Optically derived
DEM
LiDAR DEM
Depth (m)
4
Meters N
6 Patch
reef
8 250
10 200
Meters E 150
100
100
50
50
0 0
FIGURE 11.7 Comparison of remotely derived DEMs acquired passively from Worldview-2 (WV2) and
actively via LiDAR. The top is WV2 imagery from the northern Florida Keys, which has been processed
according to Stumpf et al. (2003) to yield an optically derived DEM (middle). For comparison, the bottom
shows a DEM derived from NASA-EAARL LiDAR for the same area. While the first-order trend of both DEMs
is the same, there are important differences. The color-scale for the DEMs is the same. Credit: Jeremy Kerr.
Satellite imagery: DigitalGlobe. LiDAR: NASA-EAARL. (Reproduced with permission from Purkis, S.J. and
Brock, J.C., LiDAR overview, in: J.A. Goodman, S.J. Purkis, and S.R. Phinn [eds.], Coral Reef Remote Sensing:
A Guide for Multi-level Sensing, Mapping and Assessment, Springer, Dordrecht, the Netherlands, 2013.)
the geomorphic zonation of coral reefs, while mapping at finer spatial scales typically employs
pixel-based classification techniques (Andréfouët et al. 2003a,b). More recently, object-based image
analysis applied to medium-resolution sensors, such as Landsat (Purkis et al. 2012; Schlager and
Purkis 2013), and high-resolution sensors such as WorldView-2 (Phinn et al. 2012; Roelfsema et al.
2013), integrated with ground validation data, has been shown to be effective in mapping coral reefs
at multiple hierarchical spatial scales and extents (Figure 11.8).
It is typical that high-resolution multispectral satellites also image a panchromatic band that is
delivered as a grayscale image that covers the red, green, and blue portions of the electromagnetic
(a) Reef type km N

Barrier reef
0 5 10 20
Fringing barrier reef
Fringing reef
Lagoon reefs
Lagoon water
(b) Geomorphic
Reef slope Reef flat inner
Reef crest Reef flat outer
Reef flat Shallow reef
(c) Benthic community
Coral-rock Other categories
Coral-rubble Breaking waves
Coral-rubble-sed Turbid water
Coral/algea rubble Deep reefs
Rockx-coral Deep slope
Rubble-rock Lagoon slope
Rubble-rock-coral Lagoon reefs
Seagrass
Seagrass-sed
Seagrass/algae rubble sed
Sed rubble seagrass/algae
Sed + small features
Sed rubble coral
Sed
Sed rubble
Fiji
km km
0 2.5 5 10 0 50 100 200
FIGURE 11.8 Example of (a) reef type, (b) geomorphic zone, and (c) benthic community map for large coral
reef system in Kadavu, Fiji. (Modified from Roelfsema, C.M. et al., Int. J. Remote Sens., 33(12), 1, 2013.)
spectrum with higher spatial resolution than the multispectral channels. Pan sharpening, which is
a process of merging high-resolution panchromatic and lower-resolution multispectral imagery to
create a single high-resolution color image, is an effective means of creating ultrahigh-resolution
data for mapping (Figure 11.9).
Side-scan sonar can provide high-resolution imagery of the seafloor (Goff et al. 2000; Collier
and Humber 2007) but cannot provide bathymetric depth information by itself, which is of key
importance in classifying SAV and coral reef habitats for management and ecological purposes.
Survey systems that provide both side-scan sonar and concurrent depth sensory readings are costly,
and because side-scan sonar instruments must be towed below the surface, the exact positioning of
the data can also be difficult to control and determine.
Multibeam sonar instruments are desirable for benthic mapping because they can provide both
spatially accurate bathymetry and benthic characteristic information over a continuous swath of
seafloor (Kostylev et al. 2001; Wilson et al. 2007) (Figure 11.10). Multibeam sonars have trans-
ducers that send and receive a large number of concurrent, highly accurate, and precisely located
signals in an array across the swath of the instrument beneath the vessel. The return signals are
analyzed for two-way travel time and return echo strength. The time required for the sound wave
N
km
00.51 2
Pan-sharpened (0.5 × 0.5 m)
Geomorphic Meters
0 50 100 200 300 400 500
N
Benthic community
Patch level
FIGURE 11.9 Example of geomorphic and benthic community coral reef maps derived from high-spatial-
resolution multispectral QuickBird imagery (2.4 m × 2.4 m) via object-based image analysis. Consideration
of the pan-sharpened data (0.5 m × 0.5 m) facilitates mapping down to the scale of individual coral patches.
Location is Heron Reef, Capricorn Bunker Group, Great Barrier Reef, Australia.
162°18΄W 162°00΄W N
6°00΄N 0 5 10 km Bathymetry
0m
500 m
1000 m
No data
Palmyra 1500 m
Atoll
2000 m
2500 m
5°48΄N
3000 m
3500 m
FIGURE 11.10 An NOAA multibeam bathymetric survey of Palmyra Atoll (U.S. Pacific). Spatial resolution
of the data is 40 m × 40 m, and they were collected by the NOAA Coral Reef Ecosystem Division. Multibeam
is an active ship-mounted remote sensing technology utilizing acoustic energy and is thus unlimited by water
depth. Reef geomorphology is clearly resolved. (From NOAA.)
to reflect off of the seafloor and return to the sounder is an indicator of water depth and hence
provides the bathymetry data. The strength of the echo return, termed “backscatter,” is a function
of the incident angle of the sound pulse, the roughness or surface characteristics of the seafloor
(e.g., coral, sand, or seagrass), and the composition or density of the bottom (e.g., rock or mud).
Multibeam sonar instruments are generally hull-mounted on the vessel for maximum stability. One
problem with multibeam sonar is that swath width is a function of depth and that, under varying
topographic conditions, sample width and backscatter signal characteristics along the vessel track
can vary considerably.
Benthic Characteristics of SAV: A Case Study—Moreton Bay, Australia

The benthic characteristics of SAV have been mapped at various levels of detail around the globe.
Work conducted in Moreton Bay (South East Queensland, Australia) provides a useful case study
for seagrass mapping at high spatial resolution, but also large spatial extent. In Moreton Bay, several
mapping approaches and sensor types have been trialed over time to audit seagrass species composi-
tion, percentage cover, and biomass. As the site is a coastal embayment with varying water depth
and clarity, optical remote sensing techniques alone are insufficient to map the entire area. For this
reason, an integrated approach has been devised to produce maps that completely cover Moreton
Bay, with classification of the seabed based on a blend of local expert knowledge, field data, bathym-
etry, and satellite remote sensing (Roelfsema et al. 2009). For the Eastern Banks, a 200 km2 shal-
low and optically clear precinct of Moreton Bay, the distribution of seagrass was mapped using a
blend of pixel- and object-based classification approaches applied to high- and moderate-spatial-
resolution multispectral and hyperspectral imagery (Phinn et al. 2008; Roelfsema et al. 2009; Lyons
et al. 2012). The deeper and more turbid precincts of the bay, by contrast, have been mapped using a
combination of field-based methods, including drop-camera points, towed video, and data collected
from autonomous underwater vehicles (Stevens and Connelly 2005; Davie et al. 2008). Such field
techniques are often run in parallel to acoustic surveys, which extend the ability to map seagrass
characteristics into optically deep water (Holmes et al. 2008). The culmination of the mapping
effort is the production of GIS-ready information on seagrass species composition and percentage
cover, along with biomass maps that report changes in the seagrass meadows through time (Figure
11.11) (Roelfsema et al. 2014). These high-resolution products for Eastern Banks in Moreton Bay
complement with increased detail annual percentage seagrass cover maps produced since 1988 from
public-domain Landsat imagery. In both cases, the seabed is classified using object-based methods
(Lyons et al. 2012; Roelfsema et al. 2014).
The Moreton Bay case study covers relatively small areas, and as with coral reefs, management
of SAV-dominated areas typically demands regional-scale seabed maps extending tens of thousands
of square kilometers. By contrast, most peer-reviewed publications focus on smaller areas, which
serve to create a disjoint between scientific seabed mapping and the needs of resource managers
(Roelfsema et al. 2009). This being said, work published by Rowlands et al. (2012) covers coral
reef areas at scales exceeding 20,000 km2 while still maintaining meter-scale resolution. Published
examples of regional-scale mapping of SAV also exist (e.g., Wabnitz et al. 2008), though accom-
plished at a resolution of tens of meters.
Image-Derived Bathymetry
Provided that the seafloor or lake floor is shallow and well lit, we have seen that remote sensing is
a powerful technology to map coral and SAV features over vast areas. The product of such work
is 2D (plan view) and therefore fails to capture the vertical complexity of the benthic system. This
depth component is, however, critical for interpreting the geomorphology of aquatic systems, which
in turn may offer significant insight into factors such as the distribution of fish and invertebrate life
(Andréfouët et al. 2005; Purkis et al. 2008; Mellin et al. 2009), as well as the geological history of
Cerulata spp. 0–40

Halophila ovalis 1–10
40–80
Halophila spinulosa 10–20
80–120
Halodule uninervse 20–30
Syringodium spp. 30–40 120–160
Zostera spp. 40–50
50–60
60–70
70–80
80–90
90–100
0 1 2 4
km
Meters
0 50 100 200
(a) (b) (c)
FIGURE 11.11 Examples of remotely sensed maps of seagrass species (a), percentage and cover (b) created
by combining detailed field observations with high-spatial-resolution imagery and object-based image clas-
sification. Aboveground biomass (c) is derived from the cover and biomass maps by means of an empirical
relationship derived from seafloor cores that link these variables and species diversity. (From Roelfsema, C.M.
et al., Remote Sens. Environ., 150, 172, 2014.)
the system (Boss 1996; Wilkinson and Drummond 2004; Brock et al. 2008; Purkis and Kohler 2008;
Purkis et al. 2010). In rare cases, ancillary information on water depth for a submerged area of inter-
est may be available at a resolution sufficient to be coupled with remote sensing imagery. These
depth data could, for example, have been acquired from vessel soundings or from an airborne bathy-
metric LiDAR survey. In such cases, the 2D product from the remote sensing analysis can be fused
with the separate bathymetry survey to yield a 3D dataset. Unfortunately, such data are not available
for most reef or SAV areas because of the difficulty in obtaining soundings over vast expanses of
shallow water. There exists the possibility, however, of extracting bathymetric dataset directly from
multispectral/hyperspectral satellite or airborne images such as Landsat, IKONOS, or CASI (Green
et al. 2000; Andréfouët et al. 2005). These optical bathymetric algorithms are sufficiently reliable
to now be implemented commercially. Resulting from research conducted in the 1980s, hydro-
graphic charts of French overseas territory (SHOM-SPT 1990) now include bathymetry derived
using Satellite Pour l’Observation de la Terre (SPOT) satellite data and a multiregression algorithm.
NOAA is updating bathymetric maps of the Northwestern Hawaiian Islands using a revised ratio
algorithm applied to IKONOS data (Stumpf et al. 2003).
By virtue of its limited spectral resolution, multispectral imagery is less well poised than
hyperspectral to provide a bathymetry estimate. In saying this, the strategy proposed by Stumpf
et al. (2003) is very satisfactory, provided that a sufficient number of independent soundings are
available with which to tune the algorithm’s coefficients. This empirical ratio approach capitalizes
on the differential attenuation of blue and green light by water. Lyzenga et al. (2006) achieved a
similar result using a radiative transfer model. With the additional number of bands that hyper-
spectral imagery provides, water depth can be retrieved using techniques such as optimization
(Lee et al. 1999, 2001), spectral unmixing (Hedley and Mumby 2003; Hedley et al. 2004), and
lookup-table methodology (Louchard et al. 2003; Mobley et al. 2005; Lesser and Mobley 2007),
to name but a few.
Cost Benefit
The Cost Benefit of using remote sensing to assess coral or SAV depends on the needs that will
be placed on the data, the resources available to acquire them, and the environmental conditions
under which the data must be collected (Roelfsema et al. 2004; Joyce et al. 2005; Roelfsema and
Phinn 2008; Roelfsema and Phinn 2010). The needs vary from the type of detail required, such
as presence–absence of a seabed class, to species-level seafloor audits throughout a regional-scale
study area. The following paragraphs describe the factors that should be considered when assessing
the Cost-Benefit balance of a seabed remote sensing survey.
First, the total time needed to conduct a coral or SAV mapping and monitoring project must be
considered, along with the associated costs, which vary widely per monitoring option (satellite ver-
sus an acoustic survey), biophysical variable (broad seabed categories versus a species-level assess-
ment), and environment type (closed estuary versus oceanic fringing reef) sought.
The factors that conspire to determine the duration and cost of a project are determined by the
survey mobilization and demobilization cost (considered once off) and the field-based and lab-
based effort. In general, it can be noted that the cost for a given project will be a trade-off between
the quality of field and image data used, as well as the complexity of image processing needed to
deliver the needed insight. The time required to conduct a monitoring project is influenced by fac-
tors including the duration dedicated to project planning, the time needed to conduct field work, and
the effort needed to process the remote sensing data. Each component in a project must be budgeted
separately in terms of start-up and ongoing costs. As defined by the BRG Remote Sensing Toolkit,
a project housed in the University of Queensland (see http://www.gpem.uq.edu.au/brg-rstoolkit),
start-up costs are defined as the cost of starting a project from scratch, assuming that all equipment,
software, and hardware need to be purchased and that all staff need to be hired. Equipment needs,
according to the BRG Toolkit, encompass field equipment, staff training, and all maintenance
costs. The actual cost of a project, conversely, is defined as the cost to run the project without the
start-up or general maintenance costs. These costs include the project planning, field costs, image
acquisition, as well as the expenditure to facilitate analysis of field and image data, and the cost of
reporting. It is pertinent to note that work in aquatic environments is typically considerably more
costly than equivalent work on land, because of the often remote location of the work, the need to
use boats, and the more logistically challenging workplace health and safety requirements.
CONCLUSION
Since energy in the microwave wavelengths does not penetrate water, this portion of the electro-
magnetic spectrum is of limited use for monitoring coral reef and SAV environments. By contrast,
instruments operating in the short-wavelength visible spectrum (blue and green light) penetrate the
water to a high degree and are hence of great utility. The principal goal of reef and SAV remote
sensing is to quantify the character of the seafloor. Multispectral satellites are capable of differenti-
ating between areas composed of sand, seagrass, algae, and corals. In theory at least, hyperspectral
sensors, such as CASI, are better capable of separating live coral from dead and discerning SAV
densities, provided that the water column is not too deep.
The water column overlying a submerged feature alters both the magnitude and distribution of
light with respect to wavelength. The spectral effect of submergence becomes more acute as wave-
length and water depth increase. The optical signal that exits the water surface, having interacted
with the sea or lake floor, is a fraction of the strength of the light that originally down-welled onto
the water surface. By virtue of this weak and distorted signal, remote sensing instruments have to
operate at the limits of their sensitivity in order to make meaningful observations of submerged tar-
gets. Furthermore, advanced processing using algorithms, such as spectral unmixing and/or radia-
tive transfer theory, may have to be used. The possibility of being able to visually observe coral and
SAV from a boat will be a good initial indicator as to whether optical remote sensing can be applied
to map an area of interest.
While optical remote sensing is limited to optically shallow waters, acoustic technologies can
be applied in all water types, regardless of depth and turbidity. Acoustic tools are not without limi-
tations, however, as they can only be applied in waters that offer safe passage to a vessel. When
deciding to use passive (optical) or active remote sensing to map and monitor coral and SAV, several
trade-offs need to be considered. First, the cost and skills required to operate each technology vary
and might be out of reach for a given project. Second, there is a variety of techniques that can be
applied to extract information from passive and active remote sensing sensors, but each provides a
different level of information (Roelfsema and Phinn 2008). For instance, pixel- and object-based
image analysis can both be successfully employed for aquatic mapping, and each can provide a host
of advantages and disadvantages. For example, object-based classifiers take into account image prop-
erties such as texture, location, and color, all of which can combine to deliver reliable regional-scale
maps, but the software is expensive. Decisions, such as those listed, are discussed in detail within the
BRG Remote Sensing Toolkit, which provides a framework for scientists and managers to develop
mapping projects for the aquatic environment using earth observation data (Roelfsema et al. 2010).
REFERENCES
Andréfouët, S., A. Gilbert, L. Yan, G. Remoissenet, C. Payri, and Y. Chancerelle. 2005. The remark-
able population size of the endangered clam Tridacna maxima assessed in Fangatau Atoll (Eastern
Tuamotu, French Polynesia) using in situ and remote sensing data. ICES Journal of Marine Science
62:1037–1048.
Andréfouët, S., E.J. Hochberg, C. Payri, M.J. Atkinson, F.E. Muller-Karger, and H. Ripley. 2003a. Multi-
scale remote sensing of microbial mats in an atoll environment. International Journal of Remote Sensing
24(13):2661–2682.
Andréfouët, S., P. Kramer, D. Torres-Pulliza, K.E. Joyce, E.J. Hochberg, R. Garza-Perez, P.J. Mumby et al.
2003b. Multi-site evaluation of Ikonos data for classification of tropical coral reef environments. Remote
Sensing of the Environment 88(1–2):128–143.
Andréfouët, S., F.E. Muller-Karger, J.A. Robinson, C.J. Kranenburg, D. Torres-Pulliza, S. Spraggins, and
B. Murch. 2006. Global assessment of modern coral reef extent and diversity for regional science and
management applications: A view from space. 10th International Coral Reef Symposium, International
Coral Reef Society, Okinawa, Japan.
Blaschke, T. 2010. Object based image analysis for remote sensing. Journal of Photogrammetry and Remote
Sensing 65(1):2–16.
Boss, S.K. 1996. Digital shaded relief image of a carbonate platform (northern Great Bahama Bank): Scenery
seen and unseen. Geology 24:985–988.
Brando, V. and A. Dekker. 2003. Satellite hyperspectral remote sensing for estimating estuarine and coastal
water quality. IEEE Transactions on Geosciences and Remote Sensing 41(6):1378–1387.
Brasier, M.D. 1975. An outline history of seagrass communities. Palaeontology 18:681–702.
Brock, J.C., M. Palaseanu-Lovejoy, C.W. Wright, and A. Nayegandhi. 2008. Patch-reef morphology as a proxy
for Holocene sea-level variability, Northern Florida Keys, USA. Coral Reefs 27:555–568.
Bruckner, A.W. and R.J. Bruckner. 2006. Consequences of yellow band disease (Ybd) on Montastraea annu-
laris (species complex) populations on remote reefs off Mona Island, Puerto Rico. Diseases of Aquatic
Organisms 69(1):67–73.
Bruno, J.F., E.R. Selig, K.S. Casey, C.A. Page, B.L. Willis, C.D. Harvell, H. Sweatman, and A.M. Melendy.
2007. Thermal stress and coral cover as drivers of coral disease outbreaks. PLoS Biology 5(6):e124.
Campbell, S.J., L.J. Mckenzie, and S. P. Kerville. 2006. Photosynthetic responses of seven tropical sea-
grasses to elevated seawater temperature. Journal of Experimental Marine Biology and Ecology
330(2):455–468.
Congalton, R.G. and K. Green. 2008. Assessing the Accuracy of Remotely Sensed Data: Principles and
Practices. Mapping Science. CRC Press, Boca Raton, FL.
Connell, J.H. 1973. Population ecology of reef-building corals. In: O.A. Jones and R. Endean (eds.). Biology
and Geology of Coral Reefs, Vol. 2, Biology. Academic Press, New York.
Collier, J.S. and S.R. Humber. 2007. Time-lapse side-scan sonar imaging of bleached coral reefs: A case study
from the Seychelles. Remote Sensing of Environment 108:339–356.
Correa, T.B.S., G.P. Eberli, M. Grasmueck, J.K. Reed, K. Verwer, and S.J. Purkis. 2012. Variability of cold-
water coral mounds in a high sediment input and tidal current regime, Straits of Florida. Sedimentology
59:1278–1304.
Daniel, P., T.J.M. Malthus, S.R. Phinn, C.M. Roelfsema, I. Leiper, and S. Fyfe. 2010. A comparison of spec-
tral measurement methods for substratum and benthic features in seagrass and coral reef environments.
Proceedings of the Art, Science and Applications of Reflectance Spectroscopy Symposium, ASD Inc.,
Boulder, CO, Vol. II, 15pp.
Davie, A., K. Hartmann, G. Timms, M. De Groot, and J. Mcculloch. 2008. Benthic habitat mapping with
sutonomous underwater vehicles. OCEANS 2008:1–9.
Dekker, A.G., V.E. Brando, J.M. Anstee, E.J. Botha, Y.J. Park, P. Daniel, T.J. Malthus et al. 2010. A comparison
of spectral measurement methods for substratum and benthic features in seagrass and coral reef envi-
ronments. Proceedings Art, Science and Applications of Reflectance Spectroscopy Symposium, January
2010, Boulder, CO.
Dekker, A.G., V.E. Brando, J.M. Anstee, N. Pinnel, T. Kutser, E.J. Hoogenboom, S. Peters et al. 2001. Imaging
spectrometry of water. In: F. Van Der Meer and S.M. De Jong (eds.). Remote Sensing and Digital Image
Processing, Vol. 4: Imaging Spectrometry Kluwer Publishers, Dordrecht, the Netherlands. pp. 307–359.
English, S., C.R. Wilkinson, and V. Baker. 1997. Survey Manual for Tropical Marine Resources. Australian
Institute of Marine Science, Townsville, Queensland, Australia.
Foody, G.M. 2011. Classification accuracy assessment. IEEE Geoscience and Remote Sensing Society
Newsletter 159:8–13.
Friedman, A., O. Pizarro, S.B. Williams, and M. Johnson-Roberson. 2012. Multi-scale measures of rugosity,
slope and aspect from benthic stereo image reconstructions. PLoS One 7(12): e50440.
Glynn, P.W. 1994. State of coral reefs in the Galapagos Islands: Natural vs. anthropogenic impacts. Marine
Pollution Bulletin 29(1–3):131–140.
Goff, J.A., H.C. Olson, and C.S. Duncan. 2000. Correlation of side-scan backscatter intensity with grain-size
distribution of shelf sediments, New Jersey margin. Geo-Marine Letters 20:43–49.
Green, E.P., P.J. Mumby, A.J. Edwards, and C.D. Clark. 2000. Remote sensing handbook for tropical coastal
management. Edwards, A.J. (ed.) Coastal Management Sourcebooks 3, UNESCO, Paris, France. pp. 316.
Hedley, J.D., A.R. Harborne, and P.J. Mumby. 2005. Simple and robust removal of sun glint for mapping
shallow-water benthos. International Journal of Remote Sensing 26(10):2107–2112.
Hedley, J.D. and P.J. Mumby. 2003. A remote sensing method for resolving depth and subpixel composition of
aquatic benthos. Limnology and Oceanography 48:480–488.
Hedley, J.D., P.J. Mumby, K.E. Joyce, and S.R. Phinn. 2004. Spectral unmixing of coral reef benthos under
ideal conditions. Coral Reefs 23:60–73.
Hochberg, E.J. and M.J. Atkinson. 2000. Spectral discrimination of coral reef benthic communities. Coral
Reefs 19:164–171.
Hoegh-Guldberg, O., P.J. Mumby, A.J. Hooten, R.S. Steneck, P. Greenfield, E. Gomez, C.D. Harvell et al.
2007. Coral reefs under rapid climate change and ocean acidification. Science 318(5857):1737–1742.
Holmes, K.W., K.P. Van Niel, B. Radford, G.A. Kendrick, and S.L. Grove. 2008. Modelling distribution of marine
benthos from hydroacoustics and underwater video. Continental Shelf Research 28(14):1800–1810.
Hughes, T.P., A.H. Baird, D.R. Bellwood, M. Card, S.R. Connolly, C. Folke, R. Grosberg et al. 2003. Climate
change, human impacts, and the resilience of coral reefs. Science 301(5635):929–933.
Joyce, K.E., S.R. Phinn, and C.M. Roelfsema. 2005. The costs and benefits of image acquisition and pre-
processing for coral reef remote sensing. Backscatter. Journal of the Alliance for Marine Remote Sensing
16:23–26.
Kostylev, V.E., B.J. Todd, G.B.J. Fader, R.C. Courtney, G.D.M. Cameron, and R.A. Pickrill. 2001. Benthic
habitat mapping on the Scotian Shelf based on multibeam bathymetry, surficial geology and sea floor
photographs. Marine Ecology Progress Series 219:121–137.
Kutser, T., A.G. Dekker, and W. Skirving. 2003. Modeling spectral discrimination of Great Barrier Reef benthic
communities by remote sensing instruments. Limnology and Oceanography 48(1):497–510.
Kuster, T., I. Miller, and D.L.B. Jupp. 2006. Mapping coral reef benthic substrates using hyperspectral space-
borne image and spectral libraries. Estuarine. Coastal and Shelf Science 70:449–460.
Lee, Z., K.L. Carder, C.D. Mobley, R.G. Steward, and J.S. Patch. 1999. Hyperspectral remote sensing for shallow
waters: 2. Deriving bottom depths and water properties by optimization. Applied Optics 38:3831–3853.
Lee, Z., K.L. Carder, R.F. Chen, and T.G. Peacock. 2001. Properties of the water column and bottom derived
from Airborne Visible Infrared Imaging Spectrometer (AVIRIS) data. Journal of Geophysical Research
106:11639–11651.
Leiper, I., S. Phinn, and A.G. Dekker. 2012. Spectral reflectance of coral reef benthos and substrate assem-
blages on Heron Reef, Australia. International Journal of Remote Sensing 33(12):3946–3965.
Lesser, M.P. and C.D. Mobley. 2007. Bathymetry, water optical properties, and benthic classification of coral
reefs using hyperspectral remote sensing imagery. Coral Reefs 26:819–829.
Louchard, E.M., R.P. Reid, F.C. Stephens, C.O. Davis, R.A. Leathers, and T.V. Downes. 2003. Optical remote
sensing of benthic habitats and bathymetry in coastal environments at Lee Stocking Island, Bahamas: A
comparative spectral classification approach. Limnology and Oceanography 48:511–521.
Lyons, M.B., S.R. Phinn, and C.M. Roelfsema. 2012. Long term land cover and seagrass mapping using
Landsat and object-based image analysis from 1972 to 2010 in the coastal environment of South East
Queensland, Australia. ISPRS Journal of Photogrammetry and Remote Sensing 71:34–46.
Lyzenga, D.R., N.P. Malinas, and F.J. Tanis. 2006. Multispectral bathymetry using a simple physically based
algorithm. IEEE Transactions on Geoscience and Remote Sensing 44:2251–2259.
Mellin, C., S. Andréfouët, M. Kulbicki, M. Dalleau, and L. Vigliola. 2009. Remote sensing and fish–habi-
tat relationships in coral reef ecosystems: Review and pathways for multi-scale hierarchical research.
Marine Pollution Bulletin 58:11–19.
Mobley, C.D. 1995. Hydrolight 3.0 User’s Guide. SRI Project 5632, SRI International, Menlo Park, CA.
Mobley, C.D., L.K. Sundman, C.O. Davis, T.V. Downes, R.A. Leathers, M.J. Montes, J.H. Bowles et al. 2005.
Interpretation of hyperspectral remote-sensing imagery via spectrum matching and look-up tables.
Applied Optics 44:3576–3592.
Muller, E.M., C.S. Rogers, A.S. Spitzack, and R. VanWoesik. 2008. Bleaching increases likelihood of disease
on Acropora palmata (Lamarck) in Hawksnest Bay, St John, US Virgin Islands. Coral Reefs 27:191–195.
Phinn, S.R., E.J. Hochberg, and C.M. Roelfsema. 2013. Visible and infrared overview. In: J. Goodman,
S. Purkis, and S.R. Phinn (eds.). Coral Reef Remote Sensing: A Guide for Multi-Level Sensing Mapping
and Assessment. Springer, Dordrecht, the Netherlands, pp. 3–28.
Phinn, S.R., C.M. Roelfsema, A. Dekker, V. Brando, and J. Anstee. 2008. Mapping seagrass species, cover and
biomass in shallow waters: An assessment of satellite multi-spectral and airborne hyper-spectral imaging
systems in Moreton Bay (Australia). Remote Sensing of the Environment 112(8):3413–3425.
Phinn, S.R., C.M. Roelfsema, and P.J. Mumby. 2012. Multi-scale image segmentation for mapping coral reef
geomorphic and benthic community zone. International Journal of Remote Sensing 33(12):3768–3797.
Purkis, S.J. 2005. A “reef-up” approach to classifying coral habitats from Ikonos imagery. IEEE Transactions
on Geoscience and Remote Sensing 43(6):1375–1390.
Purkis, S.J. and J.C. Brock. 2013. LiDAR overview. In: J.A. Goodman, S.J. Purkis, and S.R. Phinn (eds.). Coral
Reef Remote Sensing: A Guide for Multi-level Sensing, Mapping and Assessment. Springer, Dordrecht,
the Netherlands.
Purkis, S.J., P.M. Harris, and J. Ellis. 2012. Patterns of sedimentation in the contemporary Red Sea as an analog
for ancient carbonates in rift settings. Journal of Sedimentary Research 82:859–870.
Purkis, S.J. and V. Klemas. 2011. Remote Sensing and Global Environmental Change. Wiley-Blackwell,
Oxford, U.K.
Purkis, S.J. and K.E. Kohler. 2008. The role of topography in promoting fractal patchiness in a carbonate shelf
landscape. Coral Reefs 27:977–989.
Purkis, S.J. and B. Riegl. 2005. Spatial and temporal dynamics of Arabian Gulf coral assemblages quantified
from remote-sensing and in situ monitoring data. Marine Ecology Progress Series 287:99–113.
Purkis, S.J., G.P. Rowlands, B.M. Riegl, and P.G. Renaud. 2010. The paradox of tropical karst morphology in
the coral reefs of the arid Middle East. Geology 38:227–230.
Roelfsema, C.M., K.E. Joyce, and S.R. Phinn. 2004. Evaluation of benthic survey techniques for validating
remotely sensed images of coral reefs. 10th International Coral Reefs Symposium, Okinawa, Japan.
Roelfsema, C.M., M. Lyons, E.M. Kovacs, P. Maxwell, M.I. Saunders, J. Samper-Villarreal, and S.R. Phinn.
2014. Multi-temporal mapping of seagrass cover, species and biomass: A semi-automated object based
image analysis approach. Remote Sensing of Environment 150:172–187.
Roelfsema, C.M. and S.R. Phinn. 2008. Evaluating eight field and remote sensing approaches for mapping
the benthos of three different coral reef environments in Fiji. Proceedings of SPIE Asia-Pacific Remote
Sensing, Noumea, New Caledonia.
Roelfsema, C.M. and S.R. Phinn. 2010. Integrating field data with high spatial resolution multispectral satellite
imagery for calibration and validation of coral reef benthic community maps. Journal of Applied Remote
Sensing 4(1):1–28.
Roelfsema, C.M., S.R. Phinn, S. Jupiter, J. Comley, and S. Albert. 2013. Mapping coral Rreefs at reef to reef-
system scales (10–600 km2) using OBIA driven ecological and geomorphic principles. International
Roelfsema, C.M., S.R. Phinn, D. Tracey, J. Speirs, M. Hewson, and K. Johansen. 2010. A web based toolkit for
using remote sensing to map and monitor terrestrial, marine and atmospheric environments. http://www.
gpem.uq.edu.au/brg-rstoolkit/. Accessed September 1, 2014.
Roelfsema, C.M., S.R. Phinn, N. Udy, and P. Maxwell. 2009. An integrated field and remote sensing approach
for mapping seagrass cover, Moreton Bay, Australia. Journal of Spatial Science 54(1):45–62.
Rosenberg, E., A. Kushmaro, E. Kramarsky-Winter, E. Banin, and L. Yossi. 2009. The role of microorganisms
in coral bleaching. ISME Journal 3:139–146.
Rowlands, G., S.J. Purkis, B. Riegl, L. Metsamaa, A. Bruckner, and P. Renaud. 2012. Satellite imaging coral reef
resilience at regional scale. A case-study from Saudi Arabia. Marine Pollution Bulletin 64:1222–1237.
Schlager, W. and S.J. Purkis. 2013. Bucket structure in carbonate accumulations of the Maldive, Chagos and
Laccadive archipelagos. International Journal of Earth Sciences 101:2225–2238.
SHOM-SPT. 1990. Les dossiers de spatio-préparation des campagnes hydrographiques aux Tuamotu-Gambier.
In: Proceedings of Pix’Iles 90: International Workshop on Remote Sensing and Insular Environments in
the Pacific: Integrated approaches, Noméa-Tahiti, pp. 593–595, 6pl.
Short, F., T. Carruthers, W. Dennison, and M. Waycott. 2007. Global seagrass distribution and diversity:
A bioregional model. Journal of Experimental Marine Biology and Ecology 350(1–2):3–20.
Stevens, T. and R.M. Connelly. 2005. Local-scale mapping of benthic habitats to access representation in
marine protected area. Marine and Freshwater Research 56:111–123.
Stumpf, R.P., K. Holderied, and M. Sinclair. 2003. Determination of water depth with high-resolution satellite
imagery over variable bottom types. Limnology and Oceanography 48:547–556.
Tanner, J.E., T.P. Hughes, and J.H. Connell. 1996. The role of history in community dynamics: A modelling
approach. Ecology 77(1):108–117.
Wabnitz, C.C., S. Andréfouët, D. Torres-Pulliza, F.E. Muller-Karger, and P.A. Kramer. 2008. Regional-scale
seagrass habitat mapping in the wider Caribbean region using Landsat sensors: applications to conserva-
tion and ecology. Remote Sensing of Environment, 112:3455–3467.
Waycott, M., C.M. Duarte, T.J.B. Carruthers, R.J. Orth, W.C. Dennison, S. Olyarnik, A. Calladine et al.
2009. Accelerating loss of seagrasses across the globe threatens coastal ecosystems. Proceedings of the
National Academy of Sciences of the United States of America 106(30):12377–12381.
Wilson, M.F.J., B. O’Connell, C. Brown, J.C. Guinan, and A.J. Grehan. 2007. Multiscale terrain analysis of
multibeam bathymetry data for habitat mapping on the continental slope. Marine Geodesy 30:3–35.
Wilkinson, B.H. and C.N. Drummond. 2004. Facies mosaics across the Persian Gulf and around Antigua—
stochastic and deterministic products of shallow-water sediment accumulation. Journal of Sedimentary
Wulder, M.A., J.G. Masek, W.B. Cohen, T.R. Loveland, and C.E. Woodcock. 2012. Opening the archive: How
free data has enabled the science and monitoring promise of Landsat. Remote Sensing of the Environment
122: 2–10.
12 Remote Sensing of Mangroves
Victor V. Klemas
CONTENTS
Introduction..................................................................................................................................... 243
Mangrove Characteristics...............................................................................................................244
Mangrove Remote Sensing............................................................................................................. 245
Advanced Remote Sensing Techniques.......................................................................................... 249
Introduction................................................................................................................................ 249
High-Spatial-Resolution Satellite Techniques........................................................................... 249
Hyperspectral Imaging............................................................................................................... 251
Mangrove Canopy Modeling..................................................................................................... 253
Radar Applications..................................................................................................................... 253
Data Fusion Techniques............................................................................................................. 254
General Considerations.............................................................................................................. 255
Case Study: Object-Based Mapping and Change Analysis of Mangroves in Senegal................... 255
Conclusion...................................................................................................................................... 257
References....................................................................................................................................... 258
INTRODUCTION
Mangrove forests are highly productive ecosystems that dominate the intertidal zone of many of
the world’s low wave energy tropical and subtropical coastlines. Mangroves and the organisms they
support are of significant ecological and economic value. Mangroves help reduce the erosional
impact of storms, serve as breeding and feeding grounds for juvenile fish and shellfish, trap silt that
could smother offshore coral reefs, and cleanse near-shore water through the uptake of nutrients and
pollutants. In many coastal countries, they provide people with food and wood for heating, cooking,
and house building (Heumann 2011; Pinet 2009; Tomlinson 1986).
Mangroves, which once occupied 75% of tropical and subtropical coastlines, are now seriously
threatened by coastal development projects and climate change, including accelerating sea level
rise (Gilman et al. 2008; Guanawardena and Rowan 2005; IPCC 2001; McInnes et al. 2003). In
many countries, mangrove swamps are being cut down to provide firewood or building material and
are being destroyed by the development of shrimp ponds (Alongi 2002, 2008; Erftemeijer 2002;
Heumann 2011; Wang and Sousa 2010). For instance, in Sri Lanka, shrimp sales from ponds in
mangrove areas are an important source of foreign exchange and account for over 40% of total
aquaculture export.
Rapid losses of mangroves are compelling coastal managers to inventory and monitor the remain-
ing mangroves in order to protect them from harmful development (Blasco et al. 2001; Nfotabong-
Atheull et al. 2013; Terchunian and Klemas 1986). Mapping and quantifying the structure and
biomass of mangrove ecosystems on a large scale is also important for studies of carbon storage, bio-
diversity, forest quality, and habitat suitability. Remote sensing has had a crucial role in monitoring
mangroves, but the majority of applications, using airborne and medium-resolution satellite sensors,
have been limited to mapping areal extent and patterns of change at local scales. Recent advances in
sensors and image analysis techniques have improved the accuracy of land cover classification and
243
change detection. Furthermore, they offer scientists the capacity to observe stand characteristics such
as leaf area, canopy closure, species composition, canopy height, and standing biomass (Al-Tahir
et al. 2006; Everitt et al. 2007; Heumann 2011; Kuenzer et al. 2011; Lang and McCarty 2008; Tiner
et al. 2002). The objectives of this chapter are to review practical techniques for remotely monitoring
and mapping mangroves and to use a case study to illustrate their effectiveness.
MANGROVE CHARACTERISTICS
Mangrove forests are among the most productive and biologically important ecosystems in the
world, including trees and shrubs, which grow in the tropical and subtropical tidal zones world-
wide. They have adapted to the harsh conditions of high salinity, warm air and water tempera-
tures, extreme tides, muddy sediment-laden waters, and oxygen-depleted soils. Mangrove plants are
large, woody, and treelike and usually have a thick, partially exposed network of intertwined roots.
Extensive prop roots penetrate deep into the anaerobic mud, bringing oxygen to deeper portions of
the root and providing surfaces for the attachment of clams, oysters, barnacles, and other marine
animals. They are among the few emergent woody plants that tolerate the high salinity of the open
sea. Their wood is very hard and commercially valuable. In some coastal countries, mangroves
provide local people with food and wood for heating, cooking, and house building (Hogarth 2007;
Mitsch and Gosselink 2007; Odum 1993).
More than 50 species of mangroves exist worldwide. Mangrove species are usually distributed
with distance from the water’s edge, forming zones of differing species composition perpendicular
to the intertidal gradient. The red mangrove, Rhizophora mangle, is probably the best known. It
typically grows along the water’s edge and is easily identified by its tangled roots. These roots have
earned mangroves the title “walking trees,” because the red mangrove appears to be standing or
walking on the surface of the water (Pinet 2009).
The other common species include the black mangrove and the white mangrove. The black man-
grove, Avicennia germinans, usually grows at a slightly higher elevation upland from the red man-
grove. The black mangrove can be identified by many fingerlike projections called pneumatophores
that protrude from the soil around its trunk (Florida D.N.R. 1987).
The white mangrove, Laguncularia racemosa, usually occupies the highest elevations farther
upland than the red or the black mangroves. In contrast to the red or black mangroves, the white
mangrove has no visible aerial root system. The white mangrove can be identified by its leaves,
which are elliptical and light yellow green and have two distinguishing glands at the base of the leaf
blade where the stem starts.
All three of these mangrove species use a remarkable method of propagation. Seeds sprout while
still on the trees and drop into the soft bottom around the base of the trees or are transported by
currents and tides to other suitable locations. The tropical and subtropical species are sensitive to
extreme temperature fluctuations as well as subfreezing temperatures. Salinity, water temperature,
tidal fluctuations, and soil types also affect their growth and distribution (Florida D.N.R. 1987).
Mangrove forests dominate the intertidal zone of about 70% of the world’s low wave energy
tropical and subtropical coastlines. In terms of net primary production, healthy mangroves are con-
sidered the most productive, followed by sea grasses, marsh grasses, and other coastal ecosystems.
Mangrove litter provides large amounts of carbon to coastal and offshore marine ecosystems and
contributes a significant amount of the dissolved organic carbon to ocean sediments worldwide.
From an ecological point of view, mangroves (1) provide a vital habitat for a wide variety of animals
and plants; (2) function as nursery and feeding grounds for many species of commercially valuable
fishes, crustaceans, and mollusks; (3) are an important source of carbon for detritus-based food
webs in adjacent coastal waters; (4) reduce shoreline erosion by stabilizing deposited sediments;
(5) buffer the impact of storm waves and floods on inland areas; and (6) trap nutrients and sediments
in runoff from upland areas, helping to maintain the quality of estuarine and near-shore waters
(Mitsch and Gosselink 2007; Odum 1993; Wang and Sousa 2010).
Remote Sensing of Mangroves 245
FIGURE 12.1 Shrimp ponds within mangrove areas in Belize. (Photo courtesy of Ilka Feller.)
Mangroves can readily survive normal storms but suffer widespread structural damage from
powerful hurricane winds and waves, which can defoliate them and/or tear them up (Doyle et al.
2009; Klemas 2009; Pinet 2009). More importantly, in many countries, mangrove forests are threat-
ened by coastal development, including shrimp aquaculture, timber cutting, and establishment of
tourist resorts. It is estimated that one-third of the world’s mangrove forests have been lost in the
past 50 years. For instance, in the Caribbean, the rate of mainland mangrove deforestation is esti-
mated at 1.5% annually (Pannier 1979; Wang and Sousa 2010).
The development of shrimp ponds has been a major problem, since the excavation to create
the ponds frequently occurs within mangrove swamps. Shrimp ponds are artificially bermed hold-
ing ponds constructed to raise shrimp to a marketable size for human consumption (Figure 12.1).
Intertidal areas are chosen to take advantage of natural tidal flushing. In some cases, shrimp ponds
are fertilized to accelerate shrimp growth, and when eutrophic conditions occur because of overfer-
tilization, producers find it less expensive to create additional shrimp ponds in new mangrove areas
rather than clean up the old ponds. In some coastal countries, such as Ecuador, the Philippines, and
Sri Lanka, shrimp pond construction has devastated major mangrove areas.
MANGROVE REMOTE SENSING

Similar to other vegetation types, the mangrove reflectance spectrum has two primary chlorophyll
absorption bands in the blue and red regions that are due to pigments such as chlorophyll a and b and
beta-carotene, which reside in the upper layers of the leaf (“palisade parenchyma”) (Figure 12.2).
By contrast, there is strong reflectance at near-infrared (NIR) wavelengths, caused by scattering in
the “spongy mesophyll,” a deeper layer of cells within the leaf, where the internal cell wall structure
acts as a strong diffuse reflector of NIR radiation. Much of the NIR radiation that is scattered within
the leaf is reflected back through the leaf surface, with another portion of the radiation penetrating
through the leaf potentially to the next leaf layer. Thus, the NIR can help determine canopy density
and biomass (Diaz and Blackburn 2003; Kamaruzaman and Kasawani 2007; Klemas 2011; Purkis
and Klemas 2011).
In the middle-infrared region, three water absorption bands show up depending on how hydrated
the leaf tissue is (Figure 12.2). Thus, if mangroves are stressed or are drying up, there will be less
absorption by the blue and red chlorophyll bands and water bands, as well as less reflection in the NIR.
Dominant factor causing leaf reflectance:
Leaf pigments Cell structure Water content
Visible Near-IR Shortwave-IR
en
e
d
re
u
Re
Bl
80 G Absorption
spectrum
70
of water
60 Chlorophyll Water
absorption
Reflectance (%)
50 absorption
40
30
Reflectance
20 spectrum of
green
10 vegetation
0
400 800 1,200 1,600 2,000 2,400
λ (nm)
FIGURE 12.2 Typical reflection and absorption spectra of vegetation, including mangroves. (From Purkis, S.
and Klemas, V., Remote Sensing and Global Environmental Change, Wiley-Blackwell, Oxford, U.K., 2011.)
Also, as canopy density decreases, the reflectance spectra of soil and/or water in the gaps between
canopies start intermixing with the mangrove spectral signatures. The variability of spectral reflec-
tance of the mangroves in the NIR region has been found to be higher than in the visible portion
of the spectrum. This may be due to the fact that in the NIR region the internal structure of leaves,
such as size, shape, and distribution of air–water interfaces within the “mesophyll” layer, exerts a
variable influence on the reflectance (Ajithkumar et al. 2008).
Large-scale mangrove mapping has been performed in the past using medium-resolution sat-
ellites, such as Landsat TM and SPOT (Figure 12.3) (Gao 1998; Giri et al. 2007; Guebas 2002;
James et al. 2007; Kovacs et al. 2001; Laba et al. 1997; Long and Skewes 1996; Manson et al. 2001;
Nayak and Bahuguna 2001; Saito et al. 2003). SPOT multispectral data were shown to be suitable
for mapping dense mangroves. Sparse mangroves were less accurately mapped, due to the spectral
interference of surrounding mudflats (Gao 1998). Overall mapping accuracy was sometimes below
85% due to difficulties in detecting small mangrove areas with satellite imagery. Therefore, to
obtain the required high resolution at local scales, researchers have often used airborne multispec-
tral scanners and aerial photography (Green et al. 1998a,b). One advantage of aerial photography
is that historical photos allow for detecting changes that took place well before the availability of
satellite remote sensing.
Some success in discriminating several mangrove physiognomic subclasses using SPOT and
Landsat TM data has been reported (Blasco et al. 1998). However, in most studies, satellite imagery
was used primarily to distinguish mangrove from nonmangrove habitats, without regard to the spe-
cies or condition of the mangroves (Aschbacher et al. 1995; Everitt et al. 1996; Rasolofoharinoro
et al. 1998).
A good example of mangrove mapping using Landsat TM is the USGS/NASA project designed
to determine mangrove cover on a global scale (Giri et al. 2010). The satellite imagery, with a spatial
resolution of 30 m, was used to produce the most comprehensive and accurate data on the extent,
distribution, and decline of mangrove forests available worldwide. More than 1000 Landsat scenes
were analyzed using hybrid supervised and unsupervised digital image classification techniques.
FIGURE 12.3 Portion of Landsat 7 ETM+ image of mangroves and shrimp farms in Cambodia
(December 19, 2013). Mangroves, green; shrimp farms, blue to purple rectangular features. (From U.S.
Geological Survey.)
The time series of the imagery revealed that in the year 2000 only 137,760 km2 of mangroves still
existed. This meant that the remaining area of mangrove forest in the world was less than previ-
ously thought and 12.3% smaller than the estimate by the Food and Agriculture Organization of the
United Nations (FAO). The results also showed that 75% of remaining mangrove forest was found
in just 15 countries, of which only 6.9% is protected under the existing protected areas network.
Based on the produced maps, researchers were able to determine that Asia had 42% of the world’s
mangroves, Africa 21%, North and Central America 15%, Oceania 12%, and South America 11%.
After the launch in 1986 of the first SPOT satellite, operating with a resolution of 20 m in the
multispectral mode and 10 m in the panchromatic mode, the possibility for acquiring more accurate
data for all mangroves of the world significantly improved. This is how the first World Mangrove
Atlas was prepared and published (Blasco et al. 2005; Spalding et al. 1997), followed by the first
worldwide mangrove inventory carried out by the European Community (Aizpuru et al. 2000).
A thematic study of the world’s mangroves from 1980 to 2005 was prepared by the FAO in 2007.
That study was performed in the framework of the Global Forest Resources Assessment 2005 (FAO
2007). The most recent, updated mangrove atlas, the World Atlas of Mangroves, was published
in 2010 (Spalding et al. 2010). This atlas documents 98.6% of all mangrove forests and greatly
supplements the global coverage estimates provided by the FAO atlas of 2007. The atlas shows that
globally mangroves occupy a little over 150,000 km2, span over 123 countries, and contain about
73 species or hybrids.
Green et al. (1998a) proposed resampling the 30 m resolution Landsat TM multispectral bands
using the 10 m spatial resolution SPOT satellite panchromatic band for mangrove mapping. Merging
the two images would create a seven-band image with a spatial resolution of 10 m yet retain the
spectral resolution of the TM sensor. However, this approach has not been used much, especially
since higher-resolution imagery became available.
A good example of the use of medium-resolution satellite imagery and aerial photos is a study
performed along the French Guiana coast to map mangrove changes due to mudbank shifts. Along
the coast of the Guianas, wide mudbanks shift toward the northwest as a consequence of the huge
particulate discharge of the Amazon River. Because of this movement, the coastline is unstable
and continuously changing. Such changes determine the structure and composition of mangrove
forests, the only type of vegetation adapted to this dynamic environment. A combination of aerial
photographs, SPOT satellite images, and field surveys was used to identify coastal changes that took
place over the last 50 years and relate them to natural processes of turnover and replenishment of
mangrove forests (Fromard et al. 2004).
Investigators have studied natural and man-made mangrove changes using remote sensing tech-
niques and various change detection methods in Amazonia, Tanzania, the U.S. Gulf Coast, and
other coastal areas (Cohen and Lara 2003; Wang et al. 2003). For instance, a decision tree method
was used to identify mangrove forest changes in the Pearl River Estuary, China, by integrating
Landsat TM data and ancillary GIS data (Liu et al. 2008).
Remote sensing was widely used to assess the immense devastation of coastal ecosystems in
India and other Indian Ocean countries caused by the tsunami of December 26, 2004. Mangrove
forest ecosystems were the most affected. The Linear Imaging Self-Scanner (LISS) on the Indian
Remote Sensing Satellites (IRS) was used effectively to detect, assess, and monitor changes in
the mangroves for the pre- and post-tsunami periods using a time series of multispectral data. So
many vast pristine tracts of mangrove were destroyed that it may pose a long-term threat to the
region, not only in terms of forest and biodiversity conservation, but also in the ability of the eco-
system to support the livelihoods of coastal communities (Chatterjee et al. 2008; Olwig et al. 2007;
Sirikulchayanon et al. 2008).
Along the U.S. Gulf Coast, hurricanes and oil spills have devastated large areas of mangrove and
other coastal vegetation (Klemas 2009, 2010). During the storms, the vegetation was ripped out and
sand washed in, scouring and damaging the mangrove roots and harming the animals that live there.
To study the impact of the 2010 Gulf of Mexico oil spill and other stressors, such as winter freeze,
erosion, and land subsidence affecting Louisiana mangrove ecosystems, information on the present
condition, historical status, and dynamics of the mangroves was needed. Aerial photographs at 1 m
spatial resolution and 30 m resolution Landsat TM satellite data were used to prepare mangrove
distribution maps from prior to the oil spill (Giri et al. 2011). Image classification was performed
using a decision tree classification approach. Then land cover change maps were prepared for every
2 years from 1984 to 2010 depicting ecosystem shifts (e.g., expansion, retraction, and disappearance).
This new spatial/temporal information was used to assess short-term and long-term impacts of the
oil spill on mangroves (Giri et al. 2011). Spills by oil tanker accidents, illegal tanker cleaning, and
oil leaks from drilling platforms or wells are also causing damage nearly every year to mangrove
ecosystems of Nigeria, the Caribbean, the South China Sea, and other coastal countries and regions.
Before the launch of very-high-resolution satellites, most of the high-resolution studies were
performed using airborne film cameras and airborne multispectral and hyperspectral scanners, such
as the Compact Airborne Spectrographic Imager (CASI) (Tiner 1996). For instance, Green et al.
(1998b) classified mangroves from the Turks and Caicos Islands using CASI, Landsat TM, and
SPOT data. As expected, the classification accuracy of CASI data was higher than Landsat TM
for all classification methods tested. Merging Landsat TM and SPOT XP data improved visual
interpretation of images, but did not enhance discrimination of different mangrove types. Most
importantly, the authors identified the most accurate combinations of sensor and image-processing
methods for mapping the mangroves of the eastern Caribbean islands.
Landsat TM, SPOT, airborne imagers, optical models, and high-resolution field data have also
been used to study and map coral reefs and associated coastal ecosystems (Purkis et al. 2002, 2006;
Purkis and Pasterkamp 2004). In many parts of the world, the health of coral reefs has been declining.
Among the documented impacts on corals are global climate change (e.g., increases in sea-surface
temperature, sea level, CO2 saturation, and frequency and intensity of storms), shifts in water qual-
ity, impacts due to increased loading of sediment, overfishing, contaminants, and nutrients reaching
coastal environments (Purkis et al. 2002). Ideally, mapping submerged aquatic vegetation (SAV),
coral reefs, and general bottom characteristics requires high-resolution (1–4 m) multispectral/
hyperspectral imagery (Mumby and Edwards 2002; Purkis et al. 2002). Aerial hyperspectral scan-
ners and high-resolution multispectral satellite systems, such as IKONOS and QuickBird, have been
used in the past to map SAV with accuracies of about 75% for classes including high-density sea
grass, low-density sea grass, and unvegetated bottom (Akins et al. 2010; Mishra et al. 2006; Wolter
et al. 2005). Advanced analysis methods for hyperspectral imagery should improve SAV and coral
reef mapping (Maeder et al. 2002; Purkis et al. 2008).
ADVANCED REMOTE SENSING TECHNIQUES

Introduction
A major challenge with mangrove remote sensing is the fact that the spectral signal of these plant
communities is strongly influenced by tidal effects and physical properties of the soils. It is also
affected by the physiological status of the plants and morphological properties of each species.
As a result, leaf reflectance alone is often not sufficient for discriminating mangrove species with
medium-resolution multispectral sensor systems, such as Landsat TM or SPOT (Blasco et al. 1998).
To alleviate this problem, researchers are turning to more advanced remote sensing techniques,
including high-spatial-resolution imagers, hyperspectral sensors, LiDAR, and radar.
High-Spatial-Resolution Satellite Techniques

Most mangrove areas are patchy and small and have gaps in the canopy that expose moist soil or
water, depending on the tidal conditions. Therefore, high spatial resolution is required for mapping
them. With the launch of very-high-resolution satellites, such as those in Table 12.1, new opportuni-
ties arose to map mangroves, coral reefs, and other patchy coastal features (Chauvaud et al. 1998;
Jensen et al. 1991; Purkis and Klemas 2011; Wang and Sousa 2010). QuickBird and IKONOS satel-
lites have been in orbit for nearly a decade, providing panchromatic imagery of 0.6 and 1.0 m spatial
resolution and multispectral imagery of 2.4 and 4.0 m resolution, respectively.
High-resolution imagery, such as those from IKONOS and QuickBird, can reduce the number of
mixed pixels, thus improving classification accuracy. QuickBird and IKONOS have blue, green, red,
and NIR spectral bands that are somewhat similar to those of Landsat TM and SPOT. QuickBird and
IKONOS have also had a long mission life and substantial archived imagery, needed for monitoring
mangrove losses and other changes (Al-Tahir et al. 2006; Blasco et al. 2005; Lee and Yeh 2009).
Wang et al. (2004b) compared IKONOS and QuickBird images for mapping mangrove species
on the Caribbean coast of Panama. In their study, they used combinations of spectral bands and
texture information in the higher-resolution panchromatic bands to successfully classify mangrove
species. Results indicate that inclusion of image texture information enhanced classification accu-
racy significantly and both IKONOS and QuickBird images produced promising results in classify-
ing mangrove species. Incorporating texture information in the analysis of IKONOS imagery has
helped improve thematic accuracy of coastal habitats, including mangroves, in other studies as well
(Mumby and Edwards 2002). IKONOS satellite and LAI-2000 field sensors have also been used to
map the mangrove leaf area index (LAI) at the species level (Kovacs et al. 2004, 2005).
A real challenge has been the development of methods for mapping mangrove forests at the
species level using high-spatial-resolution imagery. High-resolution imagery is more sensitive to
within-class spectral variance, making the separation of spectrally mixed land cover types more
250
TABLE 12.1
High-Resolution Satellite Parameters and Spectral Bands
IKONOS QuickBird OrbView-3 WorldView-1 GeoEye-1 WorldView-2
Sponsor Space imaging DigitalGlobe Orbimage DigitalGlobe GeoEye DigitalGlobe
Launched Sept. 1999 Oct. 2001 June 2003 Sept. 2007 Sept. 2008 Oct. 2009
Spatial resolution (m) Panchromatic 1.0 0.61 1.0 0.5 0.41 0.5
Multispectral 4.0 2.44 4.0 n/a 1.65 2
Spectral range (nm) Panchromatic 525–928 450–900 450–900 400–900 450–800 450–800
Coastal blue n/a n/a n/a n/a n/a 400–450
Blue 450–520 450–520 450–520 n/a 450–510 450–510
Green 510–600 520–600 520–600 n/a 510–580 510–580
Yellow n/a n/a n/a n/a n/a 585–625
Red 630–690 630–690 625–695 n/a 655–690 630–690
Red edge n/a n/a n/a n/a n/a 705–745
NIR 760–850 760–890 760–900 n/a 780–920 770–1040
Swath width (km) 11.3 16.5 8 17.6 15.2 16.4
Off nadir pointing (°) ±26 ±30 ±45 ±45 ±30 ±45
Revisit time (days) 2.3–3.4 1–3.5 1.5–3 1.7–3.8 2.1–8.3 1.1–2.7
Orbital altitude (km) 681 450 470 496 681 770
Sources: DigitalGlobe, Quickbird Imagery Products and Product Guide (revision 4), Digital Globe, Inc., Longmont, CO, 2003; Orbimage,
OrbView-3 Satellite and Ground Systems Specifications, Orbimage Inc., Dulles, VA, 2003; Parkinson, C.L., IEEE Trans. Geosci.
Remote Sens., 41, 173, 2003; Space Imaging, IKONOS Imagery Products and Product Guide (version 1.3), Space Imaging LLC,
Thornton, CO, 2003.
Remote Sensing of Wetlands: Applications and Advances
difficult than when using medium-resolution imagery. Therefore, pixel-based techniques are some-
times replaced by object-based image analysis (OBIA) methods. OBIA is a classification technique
that uses objects rather than individual pixels for image analysis. Objects are contiguous pixels that
are grouped on the basis of similar properties through an image segmentation process. Objects
can be created at different levels. For instance, lower-level objects could represent individual tree
crowns; midlevel objects could represent a group of tree crowns of the same species and age; and
high-level objects could represent an entire forest patch (Krause et al. 2004). OBIA methods incor-
porate spatial neighborhood properties by segmenting or partitioning the image into a series of
closed objects that coincide with the actual spatial pattern and then classifying the image. “Region
growing” is among the most commonly used segmentation methods. This procedure starts with the
generation of seed points over the whole scene, followed by the grouping of neighboring pixels into
an object under a specific homogeneity criterion. Thus, the object keeps growing until its spectral
closeness metric exceeds a predefined break-off value (Kelly and Tuxen 2009; Shan and Hussain
2010; Wang et al. 2004).
Wang et al. (2004) developed an integrated pixel-based and object-based method, to classify
the canopies of three mangrove species using IKONOS 1 m panchromatic and 4 m multispectral
images. They wanted to determine if the spectral separability among mangrove species would
be enhanced by taking the object as the basic spatial unit as opposed to the pixel. Three dif-
ferent classification methods were compared: maximum likelihood classification (MLC) at the
pixel level, nearest neighbor (NN) classification at the object level, and a hybrid classification
(MLCNN) that integrates the pixel- and object-based methods. Among the three classification
methods, MLCNN attained the best average accuracy of 91.4%. The team also compared the abil-
ity to discriminate canopies of different mangrove species using various combinations of spectral
and textural information inherent in IKONOS and QuickBird imagery (Wang et al. 2004; Wang
and Sousa 2010).
Multitemporal information can be helpful in discriminating the canopies of different forest spe-
cies (Jensen 2004, 2007). Multiseasonal imagery can facilitate species-level classification of man-
grove forests. Wang and Sousa (2010) found that an IKONOS image acquired during the early rainy
season captured more effectively the difference among mangrove species, than one taken during the
dry season. They attribute this difference to possible phenological and physiological changes that
affect the reflectance of tree canopies. At their study site in Panama, mangroves display new leaves
during the early wet season while experiencing stress from drought and high soil salinity during
the dry season.
Hyperspectral Imaging
More recently, airborne hyperspectral imagers are being used for mapping mangroves. Hyperspectral
imagers may contain hundreds of narrow spectral bands located in the visible, NIR, mid-infrared,
and sometimes thermal portions of the electromagnetic (EM) spectrum (Jensen 2007). For instance,
NASA’s Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) has 224 narrow (10 nm) spec-
tral bands from 400 to 2500 nm, a typical pixel size of 20 m, and swath width of 10.5 km. Airborne
and laboratory hyperspectral data used in mangrove studies have shown that species identification
is possible (Heumann 2011; Kuenzer et al. 2011; Panigrahy et al. 2012; Wang and Sousa-Filho 2009;
Yang et al. 2009). Techniques for optimizing hyperspectral band selection for mangrove mapping
have been developed (Vaiphasa et al. 2007). Hyperspectral imagers offer some operational chal-
lenges, including large data volume, more complex image-processing procedures, and high price of
commercial data (Hirano et al. 2003).
Hyperspectral imaging systems are now available not only for airborne applications but also
in space, such as the satellite-borne Hyperion system, which can detect fine differences in spec-
tral reflectance, potentially enabling species discrimination on a global scale (Blasco et al. 2005;
Heumann 2011; Papes et al. 2010). The Hyperion sensor provides imagery with 220 spectral
bands at a spatial resolution of 30 m. Although there have been few a studies using satellite-based
hyperspectral remote sensing to detect and map mangrove species, results so far have shown that
discrimination between multiple species is possible (Christian and Krishnayya 2009; Vaiphasa
et al. 2005).
The advantages and problems associated with hyperspectral mapping have been clearly dem-
onstrated by Hirano et al. (2003) who used AVIRIS hyperspectral data to map the vegetation for a
portion of the Everglades National Park in Florida. After removing atmospheric effects, the authors
improved the signal-to-noise (S/N) ratio by reducing inherent noise in the image data by employ-
ing the minimum noise fraction (MNF) transform implemented in the ENVI (Research Systems,
Inc.) image-processing software package. Applying principal component analysis (PCA) to the
Everglades AVIRIS image yielded the first 12 bands from the MNF image as containing the most
useful signal information. There was also a need to rectify the image in order to remove geometric
distortions caused by variations in aircraft tilt.
A comparison of the geographic locations of spectrally pure pixels in the AVIRIS image with
dominant vegetation polygons of the Everglades vegetation database identified spectrally pure pix-
els as 10 different vegetation classes, plus water and mud. An adequate number of pure pixels were
identified to permit the selection of training samples used in the automated classification proce-
dure. The spectral signatures from the training samples were then matched to the spectral signa-
tures of each individual pixel. Image classification was undertaken using the ENVI spectral angle
mapper (SAM) classifier in conjunction with the spectral library created for the Everglades study
area. The SAM classifier examines the digital numbers (DNs) of all bands from each pixel in the
AVIRIS dataset to determine similarity between the direction of the spectral signature (i.e., color)
of the image pixel and that of a specific class in the spectral library. A coincident or small spectral
angle between the vector for the unknown pixel and that for a vegetation class training sample
indicates that the image pixel likely belongs to that vegetation class. In the case of spectrally mixed
pixels, the relative probability of membership (based on the spectral angle) to all vegetation classes
is calculated. Mixed pixels are then assigned to the class of the greatest probability of membership
(Hirano et al. 2003).
The hyperspectral data proved effective in discriminating spectral differences among major
Everglades vegetation plants such as red, black, and white mangrove communities and enabled the
detection of exotic invasive species. The overall classification accuracy for all vegetation pixels was
65.7%, with different mangrove tree species ranging from 73.5% to 95.7%. Limited spatial resolu-
tion was a problem, resulting in too many mixed pixels. Another problem was the complexity of
image-processing procedures that are required before the hyperspectral data can be used for auto-
mated classification of wetland vegetation. To obtain improved results, all 224 hyperspectral bands
had to be used in the analysis. The tremendous volume of hyperspectral image data necessitated
the use of specific software packages, large data storage, and extended processing time (Hirano
et al. 2003).
A particular problem with mangrove species discrimination using hyperspectral satellite sensors
is that unless the canopy is very dense, there will be gaps exposing moist soil or water, which will
“contaminate” the pure spectral reflectance signatures of the mangrove. The low spatial resolution
(30 m) of hyperspectral satellite systems, such as Hyperion, will increase the number of mixed
pixels, making classification more difficult. The classifications of wetlands from hyperspectral data
have been shown to be somewhat superior to those from multispectral data, but results of feature
reduction experiments have shown that spatial resolution affects classification accuracy much more
than spectral resolution. Hyperspectral data at 2.6 m resolution have provided the same or poorer
performance than multispectral data at 1 m resolution (Belluco et al. 2006). The higher resolution
also has the advantage of providing a larger number of reference pixels to be used in classifier train-
ing and reducing within-pixel heterogeneity, thus increasing their spectral separability. An overall
balance must be obtained when adopting a higher spatial resolution with a small number of bands
(e.g., four) at the cost of a reduced spectral resolution (Belluco et al. 2006).
Mangrove Canopy Modeling

Over the past two decades, some useful modeling of mangrove canopy structure, distribution, and
reflectivity has been conducted by several investigators. Ramsey and Jensen (1995) have carried
out major studies to model mangrove canopy reflectance using a light interaction model and an
optimization technique. At sites containing mixtures of black, white, and red mangroves, canopy
reflectance spectra were derived from high-resolution data taken from a helicopter. Canopy charac-
teristics were predicted from the canopy reflectance spectra by using measured and estimated data
as inputs into the light–canopy interaction model within an optimization routine. Furthermore, the
authors were able to relate mangrove canopy spectra to site-specific data, using remote sensing
as a tool to describe the spectral and structural changes within and between mangrove species
and community types. High-resolution canopy reflectance spectra, leaf spectra, canopy closure,
height, and species composition data were used in the study. One of the many interesting results
was that 84% of the LAI variance was explained by using the normalized difference vegetation
index (NDVI); however, species composition was not correlated to any combination of reflectance
bands or vegetation index (Ramsey and Jensen 1996). A major challenge all along has been to
obtain an in-depth understanding of the factors affecting the interaction between EM radiation
and wetland vegetation in a particular environment, selecting the best spatial and spectral resolu-
tions and choosing cost-effective processing techniques for discriminating the vegetation types
(Adam et al. 2010).
Radar Applications
Synthetic aperture radar (SAR) can penetrate cloud cover and be used throughout the year in tropi-
cal areas, whereas multispectral optical images can usually be acquired only during dry season
(Lang et al. 2008). SAR imagery from ERS-1, ERS-2, JERS-1, Radarsat-1, and SIR-C has been
successfully used to delineate mangrove extent (Fromard et al. 2004; Lucas et al. 2007; Pasqualini
et al. 1999; Simard et al. 2010). Radarsat images have been used to estimate mangrove wetland
biomass in South China using regression and analytical models (Li et al. 2007). The influence of
canopy structure on radar backscattering of mangrove forests has been investigated (Proisy et al.
2002). One of the challenges seems to be how to suppress the random noise (speckle) associated
with SAR data, which limits the delineation of coastal units and land cover types (Blasco et al.
2005; Mougin et al. 1999; Proisy et al. 2000).
In recent years, the combination of radar and LiDAR has yielded valuable results well beyond
determining mangrove cover alone. Polarimetric and interferometric radar and airborne LiDAR
systems have been used to study the mangrove canopy 3D structure and ecosystem productivity
(Klemas 2013; Kovacs et al. 2008; Kushwaha et al. 2000; Lucas et al. 2007). The LiDAR and radar
data can be used to characterize mangrove canopy structure in three dimensions. The LiDAR pro-
vides an explicit representation of mangrove canopy 3D structure within small areas, and the radar
enables mapping of canopy height at landscape scales (Simard et al. 2006).
The 3D modeling of mangrove forests was significantly advanced by the Shuttle Radar
Topography Mission (SRTM) data. Maps of mean mangrove height and biomass in the Everglades
National Park have been produced using elevation data from the SRTM (Figure 12.4) (Simard et al.
2006). The SRTM data were calibrated using airborne LiDAR data and a high-resolution USGS
digital elevation model (DEM). The mangrove height map had a mean tree height error of 2.0 m
(RMSE) over a single pixel of 30 m. Field data were used to derive a relationship between mean
forest stand height and biomass in order to map the spatial distribution of standing biomass of
mangroves for the entire national park (Simard et al. 2006). The 3D rendition was validated with
airborne LiDAR and field data to provide large-scale canopy height and biomass estimates of man-
grove forests. Estimates of forest biomass provide valuable insights into carbon storage and cycling
in forests (Lang and Kasischke 2008; Simard et al. 2006, 2010).
81°30'W
25°45'N 81°15'W 81°W 80°45'W 80°30'W
25°45'N
25°30'N
25°30'N
Mangrove standing biomass (Mg/ha)
200
175
150
125
100
75
25°15'N
25°15'N
50
25
0
0 10 20
km
81°30'W 81°15'W 81°W 80°45'W 80°30'W
FIGURE 12.4 Mangrove forest biomass map near Everglades National Park in Florida produced using the
elevation data from the SRTM. (From Simard, M. et al., Photogramm. Eng. Remote Sens., 72, 299, 2006.)
Data Fusion Techniques

Data fusion techniques can improve classification accuracy by using different data sources to
increase the dimensionality of available information. In mangrove studies, SAR data have been
processed in combination with multispectral, hyperspectral, or LiDAR data (Lang and McCarty
2008; Ramsey et al. 1998; Waleska et al. 2011). For instance, Waleska et al. (2011) mapped 8 classes
of land cover near the mouth of the Amazon River using supervised classification of Landsat ETM
and a merged ETM–SAR product. The integrated product increased the accuracy and provided
additional information, permitting a more efficient identification and mapping of tropical coastal
wetlands.
Higher classification accuracies for different habitats and mangrove forest types were also
achieved when hyperspectral and radar data were used in combination. Held et al. (2003) combined
a high-spatial/spectral-resolution airborne scanner, the CASI, with NASA’s polarimetric radar,
AIRSAR, for mapping and monitoring mangrove estuaries and investigated difficulties associated
with multisource data integration. The radar provided general structural information in relation to
mangrove zonation, while the high-resolution hyperspectral scanner allowed for finer-detail analy-
sis and green biomass information. MLCs of both the individual and integrated datasets were per-
formed, with the latter producing more accurate results.
Held et al. (2003) explored a hierarchical neural network classification, where the more general
mangrove zones were separated first based on structural information, before species complexes
were extracted using spectral differences. Neural networks have also been used to classify man-
grove species from multiseasonal IKONOS and other high-resolution imagery. Neural networks
were shown to gain discrimination power when textural information was added to the analysis
(Wang et al. 2008; Wang and Sousa 2010).
General Considerations
Some of the more advanced techniques used in mangrove and other coastal habitat remote sensing
are described in several articles (Gao et al. 2004; Heumann 2011; Kuenzer et al. 2011; Phinn et al.
2000; Wang and Sousa 2010). The accuracy of most mangrove maps ranges between 75% and 90%.
Significant progress is being made in distinguishing mangrove stand density and height classes.
According to Blasco et al. (2005), some of the most useful classes that can currently be detected
from space include the following:
1. Dense natural mangroves (multispecific, coverage over 80%, often in protected areas)
2. Degraded mangroves (ground coverage 50%–80%, combined soil/plant reflectance)
3. Fragmented mangroves (ground coverage 25%–50%, moist soils dominate reflectance)
4. Leafless mangroves (low reflectance in NIR due to mortality or disease)
5. Deforested or clear-felled mangroves (opening in mangrove canopy where mangrove pix-
els are replaced by water or mud pixels)
6. Mangrove converted to other uses (mangrove converted to shrimp ponds or paddy fields,
showing very different reflectance)
7. Restored mangrove areas (these monospecific planted stands show strong chlorophyll
absorption and strong response in NIR)
The cost-effectiveness of various remote sensing systems for tropical coastal resource assessment
and management has been studied and is also summarized in several articles (Green et al. 1996;
Mumby et al. 1999). Some of these articles caution that medium-resolution satellite imagery is suit-
able for coarse habitat mapping where overall accuracies of 70% are acceptable but is inadequate
for fine-detail mapping. The high cost of covering large areas with very-high-resolution imagery
is emphasized. It is suggested that large vegetated areas be surveyed with medium-resolution sys-
tems to identify rapidly changing sites. Then only the critical or rapidly changing sites, which often
contain only 5%–10% of the scene pixels, should be imaged at very high resolution (Klemas 2011).
CASE STUDY: OBJECT-BASED MAPPING AND CHANGE

ANALYSIS OF MANGROVES IN SENEGAL
Object-based methods for image analysis possess the advantage of incorporating the spatial con-
text and mutual relationships between objects. In this study, an object-based method was applied
to SPOT XS data to map land cover in the mangrove ecosystem of Low Casamance, Senegal. The
method was also used to analyze changes in mangrove area between 1986 and 2006 (Conchedda
et al. 2007).
The study area of Low Casamance in Southwestern Senegal is the downstream part of the catch-
ment formed by the river Casamance. The estuary is composed of tidal floodplains with a multitude
of mangrove-fringed channels and contains large bare zones (tannes) behind and within mangrove
stands. These occur on slightly higher ground between channels and are halophytes or salt crusts.
Since the 1960s, the rainfall shortage linked to the “Sahel drought” contributed to changes in the
physical/chemical conditions of groundwater and river water. Thus, these mangroves grow in a sys-
tem with low freshwater input and are influenced mainly by seawater (Diop et al. 1997).
The Low Casamance land cover mapping and change detection analysis used 20 m resolution,
multispectral SPOT 1 and 2 images covering a 20-year period from 1986 to 2006. All images
were taken during the dry season and were totally cloud-free. The object-based method used for
this study applied multiresolution segmentation and implemented class-specific rules that incor-
porate spectral properties and relationships between image objects at different hierarchical levels
(Conchedda et al. 2007). The segmentation process divides the image into spatially continuous and
homogeneous regions called image objects.
The software used in this study, Definiens Professional 5, provides a complete tool for OBIA
with various segmentation algorithms (Definiens 2006). The segmentation process aggregates adja-
cent pixels in image objects by considering spectral and shape characteristics throughout a pair-
wise clustering process. Thus, the segments have not only spectral properties but also region-based
metrics such as shape, texture, structure, size, and context. This segmentation represents an optimi-
zation process that produces regions with minimum internal heterogeneity. In each step of the pair-
wise clustering process, image objects are merged. This results in the smallest growth of internal
heterogeneity. Once the smallest growth exceeds the threshold defined by the scale parameters, the
process stops (Benz et al. 2004; Conchedda et al. 2007).
Scale parameters refer to primary object features, such as shape, color, and their weight, in the
definition of heterogeneity. The segmentation can be applied with different scale parameters to form
a hierarchical network of image objects. The relationship between image objects defined at differ-
ent scales can be used for classification. The application of classification algorithms (supervised or
not) in order to image objects offers several advantages compared to pixel-based methods. Within
object-based analyses, spectral, textural, contextual, and scale information can be integrated into
the classification hierarchical rule set or to the classification feature space of supervised classifica-
tion. This information usually increases the accuracy of classification (Benz et al. 2004).
As shown in Figure 12.5, the following land cover classes were mapped: mangroves, salt mud
flats, tannes, herbaceous crops, tree savannah, and tree cover. The tree cover class corresponds
to an open forest of palm trees interspersed by fruit trees. The mapped area of mangroves was
estimated at 76,550 ha in 2006.The classification was validated using reference data from ground
truth points in the field and from points collected from visual inspection of high-resolution
images. The object-based method clearly discriminated the different land cover classes within
the mangrove ecosystem with an overall accuracy of 86%. The overall accuracy of classification
suggests that the spectral and spatial resolution of SPOT data is suitable for mapping mangroves
in the study area.
The object-based approach for change analysis was conducted using the multidate composites
from 1986 to 2006 and applied an NN classifier. The image-to-image, object-based approach to
change analysis captured the fragmented and scattered pattern of change that prevailed in the study
area. In some parts of the test site, there was actually an increase in mangrove area due to the
improved rainfall conditions after the droughts of the 1970s and 1980s. Thus, many of the changes
were natural and not due to exploitation by man (Conchedda et al. 2007). The maps produced as
part of this study represent an important updated reference that supports mangrove management in
Senegal.
Reference points based on visual inspection of the 1986 and 2006 SPOT images were used to
assess the accuracy of the change detection analysis, since no field data were available for 1986. The
user’s accuracy for the change analysis was 85%. The overall accuracy was somewhat lower due to
the method’s difficulties in detecting small areas of change.
This raises the question of how much the accuracy in this case study could have been improved
if higher-resolution imagery had been used. For instance, using IKONOS 1 m panchromatic and
4 m multispectral images, Wang et al. (2004) were able to get average accuracies of 91.4% for map-
ping three different mangrove canopy types and four other cover types at their study sites on the
Senegal
Low Casamance
Herbaceous crops
Tree cover
Mangroves
Salt mud flats
Tree savannah
Kilometers N
10 5 0 Tannes
Water
FIGURE 12.5 Land cover classification from SPOT XS of March 2006 of the Low Casamance Estuary in
Senegal. The mangrove areas are shown in dark green. (From Conchedda, G. et al., Object-based monitor-
ing of land cover changes in mangrove ecosystems of Senegal, The Fourth International Workshop on the
Analysis of Multi-Temporal Remote Sensing Image, Multitemp 2007, Katholieke Universiteit Leuven, Leuven,
Belgium, July 18–20, 2007, IEEE, JRC Publication Nr: JRC37716, pp. 1–6, 2007.)
Caribbean coast of Panama. However, they used a hybrid classification scheme that integrates MLC
at the pixel level and NN classification at the object level. Results showed that the classification of
two spectrally mixed classes, red mangrove and black mangrove canopies, was improved by object-
based classification as compared to pixel-based classification. The integrated classification method
outperformed the MLC and NN methods used individually.
CONCLUSION
Mangrove forests are highly productive ecosystems that dominate the intertidal zone of about
70% of the world’s low wave energy tropical and subtropical coastlines. They help reduce the ero-
sional impact of storms, serve as breeding and feeding grounds for juvenile fish, trap silt that could
smother offshore coral reefs, and cleanse near-shore water by the uptake of nutrients and pollut-
ants. Unfortunately, mangroves are threatened by coastal development projects and climate change,
including accelerating sea level rise. The rapid losses of mangroves are compelling coastal manag-
ers to map and protect the remaining mangrove forests.
Remote sensing plays a crucial role in monitoring mangroves, but the majority of applica-
tions, mainly with airborne and medium-resolution satellite sensors, have been limited to mapping
areal extent and change. Recent advances in remote sensing of mangroves have provided more
effective methods for improving classification accuracy, estimating leaf area, mapping individual
species, measuring canopy height, and mapping mangrove extent at regional and global scales.
New sensors and data analysis techniques include high-resolution satellites, such as IKONOS and
QuickBird, polarimetric SAR (PolSAR), interferometric SAR (InSAR), hyperspectral and LiDAR
systems, as well as OBIA, image texture analysis, and machine-learning algorithms. These new
developments have enabled scientist to produce detailed characterizations of mangrove forests
(Heumann 2010).
Since mangroves often grow in narrow, small patches, fine-spatial-resolution systems, such as
IKONOS or QuickBird, are required to minimize the number of mixed pixels. However, high-
resolution imagery is more sensitive to within-class spectral variance, making separation of spec-
trally mixed land cover types more difficult than when using medium-resolution imagery. Therefore,
pixel-based techniques have sometimes been replaced by object-based methods, which incorporate
spatial neighborhood properties by segmenting or partitioning the image into a series of closed
objects that coincide with the actual spatial pattern and then classifying the image.
Most investigators have found that multispectral bands alone are not good enough to distinguish
mangroves at the species level, since the reflectance data indicate major spectral overlaps among
the species. Airborne hyperspectral imagers have been used successfully to discriminate species
at the local scale. On a global scale, satellite hyperspectral imagery, such as that provided by the
Hyperion satellite, may enable scientists to identify the species, canopy density, and health (stress)
of mangroves and coral reefs in the future. Although hyperspectral classifications are somewhat
superior to those from multispectral data, results have shown that when mapping mangroves, spatial
resolution affects classification accuracy more than spectral resolution.
Data fusion techniques can improve classification accuracy by increasing the dimensionality of
available information. In mangrove studies, SAR data have been processed in combination with
multispectral, hyperspectral, or LiDAR data. Higher classification accuracies for different habi-
tats and mangrove forest types were often achieved when hyperspectral and radar data were used
together. Polarimetric and interferometric radar and airborne LiDAR systems have been used to
study the 3D structure of the mangrove canopy and ecosystem productivity. The 3D modeling of
mangrove forests was made possible over very large areas by the SRTM data.
In the case study, an object-based method was applied to SPOT XS data to map land cover in the
mangrove ecosystem of Low Casamance, Senegal. The maps produced represented an important
updated reference suitable to inform coastal management in Senegal. The object-based method
clearly discriminated the different land cover classes within the mangrove ecosystem with an over-
all accuracy of 86%. The accuracy of classification suggests that the spectral and spatial resolution
of SPOT data is suitable for mapping mangroves in the study area. The object-based mapping
approach used should be valid for similar studies in other regions, especially if high-resolution
imagery is available.
The accuracy of most mangrove maps ranges between 75% and 90%. Up to seven useful man-
grove classes have been detected from space. The high cost of covering large areas with fine-
resolution imagery is prohibitive. Large vegetated areas should be surveyed with medium-resolution
satellites to identify rapidly changing sites. Only rapidly changing or otherwise critical sites, which
often contain only 5%–10% of the scene pixels, should be imaged at very fine resolution.
REFERENCES
Adam, E., O. Mutanga, and D. Rugege. 2010. Multispectral and hyperspectral remote sensing for identification
and mapping of wetland vegetation. Wetlands Ecology and Management 18:281–296.
Aizpuru, M., F. Achard, and F. Blasco. 2000. Global assessment of cover change of the mangrove forest using
satellite imagery at medium to high resolution. In: EEC Research Project No 15017-1999-05 FIED ISP
FR. Joint Research Center, Ispra, Italy.
Ajithkumar, T.T., T. Thangaradjou, and L. Kannan. 2008. Spectral reflectance properties of mangrove species
of the Muthupettai mangrove environment, Tamil Nadu. Journal of Environmental Biology 29:785–788.
Akins, E.R., Y. Wang, and Y. Zhou. 2010. EO-1 advanced land imager data in submerged aquatic vegetation
mapping. In: Remote Sensing of Coastal Environments, Y. Wang, ed., pp. 297–312. New York: CRC Press.
Alongi, D.M. 2002. Present state and future of the world’s mangrove forests. Environmental Conservation
29:331–349.
Alongi, D.M. 2008. Mangrove forests: Resilience, protection from tsunamis, and responses to global climate
change. Estuarine, Coastal and Shelf Science 76:1–13.
Al-Tahir, A., S.M.J. Baban, and B. Ramlal. 2006. Utilizing emerging geo-imaging technologies for the man-
agement of tropical coastal environments. West Indian Journal of Engineering 29:11–22.
Aschbacher, J., R. Ofren, J.-P. Delsol, and T.B. Suselo. 1995. An integrated comparative approach to man-
grove vegetation mapping using advanced remote sensing and GIS technologies: Preliminary results.
Hydrobiologia 295:285–294.
Belluco, E., M. Camuffo, S. Ferrari et al. 2006. Mapping salt marsh vegetation by multispectral and hyperspec-
tral remote sensing. Remote Sensing of the Environment 105:54–67.
Benz, U.C., P. Hofmann, G. Willhauk, I. Lingenfelder, and M. Heynen. 2004. Multi-resolution, object-oriented
fuzzy analysis of remote sensing data for GIS-ready information. Journal of Photogrammetry and
Blasco, F., M. Aizpuru, and D. Din Ndongo. 2005. Mangroves, remote sensing. In: Encyclopedia of Coastal
Science, M.L. Schwartz, ed., pp. 614–617. Dordrecht, the Netherlands: Springer.
Blasco, F., M. Aizpuru, and C. Gers. 2001. Depletion of the mangroves of Continental Asia. Wetlands Ecology
and Management 9:245–256.
Blasco, F., T. Gauquelin, M. Rasolofoharinoro, J. Denis, M. Aizpuru, and V. Caldairou. 1998. Recent advances
in mangrove studies using remote sensing data. Marine and Freshwater Research 49:287–296.
Chatterjee, B., M.C. Porwal, and Y.A. Hussin. 2008. Assessment of tsunami damage to mangrove in India using
remote sensing and GIS. The International Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences 37:283–288 (Part B8, Beijing, 2008).
Chauvaud, S., C. Bouchon, and R. Maniere. 1998. Remote sensing techniques adapted to high resolution map-
ping of tropical coastal marine ecosystems (coral reefs, seagrass beds and mangrove). International
Christian, B. and N.S.R. Krishnayya. 2009. Classification of tropical trees growing in a sanctuary using
Hyperion (EO-1) and SAM algorithm. Current Science 96:1601–1607.
Cohen, M.C. and R.J. Lara. 2003. Temporal changes of mangrove vegetation boundaries in Amazonia:
Application of GIS and remote sensing techniques. Wetlands Ecology and Management 11:223–231.
Conchedda, G., L. Durieux, and P. Mayaux. 2007. Object-based monitoring of land cover changes in mangrove
ecosystems of Senegal. The Fourth International Workshop on the Analysis of Multi-Temporal Remote
Sensing Images. Multitemp 2007, Katholieke Universiteit Leuven, Leuven, Belgium, July 18–20, 2007.
IEEE. JRC Publication Nr: JRC37716. pp. 1–6.
Definiens. 2006. Definiens Professional 5 Users Guide. Document version 5.0.6.1. Munchen, Germany:
Definiens AG.
Diaz, B.M. and G.A. Blackburn. 2003. Remote sensing of mangrove biophysical properties: Evidence from
laboratory simulation of the effects of background variations on spectral vegetation indices. International
Digital Globe. 2003. Quickbird Imagery Products and Product Guide (revision 4). Longmont, CO: Digital
Globe, Inc.
Diop, E.S., A. Soumare, N. Diallo, and A. Guisse. 1997. Recent changes of the mangroves of the Saloum River
Estuary, Senegal. Mangroves and Salt Marshes 1:163–172.
Doyle, T.W., K.W. Krauss, and C.J. Wells. 2009. Landscape analysis and pattern of hurricane impact and circu-
lation in mangrove forests of the Everglades. Wetlands 29:44–53.
Erftemeijer, P.L.A. 2002. A new technique for rapid assessment of mangrove degradation: A case study of
shrimp pond encroachment in Thailand. Trees-Structure and Function 16:204–208.
Everitt, J.H., F.W. Judd, D.E. Escobar, and M.R. Davis. 1996. Integration of remote sensing and spatial infor-
mation technologies for mapping black mangrove on the Texas Gulf coast. Journal of Coastal Research
12:64–69.
Everitt, J.H., C. Yang, K.R. Summy, F.W. Judd, and M.R. Davis. 2007. Evaluation of color-infrared photogra-
phy and digital imagery to map black mangrove on the Texas Gulf coast. Journal of Coastal Research
23:230–235.
FAO. 2007. The world’s mangroves 1980–2005. A thematic study prepared in the framework of the Global
Forest Resources Assessment 2005. FAO Forestry Paper 153. Food and Agriculture Organization of the
United Nations, Rome, Italy.
Florida, D.N.R. 1987. Florida’s Mangroves: Walking Trees, pp. 1–8. St. Petersburg, FL: Florida Department of
Natural Resources. Bureau of Marine Research Publication.
Fromard, F., C. Vega, and C. Proisy. 2004. Half a century of dynamic coastal change affecting mangrove shore-
lines of French Guiana. A case study based on remote sensing data analysis and field surveys. Marine
Geology 2008:265–280.
Gao, J. 1998. A hybrid method toward accurate mapping of mangroves in a marginal habitat from SPOT mul-
tispectral data. International Journal of Remote Sensing 19:1887–1899.
Gao, J., H. Chen, Y. Zhang, and Y. Zha. 2004. Knowledge-based approaches to accurate mapping of mangroves
from satellite data. Photogrammetric Engineering and Remote Sensing 70:1241–1248.
Gilman, E.L., J. Ellison, N.C. Duke, and C. Field. 2008. Threats to mangroves from climate change and adapta-
tion options: A review. Aquatic Botany 89:237–250.
Giri, C., J. Long, and L. Tieszen. 2011. Mapping and monitoring Louisiana’s mangroves in the aftermath of the
2010 Gulf of Mexico oil spill. Journal of Coastal Research 27(6):1059–1064.
Giri, C., E. Ochieng, L.L. Tieszen et al. 2010. Status and distribution of mangrove forests of the world using
earth observation satellite data. Global Ecology and Biogeography 20(1):154–159. DOI: 10.1111/j.1466-
8238.2010.00584.x. (Accessed April 7, 2011)
Giri, C., B. Pengra, A. Singh, and L.L. Tieszen. 2007. Monitoring mangrove forest dynamics of the Sundarbans
in Bangladesh and India using multi-temporal satellite data from 1973 to 2000. Estuarine Coastal and
Shelf Science 73:91–100.
Green, E.P., C.D. Clark, P.J. Mumby, A.J. Edwards, and A.C. Ellis. 1998a. Remote sensing techniques for man-
grove mapping. International Journal of Remote Sensing 19:935–956.
Green, E.P., P.J. Mumby, A.J. Edwards, and C.D. Clark. 1996. A review of remote sensing for tropical coastal
resources assessment and management. Coastal Management 24:1–40.
Green, E.P., P.J. Mumby, A.J. Edwards, C.D. Clark, and A.C. Ellis. 1998b. The assessment of mangrove areas
using high resolution multispectral airborne imagery (CASI). Journal of Coastal Research 14:433–443.
Guanawardena, M. and J.S. Rowan. 2005. Economic valuation of a mangrove ecosystem threatened by shrimp
aquaculture in Sri Lanka. Environmental Management 36:535–550.
Guebas, F.D. 2002. The use of remote sensing and GIS in the sustainable management of tropical coastal eco-
systems. Environment Development and Sustainability 4:93–112.
Held, A., C. Ticehurst, L. Lymburner, and N. Williams. 2003. High resolution mapping of tropical mangrove
ecosystems using hyperspectral and radar remote sensing. International Journal of Remote Sensing
24:2739–2760.
Wetlands 23:436–448.
Hogarth, P. 2007. The Biology of Mangroves and Seagrasses. Oxford, U.K.: Oxford University Press.
IPCC. 2001. Climate Change 2001—The Scientific Basis. Cambridge, U.K.: Cambridge University Press.
James, G.K., J.O. Adegoke, E. Saba, P. Nwilo, and J. Akinyede. 2007. Satellite-based assessment of the extent
and changes of the mangrove ecosystem of the Niger Delta. Marine Geodesy 30:249–267.
Jensen, J.R. 2004. Introductory Digital Image Processing: A Remote Sensing Perspective, 3rd edn. Upper
Saddle River, NJ: Prentice-Hall.
Jensen, J.R. 2007. Remote Sensing of the Environment: An Earth Resource Perspective. Upper Saddle River,
NJ: Prentice-Hall.
Jensen, J.R., H. Lin, X. Yang, E. Ramsey, B.A. Davis, and C.W. Thoemke. 1991. The measurement of mangrove
characteristics in Southwest Florida using SPOT multispectral data. Geocarto International 2:13–21.
Kamaruzaman, J. and I. Kasawani. 2007. Imaging spectrometry on mangrove species identification and map-
ping in Malaysia. WSEAS Transactions on Biology and Biomedicine 4:118–126.
Kelly, M. and K. Tuxen. 2009. Remote sensing support for tidal wetland vegetation research and management.
In: Remote Sensing and Geospatial Technologies for Coastal Ecosystem Assessment and Management,
X. Yang, ed. Berlin, Germany: Springer-Verlag.
Klemas, V. 2009. The role of remote sensing in predicting and determining coastal storm impacts. Journal of
Coastal Research 25:1264–1275.
Klemas, V. 2010. Tracking oils slicks and predicting their trajectories using remote sensors and models:
Case studies of the Sea Princess and Deepwater Horizon oil spills. Journal of Coastal Research
26:789–797.
Klemas, V. 2013. Remote sensing of wetland biomass: An overview. Journal of Coastal Research
29:1016–1028.
Kovacs, J.M., F. Flores-Verdugo, J.F. Wang, and L.P. Aspden. 2004. Estimating leaf area index of a degraded
mangrove forest using high spatial resolution satellite data. Aquatic Botany 80:13–22.
Kovacs, J.M., C.V. Vandenberg, and F. Flores-Verdugo. 2008. The use of multipolarised spaceborne SAR
backscatter for monitoring the health of a degraded mangrove forest. Journal of Coastal Research
24:248–254.
Kovacs, J.M., J. Wang, and M. Blanco-Correa. 2001. Mapping disturbances in a mangrove forest using multi-
date Landsat TM imagery. Environmental Management 27:763–776.
Kovacs, J.M., J. Wang, and F. Flores-Verdugo. 2005. Mapping mangrove leaf area index at the species level
using IKONOS and LAI-2000 sensors for the Agua Brava Lagoon, Mexican Pacific. Estuarine, Coastal
and Shelf Science 62:377–384.
Krause, G., M. Bock, S. Weiers, and G. Braun. 2004. Mapping land-cover and mangrove structures with remote
sensing techniques: A contribution to a synoptic GIS in support of coastal management in North Brazil.
Kuenzer, C., A. Bluemel, S. Gebhardt, T. Vo Quoc, and S. Dech. 2011. Remote sensing of mangrove ecosys-
tems. Remote Sensing 3:878–928.
Kushwaha, S.P.S., R.S. Dwivedi, and B.R. Rao. 2000. Evaluation of various digital image processing tech-
niques for detection of coastal wetlands using ERS-1 SAR data. International Journal of Remote Sensing
21:565–579.
Laba, M., S.D. Smoth, and S.D. Degloria. 1997. Landsat-based landcover mapping in the lower Yuna River
watershed in the Dominican Republic. International Journal of Remote Sensing 18:3011–3025.
Lang, M., G. McCarty, and M. Anderson. 2008. Monitoring wetland hydrology at a watershed Scale: Dynamic
information for adaptive management. Journal of Soil and Water Conservation 63:49A.
Lang, M.W. and E.S. Kasischke. 2008. Using C-band synthetic aperture radar data to monitor forested wetland
46:535–546.
Lang, M.W. and G.W. McCarty. 2008. Remote sensing data for regional wetland mapping in the United States:
Trends and future prospects. In: Wetlands: Ecology, Conservation and Restoration, R.E. Russo, ed.
Hauppauge, NY: Nova Science Publishers, Inc.
Lee, T.M. and H.C. Yeh. 2009. Applying remote sensing techniques to monitor shifting wetland vegeta-
tion: A case study of Danshui River estuary mangrove communities, Taiwan. Ecological Engineering
35:487–496.
Li, X., A.G.O. Yeh, S. Wang, K. Liu, X. Liu, and J. Qian. 2007. Regression and analytical models for estimat-
ing mangrove wetland biomass in South China using Radarsat images. International Journal of Remote
Sensing 28:5567–5582.
Liu, K., X. Li, and S.G. Wang. 2008. Monitoring mangrove forest changes using remote sensing and GIS data
with decision-tree learning. Wetlands 28:336–346.
Long, B.G. and T.D. Skewes. 1996. A technique for mapping mangroves with Landsat TM satellite data and
Geographic Information System. Estuarine Coastal and Shelf Science 43:373–381.
Lucas, R.M., A.L. Mitchell, A. Rosenqvist, C. Proisy, I. Melius, and C. Ticehurst. 2007. The potential of
L-band SAR for quantifying mangrove characteristics and change: Case studies from the tropics. Aquatic
Conservation: Marine and Freshwater Ecosystems 17:245–264.
Maeder, J., S. Narumalani, D. Rundquist et al. 2002. Classifying and mapping general coral-reef structure using
Ikonos data. Photogrammetric Engineering and Remote Sensing 68:1297–1305.
Manson, F.J., N.R. Loneragan, I.M. McLeod, and R.A. Kenyon. 2001. Assessing techniques for estimating
the extent of mangroves: Topographic maps, aerial photographs, and Landsat TM images. Marine and
Freshwater Research 52:787–792.
McInnes, K.L., K.J.E. Walsh, G.D. Hubbert, and T. Beer. 2003. Impact of sea-level rise and storm surge on a
coastal community. Natural Hazards 30:187–207.
Mishra, D., S. Narumalani, D. Rundquist, and M. Lawson. 2006. Benthic habitat mapping in tropical marine
environments using QuickBird multispectral data. Photogrammetric Engineering and Remote Sensing
72:1037–1048.
Mitsch, W.J. and J.G. Gosselink. 2007. Wetlands. Hoboken, NJ: John Wiley & Sons, Inc.
Mougin, E., C. Proisy, G. Marty et al. 1999. Multifrequency and multipolarization radar backscattering from
mangrove forests. IEEE Transactions on Geoscience and Remote Sensing 34:94–102.
Mumby, P.J. and A.J. Edwards. 2002. Mapping marine environments with IKONOS imagery: Enhanced spatial
resolution can deliver greater thematic accuracy. Remote Sensing of the Environment 2:248–257.
Mumby, P.J., E.P. Green, A.J. Edwards, and C.D. Clark. 1999. The cost-effectiveness of remote sensing for trop-
ical coastal resources assessment and management. Journal of Environmental Management 55:157–166.
Nayak, S. and A. Bahauguna. 2001. Application of remote sensing data to monitor mangroves and other coastal
vegetation in India. Indian Journal of Marine Sciences 30:195–213.
Nfotabong-Atheull, A., N. Din, and F. Dahdouh-Guebas. 2013. Qualitative and quantitative characterization of
mangrove vegetation structure and dynamics in a peri-urban setting of Douala (Cameroon): An approach
using airborne imagery. Estuaries and Coasts 36:1181–1192.
Odum, E.P. 1993. Ecology and Our Endangered Life-Support Systems. 2nd edn. Sunderland, MA: Sinauer
Associates, Inc.
Olwig, M.F., M.K. Sorensen, M.S. Rasmussen, F. Danielsen, V. Selvam, and L.B. Hansen. 2007. Using remote
sensing to assess the protective role of coastal woody vegetation against tsunami waves. International
Orbimage. 2003. OrbView-3 Satellite and Ground Systems Specifications. Dulles, VA: Orbimage Inc.
Panigrahy, S., T. Kumar, and K.R. Manyunath. 2012. Hyperspectral leaf signature as an added dimension
for species discrimination: Case study of four tropical mangroves. Wetlands Ecology and Management
20(2):101–110. DOI 10.1007/s11273-011-9245-z.
Pannier, F. 1979. Mangroves impacted by human-induced disturbances: A case study of the Orinoco Delta
mangrove ecosystem. Environmental Management 3:205–216.
Papes, M., R. Tupayachi, P. Martinez, A.T. Peterson, and G.V.N. Powell. 2010. Using hyperspectral satellite
imagery for regional inventories: A test with tropical emergent trees in the Amazon Basin. Journal of
Vegetation Science 21:342–354.
Parkinson, C.L. 2003. Aqua: An Earth-observing satellite mission to examine water and other climate variables.
IEEE Transactions on Geoscience and Remote Sensing 41:173–183.
Pasqualini, V., J. Iltis, N. Dessay, M. Lointier, O. Guelorget, and L. Polidori. 1999. Mangrove mapping in north-
western Madagascar using SPOT-XS and SIR-C radar data. Hydrobiologia 413:127–133.
Phinn, S.R., C. Menges, G.J.E. Hill, and M. Stanford. 2000. Optimizing remotely sensed solutions for moni-
toring, modeling and managing coastal environments. Remote Sensing of the Environment 73:117–132.
Pinet, P.R. 2009. Invitation to Oceanography, 5th edn. Sudbury, MA: Jones & Bartlett.
Proisy, C., E. Mougin, F. Fromard, and M.A. Karam. 2000. Interpretation of polarimetric radar signatures of
mangrove forests. Remote Sensing of the Environment 71:56–66.
Proisy, C., E. Mougin, F. Fromard, V. Trichon, and M.A. Karam. 2002. On the influence of canopy structure on
the radar backscattering of mangrove forests. International Journal of Remote Sensing 23:4197–4210.
Purkis, S.J., N.A.J. Graham, and B.M. Riegl. 2008. Predictability of reef fish diversity and abundance using
remote sensing data in Diego Garcia (Chagos Archipelago). Coral Reefs 27:167–178.
Purkis, S.J., J.A.M. Kenter, E.K. Oikonomou, and I.S. Robinson. 2002. High-resolution ground verification,
cluster analysis and optical model of reef substrate coverage on Landsat TM imagery (Red Sea, Egypt).
International Journal of Remote Sensing 23:1677–1698.
Purkis, S.J. and V. Klemas. 2011. Remote Sensing and Global Environmental Change. Oxford, U.K.:
Wiley-Blackwell.
Purkis, S.J., S. Myint, and B. Riegl. 2006. Enhanced detection of the coral Acropora cervicornis from satellite
imagery using a textural operator. Remote Sensing of the Environment 101:82–94.
Purkis, S.J. and R. Pasterkamp. 2004. Integrating in situ reef-top reflectance spectra with Landsat TM imagery
to aid shallow-tropical benthic habitat mapping. Coral Reefs 23:5–20.
Ramsey, E.W. and J.R. Jensen. 1995. Modeling mangrove canopy reflectance using a light interaction model
and an optimization technique. In: Wetland and Environmental Applications of GIS, J.G. Lyon, and
J. McCarthy, eds., pp. 61–81. Boca Raton, FL: Lewis Publishers.
Ramsey, E.W. and J.R. Jensen. 1996. Remote sensing of mangrove wetlands: Relating canopy spectra to site-
specific data. Photogrammetric Engineering and Remote Sensing 62:939–948.
Ramsey, E.W., G.A. Nelson, and S.K. Sapkota. 1998. Classifying coastal resources by integrating optical and
radar imagery and color infrared photography. Mangroves and Salt Marshes 2:109–119.
Rasolofoharinoro, M., F. Blasco, M.F. Bellan, M. Aizpuru, T. Gauquelin, and J. Denis. 1998. A remote sens-
ing based methodology for mangrove studies in Madagascar. International Journal of Remote Sensing
19:1873–1886.
Saito, H., M.F. Bellan, A. Al-Habshi, M. Aizpuru, and F. Blasco. 2003. Mangrove research and coastal eco-
system studies with SPOT-4 HRVIR and TERRA ASTER in the Arabian Gulf. International Journal of
Shan, J. and E. Hussain. 2010. Object-based data integration and classification for high-resolution coastal map-
ping. In: Remote Sensing of Coastal Environment, J. Wang, ed. Boca Raton, FL: CRC Press.
Simard, M., L.E. Fatoyinbo, and N. Pinto. 2010. Mangrove canopy 3D structure and ecosystem productivity
using active remote sensing. In: Remote Sensing of Coastal Environments, Y. Wang, ed., pp. 61–78. Boca
Raton, FL: CRC Press, Taylor & Francis Group, LLC.
Simard, M., K. Zhang, V.H. Rivera-Monroy et al. 2006. Mapping height and biomass of mangrove forests in
Everglades National Park with SRTM elevation data. Photogrammetric Engineering and Remote Sensing
72:299–311.
Sirikulchayanon, P., W.X. Sun, and T.J. Oyana. 2008. Assessing the impact of the 2004 tsunami on mangroves
using remote sensing and GIS techniques. International Journal of Remote Sensing 29:3553–3576.
Space Imaging. 2003. IKONOS Imagery Products and Product Guide (version 1.3). Thornton, CO: Space
Imaging LLC.
Spalding, M., F. Blasco, and C. Field. 1997. World Mangrove Atlas. Okinawa, Japan: ISME.
Spalding, M., M. Kainuma, and L. Collins. 2010. World Atlas of Mangroves. London, U.K.: Earthscan.
Terchunian, A. and V. Klemas. 1986. Mangrove mapping in Ecuador: The impact of shrimp pond construction.
Tiner, R.W. 1996. Wetlands. In: Manual of Photographic Interpretation, 2nd edn. Falls Church, VA: American
Society for Photogrammetry and Remote Sensing.
Tiner, R.W., H.C. Bergquist, G.P. DeAlessio, and M.J. Starr. 2002. Geographically Isolated Wetlands: A
Preliminary Assessment of their Characteristics and Status in Selected Areas of the United States.
Hadley, MA: U.S. Department of the Interior, Fish and Wildlife Service, Northeast Region.
Tomlinson, P.B. 1986. The Botany of Mangroves. Cambridge, U.K.: Cambridge University Press.
Vaiphasa, C., S. Ongsomwang, T. Vaiphasa, and A.K. Skidmore. 2005. Tropical mangrove species discrimina-
tion using hyperspectral data: A laboratory study. Estuarine, Coastal and Shelf Science 65:371–379.
Vaiphasa, C., A.K. Skidmore, W.F. de Boer, and T. Vaiphasa. 2007. A hyperspectral band selector for plant spe-
cies discrimination. ISPRS Journal of Photogrammetry and Remote Sensing 62:255–235.
Waleska, S., P. Rodrigues, P. Walfir, and M. Souza-Filho. 2011. Use of multi-sensor date to identify and map
tropical coastal wetlands in the Amazon of Northern Brazil. Wetlands 31:11–23.
Wang, L., J.L. Silvan-Cardenas, and W.P. Sousa. 2008. Neural Network classification of mangrove species from
multi-seasonal Ikonos imagery. Photogrammetric Engineering and Remote Sensing 74:921–927.
Wang, L. and W.P. Sousa-Filho. 2009. Distinguishing mangrove species with laboratory measurements of
hyperspectral leaf reflectance. International Journal of Remote Sensing 30:1267–1281.
Wang, L. and W.P. Sousa. 2010. Remote sensing of coastal mangrove forest. In: Remote Sensing and Geospatial
Technologies for Coastal Ecosystem Assessment and Management, X. Yang, ed., pp. 323–340. Berlin,
Germany: Springer-Verlag.
Wang, L., W.P. Sousa, and P. Gong. 2004a. Integration of object-based and pixel-based classification for map-
ping mangroves with Ikonos imagery. International Journal of Remote Sensing 25:5655–5668.
Wang, L., W.P. Sousa, P. Gong, and G.S. Biging. 2004b. Comparison of IKONOS and QuickBird images
for mapping mangrove species on the Caribbean coast of Panama. Remote Sensing of the Environment
91:432–440.
Wang, Y., G. Bonynge, J. Nugranad et al. 2003. Remote sensing of mangrove change along the Tanzania coast.
Marine Geodesy 26:35–48.
Wolter, P.T., C.A. Johnston, and G.J. Niemi. 2005. Mapping submerged aquatic vegetation in the US Great
Lakes using Quickbird satellite data. International Journal of Remote Sensing 26:5255–5274.
Yang, C.H., J.H. Everitt, R.S. Fletcher, R.R. Jensen, and P.W. Mausel. 2009. Evaluating AISA plus hyper-
spectral imagery for mapping black mangrove along the South Texas Gulf Coast. Photogrammetric
13 Tidal Marsh Classification
Approaches and Future Marsh
Migration Mapping Methods
for Long Island Sound,
Connecticut, and New York
Mark Hoover and Adam Walton Whelchel
CONTENTS
Introduction..................................................................................................................................... 265
Current Tidal Marsh Classification Techniques and Methods........................................................266
Classification Process.................................................................................................................266
Image Segmentation.............................................................................................................. 267
Training Data......................................................................................................................... 268
Rule Development and Implementation................................................................................ 268
Manual Edit........................................................................................................................... 269
Accuracy Assessment............................................................................................................ 269
Future Tidal Marsh Migration Mapping Techniques and Methods................................................ 270
Development of Tidal Marsh Migration Tool—Inputs.............................................................. 271
Digital Elevation Models....................................................................................................... 271
Image Classification.............................................................................................................. 272
Current Tide Levels............................................................................................................... 272
Current Accretion Rates........................................................................................................ 274
Model Process....................................................................................................................... 274
Results............................................................................................................................................. 275
Suggested Applications.............................................................................................................. 278
Limitations of Study...................................................................................................................280
Resource Requirements: Equipment, Software, and Expertise..................................................280
References....................................................................................................................................... 281
INTRODUCTION
Tidal marshes are unique ecosystems that provide essential habitat and protection for wildlife and
people, respectively (Titus 1988; Dreyer and Niering 1995; NECIA 2007; Tiner 2013). Historically,
tidal marshes have been poorly managed and viewed as expendable resulting in loss and degradation
in the United States. In the State of Connecticut, 30% of the coastal wetlands have been destroyed
through human activities (Dreyer and Niering 1995) and virtually all remaining areas were ditched
265
to control mosquito populations (Tiner et al. 2013). Since the passage of the Connecticut Tidal
Wetlands Act of 1969 and New York Tidal Wetlands Act of 1973, there has been a concerted effort
to protect existing tidal marshes and mitigate for unavoidable losses. Although many of the historic
risks to marshes from development have been reduced through regulation, a growing threat has
emerged. Accelerated rates of sea level rise are now having a detrimental effect on tidal marsh
ecosystems (Titus 1988; Warren and Niering 1993; Nicholls et al. 1999; Morris et al. 2002; Cooper
et al. 2005; Slovinksy and Dickson 2006; Tiner 2013). To mitigate this impact, an accurate classifi-
cation of existing marsh habitat and projections of future marsh conditions under various sea level
rise scenarios are needed.
Presented here is a case study for the Connecticut and New York coastlines on the Long Island
Sound that will (1) illustrate an advanced approach to classifying coastal wetland vegetation to
assess the current distribution and abundance of tidal marsh habitat, (2) provide mapping techniques
and methods to assess future marsh habitat, and (3) link present classification and future mapping
approaches to management implications in coastal Connecticut.
CURRENT TIDAL MARSH CLASSIFICATION TECHNIQUES AND METHODS

Classification of coastal wetland vegetation requires high-resolution imagery, which when employ-
ing computer-assisted, pixel-based spectral classifiers can create many errors. These errors can be
attributed to high-resolution imagery generally having fewer spectral bands, greater spectral varia-
tion between classes, and increased shadow abundance (Laliberte et al. 2004; Xiaoxia et al. 2005).
Pixel-based classifications can also result in “salt and pepper” outputs because of the size of the unit
classified (Xiaoxia et al. 2005).
To overcome shortfalls associated with pixel-based classifiers applied to high-resolution imag-
ery, an object-oriented approach to image classification has been developed. Object-oriented clas-
sification segments an image into meaningful objects prior to classification (Laliberte et al. 2004;
Xiaoxia et al. 2005; Platt and Rapoza 2008). The classification of the objects can be based on
spectral information as well as textural, spatial, contextual information and ancillary data, such as
elevation (Laliberte et al. 2004; Xiaoxia et al. 2005).
Several studies have found greater accuracy with object-oriented classification over traditional
pixel-based approaches. Platt and Rapoza (2008) compared classification strategies on a 4 m, four-
band IKONOS image. They classified the image into seven categories and found that the object-
oriented classification produced 14% more accurate results (78% versus 64%). Xiaoxia et al. (2005)
compared classification strategies using QuickBird imagery. The object-oriented approach led to a
20% increase in accuracy (83% versus 63%). The increased accuracy can be attributed both to clas-
sifying objects instead of individual pixels and increased use of ancillary data in the classification
process.
Classification Process
The implementation of an object-oriented approach to tidal marsh classification involved five steps:
image segmentation, selection of training data, rule development and implementation, manual cor-
rection, and accuracy assessment. For the purpose of processing power and to preserve similar
marsh characteristics and reflectance properties, the study region was divided into 40 individually
classified sites.
Two classifications were created. The first broke the image into broad classifications of “urban,”
“agrigrass” (a combined agricultural and grassland category), “forest,” “wetland,” and “water.”
This classification was developed as a first cut to classify the upland habitat for use in the future
marsh migration model discussed in Section, Development of Tidal Marsh Migration Tool–Inputs.
Tidal Marsh Classification Approaches 267
Elevation (m)
High=4.5
Low=0.0
(a) (b)
FIGURE 13.1 The CIR imagery (a) flown in 2005 and DEM (b) derived from LiDAR collected in 2006 for
the Barn Island Wildlife Management Area in Stonington, CT. These were the primary data sources for the
object-oriented land cover classification.
The five-step classification process was then executed again for the areas initially classified as “wet-
land” and “water” to more specifically classify areas of “low marsh,” “high marsh,” “iva” (Iva fru-
tescens, marsh elder), “phrag” (Phragmites australis, common reed), “sand,” and “water.” We will
focus on the more specific classification, but the same procedure was implemented for both.
A 2 ft digital elevation model (DEM) and a digital, 1 ft, color-infrared (CIR) imagery were
used in the classification process (Figure 13.1). The DEM was interpolated from 3 ft posting light
detection and ranging (LiDAR) data collected in 2006. The CIR imagery was obtained from the
Long Island Sound Resource Center website (Long Island Sound Resource Center, 2005). The 2005
leaf-on CIR imagery contained near-infrared (NIR), red, and green bands. A normalized difference
vegetation index (NDVI) layer was also developed from the CIR imagery.
Image Segmentation
The five layers (DEM, NDVI, NIR, red, and green) were then loaded into eCognition (now
Definiens Developer, an object-oriented image classification software). A multiresolution seg-
mentation was then initiated to create polygons with homogenous characteristics (Figure 13.2).
This was an iterative step as different parameters, such as scale parameter, color, shape, smooth-
ness, and compactness, could be altered to create differently sized and shaped polygons. The
color, shape, smoothness, and compactness values were kept at the defaults (0.8, 0.2, 0.9, and
0.1, respectively) with the scale parameter adjusted for each site. The scale parameter was set
to create the largest possible polygons that enable detection of differences between the broad
categories of “low marsh,” “high marsh,” “iva,” “phrag,” “sand,” and “water.” The segmentation
divided the image into polygons that had similar characteristics in the five layers loaded (i.e.,
similar elevation and reflectance in the other four bands). The attributes selected were deemed
the most useful spectral and textural characteristics for image classification: mean, standard
FIGURE 13.2 The multiresolution segmentation of the tidal marsh created in eCognition. The polygons cre-
ated represent areas with similar spectral reflectance, NDVI, and elevation.
deviation, and GLCM homogeneity texture characteristic for all five layers. The attribute values
were calculated using the values of all the pixels in an individual polygon.
Training Data
Each marsh was then visited to collect in situ vegetation data for development of training data used
in the classification process. A GPS camera and printouts of the CIR imagery with overlaying seg-
mented polygons were taken to assist with data collection. The vegetation was noted on selected
polygons, and GPS-referenced digital photographs were taken to confirm the location of the spe-
cies. The areas were labeled as “low marsh” if they contained either the tall or short form Spartina
alterniflora. If they contained Iva, they were labeled “iva” and similarly areas of Phragmites labeled
“phrag.” If the area contained any other plant species including S. patens, Juncus gerardii, Distichlis
spicata, it was labeled “high marsh” (Figure 13.3).
Rule Development and Implementation

The labeled polygons were then loaded into See5 to create a classification hierarchy. See5 is a data
mining tool that extracts patterns from categories to predict future situations. Once patterns are
identified, classifiers are created and expressed as decision trees. These decision trees are a series of
if-then rules and were used to classify the rest of the image. The following is an example of a See5
decision tree (Figure 13.4). The classification tree was then entered as rules in eCognition. The rules
were then executed to classify all the segmented polygons in the image.
High marsh
Urban
Forest
Iva
Low marsh
FIGURE 13.3 Training areas for tidal marsh classification based on in situ vegetation collection. These
training areas were then used in See5, a data mining tool, to classify the rest of the image.
Manual Edit
After the classification was run in eCognition, the results were then exported and examined with
the CIR imagery individually to ensure correct classification. Areas that were clearly classified
incorrectly were manually corrected by editing the attribute table of the polygons. The result, when
combined with the upland classification, provided a finished classification for the study region
(Figure 13.5).
Accuracy Assessment
An accuracy assessment was conducted for each of the classifications. As previously mentioned, the
coastline was divided into 40 sites based on geographic extent, marsh similarity, and tidal charac-
teristics. For each site, 200 random points were generated. One hundred points were selected outside
of the marsh, in the “urban,” “agrigrass,” “forest,” and “water” classes, using an equalized random
selection process. The other 100 points were selected within the marsh, using a stratified random
process with a minimum of 5 points per each class (Figure 13.6). For the upland classes (“forest,”
“agrigrass,” and “urban”), each point was assigned a cover class name based on the interpretation
of the CIR imagery. For the 100 points inside of each marsh, each point was visited in the field. The
points were found by using printouts of the CIR imagery, with the points overlain. The GPS camera
was used to document, visually and spatially, each field-checked reference point.
Once the 200 points were given their correct identification from either imagery interpretation or
field visit, an error matrix was constructed for each marsh (Table 13.1). The error matrix displayed
FIGURE 13.4 An example of a See5 output classification tree. This tree was created by using the training
data collected in the field. The tree was then entered in eCognition as rules to classify the remaining study
region.
the classification (i.e., map data) and correct identification (i.e., reference data) for each point. The
error matrix also indicates the overall accuracy of the classification. The overall accuracy of the
40 sites ranged from 71% to 95% with an average of 85.1%.
FUTURE TIDAL MARSH MIGRATION MAPPING TECHNIQUES AND METHODS

One of the greatest challenges for coastal resource practitioners is managing tidal marsh migration
in the twenty-first century. The key concern is whether or not future habitat, both spatially and func-
tionally, will compensate for expected conversion of existing habitat to open water due to rising sea
level. As the degree, duration, and extent of inundation change, tidal marshes have three functional
responses: convert in place, keep pace, and/or advance inland. Historically, tidal marshes have,
through accretion, increased the marsh plain elevation to keep pace with sea level rise (Titus 1988;
Bricker-Urso et al. 1989; Dreyer and Niering 1995; Nydick et al. 1995; NECIA 2007). The projected
rates of sea level rise for this century, however, are predicted to be greater than recent historic rates
and will simply prevent many tidal marshes from keeping pace. Tidal marshes can advance as
higher sea levels move tides further inland creating opportunities for marsh to expand into low-lying
uplands or neighboring freshwater wetlands. Marshes unable to accrete and/or expand inland could
become permanently inundated leading to further wetland loss (Brinson et al. 1995; Tiner 2013). It
therefore becomes an imperative for resource managers to understand which marshes are most vul-
nerable to changes in sea level as well as where adjoining land use and topography are suitable for
marsh migration, so that optimal immediate and longer-term adaptation, mitigation, and restoration
efforts can be prioritized and implemented.
To effectively sustain these critical natural resources along the coast of Long Island Sound,
managers need more sophisticated tidal marsh classification techniques coupled with the ability to
Agrigrass
Forest
High marsh
Iva
Low marsh
Phrag
Sand
Urban
Water
FIGURE 13.5 Finished classification for a section of Barn Island Wildlife Management Area. The classifica-
tion for the entire site had an overall accuracy of 75%.
map future landscape position and extent. In order to project future tidal marsh advancement under
various sea level rise scenarios, high-resolution remote sensing data along with site-specific in situ
data were used across the study region to develop a marsh migration tool in ArcGIS. The tool uses
elevation data derived from LiDAR to simulate future inundation. Simulated inundations along
with aerial photographs of current marsh condition and in situ vegetation and accretion data enable
the tool to project tidal marsh migration, marsh loss through conversion, and the redistribution of
vegetation types by the year 2100.
Development of Tidal Marsh Migration Tool—Inputs

The tool was written in Python computer language and designed to be a stand-alone tool in
ArcGIS. The tool appears like other built-in tools and has similar functionality and descriptive
help. For the tool to run, seven site-specific inputs must be loaded: high-resolution DEM, unique
land cover classification (i.e., object-oriented approach described earlier), tide levels, low and high
marsh accretion rate, and root mean square error (RMSE) of the DEM.
Digital Elevation Models

The DEMs were generated from LiDAR point clouds with points interpolated using ordinary
kriging. DEMs were further postprocessed to remove artificial impediments to flow due to the
inability of LiDAR to penetrate pavement resulting in ground returns on bridge and overpass
In marsh
Outside of marsh
FIGURE 13.6 Two-hundred points used for an accuracy assessment on the classification of Barn Island
Wildlife Management Area. The points outside the marsh were checked via satellite imagery and all points
within the marsh were visited in the field for verification.
surfaces. Aerial imagery was used to identify and mark all bridges and overpasses in a raster for-
mat, which was then reclassified giving impediments a value of 0 and all other areas a value of 1.
The resulting reclassification was then multiplied by the DEM producing a raster with elevation
of 0 (i.e., sea level) for all bridges and overpasses. The RMSE was calculated by comparing the
DEM elevations to established National Geodetic Survey benchmarks and microrelief plots within
marshes in the study region (0.12 m RMSE for Connecticut DEM).
Image Classification
The classifications created for current tidal marsh conditions were previously described (see “Current
Tidal Marsh Classification Techniques and Methods” section). For the tool to function properly, the
classification had to include the following classes: “forest,” “agrigrass,” “urban,” “water,” “phrag,”
“high marsh,” “low marsh,” “iva,” and “sand.”
Current Tide Levels

Two generalized marsh zones—low and high—as defined by frequency of inundation (Lefor et al.
1987; Tiner 2013) were identified across the study region. The low marsh is below the mean high
water (MHW) line and flooded at least once daily. High marsh is above the MHW and flooded
less often, mostly by spring high water (SPHW) and storm events. Proximate NOAA tide gauge
data were used to generate site-specific tide values for MHW and SPHW for the year of imagery
TABLE 13.1
Error Matrix for Barn Island Wildlife Management Area in Southeastern Connecticut
Barn Island Reference Data
Low High Commission Users
Map Data Forest Urban Water Agrigrass Marsh Marsh Iva Phrag Sand Total Error (%) Accuracy (%)
Tidal Marsh Classification Approaches
Forest 35.023 1 0 0 0 0 0 1 0 25 8.7 91.3

Urban 3 15 0 6 0 0 0 0 0 25 36.0 64.0
Water 0 0 25 0 0 0 0 0 0 25 0.0 100.00
Agrigrass 6 0 0 19 0 0 0 0 0 25 24.0 76.0
Low marsh 0 0 5 0 29 4 0 4 0 42 31.0 69.1
High marsh 0 0 0 0 2 21 0 0 0 23 8.7 91.3
Iva 0 0 0 0 0 5 1 2 1 9 88.9 11.1
Phrag 4 0 0 1 0 2 1 13 0 21 38.1 61.9
Sand 0 0 0 0 0 1 1 0 3 5 40.0 60.0
Total 36 17 30 26 31 33 3 20 4 200
Omision error (%) 19.4 5.9 16.7 23.1 6.5 36.4 66.7 35.0 25.0
Producers accuracy (%) 80.6 94.1 83.3 76.9 93.6 63.6 33.3 66.0 75.0
Note: It was created through field validation of 200 random points. The matrix shows an overall accuracy of 75%.
273
retrieval (2005 for CT; 2007 for NY). The average mean higher high water for each month was used
as MHW. This value was used as it represents the average of the higher high water mark from each
tidal day. This way, the largest possible extent of area inundated daily was captured. The SPHW was
calculated by averaging the highest tide for all 12 months because there was no available spring tide
values at the tide gauges used in this study. The MHW and SPHW values were linearly interpolated
to generate site-specific tide conditions.
Current Accretion Rates

Available accretion rates for low and high marsh were retrieved from peer-reviewed publications
and surface elevation tables to generate site-specific rates via linear interpolation (Harrison and
Bloom 1977; Orson et al. 1998). Marshes with greater tidal amplitude and frequency of inundation
had higher accretion rates due to increased access to sediment (Harrison and Bloom 1977).
Model Process
The process is a sequential series of steps that integrates the model inputs. The first step requires the
creation of a “source” layer, which is the raster from which the flood will originate (i.e., ocean). This
layer is created using “map algebra” to select all elevations under 0 m in the DEM. The next step is
the creation of six adjusted DEMs spanning the 95% confidence interval for the RMSE. The RMSE
value is multiplied by the z-value of the corresponding percentiles that are then added to the original
DEM to create six additional DEMs. Each DEM is then used separately in the following procedure
to create output shapefiles.
Once the base layers are created, accretion rates for the first time step (34 years—2006–2040)
are generated. The object-oriented classification discussed earlier is then assigned a new field name
called “accretion” to the attribute table. Areas classified as “low marsh” are assigned low marsh
accretion rates, while areas classified as “high marsh,” “iva,” or “phrag” are assigned high marsh
accretion rates. Due to frequency of inundation, low marsh accretion rates are higher than high
marsh accretion rates. The accretion rates are then multiplied by the 34-year time step, and added
to the DEM. The new DEM with adjustments for accretion is the projected elevation for 2040 and
is used to model sea level rise scenarios (Figure 13.7).
The flooding is simulated using two sequential processes: level slicing and the “cost distance”
tool from ArcGIS. The projected sea level height in 2040 is used to calculate and reclassify all areas
in the 2040 DEM under that height using the “raster calculator” algorithm (i.e., if 2040 DEM ≤ 2040
sea level height, then 1, else 0). The “cost distance” tool is then used to simulate flooding. The “cost
raster” is the reclassified level slice, the “source data” is the “source” layer created initially, and the
“maximum threshold” is set to 0.5. The “cost distance” tool grows from the ocean and sums the
values of the pixels encompassed. Since the maximum threshold is 0.5, pixels with a value of 1 are
not included resulting in the generation of one connected flooded region raster that is then converted
to a polygon. This process is repeated for two more flood scenarios in the 2040 time step: 2040 plus
the current MHW and SPHW, respectively. Overlapping regions from the final flooded area outputs
are then deleted. First, the 2040 flood scenario plus MHW is erased from the 2040 plus SPHW to
generate a polygon representing the area between MHW and SPHW (high marsh). Next, the 2040
flood scenario polygon is erased from the 2040 plus MHW to generate a polygon between the flood
scenario and MHW (low marsh).
For the 2070 time step, the low and high marsh accretion rates were entered into the “accretion”
field of the 2040 polygons and converted to raster. These were the same accretion rates as the 2006
time step. While increased inundation may lead to increased accretion, there was too much uncer-
tainty to properly estimate this change; thus, the accretion values were held constant. The outputs
were multiplied by 30 to generate 2070 accretion and added to the 2040 DEM to generate the 2070
DEM. The 2070 elevation and respective flooding scenarios for 2070 are used to simulate 2070 con-
ditions. The tidal conditions also remain constant throughout the model. Tidal values may fluctuate
with increased sea level, but the relationship is too uncertain to properly model, so again it was
Agrigrass
Forest
High marsh
Iva
Low marsh
Phrag
Sand
Urban
Water
Elevation (m)
High =0.1
Low =0
(a) (b)
FIGURE 13.7 Classification for Barn Island Wildlife Management Area (a) and corresponding accretion
for 2040 (b). The accretion was derived from applying current yearly accretion rates to the current land cover
classification and multiplying it by the first time step (34 years).
held constant. Subsequently, the projected low and high marsh areas for 2070 are used to simulate
accretion in 2100 with a repeat of the process to create the final two polygons: low marsh and high
marsh areas in 2100. The two outputs reflect the different flood scenarios utilized in this analysis
(IPCC 2007 [0.59 m by 2100]; Rahmstorf 2007 [1.2 m by 2100]) resulting in two projections of three
time steps for each marsh complex in the study region (Figure 13.8).
This process is iterated so that each of the seven DEMs (one original and six created based on
the RMSE) is used to generate shapefiles for projected low and high marsh by 2100. To assess the
DEM error, the results from all seven DEMs are aggregated to create probability maps. The output
shapefiles are converted to raster and given a value of 1 for areas with new marsh and 0 for areas not
projected to be marsh. Areas projected to be low or high marsh were added together to produce two
probability maps for each site under each flooding scenario. With values ranging from 1 to 7, a 95%
confidence interval of the respective marsh type is generated with higher values having a higher
probability of becoming the respective marsh type (Figure 13.9).
RESULTS
The object-oriented classification of current marsh conditions calculated 5790 ha of marsh habitat
along Long Island Sound consisting of 2537 ha of “low marsh” and 3252 ha of “high marsh,” “iva,”
or “phrag,” which are all considered high marsh in the marsh migration tool. The classification of
current marshes provided an overview for federal, state, and local managers to quantify the distri-
bution and diversity of the coastal marshes. The classification provided localized maps of marsh
habitat useful for public outreach and education.
The future marsh habitat was projected for two sea level rise scenarios: the IPCC 0.59 m
rise by 2100 and Rahmstorf 1.20 m rise by 2100. The IPCC scenario resulted in an increase to
Barn Island
2006 conditions
Water
N
Low marsh
0 0.25 0.5 1 High marsh

Miles
(a)
Barn Island Barn Island

IPCC 0.59 m by 2100 Rahmstorf 1.2 m by 2100
N N
0 0.25 0.5 1 0 0.25 0.5 1

Miles Miles
(b) (c)
FIGURE 13.8 (a) The current classification, (b) IPCC (0.59 m) and (c) Rahmstorf (1.20 m) projections for
Barn Island Wildlife Management Area. The current classification was produced using an object-oriented
image classification approach. The IPCC and Rahmstorf projections were produced using the tidal marsh
migration model.
Barn Island Barn Island

IPCC low marsh IPCC high marsh
probability map probability map
1 1
2 2
3 3
4 4
5 5
6 6
7 7
N N
0 0.25 0.5 1 0 0.25 0.5 1

Miles Miles
(a) (b)
FIGURE 13.9 Probability maps for (a) low marsh and (b) high marsh under the IPCC projection. The prob-
ability maps were created by aggregating outputs from running the model with seven DEMs that were adjusted
to span the 95% confidence interval for the RMSE of the original DEM, which was 0.1 m. The values range
from 1 to 7 with higher numbers having higher probability of having suitable flooding conditions for marsh
formation.
10,000
High marsh
9,000 Low marsh
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
0
Current 0.59 1.2
FIGURE 13.10 Current marsh habitat and projected 2100 habitat under the IPCC (0.59 m) rise in sea level
and Rahmstorf (1.20 m) rise.
8539 ha of total marsh by 2100: 5211 ha of low marsh and 3327 ha of high marsh. The Rahmstorf
scenario projected 9475 ha of total marsh by 2100: 6653 ha of low marsh and 2282 ha of high
marsh (Figure 13.10).
Both sea level rise scenarios result in a larger expansion of low marsh relative to high marsh.
This can be explained by the slope of the adjoining upland habitat. As sea level rises, areas that
8000
High marsh
7000 Low marsh
6000
5000
4000
3000
2000
1000
0
Current 0.59 1.2
FIGURE 13.11 Current marsh habitat and projected 2100 habitat under the IPCC (0.59 m) rise in sea level
and Rahmstorf (1.20 m) rise under a built shoreline land use scenario. The built shoreline scenario restricted
marsh migration into current areas classified as “urban” or “agrigrass.”
were once flooded irregularly (high marsh) begin to be flooded diurnally fostering a conversion to
low marsh habitat. Since the high marsh is relatively flat and close to the elevation of low marsh, a
small increase in sea level opens large areas for low marsh expansion. Also, sea level rise did result
in some conversion of low marsh to open water but not enough to cancel out the expansion of low
marsh into current high marsh areas.
High marsh on the other hand migrates almost entirely onto adjoining upland habitat. The upland
has a much steeper slope, so a small increase in sea level opens up proportionally smaller areas for
high marsh advancement. This explains why low marsh increases while high marsh remains rela-
tively constant under both sea level rise scenarios.
Figure 13.10 shows the best case scenarios for future marsh habitat. These scenarios project the
conversion of all inundated upland habitat to marsh habitat regardless of current land cover and own-
ership. To simulate a more realistic scenario, adjoining upland areas currently classified as “urban”
or “agrigrass” were prevented from converting to marsh habitat by artificially adding 50 ft to the
DEM used in the original 2006 classification. The model was then run with this altered DEM to sim-
ulate a scenario that represented the maximum extent of built shoreline. Under this scenario, future
marsh habitat, particularly high marsh, was greatly reduced. For the IPCC built shoreline scenario,
the low marsh decreased to 4919 ha from 5211 ha. In addition, the amount of high marsh decreased
from 3327 ha to 1906 ha. For the Rahmstorf built shoreline scenario, the amount of low marsh
decreased to 4547 ha from 6643 ha under the Rahmstorf nonbuilt scenario. The high marsh under
the Rahmstorf built shoreline scenario was the most impacted resulting in only 968 ha remaining as
compared to the 2282 ha in the original Rahmstorf nonbuilt scenario projection (Figure 13.11). In
more urbanized environments, the built shoreline restrictions have even more significant impacts.
In Stratford (CT), high marsh expanded by 130% when modeled with no built shoreline; however, it
was reduced by 70% when built shoreline practices were modeled (Figure 13.12).

Suggested Applications
The marsh migration model was designed to generate a planning-level assessment of future marsh
conditions. Results can help decision makers make best use of limited resources by focusing on
specific marsh systems of concern. While overall marsh habitat increases under most sea level and
land use scenarios, the composition of the marsh changes drastically. Currently, there is slightly
Ocean
Low marsh
High marsh
Built shoreline
No built shoreline
FIGURE 13.12 Marsh projections under a Rahmstorf (1.20 m) rise in sea level for Stratford, CT. The image
on the left projects marsh advancement with no restrictions, while the image on the right simulates a “built
shoreline” scenario. Under the built shoreline scenario, all “urban” and “agrigrass” land covers were artifi-
cially raised by altering the underlying DEM by 50 ft. This mimics the construction of walls and had signifi-
cant impacts on marsh projections.
more high marsh than low marsh. However, under all sea level scenarios, low marsh expands while
high marsh either remains constant or declines (70% reduction of high marsh for Rahmstorf built
shoreline scenario).
This study makes it clear that the conservation of properly functioning tidal marshes is dependent
on the ability of high marsh to advance inland. High marsh is an essential habitat for a variety of ani-
mals, especially as a bird nesting area as well as providing risk reduction through wave attenuation
for people and property. Identifying suitable and unsuitable inland areas for tidal marsh advancement
is an essential strategy for resource managers looking to maximize the persistence of this habitat
type in the landscape. For the study region, advancement zones were identified by overlaying the
outputs from the tidal marsh migration tool with the object-oriented derived land cover classification.
Areas where the marsh projections overlapped the “urban” land cover were marked as unsuitable or
“no advancement.” These areas were labeled “no advancement” because even if they were acquired
for conservation, they would not convert to marsh without extensive intervention. Areas where marsh
projections overlapped the “forest” or “agrigrass” land cover type were labeled as “advancement”
because if they were conserved as open space, they would naturally convert to marsh.
Maps showcasing advancement zones (Figure 13.13) for each of the 40 sites are enabling federal,
state, municipal, and private resource managers and decision makers to prioritize and focus their
limited capacity on areas with the greatest potential to accommodate future tidal marsh over time.
Of particular concern is the longer-term loss of high marsh habitat through conversion to low marsh
and landward impediments to advancement, principally current development and/or topography.
The longitudinal compression and eventual loss of high marsh in Long Island Sound and other
estuaries around the globe with similar constraints will result in the reduction of critical habitat and
protective functions for wildlife and people and property from extreme weather events and acceler-
ating rates of sea level rise.
Advancement
N
No advancement
W E
0.5 0.25 0 0.5 miles
S
FIGURE 13.13 Advancement zones for the Barn Island Wildlife Management Area. Areas of “advance-
ment” were upland habitat that were projected to be marsh and would convert to marsh habitat. Areas of “no
advancement” are those regions that will be flooded but are currently developed.
Limitations of Study
While the marsh migration model offers projections of future tidal marsh advancement, it is not
without limitations. Several aspects of the tool are simplistic models of complex marsh processes.
Tidal and accretion rates were generated by interpolating between the few available datasets. In
addition, tidal conditions and accretion rates were held constant throughout each site and through
time. The DEM also had an RMSE of 0.12 m that impacted the results.
The purpose of this project was to generate an overall assessment of current marsh distribution
and diversity and project possible areas of future marsh habitat. The model is not accurate enough
to decipher specific low marsh/high marsh boundaries or project absolutely where future marsh
habitat will reside. Rather, it is designed to give decision makers the best available estimate of which
marshes will lose the most marsh habitat and which upland habitat has high probability for inunda-
tion, thus most critical for conservation.
Resource Requirements: Equipment, Software, and Expertise

This project used a new approach to classify current marsh habitat as well as an innovative approach
to project future marsh. The outputs provide the ability to assess current management practices as
well as implement mitigation strategies for future marsh conservation. The methods described here
are readily applicable to other regions and require an intermediate understanding of remote sensing
processing as well as GIS analysis. For the current classification, the user would require high-
resolution remotely sensed imagery as well as a DEM for high-accuracy marsh classification. GPS
equipment is also required for in situ verification. The required software to duplicate the current
classification methods is ArcGIS and Definiens Developer. The tidal marsh migration tool is avail-
able as a tool in ArcGIS. The tool can be loaded and run by just providing the input data. The data
required to run the tool are a current classification, DEM, tidal and accretion rates, and the RMSE
for the DEM. The outputs for the tool are shapefiles for future low marsh and high marsh zones.
Further analysis such as identifying advancement zones would have to be performed post analysis
and would require a higher understanding of GIS analysis.
REFERENCES
Bricker-Urso, S., S.W. Nixon, J.K. Cochran, D.J. Hirschber, and C. Hunt. 1989. Accretion rates and sediment
accumulation in Rhode Island salt marshes. Estuaries 12(4):300–317.
Brinson, M.M., R.R. Christian, and L.K. Blum. 1995. Multiple states in the sea-level induced transition from
terrestrial forest to estuary. Estuaries 18(4):648–659.
Civco, D., J. Hurd, M. Gilmore, and S. Prisloe. 2006. Characterization of coastal wetland systems using mul-
tiple remote sensing data types and analytical techniques. IEEE International Geoscience and Remote
Sensing Symposium, pp. 3442–3446. doi: 10.1109/IGARSS.2006.883.
Civco, D.L., M.S. Gilmore, E.H. Wilson, N. Barrett, S. Prisloe, J.D. Hurd, and C. Chadwick. 2008. Multitemporal
spectroradiometry-guided object-oriented classification of salt marsh vegetation. Proceedings of the
SPIE Europe Remote Sensing, Cardiff, Wales, U.K.
Cooper, M., M D. Beevers, and M. Oppenheimer. 2005. Future sea level rise and the New Jersey Coast. Science
Technology and Environmental Policy Program, Princeton University, Princeton, NJ.
Dreyer, G.D. and W.A. Niering. 1995. Tidal Marshes of Long Island Sound. The Connecticut College
Arboretum, New London, CT.
Harrison, E.Z. and A.L. Bloom. 1977. Sedimentation rates on tidal marshes in Connecticut. Journal of
Sedimentary Petrology 47(4):1484–1490.
Intergovernmental Panel on Climate Change (IPCC). 2007. Climate Change 2007: A Synthesis Report.
Intergovernmental Panel on Climate Change, Valencia, Spain.
Laliberte, A.S., A. Rango, K.M. Havstad, J.F. Paris, R.F. Beck, R. McNeely, and A.L. Gonzalez. 2004. Object-
oriented image analysis for mapping shrub encroachment from 1937 to 2003 in southern New Mexico.
Remote Sensing of the Environment 93:198–210.
Lefor, M.W., W.C. Kennard, and D.L. Civco. 1987. Relationships of salt-marsh plant distributions to tidal levels
in Connecticut, USA. Environmental Management 11(1):61–68.
Long Island Sound Resource Center. 2005. http://www.lisrc.uconn.edu/lisrc/catalogimages.asp.
Morris, J.T., P.V. Sundareshwar, C.T. Nietch, B. Kjerfve, and D.R. Cahoon. 2002. Responses of coastal wet-
lands to rising sea level. Ecology 83(10):2869–2877.
Nicholls, R.J., F.M.J. Hoozemans, and M. Marchand. 1999. Increasing flood risk and wetland losses due to
global sea-level rise: Regional and global analysis. Global Environmental Change 9:S69–S87.
Northeast Climate Impact Assessment (NECIA). 2007. Confronting Climate Change in the U.S. Northeast:
Science, Impacts, and Solutions. Northeast Climate Impact Assessment, Cambridge, MA.
Nydick, K.R., A.B. Bidwell, E. Thomas, and J.C. Varekamp. 1995. A sea-level rise curve from Guilford,
Connecticut, USA. Marine Geology 124:137–159.
Orson, R.A., R.S. Warren, and W.A. Niering. 1998. Interpreting sea level rise and rates of vertical
marsh accretion in a southern New England tidal salt marsh. Estuarine, Coastal and Shelf Science
47:419–429.
Platt, R.V. and L. Rapoza. 2008. An evaluation of an object-oriented paradigm for land use/land cover classifi-
cation. The Professional Geographer 60(1):87–100.
Rahmstorf, S. 2007. A semi-empirical approach to projecting future sea level rise. Science 135:368–370.
Slovinsky, P.A. and S.M. Dickson. 2006. Impacts of Future Sea Level Rise on the Coastal Floodplain. Maine
Geological Survey, Augusta, ME.
Tiner, R.W., K. McGuckin, and J. Herman. 2013. Potential Wetland Restoration Sites for Connecticut: Results
of a Preliminary Statewide Survey. U.S. Fish and Wildlife Service, Northeast Region, Hadley, MA.
Titus, J.G. 1988. Sea level rise and wetland loss: An overview. U.S. Environmental Protection Agency Report,
203-05-8b-013, Washington, DC.
Warren, R.S. and W.A. Niering. 1993. Vegetation change on a northeast tidal marsh: Interaction of sea-level rise
and marsh accretion. Ecology 74(1):96–103.
Xiaoxia, S., Z. Jixian, and L. Zhengjun. 2005. A comparison of object-oriented and pixel based classification
approaches using Quickbird imagery. International Symposium on Spatio-Temporal Modeling, Spatial
Reasoning, Analysis, Data Mining, Data Fusion, Beijing, China, August 27–29.
14 Using Moderate-Resolution
Satellite Sensors for
Monitoring the Biophysical
Parameters and Phenology
of Tidal Marshes
Deepak R. Mishra and Shuvankar Ghosh
CONTENTS
Introduction.....................................................................................................................................284
Recent Applications of Remote Sensing for Tidal Wetland Mapping............................................ 285
Monitoring Tidal Marshes for the Northern Coast of the Gulf of Mexico..................................... 287
Study Area....................................................................................................................................... 287
Methods........................................................................................................................................... 288
In Situ Data Collection............................................................................................................... 288
Top of Canopy Reflectance (Rrs)........................................................................................... 289
Leaf Chlorophyll Content...................................................................................................... 289
Vegetation Fraction................................................................................................................ 289
Leaf Area Index..................................................................................................................... 291
Green Biomass...................................................................................................................... 291
Satellite Data.............................................................................................................................. 291
Landsat.................................................................................................................................. 291
MODIS.................................................................................................................................. 292
Model Calibration and Validation.............................................................................................. 292
Map Preparation and Phenology Extraction.............................................................................. 292
Results and Discussion................................................................................................................... 293
Canopy Reflectance Analysis..................................................................................................... 293
Landsat Analysis........................................................................................................................ 294
Oil Spill Change Analysis Using Landsat Biophysical Products............................................... 297
MODIS Analysis........................................................................................................................ 298
Application of These Technologies............................................................................................ 301
Conclusion......................................................................................................................................302
References....................................................................................................................................... 311
283
INTRODUCTION
Tidal marshes are some of the most ecologically and economically productive as well as vul-
nerable ecosystem of the world (Mitsch and Gosselink 2007; Tiner 2013). They perform vari-
ous ecological functions such as water quality protection through nutrient retention, flooding and
erosion control, streamflow maintenance, groundwater recharge, providing habitats for numer-
ous species, and storm surge protection. Their high carbon sequestration potential makes tidal
marshes critical habitats, considering the exponential increment of greenhouse gases in the atmo-
sphere since the last few decades (Gallagher et al. 1980; Connor et al. 2001; Chmura et al. 2003).
However, these wetlands are severely threatened by natural processes and anthropogenic activities
including global warming–induced sea-level rise (Tiner 2013), land-use change (Kennish 2001;
Silliman et al. 2009), soil erosion (Sugumaran et al. 2004; Ravens et al. 2009), natural and man-
made disasters and associated cleanup efforts (Gilfillan et al. 1995; Hester and Mendelssohn 2000;
Ramsey and Rangoonwala 2010; Mishra et al. 2012), and succession by other species (Artigas and
Pechmann 2010).
Given the significance of tidal marshes, government agencies interested in conserving coastal
wetlands are initiating programs to track changes in their extent and condition. A sound wetland
monitoring program for sustainable management should be comprehensive in nature covering
important parameters, especially wetland distribution, composition, characteristics, and health and
productivity (Adam et al. 2010). Monitoring efforts based on traditional field sampling is often
costly, time consuming, and inadequate to provide a broad regional trend of wetland status. Remote
sensing has long been used as the preferred tool for conducting wetland assessment at multiple
scales. Satellite and airborne remote sensors provide cost-efficient alternatives to intensive field sur-
veys in monitoring and assessing coastal ecosystems and their dynamics at appropriate scales and
resolutions. Both remote sensing and geographic information system (GIS) technology have been
used by researchers and government agencies for more than 30 years to map and monitor wetland
resources (Tiner 1996; Dahl 2006). For example, the U.S. Fish and Wildlife Service (FWS) has been
using remote sensing and on-ground observations to delineate the biological extent of wetlands for
the past three decades and has made these data available to federal, state and private agencies, and
citizens in general. Several U.S. states have also conducted their own wetland resource inventories,
using both color and color-infrared aerial photographs along with fine- and medium-resolution mul-
tispectral satellite imageries. With the recent advances in remote sensing technology, researchers
have been applying a wide range of techniques using both passive and active remote sensing includ-
ing hyperspectral sensors and LiDAR to map wetlands (Lefebvre et al. 2010; Yang and Artigas
2010; Zhang and Xie 2012; Evans and Costa 2013).
Wetland remote sensing is challenging compared to remote sensing of terrestrial vegetation
because the water or moist soil interface reduces the intensity of the near-infrared (NIR) signal.
Wetland habitats also exhibit high spectral and spatial variations due to the abrupt change of envi-
ronmental conditions that produce short ecotones, causing difficulty in vegetation community
boundary identification (Schmidt and Skidmore 2003). Different vegetation species growing in the
wetland habitats produce different vegetative groups that have various phenological cycles, physi-
ological structures, growth requirements, and biochemical composition, which eventually lead to
diverse spectral behavior and make mapping a difficult task (Rosso et al. 2005; Zomer et al. 2009).
One of the challenges for wetland remote sensing studies is the lack of available spectral libraries
for the large number of species that need to be characterized (Zomer et al. 2009). Patchiness and the
fine scale of heterogeneity in many wetlands lead to the inability of medium-resolution sensors such
as Landsat and SPOT to capture the species-level variability, while the use of aerial photography is
expensive and inefficient. Nonetheless, Landsat and SPOT data have been traditionally used for wet-
land cover mapping (Rundquist et al. 2001; Klemas 2011). Their respective spatial resolutions of 30
and 20 m have proven to be reasonably good for mapping the presence/absence and distribution of
more readily identifiable wetland habitats (Lunetta and Balogh 1999; Houhoulis and Michener 2000;
Using Moderate-Resolution Satellite Sensors 285
Harvey and Hill 2001). Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus
(ETM+) have also been extensively used to study water depth and turbidity in wetlands, as well
as the seasonal dynamics of inundation, and vegetation cover (Bustamante et al. 2009; Ward et al.
2013). Both supervised and unsupervised classification techniques or their combination have been
used in conjunction with multitemporal imagery across varying spatial resolutions by numerous
researchers for actual classification of the wetland cover (Jensen 1996; Ozesmi and Bauer 2002;
McCarthy et al. 2005; Campbell 2007; Lillesand et al. 2008). Existing aerial photographs, maps,
and field samples have served as ancillary data whenever available. As discussed earlier, spatial res-
olution of the sensors has been regarded as the most important factor in wetland mapping. Imagery
from TM, SPOT, and other moderate-resolution sensors provides decent classification accuracies
when wetland cover dominates extensive areas; however, classification using MODIS or AVHRR
(250 m, 500 m, 1 km) results in frequent misclassification of pixels (Santos-Gonzalez 2002; Fang
et al. 2013).
Today, given significant improvements in wetland conservation via government laws and policies
in the United States, there is increasing interest in evaluating and monitoring wetland condition or
health (EPA 2011). This chapter provides a review of the use of remote sensing for tidal wetland
mapping and presents a sample application of moderate-resolution satellite imagery for assessing
the condition or health of those wetlands.
RECENT APPLICATIONS OF REMOTE SENSING FOR TIDAL WETLAND MAPPING

Since the last decade, the availability of high temporal and spectral resolution satellite data has sig-
nificantly enhanced the capability of mapping tidal wetland ecosystems (Ozesmi and Bauer 2002;
Jensen et al. 2007; Laba et al. 2008; Wang et al. 2010). Landsat, ASTER, and MODIS imagery is
open-sourced for agencies and the public in general. High-spatial-resolution imagery can now be
obtained from satellites such as IKONOS and QuickBird. Major plant species within a heteroge-
neous tidal wetland complex have been classified using multitemporal high-resolution QuickBird
images, field reflectance spectra, and LiDAR height information (Gilmore et al. 2010). Phragmites
australis, Typha spp., and Spartina patens were spectrally distinguishable at particular times of the
year, likely due to the differences in biomass and pigments and the rate at which they changed dur-
ing the growing season (Ghioca-Robrecht et al. 2008; Gilmore et al. 2008, 2010). QuickBird was
also used to map plant communities and invasive species in the Hudson River National Estuarine
Research Reserve (HRNERR), an area occupied by four diverse tidal wetlands as well as invasive
species Trapa natans, P. australis, and Lythrum salicaria (Laba et al. 2008). In the HRNERR
study, the standard maximum likelihood classification (MLC) and a traditional confusion matrix
were employed as well as a fuzzy set analysis as the basis for accuracy assessment. The latter analy-
sis represented the nature, frequency, magnitude, and source of errors corresponding to the plant
communities assigned. The overall accuracies varied from 64.9% to 73.6% for the four study wet-
lands. The lowest user’s accuracy was associated with vegetated lower intertidal Typha angustifolia,
Scirpus spp., L. salicaria, Scirpus pungens, wooded swamp, and Acorus calamus. In general, the
water/tidal channels class was the most accurately mapped, while the Scirpus spp. community was
the least accurately represented.
IKONOS data were used to monitor the response of estuaries and tidal wetlands to environmen-
tal change in HRNERR (Laba et al. 2010). The authors investigated the effect of textural informa-
tion on land cover map accuracy. An MLC algorithm with various methods of enhancement was
used to classify the image. The inclusion of local texture information (variance) of different moving
window sizes (3 × 3 and 5 × 5) was applied. Additional methods of using variance-based, edge-
preserving smoothing, and segmentation algorithm combination were also used. Overall results
showed that the method using an MLC to four bands of IKONOS produced an accuracy of 76.2%.
Confusion increased due to spectral properties of pixels corresponding to vegetated lower intertidal,
submerged aquatic vegetation, and water.
Multitemporal SPOT-5 images have been used to study wetlands in the Camargue (Rhone
delta, France)—home to mixed semibrackish water, submerged macrophytes (Potamogeton spp.,
Ruppia spp., Chara spp.), and emergent halophytes (common reed, P. australis, and club rush,
Bolboschoenus maritimus) (Davranche et al. 2010). They evaluated the classification tree method
to map the presence of dominant emergent and submerged macrophytes of these wetlands. Data
fragmentation and two cross-validation procedures were applied in the pruning stages. Mean reflec-
tance values of each band and multitemporal vegetation indices (VIs) such as simple ratio (SR),
differential vegetation index (DVI), moisture stress index (MSI), soil-adjusted vegetation index
(SAVI), optimized SAVI (OSAVI), normalized difference water index (NDWI), NDWI of Mac
Feeters (NDWIF), normalized difference vegetation index (NDVI), and difference between NDVI
and NDWI (DVW) were extracted for plots of reeds, aquatic beds, and other land covers. Results
showed that the method accurately mapped common reed and submerged macrophytes with accu-
racies of 97% and 86%, respectively. The discrimination of common reed was based on the differ-
ences in reflectance of band 3 (red) between images acquired in March and June. This difference
occurred due to increase chlorophyll production with the progress of growing season. Similarly,
NDWIF, which classifies water in positive and chlorophyll/turbidity in negative values, was used to
differentiate submerged macrophytes from open marshes.
Apart from multispectral satellite sensors, low-altitude high-resolution airborne sensors can be
used to study relatively small wetland patches (McCoy 2005; Jensen 2007; Klemas 2011). Airborne
digital cameras providing color and color-infrared digital imagery can be integrated with global
positioning system (GPS) and used in a GIS for a wide range of modeling applications (Lyon and
McCarthy 1995). Airborne hyperspectral imagers, such as the advanced visible infrared imag-
ing spectrometer (AVIRIS) and the Compact Airborne Spectrographic Imager (CASI) have been
extensively used for mapping coastal wetlands (Thomson et al. 1998; Ozesmi and Bauer 2002;
Schmidt and Skidmore 2003; Li et al. 2005; Rosso et al. 2005). Hyperspectral imagers provide
hundreds of narrow spectral bands located in the visible, NIR, mid-infrared (MIR), and sometimes
thermal portions of the electromagnetic (EM) spectrum (Jensen et al. 2007) and are effective in
discriminating between different land covers and vegetation communities. However, the volumi-
nous data of hyperspectral image often require the use of specific software packages, large data
storage devices, and extended processing times (Hirano et al. 2003). Lopez et al. (2004) performed
a detailed accuracy assessment of airborne hyperspectral data for mapping plant species in fresh-
water coastal wetlands.
In summary, the majority of the research efforts for monitoring wetlands using remote sens-
ing have been made to delineate the extent of wetland ecosystems and classify plant communities
using both active (sensors with their own source of illumination) and passive (sensors using natural
source of illumination) sensors and numerous image processing techniques (Gross and Klemas
1986; Malthus and George 1997; Yan et al. 2006; Marani et al. 2006; Sadro et al. 2007; Gilmore
et al. 2008; Lucas and Carter 2008; Artigas and Pechmann 2010; Collin et al. 2010; Davranche et al.
2010; Goudie 2013). Such mapping provides crucial information on the presence/absence of wetland
patches, previous and current spatial extents, and the dynamics of wetland cover. Unfortunately,
remote sensing studies trying to monitor and analyze the biophysical properties of wetlands are
very limited and intermittent (e.g., Hardisky et al. 1983, 1984, 1986; Kearney et al. 2009; Ramsey
and Rangoonwala 2010; Mishra et al. 2012). Biophysical parameters of wetlands are primary indi-
cators of the physiological status, photosynthetic capacity, and nitrogen content. Remote estima-
tion of biophysical parameters such as leaf area index (LAI), canopy chlorophyll (CHL) content,
vegetation fraction (VF), and aboveground green biomass (GBM) should be performed regularly
to assess the health of wetland ecosystems. Monitoring these parameters not only helps in assess-
ing the overall dynamics of wetlands but also facilitates prioritization of restoration in areas that
require immediate restoration and conservation measures. A robust biophysical mapping protocol
is also critical in assessing the success/failure of previous restoration efforts (Hinkle and Mitsch
2005; Friess et al. 2012).
MONITORING TIDAL MARSHES FOR THE NORTHERN

COAST OF THE GULF OF MEXICO
The following case study represents an example of the application of moderate-resolution satellite
data for calibrating and fine-tuning a suite of existing and well-established VIs for tidal marshes.
Topics for discussion include (1) calibration and validation of Landsat and MODIS-based VIs for
predicting biophysical parameters (green LAI [GLAI], CHL, percent green VF, and aboveground
GBM for Gulf Coast marshes), (2) development and analysis of time-series composites of these bio-
physical parameters, and (3) phenological analysis of the marsh biophysical products. The existing
VIs that have been traditionally targeted at terrestrial vegetation need new calibration and adjust-
ment before they can be used on marsh vegetation. Landsat TM data analysis presented in this
chapter is a case study related to the oil spill damage assessment and demonstrates the utility of
the sensor to map the marsh biophysical parameters and detect short-term changes in tidal marshes
due to natural or anthropogenic events. The MODIS analysis presented in this chapter spans over
10 years (2000–2010) and involves the majority of northern Gulf Coast tidal marshes. An efficient
and nondestructive biophysical mapping protocol for emergent wetlands can be an invaluable tool
for restoration decision making, thus providing government regulators with important information
for estimating mitigation needs.
STUDY AREA
Tidal marshes occupy the majority of the northern coast of the Gulf of Mexico (>8600 km 2)
(National Wetlands Inventory, U.S. Fish and Wildlife Service, http://www.fws.gov/wetlands/)
(Figure 14.1). They are mostly dominated by smooth cordgrass (Spartina alterniflora), salt meadow
cordgrass (S. patens), and black needlerush (Juncus roemerianus) with occasional presence of
perennial glasswort (Salicornia virginica), saltwort (Batis maritima), and salt grass (Distichlis
spicata). The highly saline and anaerobic nature of the soil allows only relatively few species to
thrive, and as such, floral diversity is remarkably low (Weis 2010; Tiner 2013). The region expe-
riences a tropical to subtropical climate characterized by hot summers, with occasional tropical
storms and moderately cold winters. Average annual temperature varies between 15°C and 25°C,
while annual precipitation ranges from 80 to 100 cm (Weather Underground, http://www.wun-
derground.com/). Louisiana alone comprises more than half of the tidal wetland habitat in the
W E
Survey sites 2010

Survey sites 2011
0 100 200 300 400

km
FIGURE 14.1 Tidal marsh extent in the northern Gulf of Mexico (green), spanning over Louisiana,
Mississippi, Alabama, and Florida. The digital vector boundary acquired from NWI is overlaid on Band 2
(NIR) of a 500 m 8-day surface reflectance MODIS image. Point locations represent the sampling sites during
2010 (blue) and 2011 (red).
W E
0 25 50 75 100
km S
FIGURE 14.2 Tidal marsh extent in southeast Louisiana (green). The digital vector boundary acquired from
NWI is overlaid on Band 2 (NIR) of a 500 m 8-day surface reflectance MODIS image.
northern Gulf of Mexico (Figure 14.2). For Landsat mapping, the analysis was restricted to the
wetland extent in southeast Louisiana because of the sporadic availability of cloud-free images for
the entire Gulf Coast. Most of the northern Gulf of Mexico was used for MODIS analysis because
of the availability of the cloud-free 8-day 250 and 500 m composite images.
METHODS
The study involved a combination of in situ data collected over multiple field trips and Landsat and
MODIS satellite data acquired from National Aeronautics and Space Administration (NASA) and
U.S. Geological Survey (USGS) archives. The in situ data included four biophysical parameters
(GLAI, CHL, VF, and GBM) and hyperspectral reflectance readings, acquired from roughly 200
study plots during field campaigns spanning over 2010–2011. These data were used for model cali-
bration and validation using Landsat and MODIS in order to create time-series maps to analyze the
salt marsh habitat dynamics.
In Situ Data Collection

Field data collection is crucial for calibrating the satellite-based VIs for accurate marsh health
mapping. Several metrics are recognized as important for this work such as top of canopy (TOC)
spectral reflectance, GLAI, CHL, percent green VF, and aboveground GBM. These parameters
have the potential to serve as suitable indicators of marsh health and productive capacity. Spectral
signature of healthy vegetation is characterized by high absorption in the visible and high scatter-
ing in the infrared regions of the EM spectrum. Progressive stages of senescence and stress lead
to reduced absorption in the visible regions, with decreasing concentrations of the photosynthetic
pigment chlorophyll, and increase in the concentrations of accessory pigments such as carotene and
xanthophyll. This decrease in chlorophyll concentration both at the leaf and at canopy level directly
influences the photosynthetic capacity of marsh vegetation.
Productive capacity of vegetation is a direct function of the canopy structure, leaf, and healthy
plant distribution, which influences light use efficiency (LUE) and fraction of photosynthetic
active radiation (fPAR). GLAI (the ratio of total green foliage area to ground) has a direct rela-
tionship with both LUE and fPAR. Greater GLAI usually indicates healthy and productive vegeta-
tion. Green VF defines the distribution of green vegetation from a horizontal perspective. VF is
related to the spectral signature of the marsh surface based on the proportion of soil/dead biomass
and healthy vegetation. Finally, monosaccharide carbohydrates produced through photosynthe-
sis are stored in the plant body usually in the form of long-chain polysaccharide carbohydrates
(e.g., starch and cellulose), which accumulate to increase the dry weight of the plant. Therefore,
aboveground GBM becomes a good potential indicator of the photosynthetic capacity of marsh
vegetation.
Top of Canopy Reflectance (Rrs)

A dual-fiber system with two intercalibrated Ocean Optics USB4000 hyperspectral radiometers
(Ocean Optic Inc., Dunedin, FL) mounted on an aluminum frame was used to acquire the TOC
spectral reflectance (R rs) data in the range of 200–1100 nm with a sampling interval of 0.3 nm
(Rundquist et al. 2004). The radiometer #1 with a field of view (FOV) of 25° pointed downward
to acquire upwelling radiance (L; W m 2/sr), while the radiometer #2 equipped with a cosine cor-
rector pointed upward to acquire downwelling irradiance (E; W/m2) simultaneously. Four scans
of radiance and irradiance were acquired and converted to four R rs readings by dividing L over
E and then averaged out to obtain composite spectra of the study plot. Based on the FOV and the
height of the frame (5 m), the spatial resolution (IFOV) of the sensor was calculated to be 1.83 m
(Figure 14.3a and b):
⎧ ⎛ α ⎞⎫
d = 2 ⎨h × ⎜ tan ⎟ ⎬ (14.1)
⎩ ⎝ 2 ⎠⎭

where
d is the diameter of the IFOV
h is the height of the sensor from the target
α is the FOV of the sensor
Intercalibration of the radiometers was accomplished by measuring the upwelling radiance of a

99% white Spectralon reflectance standard (Labsphere, Inc., North Sutton, NH) simultaneously
with incident irradiance. In case of changing sky conditions, the sensor was recalibrated at regular
intervals (Figure 14.3c and d).
Leaf Chlorophyll Content

Minolta 502 SPAD Chlorophyll Meter (Spectrum Technologies Inc., East Plainfield, IL) was used
to measure the in situ leaf-level chlorophyll content (Figure 14.3e). A total of 20 stratified random
SPAD readings were acquired from each sampling location across varying chlorophyll levels inside
the IFOV of the sensor. The 20 readings were averaged and converted to absolute chlorophyll values
(mg/m2) by using coefficients derived from a calibration experiment conducted in the beginning of
the field season. The calibration procedure involved laboratory-based analytical extraction of chlo-
rophyll a from a few leaf samples and development of a statistical relationship between the analyti-
cal chlorophyll a and the corresponding SPAD readings (Gitelson et al. 2005).
Vegetation Fraction
Percentage green VF was estimated from a circular crop of vertical digital photographs of the
study plots acquired by OLYMPUS E-400 digital SLR camera (Olympus America Inc., Center
16 ft
(c)
~2 m
(a) (b) (d)
(e) (f) (g)
(h) (i) (j)
FIGURE 14.3 Field data collection: (a) a dual-headed Ocean Optics sensor setup mounted on a 5 m (16 ft)
high frame; (b) instantaneous FOW (spatial resolution) of the sensor (diameter, 2.2 m); (c, d) in situ calibra-
tion of the Ocean Optics sensor using a 99% Barium sulfate reflectance panel (Spectralon panel); (e) leaf-level
chlorophyll measurements using SPAD chlorophyll meter; (f, g) percent VF estimation from vertical digital
photograph acquired by Olympus E-400 DSLR camera; (h, i) LAI measurement using both below- and above-
canopy incident solar radiation readings with LiCOR’s LAI-200 Plant Canopy Analyzer and AccuPAR LP-80
Ceptometer; and (j) biomass sample collection from study plots for further sorting, oven drying, and subse-
quent dry weight measurements.
Valley, PA). The camera was installed on the frame along with a laser pointer next to the hyperspec-
tral radiometer. The laser pointer marked the center of the digital photograph and the IFOV. The
digital photograph was cropped to match the IFOV of the hyperspectral radiometer and the VF was
estimated by the ratio of the number of green pixels to the total number of pixels in each photograph
(Figure 14.3f and g).
Leaf Area Index

LAI was measured using an LAI Plant Canopy Analyzer 2000 (LICOR Biosciences Inc., Lincoln,
NE) (Gitelson 2004) and AccuPAR LP-80 Ceptometer (Decagon Devices Inc., Pullman, WI)
(Delalieux et al. 2008, Kovacs et al. 2009) (Figure 14.3h and i). The median of four LAI readings
taken in each study plot was used as the LAI of the study plot. Each LAI measurement involved
one above-canopy and four below-canopy readings. GLAI was calculated as the product of LAI and
CHL (mg/m2) was calculated as the product of LAI and LLC based on Gitelson et al. (2002):
GLAI = LAI × VF (14.2)

CHL = LLC × LAI (14.3)
Green Biomass
Aboveground GBM data were collected by destructive sampling from a 0.09 m2 (1 ft2) subplot
within each study plot using a PVC frame and clippers (Figure 14.3j). Each biomass sample was
sorted to separate the green from brown, oven dried at 65°C overnight (~24 h) to get rid of the mois-
ture, and then the dry weight was recorded using a standard measuring balance. Precautions were
taken to avoid moisture absorption by the dried GBM during dry weight measurement as much as
possible. The dry GBM weights (g/ft2) were then rescaled to g/m2.
Satellite Data
Landsat
Multitemporal Landsat-5 (TM) 30 m datasets for the southeastern Louisiana coast were acquired
from the USGS archives (USGS Global Visualization Viewer, http://glovis.usgs.gov) for the grow-
ing seasons (April–October) of 2009–2011. Atmospheric scattering was removed from each image,
and the digital numbers (DN) were converted to on-ground reflectance (ρ) values using the COST
atmospheric correction model (Chavez 1996). COST atmospheric correction is based on a revised
dark-object model, where atmospheric transmittance is estimated from the cosine of zenith angle
(cos(TZ)). The method is employed to derive the multiplicative transmittance correction coefficients
without in situ field measurements for bands 1–5 and band 7 of Landsat sensors (ARSC 2002).
Details of the COST correction procedure for the Louisiana coast Landsat imageries are avail-
able (Mishra et al. 2012). Landsat scenes acquired by the sensor within ±7 days were mosaicked
and the subsets of marsh extents were prepared in ERDAS IMAGINE 2010 (Leica Geosystems
Inc., Heerbrugg, Switzerland) for southeastern Louisiana. Wetland vector boundaries acquired from
National Wetlands Inventory (NWI) archives (National Wetlands Inventory, U.S. Fish and Wildlife
Service, http://www.fws.gov/wetlands/) were used to subset the Landsat scenes. Cloud cover in the
Landsat datasets was eliminated using a threshold NDVI mask. The NDVI mask, a binary image,
was created from the monthly subsets. A rescaled NDVI threshold value of 135 was used to mask
out cloud cover areas from the image. Individual binary cloud cover masks were generated for each
monthly subset. These masks were then applied to the corresponding monthly subset to create
cloud-free monthly images.
MODIS
Multitemporal 8-day Level 1B atmospherically corrected surface reflectance composites for the
Northern Gulf of Mexico were acquired from NASA (MODIS Land, http://modis-land.gsfc.nasa.
gov) for the growing season (April–October) from 2000 through 2011. Both 250 and 500 m scenes
from MODIS sensor were downloaded for the Gulf Coast and mosaicked. The mosaic scenes were
cropped to isolate the marsh habitats by using the vector boundaries obtained from NWI database.
Model Calibration and Validation

Following the initial preprocessing of satellite data, in situ sampling locations were used to extract
pixel values from the Landsat and MODIS images. Scenes were chosen based on the proximity
of the dates between the image acquisition and field data collection. Several preexisting and well-
established VIs were derived from the extracted pixel values for model calibration for each bio-
physical parameter (Table 14.1). For a pixel containing multiple sampling locations, the average
value of the individual biophysical parameter was calculated and used in model calibration. During
this process, roughly 100 sampling plots acquired in 2010 were reduced to 10–15 MODIS pixels and
around 30 Landsat pixels. Further, an independent dataset was acquired during the field campaigns
in 2011 containing another 100 sampling plots that were subsequently reduced to 10–12 MODIS and
15–20 Landsat pixels and used for model validation. Model calibrations were performed for Landsat
and MODIS followed by validations. Performance uncertainties were analyzed based on percent
root mean square error (%RMSE) and residuals (measured/predicted).
Map Preparation and Phenology Extraction

The best-fit models selected after successful calibration and validation were used on Landsat and
MODIS datasets to develop time-series composites. Since Landsat 5 has a temporal resolution of
16 days, acquiring monthly cloud-free datasets for the Louisiana marshes was challenging. Multiple
masking procedures involved in Landsat image processing in order to retain a common area for
time-series analysis restricted the study area to southeastern Louisiana. MODIS, on the other hand,
has much frequent temporal coverage compared to Landsat, and the MODIS 8-day surface reflec-
tance products eliminated the problem of cloud cover. Therefore, MODIS analysis covered the
majority of northern Gulf Coast marsh habitats at a temporal frequency of 8 days.
Site-specific phenology charts were extracted from the time-series composites prepared using
the best-fit model to analyze the spatiotemporal dynamics of the biophysical parameters, health, and
TABLE 14.1
List of Satellite Image–Derived Vegetation Indices for Calibration and
Validation of the Biophysical Parameters
Vegetation Index Formula Reference
NDVI (RNIR − RRed)/(RNIR + RRed) Rouse et al. (1974)
EVI2 {2.5 × (RNIR − RRed)/(RNIR + 2.4 × RRed + 1)} Huete et al. (2002)
Chlorophyll index red (CIRed)a (RNIR − RRed)/RRed Gitelson et al. (2006)
WDRVI (α × RNIR − RRed)/(α × RNIR + RRed) Gitelson (2004)
SAVI (RNIR − RRed) × (I + L)/(RNIR + RRed + L) Huete (1988)
Chlorophyll index green (CIGreen) (RNIR −RGreen)/RGreen Gitelson et al. (2006)
VARI (RGreen − RRed)/(RGreen + RRed) Gitelson et al. (2002)
a RRed was used instead of Rred-edge due to the unavailability of a red-edge band in LANDSAT and MODIS.
physiology. Random points were overlaid on specific sites (e.g., restoration sites, oil spill-affected
sites, fringing marsh sites, and interior marsh sites), and the modeled values of each biophysical
parameters were extracted from the time-series composites. Monthly average of the extracted bio-
physical parameters was plotted for the growing season in order to generate a phenology chart for a
particular marsh patch. For Landsat, common cloud-free areas across different months were chosen
for phenological analysis. When common cloud-free areas were hard to find for similar months
across multiple growing seasons, adjacent months were considered for analysis and comparison
between different years. For MODIS, three to four time-series composites were generated for each
growing season month using the 8-day reflectance products. Therefore, monthly phenology was
derived as the average of the three to four biophysical values extracted from the corresponding
composites for specific sites.
RESULTS AND DISCUSSION

Canopy Reflectance Analysis
In situ canopy reflectance spectra, collected from sampling locations after the oil spill, were
acquired along a gradient of oil contamination and grouped into three categories for analysis: no
contamination, moderate contamination, and high contamination (Figure 14.4). As evident in the
spectral response, the characteristic 675 nm red absorption feature was less prominent in highly
contaminated marsh patches when compared to the noncontaminated areas (Mishra et al. 2012).
The overall visible reflectance (400–700 nm) was higher in contaminated vegetation, which implies
reduced effect of the photosynthetic pigments such as chlorophylls a and b. NIR reflectance, on the
other hand, did not show major fluctuations with the varying degree of contamination, implying that
the foliar cell structure of the marsh plants was still intact during the time of data collection. With
time, progressive reduction in photosynthetic pigments, and increased physiological stress, tidal
0.25
Remote sensing reflectance (Rrs) (sr–1)
0.2
0.15
No contamination
Moderate contamination
0.1
High contamination
0.05
0
400 450 500 550 600 650 700 750 800 850 900
Wavelength (nm)
FIGURE 14.4 Remote sensing reflectance (R rs) calculated from selected sampling plots containing
S. alterniflora with similar biophysical parameters across different levels of oil contamination. Solid line
represents vegetation plots with no contamination, while dotted lines represent moderate and high levels of
contamination. The decreasing absorption trends between 600 and 700 nm, with progressive levels of con-
tamination, occur due to decrease in photosynthetic activity due to oil contamination. R rs was estimated as
the ratio of upwelling radiance to downwelling irradiance. (From Mishra, D.R. et al., Remote Sens. Environ.,
118, 176, 2012.)
marsh vegetation developed chlorosis (discoloration) accompanied by progressive defoliation. In

general, it was observed that with oiling, marsh plants first went through the visual sign of contami-
nation (oil coating), followed by reduced photosynthetic activity and increased physiological stress
(yellowing/browning up), and finally cell damage (dead biomass).
Landsat Analysis
VIs derived from Landsat were correlated with the average value of the biophysical parameters
from all sampling locations. Based on comparison of correlation coefficients (R 2) (Figure 14.5),
%RMSE value and (Tables 14.2 and 14.3), and the residual trends (Figure 14.6) of various well-
established VIs, three were selected for the final biophysical model calibration. Enhanced vegeta-
tion index 2 (EVI2) (Huete et al. 2002), wide dynamic range vegetation index (WDRVI; α = 0.1)
(Gitelson 2004), and NDVI (Rouse et al. 1973; Tucker 1979) were chosen as appropriate mod-
els for predicting GBM, GLAI, and VF, respectively. It must be noted that the indices chosen
for comparison for model development use only the red and NIR bands because blue and green
band-based VIs did not produce good results during model calibration. The background water
scattering and residual atmospheric scattering present in blue and green bands of Landsat could
have contributed to the poor calibration results. Time-series composites of biophysical param-
eters were developed using monthly subsets of Landsat data for the southeastern Louisiana tidal
marsh extents. After elimination of cloud cover, nine monthly subsets were reduced to five to six
1600 GBM 1
GLAI
1400 0.9
0.8
Green biomass (g/m 2)
1200
0.7
1000 0.6
GLAI
800 0.5
600 0.4
0.3
400
0.2
200 0.1
R2 = 0.648 R2 = 0.754
0 0
0 0.1 0.2 0.3 0.4 0.5 –1 –0.8 –0.6 –0.4 –0.2 0
(a) EVI2 (b) WDRVI (α = 0.01)
100
90 VF (%)
Vegetation fraction (%)
80
70
60
50
40
30
20
10
R2 = 0.705
0
0 0.2 0.4 0.6 0.8 1
(c) NDVI
FIGURE 14.5 VIs derived from Landsat spectral bands calibrated against corresponding in situ measure-
ments of GBM, GLAI, and VF. Based on linear regression coefficients, different indices were chosen for each
biophysical parameter. EVI2, WDRVI (α = 0.01), and NDVI showed best linear relations with (a) GBM, (b)
GLAI, and (c) VF, respectively.
TABLE 14.2
Correlation Coefficients between Landsat-Derived Vegetation Indices
and Different Biophysical Parameters
WDRVI WDRVI WDRVI
R2 NDVI EVI2 CIred (α = 0.05) (α = 0.1) (α = 0.2) SAVI
GLAI 0.756 0.753 0.751 0.753 0.754 0.756 0.752
VF (%) 0.705 0.740 0.711 0.715 0.716 0.717 0.741
GBM (g/m2) 0.583 0.648 0.647 0.645 0.641 0.631 0.647
TABLE 14.3
Percent Root Mean Square Errors Exhibited by Landsat-Derived
Biophysical Models for Different Biophysical Parameters
WDRVI WDRVI WDRVI
%RMSE NDVI EVI2 CIred (α = 0.05) (α = 0.1) (α = 0.2) SAVI
GLAI 29.611 35.613 26.176 26.477 26.775 27.312 36.431
VF (%) 18.160 21.361 20.838 20.162 19.717 19.168 21.401
GBM (g/m2) 32.702 24.800 26.083 53.606 38.204 32.165 24.812
40 0.3
EVI2 linear residuals for GBM WDRVI residuals for GLAI
30
0.2
20
0.1
10
0 0
15.21
23.75
29.35
34.10
38.13
40.15
41.92
46.17
49.49
49.77
54.56
57.48
60.31
65.52
70.38
75.58
0.04
0.06
0.06
0.06
0.09
0.09
0.11
0.14
0.15
0.15
0.16
0.20
0.21
0.21
0.25
0.31
0.31
0.34
0.41
0.44
0.46
0.47
0.60
–10
–0.1
–20
–0.2
–30
–40 –0.3
(a) (b)
40
NDVI residuals for VF
30
20
10
0
3.72
4.84
5.84
7.19
9.59
10.45
11.64
19.24
20.73
22.28
27.19
32.42
35.81
41.95
47.54
55.57
66.62
–10
–20
–30
–40
(c)
FIGURE 14.6 Residual plots created for Landsat-derived best-fit biophysical models using an independent
set of validation data for (a) GBM, (b) GLAI, and (c) VF.
June 2009 June 2010 June 2011
August 2009 July 2010 August 2011
September 2009 September 2010 September 2011

N
Green leaf area index
0–0.5 0 50 100 150 200 W E
0.5–1 km
1–1.5 S
>1.5
FIGURE 14.7 Monthly composites of GLAI for tidal marsh extents in southeastern LA. These maps were
created by applying the WDRVI (α = 0.01)-based calibration coefficients on USGS’s Landsat 30 m mosaics for
growing seasons of 2009, 2010, and 2011. Composites were created after masking out the cloud cover areas
using an NDVI threshold. The impact of the oil spill can be seen in the September 2010 map, which shows an
increase in highly stressed (red) areas.
cloud-free subsets per growing season for map preparation (Figure 14.7). Visual analysis of the
GLAI composites showed the sharp drop in GLAI during the growing season of 2010, particularly
in the month of September as a result of the spill-induced browning. In 2011, the spatiotemporal
distribution of GLAI appeared to have achieved a normal growth pattern. These map composites
can be used to perform localized or site-specific marsh monitoring such as tracking the progress
of a specific restoration project, recovery of the dieback sites, and growth after human-induced
event (e.g., controlled burning or arson). For example, phenological analysis of an interior marsh
patch in Plaquemines Parish shows very similar growth pattern between the growing seasons of
2009 and 2010, with no sign of damage due to the oil spill in 2010 (Figure 14.8). The normal marsh
phenology is characterized by peak growth and photosynthetic activity in July/August, followed
by onset of dormancy and senescence (reduced photosynthesis) in late September. However, short-
term degradation associated with the oil spill and recovery efforts is reflected in the 2011 growing
season, when the peak growth was delayed to September. Further investigation revealed that this
temporal variation in the peak growth was not attributed to change in other factors as temperature
and precipitation levels remained fairly consistent over the three growing seasons (Mishra et al.
2012).
1100 1.40
2009 2009
1000 2010 1.20 2010
Green biomass (g/sq.m)

900 2011 2011
1.00
800
0.80
700
0.60
600
500 0.40
400 0.20
(a) June July/August September October (b) June July/August September October
90
2009
80 2010
2011
70
60
50
40
30
(c) June July/August September October
FIGURE 14.8 Phenological plots showing seasonal variations in the levels of (a) GBM, (b) GLAI, and (c) VF
in an interior tidal marsh patch of Plaquemines Parish, Louisiana, from 2009 to 2011. The peak growth in 2011
was achieved in September almost a month later than the peaks observed in 2009 and 2010 due to degradation
of salt marshes by the oil spill and subsequent cleanup efforts.
Oil Spill Change Analysis Using Landsat Biophysical Products

The oil spill change detection case study presented here was based on the Landsat-derived bio-
physical models described previously (Mishra et al. 2012). This study was the first quantitative
assessment on the ecological impact of the Deepwater Horizon oil spill, dispersant, and cleanup
efforts on southeastern Louisiana wetlands in terms of their photosynthetic capacity and physi-
ological status. Landsat-derived GBM and CHL map products were analyzed and monitored dur-
ing much of the tidal marsh growing season (May–October) of 2009 (prespill) and 2010 (during
spill) in order to compare and isolate the spill-impacted areas. The initial assessment showed a
significant post-spill increase in areas with reduced biomass and CHL (>400 km 2) during the
2010 growing season compared to just 50–65 km 2 of such reductions during the 2009 growing
season (Figure 14.9). Phenological analysis of growing season revealed a significant decrease
in the magnitude of biomass and CHL during the peak of the growing season. In terms of tidal
marsh health, June was consistently found to be the worst month in 2010 followed by initial
signs of recovery along the fringing marsh (i.e., shoreline area first impacted by oil) toward the
end of the growing season. Interior marsh patches, on the other hand, exhibited persistent signs
of stress toward the end of the growing season, probably due to the time lag effect of the oil
reaching the interior marsh zones. These interior areas also took longer to recover in terms of
their biophysical characteristics because of reduced flushing compared to the shoreline marshes
(Figure 14.10). The marsh change products generated from this study isolated the areas impacted
the most by the spill and provided useful information to coastal managers during recovery and
restoration efforts.
CHL decrease 2009–2010 (>20 mg/m2)
0 9 18 27 36
km
N
GBM decrease 2009–2010 (>200 g/m2)
500
2009
450
2010
400
Decrease in area (sq.km)
350
300
0 9 18 27 36 250
km 200
150
100
50
0
(a) (b) CHL GBM
FIGURE 14.9 Changes observed in biophysical parameters between 2009 and 2010: (a) comparison of
areas experiencing biomass and CHL degradation between June 2009 (pre-oil spill) and 2010 (oil spill).
The areas highlighted in red denote areas experiencing decrease in CHL and GBM levels by more than 20
mg/m 2 and 200 g/m 2, respectively; (b) estimated tidal marsh areas experiencing biomass and CHL reduc-
tion by more than 20 mg/m 2 and 200 g/m 2, respectively. (From Mishra, D.R. et al., Remote Sens. Environ.,
118, 176, 2012.)
MODIS Analysis
After extensive testing of numerous VIs on MODIS data, the WDRVI (α = 0.1) (Gitelson 2004) was
selected for estimating the biophysical parameters (GLAI, CHL, VF, GBM) for the 250 m dataset,
whereas the visible atmospheric resistant index (VARI) (Gitelson et al. 2002) was selected for pre-
dicting the biophysical parameters for the 500 m dataset (Tables 14.4 through 14.7; Figures 14.11
through 14.14). The performance of several red–NIR-based VIs on 250 m data was fairly close to
each other, yet WDRVI was selected for map preparation because it provided the best combination
of R2 and %RMSE for all biophysical parameters. For the 500 m data, VARI outperformed all other
VIs by a large margin. No significant pattern of overestimation or underestimation was observed
in the model residuals (Figures 14.12 and 14.14). Time-series composites were generated for each
biophysical parameter and for both resolutions separately, using the best-fit models for growing
seasons from 2000 to 2011 (see Figures 14.15 through 14.18 for examples). Phenology charts for
160 160
CHL
140 140
GBM
Canopy chlorophyll (mg/m2)

120 120
Biomass (g/m2)
100 100
80 80
60 60
40 40
20 20
0 0
(a) Apr May Jun Jul/Aug Oct
140 120
CHL
120 100
GBM
100
80
Biomass (g/m2)
80
60
60
40
40
20 20
0 0
(b) Apr May Jun Jul/Aug Oct
FIGURE 14.10 Phenological variations of CHL and GBM during the growing season in the (a) fringe and
(b) interior marshes after the BP oil spill. The fringe marshes show a decline in CHL and biomass in June, the
middle of the growing season, due to the effects of the crude oil reaching ashore. The interior marshes showed
a similar decline in the month of July/August due to the time lag of the oil reaching the interior patches.
However, both showed signs of recovery toward the end of the growing season in October. (From Mishra, D.R.
et al., Remote Sens. Environ., 118, 176, 2012.)
TABLE 14.4
Correlation Coefficients between MODIS (250 m)-Derived
Vegetation Indices and Different Biophysical Parameters
WDRVI WDRVI WDRVI
R2 NDVI EVI2 CIred (α = 0.05) (α = 0.1) (α = 0.2) SAVI
GLAI 0.867 0.854 0.597 0.800 0.821 0.846 0.867
VF (%) 0.860 0.847 0.620 0.832 0.845 0.859 0.860
GBM (g/m2) 0.858 0.842 0.560 0.838 0.850 0.864 0.858
CHL (mg/m2) 0.778 0.758 0.390 0.829 0.837 0.843 0.778
Note: Numbers in bold indicate the coefficient of determination values for the selected model.
TABLE 14.5
Percent Root Mean Square Errors Exhibited by MODIS (250 m)-Derived
WDRVI WDRVI WDRVI
%RMSE NDVI EVI2 CIred (α = 0.05) (α = 0.1) (α = 0.2) SAVI
GLAI 19.164 17.441 15.354 14.443 14.967 15.957 19.166
VF (%) 25.040 24.066 29.432 21.769 22.055 22.636 25.036
GBM (g/m2) 21.489 24.969 40.174 19.816 19.616 19.425 21.492
CHL (mg/m2) 28.954 30.729 41.065 25.894 25.817 25.789 28.952
Note: Numbers in bold indicate the percent root mean square error values for the selected model.
TABLE 14.6
Correlation Coefficients between MODIS (500 m)-Derived Vegetation Indices and
Different Biophysical Parameters
WDRVI WDRVI WDRVI CI
R2 NDVI EVI2 CIred (α = 0.05) (α = 0.1) (α = 0.2) SAVI Green VARI
GLAI 0.097 0.128 0.214 0.201 0.184 0.159 0.097 0.000 0.910
VF (%) 0.165 0.192 0.236 0.264 0.249 0.227 0.165 0.000 0.980
GBM (g/m2) 0.477 0.548 0.658 0.644 0.632 0.615 0.478 0.206 0.938
CHL (mg/m2) 0.000 0.000 0.000 0.422 0.446 0.482 0.000 0.000 0.864
Note: Numbers in bold indicate the coefficient of determination values for the selected model.
TABLE 14.7
Percent Root Mean Square Errors Exhibited by MODIS (500 m)-Derived
%RMSE VARI
GLAI 32.474
VF (%) 24.344
CHL (mg/m2) 21.998
GBM (g/m2) 17.340
the last decade were extracted for selected marsh patch locations in the northern Gulf of Mexico
(Figures 14.19 and 14.20). The map composites and the phenology products when used together
can be a powerful tool to analyze the long-term health trend of specific marshes. For example,
the effects of different natural and anthropogenic hazards, such as Hurricanes Katrina (2005) and
Gustav (2008) in Plaquemines Parish and Hurricanes Isidore (2002) and Gustav and the Deepwater
Horizon oil spill (2010) in the Terrebonne Parish in Louisiana, are very much evident from the map
composites and phenology trend (e.g., fluctuations visible in the phenology charts occurring over
the selected time frame of the natural and man-made disturbances are highlighted in Figures 14.19
and 14.20). A sharp decrease was observed in all four biophysical parameters after these disasters
followed by a recovery in the next year’s growing season.
2.5 900
GLAI CHL
800
2 700
CHL (mg/m2 )
600
1.5
500
GLAI
400
1
300
0.5 200
R2 = 0.8206 R2 = 0.84 100
0 0
–1 –0.8 –0.6 –0.4 –0.2 0 –1 –0.8 –0.6 –0.4 –0.2 0
(a) WDRVI (α = 0.1) (b) WDRVI (α = 0.1)
100 1200
VF GBM
90
1000
80
70 800
GBM (g/m2 )
60
VF (%)
50 600
40
30 400
20 200
R2 = 0.845 10 R2 = 0.8502
0 0
–1 –0.8 –0.6 –0.4 –0.2 0 –1 –0.8 –0.6 –0.4 –0.2 0
(c) WDRVI (α = 0.1) (d) WDRVI (α = 0.1)
FIGURE 14.11 WDRVI derived from MODIS 8-day 250 m surface reflectance images calibrated against (a)
GLAI, (b) CHL, (c) percent VF, and (d) GBM. WDRVI (α = 0.01) was chosen as the best index for creating the
biophysical models based on comparisons between regression coefficients and MSE values of numerous VIs.
Application of These Technologies

The time-series composites along with phenological charts provide the end user with both qualita-
tive and quantitative information of the marsh health and physiological status. The biophysical map-
ping methodology developed through this study can be used for identifying critical “hotspots” of
marsh degradation due to factors other than natural disasters (e.g., developmental activities, local-
ized drought, and urban runoff). These products have the potential to facilitate prioritization of
restoration efforts in various ways including identifying areas in need of immediate attention and
monitoring the effectiveness of specific restoration projects by comparing the productivity trend and
species composition pre- and postrestoration. They will provide government regulators and restora-
tion managers with important information regarding mitigation needs. The time-series composites
can be further used to detect progressive change in marsh habitats both inter- and intraseasonally.
Further, the long-term phenology plots provide documentation of the progressive improvement or
decline in their health status following natural or human-induced disasters.
Using similar change detection techniques shown for Landsat data analysis, these time-series
composites can be used to generate valuable quantitative change information occurring both in spe-
cific marsh patches or the entire wetland extent along a particular state or across the northern Gulf
Coast. They can also be used to predict potential future changes that can take place in the ecosys-
tem, especially with the climate change–induced sea-level rise, land-use changes, and developmen-
tal pressures. That information can enable the policy makers and regulators to formulate effective
measures in advance for tidal marsh conservation.
0.6 20
GLAI residuals CHL residuals
15
0.4
10
0.2
5
0 0
0.24
0.43
0.63
0.69
0.78
0.90
1.22
1.31
1.78
20.40
23.60
26.24
34.95
43.81
51.98
56.42
–5
–0.2
–10
–0.4
–15
–0.6 –20
(a) (b)
20 250
VF residuals GBM residuals
200
15
150
10
100
5 50
0 0 268.34
387.12
410.43
435.08
450.15
451.22
579.48
648.90
661.77
724.68
771.13
778.96
782.24
827.02
29.13
34.28
35.81
42.12
52.27
54.68
64.77
71.64
75.83
–5 –50
–100
–10
–150
–15 –200
–20 –250
(c) (d)
FIGURE 14.12 Residual plots for MODIS 250 m-based biophysical models (WDRVI) for (a) GLAI, (b) CHL,
(c) percent VF, and (d) GBM. The residual plots show no specific trend of over- or underprediction.
CONCLUSION
Research projects or surveys covering emergent wetlands have been often limited to habitat
delineation, species mapping, and area loss/gain analysis. Although these surveys have been
efficient in providing information on the presence/absence of wetlands and their change in extent,
assessment of their health and productivity using remote sensing has received little attention.
Detailed information on wetland health condition is required for assessment of previous restora-
tion efforts and prioritization of future actions. The estimated time-series maps and phenological
charts derived from Landsat TM and MODIS imagery offer great potential for assessing marsh
health and productivity. These products can be used in conjunction with different hydrologi-
cal, meteorological, and land-use data to assess the temporal variations of marsh photosynthetic
capacity and carbon sequestration potential. Coastal resource managers and policy makers now
have access to the large-scale maps of the Gulf Coast marsh productivity and can use these prod-
ucts to identify problem areas that should be high priorities for restoration, to evaluate the relative
success of prior restoration efforts, and to locate marshes most likely to be affected by coastal
developmental activities.
The biophysical parameters analyzed in this chapter (e.g., GLAI, VF, CHL, and GBM) are
suitable proxies for photosynthetic capacity, nitrogen content, and physiological status of marsh
vegetation since they are sensitive to natural processes and anthropogenic activities occurring
across the region. These characteristics can differ between species of marsh vegetation due
GLAI CHL
2 1000
1.8 900
1.6 800
1.4 700
CHL (mg/m )
2
1.2 600
GLAI
1 500
0.8 400
0.6 300
0.4 200
0.2 2 100 2
R = 0.91 R = 0.86
0 0
–0.1 –0.05 0 0.05 0.1 0.15 –0.1 –0.05 0 0.05 0.1 0.15
(a) VARI (b) VARI
1200
100 GBM
VF 1100
90 1000
80 900
70 800
60 700
600
VF (%)
50
500
40
400
30
300
20 200
10 2 100 2
R = 0.98 R = 0.9383
0 0
–0.1 –0.05 0 0.05 0.1 0.15 –0.1 –0.05 0 0.05 0.1 0.15
(c) VARI (d) VARI
FIGURE 14.13 VARI derived from MODIS 8-day 500 m surface reflectance images calibrated against (a)
GLAI, (b) CHL, (c) percent VF, and (d) GBM. VARI was chosen as the best index for creating the biophysical
models based on comparisons between regression coefficients and RMSE values of numerous VIs.
to differences in foliar and canopy structures, occurrence and habitat preference, and inher-
ent physiology. Species-dependent biophysical models are generally more accurate and can be
implemented only when marshes in the study area are classified into individual plant communi-
ties. Classification of the wetland habitats into homogenous species communities is impracti-
cal as it requires significant ground data and information, and it is often difficult when using
moderate- and coarse-spatial-resolution sensors such as Landsat TM and MODIS because of
mixed pixel issues. Therefore, using species-invariant biophysical models (models that are not
influenced by species type, structure, and composition), even though less accurate compared to
species-specific models (models for single species homogenous marsh patches), is the best option
to generate high-frequency time-series composites. This study has demonstrated the value of
species-invariant biophysical models, which can be implemented for any marsh habitat irrespec-
tive of its community composition.
Most of the aboveground growth of marsh vegetation in the northern Gulf Coast occurs over
a 7-month period (April–October). Therefore, it serves very little purpose to procure imagery of
marshes between November and March as the vegetation usually enters a period of senescence.
Both moderate- and coarse-resolution satellite sensors such as Landsat and MODIS provide histori-
cal coverage in the form of data archives that can be used in the analysis of interseasonal health and
photosynthetic activity of marsh vegetation. However, it is clear from our work that the accuracy of
the coarse-resolution satellite–based biophysical models for tidal marshes depends primarily on the
accuracy of in situ data collection. One of the primary sources of error is the lack of variability in
0.6 GLAI residuals 250

CHL residuals
200
0.4 150
100
0.2
50
0 0
268.09
326.67
408.13
414.30
420.46
484.91
506.07
723.51
562.13
0.30
0.43
0.46
0.88
0.97
1.31
1.51
1.53
–50
–0.2
–100
–150
–0.4
–200
–0.6 –250
(a) (b)
25 200
VF residuals
20 150 GBM residuals
15
100
10
5 50
0 0
8.35
19.24
22.28
28.91
29.00
33.38
37.30
45.54
48.36
48.55
48.62
48.83
51.63
55.81
322.59
394.17
432.17
470.60
486.21
532.71
606.58
648.90
685.02
707.40
723.17
724.68
841.31
–5 –50
–10
–100
–15
–20 –150
–25 –200
(c) (d)
FIGURE 14.14 Residual plots for MODIS 500 m–based biophysical models (VARI) for (a) GLAI, (b) CHL,
(c) percent VF, and (d) GBM. The residual plots show no specific trend of over- or underprediction.
the ground data representing a coarse pixel of Landsat TM or MODIS sensors. Often the number
of sampling points averaged to represent a pixel is either not sufficient or does not cover all species
present in the pixel, thus introducing errors into the model calibration. Therefore, it is crucial that
a representative and exhaustive ground sampling design be selected and finalized with the help of
either high-resolution aerial photos or Google Earth before the actual field data collection. Other
sampling errors are inherent and unavoidable in any field campaign, which may range from equip-
ment errors to human errors. Careful operation and periodic calibration and maintenance of equip-
ment help reduce those errors.
Finally, the accuracy of such analysis also depends on the satellite data quality and cloud cover.
A critical drawback for Landsat data in particular is the extensive cloud cover often present in the
data during the summer season in the northern Gulf Coast. Sporadic coverage of Landsat coupled
with presence of cloud cover makes it increasingly difficult to obtain cloud-free, clear imagery
for that area for a time-series analysis. Removing cloud cover can be an extremely painstaking
process, as both the cloud cover and the associated shadows need to be eliminated before any sort
of analytical process is performed. These factors reduce the applicability of Landsat TM data to
high-temporal-frequency long-term marsh monitoring studies. On the other hand, MODIS provides
8-day cloud-free surface reflectance products, which is the ideal choice for such analysis despite its
coarser spatial resolution. Nevertheless, the methods employed in this case study clearly demon-
strate the utility of multispectral sensors such as Landsat TM and MODIS for the effective monitor-
ing of tidal marshes.
N
August 5, 2005
W E

0–200
200–400
400–600
600–800
800–1000 0 50 100 150 200
>1000 km
(a)
N
August 5, 2005
W E
Green biomass (g/m2)

0–300
300–600
600–900
900–1200
1200–1500 0 50 100 150 200
>1500 km
(b)
N
August 5, 2005
W E

0–0.5
0.5–1
1–1.5
1–2
2–2.5 0 50 100 150 200
>2 km
(c)
N
August 5, 2005
W E

0–20
20–40
40–60
60–80 0 50 100 150 200
80–100 km
(d)
FIGURE 14.15 Map composites showing the spatial distribution of (a) CHL, (b) GBM, (c) GLAI, and (d) VF
derived from the WDRVI (α = 0.05)-based biophysical model applied to MODIS 250 m 8-day surface reflec-
tance data. Area enclosed by the brown ellipse shows the marsh status pre-Hurricane Gustav in 2008. Red
pixels indicate areas under stress in terms of their biophysical parameters.
N
September 14, 2005
W E

0–200
200–400
400–600
600–800
800–1000 0 50 100 150 200
>1000 km
N
(a)
N
September 14, 2005
W E

0–300
300–600
600–900
900–1200
1200–1500 0 50 100 150 200
>1500 km
(b)
N
September 14, 2005
W E

0–0.5
0.5–1
1–1.5
1–2
2–2.5 0 50 100 150 200
>2 km
(c)
N
September 14, 2005
W E

0–20
20–40
40–60
60–80 0 50 100 150 200
80–100 km
(d)
derived from the WDRVI (α = 0.05)-based biophysical model applied to MODIS 250 m 8-day surface reflec-
tance data. Area enclosed by the brown ellipse shows marsh status post-Hurricane Gustav in 2008. A sig-
nificant increase in red pixels was observed, which indicates severe hurricane-induced damage to the marsh
habitats.
N
August 5, 2008
W E

0–200
200–400
400–600
600–800
800–1000 0 50 100 150 200
>1000 km
(a)
N
August 5, 2008
W E

0–300
300–600
600–900
900–1200
1200–1500 0 50 100 150 200
>1500 km
(b)
N
August 5, 2008
W E

0–0.5
0.5–1
1–1.5
1–2
2–2.5 0 50 100 150 200
>2 km
(c)
N
August 5, 2008
W E

0–20
20–40
40–60
60–80 0 50 100 150 200
80–100 km
(d)
derived from the VARI-based biophysical model applied to MODIS 500 m 8-day surface reflectance data.
Area enclosed by the brown ellipse shows the marsh status pre-Hurricane Katrina in 2005. Red pixels indicate
areas under stress in terms of their biophysical parameters.
N
September 14, 2008
W E

0–200
200–400
400–600
600–800
800–1000 0 50 100 150 200
>1000 km
(a)
N
September 14, 2008
W E

0–300
300–600
600–900
900–1200
1200–1500 0 50 100 150 200
>1500 km
(b)
N
September 14, 2008
W E

0–0.5
0.5–1
1–1.5
1–2
2–2.5 0 50 100 150 200
>2 km
(c)
N
September 14, 2008
W E

0–20
20–40
40–60
60–80 0 50 100 150 200
80–100 km
(d)
derived from the VARI-based biophysical model applied to MODIS 500 m 8-day surface reflectance data.
Area enclosed by the brown ellipse shows the marsh status post-Hurricane Katrina in 2005. A significant
increase in red pixels was observed, which indicates severe hurricane-induced damage to the marshes.
Vegetation fraction (%) Green leaf area network Green biomass (g/sq.ft) Canopy chlorophyll (mg/sq.m)
(b)
(a)
(d)
(c)
0
10
20
30
40
50
60
70
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0
10
20
30
40
50
60
70
0
50
100
150
200
250
300
350
400
April April April April
May May May May
June June June June
July July July July
2000
2000
2000
2000
August August August August
September September September September
October October October October
May May May May
June June June June
July July July July
2001
2001
2001
2001
May May May May
June June June June
July July July July
2002
2002
2002
2002
May May May May
June June June June
July July July July
2003
2003
2003
2003
May May May May
June June June June
July July July July
2004
2004
2004
2004
Using Moderate-Resolution Satellite Sensors

May May May May
June June June June
July July July July
VF
2005
2005
2005
2005

CHL
GBM
GLAI
May May May May
June June June June
July July July July
2006
2006
2006
2006
Katrina

Hurricane

May May May May
June June June June
July July July July
2007
2007
2007
2007


May May May May
June June June June
July July July July
2008
2008
2008
2008
May May May May
June June June June
July July July July
2009
2009
2009
2009
Gustav

Hurricane

May May May May
June June June June
July July July July
2010
2010
2010
2010


May May May May
2011
2011
2011
2011
June June June June
reflectance data). The effect of Hurricane Katrina in 2005 and Gustav in 2008 is highlighted by dotted lines. A
clear decline in the levels of the biophysical parameters was observed during the respective growing seasons.
(d) VF in tidal marshes of Plaquemines Parish, Louisiana, from 2000 to 2010 (250 m MODIS 8-day surface
309
FIGURE 14.19 Phenological variations observed in the magnitude of (a) CHL, (b) GBM, (c) GLAI, and
Vegetation fraction (%) Green leaf area index Green biomass (g/sq.ft) Canopy chlorophyll (mg/sq.m)
(c)
(a)
(b)
(d)
310
0
10
20
30
40
50
60
70
80
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0
10
20
30
40
50
60
70
80
90
0
100
200
300
400
500
600
700
800
900
May May May May
June June June June
July July July July
2000
2000
2000
2000
May May May May
June June June June
July July July July
2001
2001
2001
2001
May May May May
June June June June
July July July July
2002
2002
2002
2002
ACKNOWLEDGMENTS
May May May May
June June June
July July June July
2003
2003
2003
2003

September September September
May May May May
Bill
June June June June

July July July July
2004
2004
2004
2004

Hurricane Isidore Tropical Storm

May May May May
June June June June
July July July July
VF
2005
2005
2005
2005

CHL
GBM
GLAI
May May May May
June June June June
July July July July
2006
2006
2006
2006

May May May May
June June June June
July July July July
2007
2007
2007
2007

parameters was observed in all cases during the respective growing seasons.
May May May May
June June June June
July July July July
2008
2008
2008
2008

of Mexico Research Initiative (GoMRI), and NASA Gulf of Mexico program.

May May May May
June June June June
Hurricane Gustav
July July July July

2009
2009
2009
2009

May May May May
June June June June
July July July July
2010
2010
2010
2010

BP oil spill


May May May May
2011
2011
2011
2011
June June June June
in literature review. The research was funded by the Division of Environmental Biology of the
Mike Brown (University of New Orleans); Christina Mohrman (Grand Bay National Estuarine
We would like to acknowledge Chris Downs, Mike Bryant, Calista Guthrie, and Gary Alon
tance data). The effect of Hurricanes Isidore and Gustav in 2002 and 2008, respectively, Tropical Storm Bill
activities spanning over 3 years. We also thank Ike Astuti (University of Georgia) for assistance
Research Reserve); and Charles Jordan (University of Georgia) for their help in field data collection
Kirui (Jackson State University); Sarah Moore, Ross Del Rio, Lindsay Dunaj, Phil McCarty, and
FIGURE 14.20 Phenological variations observed in the magnitude of (a) CHL, (b) GBM, (c) GLAI, and (d)
Blakeney (Mississippi State University); Paul Merani (University of Nebraska-Lincoln); Philemon

VF in tidal marshes of Terrebonne Parish, Louisiana, from 2000 to 2010 (500 m MODIS 8-day surface reflec-
National Science Foundation, the Northern Gulf Institute at Mississippi State University, the Gulf
in 2003, and BP oil spill in 2010 is highlighted by dotted lines. A clear decline in the levels of the biophysical
REFERENCES
and mapping of wetland vegetation: A review. Wetlands Ecology and Management 18:281–296.
Arizona Remote Sensing Center (ARSC). 2002. Landsat 5 atmospheric and radiometric correction.
Arizona Remote Sensing Center, University of Arizona. Adaptation by S.M. Skirvin, 1996. http://arsc.arid.
arizona.edu/resources/image_processing/Landsat/ls5-atmo.html.
Artigas, F. and I. C. Pechmann. 2010. Balloon imagery verification of remotely sensed Phragmites
australis expansion in an urban estuary of New Jersey, USA. Landscape and Urban Planning
95:105–112.
Bustamante, J., F. Pacios, R. Díaz-Delgado, and D. Aragonés. 2009. Predictive models of turbidity and
water depth in the Doñana marshes using Landsat TM and ETM+ images. Journal of Environmental
Management 90:2219–2225.
Campbell, J. B. 2007. Introduction to Remote Sensing. Guilford, New York, New York, USA.
Chavez, P. S. 1996. Image-based atmospheric corrections-revisited and improved. Photogrammetric
Chmura, G. L., S. C. Anisfeld, D. R. Cahoon, and J. C. Lynch. 2003. Global carbon sequestration in tidal, saline
wetland soils. Global Biogeochemical Cycles 17:1111.
Collin, A., B. Long, and P. Archambault. 2010. Salt-marsh characterization, zonation assessment and mapping
through a dual-wavelength LiDAR. Remote Sensing of the Environment 114:520–530.
Connor, R. F., G. L. Chmura, and C. B. Beecher. 2001. Carbon accumulation in Bay of Fundy salt marshes:
Implications for restoration of reclaimed marshes. Global Biogeochemical Cycles 15:943–954.
Dahl, T. E. 2006. Status and Trends of Wetlands in the Conterminous United States 1998 to 2004. U.S. Fish and
Wildlife Service, Washington, DC.
Davranche, A., G. Lefebvre, and B. Poulin. 2010. Wetland monitoring using classification trees and SPOT-5
seasonal time series. Remote Sensing of the Environment 114:552–562.
Delalieux, S., B. Somers, S. Hereijgers, W. W. Verstraeten, W. Keulemans, and P. Coppin. 2008. A near-infrared
narrow-waveband ratio to determine leaf area index in orchards. Remote Sensing of the Environment
112:3762–3772.
EPA. 2011. Potential framework for reporting on ecological condition and ecosystem services for the 2011
national wetland condition assessment. U.S. Environmental Protection Agency, National Health
and Environmental Effects Research Laboratory, Western Ecology Division, Corvallis, OR. Report
EPA/600/R‐11/104.
Evans, T. L. and M. Costa. 2013. Landcover classification of the Lower Nhecolândia subregion of the Brazilian
Fang, H., W. Li, and R. Myneni. 2013. The impact of potential land cover misclassification on MODIS leaf area
index (LAI) estimation: A statistical perspective. Remote Sensing 5:830–844.
Friess, D. A., T. Spencer, G. M. Smith, I. Möller, S. M. Brooks, and A. G. Thomson. 2012. Remote sensing of
geomorphological and ecological change in response to saltmarsh managed realignment, The Wash, UK.
International Journal of Applied Earth Observation and Geoinformation 18:57–68.
Gallagher, J. L., R. J. Reimold, R. A. Linthurst, and W. J. Pfeiffer. 1980. Aerial production, mortality, and
mineral accumulation-export dynamics in Spartina alterniflora and Juncus roemerianus plant stands in a
Georgia salt marsh. Ecology 6:303–312.
Ghioca-Robrecht, D. M., C. A. Johnston, and M. G. Tulbure. 2008. Assessing the use of multiseason Quickbird
imagery for mapping invasive species in a Lake Erie coastal marsh. Wetlands 28:1028–1039.
Gilfillan, E. S., N. P. Maher, C. M. Krejsa, M. E. Lanphear, C. D. Ball, J. B. Meltzer, and D. S. Page. 1995.
Use of remote sensing to document changes in marsh vegetation following the Amoco Cadiz oil spill
(Brittany, France, 1978). Marine Pollution Bulletin 30:780–787.
Gilmore, M. S., D. L. Civco, E. H. Wilson, N. Barrett, S. Prisloe, J. D. Hurd, and C. Chadwick. 2010. Remote
sensing and in situ measurements for delineation and assessment of coastal marshes and their constituent
species. Chapter 13. In: Y. Wang (ed.). Remote Sensing of Coastal Environment. CRC Press, Boca Raton,
FL, pp. 1–24.
Gilmore, M. S., E. H. Wilson, N. Barrett, D. L. Civco, S. Prisloe, J. D. Hurd, and C. Chadwick. 2008. Integrating
multi-temporal spectral and structural information to map wetland vegetation in a lower Connecticut
River tidal marsh. Remote Sensing of the Environment 112:4048–4060.
Gitelson, A. A. 2004. Wide dynamic range vegetation index for remote quantification of crop biophysical char-
acteristics. Journal of Plant Physiology 161:165–173.
Gitelson, A. A., Y. J. Kaufman, R. Stark, and D. Rundquist. 2002. Novel algorithms for remote estimation of
vegetation fraction. Remote Sensing of the Environment 80:76–87.
Gitelson, A. A., G. P. Keydan, and M. N. Merzlyak. 2006. Three‐band model for noninvasive estimation of
chlorophyll, carotenoids, and anthocyanin contents in higher plant leaves. Geophysical Research Letters
33:L11402.
Gitelson, A. A., A. Vina, V. Ciganda, and D. C. Rundquist. 2005. Remote estimation of canopy chlorophyll
content in crops. Geophysical Research Letters 32:L08403.
Goudie, A. 2013. Characterising the distribution and morphology of creeks and pans on salt marshes in England
and Wales using Google Earth. Estuarine, Coastal and Shelf Science 129:112–123.
Gross, M. F. and V. Klemas. 1986. The use of airborne imaging spectrometer (AIS) data to differentiate marsh
vegetation. Remote Sensing of the Environment 19:97–103.
Hardisky, M. A., F. C. Daiber, C. T. Roman, and V. Klemas. 1984. Remote sensing of biomass and annual net
aerial primary productivity of a salt marsh. Remote Sensing of the Environment 16:91–106.
Hardisky, M. A., M. F. Gross, and V. Klemas. 1986. Remote sensing of coastal wetlands. Bioscience
36:453–460.
Hardisky, M. A., R. M. Smart, and V. Klemas. 1983. Seasonal spectral characteristics and aboveground biomass
of the tidal marsh plant Spartina alterniflora. Photogrammetric Engineering and Remote Sensing 49:85–92.
Harvey, K. R. and G. J. E. Hill. 2001. Vegetation mapping of a tropical freshwater swamp in the Northern
Territory, Australia: A comparison of aerial photography, Landsat TM and SPOT satellite imagery.
International Journal of Remote Sensing 22:2911–2925.
Hester, M. W. and I. A. Mendelssohn. 2000. Long-term recovery of a Louisiana brackish marsh plant commu-
nity from oil-spill impact: Vegetation response and mitigating effects of marsh surface elevation. Marine
Environmental Research 49:233–254.
Hinkle, R. L. and W. J. Mitsch. 2005. Salt marsh vegetation recovery at salt hay farm wetland restoration sites
on Delaware Bay. Ecological Engineering 25:240–251.
Houhoulis, P. F. and W. K. Michener. 2000. Detecting wetland change: A rule-based approach using NWI and
SPOT-XS data. Photogrammetric Engineering and Remote Sensing 66:205–211.
Huete, A. R. 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of the Environment 25:295–309.
Huete, A. R., K. Didan, K. T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the
radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of the
Jensen, J. R. 1996. Introductory Digital Image Processing: A Remote Sensing Perspective. Prentice-Hall Inc.,
New York.
Jensen, J. R. 2007. Remote Sensing of the Environment: An Earth Resource Perspective. Prentice-Hall Inc.,
New York.
Jensen, R., P. Mausel, N. Dias, R. Gonser, C. Yang, J. Everitt, and R. Fletcher. 2007. Spectral analysis of coastal
vegetation and land cover using AISA+ hyperspectral data. Geocarto International 22:17–28.
Kearney, M. S., D. Stutzer, K. Turpie, and J. C. Stevenson. 2009. The effects of tidal inundation on the reflec-
tance characteristics of coastal marsh vegetation. Journal of Coastal Research 25:1177–1186.
Kennish, M. J. 2001. Coastal salt marsh systems in the US: A review of anthropogenic impacts. Journal of
Coastal Research 17:731–748.
Kovacs, J. M., J. M. L. King, F. F. de Santiago, and F. Flores-Verdugo. 2009. Evaluating the condition of
a mangrove forest of the Mexican Pacific based on an estimated leaf area index mapping approach.
Environmental Monitoring and Assessment 157:137–149.
Laba, M., B. Blair, R. Downs, B. Monger, W. Philpot, S. Smith, P. Sullivan, and P. C. Baveye. 2010. Use of
textural measurements to map invasive wetland plants in the Hudson River National Estuarine Research
Reserve with IKONOS satellite imagery. Remote Sensing of the Environment 114:876–886.
Laba, M., R. Downs, S. Smith, S. Welsh, C. Neider, S. White, M. Richmon, W. Philpot, and P. Baveye. 2008.
Mapping invasive wetland plants in the Hudson River National Estuarine Research Reserve using
Quickbird satellite imagery. Remote Sensing of the Environment 112:286–300.
Lefebvre, A., T. Corpetti, V. Bonnardot, H. Quénol, and L. Hubert-Moy. 2010. Vineyard identification and
characterization based on texture analysis in the Helderberg Basin (South Africa). Proceedings of 30th
IEEE International Geoscience and Remote Sensing Symposium, Honolulu, Hawaii, 25–30 July, 2010.
Institute of Electrical and Electronics Engineers, New York, pp. 2852–2855.
Li, L., S. L. Ustin, and M. Lay. 2005. Application of multiple end-member spectral mixture analysis (MESMA)
to AVIRIS imagery for coastal salt marsh mapping: A case study in China Camp, CA, USA. International
Lillesand, T. M., R. W. Kiefer, and J. W. Chipman. 2008. Remote Sensing and Image Interpretation. John Wiley &
Sons Ltd., Hoboken, NJ.
Lopez, R. D., C. M. Edmonds, A. C. Neale, T. S. Slonecker, K. B. Jones, D. T. Heggem, J. G. Lyon, E. Jaworski,
D. Garofalo, and D. Williams. 2004. Accuracy assessments of airborne hyperspectral data for mapping
opportunistic plant species in freshwater coastal wetlands. In: R. S. Lunetta and J. G. Lyon (eds.). Remote
Sensing and GIS Accuracy Assessment. CRC Press, Boca Raton, FL, pp. 253–267.
Lucas, K. L. and G. A. Carter. 2008. The use of hyperspectral remote sensing to assess vascular plant species
richness on Horn Island, Mississippi. Remote Sensing of the Environment 112:3908–3915.
Lunetta, R. S. and M. E. Balogh. 1999. Application of multi-temporal Landsat 5 TM imagery for wetland iden-
tification. Photogrammetric Engineering and Remote Sensing 65:1303–1310.
Lyon, J. J. G. and J. MacCarthy. 1995. Wetlands and Environmental Applications of GIS. CRC Press, Boca
Raton, FL.
Malthus, T. J. and D. G. George. 1997. Airborne remote sensing of macrophytes in Cefni Reservoir, Anglesey,
UK. Aquatic Botany 58:317–332.
Marani, M., S. Silvestri, E. Belluco, N. Ursino, A. Comerlati, O. Tosatto, and M. Putti. 2006. Spatial orga-
nization and ecohydrological interactions in oxygen‐limited vegetation ecosystems. Water Resources
Research 42:W06D06.
McCarthy, J., T. Gumbricht, and T. S. McCarthy. 2005. Ecoregion classification in the Okavango Delta,
Botswana from multitemporal remote sensing. International Journal of Remote Sensing 26:4339–4357.
McCoy, R. M. 2005. Field Methods in Remote Sensing. Guilford Press, New York.
Mishra, D. R., H. J. Cho, S. Ghosh, A. Fox, C. Downs, P. B. T. Merani, P. Kirui, N. Jackson, and S. Mishra.
2012. Post-spill state of the marsh: Remote estimation of the ecological impact of the Gulf of Mexico oil
spill on Louisiana Salt Marshes. Remote Sensing of the Environment 118:176–185.
Mitsch, W. J. and J. G. Gosselink. 2007. Wetlands. John Wiley & Sons, Hoboken, NJ.
MODIS Land, National Aeronautics and Space Administration, Retrieved on December 21, 2012, from http://
modis-land.gsfc.nasa.gov.
National Wetlands Inventory, U.S. Fish and Wildlife Service, Retrieved on August 21, 2012, from http://www.
fws.gov/wetlands/.
Ozesmi, S. L. and M. E. Bauer. 2002. Satellite remote sensing of wetlands. Wetlands Ecology and Management
10:381–402.
Ramsey III, E. and A. Rangoonwala. 2010. Mapping the onset and progression of marsh dieback. Chapter 6.
In: Y. Wang (ed.). Remote Sensing of Coastal Environments. CRC Press, Boca Raton, FL, pp. 123–149.
Ravens, T. M., R. C. Thomas, K. A. Roberts, and P. H. Santschi. 2009. Causes of salt marsh erosion in Galveston
Bay, Texas. Journal of Coastal Research 25:265–272.
Rosso, P. H., S. L. Ustin, and A. Hastings. 2005. Mapping marshland vegetation of San Francisco Bay,
California, using hyperspectral data. International Journal of Remote Sensing 26:5169–5191.
Rouse, J. W., R. H. Haas Jr., J. A. Schell, and D. W. Deering. 1974. Monitoring vegetation systems in the Great
Plains with ERTS. Third ERTS-1 Symposium. NASA Special Publication, NASA SP-351, pp. 309–317.
Rundquist, D., R. Perk, B. Leavitt, G. Keydan, and A. Gitelson. 2004. Collecting spectral data over cropland
vegetation using machine-positioning versus hand-positioning of the sensor. Computers and Electronics
in Agriculture 43:173–178.
Rundquist, D. C., S. Narumalani, and R. M. Narayanan. 2001. A review of wetlands remote sensing and defin-
ing new considerations. Remote Sensing Reviews 20:207–226.
Sadro, S., M. Gastil-Buhl, and J. Melack. 2007. Characterizing patterns of plant distribution in a southern
California salt marsh using remotely sensed topographic and hyperspectral data and local tidal fluctua-
tions. Remote Sensing of the Environment 110:226–239.
Santos-González, C. 2002. Suitability of NDVI AVHRR data for wetland detection. A case study: Kakadu
National Park, Australia. Master thesis. School of Biological, Earth and Environmental Sciences,
University of New South Wales, Paddington, New South Wales, Australia.
Schmidt, K. S. and A. K. Skidmore. 2003. Spectral discrimination of vegetation types in a coastal wetland.
Silliman, B. R., E. D. Grosholz, and M. D. Bertness. 2009. Human Impacts on Salt Marshes: A Global
Perspective. University of California Press, Berkeley, CA.
Sugumaran, R., J. C. Meyer, and J. Davis. 2004. A web-based environmental decision support system (WEDSS)
for environmental planning and watershed management. Journal of Geographical Systems 6:307–322.
Thomson, A. G., R. M. Fuller, T. H. Sparks, M. G. Yates, and J. A. Eastwood. 1998. Ground and airborne
radiometry over intertidal surfaces: Waveband selection for cover classification. International Journal of
Tiner, R. W. 1996. Wetland definitions and classifications in the United States. In: National Water Summary
on Wetland Resources, US Geological Survey, Water-Supply Paper 2425, Washington, DC, pp. 27–34.
Tiner, R. W. 2013. Tidal Wetlands Primer: An Introduction to their Ecology, Natural History, Status, and
Conservation. The University of Massachusetts Press, Amherst, MA.
Tucker, C. J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote
Sensing of Environment, 8:127–150.
USGS Global Visualization Viewer, Retrieved November 21, 2012, from http://glovis.usgs.gov.
Wang, Y., M. Christiano, and M. Traber. 2010. Mapping salt marshes in Jamaica Bay and terrestrial vegetation
in Fire Island National Seashore using QuickBird satellite data. Chapter 9. In: Y. Wang (ed.). Remote
Sensing of Coastal Environments. CRC Press, Boca Raton, FL, pp. 191–208.
Ward, D. P., S. K. Hamilton, T. D. Jardine, N. E. Pettit, E. K. Tews, J. M. Olley, and S. E. Bunn. 2013. Assessing
the seasonal dynamics of inundation, turbidity, and aquatic vegetation in the Australian wet–dry tropics
using optical remote sensing. Ecohydrology 6:312–323.
Weather Underground, Retrieved October 21, 2012, from http://www.wunderground.com/.
Weis, J. S. 2010. Salt marsh. In: C. J. Cleveland (ed.). Encyclopedia of Earth. Environmental Information
Coalition, National Council for Science and the Environment, Washington, DC.
Yan, G., J. F. Mas, B. H. P. Maathuis, Z. Xiangmin, and P. M. Van Dijk. 2006. Comparison of pixel‐based and
object‐oriented image classification approaches—A case study in a coal fire area, Wuda, Inner Mongolia,
China. International Journal of Remote Sensing 27:4039–4055.
Yang, J. and F. J. Artigas. 2010. Mapping salt marsh vegetation by integrating hyperspectral and LiDAR remote
sensing. Chapter 8. In: Y. Wang (ed.). Remote Sensing of Coastal Environments. CRC Press, Boca Raton,
FL, pp. 173–187.
mapping in the Everglades from hyperspectral imagery. Remote Sensing of the Environment 124:310–320.
Zomer, R. J., A. Trabucco, and S. L. Ustin. 2009. Building spectral libraries for wetlands land cover classifica-
tion and hyperspectral remote sensing. Journal of Environmental Management 90:2170–2177.
15 Great Lakes Coastal
Wetland Mapping
Laura L. Bourgeau-Chavez, Zachary M. Laubach,
Anthony J. Landon, Elizabeth C. Banda,
Michael J. Battaglia, Sarah L. Endres, Mary Ellen Miller,
Robb D. Macleod, and Colin N. Brooks
CONTENTS
Introduction..................................................................................................................................... 316
Why Inventory and Monitor Great Lakes Coastal Wetlands?......................................................... 318
Remote Sensing of Great Lakes Wetlands...................................................................................... 320
Case Study 1: Mapping an Invasive Species Using Synthetic Aperture Radar.............................. 321
Introduction................................................................................................................................ 321
Mapping Methods...................................................................................................................... 321
Results for Case Study 1............................................................................................................ 324
Use of Data for Management and Decision Support................................................................. 326
Case Study 2: Mapping Coastal Wetlands Using a Hybrid Radar/Multispectral Approach........... 327
Introduction................................................................................................................................ 327
Mapping Approach..................................................................................................................... 327
Case Study 2: Summary and Significance................................................................................. 328
Case Study 3: Forested Wetland Mapping with SAR..................................................................... 329
Introduction................................................................................................................................ 329
Methodology.............................................................................................................................. 331
Case Study 3: Results................................................................................................................. 333
Summary and Next Steps........................................................................................................... 335
Case Study 4: Advanced Multisensor Approach for Updating NWI in Minnesota........................ 336
Introduction................................................................................................................................ 336
Improved Data Sources and Software........................................................................................ 337
Mapping Methodology for NWI Updates.................................................................................. 337
Image Segmentation.............................................................................................................. 337
Field Training Data................................................................................................................ 339
Potential NWI Computer-Generated Classification Process................................................. 339
Photo Interpretation: Object Editing/Manual Delineation....................................................340
Case Study 4: Summary.............................................................................................................340
Conclusion......................................................................................................................................340
References....................................................................................................................................... 341
315
INTRODUCTION
Coastal wetlands of the Great Lakes are dynamic biophysical systems that are connected to the
lakes through linkages of surface water, groundwater, or both. Coastal wetlands include wetlands
adjacent to the lakeshore as well as those complexes found along tributary rivers and streams that
function in concert with the Great Lakes water regimes (Nature Conservancy 1994). Nearly 20% of
all freshwater on earth is found in the Great Lakes Basin (245,759 km2 of surface water; National
Oceanic and Atmospheric Administration [NOAA] GLERL—www.glerl.noaa.gov), which is com-
prised of the five Great Lakes (Superior, Michigan, Huron, Erie, and Ontario), its connecting chan-
nels (St. Mary’s, St. Clair, Detroit, St. Lawrence, and Niagara Rivers), and Lake St. Clair (between
Lakes Erie and Huron). With an overall shoreline length of 17,017 km, the Great Lakes Basin ulti-
mately connects to the Atlantic Ocean through the St. Lawrence River system (Figure 15.1). The
basin includes watersheds within two Canadian provinces (Ontario and Quebec) and eight U.S.
states (Minnesota, Wisconsin, Illinois, Indiana, Michigan, Ohio, Pennsylvania, and New York)
making it an internationally important and diverse freshwater system.
The influence of large lake processes including water-level fluctuations, wave activity, and seiches
(wind tides) distinguishes the Great Lakes coastal wetlands from others in the contributing water-
sheds (Environment Canada 2002). Coastal wetlands are hydrologically connected to the Great
Lakes, and even small changes in lake levels can have a direct effect on patterns of inundation. Lake-
driven fluctuations in wetland hydrology can occur seasonally or over longer durations, causing
wetland vegetation types to shift in relation to changes in water depth (Environment Canada 2002).
FIGURE 15.1 Map of the Great Lakes Basin including eight states (Minnesota, Wisconsin, Illinois, Indiana,
Ohio, Michigan, Pennsylvania, and New York) and one province (Ontario). Yellow lines denote watershed
boundaries for each lake.
Great Lakes Coastal Wetland Mapping 317
Present-day lake levels typically vary seasonally within 0.5 m range, with fluctuations of 1–2 m
occurring during extended dry or wet periods (Gronewold et al. 2013). Intraseasonal fluctuations
of water levels reflect the annual climate-driven hydrologic cycle, which is characterized by higher
water levels during the spring and early summer and lower water levels during the remainder of the
year. Monthly mean water-level deviations from annual means are shown in Figure 15.2. The high-
est lake levels usually occur in June on Lakes Ontario and Erie, in July on Lake Michigan–Huron,
and in August or September on Lake Superior (Wilcox et al. 2007). The lowest lake levels usually
occur in December on Lake Ontario, in February on Lakes Erie and Michigan–Huron, and in
March on Lake Superior. Based on the monthly average water levels, the magnitudes of unregulated
seasonal fluctuations are relatively small, averaging about 0.4 m on Lakes Superior and Michigan–
Huron, about 0.5 m on Lake Erie, and about 0.6 m on Lake Ontario (Wilcox et al. 2007). The long-
term variations in water levels by lake are presented in Figure 15.3 from 1860 to the present (see the
Great Lakes Water Level Dashboard at http://www.glerl.noaa.gov/data/now/wlevels/levels.html for
more information).
Since wetland plants have species-specific adaptations to water depth ranges and seasonality
of flooding, a zonation of ecological associations typically forms along the lake shore from deep
water to dry land (Figure 15.4). Not all zones depicted in Figure 15.4 are present or well developed
in every coastal wetland complex, as variation is contingent on site-specific conditions. In addition,
when hydrologic regimes change, there is typically a shift in species composition within these zones
in response to these changes, and it is not uncommon to have interannual variation in the physical
location of the zones within a given wetland complex.
Besides hydrology (i.e., the frequency, duration, and timing of flooding), other factors also con-
tribute to regionalized patterns of vegetative zonation including shoreline configuration, glacial and
bedrock geology, climate, and land use (Albert 2003) (for a geological history of the Great Lakes, see
Hough 1958; Cvancara and Melik 1961; Wold et al. 1981; Larsen and Schaetzl 2001). Consequently,
coastal wetlands of the Great Lakes represent a diverse set of ecosystems ranging from marshes,
freshwater estuaries, deltas, lagoons, and lake plain prairie to fens, bogs, shrub or treed swamps,
and forested dune and swale complexes (Albert 2003). With such a diverse set of wetland types and
factors that influence them, mapping and monitoring Great Lakes coastal wetlands with a single
Great Lakes seasonal water-level fluctuations

0.4
Water-level difference from mean lake level (m)
Superior
0.3 Michigan–Huron
Erie
0.2 Ontario
0.1
–0.1
–0.2
–0.3
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
FIGURE 15.2 Great Lakes seasonal water-level fluctuations by lake basin (Superior, Michigan–Huron com-
bined, Erie, and Ontario) plotted as the water-level difference from the mean lake levels as calculated over the
historic record (1860–2010). (From GLERL database; Gronewold, A.D. et al., Environ. Model. Softw., 49, 34,
2013, http://dx.doi.org/10.1016/j.envsoft.2013.07.003.)
185
Superior
184
183
182
181
176
Erie
Water elevation above sea level (m)
175
174
173
172
76
Ontario
75
74
73
72
178
Michigan–Huron
177
176
175
174
1860
1868
1876
1885
1893
1901
1910
1918
1926
1935
1943
1951
1960
1968
1976
1985
1993
2001
2010
FIGURE 15.3 Historic Great Lakes water levels from 1860 to 2010 by lake basin (Superior, Michigan–
Huron combined, Erie, and Ontario). (From GLERL database; Gronewold, A.D. et al., Environ. Model. Softw.,
49, 34, 2013, http://dx.doi.org/10.1016/j.envsoft.2013.07.003.)
data source presents difficulty. Advanced remote sensing methods that use a combination of data
sources and methods from multiple seasons are necessary to capture the hydrologic and phenologi-
cal variation that characterizes the diversity of the region’s wetlands.
This chapter reviews the need for wetland mapping of Great Lakes coastal wetlands and summa-
rizes applications of remote sensing for this purpose. It includes presentation of some novel remote
sensing approaches to mapping Great Lakes coastal wetlands through description of salient features
of selective case studies.
WHY INVENTORY AND MONITOR GREAT LAKES COASTAL WETLANDS?

Coastal wetlands of the Great Lakes form a transition zone between the lakes and the upland zone,
acting as a vital link between land and water. These coastal wetlands serve major ecological and
economic roles contributing to the overall health and maintenance of the Great Lakes. Coastal
wetlands act as natural filtration systems and are locations of important biogeochemical transforma-
tions (Mitsch and Gosselink 2000; Zedler and Kercher 2004). They store, produce, and transform
nutrients and organic material. These highly productive wetlands provide habitat, sources of food,
and breeding grounds for many common and regionally rare bird, mammal, reptile, amphibian,
and invertebrate species (Albert 2003). Despite their importance, more than two-thirds of wetlands
in the Great Lakes region have been drained for agriculture and other development (Dahl 1990),
and the remaining wetlands are subject to a number of threats, including land use change, climate
Upland Coastal wetland Lake
Swamp Marsh
Dense Sparse Sub-

Forest Shrub Meadow
emergent emergent emergent
a
b
a - High water level of lake c
b - Average water level of lake
c - Low water level of lake
Ground-water levels
FIGURE 15.4 Typical ecosystem zonation pattern that arises in Great Lakes coastal wetlands due to
variations in patterns of flooding and wet soils from the open water inland to dry land. (From http://www.
ijc.org/loslr/en/background/w_wetlans.php; Environment Canada [Wilcox, D.A. et al.], Where Land
Meets Water—Understanding Wetlands of the Great Lakes, Canadian Wildlife Service, Toronto, Ontario,
Canada, 2002.)
change, and invasive species. Although wetlands serve vital ecological functions, due to their lim-
ited capacity for adaptation (high levels of endemic specialization and niche adaptation), they are
highly vulnerable to both climatic (IPCC 2008) and anthropogenic changes. Threats to wetlands
and wetland processes put vital ecosystem services at risk, thus making wetlands one of the most
sensitive and crucial land cover types subject to climate change.
Over the past several decades, three critical changes in the landscape of the Great Lakes region
have strongly influenced changes in wetland condition: (1) increases in urbanization and suburban
sprawl (Brown et al. 2005), (2) changes in agricultural practices, and (3) widespread invasion by
exotic wetland plant species. Land use change from forests and farmland into large expanses of
residential land with impervious surfaces has resulted in loss of wetland area and increases in runoff
volume and intensity (Churkina et al. 2010). Changes in agricultural practices, such as no-till farm-
ing, have reduced runoff volume and sediment transport; however, such no-till practices require that
increased concentrations of herbicides and pesticides be used. Other changes in farming practices
that influence wetland condition include increased use of manure spreading, larger animal opera-
tions, changes in timing of fertilizer application, and efforts to increase stream buffers and other
soil conservation measures. The widespread invasion of exotic wetland plant species has resulted in
large, dense monocultures that exclude native vegetation and alter ecosystem function.
Complexity in the Great Lakes attributed to humans, climate, and natural variation underlies the
importance of mapping and monitoring not only wetlands but adjacent land cover and land uses.
In order to evaluate and manage a complex ecosystem, the size of the Great Lakes coastal zone,
TABLE 15.1
List of Existing Wetland Maps Created for Portions of the Great Lakes
Map Reference(s) Description URL
National Wetlands Ducks Unlimited (2008), Divides wetland and deepwater http://www.fws.gov/
Inventory (NWI) Wilen and Tiner (1989), habitats into systems, subsystems, wetlands/Wetlands-
Cowardin et al. (1979) and classes Mapper.html
Wisconsin Wetland Wisconsin Department Divides wetland habitats into http://dnr.wi.gov/topic/
Inventory (WWI) of Natural Resources systems, subsystems, and classes wetlands/inventory.html
NOAA C-CAP Dobson et al. (1995), Includes wetland and upland classes http://www.csc.noaa.gov/
Klemas et al. (1993) of coastal watersheds dataviewer/index.html#
Canadian Wetland Environment Canada Uses the five major classes of the http://maps.ducks.ca/cwi/
Inventory (CWI) (2009), Reimer (2009), CWCS system for wetlands (bog,
Grenier et al. (2007), fen, swamp, marsh, shallow water)
Fournier et al. (2007)
Great Lakes Coastal Ingram et al. (2004) Uses hydrogeomorphic classification; http://projects.glc.org/
Wetland Inventory compiled from the NWI, Ohio wetlands/inventory.html
Wetland Inventory, USFWS reports
and hard copy maps, and Ontario
Great Lakes Coastal Wetland Atlas
Note: Only the Great Lakes Coastal Wetland Inventory is comprehensive for the entire coastal Great Lakes.
annual mapping and monitoring through remote sensing is necessary. Several maps of wetlands or
land use, land cover (LULC) exist on a state-by-state or provincial basis from various time periods
(Table 15.1). There are also several national maps created for the United States and Canada, but
none of these maps cross the binational political border. Only the Great Lakes Coastal Wetland
Inventory stretches across the binational political boundary to create a basin-wide coastal inven-
tory map. However, this map is actually a fusion of maps from multiple sources, resolutions, and
methodologies (Table 15.1).
REMOTE SENSING OF GREAT LAKES WETLANDS

Maps and other spatial data inventories are restricted in dimension and represent static conditions
at a given point in time. However, environmental changes can occur rapidly, and regular updates
are needed to maintain the accuracy of cross-sectional data. Updating high-resolution maps on a
routine basis is generally cost and time prohibitive. Completing an area, the size of the Great Lakes
Basin can require multiple years of effort at huge expense. Therefore, a trade-off exists between
resolution and timeliness, such that timely change maps can be efficiently created at a coarser reso-
lution over the entire basin, while high spatial resolution maps can generally be completed in the
same time frame for a subset of the basin area. A variety of methods have been developed in an
attempt to map with higher accuracy at coarser spatial resolution or with greater efficiency at finer
resolution. Advances in sensing instruments and image processing software have allowed signifi-
cant improvement in mapping capability, efficiency, and accuracy.
Wetlands have historically been one of the most difficult ecosystems to classify using remotely
sensed data. This challenge arises in part due to the high variability in wetland morphology, sea-
sonality, and biodiversity. Transitional patterns of vegetative growth driven by hydrologic regimes
contribute to a patchwork of structural and spectral variation within wetlands. Effective use of
remote sensing for any environmental application requires an understanding of both the sensor’s
characteristics and interactions between the sensor and the environment, a relationship not to be
understated, especially when mapping wetlands.
Researchers have begun to utilize multiple sensors (thermal, optical, light detection and ranging
[LiDAR], and SAR) and multiple seasons of imagery to better detect wetlands, such as the Canadian
Wetlands Inventory and the Great Lakes Coastal Wetlands Mapping program (see “Case Study 2”)
funded by the U.S. Environmental Protection Agency (EPA) under the Great Lakes Restoration
Initiative (GLRI) (http://greatlakesrestoration.us/). The use of sensor fusion techniques across mul-
tiseason datasets has enabled taxonomic-specific mapping of vegetation, which is otherwise often
assumed to be best conducted with high-resolution hyperspectral data. Using multiple seasons of
data allows improved discrimination of wetland types by assessing the seasonal variations in phe-
nology related to changes in plant structure and chlorophyll concentrations as well as seasonal pat-
terns of inundation. A review of advanced methodologies is discussed in case studies later in this
chapter as well as in other chapters of this book. Sensor fusion and multiseason data combination
methodologies do not necessarily replace traditional remote sensing approaches, but augment and
improve upon them.
CASE STUDY 1: MAPPING AN INVASIVE SPECIES

USING SYNTHETIC APERTURE RADAR
Introduction
Throughout the past two centuries, a variety of nonnative species have invaded the Great Lakes
region. Many of these species are benign while others exhibit an aggressive, noxious, and persistent
effect. Understanding the current extent of problematic invasive species is critical for management
and control, for determining areas at risk from invasion, and for assessing potential impacts on eco-
system services. One particular invasive wetland plant species—Phragmites australis—has been a
severe detriment to Great Lakes coastal wetlands. The invasive form of Phragmites is now prevalent
in coastal marshes, exploiting rapid water-level changes and vast areas of exposed habitat. Once
established, this invasive plant is capable of forming dense, tall (up to 5 m) monocultures that are dif-
ficult to control without continuous management (Figure 15.5). In the Great Lakes region, invasion
by Phragmites has had several negative impacts on ecosystem services including displacement of
native wetland vegetation and reduction of habitat quality and biologic diversity (Findlay et al. 2002).
Management and control of such a pervasive and tenacious invader requires knowledge of the
species distribution. Numerous environmental organizations have been working to locate and record
the extent of Phragmites on local plots of land, but a U.S. basin-wide dataset was not available until
mapping began in 2010 by Bourgeau-Chavez et al. (2013).
Mapping Methods
Traditional vegetation mapping relies on differences in spectral signatures between plant species in
the optical and infrared (IR) frequencies. Optical–IR sensors operate at wavelengths on the microm-
eter scale and measure energy naturally reflected from earth surfaces. Optical and IR shortwave
radiant energy reflectance from vegetation varies depending on features at the cellular level (e.g.,
chlorophyll and leaf moisture) as well as variations in surface or background reflectance (e.g., soil
type and water). Sensors that utilize microwave frequencies have a much longer wavelength (mm to
cm scale) than optical and IR sensors. The longer wavelength microwave energy backscattered and
received by a synthetic aperture radar (SAR) sensor is sensitive to vegetative structure and biomass,
dielectric properties (i.e., moisture content) of vegetation and soils, surface roughness, and patterns
of inundation (Bourgeau-Chavez et al. 2009). In addition, microwave energy is capable of penetrat-
ing a vegetation canopy to detect the presence or absence of a flooded surface, thereby allowing for
the distinction between wetland and upland ecosystem types. These SAR characteristics allow for
distinguishing between wetland plants; for example, the high biomass, tall, dense Phragmites stands
from the typical lower biomass native herbaceous wetland vegetation.
FIGURE 15.5 Photo of a dense monoculture of invasive Phragmites australis along Lake Michigan (near
Muskegon, MI). Note: Person in photo is 5 ft tall. (From Bourgeau-Chavez, L.L. et al., J. Great Lakes Res.
Suppl., 39, 65, 2013; Photo by Elizabeth Banda.)
SAR is an active sensor, emitting microwave radiation and measuring the energy backscattered
from the elements being imaged (e.g., wetland plants). Phased array–type L-band synthetic aper-
ture radar (PALSAR) is an L-band (~24 cm wavelength) radar that orbited the earth onboard the
Japanese platform Advanced Land Observation System (ALOS) from 2006 to 2011. The PALSAR
sensor operated in different modes including single, dual, and quadrature polarization. The fine
beam dual (FBD) polarization data of ALOS PALSAR were determined to be the most valu-
able for mapping on a regional basis due to their moderate resolution (20 m) and two polarimetric
channels—horizontal send and receive (L-HH) and horizontal send and vertical receive (L-HV).
The L-HH-polarized data have been found to be most useful for detection of flooding beneath vege-
tation canopy (Hess et al. 1995), while L-HV is more sensitive to differences in biomass (Bourgeau-
Chavez et al. 2009). Figure 15.6 shows how seasonal PALSAR imagery was used to detect invasive
Phragmites and distinguish it from other wetland and nonwetland cover types. The three seasons
of data allowed for distinguishing between the high biomass Phragmites with standing dead stems
that persist across seasons, as opposed to other wetland vegetation that have lower biomass and
readily decompose over winter. The three image dates also show differences in inundation and
biomass. While any single date image provides useful information, combining all three dates into a
false-color composite allows clearer distinction of not only the Phragmites dominated sites but other
wetland cover types (Figure 15.7).
Bourgeau-Chavez et al. (2013) used a combination of dual polarization, L-band (~24 cm wave-
length), ALOS PALSAR data, and field documentation for mapping large monotypic stands of
PALSAR Jun 08 PALSAR Oct 09 PALSAR Apr 10
Agriculture
(row crop) Sedge/grasses Typha
Floating
aquatic
Phragmites
dominant (moist)
Phragmites
dominant (wet) Lake Erie
Forest
6.5 km
7.5 km
FIGURE 15.6 Summer (red, June 2008), fall (green, October 2009), and spring (blue, April 2010) PALSAR
L-HH images of PALSAR L-HH data (top black and white images) collected over a diked wetland at Pointe
Mouillee State Game Area on Lake Erie were used to create a red, green, and blue false-color composite
(bottom images). The bottom-left image is a subset of the full PALSAR scene (shown in bottom right). This
false-color image highlights the differences in water levels and phenological condition of the vegetation over
the seasons allowing the image analyst to distinguish vegetation types (e.g., Phragmites is primarily orange in
this composite). (From Bourgeau-Chavez, L.L. et al., J. Great Lakes Res. Suppl., 39, 65, 2013.)
22 km
22 km
30 km 30 km
FIGURE 15.7 Three-date false-color composite of PALSAR L-HH images (red, June 2008; green, October
2009; blue, April 2010) over Pointe Mouillee on Lake Erie (left image) used in the classification scheme to
detect and map Phragmites australis (shown in green in right image).
Phragmites. Approximately two hundred eighty 70 km × 70 km PALSAR images collected between
2008 and 2010 and field data collected from over 1175 locations were used to create a U.S. Great
Lakes Basin–wide map of invasive P. australis. Only the large stands of the invasive variety were
mapped, and the native, noninvasive variety does not form large dense stands of high biomass and
thus was not mapped.
To train and validate the Phragmites map, a large field campaign was carried out to collect
information on wetland type and dominant cover at randomly selected locations within coastal
emergent wetlands across the U.S. coastal Great Lakes Basin (field sites were selected based on
the emergent classes of the U.S. Fish and Wildlife Service’s National Wetlands Inventory [NWI]
maps). To match the minimum mapping unit (mmu) of the map product (0.2 ha, 0.5 acre), all sites
were sampled in 0.2 ha (40 m × 50 m) increments, which were critical to the mapping particu-
larly because of inherent speckle noise in the SAR imagery. SAR scattering from two or more
individual components within a resolution cell results in interference of the backscattered signals
(Raney 1998). The result of this scattering is random constructive and destructive interference,
manifesting itself in bright and dark neighboring pixels producing a “salt and pepper” effect. Such
coherent addition of backscatter from multiple scatterers in the same resolution cell is referred to
as speckle (Lee and Pottier 2009). Due to the speckle issue, a single pixel in SAR imagery cannot
be used to measure features on the ground; instead a spatial average is often used. Speckle can
be minimized by multilook filtering, by averaging adjacent pixels, or alternatively by applying
speckle filters. Filtering or averaging results in improved radiometric resolution at the expense of
reduced spatial resolution.
While the mapping was focused on PALSAR imagery alone, the utility of multiple sources
of data (Landsat and air photos) was incorporated into the overall mapping project. Landsat data
were used for checks on georectification, while aerial imagery from sources including the National
Agricultural Imagery Program (NAIP) was used for field planning and data collection as well as for
identification of Phragmites and interpretation of wetlands.
Each area of interest (AOI—defined in part by the 70 km × 70 km PALSAR scene extent) was
mapped separately using field data and air photo comparison for training. An unsupervised clas-
sification algorithm was used to group pixels together that have statistically similar cell values
in the six-band datasets. The six-band seasonal datasets (spring, summer, fall L-HH, and L-HV)
were processed through the isodata unsupervised classification routine (ERDAS 2010) with a range
of classes (32–64), multiple iterations (20 maximum), and a convergence threshold of 0.95. This
was followed by the application of group partitioning of similar pixel clusters. For each image
set, a manual analysis of the unsupervised classifications was performed for the ground-truthed
areas. Around each field location, the unsupervised class was recorded for association with the
wetland type observed in the field. Classes that contained potential Phragmites were extracted and
run through a supervised classifier. The selected process was based on the work of Bourgeau-
Chavez et al. (2009) using PALSAR to map wetlands near Lake St. Clair and is fully described in
Bourgeau-Chavez et al. (2013).
In an effort to minimize exclusion of Phragmites, classification was conducted to err on the side
of commission rather than omission while filtering out obvious upland and other confusion areas
in postprocessing. Areas smaller than the project’s mmu produced by the supervised classification
were grouped with their surrounding cells (12 minimum) and assigned the same class. Once the
classification was complete, agricultural confusion pixels were filtered out using selected cover
types from 2006 NOAA Coastal Change Analysis Program (C-CAP) and 2009 Cropland Data
Layer (CDL) products in ArcGIS. The final map is presented in Figure 15.8.
Results for Case Study 1

The final maps are binary and thus only two classes were evaluated in the accuracy assessments,
“Phragmites” and “other” land cover types (primarily wetlands). Accuracy was determined using
N
W E
S
Potential invasive Phragmites
0 75 150 300 Miles
0 87.5 175 350 km
FIGURE 15.8 Overview of the distribution of invasive Phragmites along the U.S. Great Lakes coastline,
with a minimum mapping unit (mmu) of 0.2 ha, as mapped with PALSAR L-band data from 2008 to 2010
(white). The area is labeled as “potential Phragmites” recognizing that although this map represents the range
of PALSAR spectral signatures observed in monocultures of invasive Phragmites with a 0.2 ha mmu, there
may be some confusion with other types (e.g., high biomass Typha spp.) or omission due to ongoing control
efforts and rapid spread beyond the temporal resolution of this work. (From Bourgeau-Chavez, L.L. et al.,
J. Great Lakes Res. Suppl., 39, 65, 2013.)
the randomly selected validation sites, withheld from the original mapping process, in comparison
with the final mapped invasive Phragmites products. Assessments were conducted basin wide as
well as for individual lakes (see Bourgeau-Chavez et al. 2013 for details). Three estimates of accu-
racy were calculated, user’s accuracy, producer’s accuracy, and overall accuracy. User’s accuracy
is a measure of how accurately a classification performed in the field (errors of commission), while
producer’s accuracy is a measure of how accurately the analyst classified the image data (errors of
omission) (Congalton and Green 2008). The overall accuracy provides the summary of correctly
classified validation locations. For the entire Great Lakes Basin, the overall accuracy was 87%, with
86% producer’s accuracy (i.e., errors of omission) for Phragmites with greater than 90% ground
cover and 43% user’s accuracy (i.e., errors of commission), thus a high commission error. Although
the target was invasive Phragmites-dominant stands with greater than 90% cover, it was determined
that many stands of at least 50% cover were detectable. Where Phragmites had greater than 50%
ground cover, the producer’s accuracy dropped to 70%, and the user’s accuracy increased to 58%
(Table 15.2).
The area of Phragmites mapped was greatest in Lake Huron (10,395 ha) followed by Lake Erie
(8,233 ha) and Lake Michigan (6,002 ha). Very little Phragmites was detected on the shores of Lake
Ontario (13 ha), and no large monotypic stands of Phragmites were detected during field activities
or image analysis in coastal Lake Superior.
TABLE 15.2
Accuracy Assessment for the Entire U.S. Great Lakes Basin for Invasive
Phragmites Stands with Greater than 90% Cover
PALSAR Class
Phragmites Other Total Producer’s Accuracy
Field Observation
(Omission Error)
Phragmites 57 9 66 86
Other 75 527 602 88
Total 132 536 668
User’s accuracy (commission error) 43 98 87
Source: Bourgeau-Chavez, L.L. et al., J. Great Lakes Res. Suppl., 39, 65, 2013.
Notes: User’s accuracy is a measure of how accurately a classification performed in the field (errors
of commission) while producer’s accuracy is a measure of how accurately the analyst classi-
fied the image data (errors of omission; Congalton and Green, 2008).
Use of Data for Management and Decision Support

The PALSAR-derived map representing the distribution of large, mature, monotypic invasive
Phragmites stands along the U.S. coast of the Great Lakes is the first U.S. basin-wide coastal map
of its kind. This highly accurate dataset provides a benchmark that allows national, regional, and
local managers to visualize the extent of Phragmites invasion in the Great Lakes and strategically
plan efforts to manage existing populations and minimize new colonization through a web-based
decision support tool developed by the U.S. Geological Survey (USGS) (http://cida.usgs.gov/glri/
phragmites/).
Phragmites treatment and control operations are underway across the Great Lakes region by a
diverse suite of stakeholders, though management is often limited in geographic extent. Federal,
state, local, and private organizations charged with conserving habitat for wildlife, biodiversity,
recreational use, and public safety invest significant financial resources annually to control the rapid
expansion of this species into critical wetland areas. Approximately $6 million has been supplied
for the first 3 years of the GLRI for remediating invasive Phragmites in the Great Lakes Basin. In
the State of Michigan, it is estimated that the Department of Transportation spends 10% of its main-
tenance budget on managing Phragmites in roadside ditches, with similar budgets in neighboring
states. While management can be effective, the current level of investment is neither sustainable nor
ecologically desirable.
Management is further challenged by a lack of coordination and communication among
stakeholders. The Great Lakes Commission (GLC), in partnership with the USGS-Great Lakes
Science Center (GLSC), initiated the Great Lakes Phragmites Collaborative (GLPC) (http://
greatlakesphragmites.net/) in 2012 to improve communication and collaboration. More specifically,
this collaboration aims to facilitate coordinated, efficient, and strategic approaches to Phragmites
management, restoration, and research across the Great Lakes Basin. The GLPC serves as a com-
munication conduit via an interactive website, a webinar series, and social media to facilitate access
to information and resources and encourage technology transfer and network building among habi-
tat managers, governmental agencies, and private landowners.
The Phragmites distribution maps produced by this project are useful to inform current manage-
ment and policy efforts as well as provide a baseline characterization for monitoring Phragmites
distribution over time. Unfortunately, the ALOS PALSAR failed in spring of 2011, but PALSAR-2
is planned for launch in 2014 and should provide the capability to track future vegetation changes.
The circa 2008–2010 Phragmites distribution maps are publically available for the U.S. portion of
the Great Lakes coastline (http://cida.usgs.gov/glri/phragmites/) along with the spatially explicit
field data (1145 sites) and associated attributes collected as part of this mapping effort (http://mtri.
org/Phragmites.html).
CASE STUDY 2: MAPPING COASTAL WETLANDS USING

A HYBRID RADAR/MULTISPECTRAL APPROACH
Introduction
The use of seasonally variable imagery and multiple-sensor or hybrid approaches has proven effec-
tive in detection of various LULC classes in the Great Lakes Basin (Bourgeau-Chavez et al. 2008,
2012). This case study summarizes an EPA-funded project that is currently focused on mapping
LULC classes within a 10 km buffer of the Great Lakes coastline both in the United States and
Canada. The objective was to specifically map wetlands and adjacent land use for the Coastal Great
Lakes Basin with the use of multiple imagery types and field data sources. This effort represents
the first comprehensive wetland delineation of the binational coastal Great Lakes using a consis-
tent mapping technique. It includes detection of several invasive plant species (e.g., Typha spp. and
P. australis). Maps are being developed in cooperation with the overarching Great Lakes Consortium
plan to provide a comprehensive regional baseline map suitable for coastal wetland assessment and
management by agencies at the local, tribal, state, and federal levels. The project is focused on pro-
viding not only LULC baseline data but also a repeatable methodology to monitor future change.
Mapping Approach
The study area consists of a 10 km buffer around the Great Lakes, which encompasses approxi-
mately 92,000 km2 of diverse cover types. Monitoring a landscape of this size requires remote
sensing integrated with field data and GIS to adequately capture the heterogeneity of the study area.
For this project, a fusion of contemporary (2007–2011) satellite imagery of moderate resolution
(10–30 m) is used from optical–IR and SAR sensors. Each sensor provides a unique set of informa-
tion, which, when used in combination, provides a powerful means to distinguish different land
cover types. Imagery from both optical and SAR sensors was collected in three time frames (spring,
summer, and fall). Seasonal date cutoffs for the sensor data collection were an approximation of first
growth for spring (April–May), peak growth for summer (June–August), and early senescence for
fall (September–October).
A hybrid classification scheme was developed by combining Anderson Level I upland classes
(Anderson et al. 1976) and the NWI classes as well as additional classes that aid in improving map
accuracy. The identification and delineation of upland classes is done using image interpretation
techniques. Field data provide calibration/validation data for specific wetland classes. These data
were collected by multiple institutions at opportunistic intervals. The base field dataset relied on
information collected under the Phragmites mapping project (described earlier). Additional field
datasets were compiled from independent projects throughout the time frame of interest (Figure
15.9). Wetland vegetation types are identified using field data and a select proportion (20%) of that
data is reserved for validation. Additional delineation of wetland and upland classes for training
data is accomplished by image interpretation of current aerial photographs.
The data processing and map generation flow is depicted in Figure 15.10. Field data of known
wetland class type are combined with the imagery stacks in order to generate training data for
Random Forests (Breiman 2001), a machine learning algorithm. Random Forests is an ensemble
classifier consisting of multiple decision trees that were each generated from a random subset of
training data sites and bands from the image stacks. Once the forest of decision trees is created, an
individual pixel’s classification is determined by which class receives the most “votes” from each
decision tree. Random Forests was chosen because of its high classification accuracy and for its
Legend
Aquatic bed
Wetland
Typha
Phragmites
Schoenoplectus
Mixed
N
0 100 200 400 km
W E
0 100 200 400 Miles S
FIGURE 15.9 Field data collection sites for the Great Lakes Coastal Mapping GLRI project color-coded
by dominant cover type around the Great Lakes Basin. The points labeled “wetland” are emergent wetlands
with various codominant species, and the points labeled “mixed” are a combination of peatland, shrub, and
forested wetland locations.
strengths that include handling datasets with a small number of observations and a large number
of attributes. It is also well suited to parallel processing and relatively insensitive to nonpredictive
inputs (Liaw and Wiener 2002). Additionally, the algorithm can easily handle missing attributes
(e.g., cloud-obscured pixels) as decision trees built without the missing attributes can be used to
classify the compromised data. Random Forests generates an “out of bag” estimate of classification
accuracy using the subset of training data not used in generating each tree. These data are, however,
used in generating other trees; therefore, to ensure a robust validation, 20% of the training data are
reserved for validation prior to running Random Forests. The classifier is teachable and accuracy
can be increased if needed by adding additional training data.
Initial classification began on the U.S. side of Lake Huron and is currently ongoing across the
basin as of early 2014. An example of the mapped product is shown in Figure 15.11 for Lake St.
Clair area in Michigan and Canada. Overall accuracy goals are 90% and individual class accuracies
are 70%. Final maps are expected to be completed by summer of 2014.
Case Study 2: Summary and Significance

The final map products will represent a current, comprehensive basin-wide inventory of coastal wet-
lands, as defined by U.S. Fish and Wildlife Service NWI types with additional classes for invasive plant
species—P. australis and Typha spp.—and adjacent land use classes as defined in the Great Lakes
Coastal Wetland Consortium (GLCWC) monitoring plan protocol (Bourgeau-Chavez et al. 2008).
This map will provide not only information on wetland extent and type but also contemporary infor-
mation on potential wetland stressors (e.g., invasive plant species and level and type of development
Classification schematic
Fall
Summer
Spring Landsat TM/
PALSAR stack
Random
Seasonal Landsat TM stack
forests
Fall Supervised
Summer data
Spring
Seasonal PALSAR stack
Field Aerial image

Classified image
data interpretation
Land cover
FIGURE 15.10 Processing flow chart for Random Forests classification methodology used for the Great
Lakes Coastal Mapping GLRI project.
surrounding the wetlands). More specifically, the map will assist in identifying indicators of wetland
health defined through the State of the Lakes Ecosystem Conference (SOLEC) including (1) land
cover adjacent to coastal wetlands, (2) land cover/land conversion, (3) urban density, (4) nonnative
terrestrial species, and (5) wetland extent and composition (http://www.epa.gov/solec/sogl2009/). It
will also provide reference and input for the Great Lakes Instrumentation Collaboratory (GLIC),
which has a 5-year plan for collection of biologic and other field-based indicators of wetland health
throughout the Great Lakes (see Albert 2008; Grabas et al. 2008; Timmermans et al. 2008; Uzarski
and Otieno 2008; Uzarski et al. 2008a,b,c).
Monitoring at a synoptic scale is necessary for effective coastal land and water management to
understand and mitigate the increasing risk of the Great Lakes to many anthropogenic influences
(e.g., drainage, dredging, filling, shoreline modification, water-level regulation, nutrient enrichment,
introduction of nonnative species, road development, and climate change). The GLCWC monitoring
plan (http://www.glc.org/wetlands/final-report.html) is designed not only to assess the health and
quality of the ecosystems but also to provide a baseline for assessing the effects of climate change
and to provide key inputs to decision support for coastal management into the future. The mapping
methodology used in this case study is reproducible, allowing for the continual development of
future maps for monitoring and detecting change in the Great Lakes Basin.
CASE STUDY 3: FORESTED WETLAND MAPPING WITH SAR

Introduction
Forested wetlands are one of the most difficult wetland types to map remotely due to the can-
opy cover that typically obscures the ground surface. It is particularly difficult to detect forested
Legend
Urban
Suburban
Urban grass
Urban road
Agriculture
Fallow field
Forest
Shrub
Barren light
Barren dark
Water
Aquatic bed
Wetland
Schoenoplectus
Typha
Phragmites
Wetland shrub
Forested wetland
N
0 5 10 km
W E
0 5 10 Miles
S
FIGURE 15.11 Example of Random Forests wetland mapping classification results based on spring, sum-
mer, and fall PALSAR and Landsat data for the St. Clair River Delta, Michigan (Great Lakes Coastal Mapping
GLRI project). Classes include wetland and upland categories for assessment of wetland ecosystem health.
wetlands under dense, closed canopy conditions with optical–IR imagery. Even with high-resolution
(e.g., 30 cm resolution) aerial photos, delineation of forested wetlands is difficult (see Chapter 3).
For deciduous forests, imagery collected in early spring during “leaf off” periods aids in forested
wetland mapping, but for coniferous forests it remains difficult given persistence of the needle
leaf canopy. SAR presents an advanced tool to aid in forested wetland mapping. The ability of
microwave SAR energy to penetrate the forest canopy makes forested wetland detection possible
despite cloud or canopy cover. Inundation beneath a forest canopy can be detected with L-band
HH-polarization SAR data (e.g., Hess et al. 1995; Bourgeau-Chavez et al. 2008) and in some cases
with C-HH-polarization imagery (Bourgeau-Chavez et al. 2001; Townsend et al. 2002; Lang et al.
2008) through enhanced backscatter returns.
For mapping forested wetlands with SAR, the wavelength needs to be long enough to reliably
penetrate the canopy layer to detect the inundated surface. This forest floor interaction with the
SAR energy is the key to mapping forested wetlands (Figure 15.12). Scattering from the smooth
water surface to the tree trunks and back to the sensor creates an enhanced “double-bounce” effect
particularly for HH polarization, making detection possible. ALOS PALSAR has a wavelength of
~24 cm that enables canopy penetration in temperate forests, and the satellite was in operation from
2006 to 2011 providing excellent coverage of the Great Lakes during multiple seasons. C-band SAR
sensors (e.g., RADARSAT-2) transmit energy at a wavelength near 5.7 cm. The shorter wavelength
of C-band SAR exhibits strong interaction and reflection of the signal at the forest canopy, and it
thus is less likely to penetrate to the forest floor in dense canopy conditions. Other polarizations
(HV and vertical send and receive [VV]) do not exhibit the strong double-bounce effect that is
Scattering of radar energy
Flooded vegetation versus Nonflooded vegetation

L-band
C-band C-band L-band

C-band
Flooded forest Flooded grassland/shrubland
L-band
C-band
C-band L-band
Nonflooded forest Nonflooded grassland/shrubland
FIGURE 15.12 Theoretical scattering of L-band (~24 cm wavelength) and C-band (~5.7 cm wavelength)
SAR energy given different vegetation cover and inundation conditions.
apparent in the HH-polarized data (Figure 15.13), but under some circumstances, backscatter from
these polarizations may be enhanced by inundation or high levels of soil moisture.
It should be noted that when the ground is dry, energy is not reflected as strongly from the for-
est floor due to signal attenuation through soil absorption. When the soil is moist, signal reflection
increases.
Another improvement in wetland detection, especially for forested wetlands, is the inclusion of
multiseasonal images collected in spring, summer, and fall. The high availability of multiseason
SAR data from satellite sensors with consistent imaging geometry facilitates wetland detection.
Multiseason data enable areas suspected to be forested wetlands to be monitored throughout an
entire growing season. Seasonal changes in water availability and ground hydrology correspond
with the intensity of radar return signals. Spring and fall scenes normally have increased backscat-
ter due to higher soil moisture content or inundation, whereas summer scenes can have lower returns
because of a lack of inundation and/or lower soil moisture.
Methodology
Spring and fall PALSAR imageries were used to create a “potential” forested wetland product
using a simple thresholding technique for the Upper Peninsula of Michigan to aid in an update of
the NWI conducted by Ducks Unlimited. Spring and fall seasons were chosen since they represent
relatively wetter times of the year. A simple thresholding technique was used since it allows the
backscatter signal range characteristic of forested wetlands to be extracted. While this range is
somewhat unique to forested wetlands, there are other land cover types that may also exhibit such
high backscatter signatures (e.g., tree plantations, row structure of agricultural fields, and urban
Duke Forest flooded cypress SIR-C imagery

April 17, 1994
C-band L-band
HH
Cypress
Swamp
HV
VV
FIGURE 15.13 Shuttle Imaging Radar-C C-band (~5.7 cm wavelength) and L-band (~24 cm wavelength)
images from April 17, 1994. Images from HH (horizontal send and receive), HV, and VV polarizations show
the enhanced brightness of the inundated cypress swamp due to double-bounce effects at HH and less vis-
ibility of the forested wetland area at HV and VV polarizations. L-band shows improved canopy penetration
to detect a greater extent of the inundation over the shorter C-band data.
areas). Figure 15.14 shows the average PALSAR HH-polarization backscatter (in decibels [dB])
values sampled from many different land cover types. The lower the dB value, the darker the area
will be in the scene. Waterways have the lowest values and appear near black in the scenes due to
specular reflectance, while highly developed areas (e.g., urban areas) exhibit the highest energy
returns due to double bounce from urban structures. Notably, forested wetlands are characteristi-
cally brighter than other types of forests, again due to the double bounce. Using the somewhat
unique backscatter signature associated with forested wetlands in the radar imagery, a “baseline”
for detection was established. For inclusion of all forested wetlands, the upper backscatter values
were left open ended. The brightest returns are often due to terrain slope or urban backscatter, both
of which can be easily filtered in postprocessing via existing cover-type maps (e.g., NOAA C-CAP
or NLCD) and high-resolution digital elevation models (DEMs). Simple thresholding allows for
quick identification of areas of potential forested wetlands. Under many circumstances the thresh-
olding produces the desired product relatively quickly without large amounts of training data.
For each spring/fall image set, the simple thresholding classification was conducted to obtain
only the “bright areas” of the imagery, thereby excluding areas that were not likely to be forested
wetlands. Known forested areas, exhibiting strong backscatter, were used as a guide in determining
the threshold value for each PALSAR image since the threshold values vary by image acquisition
date due to variations in moisture conditions in the vegetation and ground. After thresholding,
additional GIS layers can be used to remove areas of confusion such as agricultural lands, urban
–24
–20
L-HH mean backscatter (dB)
–16
–12
–8
–4
0
Deciduous forest
Evergreen forest
Forested wetland
Development high intensity

Mixed forest
Palustrine wetland forest

Riverine water
Palustrine shrub
Upland scrub
Development medium intensity

Grassland pasture
SAR forested wetlands

Open water
Estuary emergent
Development open
Bare land
Land cover category
FIGURE 15.14 Bar chart showing the typical L-HH (~24 cm wavelength) backscatter (in dB) from a range
of cover types that were chosen based on NOAA C-CAP classes. “SAR Forested Wetlands” category is the
average backscatter for those areas depicted from a thresholding method. This compares to backscatter values
from the “Forested Wetland” class of NOAA C-CAP well.
developed areas, and sloped terrain. Confusion with agricultural lands or tree plantations arises
from the planted row structure that creates a bright return similar to that of a forested wetland due
to Bragg scattering. Bragg scattering is the coherent combination of signals reflected from features
with periodic occurrence in the direction of wave propagation and whose spacing is equal to ½ the
wavelength (Henderson and Lewis 2008).
Masks used to filter confusion areas were made from standard mapping products including the
C-CAP from the NOAA and DEMs from the USGS via their National Map Viewer and Download
Platform (http://viewer.nationalmap.gov/viewer/). NOAA C-CAP was used to filter out urban classes
from the wetland product. Using ArcMap and the Slope tool, a DEM was used to create a mask that
would eliminate areas from the forested wetland class that exhibit a slope angle that is greater than
3°, a slope above which occurrence of forested wetlands is unlikely, but bright signatures in the SAR
imagery are expected (Sader et al. 1995; Whitcomb et al. 2009).
Case Study 3: Results

The resulting product for the Upper Peninsula of Michigan (presented in Figure 15.15) includes
nonwetland areas, which are primarily agricultural fields. This product was intended to be used
as an ancillary dataset meant to aid image interpreters in manually delineating wetlands for the
NWI. Therefore, error of inclusion (commission) was preferred over exclusion (omission). Since
agricultural areas are quite visible in air photos, they are not easily mistaken for potential forested
wetlands, so further filtering was not conducted.
FIGURE 15.15 SAR-derived “potential” forested wetland inundation product for the entire Upper Peninsula
of Michigan after removing slopes that had a gradient greater than 3° based on a 10 m resolution NED from
USGS.
An accuracy assessment of the potential forested wetland product revealed a 90% agreement
between a small set of field data collections (20 field locations) and the SAR-based map products
(Figure 15.16). A full validation field campaign was not conducted since the product was to be used
as ancillary information in an NWI update and labeled as “potential forested wetland” to aid in
manual interpretation of aerial imagery. This product was very helpful for Ducks Unlimited in their
update of the NWI for this region, and a similar PALSAR product was demonstrated as useful in
“Case Study 4.”
Figure 15.17 shows the SAR-derived “potential forested wetland” areas from spring and fall
time periods as well as the SAR-detected co-occurrence of “inundation” overlaid onto the 1970s
NWI forested wetland boundaries. There was not complete agreement between the NWI-identified
forested wetland areas and that of the SAR-derived potential forested wetland product. This is to be
expected because areas that were not inundated during the timing of the satellite overpass may not
have exhibited the high backscatter returns characteristic of forested wetlands. In addition, there
were areas mapped from the SAR technique that were not detected by air photo interpretation of
NWI due to canopy closure.
It is important to recognize that the NWI data include bog and fen peatlands in the forested
wetland class, but since these sites are not typically inundated with water, they are not detected
using the aforementioned thresholding technique. Although still detectable by SAR techniques,
they are not captured by a thresholding technique. A map of peatland of the Upper Peninsula of
Michigan is being created from a fusion of PALAR and optical–IR data using similar methods to
those described in “Case Study 2” using Random Forests. The details on how the L-band SAR mul-
tidate data work to detect bogs and fens are described in Bourgeau-Chavez et al. (2009).
Michigan
Wisconsin
W E
S
Potential forested wetlands
Fall inundated
Spring inundated
Annually inundated
NWI forested wetlands
0 0.5 1 2 Miles
0 1 2 4 km
FIGURE 15.16 Potential forested wetland SAR-derived product for Houghton, MI, based on PALSAR
thresholding technique showing fall inundated areas as yellow, spring inundated areas as green, and the
co-occurrence of inundation during these time frames as red. Light blue areas are those mapped as palustrine
forested wetlands by the NWI in the 1970s.
Summary and Next Steps

Thresholding of multiple-date L-band SAR imagery was an effective remote sensing approach for
the identification of potential forested wetlands across the Upper Peninsula of Michigan, and such
a method would likely have similar success in other temperate areas. Indeed, similar results were
found for studies in Maine, Iowa, and Minnesota; see “Case Study 4.” The continuity of SAR imag-
ery collection into the future is needed to provide the opportunity not only to map forested wetlands
but also to refine classification techniques while building a data archive that would facilitate change
analysis. This type of analysis with multiseason as well as multiyear datasets will better enable
understanding of the observed variation that is inherent in forested wetlands and provide insight
into longer-term trends. With the planned launch of a constellation of RADARSAT satellites,
PALSAR-2, and planning discussions for NASA’s DESDynI and German DLR’s TerraSAR-L, the
potential utility of the methods explored in this case study for long-term monitoring of our forested
wetlands should have long-term applicability for a wide range of users (e.g., hydrologists, ecologists,
managers of wildlife habitat, carbon cycle modelers, and climate change modelers). Future sensors
W E
0 3.5 7 14 Miles
0 5 10 20 km
Legend
Both forested wetlands
NWI forested wetlands
Remotely sensed forested wetlands
FIGURE 15.17 Comparison of NWI classified forested wetlands (blue) and remotely sensed forested wet-
lands (red) near Newberry, in the Upper Peninsula of Michigan. Blue areas are likely peatlands that would
not be mapped by the methodology used to detect inundated woody wetlands. Purple areas show forested
wetlands mapped using SAR and NWI, and red areas detected by SAR indicate potential forested wetlands.
Notice pine plantations and orchards that are misidentified as “potential wetland” (red areas with geometric
shapes in the center of the image).
like PALSAR-2 (the successor to PALSAR, planned for launch in 2014) will provide similar cover-
age and resolution in the L-band wavelength to that used for this case study. The planned launch
of the C-band (~5.7 cm) RADARSAT constellation may have less utility for Great Lakes forested
wetlands than the L-band (~24 cm) systems. While Townsend (2001, 2002) and Lang et al. (2008a,c)
found C-band HH-polarized data to be useful for mapping of inundation in the southeastern and
mid-Atlantic states, the characteristic enhanced signatures in forested wetlands of the Great Lakes
have not been observed. Further investigation into the structural characteristics of forested wetlands
in these different regions is needed to determine the limitations of C-band frequencies.
CASE STUDY 4: ADVANCED MULTISENSOR APPROACH

FOR UPDATING NWI IN MINNESOTA
Introduction
The existing NWI data for much of Minnesota are from the 1980s, and significant changes to
wetlands have occurred over the past 20–30 years. There were also limitations in the original
source data, technology, and methods that resulted in undermapping of some wetland types,
especially emergent and forested wetlands (Tiner 2009). Due to this underrepresentation of cer-
tain wetland classes and the age of the NWI data, a State of Minnesota interagency partnership
identified a need for improving the classification and updating of NWI data. This motivated
the Minnesota Department of Natural Resources (MN DNR) to commission the University of
Minnesota Remote Sensing and Geospatial Analysis Laboratory (RSGAL) to develop advance-
ments in wetland mapping methods for the update (Corcoran et al. 2011, 2013; Knight et al. 2013;
Rampi et al. 2014). The MN DNR then contracted Ducks Unlimited and Equinox Analytics,
Inc. (DU/EA) to apply the methods developed by RSGAL and continue to develop a protocol
to enable broadscale wetland mapping across the state. The goal of this effort was to provide
a robust, semiautomated method with improved efficiency and accuracy to update NWI data,
taking advantage of both computer segmentation, automated classification, and human-based
image interpretation. The initial project focused on the 13 counties of east central Minnesota
that represented an area that has seen the most urban development and potential for the greatest
change in the original NWI as well as a high demand for an updated product. The effort by the
Minnesota consortium effectively addressed shortcomings of the existing wetland data and pro-
vided an improved classification methodology that utilized multiple data sources and advanced
GIS techniques as described later.
Improved Data Sources and Software

Many datasets and computer technologies that are readily available now were nonexistent when the
original NWI (1970s) or the first round of NWI updates (1980s) were compiled. Since the 1980s,
multiple datasets have been determined as essential to improving wetland classification, including
SAR and LiDAR technologies, multiband digital high-resolution aerial and commercial satellite
imagery, and digital soil survey data. The Minnesota NWI update included the following source
datasets: PALSAR imagery, LiDAR-derived DEMs, high-resolution digital multiband aerial pho-
tography, and digital Soil Survey Geographic (SSURGO) data.
In addition to improvements in data sources since the 1980s, computer hardware and software
technological advances have better processing capabilities and more storage capacity. Software
has become increasingly capable of handling the sophisticated image and data processing tasks
necessary to complete complex classification processes at a regional or statewide level. Image
processing and mapping software used for the Minnesota project included ESRI ArcGIS Desktop,
ERDAS Imagine, Random Forests, Trimble eCognition, and Alaska Satellite Facility’s (ASF)
MapReady.
Mapping Methodology for NWI Updates

The update to the Minnesota NWI classifications, using the Cowardin et al. (1979) wetland classifi-
cation system, was based on the process flow outlined in Figure 15.18 (Macleod et al. 2013). The pri-
mary processes are indicated as rectangles in the process flow chart: image segmentation, Random
Forest classification, photo interpretation/object editing/manual delineation, quality assurance/
quality control (QA/QC), and accuracy assessment. The protocol included advanced processing
using eCognition software, Random Forests, and multiple data sources to develop classifications
and polygons (from image objects) to assist the air photo interpreters to more efficiently and accu-
rately update the maps. The eCognition software was used to create image objects that provide the
foundation of the NWI update process.
Image Segmentation
Image segmentation was based on the digital aerial imagery, LiDAR-derived DEMs, PALSAR,
and SSURGO soils data to segment the images into polygons based upon groupings of similar
pixels from the input dataset. Trimble eCognition was the main tool for this process, providing a
ECMN
Summer
leaf-off Data analysis workflow
imagery
CIR imagery for Minnesota NWI update
CIR imagery NAIP

PALSAR quarter quads forestry CIR
other imagery
radar data
Image objects
Wooded Image objects with potential
wetlands Image with features Random forest NWI classification
segmentation classification
SSURGO
soils database Photo interpretation
object editing
manual delineation
Hydric Reference data
soils
Wetland classification
Elevation based training database Initial NWI delineation
and classification
databases
Wetland classification QA/QC

DEM
accuracy assessment and
and derivatives
database accuracy assessment
Final NWI database
FIGURE 15.18 Process flow chart for NWI updating using the improved wetlands classification methodol-
ogy for the Minnesota wetlands mapping project. (From Smith, A. et al., Updating the National Wetland
Inventory in Minnesota by integrating air photo-interpretation, object-oriented image analysis and multi-
source data fusion, ASPRS Annual Conference Proceedings, Sacramento, CA, 2012.)
segmented image object file with feature polygons (Figure 15.19). The rule set used for this process-
ing included over 250 operations, utilizing the data inputs that are described later in further detail.
PALSAR-Derived Wooded Wetlands

PALSAR L-band data were used to create a binary layer that identifies wooded wetlands based on
similar thresholding methods as described in “Case Study 3” that were developed by Bourgeau-
Chavez and shared with DU/EA. For the DU/EA NWI update protocol, the wooded wetland product
is utilized in the image segmentation process but is also available to the image interpreter as a sup-
porting data source for final classification.
Hydric Soils
SSURGO data, available from the Natural Resources Conservation Service (NRCS), were pro-
cessed into a layer containing the predominant water regime and an estimate of the proportion and
location of hydric soils.
LiDAR DEM and Derivatives

LiDAR DEMs were used for a majority of the area (3 m resolution). In areas where LiDAR was
unavailable, the National Elevation Dataset (NED) DEM was used (10 m resolution). A topographic
position index (TPI) (Weiss 2001) and the compound topographic index (CTI) (Moore et al. 1991)
were calculated from the DEM data for use in the image segmentation process. The DEM was also
used in the Random Forests classification.
Digital Aerial Imagery

Spring leaf off 50 cm resolution color IR imagery (2010–2011) was used in the image segmentation
and was the base of the photo interpretation.
Stream bed wetlands

Wooded wetlands boundary derived from
identified with PALSAR hydrologic flow modeling
and LiDAR DEM
Topographic contour lines

in forested areas
derived from LiDAR DEM
L1/L2 (6¢ depth)

Forest/nonforest
boundary derived
boundary derived from
from in situ
spectral infromation (NDVI)
SONAR bathymetry
and image object texture
FIGURE 15.19 A polygon segmented image based on multiple input datasets (LiDAR DEM, PALSAR,
aerial imagery, SSURGO) for the Minnesota wetlands mapping project. (From Macleod, R.D. et al., Updating
the National Wetland Inventory in East-Central Minnesota: Technical documentation, Minnesota Department
of Natural Resources, Division of Ecological and Water Resources, St. Paul, MN, retrieved from http://files.
dnr.state.mn.us/eco/wetlands/nwi_ecmn_technical_documentation.pdf, 2013.) Purple areas are wooded
wetlands identified with PALSAR, dark green areas indicate nonwetland forested areas (based on Normalized
Difference Vegetation Index [NDVI]), light blue areas are stream bed wetlands from LiDAR DEM and flow
modeling, yellow areas are high NDVI brightness areas that are indicative of bare soil (or senescent wetland
vegetation), and dark blue areas are water bodies greater than 2 m deep based on sonar bathymetry.
Field Training Data

As required for most remote sensing mapping projects, a significant field training dataset was
needed. For this project, 3350 validation and training data locations were evaluated. Field data
were collected by the project team in 12 representative quadrangles, including urban, residential,
and rural areas. The training data were used in Random Forests to provide a classification that was
utilized by the air photo interpreter as an aide for areas not readily identifiable.
Potential NWI Computer-Generated Classification Process

The Random Forests classification utilizes the procedure described by Breiman (2001) and adds
potential NWI classification attributes to the image objects created in the previous steps. Inputs to
Random Forests are the point file and training data generated by eCognition in a previous step in the
process. Random Forests classifies each point in the input file and assigns a confidence value to the
classification. The unique identification shared by the point file and attributes and polygons gener-
ated by eCognition is used to join the image segments and the classification suggested by Random
Forests. The classified image segments are used to enhance the traditional photointerpretation and
are not the final product. The NWI process is based on human interpretation, which is recognized
as having greater accuracy given all the input data than any computer-generated classification.
Image interpreters are capable of identifying associations, shapes, patterns, tones, etc., simultane-
ously to distinguish cover types and delineate boundaries.
Photo Interpretation: Object Editing/Manual Delineation

The photo interpretation step included object editing and manual delineation, which is the primary
human-interface portion of the process. Image interpreters work with the output of the image seg-
mentation process that includes an initial classification. These initial boundaries are reviewed and
adjusted or redrawn, as necessary. The initial classification is reviewed and accepted or altered. All
input datasets are available to the interpreter as an aide to perform the boundary and classification
review. The output from the Random Forests algorithm is used as a secondary dataset for areas not
readily classified. The NWI classification with boundary delineation is the output of this process
and then subject to a QA/QC review.
Upon completion of the initial delineation and classification, the image interpreter runs an auto-
mated QA/QC protocol designed to identify topological errors and attribute inconsistencies. A sec-
ond step of the QA/QC process includes a second interpreter evaluating 10% of the completed
products to verify consistency between interpreters. This yields the final NWI product. This data-
base is then assessed for accuracy by a third party, using an independent field reference source.
Case Study 4: Summary

The integration of image objects, machine learning, and traditional image interpretation methods
with multiseason and multisensor data (including soils and DEM data) has produced an advanced
mapping method for updating the NWI described in “Case Study 4.” This new methodology takes
advantage of the human image interpretation skills that are unparalleled by a computer and uses
new technology to improve the accuracy and efficiency of mapping. In particular, it aids image
interpreters by creating polygons (through image segmentation to create image objects in eCogni-
tion), leading to less time digitizing boundaries. It is more efficient for the interpreter to edit an
existing boundary than to create one from scratch. The new approach also provides added infor-
mation to the interpreters in the GIS database, allowing them to scan multiple data sources at the
touch of a finger and provides the Random Forests classification that is based on all data layers as
backup.
CONCLUSION
In this chapter, biophysical aspects of Great Lakes coastal wetlands and what makes them unique
have been reviewed, as well as an overview of traditional and advanced methods in remote sens-
ing to improve wetland inventories and classification. The discussion has focused on traditional
optical–IR and SAR sensors; however, hyperspectral and LiDAR sensors are also of high value
for improving wetland mapping. Both multisensor and multiseasonal data fusion improve mapping
accuracies and efficiency. Furthermore, data fusion across sensor platforms provides a more robust
detection capacity, whether it be SAR with optical data fusion, optical with LiDAR, thermal, or
hyperspectral data. Different sensor frequencies have distinct capabilities that provide complemen-
tary information about wetland characteristics and together increase detection accuracy especially
for discriminating specific wetland habitat types or species. Various methods have been described
that have taken advantage of new sensor data availability and sensor fusion, while maintaining the
traditional air photo interpretation methods as one of the fundamental steps. Whether using multi-
sensor information to aid in traditional air photo interpretation (as in “Case Study 4”) or using the
multisensor data to directly create the map classification (“Case Study 2”), data fusion is becoming
the standard.
A fundamental tool necessary for the management and remediation of wetlands is accurate spatial
information. Mapping of wetlands across time and space is essential to monitoring invasive species
trends, changes in biodiversity, and evaluating ecosystem function and health. Such maps also pro-
vide a foundation upon which to build wetland science. Vast areas of land in the Great Lakes Basin
are undergoing rapid land cover and land use change. Land transformations have altered groundwa-
ter levels and stream flows, increased nutrient loading, and contributed to anthropogenic contamina-
tion that impacts ecosystems at a variety of scales (Richards 1993; Turner 1993; Foley et al. 2005).
Climate change has also had significant impacts in the region, with rising maximum summer and
winter temperatures since the 1970s leading to increases in evapotranspiration. Precipitation has
also increased over this period (Assel et al. 2004). Future climate projections suggest increased pre-
cipitation across the region (Sousounis and Bisanz 2000), with likely increases in the intensity and
frequency of precipitation events. Such extreme precipitation events and subsequent flooding can
increase runoff from urban and agricultural areas (Lipp et al. 2001), with accompanying increases
in nutrient loads to wetlands. Thus, changes in land use/land cover and climate (both annual fluc-
tuations in weather patterns and long-term trends) affect the quantity and quality of water flowing
from uplands to the coasts.
With such dynamic landscape and climate changes occurring over relatively short time scales,
it is important to monitor the health of coastal wetland ecosystems, both from a synoptic remote
sensing point of view and from field surveys. Monitoring efforts are underway for Great Lakes
coastal wetlands through the GLCWC with funding from the GLC and the U.S. EPA. The monitor-
ing plan is designed not only to assess the health and quality of the ecosystems but also to provide a
baseline for assessing effects of climate change, while providing key inputs to decision support for
coastal management (http://projects.glc.org/wetlands/final-report.html). The monitoring protocol
represents a scientifically validated sampling design and suite of indicators and metrics to assess
coastal wetland condition. The preservation and maintenance of Great Lakes coastal wetlands will
largely depend on the success of this and other environmental management efforts, and applications
of remote sensing will provide critical information for decision makers.
ACKNOWLEDGMENTS
The work presented in Case Studies 1 through 3 were supported by the GLRI through grants
from the USGS, EPA, and U.S. Fish and Wildlife Service. We would like to acknowledge Kirk
Scarbrough for his contributions to this chapter and research on the three case studies. We would
also like to thank Arthur Endsley, Brian Huberty, and Zach Raymer for their assistance with review
and editing. Lastly, we acknowledge Maria Chavez, Dante Mann, Rachel Posavetz, Aaron Smith,
and Kaitlyn Smith for their help with figures and formatting.
REFERENCES
Albert, D.A. 2003. Between Land and Lake: Michigan’s Great Lakes Coastal Wetlands. Michigan Natural
Features Inventory, Michigan State University Extension, East Lansing, MI. Extension Bulletin E-2902.
Albert, D.A. 2008. Chapter 3. Vegetation community indicators. In: Great Lakes Coastal Wetlands Monitoring
Plan, Great Lakes Coastal Wetlands Consortium, A project of the Great Lakes Commission, funded
by the U.S. EPA GLNPO, March 2008, pp. 32–58, http://projects.glc.org/wetlands/final-report.html.
Albert, D.A., Wilcox, D.A., Ingram, J.W., and Thompson, T.A. 2005. Hydrogeomorphic classification for Great
Lakes coastal wetlands. Journal of Great Lakes Research 31:129–146.
Anderson, J.R., Hardy, E.E., Roach, J.T., and Witmer, R.E. 1976. A land use and land cover classification sys-
tem for use with remote sensor data. USGS professional paper 964.
Anderson, M., Bourgeron, P., Bryer, M.T., Crawford, R., Engelking, L., Faber-Langendoen, D., Gallyoun, M.,
and Weakley, A.S. 1998. International classification of ecological communities: Terrestrial vegetation of
the United States. Volume II. The National Vegetation Classification System: List of Types. The Nature
Conservancy, Arlington, VA, 502pp.
Assel, R.A., Quinn, F.H., and Sellinger, C.E. 2004. Hydroclimatic factors of the recent record drop in Laurentian
Great Lakes water levels. Bulletin of the American Meteorological Society 85(8):1143–1151.
Bourgeau-Chavez, L.L., Kasischke, E.S., Brunzell, S.M., Mudd, J.P., Smith, K.B., and Frick, A.L. 2001.
Analysis of spaceborne SAR data for wetland mapping in Virginia riparian ecosystems. International
Bourgeau-Chavez, L.L., Kowalski, K.P., Carlson Mazur, M.L., Scarbrough, K.A., Powell, R.B., Brooks,
C.N., Huberty, B. et al. 2013. Mapping invasive Phragmites australis in the coastal Great Lakes with
ALOS PALSAR satellite imagery for decision support. Journal of Great Lakes Research Supplement
39:65–77. http://dx.doi.org/10.1016/j.jglr.2012.11.001. http://www.sciencedirect.com/science/article/
pii/S0380133012002122
Bourgeau-Chavez, L.L., Lopez, R.D., Trebitz, A., Hollenhorst, T., Host, G.E., Huberty, B., Gauthier, R.L., and
Hummer, J. 2008. Chapter 8. Landscape-based indicators. In: Great Lakes Coastal Wetlands Monitoring
Bourgeau-Chavez, L.L., Riordan, K., Powell, R., Miller, N., and Nowels, M. 2009. Chapter 33. Improving
wetland characterization with multi-sensor, multi-temporal SAR and optical/infrared data fusion. In:
G. Jedlovec (ed.). Advances in Geoscience and Remote Sensing. InTech, Vukovar, Croatia, pp. 679–708.
Bourgeau-Chavez, L.L., Scarbrough, K., Miller, M.E., Laubach, Z., Endres, S., Banda, E., Battaglia, M. et al.
2012. Mapping coastal Great Lakes wetlands and adjacent land use through hybrid optical-infrared and
radar image classification techniques. Intecol Wetlands Conference, Orlando, FL, June 6, 2012.
Brown, D.G., Johnson, K.M., Loveland, T.R., and Theobald, D.M. 2005. Rural land-use trends in the contermi-
nous United States, 1950–2000. Ecological Applications 15(6):1851–1863.
Churkina, G., Brown, D.G., and Keoleian, G. 2010. Carbon stored in human settlements: The conterminous
United States. Global Change Biology 16:135–143.
Congalton, R. and Green, K. 2008. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices.
CRC Press, Boca Raton, FL.
Corcoran, J.M., Knight, J.F., Brisco, B., Kaya, S., Cull, A., and Murhnaghan, K. 2011. The integration of
optical, topographic, and radar data for wetland mapping in northern Minnesota. Canadian Journal of
Corcoran, J.M., Knight, J.F., and Gallant, A.L. 2013. Influence of multi-source and multi-temporal remotely
sensed and ancillary data on the accuracy of random forest classification of wetlands in northern
Minnesota. Remote Sensing 5(7):3212–3238.
Cowardin, L.M., Carter, V., Golet, F.C., and LaRoe, E.T. 1979. Classification of Wetlands and Deepwater
Habitats of the United States. U.S. Fish and Wildlife Service, Washington, DC. FWS/OBS-79/31.
Cvancara, A.M. and Melik, J.C. 1961. Bedrock geology of Lake Huron. In: Great Lakes Research Division
Publication 7. Institute for Science and Technology, University of Michigan, Ann Arbor, MI,
pp. 116–125.
Dahl, T.E. 1990. Wetlands Losses in the United States, 1780s to 1980s. U.S. Department of the Interior,
Washington, DC, http://www.npwrc.usgs.gov/resource/wetlands/wetloss/. Accessed November 5, 2014.
Dobson, J.E., Bright, E.A., Ferguson, R.L., Field, D.W., Wood, L.L., Haddad, K.D., Iredale, H. et al. 1995.
NOAA coastal change analysis program (CCAP): Guidance for regional implementation. NOAA
Technical Report NMFS 123, U.S. Department of Commerce, Seattle, WA.
Ducks Unlimited. 2008. Updating the national wetlands inventory (NWI) for the great lakes. October 15, 2008.
http://www.ducks.org/media/Conservation/GLARO/_documents/_library/_gis/NWI_Workplan.pdf.
Environment Canada. 2009. Canada’s 4th National Report to the United Nations Convention on Biological
Diversity. Ottawa, Ontario, 71pp.
Environment Canada. (Wilcox, D.A., Patterson, N., Thompson, T.A., Albert, D., Weeber, R., McCracken, J.,
Whillans, T., and Gannon, J., contributors). 2002. Where Land Meets Water—Understanding Wetlands of
the Great Lakes. Canadian Wildlife Service, Toronto, Ontario, Canada.
ERDAS. 2010. ERDAS Field Guide (vol. June 2010 edn.). ERDA Inc., Norcross, GA.
Findlay, S., Dye, S., and Keuhn, K. 2002. Microbial growth and nitrogen retention in litter of Phragmites aus-
tralis compared to Typha angustifolia. Wetlands 3(22):16–625.
Foley, J.A., DeFries, R., Asner, G.P., Barford, C., Bonan, G., Carpenter, S.R., and Snyder, P.K. 2005. Global
consequences of land use. Science 309(5734):570–574.
Fournier, R.A., Grenier, M., Lavoie, A., and Hélie, R. 2007. Towards a strategy to implement the Canadian
Wetland Inventory using satellite remote sensing. Canadian Journal of Remote Sensing 33(Suppl. 1):
S1–S16. doi:10.5589/m07-051.
Grabas, G.P., Crewe, T.L., and Timmermans, S.T.A. 2008. Chapter 7. Bird community indicators. In: Great
Lakes Coastal Wetlands Monitoring Plan, Great Lakes Coastal Wetlands Consortium, A project of the
Great Lakes Commission, funded by the U.S. EPA GLNPO, March 2008, pp. 116–142, http://projects.
glc.org/wetlands/final-report.html. Accessed November 5, 2014.
Great Lakes Coastal Wetland Inventory. 2004. http://glc.org/projects/habitat/coastal-wetlands/cwc-inventory/.
Grenier, M., Demers, A.M., Labrecque, S., Benoit, M., Fournier, R.A., and Drolet, B. 2007. An object-based
Canada. Canadian Journal of Remote Sensing 33(Suppl. 1):S28–S45. doi:10.5589/m07-048.
Gronewold, A.D., Clites, A.H., Smith, J.P., and Hunter, T.S. 2013. A dynamic graphical interface for visu-
alizing projected, measured, and reconstructed surface water elevations on the earth’s largest lakes.
Environmental Modelling and Software 49:34–39. http://dx.doi.org/10.1016/j.envsoft.2013.07.003.
Henderson, F.M. and Lewis, A.J. 2008. Radar detection of wetland ecosystems: A review. International Journal
Herdendorf, C.E., Hartley, S.M., and Barnes, M.D. 1981. Fish and Wildlife Resources of Great Lakes Coastal
Wetlands within the United States, Vol. 1. U.S. Department of the Interior, Washington, DC.
Hess, L., Melack, J., Filoso, S., and Wang, Y. 1995. Delineation of inundated area and vegetation along the
Amazon floodplain with the SIR-C synthetic aperture radar. IEEE Transactions on Geoscience and
Hough, J.L. 1958. Geology of the Great Lakes. University Illinois Press, Urbana, IL.
Ingram, J., Holmes, K., Grabas, G., Watton, P., Potter, B., Gomer, T., and Stow, N. 2004. Development of
a coastal wetlands database for the Great Lakes Canadian shoreline. Final report to the Great Lakes
Commission (Wetlands2-EPA-03).
IPCC. 2008. Technical paper of climate change and water. IPCC-XXVIII/Doc.13 (8.IV.2008), IPCC Secretariat,
Geneva, Switzerland.
Klemas, V.V., Dobson, J.E., Ferguson, R.L., and Haddad, K.D. 1993. A coastal land cover classification system
for the NOAA coastwatch change analysis project. Journal of Coastal Research 9(3):862–872.
Knight, J.F., Tolcser, B., Corcoran, J., and Rampi, L. 2013. The effects of data selection and thematic detail
on the accuracy of high spatial resolution wetland classifications. Photogrammetric Engineering and
Lang, M.W. and Kasischke, E.S. 2008. Using C-band synthetic aperture radar data to monitor forested wetland
46(2):535–546. doi:10.1109/TGRS.2007.909950.
Lang, M.W., Townsend, P., and Kasischke, E. 2008. Influence of incidence angle on detecting flooded for-
ests using C-HH synthetic aperture radar data. Remote Sensing of the Environment 112(10):3898–3907.
doi:10.1016/j.rse.2008.06.013.
Larsen, G. and Schaetzl, R. 2001. Origin and evolution of the Great Lakes. Journal of Great Lakes Restoration
24(4):518–546.
Lee, J.S. and Pottier, E. 2009. Polarimetric Radar Imaging: From Basics to Applications. CRC Press, Boca
Raton, FL.
Liaw, A. and Wiener, M. 2002. Classification and regression by random forests. R News 2(3):18–22.
Lillesand, T., Kiefer, R., and Chipman, J. 2007. Remote Sensing and Image Interpretation, 6th edn. John
Wiley & Sons, Inc., Hoboken, NJ.
Lipp, E.K., Kurz, R., Vincent, R., Rodriguez-Palacios, C., Farrah, S.R., and Rose, J.B. 2001. The effects of
seasonal variability and weather on microbial fecal pollution and enteric pathogens in a subtropical estu-
ary. Estuaries 24(2):266–276.
Macleod, R.D., Paige, R.S., and Smith, A.J. 2013. Updating the National Wetland Inventory in East-Central
Minnesota: Technical Documentation. Minnesota Department of Natural Resources, Division of
Ecological and Water Resources, St. Paul, MN. Retrieved from http://files.dnr.state.mn.us/eco/wetlands/
nwi_ecmn_technical_documentation.pdf. Accessed November 5, 2014.
Mitsch, W.J. and Gosselink, J.G. 2000. The value of wetlands: Importance of scale and landscape setting.
Ecological Economics 35:25–33.
Moffett, M.F., Dufour, R.L., and Simon, T.P. 2007. An inventory and classification of coastal wetlands of the
Laurentian Great Lakes. In: T.P. Simon and P.M. Stewart (eds.). Coastal Wetlands of the Laurentian
Great Lakes: Health, Habitat, and Indicators. Author House Press, Bloomington, IN, pp. 17–99.
Moore, I.D., Grayson, R.B., and Ladson, A.R. 1991. Digital terrain modeling: A review of hydrological, geo-
morphological, and biological applications. Hydrological Processes 5:3–30.
Nature Conservancy. 1994. The conservation of biological diversity in the Great Lakes ecosystem: Issues and
opportunities. The Nature Conservancy Great Lakes Program, Chicago, IL.
Owens, T. and Hop, K.D. 1995. Long Term Resource Monitoring Program Standard Operating Procedures:
Field Station Photo Interpretation. U.S. Fish and Wildlife Service, National Biological Survey,
Environmental Management Technical Center, Onalaska, WI.
Rampi, L.P., Knight, J.F., and Lenhart, C.F. 2014. Comparison of flow direction algorithms in the application
of the CTI for mapping wetlands in Minnesota. Wetlands 34(3):513–525.
Raney, R.K. 1998. Chapter 2. Radar fundamentals: Technical perspective. In: F.M. Henderson and A.J. Lewis
(eds.). Principles and Applications of Imaging Radar: Manual of Remote Sensing, vol. 2, 3rd edn. John
Wiley & Sons, New York, pp. 9–124.
Reimer, K. 2009. The need for a Canadian wetland inventory. Conservator 30(1):36–45.
Richards, J.A. 1993. Remote Sensing Digital Image Analysis: An Introduction, 2nd edn. Springer-Verlag,
Berlin, Germany.
Sader, S.A., Ahl, D., and Liou, W.S. 1995. Accuracy of landsat-TM and GIS rule-based methods for forest
wetland classification in Maine. Remote Sensing of the Environment 53:133–144.
Smith, A., Macleod, R., Paige, R., and Kloiber, S. 2012. Updating the National Wetland Inventory in Minnesota
by integrating air photo-interpretation, object-oriented image analysis and multi-source data fusion. In:
ASPRS Annual Conference Proceedings, Sacramento, CA, 2012.
Smith, R.D., Ammann, A., Bartoldus, C., and Brinson, M.M. 1995. An Approach for Assessing Wetland
Functions using Hydrogeomorphic Classification, Reference Wetlands, and Functional Indices. U.S.
Army Engineer Waterways Experiment Station, Vicksburg, MS. WES/TR/WRP-DE-9.
Sousounis, P.J. and Bisanz, J.M. 2000. Preparing for a Changing Climate: The Potential Consequences of
Climate Variability and Change. Great Lakes Overview. A Summary by the Great Lakes Regional
Assessment Group, U.S. Global Change Research Programme, University of Michigan, Ann Arbor, MI.
Timmermans, S.T.A., Crewe, T.L., and Grabas, G.P. 2008. Chapter 6. Amphibian community indicators. In:
Great Lakes Coastal Wetlands Monitoring Plan, Great Lakes Coastal Wetlands Consortium, A project of
the Great Lakes Commission, funded by the U.S. EPA GLNPO, March 2008, pp. 92–115, http://projects.
glc.org/wetlands/final-report.html. Accessed November 5, 2014.
Tiner, R.W. (ed.). 2009. Status Report for the National Wetlands Inventory Program: 2009. U.S. Fish and
Wildlife Service, Division of Habitat and Resource Conservation, Branch of Resource and Mapping
Support, Arlington, VA.
Townsend, P.A. 2001. Mapping seasonal flooding in forested wetlands using multi-temporal RADARSAT
SAR. Photogrammetric Engineering & Remote Sensing 67(7):857–864.
wetlands using C- band SAR. International Journal of Remote Sensing 23(3):443–460.
Turner, R.K. 1993. Sustainability: Principles and practice. In: R.K. Turner (ed.). Sustainable Environmental
Economics and Management: Principles and Practice. Belhaven Press, London, U.K., pp. 3–36.
Uzarski, D.G., Burton, T.M., Brazner, J.C., and Ciborowski, J.J.H. 2008a. Chapter 4. Invertebrate commu-
nity indicators. In: Great Lakes Coastal Wetlands Monitoring Plan, Great Lakes Coastal Wetlands
Consortium, A project of the Great Lakes Commission, funded by the U.S. EPA GLNPO, March 2008,
pp. 59–76, http://projects.glc.org/wetlands/final-report.html. Accessed November 5, 2014.
Uzarski, D.G., Burton, T.M., Brazner, J.C., Ciborowski, J.J.H. 2008b. Chapter 5. Fish community indicators.
In: Great Lakes Coastal Wetlands Monitoring Plan, Great Lakes Coastal Wetlands Consortium, A project
of the Great Lakes Commission, funded by the U.S. EPA GLNPO, March 2008, pp. 77–91, appendix 5-1,
5-2, http://projects.glc.org/wetlands/final-report.html. Accessed November 5, 2014.
Uzarski, D.G., Burton, T.M., and Ciborowski, J.J.H. 2008c. Chapter 2. Chemical/physical and land use/cover
measurements. In: Great Lakes Coastal Wetlands Monitoring Plan, Great Lakes Coastal Wetlands
Consortium, A project of the Great Lakes Commission, funded by the U.S. EPA GLNPO, March 2008,
pp. 29–31, http://projects.glc.org/wetlands/final-report.html. Accessed November 5, 2014.
Uzarski, D.G. and Otieno, S. 2008. Chapter 1. Statistical design. In: Great Lakes Coastal Wetlands Monitoring
Wei, A. and Chow-Fraser, P. 2006. Synergistic impact of water level fluctuation and invasion of Glyceria on
Typha in a freshwater marsh of Lake Ontario. Aquatic Botany 84(1):63–69.
Weiss, A.D. 2001. Topographic positions and landforms analysis (Conference Poster). In: ESRI International
User Conference, San Diego, CA, July 9–13.
Whitcomb, J., Moghaddam, M., McDonald, K., Kellndorfer, J., Podest, E. 2009. Mapping vegetated wetlands
of Alaska using L-band radar satellite imagery. Canadian Journal of Remote Sensing 35(1):54–72.
Wilcox, D.A., Thompson, T.A., Booth, R.K., and Nicholas, J.R. 2007. Lake-Level Variability and Water
Availability in the Great Lakes. U.S. Geological Survey Circular 1311, Reston, VA.
Wilen, B.O. and Tiner, R.W. 1989. The National Wetlands Inventory—The first ten years. In: Wetland Concerns
and Successes. American Water Resources Association, Middleburg, VA (September 1989), http://www.
fws.gov/wetlands/Documents/The-National-Wetlands-Inventory-The-First-Ten-Years.pdf. Accessed
November 5, 2014.
Wold, R.J., Paull, R.A., Wolosin, C.A., and Friedel, R.J. 1981. Geology of central Lake Michigan. American
Association Petroleum Geology 65:1621–1632.
Zedler, J.B. and Kercher, S. 2004. Causes and consequences of invasive plants in wetlands: Oppor
tunities, opportunists, and outcomes. Critical Reviews in Plant Sciences 23(5):431–452. doi:
10.1080/07352680490514673.
16 Mapping Wetlands and Surface
Water in the Prairie Pothole
Region of North America
Jennifer Rover and David M. Mushet
CONTENTS
Introduction..................................................................................................................................... 347
Prairie Pothole Region............................................................................................................... 347
Prairie Pothole Wetland Classification....................................................................................... 350
Remote Sensing of Prairie Pothole Wetlands.................................................................................. 352
Technologies Applied................................................................................................................. 354
Aerial Photography............................................................................................................... 354
Multispectral and Hyperspectral Data................................................................................... 355
LiDAR and Radar.................................................................................................................. 358
Case Study...................................................................................................................................... 359
Introduction................................................................................................................................ 359
Data and Analysis....................................................................................................................... 359
Results and Discussion...............................................................................................................360
Case Study Conclusion.............................................................................................................. 361
Limitations and Potential of Remote Sensing for Mapping Pothole Wetlands............................... 362
Conclusion...................................................................................................................................... 363
Acknowledgments...........................................................................................................................364
References.......................................................................................................................................364
INTRODUCTION
The Prairie Pothole Region (PPR) is one of the most highly productive wetland regions in the world.
Prairie Pothole wetlands serve as a primary feeding and breeding habitat for more than one-half of
North America’s waterfowl population, as well as a variety of songbirds, waterbirds, shorebirds, and
other wildlife. During the last century, extensive land conversions from grassland with wetlands to
cultivated cropland and grazed pastureland segmented and reduced wetland habitat. Inventorying
and characterizing remaining wetland habitat is critical for the management of wetland ecosystem
services. Remote sensing technologies are often utilized for mapping and monitoring wetlands.
This chapter presents background specific to the PPR and discusses approaches employed in map-
ping its wetlands before presenting a case study.
Prairie Pothole Region

The PPR of North America spans from south-central Canada to the north-central United States
(Figure 16.1). This vast area, encompassing over 800,000 km2, contains millions of wetlands
formed by Late Pleistocene glacial processes. The glacial processes resulted in a diverse landscape
347
Alberta Saskatchewan Manitoba Ontario
Prairie
Pothole
Montana Region
North Dakota
Minnesota
Canada
South Dakota Wisconsin
United States
Iowa
Nebraska
0 1,000 2,000
km
FIGURE 16.1 The PPR is located in central North America. (U.S. Geological Survey. 2003; Data for base
map is available at http://nationalatlas.gov. Accessed January 10, 2014.)
of depressions, which overlie glacial drift (Figure 16.2). Water accumulates in these depressions
and forms what are known as prairie pothole wetlands. Water retention periods vary among prairie
pothole wetlands and are driven by evapotranspiration, precipitation, water table conditions, land
use, and the lithology and structure of the underlying glacial till (Kantrud et al. 1989a,b; Winter
1989). Vegetation is often arranged in concentric plant community zones within the depressions
(Figure 16.2). Prairie pothole wetlands contain one or more vegetative zones (e.g., wet meadow,
shallow marsh, and deep marsh) that can be used to distinguish major wetland classes (Stewart and
Kantrud 1971). These vegetative zones are dynamic, and spatial locations and plant community
compositions of zones change in response to water-level changes occurring during wet and dry
periods (Kantrud et al. 1989b; Tiner 1996). During these periods, wetlands can shift from one class
to another class.
The PPR climate exhibits high spatial and temporal variability. Temperatures can average nearly
−20°C during the winter months, while summer temperatures can average over 20°C (Kantrud et al.
1989a). Precipitation is variable, with western and northern portions of the region generally receiv-
ing less precipitation than eastern and southern portions, averaging nearly 30 and 60 cm, respec-
tively (Kantrud et al. 1989a; Millett et al. 2009). The climate results in a negative atmospheric water
balance ranging from −60 cm in the northwestern PRR to −10 cm in the southeastern PPR (Winter
1989). Occasionally, periods of extreme drought and deluge occur (Winter and Rosenberry 1998).
For example, the Palmer Hydrological Drought Index (PHDI) for central North Dakota, in Figure
16.3, indicates an extended dry period in the late 1970s and again in the late 1980s into the early
1990s, followed primarily by wet years.
The PPR is well known for its importance to waterfowl and other wetland-dependent wildlife. On
average, 50% of the North American game duck population is produced in the PPR during any given
year (Smith 1995). Waterfowl present during spring and summer months are supported by habitat
Mapping Wetlands and Surface Water in the Prairie Pothole Region of North America 349
FIGURE 16.2 An oblique air photo shows prairie pothole wetlands and hummocky terrain formed by glacial
processes. Plant community zones are visible as concentric rings within many wetlands.
2
PHDI
–2
–4
–6
1975 1980 1985 1990 1995 2000 2005 2010
FIGURE 16.3 PHDI is an indicator of long-term cumulative hydrological conditions. Recent periods of
drought and deluge between 1975 and 2013 are apparent in the PHDI for central North Dakota. (National
Oceanic and Atmospheric Administration; Data downloadable at http://www.ncdc.noaa.gov/cag/. Accessed
January 23, 2014.)
provided by the vast number of prairie pothole wetlands distributed across the landscape. In addi-
tion to waterfowl, prairie pothole wetlands are habitat for other wetland-dependent birds (Igl and
Johnson 1998), amphibians (Larson et al. 1998), small mammals (Fritzell 1989), and a wide diversity
of aquatic invertebrates (Euliss et al. 1999). Fish are largely absent from prairie pothole wetlands
due to the fluctuating and often transitory nature of their standing water (Kantrud et al. 1989a).
However, this lack of fish contributes to the value of prairie pothole wetlands to species that flourish
in their absence (e.g., amphibians) (Mushet et al. 2012). The grassland habitats surrounding many
prairie pothole wetlands provide habitat for additional species, including several rare grassland
birds (Swengel and Swengel 1998).
The native land cover in the PPR is mixed-grass prairie in the western region and tallgrass prai-
rie in the eastern region. As permanent settlement began over a century ago, large areas of prairie
were cultivated for crop production, impacting the general condition of wetlands (Dahl 1990; Mita
et al. 2007). As of 2000, only about 40% of the mixed-grass prairie and 2% of the tallgrass prairie
remained intact in the northern Great Plains while the majority of land was converted to cropland
(Higgins et al. 2002). With the expansion of cultivated agriculture, wetlands are drained by surface
ditching and/or the installation of underground drainage tile (Anteau 2012). Miller et al. (2012)
reported only 3%–4% of wetlands remain in the Iowa portion of the PPR due to drainage.
Prairie Pothole Wetland Classification

Since wetland mapping efforts often strive to delineate wetland types across a geographic area of
interest, these efforts have been closely linked to the development of wetland classification systems.
The first inventory of wetlands in the PPR was conducted in 1954 by the U.S. Fish and Wildlife
Service as part of a national effort to assess wetland resources (Shaw and Fredline 1956). This early
wetland assessment utilized a classification system developed by the U.S. Bureau of Sport Fisheries
and Wildlife (Martin et al. 1953) to distinguish different wetland types and report wetland acreage
available as waterfowl habitat. The classification system of Martin et al. (1953) was based on water
depth during the growing season, cover interspersion, and plant species composition. Unlike many
regions of the United States where wetlands <1.6 ha were not included in the survey, the north-
central states region had nearly complete coverage given the region’s importance as a waterfowl
breeding area. Although the inventory in the north-central states was thorough, the classification
of the majority of the PPR wetlands included only a few of the 20 types of wetlands inventoried at
the national level.
Recognizing that the classification system of Martin et al. (1953) was too generalized to be
useful for wetland researchers and managers in the PPR, Stewart and Kantrud (1971) developed a
system specific to the glaciated prairie region. Their system recognizes the strong influence of water
permanence, water depth, and water chemistry on prairie wetlands and groups wetlands into seven
distinct classes. For example, Stewart and Kantrud (1971) classify wetlands or lakes as ephemeral,
temporary, seasonal, semipermanent, permanent, alkali, or fen. In addition, subclass modifiers indi-
cate differences in water chemistry (i.e., salinity) within a class. The photos in Figure 16.4 provide
examples of common wetland classes and subclasses. Within a class and subclass, wetlands can be
further distinguished by phases that reflect the influence of surrounding land use on wetland plant
communities. Vegetation zones play a key role in the Stewart and Kantrud (1971) classification sys-
tem, making this regional system well suited for remote sensing applications as vegetation zones in
prairie pothole wetlands change in response to water permanence. The Stewart and Kantrud (1971)
system is still utilized by many wetland researchers and managers in the U.S. portion of the PPR.
Millar (1976) developed a similar, region-specific classification system for the prairie wetlands of
western Canada.
The first standard wetland classification maps for the U.S. PPR were created in 1975 by the
Fish and Wildlife Service using color-infrared (IR) aerial photography flown by the National
Aeronautics and Space Administration (Tiner 1996). For this project—the National Wetlands
Inventory (NWI)—wetlands in the conterminous United States were classified using a nationwide
classification system developed by Cowardin et al. (1979), in which the majority of PPR wetlands
are grouped into a single class (palustrine). However, water chemistry and water regime modifiers
allowed for wetlands to be subdivided into groupings similar to those of the Stewart and Kantrud
(1971) system. A key difference between the two systems is that the Cowardin et al. (1979) system
(a) (b)
(c) (d)
(e)
FIGURE 16.4 Examples of pothole wetlands and water bodies: (a) slightly brackish temporary wetland, (b)
moderately brackish seasonal wetland, (c) slightly brackish semipermanent wetland, (d) brackish permanent
lake, and (e) saline permanent lake.
allows for portions of a single wetland to be classified separately, while the Stewart and Kantrud
(1971) system utilizes the central, deepest zone to assign a single classification to the entire wetland.
However, the complex multipolygon wetlands in NWI maps based on the Cowardin et al. (1979)
system can be simplified to create a Stewart and Kantrud (1971) classification by merging complex
multipolygons for each wetland into a single wetland polygon. The resulting single wetland polygon
can be assigned a classification reflective of the most permanent water regime present in the basin
(e.g., Johnson and Higgens 1997; Niemuth et al. 2010).
The Canadian Wetland Inventory (CWI) is a classification system that groups wetlands by class
(i.e., bogs, fens, marshes, swamps, and shallow/open water), form (based on water and soil charac-
teristics), and type (based on plant communities). While the Canadian wetland classification system
(National Wetlands Working Group 1997) provides general information similar to the Cowardin
et al. (1979) system used in the United States, it does not provide classifications as detailed as the
region-specific system of Stewart and Kantrud (1971) or Millar (1976). However, similar to the NWI
of the United States, the CWI is an invaluable tool in the monitoring and management of Canada’s
valuable wetland resources.
In order to facilitate functional assessments of wetlands, Brinson (1993) developed a hydrogeo-
morphic (HGM) classification for wetlands. The Brinson (1993) classification utilizes geomorphic
setting, water source, and hydrodynamics to place wetlands into functional classes. This classifi-
cation lays the foundation for assessments of the physical, chemical, and biological functions of
wetlands. The HGM approach focuses on the abiotic features of wetlands in a manner designed to
provide greater insights into the relationships between organisms and their environment.
Most recently, Euliss et al. (2004) proposed a framework that, rather than placing wetlands into
specific static categories, emphasizes the continuous nature of the spatial and temporal gradients
influencing wetland abiotic processes and biotic communities. The “wetland continuum” described
by Euliss et al. (2004) places wetlands along two continuous axes representing a wetland’s unique
hydrologic relation to groundwater and temporal position as it relates to atmospheric water inputs.
The placement and analysis of wetlands along continuous gradients reflective of their occurrence
in nature may represent the next step forward in the classification and mapping of PPR wetlands.
REMOTE SENSING OF PRAIRIE POTHOLE WETLANDS

Steward and Kantrud (1971) discussed various approaches developed for classifying wetlands spe-
cific to the PPR. Many classification approaches require water chemistry, identification of wet-
land plant species and their composition, or even water depth to characterize individual wetlands;
however, these parameters are not often directly measurable from remotely sensed data (Adam
et al. 2009). At the same time, several classification approaches use wetland characteristics such as
water permanence, wetland density, wetland distribution, basin area, basin capacity, basin drainage,
and cover interspersion that can be measured with remote sensing or in combination with other
geographic data. In addition, Steward and Kantrud (1971) found that waterfowl are influenced by
water depth, water chemistry, water permanence, and land use; the last two factors listed can be
determined directly from remotely sensed data. Biologists agree that water permanence and wet-
land density is an indicator of waterfowl breeding populations (Crissey 1969; Brewster et al. 1976;
Johnson and Grier 1988). A recent study also found greater probabilities of duck brood occupancy
in wetlands with larger wet areas and herbaceous perennial vegetation (Walker et al. 2013).
Given the important role of wetlands in the region, the dynamic nature of prairie pothole wet-
lands, and land use changes occurring in the uplands, understanding and assessing wetland condi-
tions requires high-resolution data. Unfortunately, the remotely sensed data currently available for
the region are temporally, spatially, or spectrally limited (Adam et al. 2009). Even with the techno-
logical limitations, sensors and wetland mapping science have rapidly advanced over the last few
decades. Initial studies were limited to aerial photography and airborne digital sensors for evaluat-
ing the effectiveness of remote sensing of wetlands in the PPR. Current technologies now include
sensors onboard satellites and aircraft with high spatial resolution, high-spectral resolution, and
high-temporal resolution. Although many sensors collect high-resolution data, each sensor typically
provides only one or two aspects (spatial, spectral, or temporal) of the resolution required for most
remote sensing wetland applications.
Landsat sensors and aerial photography are the most common source of remotely sensed data
for mapping wetlands in the PPR. Aerial photography was the primary data source for the NWI
mapping in the PPR (Tiner 1996). Studies requiring high-resolution data may utilize airborne data
for small geographic study areas due to the development of the technology, ease of use, and afford-
ability. Likewise, the use of hyperspectral data or high-resolution satellite data such as Satellite
Pour l’Observation de la Terre (SPOT) and IKONOS for small regions or low-temporal frequency
mapping is common due to their smaller footprint size and cost. The multispectral data provided
by space-based sensors are advantageous over sensors with limited spectral range. For instance,
automatic or semiautomatic classification approaches such as maximum likelihood, clustering, or
classification and regression tree (CART) have been shown to be useful for delineating water, wet-
lands, and upland vegetation. In addition, operational programs, such as Landsat, provide data for
multitemporal analysis of wetlands and their uplands. The differences in moderate-resolution mul-
tispectral Landsat data and high-resolution natural color digital aerial photography are apparent in
Figure 16.5. Landsat data in Figure 16.5a and b illustrate the impact of weather variations on the
number and size of wetlands for the same geographic extent in North Dakota (see PHDI in Figure
16.3 for drought indication). The third image, Figure 16.5c, was created using an aerial photograph
acquired during peak growing season (USDA-FSA Aerial Photography Field Office 2012).
The spectral data collected by space-based platforms generally provide information regarding
the presence and quality of water, the presence and condition of vegetation, and wetness of soils.
Water, vegetation, and soils have distinct spectral characteristics. Those distinct characteristics
change in response to plant phenology, local climate, soil water content, topography, and vegetation
0 2 4 km
(a) (b) (c)
FIGURE 16.5 Landsat data and NAIP natural color aerial photography of the same area in Stutsman County,
North Dakota, illustrate the dynamic nature of prairie pothole wetlands in both (a) dry and (b and c) wet condi-
tions. The satellite data were acquired on September 2, 1992, with Landsat 5 (a; bands 5, 4, and 3) and May
23, 2013, with Landsat 8 (b; bands 6, 5, and 4; U.S. Geological Survey; data available at http://earthexplorer.
usgs.gov. Accessed October 16, 2012). The (c) NAIP data were collected near the peak of the growing season
on July 31, 2012. (USDA-FSA Aerial Photography Field Office 2012; Data available at http://www.nd.gov/gis/
mapsdata/download/. Accessed January 28, 2014.)
stress (Adam et al. 2009). Water typically absorbs most of the energy in the near- to mid-IR wave-
lengths (~0.08–3 µm), while soil and vegetation tend to reflect energy in these wavelengths (Adam
et al. 2009). Unfortunately, mixes of vegetation and water commonly found in wetland zones can be
difficult to decipher, but Phillips et al. (2005) noted separability of plant community zones late in
the growing season. Even suspended sediments and chlorophyll in PPR wetlands can add complex-
ity due to similar spectral signatures between wetlands and uplands (Ritchie et al. 2003). Adam
et al. (2009) reviewed applications for mapping wetland vegetation using multispectral and hyper-
spectral remote sensing. The authors found that estimated biophysical and biochemical properties
and spectral characteristics of wetland vegetation fluctuated with channel width and resolution of
the data, limiting current technology with insufficient spatial and spectral resolutions for distin-
guishing narrow vegetation zones found in most wetlands (Ozesmi and Bauer 2002; Adam et al.
2009). In addition, canopy and leaf properties are similar for many vegetation species and produce
correlations in spectral reflectance. Environmental factors introduce variance at nearly the same
rate as differences in biophysical variables.
Data from passive sensors are utilized for most remote sensing applications in the PPR. This
may be driven primarily by data availability, but the physical conditions of PPR wetlands and their
surrounding uplands tend to be more conducive than forested wetland regions to passive data collec-
tion. Typically, a substantial number of days are clear during the growing season, sun angle is suffi-
cient, and land cover is primarily cropland, grassland, and pasture with limited tree cover. However,
during winter months or wet cycles, cloud cover can often reduce the quality of the image and snow
cover tends to obscure most of the underlying landscape. This limits most passive sensor mapping
applications to the growing season, which ranges from March, April, or May to September, October,
or November across the region.
For applications where atmospheric conditions are not conducive for passive sensors, informa-
tion on wetness is critical, or where data are needed regarding vegetation and land surface height,
active sensors such as synthetic aperture radar (SAR) and Light Detection and Ranging (LiDAR)
can provide useful information for wetland mapping. Although data from active sensors are comple-
mentary to passive data, the use of radar systems for wetland remote sensing in the PPR has been
limited. However, radar products show potential for mapping water area, soil moisture, and specific
vegetation features (Bhang et al. 2010). Data from active sensors, including SAR, are also being
investigated for their ability to resolve some of the spectral confusion found in data from passive sen-
sors (Henderson and Lewis 2008). Within the PPR, a study evaluating the use of SAR for the devel-
opment of products for the CWI found the inclusion of radar data promising (Brisco et al. 2011).
Airborne active sensors, such as LiDAR, also provide a third dimension to spectral data. LiDAR
data can discern vegetation structure, as well as topology; however, initial data collection and data
processing with current technology is complex (Renslow 2012). High-resolution topographic infor-
mation derived from existing LiDAR datasets collected in the PPR offered Huang et al. (2011b) and
Shook and Pomeroy (2011) an opportunity to model the water storage capacities of prairie wetland
basins.
Technologies Applied
Aerial Photography
Aerial photography provided the earliest remotely sensed data for wetland mapping in the PPR.
Several government agencies have aerial data collections, with the first acquisitions during the 1940s
focused primarily on documenting agricultural croplands. This historical imagery is archived in
collections maintained by the National Archives and Records Administration, the U.S. Geological
Survey, the U.S. Department of Agriculture Aerial Photography Field Office, and the National Air
Photo Library of Natural Resources Canada. Although aerial photography collections provide spa-
tially consistent and unique historical perspectives, most are limited in temporal and spectral reso-
lution and can be difficult to obtain. Using aerial photography for reliable identification of wetlands
can be problematic as most federally funded aerial photos collected in the PPR were acquired dur-
ing the middle of the growing season, a time when crop types were best distinguishable but when
many small wetlands had already dried (Tiner 1999). Currently, aerial photography provides the
most suitable high-resolution data for accurate wetland mapping, but aerial photos may have limited
spectral resolution that necessitates manual classification approaches to delineate wetlands and their
features (Adam et al. 2009).
Although the use of aerial photography is often limited by the scale of a project or seen as a
complement to satellite data, aerial photography has been a primary data source for several prai-
rie pothole wetland studies. For example, eight sets of aerial photography acquired over a 3-year
period and in situ data gathered from field visits were used to evaluate wetland characteristics at a
plot level in North Dakota (Cowardin et al. 1981). Class from Stewart and Kantrud (1971), density,
and area of wetland basins were estimated from the aerial photography. The majority of wetlands
were classed as temporary or seasonal. Semipermanent wetlands were limited in numbers but con-
tributed more to the total wetland area as they are typically larger. Water permanence and class,
according to Stewart and Kantrud (1971), tended to correlate with basin area, prompting further
study. The authors attempted to classify wetlands based solely on size and wetness. Unfortunately,
the approach resulted in a 33% misclassification, with the identification of seasonal wetlands most
problematic (Cowardin et al. 1981). The study concluded that acquisition timing of aerial photogra-
phy is essential for mapping the number and area of basins, and when water is present, plant com-
munities are indiscernible.
A recent study by Niemuth et al. (2010) collected aerial and video photography in May of each
year from 1987 to 2006 to document the proportion of wet area within wetland basins. The study
area consisted of hundreds of sampling blocks located across the PPR of Montana, North Dakota,
and South Dakota. From the data, spatial and interannual patterns of wetness and their variations
were mapped using supervised classification and photo interpretation. The interannual patterns of
wetness were compared to the NWI water regimes assigned to each of the basins. Niemuth et al.
(2010) found that temporary basins had the lowest percentage of water, the highest coefficient of
variation, and lowest correlation with prior year wetness. In contrast, lakes had higher correlations
with the prior year wetness and lower coefficients of variation. The study did not find an east-to-west
gradient in wetness as suggested by the regional temperature and precipitation gradient.
Multispectral and Hyperspectral Data

A new era of mapping wetlands in the PPR with remote sensing began in the late 1960s when air-
borne multispectral scanners (MSS) and computer processing techniques were investigated (Burge
and Brown 1970; Nelson et al. 1970; Work et al. 1974). In the 1970s when MSS satellite data became
publically available, the data were initially limited to a few organizations due to the equipment
and trained personnel required to process the data into information. Several early studies using
Landsat 1, formerly known as Earth Resources Technology Satellite (ERTS-1), used MSS data
and automated mapping approaches to detect open water and classify vegetation in prairie pot-
holes (Work and Gilmer 1976; Best and Moore 1979; Gilmer et al. 1980). The primary objective of
those early investigations was to develop remote sensing applications for the management of migra-
tory waterfowl. Pond numbers, area, and distribution estimated from remotely sensed data were
expected to be useful for determining waterfowl production and wetland habitat carrying capacity
(Gilmer et al. 1974; Work et al. 1974).
Landsat 1 MSS data were used for many of the first space-based wetland mapping studies in
the PPR (Gilmer et al. 1974; Work et al. 1974; Work and Gilmer 1976; Gilmer et al. 1980). Gilmer
et al. (1974) used the near-IR band (band 7, 0.8–1.1 μm) from a mid-season Landsat 1 MSS image
to detect open surface water and determine water body size and vegetation classes. Areas of open
water <1.6 ha were not detected using the single near-IR band. Additional work found multispec-
tral data from MSS aided in the identification of basic land cover classes including aquatic vegeta-
tion, pastures, cultivated land, and bare soil (Gilmer et al. 1974). Gilmer et al. (1974) also reported
confusion between spectral classes may be improved with multitemporal data that distinguishes
vegetation classes based on phenological differences. Work et al. (1974) also found pixels could
be classified as water using a single near-IR channel threshold, although water bodies <0.9 ha in
area were identified inconsistently. Using estimates of fractions of water, as a partial component
of each pixel, multichannel data were a more accurate method for estimating water extent (Work
et al. 1974).
In a later study, Work and Gilmer (1976) expanded their prior research using the near-IR Landsat
band to identify pixels containing surface water. With a thresholding technique, changes in wetness
in the spring and summer for a 2-year period were investigated for an area in eastern North Dakota.
As the authors anticipated, water area in wetlands decreased between the spring and summer sea-
son. The approach identified almost all wet areas >1.6 ha while inconsistently recognizing wet
areas with sizes ranging from 0.4 ha, the minimum size detected, to 1.6 ha. Further analysis of the
subpixel components (i.e., water, bare soil, and green vegetation) permitted consistent delineation
of wet areas >0.5 ha, while the smallest wetlands identified were 0.13 ha.
Vegetation and water quality parameters were also studied in the PPR using multispectral data
from the MSS sensor and low-altitude multispectral aerial photography (Best and Moore 1979).
Landsat MSS data were used to classify lakes based on their spectral properties and stratified
with ground sampling. Basin volumes were estimated using relationships derived from modeled
regressions of field-based depth profiles and morphometric measurements from the imagery.
Multitemporal analysis of composited MSS data acquired in August and in January also detected
areas with emergent vegetation remaining visible in the snow-covered landscape.
In the 1980s, Gilmer et al. (1980) incorporated high-resolution aerial photography with a double
sampling approach to improve estimates of small wetlands (as small as 5 m diameter) from Landsat
MSS data. Open surface water was delineated using an earlier established thresholding approach
(Work and Gilmer 1976). Wetlands were counted and plotted for areas with coincident Landsat
and aerial photo coverage for two dates in May and July 1975. Linear regression was then used to
estimate the total number of wetlands. The coefficients of determination (R2) were 0.65 for the May
data and 0.74 for the July data.
An early study using data from Landsat Thematic Mapper (TM) evaluated the technology by
comparing data from TM to NWI. Jacobson et al. (1987) compared a TM scene from May 1986 and
NWI data developed from aerial photography acquired in May 1979. They found omission errors
were high for small wetlands (78%) and commission errors were low for all size classes (4%). The
30 m Landsat pixel size, mixed pixel effects, and differences in the acquisition dates were likely
contributors to the high omission errors. The authors also noted a discrepancy between the size of
coincident wetlands estimated from the aerial photography and those from the classified Landsat
data. The Landsat TM classification increasingly underestimated the areal extents of wetlands as
the size class increased (Jacobson et al. 1987).
By the 1990s, use of remotely sensed satellite data for large-scale wetland mapping investigations
expanded with advancements in classification approaches and high-resolution satellite data. Dechka
et al. (2002) investigated IKONOS-2 imagery acquired in the spring and summer and classified wet-
lands based on the Stewart and Kantrud (1971) wetland habitat class system. Supervised and unsu-
pervised classification approaches were tested, although the study only included a 10 km × 10 km
area. Linear discriminant analysis and unsupervised ISODATA classification methods resulted in
low accuracies (~47%). The authors concluded that the low accuracies were a result of applying a
ground-based classification system to remotely sensed data. Once field data collection was modified
to account for dominant communities, the accuracies increased (84%).
Sethre et al. (2005) estimated changes in the water area of wetlands in North Dakota between
1991 and 2002 using Landsat TM and Enhanced Thematic Mapper Plus (ETM+) mid-IR (band 5,
1.55–1.75 μm) density-sliced data and a subpixel classification method. A spectral unmixing tech-
nique, Applied Analysis Spectral Analytical, was applied to detect the proportion of open water
in a pixel. The accuracy of the subpixel classification, albeit difficult to measure, was found to
be inconsistent for small wetlands. The subpixel classification and the density-sliced method had
similar accuracies for water areas >0.3 ha, but the subpixel classification was most useful for the
detection of small water bodies and delineating edges of larger water bodies.
Landsat data and ground-based spectral data were investigated by Beeri and Phillips (2007).
The authors first estimated wetland hydroperiod using Landsat data acquired during the spring,
midsummer, and late summer in the years 1997–2005 (Beeri and Phillips 2007). Water pixels were
grouped by water occurrence frequency during the growing season each year while the mean and
standard distances were calculated across years. Omission errors only occurred where the water
body was less than one pixel in size. As part of the study, ground-based spectral data and turbidity
were also investigated. Landsat pixels were classified as water, soil, aquatic plants, and terrestrial/
emergent vegetation by applying a threshold to specific indices developed from the ground-based
spectral data. Beeri and Phillips (2007) found that turbidity influenced spectral data in the 0.45–
0.70 μm range, therefore rendering Landsat visible band data for separation of turbid water and
vegetation ineffective. In the near-IR region of the spectrum, 0.75–0.90 μm, turbid water and soil
also had similar properties. The mid-IR region (band 5 and 7, 1.55–1.75 μm and 2.08–2.35 μm)
had the greatest spectral separation between turbid water, vegetation, and soil. The study found
that two normalized difference indices, one using Landsat bands 5, 7, 2, and 3 and the other using
bands 5, 7, and 1, were useful for the separation of turbid water from the surrounding land cover
in the PPR region.
Zhang et al. (2009b) also used Landsat data to classify and analyze water area in northeastern
South Dakota. Water area was estimated from multidate imagery acquired in 1985–2002 using a
CART approach and a digital orthophoto quarter quadrangle to estimate the fraction of water in
mixed Landsat pixels. Analysis of the regression tree structure showed that bands in the near-IR and
visible red portion of the spectrum (bands 4, 5, 7, and 1) were most useful for detecting water. The
approach was found to be more accurate than an unsupervised method at estimating water area for
small wetlands (20% error for water <0.5 ha; <10% error for water ≥1 ha). Power laws were devel-
oped for the years 1990, 1992, 1997, and 2002 using the number of water bodies, their size class, and
a Bayesian hierarchical linear regression model Zhang et al. (2009a). Interannual distributions had
similar slopes while intra-annual distributions did not because small temporary wetlands typically
dry out during the growing season. The study also examined slopes for water bodies delineated
from aerial photography acquired during a drought in July 1939. The severe drought was evident
in the patterns of variability in 1939 aerial photography, but the smaller water bodies were more
impacted by a less severe drought ending in the early 1990s. Simulations of wetland hydrologic
response to climate in North Dakota were also investigated using the Landsat water masks devel-
oped by Zhang et al. (2009a) and Liu and Schwartz (2011, 2012).
Landsat TM and ETM+ data were also used to classify wetlands in central North Dakota (Rover
et al. 2011). Following the classification approach suggested by Euliss et al. (2004), the authors
assigned wetlands to classes based on changes in water area within a wetland for a drought to deluge
period evident in seven Landsat acquisitions from 1989 to 2005. Water area was delineated using
a decision-tree approach and digital elevation model (DEM) data. The delineations were classified
into four functional groups (open and closed basin discharge, flow-through, and recharge) based on
water dynamics present in each wetland using a clustering for large applications algorithm (Kaufman
and Rousseeuw 2005). The final classification accuracies were highest for closed-basin wetlands
receiving groundwater discharge (user’s accuracy, 97%; producer’s accuracy, 71%). Recharge, flow-
through, and open-basin discharge wetlands were more difficult to classify accurately.
Although studies utilizing high-resolution data for wetland mapping in the PPR are limited,
a few recent studies have investigated data from high-spectral, spatial, and temporal resolution
sensors. One study developed models from hyperspectral, Landsat ETM+, and SPOT 5 data for
classifying four semipermanent and permanent wetlands located on the Missouri Coteau in North
Dakota (Phillips et al. 2005). A hyperspectral library in the 0.35–2.5 μm wavelength range was
created for wetland plant communities during the growing season and aggregated to channel
widths of the SPOT 5 and Landsat ETM+ data for classification and analysis. Sufficient spectral
variance in wetland plant community zones was found only during August and early September.
This coincided with lower water levels and seasonal vegetation growth peaks. Band channels in
the green, red, and mid-IR, available with Landsat, were found to account for the variance, but
distinction of narrow vegetation community zones within the 30 m pixel was not possible. With
the inclusion of SPOT 5 data, the authors found the approach accounted for more of the vegetation
community zones.
Decision-tree classifications applied to high-resolution data have also been investigated for their
use in mapping emergent vegetation. Lawrence et al. (2006) classified IKONOS data (4 m pixel
size) acquired in August 2000 and October 2000 into three hierarchical levels using a decision-tree
approach. Spectral data from the summer and fall image, a difference image, and principal compo-
nents of both images served as explanatory variables for three hierarchical wetland classes, that is,
Level 1, Level 2, and Level 3. Wetlands with emergent vegetation had high accuracies for each level.
The segregation of upland vegetation components into grass, woody, and specific species at Level 2
and Level 3 reduced the classification accuracy due to confusion between classes.
Coarse resolution remotely sensed data (~500–1000 m pixel size) are not commonly used for
wetland mapping applications in the PPR, but a recent study explored phenological metrics derived
from the Advanced Very-High-Resolution Radiometer (AVHRR). In the study, Herfindal et al.
(2012) evaluated data from the May waterfowl breeding population and habitat survey (Smith 1995)
and several environmental variables, including phenology derived from a normalized difference
vegetation index (NDVI) calculated from the AVHRR data. Using univariate analysis, the explana-
tory power of environmental variables on May pond counts was investigated. The explanatory vari-
ables from bimonthly NDVI composites for the years 1982–2001 included the start of spring day,
green-up prior to the start of spring, growing season peak NDVI, end-of-season day, and integrated
NDVI. The NDVI variables, with the exception of end-of-season day, were spatially correlated
(p < 0.05) with pond counts from the May waterfowl breeding population and habitat survey, an
indicator for several waterfowl populations.
LiDAR and Radar

Although passive sensors are the primary technology currently employed for wetland classification
and mapping applications in the PPR, the use of active sensor technology is increasing. LiDAR data
are available for some areas of the PPR, and the use of this technology will most likely increase
with availability (http://earthexplorer.usgs.gov/. Accessed November 6, 2014; http://agrg.cogs.nscc.
ca/projects/LiDAR_Metadata. Accessed November 6, 2014). LiDAR data offer advantages for wet-
land applications that are not obtainable from passive sensors. For example, Fang and Pomeroy
(2009) used LiDAR data to generate bare earth models, vegetation height measurements, and other
geographic data, to estimate the spatial distribution of snow water equivalent for an area of the PPR
in Saskatchewan. In another study, Huang et al. (2011b) estimated water storage capacity of wetlands
for a 196 km2 area in North Dakota using a DEM derived from airborne LiDAR data.
Integrating active data with passive data also provide advantages. Huang et al. (2011a) devel-
oped a wetland water area index using data from both passive and active sensors. Water area was
extracted from Landsat data using a supervised maximum likelihood classification of the Landsat
bands, NDVI, and a normalized difference water index (NDWI). Manually interpreted aerial pho-
tography, LiDAR data, and an index developed from the Palmer Drought Severity Index (PDSI)
were used to predict and model water distribution.
Airborne LiDAR data were also used to investigate common data processing errors introduced
while generating a LiDAR-derived DEM (Li et al. 2011). Exploring the impacts of smoothing,
coarsening, area thresholds, and depth thresholds revealed that a 10–20-time smoothing opera-
tion, an area threshold of 200 m2, and a depth threshold of 0.1 m were appropriate for accurately
identifying depressions in the Canadian PPR. Using LiDAR for another study in the Canadian PPR,
Shook and Pomeroy (2011) created a model from a LiDAR-derived DEM to simulate water area
and investigate hysteresis. The results indicate that modeling with nonlinear transfer functions may
account for the dynamic water fluctuations observed in prairie wetlands.
Brisco et al. (2011) evaluated the use and accuracy of airborne SAR data collected in September
as a potential data source for the CWI. The study investigated separability analysis and maximum
likelihood classification of three first-level classes (i.e., open water, deep marsh, and shallow marsh)
and seven second-level classes (i.e., open water, bulrush [Scirpus], cattail [Typha], grasses [Poaceae],
reed [Phragmites], sedge [Carex], and whitetop [Lepidium]). The classification accuracies of C-band
single polarization, dual polarization, multipolarization, and polarimetric combinations were evalu-
ated for the first- and second-level classifications. Results indicated that a polarization ratio (HH/
HV) and decomposition (Freeman and Durden 1998) were most useful for delineating open water,
deep marsh, and shallow marsh. The most accurate second-level classification (~65%) used Cloude–
Pottier eigenvalues. Subsequently, Brisco et al. (2011) suggested that vegetation species-level clas-
sification accuracies may be improved to an acceptable level for operational wetland mapping by
including optical data.
CASE STUDY
Introduction
A study area located in south-central North Dakota known as Cottonwood Lake study area (CLSA)
has served as a wetland research site for several decades. Data pertaining to water quality, water
table depth, soil permeability, and land use are readily available for investigating remotely sensed
data applicability. In particular, changes in water area within wetland basins provide useful infor-
mation regarding hydroperiod, hydrologic function, land use, and water quality. Previous analysis
using Landsat data concluded that wetland changes were detectable, and differences in water area
could be classified (Rover et al. 2011) based on the framework proposed by Euliss et al. (2004).
Further analysis shows encouraging capabilities for this region as the approach allows for the clas-
sification of the hydroperiod for portions of each wetland basin. Rather than classifying individual
pixels from Landsat data, the wetland basin is subdivided using digital elevation data.
Data and Analysis

The approach uses archived Landsat data that provide a historical perspective of wetland change
and hydroperiod. In the PPR, weather patterns tend to be variable (Millett et al. 2009). Using data
that span several decades tends to capture wet area response under multiple wet-to-dry and dry-to-
wet cycles. For the CLSA and surrounding area, 10 coinciding Landsat scenes were selected with
various acquisition dates throughout the growing season for the years 1973–2011 (Table 16.1). Scene
selection was prioritized to represent decadal-scale differences in climate and also minimal cloud
cover.
Water area was classified for each Landsat scene using a supervised decision tree, a nonpara-
metric rule-based classifier approach (Quinlan 1993; Lawrence and Wright 2001). A decision-tree
approach using radiance data from standard terrain correction (Level 1T) was found to be a reliable
method to differentiate water and nonwater pixels (Friedl and Brodley 1997). The decision-tree
model predicted whether a pixel was water or nonwater based on training data manually selected
from each scene. The training data included radiance, NDVI, NDWI, and the principle components
of the radiance data. Within the decision-tree modeling environment, the training data values serve
as independent variables for generating a decision tree for the image where the dependent water and
nonwater pixels can be predicted reliably. In many instances, additional training data were added to
represent pixels where the model showed low confidence in the prediction.
For regions like the PPR where water area fluctuates, multiple basins may converge during
wet periods. Capturing these events is critical for wetland classification but may pose issues for
TABLE 16.1
Satellite, Sensor, and Acquisition Date for Landsat Data
Included in North Dakota Case Study
Satellite and Sensor Acquisition Date
Landsat 1 MSS 10-06-1973
Landsat 5 TM5 05-21-1989
Landsat 5 TM 06-12-1991
Landsat 5 TM 07-14-1997
Landsat 7 ETM+ 07-01-2001
Landsat 5 TM 05-12-2003
Landsat 7 ETM+SLC-off 07-09-2004
Landsat 5 TM 06-18-2005
Landsat 5 TM 08-13-2008
Landsat 5 TM 07-05-2011
time-series analysis. Water variance should be analyzed for each individual wetland, or portions
of a wetland, rather than aggregating conditions for a group of coalescing wetlands. The approach
described in Rover et al. (2011) that delineated wetland basins with digital elevation data was
employed for our case study. Digital elevation data are available for the PPR at various resolutions
ranging from submeter LiDAR-derived data to 10 or 30 m topographic map-derived data (http://
ned.usgs.gov). For this example, the 10 m DEM was found to be sufficient for delineating coalescing
wetlands and capturing sufficient basin morphology.
A database for analysis was created by integrating the water maps developed from the decision-
tree classification and the basins delineated from the DEM analysis. For this process, the water
areas from each image were extracted as vector polygons, and within the DEM basins, water area
was summed at each date in the time series. To avoid including data with cloud contamination,
cloud and cloud-shadow masks were developed and used to notate acquisition dates where mea-
surements were unreliable. The summation produced a record for each basin that contained water
area at the given acquisition date that could be compared to maximum water extent, as well as the
DEM-derived basin area.
Wetlands were classified to determine general patterns in water area change at each acquisi-
tion date. Statistical packages, as well as image processing and GIS software, have multiple algo-
rithms for processing multitemporal data. For this example, an expectation–maximization (EM)
clustering algorithm in Weka (http://weka.wikispaces.com [Hall et al. 2009]) was used to generate
clusters from the database while ignoring missing observations due to cloud and shadow contamina-
tion. This unsupervised classification approach grouped water dynamics (i.e., water present at each
image date) into seven classes based on cross-validation. The resulting EM model was applied to
assign basins, with similar temporal dynamics, to the same class.
Results and Discussion

The EM classes represented differences in hydroperiod, an important characteristic of wet-
land hydrology, and a necessary component of mapping wetlands in the PPR (Winter 1989;
Euliss et al. 2004; Rover et al. 2011). Figure 16.6 exhibits the results of the EM clustering clas-
sification. Although each of the seven classes varied temporally, most classes had a dramatic
increase in water area occurring in the 1997 acquisition, which corresponds to a prior wet period
(Figure 16.3). Class 1 characterized areas with permanent water while class 2 areas became tem-
porally similar only after the late 1990s when wet conditions prevailed. During the mid-2000s,
100
80
Mean percent water Cluster
1
60 2
3
4
40 5
6
7
20
0
Jan-73
Jan-76
Jan-79
Jan-82
Jan-85
Jan-88
Jan-91
Jan-94
Jan-97
Jan-00
Jan-03
Jan-06
Jan-09
Jan-12
FIGURE 16.6 The mean percent water for wetlands and lakes within each cluster is shown at each of the 10
Landsat observations listed in Table 16.1. Water area for each wetland and lake was converted to percent water
based on the maximum extent of water present during any of the Landsat observations. The percent water
areas were clustered into seven probable groups by applying an EM algorithm.
class 1 and 2 appear to be impacted by drier conditions. Class 3 had a trajectory similar to
class 2, although with less magnitude and a delayed response to the drier conditions in the mid-
2000s. Class 4, 5, and 6 had more variability, most likely representing wetlands that function as
flow-through or local discharge under normal conditions and temporary-recharge wetlands under
dry conditions. The dynamics presented by the class 7 vector are unique with a dramatic increase
in direction in the 2011 image, suggesting that these areas responded to recent prolonged wet
conditions.
Similar to the Cowardin et al. (1979) classification approach used for NWI (Figure 16.7a), the
example EM classification can be applied spatially to produce maps with multiple classes for an
individual wetland (Figure 16.7b). Large wetlands tended to include more classes than small wet-
lands. The spatial distribution of the classes varied although specific patterns within wetlands were
evident. For example in Figure 16.7b, class 1 and 2 occurred across larger spatial extents and were
typically flanked by higher classes, that is, classes 2, 3, and 4, representing a decrease in water per-
manency. Areas with class 5 and 6 tended to be isolated or located near class 3. Although class 7
was limited in the Figure 16.7b map extent, class 7 wetlands were numerous in the eastern portion
of the region, as isolated wetlands in the drift plain of eastern North Dakota.
Case Study Conclusion

Our wetland classification example requires multitemporal data to account for differences in wet-
land response to climate and hydrological conditions. By spatially mapping the water dynam-
ics, we would expect vegetation communities to be representative of water permanence over the
example study period. Further analysis of upland conditions and trends in water conditions would
enable shifts in wetland conditions to be identified and quantified. Integrating additional data
from high-resolution sensors and multitemporal radar data would be useful for validating wetland
vegetation changes related to the hydroperiod in each basin, as well as the inclusion of ancil-
lary data, such as geology, water chemistry, and land cover. Further investigation and validation
would be necessary to confirm the associated advantages of this mapping approach. However, this
approach appears to have benefits that facilitate efforts to place wetlands along the continuous
hydrologic and climatic gradients discussed by Euliss et al. (2004) in their wetland continuum
conceptual model.
Lacustrine, littoral, aquatic bed, intermittently exposed

Palustrine, aquatic bed, semipermanently flooded
Palustrine, aquatic bed, semipermanently flooded, excavated
Palustrine, emergent/aquatic bed, semipermanently flooded
Palustrine, emergent, semipermanently flooded
Palustrine, emergent, seasonally flooded
Palustrine, emergent, seasonally flooded, party drained/ditched
Palustrine, emergent, seasonally flooded, excavated
Palustrine, emergent, temporarily flooded
Palustrine, emergent, temporarily flooded, party
drained/ditched
0 0.5 1
km
(a)
Cluster
1
2
3
4
5
6
7
0 0.5 1
km
(b)
FIGURE 16.7 (a) NWI data and (b) the results from the EM algorithm applied spatially to the area surround-
ing CLSA, Stutsman County, North Dakota.
LIMITATIONS AND POTENTIAL OF REMOTE SENSING

FOR MAPPING POTHOLE WETLANDS
Initially, remote sensing of wetlands in the PPR was primarily limited by the availability and
quality of data. The earliest satellite imagery had insufficient geometric registration, and the spa-
tial, spectral, and radiometric resolution of MSS data was not sufficient for most wetland map-
ping applications (Carter 1981). These limitations resulted in the inadequate detection of small
wetlands and vegetation communities (Dahl 2006). Other types of data, especially data collected
from airborne sensors, usually represent wetland conditions on one date and have limited spectral
resolution, limiting usage to applications when the physical conditions are conducive for detecting
wetlands. Although NWI was produced from airborne data, processing airborne data for monitor-
ing applications, where temporal frequency is required, is currently cost prohibitive. As the National
Agriculture Imagery Program (NAIP) imagery archive expands and costs for older acquisitions
from high-resolution sensors are reduced or eliminated, the applicability of the data will increase.
Data available from airborne and satellite high-resolution programs currently do not provide
data required for operational wetland monitoring. This results in the use of Landsat data for many
wetland-monitoring studies in the PPR, albeit its moderate resolution limits potential applica-
tions. The primary advantages of Landsat data are temporal resolution and cost. The Landsat
program has continually collected earth observation data via several sensors since 1972 and pro-
vides a temporal resolution of 18 days (Landsat 1, 2, and 3) and 16 days (Landsat 4, 5, 7, and 8).
Sensor technology and data processing have also improved. In the decades since the launch of
Landsat 1, advancements were made in data calibration, registration, and resolution (Markham
and Helder 2012). Recent improvements in Landsat data processing now provide a consistent,
calibrated source for historical observations of the land surface (Wulder et al. 2012). Another
recent change in the Landsat program data policy in 2008 provided users no-cost access to both
new and archived data for mapping long-term land surface processes (Woodcock et al. 2008;
Wulder et al. 2012).
Although the Landsat program provides a source for multitemporal data, the spatial resolu-
tion is not sufficient for detecting small prairie pothole wetlands or applying ground-based clas-
sification systems accurately. For example, assessing and monitoring water quality parameters,
including suspended sediment, chlorophyll, aquatic vascular plants, and temperature with current
technology, is somewhat limited due to currently available spectral and spatial resolutions (Federal
Geographic Data Committee 1992; Ritchie et al. 2003). Mapping efforts tend to require additional
processing or editing to correct misclassified pixels and are dependent on acquisition timing coin-
ciding with optimum wetland detection conditions (Federal Geographic Data Committee 1992).
In contrast, higher-resolution products, including aerial sensors, often lack the spectral resolution
required to produce wetland maps with accurate species-level vegetation classifications (Adam
et al. 2009). Future improvements in spatial and spectral resolution will provide opportunities
to investigate the classification of individual vegetation species, although in the short term, data
fusion with currently available products from both active and passive sensors is exhibiting notable
improvements. Larger-scale automated approaches that provide a consistent product, such as those
described by Verpoorter et al. (2012), also have potential for refining wetland and waterfowl moni-
toring in the PPR.
LiDAR and SAR data are available for many areas in the PPR. Most LiDAR acquisitions are tem-
porally limited, and data are often inconsistent between LIDAR-derived datasets if initial process-
ing requirements varied. LiDAR data are typically high resolution, and more recent SAR sensors,
like Radarsat-2, are capable of acquiring high spatial resolution data as well. Although processing
LiDAR and SAR data requires a slightly different set of skills, incorporating LiDAR or SAR data
with multispectral data is likely to improve mapping accuracies. Currently, the high spatial resolu-
tion of LiDAR data is beneficial for delineating wetland features while moderate-resolution SAR
provides temporal resolution required for wetland and upland change research.
CONCLUSION
Past wetland and water mapping research in the PPR has focused on detecting water and analyz-
ing basic wetland characteristics. Historically, aerial photography was the primary data source
for wetland mapping. More recent remote sensing applications commonly use moderate spatial
resolution passive sensors, such as Landsat, aided with aerial photography, higher-resolution
satellite data, and active sensors. The current availability of multitemporal data, processing
software, and computing capacity supports the development of new approaches for mapping
wetlands in the PPR. Comprehensive methods that account for vegetation, hydroperiod, and
upland conditions across broad scales will improve wetland mapping in the PPR and provide an
opportunity to model scenarios under diverse conditions. Due to the variable nature of weather
in the PPR, future studies must account for dynamic natural processes in order to detect and
quantify changes resulting from climate or land use change. In the near future, the develop-
ment of data products, that is, preprocessed data products, will reduce the technical require-
ments for users and enable incorporation of remotely sensed data by the broader community.
Recent changes in data distribution and no-cost data policies are likely to have a positive influ-
ence on of the future use of remotely sensed data.
ACKNOWLEDGMENTS
This chapter was supported by the U.S. Geological Survey Climate and Land Use Change Research
and Development Program, and the Energy and Minerals Resources Program.
REFERENCES
Adam, E., O. Mutanga, and D. Rugege. 2009. Multispectral and hyperspectral remote sensing for identi-
fication and mapping of wetland vegetation: A review. Wetlands Ecology and Management 18(3):
281–296.
Anteau, M. J. 2012. Do interactions of land use and climate affect productivity of waterbirds and Prairie-
Pothole wetlands? Wetlands 32(1):1–9.
Applied Geomatics Research Group. LiDAR MetaData Repository in Canada. http://agrg.cogs.nscc.ca/
projects/LiDAR_Metadata (accessed November 6, 2014)
Beeri, O. and R. L. Phillips. 2007. Tracking palustrine water seasonal and annual variability in agricultural
wetland landscapes using Landsat from 1997 to 2005. Global Change Biology 13(4):897–912.
Best, R. G. and D. G. Moore. 1979. Landsat interpretation of prairie lakes and wetlands of eastern South Dakota.
In: M. Deutch, D. R. Wiesnet, and A. Rango (eds.), Annual William T. Pecora Memorial Symposium
on Remote Sensing, pp. 499–506. Washington, DC: American Water Resources Association. Technical
Publication No. 5.
Bhang, K. J., F. W. Schwartz, and S. S. Park. 2010. Estimating historic lake stages from one-time snapshot, the
Shuttle Radar Topography Mission of 2000. Hydrological Processes 24(13):1834–1843.
Brewster, W. G., J. M. Gates, and L. D. Flake. 1976. Breeding waterfowl populations and their distribution in
South Dakota. The Journal of Wildlife Management 40(1):50–59.
Brinson, M. M. 1993. A hydrogeomorphic classification of wetlands. U.S. Army Engineers Waterways
Experiment Station, Vicksburg, MS. Wetlands Research Program, Technical Report WRP-DE-4, http://
www.dtic.mil/dtic/tr/fulltext/u2/a270053.pdf. Accessed December 4, 2013
Brisco, B., M. Kapfer, T. Hirose, B. Tedford, and J. Liu. 2011. Evaluation of C-band polarization diversity and
polarimetry for wetland mapping. Canadian Journal of Remote Sensing 37(1):82–92.
Burge, W. G. and W. L. Brown. 1970. A study of waterfowl habitat in North Dakota using remote sensing tech-
niques. Infrared and Optics Laboratory, Willow Run Laboratories, Institute of Science and Technology,
University of Michigan, Ann Arbor, MI, Tech. Rep. 2771-7-F.
Carter, V. 1981. Remote sensing for wetland mapping and inventory. Water International 6(4):177–185.
Cowardin, L. M., V. Carter, F. C. Golet, and E. T. LaRoe. 1979. Classification of Wetlands and Deepwater
Habitats of the United States. Washington, DC: U.S. Fish and Wildlife Service. FWS/OBS-79/31, http://
www.npwrc.usgs.gov/resource/wetlands/classwet/. Accessed December 4, 2013
Cowardin, L. M., D. S. Gilmer, and L. M. Mechlin. 1981. Characteristics of central North Dakota wetlands
determined from sample aerial photographs and ground study. Wildlife Society Bulletin 9(4):280–288.
Crissey, W. F. 1969. Prairie Potholes from a continental viewpoint. Saskatoon wetlands seminar. Prairie
Migratory Bird Research Centre, Saskatoon, Saskatchewan, Regina, Canada: Canadian Wildlife Service.
Report Series 6: pp. 161–171.
Dahl, T. E. 1990. Wetlands Losses in the United States 1780’s to 1980’s. Washington, DC: U.S. Fish and
Wildlife Service. http://www.fws.gov/wetlands/Documents/Wetlands-Losses-in-the-United-States-1780s-
to-1980s.pdf. Accessed December 4, 2013
Dahl, T. E. 2006. Status and Trends of Wetlands in the Conterminous United States 1998 to 2004. Washington,
DC: U.S. Fish and Wildlife Service. http://www.fws.gov/wetlands/Documents/Status-and-Trends-of-
Wetlands-in-the-Conterminous-United-States-1998-to-2004.pdf.
Dechka, J. A., S. E. Franklin, M. D. Watmough, R. P. Bennett, and D. W. Ingstrup. 2002. Classification of wet-
land habitat and vegetation communities using multi-temporal Ikonos imagery in southern Saskatchewan.
Canadian Journal of Remote Sensing 28(5):679–685.
Euliss, Jr., N. H., J. W. Labaugh, L. H. Fredrickson, D. M. Mushet, M. R. K. Laubhan, G. A. Swanson,
T. C. Winter, D. O. Rosenberry, and R. D. Nelson. 2004. The wetland continuum: A conceptual frame-
work for interpreting biological studies. Wetlands 24(2):448–458.
Euliss Jr., N. H., D. A. Wrubleski, and D. M. Mushet. 1999. Wetlands of the Prairie Pothole Region: Invertebrate
species composition, ecology, and management. In: D. P. Batzer, R. B. Rader, and S. A. Wissinger (eds.),
Invertebrates in Freshwater Wetlands of North America: Ecology and Management, pp. 471–514. New
York: John Wiley & Sons, Inc.
Fang, X. and J. W. Pomeroy. 2009. Modelling blowing snow redistribution to prairie wetlands. Hydrological
Processes 23(18):2557–2569.
Federal Geographic Data Committee. 1992. Application of satellite data for mapping and monitoring wetlands.
Wetlands Subcommittee, FGDC, Technical Report 1, Washington, DC.
Freeman, A. and S. L. Durden. 1998. A three-component scattering model for polarimetric SAR data. IEEE
Friedl, M. A. and C. E. Brodley. 1997. Decision tree classification of land cover from remotely sensed data.
Fritzell, E. K. 1989. Mammals in prairie wetlands. In: A. G. van der Valk (ed.), Northern Prairie Wetlands,
pp. 268–301. Ames, IA: Iowa State University Press.
Gilmer, D. S., E. A. Work Jr., J. E. Colwell, and D. L. Rebel. 1980. Enumeration of prairie wetlands with
Landsat and aircraft data. Photogrammetric Engineering and Remote Sensing 46(5):631–634.
Gilmer, D. S., E. A. Work Jr., and A. T. Klett. 1974. Utilization of ERTS-1 for Appraising Changes in
Continental Migratory Bird Habitat. Jamestown, ND: U.S. Fish and Wildlife Service, Northern Prairie
Wildlife Research Center.
Hall, M., E. Frank, G. Holmes, B. Pfahringer, P. Reutemann, and I. H. Witten. 2009. The WEKA data mining
software: An update. SIGKDD Explorations 11(1):10–18.
Henderson, F. M. and A. J. Lewis. 2008. Radar detection of wetland ecosystems: A review. International
Herfindal, I., M. C. Drever, K.-A. Høgda, K. M. Podruzny, T. D. Nudds, V. Grøtan, and B.-E. Sæther. 2012.
Landscape heterogeneity and the effect of environmental conditions on prairie wetlands. Landscape
Ecology 27(10):1435–1450.
Higgins, K. F., D. E. Naugle, and K. J. Forman. 2002. A case study of changing land use practices in the north-
ern Great Plains, U.S.A.: An uncertain future for waterbird conservation. Waterbirds: The International
Journal of Waterbird Biology 25:42–50.
Huang, S., D. Dahal, C. Young, G. Chander, and S. Liu. 2011a. Integration of Palmer Drought Severity Index
and remote sensing data to simulate wetland water surface from 1910 to 2009 in Cottonwood Lake area,
North Dakota. Remote Sensing of the Environment 115(12):3377–3389.
Huang, S., C. Young, M. Feng, K. Heidemann, M. Cushing, D. M. Mushet, and S. Liu. 2011b. Demonstration
of a conceptual model for using LiDAR to improve the estimation of floodwater mitigation potential of
Prairie Pothole Region wetlands. Journal of Hydrology 405(3–4):417–426.
Igl, L. D. and D. H. Johnson. 1998. Wetland birds in the northern Great Plains. In: M. J. Mac, P. A. Opler,
C. E. Puckett Haecker, and P. D. Doran (eds.), Status and Trends of the Nation’s Biological Resources,
pp. 454–455. Reston, VA: U.S. Geological Survey.
Jacobson, J. E., R. A. Ritter, and G. T. Koeln. 1987. Accuracy of Thematic Mapper derived wetlands as based
on National Wetland Inventory data. In: American Society Photogrammetry and Remote Sensing (ed.),
Prospecting New Horizons, ASPRS-ACSM Fall Convention, pp. 109–118. Reno, NV: American Society
Johnson, D. H. and J. W. Grier. 1988. Determinants of breeding distributions of ducks. Wildlife Monograph
100:1–37.
Johnson, R. R. and K. F. Higgens. 1997. Wetland Resources of Eastern South Dakota. Brookings, SD: South
Dakota State University. http://www.npwrc.usgs.gov/resource/wetlands/sdwet/index.htm. Accessed
December 4, 2013
Kantrud, H. A., G. L. Krapu, and G. A. Swanson. 1989a. Prairie Basin Wetlands of the Dakotas: A Community
Profile. Washington, DC: U.S. Fish and Wildlife Service.
Kantrud, H. A., J. B. Millar, and A. G. van der Valk. 1989b. Vegetation of wetlands of the Prairie Pothole
Region. In: A. G. van der Valk (ed.), Northern Prairie Wetlands, pp. 132–187. Ames, IA: Iowa State
University Press.
Kaufman, L. and P. J. Rousseeuw. 2005. Finding Groups in Data: An Introduction to Cluster Analysis, 2nd edn.
Hoboken, NJ: Wiley.
Larson, D. L., N. H. Euliss Jr., M. J. Lannoo, and D. M. Mushet. 1998. Amphibians of northern grasslands. In:
M. J. Mac, P. A. Opler, C. E. Puckett Haecker, and P. D. Doran (eds.), Status and Trends of the Nation’s
Biological Resources, pp. 450–451. Reston, VA: U.S. Geological Survey.
Lawrence, R. L., R. Hurst, T. Weaver, and R. Aspinall. 2006. Mapping Prairie Pothole communities
with multitemporal Ikonos satellite imagery. Photogrammetric Engineering and Remote Sensing
72(2):169–174.
Lawrence, R. L. and A. Wright. 2001. Rule-based classification systems using classification and regression tree
(CART) analysis. Photogrammetric Engineering and Remote Sensing 67(10):1137–1142.
Li, S., R. A. MacMillan, D. A. Lobb, B. G. McConkey, A. Moulin, and W. R. Fraser. 2011. Lidar DEM error
analyses and topographic depression identification in a hummocky landscape in the prairie region of
Canada. Geomorphology 129(3–4):263–275.
Liu, G. and F. W. Schwartz. 2011. An integrated observational and model-based analysis of the hydrologic
response of Prairie Pothole systems to variability in climate. Water Resources Research 47(2):W02504.
Liu, G. and F. W. Schwartz. 2012. Climate-driven variability in lake and wetland distribution across the Prairie
Pothole Region: From modern observations to long-term reconstructions with space-for-time substitution.
Water Resources Research 48(8):W08526.
Markham, B. L. and D. L. Helder. 2012. Forty-year calibrated record of earth-reflected radiance from Landsat:
A review. Remote Sensing of the Environment 122(0):30–40.
Martin, A. C., N. Hotchkiss, F. M. Uhler, and W. S. Bourn. 1953. Classification of wetlands of the United States.
U.S. Fish and Wildlife Service, Washington, DC. Special Scientific Report, Wildlife, No. 20, http://babel.
hathitrust.org/cgi/pt?id=coo.31924055486660;seq=7#view=1up;seq=9 (accessed December 4, 2013).
Millar, J. B. 1976. Wetland classification in Western Canada: A guide to marshes and shallow open water wet-
lands in the grasslands and parklands of the Prairie Provinces. Information Canada, Canadian Wildlife
Service Report Series 37, Ottawa, Ontario, Canada.
Miller, B. A., W. G. Crumpton, and A. G. van der Valk. 2012. Wetland hydrologic class change from prior
to European settlement to present on the Des Moines Lobe, Iowa. Wetlands Ecology and Management
20(1):1–8.
Millett, B., W. C. Johnson, and G. Guntenspergen. 2009. Climate trends of the North American Prairie Pothole
Region 1906–2000. Climatic Change 93(1–2):243–267.
Mita, D., E. DeKeyser, D. Kirby, and G. Easson. 2007. Developing a wetland condition prediction model using
landscape structure variability. Wetlands 27(4):1124–1133.
Mushet, D. M., N. H. Euliss Jr., and C. A. Stockwell. 2012. A conceptual model to facilitate amphibian conser-
vation in the northern Great Plains. Great Plains Research 22:45–58.
National Oceanic and Atmospheric Administration. Palmer Hydrological Drought Index (PHDI). National
Climatic Data Center. http://www.ncdc.noaa.gov/cag/ (accessed January 23, 2014).
National Wetlands Working Group. 1997. The Canadian Wetland Classification System. Wetlands Research
Centre, University of Waterloo, Waterloo, Ontario, Canada, http://www.env.gov.yk.ca/animals-habitat/
documents/canadian_wetland_classification_system.pdf. Accessed December 4, 2013
Nelson, H. K., A. T. Klett, and W. G. Burge. 1970. Monitoring migratory bird habitat by remote sensing
methods. In: Transcript of the North American Wildlife and National Resources Conference, Thirty-
Fifth North American Wildlife Conference, vol. 35, pp. 73–84. Chicago, IL: Wildlife Management
Institute.
Niemuth, N. D., B. Wangler, and R. E. Reynolds. 2010. Spatial and temporal variation in wet area of wetlands
in the Prairie Pothole Region of North Dakota and South Dakota. Wetlands 30(6):1053–1064.
Ozesmi, S. L. and M. E. Bauer. 2002. Satellite remote sensing of wetlands. Wetlands Ecology and Management
10:381–402.
Phillips, R. L., O. Beeri, and E. S. DeKeyser. 2005. Remote wetland assessment for Missouri Coteau prairie
glacial basins. Wetlands 25(2):335–349.
Quinlan, J. R. 1993. C4.5: Programs for Machine Learning. San Mateo, CA: Morgan Kaufmann.
Renslow, M. S., ed. 2012. Manual of Airborne Topographic Lidar. Bethesda, MD: American Society for
Ritchie, J. C., P. V. Zimba, and J. H. Everitt. 2003. Remote sensing techniques to assess water quality.
Photogrammetric Engineering & Remote Sensing 69(6):695–704.
Rover, J., C. K. Wright, N. H. Euliss Jr., D. M. Mushet, and B. K. Wylie. 2011. Classifying the hydrologic func-
tion of Prairie Potholes with remote sensing and GIS. Wetlands 31(2):319–327.
Sethre, P. R., B. C. Rundquist, and P. E. Todhunter. 2005. Remote detection of Prairie Pothole ponds in the
Devils Lake Basin, North Dakota. GIScience and Remote Sensing 42(4):277–296.
Shaw, S. P. and C. G. Fredline. 1956. Wetlands of the United States—Their Extent and their Value to Waterfowl
and Other Wildlife. Washington, DC: U.S. Fish and Wildlife Service. Circular 39, http://www.npwrc.
usgs.gov/resource/wetlands/uswetlan/index.htm. Accessed December 4, 2013
Shook, K. R. and J. W. Pomeroy. 2011. Memory effects of depressional storage in Northern Prairie hydrology.
Hydrological Processes 25(25):3890–3898.
Smith, G. W. 1995. A critical review of the aerial and ground surveys of breeding waterfowl in North America.
U.S. Fish and Wildlife Service, Washington, DC. Biological Science Report 5.
Stewart, R. E. and H. A. Kantrud. 1971. Classification of Natural Ponds and Lakes in the Glaciated Prairie
Region. Washington, DC: U.S. Fish and Wildlife Service, Resource Publication 92, http://www.npwrc.
usgs.gov/resource/wetlands/pondlake/index.htm. Accessed December 4, 2013
Swengel, A. B. and S. R. Swengel. 1998. Tall-grass prairie butterflies and birds. In: M. J. Mac, P. A. Opler,
C. E. Puckett Haecker, and P. D. Doran (eds.), Status and Trends of the Nation’s Biological Resources,
pp. 446–447. Reston, VA: U.S. Geological Survey.
Tiner, R. W. 1996. Wetlands. In: W. Philipson (ed.), Manual of Photographic Interpretation, pp. 475–494.
Bethesda, MD: American Society of Photogrammetry and Remote Sensing.
Tiner, R. W. 1999. Wetland Indicators: A Guide to Wetland Identification, Delineation, Classification, and
Mapping. Boca Raton, FL: Lewis Publishers.
USDA-FSA Aerial Photography Field Office. 2012. Stutsman, North Dakota NAIP 2012 Imagery. Salt Lake
City, UT. http://www.nd.gov/gis/mapsdata/download (accessed January 28, 2014).
US Geological Survey. 2003. Color North America Shaded Relief - 1 Kilometer Resolution. National Atlas of
the United States. http://nationalatlas.gov (accessed January 10, 2014).
US Geological Survey. Landsat Data. Earth Explorer. http://earthexplorer.usgs.gov/ (accessed November 6,
2014).
Verpoorter, C., T. Kutser, and L. Tranvik. 2012. Automated mapping of water bodies using Landsat multispec-
tral data. Limnology and Oceanography: Methods 10:1037–1050.
Walker, J., J. J. Rotella, J. H. Schmidt, C. R. Loesch, R. E. Reynolds, M. S. Lindberg, J. K. Ringelman, and
S. E. Stephens. 2013. Distribution of duck broods relative to habitat characteristics in the Prairie Pothole
Region. The Journal of Wildlife Management 77(2):392–404.
Winter, T. C. 1989. Hydrologic studies of wetlands in the northern prairie. In: A. G. van der Valk (ed.), Northern
Prairie Wetlands, pp. 16–54. Ames, IA: Iowa State University Press.
Winter, T. C. and D. O. Rosenberry. 1998. Hydrology of Prairie Pothole wetlands during drought and deluge:
A 17-year study of the Cottonwood Lake wetland complex in North Dakota in the perspective of longer
term measured and proxy hydrologic records. Climatic Change 40(2):189–209.
Woodcock, C. E., R. Allen, M. Anderson, A. Belward, R. Bindschadler, W. Cohen, F. Gao et al. 2008. Free
access to Landsat imagery. Science 320(5879):1011.
Work Jr., E. A. and D. S. Gilmer. 1976. Utilization of satellite data for inventorying prairie ponds and lakes.
Work Jr., E. A., D. S. Gilmer, and A. T. Klett. 1974. Utility of ERTS for monitoring the breeding habit of migra-
tory waterfowl. In: Third Earth Resources Technology Satellite-1 Symposium, pp. 102–115. Washington,
DC: Goddard Space Flight Center.
Wulder, M. A., J. G. Masek, W. B. Cohen, T. R. Loveland, and C. E. Woodcock. 2012. Opening the archive:
How free data has enabled the science and monitoring promise of Landsat. Remote Sensing of the
Zhang, B., F. W. Schwartz, and G. Liu. 2009a. Systematics in the size structure of Prairie Pothole lakes through
drought and deluge. Water Resources Research 45(4).
Zhang, B., F. W. Schwartz, and D. Tong. 2009b. LANDSAT sub-pixel analysis in mapping impact of climatic
variability on Prairie Pothole changes. Transactions in GIS 13(2):179–195.
17 Mapping the State and
Dynamics of Boreal
Wetlands Using Synthetic
Aperture Radar
Daniel Clewley, Jane Whitcomb,
Mahta Moghaddam, and Kyle McDonald
CONTENTS
Introduction..................................................................................................................................... 370
Datasets and Preprocessing............................................................................................................. 372
Radar Data.................................................................................................................................. 372
Preprocessing........................................................................................................................ 372
Coregistration of Data........................................................................................................... 374
Ancillary Data............................................................................................................................ 374
Training Data.............................................................................................................................. 375
Classification................................................................................................................................... 378
Random Forests.......................................................................................................................... 378
Accuracy Assessment................................................................................................................. 381
Implementation.......................................................................................................................... 381
Separation of Wetland Classes................................................................................................... 383
Wetland Maps................................................................................................................................. 386
1990s’ JERS-1-Derived Map..................................................................................................... 386
2000s’ ALOS PALSAR–Derived Map....................................................................................... 386
Accuracy Assessment................................................................................................................. 386
Wetland Dynamics..................................................................................................................... 390
Discussion....................................................................................................................................... 390
Improvements to Implementation.............................................................................................. 390
Wetland Dynamics..................................................................................................................... 393
Future Work................................................................................................................................ 393
Conclusion...................................................................................................................................... 394
References....................................................................................................................................... 394
369
INTRODUCTION
Northern peatlands are estimated to hold about 30% of the total global pool of soil carbon or 13%
of the total terrestrial carbon in the biosphere (Wieder and Vitt 2006). The warmer, drier conditions
being experienced throughout the Arctic as a consequence of global warming seem to be accelerat-
ing both aerobic and anaerobic decomposition of northern peatland soils, thereby increasing emis-
sions of methane (CH4) and carbon dioxide (CO2) (Bridgham et al. 2013). If continued, this trend
could cause northern peatlands to become major sources of atmospheric carbon. Studies of Siberian
wetlands have indicated that, even in the absence of methane emissions, this effect would present a
major positive warming feedback to the global climate system (Smith et al. 2004), while all existing
models predict large increases in CH4 emissions as CO2 levels continue to rise (Melton et al. 2013).
It thus appears that northern wetlands can act as both sinks and sources of atmospheric greenhouse
gases, with the balance between these two modalities a complex function of climatic and anthro-
pogenic forces (Oechel et al. 1993; Shurpali et al. 1995; Joiner et al. 1999; Bridgham et al. 2006;
Zhuang et al. 2006; Walter et al. 2007).
These considerations make it clear that validated high-resolution maps of northern wetland
extent and distribution are essential for estimating sources, sinks, and net fluxes of atmospheric CO2
and CH4 (Miller and Dinardo 2012). Unfortunately, currently available information on the locations,
types, and extents of northern wetlands is sparse and uncertain. Estimates of the total area of north-
ern wetlands, for example, vary by more than a factor of three (i.e., 2.6–9.0 × 106 km2) (Petrescu
et al. 2010; Melton et al. 2013), exhibiting a degree of uncertainty that has stood as a major impedi-
ment to the formulation of accurate estimates of global wetland fluxes of CO2 and CH4 (Bridgham
et al. 2013; Melton et al. 2013). Further, large temperature increases are predicted to occur at high
latitudes as a consequence of climate change, and as noted in Bridgham et al. (2013), “without a
robust estimate of the current distribution of global wetlands by type, there is little possibility of
accurately portraying how future global change will affect their CH4 emissions.”
Satellite remote sensing is the best approach to mapping the spatiotemporal dynamics of wet-
lands in the large, remote regions of high-latitude North America. Optical sensors, such as Landsat
Thematic Mapper/Enhanced Thematic Mapper (TM/ETM+) and Satellite Pour l’Observation de la
Terre (SPOT) High-Resolution Visible (HRV) instrument, have long been used to detect and moni-
tor wetland types, hydrologic regime, and landscape changes (e.g., Jensen et al. 1986; Haack 1996;
Ramsey and Laine 1997; Augusteijn and Warrender 1998; Wang et al. 1998; Lunetta and Barlogh
1999; Laine and Näslund-Landenmark 2002; Hamilton et al. 2006; Kayastha et al. 2012). However,
optical remote sensing is frequently impaired by cloud cover, such that long time periods are often
required to generate mosaics over large areas. Additionally, as optical sensors primarily interact
with the leaf layer, often information on the structure and underlying ground surface in wetland
systems must be inferred from the canopy layer.
Unlike optical sensors, microwave synthetic aperture radar (SAR) sensors are able to penetrate
vegetation canopies to capture high-resolution information on structural and moisture character-
istics of the vegetation and underlying surface. Because SAR sensors provide their own source
of illumination, they are able to operate day or night regardless of solar illumination conditions.
Additionally, wavelengths typically used by SAR sensors for earth observation (3–70 cm) are able
to penetrate clouds, allowing data to be acquired regardless of cloud cover and thereby facilitating
the generation of mosaics over a large spatial area within a limited time frame. When available,
multiple microwave frequencies can be used to distinguish between wetlands with varying depths
of vegetation cover, since shorter-wavelength (higher-frequency) radars are more sensitive to short
vegetation, while longer-wavelength (lower-frequency) radars are more sensitive to taller vegetation
(Kasischke et al. 1997; Kasischke and Bourgeau-Chavez 1997).
Due to their relatively long wavelength (~23 cm), L-band SAR signals penetrate much further
into the vegetation canopy than do C-band (~5 cm) and other higher-frequency SAR signals. When
trees are present, L-band SAR, particularly horizontally polarized, exhibits a characteristic strong
Mapping the State and Dynamics of Boreal Wetlands Using Synthetic Aperture Radar 371
backscatter response resulting from double-bounce scattering, in which the signal reflects from
tree trunks to the surface and then back to the radar or vice versa (Lucas et al. 2004; Liang et al.
2005). For soils with higher moisture content, an increased response is observed. In flooded forest,
the water produces a smooth, highly reflective surface, such that backscatter due to trunk–ground
interactions is significantly increased. This effect is much more prominent at L-band than it is at
higher frequencies, making L-band SAR remarkably effective at detecting flooded forested (FO)
wetlands (Townsend 2001, 2002; Hess et al. 2003; Rosenqvist et al. 2004a). If large branches are
also present, the L-band SAR backscatter response, particularly when transmitted and received at
different polarizations (cross-polarized; HV or VH), includes a significant component of volume
scattering (Moghaddam and Saatchi 1995; Moghaddam 2001). L-band SAR has been found to be
more sensitive to plant water content than C-band SAR (Ott et al. 1990; Pampaloni et al. 1997;
Schmullius and Evans 1997) and is capable of distinguishing many nonwoody as well as woody
vegetation types (Lee et al. 2001; Costa 2004). Overall, the L-band SAR backscattered signal thus
conveys a wealth of information about forest structure, biomass, and moisture content, as well as the
underlying ground surface, making it well suited for mapping wetlands (Rosenqvist et al. 2007a).
Discussion here has focused on L-band SAR due to the availability of data at this wavelength from
the Japanese Earth Resources Satellite (JERS-1) SAR operational during the 1990s and the suc-
cessor mission, the phased array L-band SAR (PALSAR) carried onboard the Advanced Land
Observing Satellite (ALOS) operational in the 2000s. However, many of the points raised also
apply to longer-wavelength P-band (~70 cm) data, which could be even more effective at elucidating
surface characteristics from under dense vegetation canopies (Moghaddam et al. 2000).
The data collected by SAR sensors are processed into backscattering coefficient (σ0) images that
can then serve as primary data layers to be classified in order to generate high-resolution wetland
maps. A number of classification methods are available: Unsupervised classifications (e.g., k-means,
ISODATA) group data into classes that must be subsequently labeled, while supervised classifica-
tions (e.g., maximum likelihood) utilize a set of training data and then assign each pixel to the
closest training class based on some similarity measure. The definition of wetland classes often
includes other factors, in addition to vegetation type, such as geomorphology and water regime (e.g.,
Cowardin et al. 1979), while some information is available from the SAR data due to variations in
backscattering coefficient due to soil moisture and inundation; ancillary data layers are generally
required to differentiate these. Therefore, rule-based or decision tree–type classifiers are preferable
as they are able to include heterogeneous data types (e.g., continuous and categorical). Decision
trees take an inherently hierarchical approach to classification, with a single layer used to perform
the split at each stage. As an extension to decision trees, ensemble learning algorithms, which
combine the output of multiple decision trees, have demonstrated the ability to perform better than
a single decision tree. Random forests (Breiman 2001) is an example of an ensemble learning algo-
rithm that has demonstrated good performance when applied to remote sensing data (e.g., Gislason
et al. 2006; Waske and Braun, 2009).
The application of random forests for mapping wetlands over large areas from SAR data was
demonstrated for Alaska, using JERS-1 data, by Whitcomb et al. (2009a), with subsequent stud-
ies focusing on applying the technique to PALSAR data (Whitcomb et al. 2009b) and expanding
to include Canadian wetlands (Whitcomb et al. 2009c). While these studies demonstrated that the
technique was capable of providing highly accurate (~90%) maps of wetland type and represented
a significant improvement over existing mapping in the area, the process required to generate the
maps was time consuming, making application to other areas (e.g., Canada) or application to other
data (e.g., PALSAR) difficult, placing limitations on the technique for monitoring wetland dynam-
ics. Only a small fraction of the classification domain (the State of Alaska) could be processed in
any single classification run, which necessitated dividing the area into a number of tiles. Aside from
the extra complexity, time, and effort this entailed, it also limited the wetland class training data to
that contained within the tile. Among other issues, this shortcoming resulted in a number of clas-
sification artifacts at the tile boundaries, which were often very difficult to resolve.
This chapter describes the general approach taken in Whitcomb (2009a–c), with the addition
of numerous enhancements to the implementation to resolve some of the issues identified. The
enhancements include
1. Utilization of readily available and highly developed open-source software (e.g., RSGISLib)
(Bunting et al. 2014) for much of the processing that substantially reduced the time and
effort required for preprocessing
2. Generation of an improved mosaic using updated PALSAR data with enhanced geometric
accuracy
3. Improved registration accuracy for JERS-1 data
4. Greater consistency between JERS-1 and PALSAR processing chains
5. Subsampling of training data, allowing the entire state to be processed together
These improvements are described in detail, and an updated thematic map of wetlands in Alaska
for the late 1990s, generated using JERS-1 data, and an analogous map for the late 2000s, generated
using PALSAR data, are presented. The accuracy of each map is evaluated, and changes between
the two maps are discussed before outlining plans for future work.
DATASETS AND PREPROCESSING

Radar Data
Radar data were available covering two periods, acquired from two L-band SAR instruments.
The first was a dataset from the SAR sensor carried onboard the JERS-1, which was launched
in February 1992 and operated until October 1998. As part of the Global Boreal Forest Mapping
(GBFM) project (Rosenqvist et al. 2004a), two mosaics were generated from JERS-1 SAR data for
North America using summer and winter imagery. The GBFM project was initiated in 1996, fol-
lowing the success of the Global Rain Forest Mapping (GRFM) project (Rosenqvist et al. 2000),
which generated SAR mosaics covering tropical regions for forest mapping. As part of the North
American component of the GBFM project, mosaics were generated covering Alaska and Canada.
Although failure of the onboard tape recorder caused problems acquiring a complete mosaic for
Canada, Alaska was covered by the Alaska Satellite Facility ground station, which allowed near
complete coverage for the winter of 1997 and summer of 1998 to be acquired, with data from other
years used to fill in gaps.
The second mosaic was from the PALSAR instrument carried onboard the ALOS, the succes-
sor to JERS-1. PALSAR also operated at L-band but offered a number of enhancements over its
predecessor including the ability to acquire fully polarimetric data (horizontal and vertical trans-
mit and receive) as well as enhancements to geometric and radiometric accuracy (Shimada et al.
2010). A comparison of the two instruments is shown in Table 17.1, in the operating modes used for
mosaic generation. Following the experience gained from generating large area SAR mosaics for
the GRFM and GBFM projects, a systematic observation strategy was designed and implemented
for ALOS with the aim of maximizing spatial and temporal consistency (Rosenqvist et al. 2004b).
Under the framework of the Kyoto & Carbon Initiative (Rosenqvist et al. 2010), a mosaic was gener-
ated for Alaska, primarily using data from the summer of 2007 with scenes from 2008, 2009, and
2010 used to fill in gaps. The mosaic utilized data acquired in fine beam dual (FBD) mode at HH
and HV polarization.
Preprocessing
JERS-1 data were provided as a mosaic, produced as part of the GBFM program, and calibrated to
normalized radar cross section (NRCS; σ0) using
TABLE 17.1
System Characteristics for JERS-1 SAR and ALOS PALSAR Instruments
JERS-1 SAR ALOS PALSAR
Years operational 1992–1998 2006–2011
Wavelength 23.5 cm 23.6 cm
Polarization HH HH + HV
Ground resolution (m; range × azimuth) 18 × 18 20 × 10
Source: Rosenqvist, A. et al., IEEE Trans. Geosci. Remote Sens., 45(11), 3307–3316, 2007b.
Note: For PALSAR, numerous modes were available, and only the characteristics for FBD
mode are shown, as this was used to produce the mosaic used in this study.
σ0 [dB] = 10 log10 DN 2 − 48.54, (17.1)

( )

provided in Chapman et al. (2002). PALSAR data were provided as strips, with individual scenes
used to fill in gaps if required. Data were supplied georeferenced at a spatial resolution of 1 arc sec-
ond (~30 m). From these data, a mosaic was generated for Alaska. To correct for a drop-off in power
toward the edge of the swath, the value of the pixel with maximum intensity was used in overlapping
areas. A layer providing the acquisition date was associated with each strip, and this was included
in the mosaicking routine to ensure the correct date was retained for every pixel in the final mosaic.
The PALSAR data were calibrated to σ0 (in dB) using
σ0 [dB] = 10 log10 DN 2 − 83.0, (17.2)

( )

provided in Shimada and Ohtaki (2010). A comparison of the JERS-1 summer mosaic and HH
polarization PALSAR mosaic is shown in Figure 17.1.
σ0(dB)
70°N
High: –5 160°W 150°W 140°W 130°W 170°W 160°W 150°W 140°W 130°W
Low: –15
65°N
65°N
65°N
65°N
60°N
60°N
60°N
60°N
0 125 250 160°W 150°W 140°W 160°W 150°W 140°W

km
FIGURE 17.1 Mosaic of JERS-1 (summer) and PALSAR (HH polarization) data for Alaska. Both datasets
are scaled to the same minimum and maximum backscatter. (JERS-1 Mosaic Courtesy of GRFM, © NASDA/
MITI. PALSAR Mosaic © JAXA.)
Coregistration of Data
While the geometric accuracy of PALSAR data is relatively high, along-track geometric
errors are a recognized problem in JERS-1 data (Siqueira et al. 2002), due to large onboard
clock errors that accumulated following the transmitter failure. These geometric errors cause
problems when utilizing JERS-1 data with ancillary data, as is required for the classification,
and also when comparing to PALSAR data to monitor wetland dynamics. Therefore, it was
considered necessary to improve the registration accuracy of the JERS-1 data, with coregister-
ing to the PALSAR data deemed the most appropriate way of achieving this. A simplified ver-
sion of the algorithm proposed in Bunting et al. (2010), available through the Remote Sensing
and GIS Library (RSGISLib) (Bunting et al. 2014), was used to generate tie points. The algo-
rithm maximizes the correlation between a window of pixels around a tie point in the refer-
ence (PALSAR) and floating (JERS-1) image. Tie points were initially generated with a regular
grid of one tie point every 50 pixels. Areas with insufficient features for matching (low stan-
dard deviation in image window) or low correlation between the two images were subsequently
dropped. In total, nearly 17,000 tie points were retained across the state, with such a large num-
ber required to account for nonlinear distortions and provide resilience to outliers. Across all tie
points, the root mean squared difference between JERS-1 and PALSAR data was over 3 pixels
(300 m), although differences in some areas were considerably larger than this. These findings
are consistent with the errors found in Siqueira et al. (2002) of up to 400 m. Using the tie points
generated, a cubic-spline transform was used to determine the correct location for each pixel
and the image was warped using a nearest-neighbor interpolation. Following registration, the
tie-point generation algorithm was rerun to determine the difference of the registered JERS-1
mosaic compared to the PALSAR data, which was found to be less than 0.5 pixels (50 m). The
coregistered data were also compared visually to other available data layers used within the
classification and high-resolution optical data available through Google Earth and were found
to correspond well.
Ancillary Data
In addition to the SAR backscatter coefficient, a number of additional layers were utilized within the
classification to discriminate between different wetland types; these are listed in Table 17.2. Some
of these, such as texture, are derived from SAR data. Although there are a number of measures of
image texture, including those specific to SAR data, the coefficient of variation was used in our
work, which is given as
TABLE 17.2
Data layers Used in the Classification, with Information They Provide
Data layer Purpose Derived From
SAR texture Provides measure of SAR brightness variability SAR data
SAR date of collection Allows adjustment for temporal differences SAR metadata
between swaths
Elevation Accounts for local terrain altitude NED
Slope Delineates wetland areas and accounts for the Elevation
effects of slope in the classification
Open water Masks out areas of water SAR data
Proximity to water Allows adaptation for waterside ecosystem Open water mask
Latitude/longitude Captures effect of geographic location Projected pixel location
σ
Coefficient of var = (17.3)
μ
where
σ is the standard deviation
μ is the mean
This was the measure used in the texture images provided with the GBFM mosaics (Rosenqvist
et al. 2004a) and was also used to generate texture images from the PALSAR data for consistency.
When classifying large multi-image mosaics comprised data acquired at different dates, it is
necessary to consider temporal differences between scenes due to seasonality or changes in envi-
ronmental conditions. By acquiring data from a similar time period, differences in seasonality can
be reduced. However, as SAR data are sensitive to the moisture properties of the vegetation and
surface, changes in these features between acquisition dates (i.e., due to rainfall) can cause large dif-
ferences in backscatter; this can manifest as “striping” between adjacent passes (Lucas et al. 2010).
To account for differences due to seasonality and weather, the acquisition date for each pixel was
included as a data layer within the classification.
Elevation is an important data source within the classification to help distinguish high- and low-
elevation wetland types since high-elevation wetlands often support significantly different mixes
of species than low-elevation wetlands, even within the same geographic region (Wettstein and
Schmid 1999). The National Elevation Dataset (NED) (Gesch et al. 2002; Gesch 2007) Digital
Elevation Model (DEM) was used to provide this information. A seamless mosaic was provided
for Alaska at 2 arc second (~60 m) resolution through the USGS bulk data distribution program.
In our previous work, artifacts in the topographic data that were available for Alaska have caused
problems with SAR processing and the classification. However, with the inclusion of high-resolution
data from airborne interferometric SAR (InSAR), the overall quality of the DEM is improving, and
the version used in this study (release date of August 2013) represents a significant improvement
over the DEM previously available. From the DEM, a slope layer was calculated by considering
the change in elevation over a three-pixel window. The slope calculation was performed prior to
reprojection, with the horizontal spacing (expressed in m) converted from decimal degrees on a
pixel-by-pixel basis using the pixel latitude. The slope layer was used as an input into the classifica-
tion and to differentiate wetland from nonwetland areas. Only areas with a slope less than 3.8° were
considered wetland, with this value chosen as it represented the 75th percentile of all of the slope
pixels extracted from NWI wetland polygons. This value was only a slight increase to the value of
3° used in Whitcomb et al. (2009a), which was empirically chosen as a compromise that captured
most legitimate wetlands while minimizing classification anomalies.
The water mask was generated by applying a rule-based classification to the JERS-1/PALSAR
data and slope layer. Water was assumed to have a slope of less than 3° and a backscatter less than
−12 dB for JERS-1 and −14 dB for PALSAR. From the water mask, a proximity to water layer was
generated providing the distance of each pixel to the nearest pixel of water (expressed in m). As the
distance to water layer is required as part of the classification, it was necessary to classify this layer
separately and prior to the main classification.
Data were supplied in a range of projections and at different resolutions. Prior to classification,
they were all reprojected to Albers conical equal-area projection and resampled to a pixel size of
100 m to match that of the JERS-1 Mosaic, which had the coarsest spatial resolution.
Training Data
The National Wetlands Inventory (NWI) dataset, provided by the U.S. Fish and Wildlife
Service, was used as the primary source of reference data with which to train and evaluate the
classification. Data were supplied in map quadrangles (i.e., four-sided polygons) covering 13%
of Alaska. The data are mainly based on 1:24,000 scale maps derived from color-infrared aerial
photography. Different wetland types are delineated as polygons and attributed following the
classification system of Cowardin et al. (1979). The scheme defines wetlands by plants (hydro-
phytes), soils (hydric soils), and frequency of flooding and/or saturation. No geometric accuracy
statistics are provided for the NWI data, but given that aerial photography can be registered
to ±15 m given reasonably flat terrain (Hudak and Wessman 1998), the accuracy of the data is
expected to be around this figure. Classification of the NWI is quoted at two levels, wetland
identification (correct discrimination of wetlands and nonwetlands) and wetland attribution (cor-
rect identification of wetland type). The accuracy of wetland identification has been found to be
greater than 90%, based on targeted field surveys (Stolt and Baker 1995; Kudray and Gale 2000),
with lower accuracy associated with FO wetlands. In a study considering wetland attribution
accuracy, using field surveys in southern Sierra Nevada, Werner (2004) noted accuracy ranged
from 97% (lacustrine) to 51% (riverine). Changes in the 15-year difference between imagery
used for NWI mapping and the field survey were identified as a possible cause of discrepancy.
The accuracy of the NWI mapping is largely determined by the ability to discriminate wetlands
using photointerpretation. Higher accuracies in wetland delineation were reported where there
were abrupt changes in topography, hydrology, soil, or vegetation at the wetland boundary, with
lower accuracies associated with the edges of FO wetlands, small patches, and narrow strips of
wetland (Tiner 1999).
The Cowardin system, used by the NWI, separates wetlands into the following geomorphologi-
cal types (ecological systems):
Marine—tidal, shores with open access to the ocean (M)

Estuarine—tidal, semiclosed by land with partial access to ocean (E)
Riverine—nontidal, contained within a channel of moving fresh water (R)
Lacustrine—nontidal, situated in topographic depressions, including wetlands bordering
freshwater lakes (L)
Palustrine—nontidal, dominated by vegetation with shallow water depth (P)
Within each geomorphological class, vegetation types include moss–lichen (ML), emergent (EM),
scrub–shrub (SS), and FO. EM is used to describe herbaceous plants protruding from saturated
soil or the water surface. Within Alaska, EM vegetation is typically dominated by cotton grass
(Eriophorum) and other sedges (Carex spp.). SS areas include shrubs such as dwarf alder (Alnus)
and willow (Salix) with FO areas most frequently containing white spruce and/or black spruce
(Picea glauca and Picea mariana).
For each geomorphology–vegetation class, a modifier is used to indicate hydrological changes
occurring on an intra-annual time scale. Given that only a single date was available from the non-
frozen season for JERS-1, and only a single date for PALSAR, accurate discrimination of the dif-
ferent water regimes was not considered possible. Consequently, the modifiers that were used as part
of the classification process were aggregated in the final thematic map. Since the NWI dataset was
obtained as a single geodatabase for the entire state, small polygons were removed and each class
was assigned an integer code, uniquely identifying each geomorphology, vegetation, and hydrologi-
cal regime combination. The complete coverage of NWI data for Alaska is shown in Figure 17.2.
The maps were derived from aerial photography acquired between 1974 and 2007, with the majority
using data from the late 1970s and early 1980s.
Although the primary aim was to classify wetland areas, it was necessary to include some
nonwetland areas in the training data to avoid classifying them as wetland. The National Land
Cover Database (NLCD) (Homer et al. 2007), produced from Landsat 5 TM and Landsat 7 ETM+
data, was used to provide training data for nonwetland classes. While the NLCD is less specific
than the Alaska Geospatial Data Clearinghouse (AGDC) used in Whitcomb et al. (2009a), it is
170°W 160°W 150°W 140°W 130°W
70°N
65°N
Nonwetland
65°N
Barren land
Deciduous forest
Wetland
Herbaceous
Moss-lichen
Palustrine
Emergent
Estuarine
Lacustrine
Riverine
Palustrine
60°N
Woody
60°N
Scrub/shrub
Estuarine
Palustrine
Forested
Estuarine
Palustrine
Nonvegetated
Open water 0 125 250
km 160°W 150°W 140°W
Mapping the State and Dynamics of Boreal Wetlands Using Synthetic Aperture Radar
FIGURE 17.2 Complete ground reference data used for the classification, comprising mainly of NWI data with the NLCD used to fill in nonwetland data. Polygons
show map quadrangles for the NWI data. Subsets show sections near Fairbanks (upper) and Anchorage (lower).
377
synoptic across Alaska and eliminates the need to map many different sets of local class defini-
tions into statewide upland classes. The NLCD is provided at a comparable resolution (30 m) to
the maps we are producing and was generated using data from 2001, which falls in the middle of
the period covered by JERS-1 and PALSAR data. From the NLCD, the barren land class and the
deciduous forest class were chosen as least likely to cause confusion with wetland classes. The
evergreen forest class cannot reliably be used as a nonwetland class due to the high prevalence
of black spruce wetlands in Alaska, the mixed forest class might also contain some wetlands,
and the SS and grassland/herbaceous categories could easily be confused with similar wetland
categories.
The total set of training data covered a large number of pixels (>12 million; Table 17.3). To
reduce the number of pixels used as input for the classification, a stratified random sample was
selected, with a maximum of 100,000 pixels used within each class. In this way, for classes where
a large number of pixels were available (e.g., palustrine emergent [PEM], seasonally flooded), only
a subset was used, whereas for classes with a smaller number of pixels (e.g., estuarine intertidal
EM, saltwater tidal, regularly flooded), all available pixels were used. Setting the maximum num-
ber of pixels per class to 100,000 resulted in only a relatively small number (~8.5%) of the total
pixels being used; this still represented a large amount of samples (>1 million) that were considered
sufficient to capture the variability within the data while being computationally manageable. By
subsampling the training data, the entire state could be classified at once, rather than splitting the
data into tiles and applying a separate classification to each tile, as in Whitcomb et al. (2009a).
Additionally, by using stratified random sampling, a more even distribution of samples across
classes was obtained.
CLASSIFICATION
Random Forests
The production of thematic maps from SAR and ancillary data utilized random forests, an ensem-
ble learning algorithm originally developed for medical diagnostics purposes (Breiman 2001) and
applied to map wetlands from SAR data in Alaska by Whitcomb et al. (2009a,b). The algorithm
is an extension of the classification and regression tree algorithm (Breiman et al. 1993), utilizing
multiple trees (a forest) generated from random samples within a set of training data. Each deci-
sion tree is constructed by randomly selecting N samples, where N is the total number of training
samples. The N pixels are selected with replacement (bootstrap aggregation; bagging) so as not to
alter the characteristics of the pool as selections are made. The probability that a given sample is
never selected during N selections is given as (Breiman et al. 1993)
N
⎛ 1⎞
P(pixel never selected during N selections) = ⎜ 1 − ⎟ , (17.4)
⎝ N⎠
in which the quantity in parentheses (1 − 1/N) is recognized as the probability that a pixel is not
selected in a single selection. Since (Thomas 1972)
N
⎛ 1⎞ 1
lim ⎜ 1 − ⎟ = e −1 ≈ , (17.5)

N →∞ ⎝ N⎠ 3
approximately N/3 samples never get selected. The approximately 2N/3 samples that are selected
are used as training data for the decision tree; these are placed in the tree’s root node. The approxi-
mately N/3 pixels that are not selected are used as validation data for that tree.
TABLE 17.3
Total Number of Pixels within Each Training Class and the Subset Used for Classification
of JERS-1
Total Training
Number Number
Class Description of Pixels Proportion (%) of Pixels Proportion (%) Final Code
Estuarine intertidal EM, saltwater 8,396 0.07 8,396 0.81 E2EM
tidal regularly flooded
Estuarine intertidal EM, 39,765 0.32 39,765 3.85
temporarily/irregularly flooded
Estuarine intertidal SS, saltwater tidal 7 0.00 7 0.00 E2SS
regularly flooded
Estuarine intertidal SS, 3,400 0.03 3,400 0.33
Estuarine intertidal FO, 322 0.00 322 0.03 E2FO
Riverine tidal EM, permanently/ 337 0.00 337 0.03 REM
semipermanently flooded
Riverine lower perennial EM, 9,048 0.07 9,048 0.88 L2EM
saturated
Lacustrine littoral EM, seasonally 53 0.00 53 0.01
flooded
Lacustrine littoral EM, 18 0.00 18 0.00
Palustrine ML, permanently/ 1 0.00 1 0.00 PML
semipermanently flooded
PSS, saltwater tidal regularly flooded 628 0.01 628 0.06
Palustrine ML, saturated 10,760 0.09 10,760 1.04
PEM, permanently/semipermanently 774,524 6.27 100,000 9.68 PEM
flooded
PEM, saltwater tidal regularly 87 0.00 87 0.01
flooded
PEM, seasonally flooded 2,162,011 17.51 100,000 9.68
PEM, temporarily/irregularly flooded 36,007 0.29 36,007 3.49
PEM, saturated 4,659,889 37.74 100,000 9.68
PSS, permanently/semipermanently 84,550 0.68 84,550 8.19 PSS
flooded
PSS, seasonally flooded 569,781 4.61 100,000 9.68
PSS, temporarily/irregularly flooded 184,775 1.50 100,000 9.68
PSS, saturated 2,697,665 21.85 100,000 9.68
PFO, permanently/semipermanently 1,098 0.01 1,098 0.11 PFO
flooded
PFO, seasonally flooded 16,400 0.13 16,400 1.59
PFO, temporarily/irregularly flooded 58,716 0.48 58,716 5.68
PFO, saturated 966,789 7.83 100,000 9.68
Barren land (nonwetland) 63,384 0.51 63,384 6.14 BL
Deciduous forest (nonwetland) 19,772 0.16 19,772 1.91 DF
Total 12,368,188 1,052,754
Note: Classes in gray were not used in the classification as they had insufficient samples. Note for PALSAR, there were a
slightly different number of samples in each class.
Starting at the root node, each decision tree is built in an iterative process in which the algorithm
successively splits nodes of the decision tree based on the input layers until it can no longer split
any more nodes. This occurs when all nodes are either composed of a single class or contain the
minimum number of samples per node. For each node split, random forests first randomly selects
m input data layers upon which to base the split, where m is a user-specified parameter. Due to the
number of data layers (11 for JERS-1, 10 for PALSAR), each data layer necessarily gets used for
many different node splits during the construction of a decision tree. Each split is accomplished in
such a manner as to minimize the heterogeneity of the classes in the child nodes. This is accom-
plished by maximizing the decrease in a measure of heterogeneity called the Gini impurity index
(Breiman et al. 1993):
G (n) = 1 − ∑ p ( c | n ), (17.6)
2
where
G(n) is the Gini index at node n
p(c|n) is the probability of being in class c at node n (i.e., the number of samples in class c at node
n divided by the number of samples at node n)
The algorithm calculates the decrease in the Gini index for all possible splits of the values in each
of the m selected data layers (where there are many possible splits of the values in each layer since
there are a large number of training samples) and then splits the node based on the split yielding the
greatest decrease in Gini index. The sum of all decreases in the Gini index associated with splits on
a given data layer l is stored and used to provide a measure of the importance of data layer l within
the classification, that is, layers that provide the highest decrease in Gini over all decision trees in
the forest are deemed to be more important.
A large number of trees are created, differing from one another in the following ways: (1)
Since the samples used to form each tree are selected randomly from the overall pool of training
samples, a different set of samples is used to form each tree and (2) the input data layers used to
split the nodes in each tree are selected randomly. The choice of the number of data layers (m)
used to split each decision tree’s nodes is important. Larger values of m increase the strength of
individual trees in the forest because each tree is formed based upon a larger subset of the avail-
able input data. However, larger values of m also increase the correlation between trees in the
forest because there is necessarily a higher degree of overlap between the subsets of input layers
used by different trees. This increases the statistical dependence between the classification results
of different trees. The overall accuracy of random forests is an increasing function of the strength
of individual trees in the forest and a decreasing function of the correlation between trees. This
relationship is quantified and substantiated empirically in Breiman (2001). An intermediate value
of m must therefore be chosen to optimize overall accuracy. The random forests documentation
suggests setting m to an integer close to the square root of the total number of input data layers.
For our case, this gives a value of 3 or 4 for m, with m = 4 producing lower errors when applied
to both JERS-1 and PALSAR data. Random forests performs better with a large number of trees.
However, after a certain number of trees have been added, the overall accuracy stops increasing.
Tests were performed with larger numbers of trees and it was determined that after ~200 trees the
out-of-bag error stopped decreasing. Therefore, a forest size of 300 trees was used to classify both
JERS-1 and PALSAR data.
Following the generation of a random forest from the available training data, it is applied to
the data layers to generate a thematic map of wetland classes. For each pixel, a predicted class is
generated for each decision tree by executing the sequence of comparisons of data layer values to
thresholds that constitute that decision tree. Once the pixel has been classified by all trees, the final
class is assigned based on the class with the most number of votes across all decision trees.
Accuracy Assessment
Following the generation of each decision tree in random forests, out-of-bag samples (i.e., those not
used to generate the tree) are run through the decision tree to generate a class for each sample. After
all trees have been generated, each sample is assigned a class based on the maximum number of
votes. As 2N/3 samples are used to build each tree, approximately a third of all trees will be used to
classify every sample. The classes predicted by random forests are compared to the class associated
with each sample to generate a confusion matrix from which overall accuracy and user’s/producer’s
accuracies for each class can be estimated. The accuracy of the classification was evaluated at two
levels; in the first instance, all the wetland classes used within random forests were considered.
As previously mentioned, these classes included modifiers for water regime, which were difficult
to discriminate given the limited time series. Therefore, the classes were aggregated to broader
geomorphology–vegetation classes, and the final classification accuracy was estimated from the
aggregated confusion matrix.
Implementation
The effective utilization of large datasets from different instruments necessitates an efficient pro-
cessing methodology. A number of open-source tools were utilized as part of a processing chain
developed to handle the preprocessing and classification of data. These included the Geospatial
Data Abstraction Library (GDAL; www.gdal.org) for general data handling, RSGISLib (Bunting
et al. 2014) for mosaicking and coregistration, the Raster I/O Simplification library (RIOS)
(Gillingham and Flood 2013) on which a number of custom programs were built (e.g., the slope
calculation), and R (R Core Development Team 2013), which was used for classification and sta-
tistical analysis. Proprietary software packages PCI Geomatica and ESRI ArcMap were also used,
although mainly for visualization and map production. As well as the software used, the format
used for data storage is an integral part of an efficient processing methodology as this can have
a large impact on processing time and general data management. The new KEA image format
(Bunting and Gillingham 2013) was used to store all raster data as it is capable of storing very
large images, supports lossless compression, and is fully compatible with the GDAL specification,
allowing it to be used with existing software built on the GDAL library. Due to the way PALSAR
data had been processed, the strips required to produce the mosaic would have occupied approxi-
mately 3 TB of storage space uncompressed. The ability to store and work with compressed files,
which reduced the storage requirements to about 100 GB, was therefore a distinct advantage.
For classification, the implementation of random forests available through the R package “ran-
domForest” (Liaw and Wiener 2002) was used, with the package rGDAL used to read and write
data. All training pixels were imported into R, and then a stratified random subset was selected
and used as input to random forests. The generated forest and confusion matrix (using out-of-bag
samples) were saved. The saved forest was then applied to the entire stack of data layers except the
ground reference layer, taking a subset of data at a time as input to avoid loading all of the data into
memory at once.
The entire data ingestion and classification process is shown in Figure 17.3. While funda-
mentally the same procedure was used as that in Whitcomb et al. (2009a), the advances in the
functionality available through open-source packages combined with increased computational
capacity have enabled the entire state to be processed as a single entity, rather than splitting into
tiles. Even though splitting into smaller tiles is a common approach to managing large images,
from a classification perspective, this is not ideal as it constrains the classification to artificial, and
often arbitrary, boundaries. Therefore, processing the entire dataset as a single entity or splitting
by more meaningful boundaries (e.g., biogeographic zones, as in Lucas et al. [2011]) is prefer-
able, where possible. After all of the layers have been preprocessed, resampled, and reprojected,
data stacks are created for JERS-1 and PALSAR classifications comprising all available layers.
382
Recode and Ground

NWI
rasterize reference
Resample Nonwetlands
NLCD
classes
JERS1
- summer Coregister Registered
- winter to PALSAR JERS1
Class
Generate Tie confusion
tie points points Random matrix
Data Apply forests
stack masking train
PALSAR mosaic Decision
tree
Texture forest
- summer Coregister Registered
- winter to PALSAR texture
JERS dates Coregister Registered

- summer to PALSAR dates
- winter
Random Thematic
forests map
Classify Water Calculate Proximity test
water mask proximity to water
Calculate Reproject
Elevation slope and resample Slope
Reproject Reprojected
and resample elevation
Calculate Lat/lon
lat/lon
FIGURE 17.3 The preprocessing and classification process used for producing a thematic map from JERS-1 data. The process for PALSAR is similar, although date
and texture images were generated as part of the process and individual strips were mosaicked together. PALSAR data were georeferenced well; therefore, no coregistra-
tion was required for PALSAR.
These data stacks have dimensions 24,600 × 20,000 × 13 (or 12 for PALSAR) equivalent to a 3D
array with over 6 billion elements (or just under for PALSAR). The classification process, that is,
generating the random forest from training data and applying to the image, takes approximately
4 h to run utilizing a single core on a high-end workstation (Intel i7 3.2 GHz with 64 GB RAM).
During tests, the runtime varied depending on the number of trees and maximum number of
samples in each class.
Separation of Wetland Classes

As was discussed earlier, random forests calculates a measure of the importance of each data layer
in the classification. For both JERS-1 and PALSAR, latitude was the most important layer, fol-
lowed by longitude, then elevation. After these, the order was slightly different for the JERS-1 and
PALSAR data, with slope, distance to water, and backscattering coefficient of more importance
than texture and date. The importance of the positional layers—latitude and longitude—is likely
due to the wide spatial distribution of the training samples and the differences due to this, with dif-
ferent geomorphological classes and water regimes occurring at different locations. Since wetland
classes are influenced by hydrology, it was expected that the layers associated with this (elevation,
slope, and proximity to water) would be important in the classification. While these layers provide a
coarse discrimination of broad wetland types, they are unable to distinguish between different veg-
etated wetland types and thus are unable to provide the fine-scale discrimination achieved through
the use of SAR data. Additionally, the geographic position and topography are static layers and
alone are unable to provide any information on wetland dynamics.
One of the key roles of the SAR layers (backscatter and texture) within the classification was the
discrimination of different types of vegetation. The JERS-1 data included both summer and winter
imagery, with the latter capturing different aspects of the landscape than the summer imagery. This
is a consequence of the fact that the radar reflectivities of plants and the ground are direct functions
of their relative dielectric constants that, in turn, vary according to the amount of unbound liquid
water they contain (Mironov et al. 2010). When the landscape freezes, most or all unbound liquid
water is eliminated, such that plant and ground dielectric constants, and so radar reflectivities, are
substantially reduced. The absence of unbound liquid water reduces radar absorptivity as well,
allowing the SAR signal to propagate farther through the vegetation canopy and into the ground
than would occur in a summer scene. Depending on the vegetation type overlying ground, the SAR
backscattered signal reduces by different amounts. The resulting backscatter measurements convey
information about layers lower in the vegetation canopy and/or deeper underground than would
have been the case had the imagery been collected in the summer. The winter imagery, therefore,
adds more information about vegetated wetland and upland regions that help separate them in the
classification process.
Figure 17.4 shows the distribution of backscattering coefficients for summer and winter JERS
imagery using the pixels used to train random forests. There is an increase in backscatter going
from EM to SS and then to FO. Within the estuarine geomorphological class, there is less dif-
ference between the SS and FO classes in the summer data, with a slightly higher median value
observed for SS. The difference is more pronounced in data from the winter, with lower val-
ues observed for SS. The differences in backscatter observed between the different vegetation
types are likely due to differences in structure, which causes a different scattering response. For
SS and FO, the larger branches exhibit a stronger scattering response at HH polarization than that
from EM vegetation.
Differences can also be observed between classes, for example, EM vegetation within the lacus-
trine class exhibits a lower backscatter than in the palustrine class. Some of these differences are
likely due to the presence of other species, as well as variations in soil moisture.
Only a single season of PALSAR data was available for use in the classification. However, the
PALSAR sensor also captured cross-polarized (HV) data, which provides additional information
0
JERS-1 summer (dB)
–5
–10
–15
–20
(a)
0
JERS-1 winter (dB)
–5
–10
–15
–20
Palustrine
forested
Estuarine
emergent
Estuarine
scrub–shrub
Estuarine
Riverine
emergent
Lacustrine
emergent
moss–lichen
Palustrine
emergent
scrub–shrub
Barren land
(nonwetland)
Deciduous
(nonwetland)
forested
forest
Palustrine
Palustrine
(b)
FIGURE 17.4 The distribution of JERS-1 HH backscatter from (a) summer and (b) winter data for the
geomorphology–vegetation classes used in the final, aggregated, classification. The distributions are shown
as standard Turkey box plots, the median is represented by a thick black line, and the box encloses the
lower quartile (25%) and upper quartile (75%). The whiskers represent 1.5 times the interquartile range
from the lower quartile and 1.5 times the interquartile range from the upper quartile. Values outside this
range are shown as points. Two nonwetland classes used as part of the classification have been left in for
completeness.
on vegetation structure. The distribution of PALSAR HH and HV backscattering coefficients is

shown in Figure 17.5. Values at HH polarization are similar to those for JERS-1, although observed
differences are likely due to varied environmental conditions (soil and vegetation moisture), as well
as differences in the characteristics of the two sensors. At HV polarization, FO and SS wetlands
exhibit a stronger scattering response than those from other wetland classes, although at lower levels
than are observed at HH polarization.
When considering data from all dates together, there was substantial variation within each class,
much of which was likely caused by the range of acquisition dates. In Figure 17.6, the PEM class
is shown, separated by acquisition date. There is substantial variation in the mean value across the
range of dates used to build the JERS-1 mosaic, even though they were all for the summer season.
Therefore, it is important to consider the date of acquisition as a data layer within the classification.
In the absence of the date information, the effect could be manifested as discontinuities in classes
between strips.
Even considering the acquisition date, there was a large range of values for each geomorphology–
vegetation combination and considerable overlap between classes, indicating that (1) two channels
of SAR data (dual season or dual polarization) were not sufficient to discriminate between classes
without ancillary information and (2) a single decision tree is unlikely to be accurate for all classes.
Consequently, the use of multiple decision trees (as generated by random forests) is likely to provide
a lower error than would be obtained using a single tree.
0
PALSAR HH (dB)
–10
–20
–30
–40
(a)
0
PALSAR HV (dB)
–10
–20
–30
–40
Palustrine
forested
Estuarine
emergent
Estuarine
scrub–shrub
Estuarine
Riverine
emergent
Lacustrine
emergent
moss-lichen
Palustrine
emergent
scrub-shrub
Barren land
(nonwetland)
Deciduous
(nonwetland)
forested
forest
Palustrine
Palustrine
(b)
FIGURE 17.5 The distribution of PALSAR (a) HH and (b) HV backscatter for the geomorphology–vegetation
classes used in the final, aggregated, classification. Two nonwetland classes used as part of the classification
have been left in for completeness.
0
JERS-1 summer (dB)
−5
−10
−15
171 184 202 205 500 538 541 544 547 550 553 563 568 572 576 591 601 626
Acquisition day (days since January 1, 1997)
FIGURE 17.6 The distribution of JERS-1 backscatter, from the summer mosaic for the PEM class, by acqui-
sition date. Each box plot shows a separate acquisition date, due to only considering summer imagery and the
inclusion of scenes to fill in gaps; the interval between these dates is not uniform. Note: Outliers (i.e., values
smaller than 1.5 times the interquartile range from the lower quartile or greater than 1.5 times the interquartile
range from the upper quartile) are not shown.
WETLAND MAPS
1990s’ JERS-1-Derived Map
The 1998 baseline map of wetlands in Alaska, generated from JERS-1 data, is shown in Figure 17.7.
The PEM and palustrine shrub/scrub (PSS) classes cover much of the state, with PEM occurring
toward the coast and PSS in the interior with patches of palustrine forested (PFO). Patches of estua-
rine emergent (E2EM) are common around the coast with lacustrine EM often around the edges of
lakes, although these are difficult to see at the scale used for Figure 17.7. For areas that overlap with
NWI mapping, the classification corresponds well to the NWI data. The classification follows logi-
cally in areas adjacent to NWI mapping quadrangles (i.e., no sharp boundaries moving away from
the data used for training), as determined through visual comparison with high-resolution optical
data available through Google Earth.
There were some sharp boundaries visible in the classification, such as the abrupt change from
PEM to PSS in the southwest, which did not appear to correspond to a change in vegetation type
when compared to historical optical data available through Google Earth. The consistency of classes
across the mosaic was overall very good. The inclusion of acquisition date for each pixel largely
accounted for differences in backscatter between strips.
2000s’ ALOS PALSAR–Derived Map

The map produced from PALSAR data is presented in Figure 17.8. At the statewide level, it is very
similar to the one produced from JERS-1 data, with PEM dominating coastal areas and PSS and
PFO inland. Closer inspection also revealed lacustrine EM around the edges of lakes and E2EM
along the coast. However, there were some differences identified, due to changes in wetland type
and extent over the period between the two maps.
Accuracy Assessment
The accuracy of the classification was evaluated internally within random forests using the out-of-
bag samples for each decision tree. When training random forests, geomorphology and vegetation
classes were split by water regime modifier to account for expected differences between the classes.
When considering these water regime modifiers, the overall accuracy of the JERS-1-derived clas-
sification was 76.4%. However, it was not expected that an accurate classification down to the water
regime could be achieved given only a single dataset from summer and winter from JERS-1 and
only a summer dataset from PALSAR. Therefore, to reduce uncertainty in the overall classification,
the different water regimes within each geomorphology–vegetation class were aggregated post clas-
sification. These aggregated classes were used to perform the accuracy assessment.
The confusion matrix for the classification derived from JERS-1 data is presented in Table 17.4.
The overall accuracy of the classification was 85.3%. It is worth noting the calculation of overall
accuracy differs from that in Whitcomb et al. (2009a) in that here the figure considers all data across
the state together, whereas their accuracies were estimated for each tile and the overall accuracy was
calculated from this set. This may explain the slightly lower figure than that reported in Whitcomb
et al. (2009a) of 89.6%. Besides overall accuracy, the accuracies for individual classes show the dis-
tribution of errors and are reported as user’s and producer’s accuracies. The user’s accuracy gives
the probability that a pixel classified as A is also A in the reference data. The producer’s accuracy
gives the probability that a pixel that is A in the reference data is correctly classified as A. The high-
est user’s accuracy was for E2EM (93.1%). The accuracy for the PEM class, which covered much
of the coastal areas, was high with a user’s accuracy of 85.6%. PSS, which dominated inland areas,
also had a high producer’s accuracy of 85.9%. The lowest user’s accuracy (33.5%) was for riverine
emergent (REM), which only covered a small proportion of the total area and was sometimes con-
fused with PEM.
70°N
170°W 160°W 150°W 140°W 130°W
65°N
65°N
Nonwetland
Barren land
Deciduous forest
Wetland
Herbaceous
Moss-lichen
Palustrine
Emergent
Estuarine
Lacustrine
Riverine
Palustrine
60°N
60°N
Woody
Scrub/shrub
Estuarine
Palustrine
Forested
Estuarine
Palustrine
Nonvegetated
Open water 0 125 250 160°W 150°W 140°W
km
FIGURE 17.7 Map of wetlands in Alaska (state boundary shown by polygon) generated from JERS-1 data. Subsets show sections near Fairbanks (upper) and Anchorage
(lower). The classification is displayed over a hillshade image derived from the NED DEM showing the topography of nonwetland areas.
387
388
170°W 160°W 150°W 140°W 130°W
70°N
65°N
65°N
Nonwetland
Barren land
Deciduous forest
Wetland
Herbaceous
Moss-lichen
Palustrine
Emergent
Estuarine
Lacustrine
Riverine
Palustrine
60°N
60°N
Woody
Scrub/shrub
Estuarine
Palustrine
Forested
Estuarine
Palustrine
Nonvegetated
Open water 0 125 250 160°W 150°W 140°W
km
FIGURE 17.8 Map of wetlands in Alaska (state boundary shown by polygon), for 2007 generated from PALSAR data. Subsets show sections near Fairbanks (upper)
and Anchorage (lower). The classification is displayed over a hillshade image derived from the NED DEM showing the topography of nonwetland areas.
TABLE 17.4
Aggregated Confusion Matrix for Classification Derived from JERS-1 Data by Geomorphology–Vegetation Class
Testing
User’s
Classification E2EM E2SS E2FO REM L2EM PML PEM PSS PFO BL DF Accuracy (%)
E2EM 44,858 427 27 0 5 3 1,094 276 574 874 23 93.14
E2SS 441 2888 1 0 0 0 37 21 9 3 0 84.94
E2FO 75 2 230 0 0 0 0 2 6 7 0 71.43
REM 0 0 0 113 1 0 202 21 0 0 0 33.53
L2EM 23 0 0 0 5063 0 3,452 377 68 17 48 55.96
PML 7 0 0 0 0 9660 192 937 585 2 5 84.83
PEM 2,784 210 1 27 1384 243 287,762 33,496 6,963 2085 1052 85.64
PSS 895 130 11 8 225 1152 19,544 330,330 22,525 7500 2230 85.90
PFO 570 26 5 0 12 188 2,159 13,530 157,645 1514 565 89.46
BL 1,179 41 1 0 16 10 1,978 6,683 2,067 50756 653 80.08
DF 206 3 0 0 108 30 1,678 4,742 3,018 1404 8583 43.41
Producer’s accuracy (%) 87.89 77.49 83.33 76.35 74.30 85.59 90.46 84.61 81.49 79.11 65.23 85.30
Note: See Table 17.3 for a description of class codes.

389
The unaggregated accuracy for the PALSAR classification map was 76.6%. The confusion matrix
for the PALSAR-derived classification is presented in Table 17.5. The overall accuracy was 85.4%.
Accuracies across different classes were similar to JERS-1. For both the JERS-1- and PALSAR-
derived classification, the greatest confusion was between the same vegetation types in different
geomorphological classes.
Wetland Dynamics
The availability of maps of wetland extent and type from 1998 and 2007 provides the opportunity
to investigate decadal changes over a large spatial area. While the production of change maps using
a pixel-by-pixel comparison, as in Whitcomb et al. (2009b), is possible, it is important to understand
the uncertainty associated with each map and how these uncertainties propagate to determine the
change within acceptable limits of confidence. For the present, the comparison of change between
the two classifications was limited to a qualitative assessment. A visual comparison was performed
between the two maps, with the original JERS-1 and PALSAR data also consulted.
At the statewide level, the two maps produced appear similar, yet closer inspection reveals dif-
ferences. Many of the lakes have reduced in size in the PALAR classification compared to the
JERS-1-derived one, with some disappearing altogether, as shown in Figure 17.9. These changes
were easily visible in the SAR data. A transition from EM vegetation to SS, or SS to FO, was vis-
ible. However, the confusion matrices (presented in Tables 17.4 and 17.5) indicated some confusion
between these classes, which can look similar in the available data (i.e., tall shrub can look similar
to low forest). Therefore, further validation is required to separate areas of real change from confu-
sion in the classification.
DISCUSSION
The availability of L-band SAR data for Alaska from 1998 through the GBFM program and 2007
from the Kyoto & Carbon Initiative has provided a unique opportunity to map northern wetlands
over a large region with relatively high spatial resolution. This can be used to not only document
the extent and location and type of wetlands but also provide the opportunity to investigate decadal
changes in wetlands. One of the key advantages of the approach presented is that the maps are con-
sistent over large spatial areas and also across datasets. The classifications differ only in the SAR
data and SAR-derived layers used. The ability to apply the classification to data acquired from dif-
ferent sensors, a decade apart, with practically no changes to the methodology shows the flexibility
of the technique presented and its suitability for long-term monitoring of wetland dynamics.
Improvements to Implementation
A key outcome of this study was improvements to the implementation of the method of Whitcomb
et al. (2009a) that greatly increase the level of automation and thereby make it possible to readily
incorporate data from multiple years or from even larger datasets (such as Canada) and to freely
explore new processing techniques as they become available. The difference between the process-
ing chains required to produce a classification for a single year compared to one capable of sys-
tematically processing data from multiple years, as is required for monitoring wetland dynamics,
is large. The process can be compared to the difference in manufacturing between producing a
prototype in a lab and producing items in bulk in a factory, with the latter requiring a completely
different set of tools. While Whitcomb et al. (2009a) demonstrated wetlands could be mapped
over large areas by applying random forests to SAR data and ancillary layers, the focus of the
study was the production of a single map rather than an implementation that could be applied
to systematic map production. Our current research, with the aim of applying the technique to
multiple years of data for monitoring wetland dynamics, has focused on the changes required
TABLE 17.5
Aggregated Confusion Matrix for Classification Derived from PALSAR Data by Geomorphology–Vegetation Class
Testing
User’s
Classification E2EM E2SS E2FO REM L2EM PML PEM PSS PFO BL DF Accuracy (%)
E2EM 49,262 398 19 0 9 3 1221 292 595 913 23 93.41
E2SS 372 2550 2 0 0 0 57 17 14 3 0 84.58
E2FO 84 6 225 0 0 0 1 4 12 4 0 66.96
REM 1 0 0 149 0 0 193 24 0 1 0 40.49
L2EM 72 0 0 2 6234 2 3646 413 87 60 78 58.84
PML 11 0 0 0 0 9645 199 1,004 638 8 10 83.76
PEM 3,227 148 6 32 1651 250 290350 31,642 6,609 3,049 1096 85.89
PSS 1,042 142 9 7 257 1086 18230 332,216 23,106 9,106 2428 85.70
PFO 600 23 3 0 13 131 2011 13,573 159,219 1,902 686 89.37
BL 1,151 8 0 0 31 5 2106 7,033 2,187 61,691 684 82.37
DF 171 2 1 0 92 25 1507 4776 3,276 1,397 9467 45.70
Producer’s accuracy (%) 87.98 77.82 84.91 78.42 75.23 86.53 90.87 84.97 81.34 78.96 65.42 85.43
Note: See Table 17.3 for a description of class codes.

391
0 5 10 km
(a) (b)
(c) (d)
FIGURE 17.9 Comparison of (a) JERS-1 and (b) PALSAR data with derived (c) JERS-1 and (d) PALSAR
classifications for 1998 and 2007 showing changes to the extent and, in some cases, the disappearance of lakes
between the two datasets and derived classification.
to make the technique operationally viable. For example, in Whitcomb et al. (2009a), data were
separated into tiles, with each tile being processed separately. This meant every step needed to be
carried out multiple times (once for each tile). Processing the entire state together simplified the
process but necessitated file formats that were capable of handling the large file sizes and software
capable of efficiently processing the data. Where possible, scripts were developed to automate
tasks, making it easier to apply to future datasets, reducing operator error, and providing a record
of the exact processing steps applied. The move to a largely open-source suite of software that
could be built on and customized as needed was an important part of this process. While propri-
etary software can be applied to perform general processing tasks, the ability to access source
code within open-source software allows it to be adapted for specific needs and upgraded over
time as new data and algorithms become available. Extending the manufacturing metaphor, this
is akin to using general-purpose tools to build a prototype, then commissioning custom tools for
the production line. Another advantage of open-source software is that it is not dependent on the
purchase of licenses. Working on a single workstation and in particular within an academic envi-
ronment, the availability of licenses may not be a major issue. However, when scaling the process
up to multiple machines, such as deploying on a high-performance computer (HPC), this cost
becomes a consideration.
Despite the enhancements to the software used for preprocessing, running random forests on
such a large number of training pixels (over 12 million) requires a very large amount of memory and
significant processing time. Initial tests suggested it was likely to take over a day to generate a single
tree (of the 300 used in the final classification) when all pixels were used. While random forests
can generate trees in parallel, even with access to an HPC comprising 300 nodes, each with suf-
ficient RAM to store all of the training data, it would take over a day to generate the classification.
To ensure the optimum parameters are being used within random forests, multiple runs are often
required. Therefore, a way of reducing the number of training pixels that did not require separating
into tiles was required. Taking a stratified random sample provided the best way to reduce the data
volume and also allowed for a more even distribution of samples. With the subsampling approach,
a forest, comprising 300 trees, can be generated in less than an hour and applied to an image in less
than 3 h. This runtime could be greatly reduced by building and applying decision trees in parallel,
which is an improvement planned for future iterations of the processing chain.
Work is still needed to refine the software suite used for the preprocessing and classification
process. But the fundamentally different aim of building a system suitable for generating maps
from multiple years that can be applied over different areas marks a key step toward applying the
technique to monitor wetland dynamics rather than just the production of a single baseline map.
Wetland Dynamics
A quantitative investigation into wetland dynamics was not carried out here, but a visual compari-
son of the two maps identified a number of areas where change was observed over time. The ability
of earth observation data to not only produce maps of wetland extent for a single time period but
also to update as new data become available provides a valuable source of information to moni-
tor trends in wetlands in response to climatic and anthropogenic influences. For monitoring large
areas covered by northern wetlands, an automated way of extracting information and estimating the
associated uncertainty is required. Many of the changes identified only covered a small number of
pixels, and such changes are unlikely to be picked up at coarser resolutions, highlighting the need
for mapping to be carried out at a high resolution.
Future Work
The method described has demonstrated the possibility to generate maps with small errors. For the
vast areas considered, even small errors can translate to large areas that are incorrectly classified.
Therefore, work continues on ways to enhance the classification, building on the technique and
processing chain presented here. The flexibility, repeatability, and speed of processing achieved
through implementation of the major enhancements described herein mean that it is now possible
to perform far more orderly and in-depth sensitivity studies of classification parameters, protocols,
and data sources than has been hitherto possible. For example, it was noted that discrimination
of wetland classes using only a single year of data is difficult and the inclusion of optical data at
a higher temporal resolution to provide an additional source of information is expected to bring
improvements to classification accuracy. Additionally, the inclusion of layers derived from the DEM
through hydrological modeling (e.g., flow direction and flow accumulation) is a proposed improve-
ment to address some of the confusion associated with the same vegetation type in different geo-
morphology classes. The use of a single slope threshold to mask out nonwetlands was based on the
assumption that slopes above a certain threshold (chosen in this study to be 3.8°) would be unable to
retain enough water to act as wetlands (Whitcomb et al. 2009a). This slope threshold may exclude
wetlands formed due to paludification, which can occur on steeper slopes due to peat moss (e.g.,
Sphagnum spp.) “wicking-up” water and advancing uphill (Tiner 1993). The inclusion of contextual
rules (e.g., neighbor to peat moss) may assist in including such areas in the classification and is a
further improvement for future studies.
Given the availability of JERS-1 data (albeit with some gaps) and PALSAR data over Canada,
expansion to Canadian wetlands is a logical progression of the work. While efforts are underway to
produce targeted maps of wetlands, reference data in Canada are not as readily available as the NWI
in the United States. This means alternative approaches must be sought, such as utilizing training
data from bordering U.S. states (Whitcomb et al. 2009c).
Besides expanding the mapping spatially, expanding temporally and continuing mapping using
future datasets is planned. Although the ALOS mission ended in May 2011, its successor (ALOS-2)
successfully launched in May 2014 and caries the next generation of L-band SAR sensor.
CONCLUSION
Northern wetlands are important ecosystems and play a vital role in the global CO2 and CH4 cycles.
Mapping has proven difficult using optical earth observation data and traditional classification tech-
niques. Building from the method proposed in Whitcomb et al. (2009a), a system was built that
is capable of applying the technique to multiple years of data and thereby suitable for monitoring
wetland dynamics. Maps of wetland type in Alaska using L-band SAR data from JERS-1 and
PALSAR for 1998 and 2007, respectively, were generated with an overall accuracy of 85.3% for the
JERS-1-derived map and 85.4% for the PALSAR-derived map. The improved software suite devel-
oped allows large datasets from two different sensors to be processed efficiently and consistently.
It has the capability to produce updated maps from new SAR data, thereby allowing monitoring of
wetland dynamics and responses to climate change.
ACKNOWLEDGMENTS
This work was supported through National Aeronautics and Space Administration (NASA)’s
Making Earth System Data Records for Use in Research Environments (MEaSUREs) Program.
This work has been undertaken in part within the framework of the JAXA Kyoto & Carbon
Initiative. ALOS PALSAR data were provided through the Alaska Satellite Facility. Portions of
this work were undertaken at the Jet Propulsion Laboratory, California Institute of Technology,
under contract to the NASA. The PALSAR data were processed by Bruce Chapman (JPL). Pete
Bunting and Sam Gillingham made software and the KEA file format library available and assisted
in their usage.
REFERENCES
Augusteijn, M.F. and C.E. Warrender. 1998. Wetland classification using optical and radar data and neural
network classification, International Journal of Remote Sensing 19:1545–1560.
Breiman, L. 2001. Random forests, Machine Learning 45:5–32.
Breiman, L., J.H. Friedman, R.A. Olshen, and C.J. Stone. 1993. Classification and Regression Trees, Chapman
and Hall, New York.
Bridgham, S., H. Cadillo-Quiroz, J. Keller, and Q. Zhuang. 2013. Methane emissions from wetlands: Biogeo
chemical, microbial, and modeling perspectives from local to global scales. Global Change Biology 19:
1325–1346.
Bridgham, S.D., P. Megonigal, J.K. Keller, N.B. Bliss, and C. Trettin. 2006. The carbon balance of North
American wetlands. Wetlands 26(4):889–916.
Bunting, P.J., D. Clewley, R.M. Lucas, and S. Gillingham. 2014. The Remote Sensing and GIS software library
(RSGISLib). Computers and Geosciences 60:216–226.
Bunting, P.J. and S. Gillingham. 2013. The KEA image file format. Computers and Geosciences 57:54–58.
Bunting, P.J., F. Labrosse, and R.M. Lucas. 2010. A multi-resolution area-based technique for automatic multi-
modal image registration. Image and Vision Computing 28(8):1203–1219.
Chapman, B., P. Siqueira, and A. Freeman. 2002. The JERS Amazon Multi-season Mapping Study (JAMMS):
Observation strategies and data characteristics. International Journal of Remote Sensing 23(7):1427–1446.
Costa, M.P.F. 2004. Use of SAR satellites for mapping zonation of vegetation communities in the Amazon
floodplain. International Journal of Remote Sensing 25(10):1817–1835.
Habitats of the United States, U.S. Department of the Interior, Fish and Wildlife Service, Washington, DC.
Northern Prairie Wildlife Research Center, Jamestown, ND. http://www.npwrc.usgs.gov/resource/1998/
classwet/classwet.htm (Version December 4, 1998).
Gesch, D.B. 2007. The National Elevation Dataset, in: Digital Elevation Model Technologies and Applications:
The DEM Users Manual, 2nd edn., ed. D. Maune, pp. 99–118., American Society for Photogrammetry
and Remote Sensing, Bethesda, MD.
Gesch, D., M. Oimoen, S. Greenlee, C. Nelson, M. Steuck, and D. Tyler. 2002. The National elevation dataset.
Gillingham, S. and N. Flood. 2013. Raster I/O simplification library. https://bitbucket.org/chchrsc/rios/
(accessed October 22, 2013).
Gislason, P.O., J.A. Benediktsson, and J.R. Sveinsson. 2006. Random forests for land cover classification.
Pattern Recognition Letters 27(4):294–300.
Haack, B. 1996. Environmental auditing: Monitoring wetland changes with remote sensing: An East African
example. Environmental Management 20:411–419.
Hamilton, S., J. Kellndorfer, B. Lehner, and M. Tobler. 2006. Remote sensing of floodplain geomorphology
as a surrogate for habitat diversity in a tropical river system (Madre de Dios, Peru). Geomorphology
89(1–2):23–38.
Hess, L.L., J.M. Melack, E.M.L.M. Novo, C.C.F. Barbosa, and M. Gastil. 2003. Dual-season mapping of
87:404–428.
Homer, C., J. Dewitz, J. Fry et al. 2007. Completion of the 2001 National Land cover database for the conter-
minous United States. Photogrammetric Engineering and Remote Sensing 73(4):337–341.
Hudak, A.T. and C.A. Wessman. 1998. Textural analysis of historical aerial photography to characterize woody
plant encroachment in South African savanna. Remote Sensing of the Environment 66:317–330.
Jensen, J.R., M. Hodgson, E. Christensen, H.E. Mackey, L. Tinney, and R. Sharitz. 1986. Remote sensing
of inland wetlands: A multispectral approach. Photogrammetric Engineering and Remote Sensing
52:87–100.
Joiner, D.W., P.M. Lafleur, J.H. McCaughey, and P.A. Bartlett. 1999. Interannual variability of carbon dioxide
exchanges in a boreal wetland in the BOREAS Northern Study Area. Journal of Geophysical Research
104(22):27663–27672.
Kasischke, E.S. and L. Bourgeau-Chavez. 1997. Monitoring South Florida wetlands using ERS-1 SAR imag-
Kasischke, E.S., J.M. Melack, and M.C. Dobson. 1997. The use of imaging radars for ecological applications:
A review. Remote Sensing of the Environment 59:141–156.
Kayastha, N., V. Thomas, J. Galbraith, and A. Banskota. 2012. Monitoring wetland change using inter-annual
landsat time-series data. Wetlands 32:1149–1162.
Kudray, G.M. and M.R. Gale. 2000. Evaluation of national wetland inventory maps in a heavily forested region
in the upper great lakes. Wetlands 20:581–587.
Laine, B.B. and B. Näslund-Landenmark. 2002. Wetland classification for Swedish CORINE Land Cover
adopting a semi-automatic interactive approach. Canadian Journal of Remote Sensing 28:139–155.
Lee, J.S., M.R. Grunes, and E. Pottier. 2001. Quantitative comparison of classification capability: Fully polari-
metric versus dual- and single-polarization SAR. IEEE Transactions on Geoscience and Remote Sensing
39(11):2343–2351.
Liang, P., M. Moghaddam, L. Pierce, and R.M. Lucas. 2005. Radar backscattering model for multilayer mixed
species forests. IEEE Transactions on Geoscience and Remote Sensing 43(11):2612–2626.
Liaw, A. and M. Wiener. 2002. Classification and regression by Random Forest. R News 2(3):18–22.
Lucas, R.M., J. Armston, R. Fairfax et al. 2010. An evaluation of the ALOS PALSAR L-band backscatter—
Above ground biomass relationship Queensland, Australia: Impacts of surface moisture condition and
vegetation structure. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
3(4):576,593.
Lucas, R.M., K. Medcalf, A. Brown et al. 2011. Updating the Phase 1 habitat map of Wales, UK, using satellite
sensor data. ISPRS Journal of Photogrammetry and Remote Sensing 66(1):81–102.
Lucas, R.M., M. Moghaddam, and N. Cronin. 2004. Microwave scattering from mixed species forests,
Queensland, Australia. IEEE Transactions on Geoscience and Remote Sensing 42(10):2142–2159.
Lunetta, R.S. and M.E. Barlogh. 1999. Application of multi-temporal landsat 5 TM imagery for wetland iden-
tification. Photogrammetric Engineering and Remote Sensing 65:1303–1310.
Melton, J.R., R. Wania, E.L. Hodson et al. 2013. Present state of global wetland extent and wetland meth-
ane modelling: Conclusions from a model inter-comparison project (WETCHIMP). Biogeosciences
10:753–788.
Miller, C.E. and S.J. Dinardo. 2012. CARVE: The carbon in arctic reservoirs vulnerability experiment. IEEE
Aerospace Conference, 2012, 4:1–17.
Mironov, V., R. De Roo, and I. Savin. 2010. Temperature-dependable microwave dielectric model for an Arctic
soil. IEEE Transactions on Geoscience and Remote Sensing 48(6):2544–2556.
Moghaddam, M. 2001. Estimation of comprehensive forest variable sets from multiparameter SAR data over
a large area with diverse species. IEEE Geoscience and Remote Sensing Symposium, IGARSS 2001
4:1660–1662.
Moghaddam, M. and S. Saatchi. 1995. Analysis of scattering mechanisms in SAR imagery over boreal forest:
Results from BOREAS ’93. IEEE Transactions on Geoscience and Remote Sensing 33(5):1290–1296.
Moghaddam, M., S. Saatchi, and R. Cuenca. 2000. Estimating subcanopy soil moisture with radar. Journal of
Geophysical Research 105(11):14899–14911.
Oechel, W.C., S.J. Hastings, G. Vourlitis, M. Jenkins, G. Riechers, and N. Grulke. 1993. Recent change of
Arctic tundra ecosystems from a net carbon dioxide sink to a source. Nature 361:520–523.
Ott, J., E. Kasischke, N.M. French, M.F. Gross, and V. Klemas. 1990. Preliminary evaluation of multichan-
nel SAR data set for a mid-Atlantic coastal marsh. IEEE Geoscience and Remote Sensing Symposium,
IGARSS 1990, pp. 453–456.
Pampaloni, P., G. Macelloni, S. Paloscia, and S. Sigismondi. 1997. The potential of C- and L-band SAR in
assessing vegetation biomass: The ERS-1 and JERS-1 experiments. ESA 3rd ERS Symposium, Florence,
Italy.
Petrescu, A.M.R., L.P.H. van Beek, J. van Huissteden et al. 2010. Modeling regional to global CH4 emissions
of boreal and arctic wetlands. Global Biogeochemical Cycles 24:Gb4009. doi:10.1029/2009gb003610.
R Core Team. 2014. R: A language and environment for statistical computing. R Foundation for Statistical
Computing, Vienna, Austria. http://www.R-project.org/.
Ramsey III, E.W. and S. Laine. 1997. Comparison of Landsat Thematic Mapper and high resolution photogra-
phy to identify change in complex coastal wetlands. Journal of Coastal Research 13:281–292.
Rosenqvist, A., C.M. Finlayson, J. Lowry, and D. Taylor. 2007a. The potential of long-wavelength satellite-
borne radar to support implementation of the Ramsar Wetlands Convention. Aquatic Conservation:
Marine and Freshwater Ecosystems 17:229–244.
Rosenqvist, A., M. Shimada, B. Chapman et al. 2000. The global rain forest mapping project–A review.
Rosenqvist, A., M. Shimada, B. Chapman et al. 2004a. An overview of the JERS-1 SAR Global Boreal Forest
Mapping (GBFM) project. IEEE Geoscience and Remote Sensing Symposium, 2004. IGARSS 2004
2:1033–1036.
Rosenqvist, A., M. Shimada, N. Ito, and M. Watanabe. 2007b. ALOS PALSAR: A pathfinder mission for
global-scale monitoring of the environment. IEEE Transactions on Geoscience and Remote Sensing
45(11):3307–3316.
Rosenqvist, A., M. Shimada, R.M. Lucas, B. Chapman, P. Paillou, L. Hess, and Lowry, J. 2010. The Kyoto &
Carbon initiative—A brief summary. IEEE Journal of Selected Topics in Applied Earth Observations and
Rosenqvist, A., M. Shimada, M. Watanabe, T. Tadono, and K. Yamauchi. 2004b. Implementation of systematic
data observation strategies for ALOS PALSAR, PRISM and AVNIR-2. IEEE Geoscience and Remote
Sensing Symposium, 2004. IGARSS 2004 7:4527–4530.
Schmullius, C. and D. Evans. 1997. Tabular summary of SIR-C/X-SAR results: Synthetic aperture radar fre-
quency and polarization requirements for applications in ecology and hydrology. IEEE Geoscience and
Remote Sensing Symposium, 1997. IGARSS 1997 4:1734–1736.
Shimada, M. and T. Ohtaki. 2010. Generating large-scale high-quality SAR mosaic datasets: Application to
PALSAR data for global monitoring. IEEE Journal of Selected Topics in Applied Earth Observations
Shimada, M., T. Tadono, and A. Rosenqvist. 2010. Advanced land observing satellite (ALOS) and monitoring
global environmental change. Proceedings of the IEEE 98(5):780–799.
Shurpali, N.J., S.B. Verma, J. Kim, and J.T. Arkebauer. 1995. Carbon dioxide exchange in a peatland ecosys-
tem. Journal of Geophysical Research 100:14319–14326.
Siqueira, P., S. Hensley, S. Shaffer et al. 2002. A continental-scale mosaic of the Amazon basin using JERS-1
SAR. IEEE Transactions on Geoscience and Remote Sensing 38(6):2638–2644.
Smith, L.C., G.M. MacDonald, A.A. Velichko et al. 2004. Siberian peatlands a net carbon sink and global
methane source since the early Holocene. Science 303:353–356.
Stolt, M.H. and J.C. Baker. 1995. Evaluation of National Wetlands Inventory Maps to inventory wetlands in the
Southern Blue Ridge of Virginia. Wetlands 15(4):346–353.
Thomas, G.B. 1972. Calculus and Analytic Geometry, Addison Wesley, Reading MA. Section 16-7, Example 6.
Tiner, R.W. 1993. The primary indicators method—A practical approach to wetland recognition and delinea-
tion in the United States. Wetlands 13(1):50–64.
Mapping, Lewis Publishers, Boca Raton, FL.
Townsend, P.A. 2001. Mapping seasonal flooding in forested wetlands using multitemporal Radarsat SAR.
wetlands using C-band SAR. International Journal of Remote Sensing 23(3):443–460.
Walter, K., M.E. Edwards, G. Grosse, S.A. Zimov, and F.S. Chapin. 2007. Thermokarst lakes as a source of
atmospheric CH4 during the last deglaciation. Science 318(5850):633–636.
Wang, J., J. Shang, B. Brisco, and R.J. Brown. 1998. Evaluation of multidate ERS-1 and multispectral Landsat
imagery for wetland detection in southern Ontario. Canadian Journal of Remote Sensing 24:60–68.
Waske, B. and M. Braun. 2009. Classifier ensembles for land cover mapping using multitemporal SAR imag-
ery. ISPRS Journal of Photogrammetry and Remote Sensing 64(5):450–457.
Werner, H.W. 2004. Accuracy assessment of National Wetland Inventory maps as Sequoia and Kings Canyon
National Parks. Park Science 23(1):19–22.
Wettstein, W. and B. Schmid. 1999. Conservation of arthropod diversity in montane wetlands: Effect of altitude,
habitat quality and habitat fragmentation on butterflies and grasshoppers. Journal of Applied Ecology
36(3):363–373.
Whitcomb, J., M. Moghaddam, K.C. McDonald, J. Kellndorfer, and E. Podest. 2009a. Mapping vegetated wet-
lands of Alaska using L-band radar satellite imagery. Canadian Journal of Remote Sensing 35(1):54–72.
Whitcomb, J., M. Moghaddam, K.C. McDonald, and E. Podest. 2009b. Decadal change in northern wet-
lands based on differential analysis of JERS and PALSAR data. IEEE Geoscience and Remote Sensing
Symposium, 2009. IGARSS 2009 3:951–954.
Whitcomb, J., M. Moghaddam, K.C. McDonald, and E. Podest. 2009c. Mapping Canadian wetlands using
L-band radar satellite imagery. IEEE Geoscience and Remote Sensing Symposium, 2009. IGARSS 2009
2:1032–1035.
Wieder, R.K. and D.H. Vitt. Eds. 2006. Boreal Peatland Ecosystems, Ecological Studies, Vol. 188, Springer-
Verlag, New York.
Zhuang, Q., J.M. Melillo, M.C. Sarofim et al. 2006. CO2 and CH4 exchanges between land ecosystems and the
atmosphere in northern high latitudes over the 21st century. Geophysical Research Letters 33(17):1–5.
18 Fusion of Multispectral
Imagery and LiDAR
Digital Terrain Derivatives
for Ecosystem Mapping
and Morphological
Characterization of a
Northern Peatland Complex
Antonio Difebo, Murray Richardson, and Jonathan Price
CONTENTS
Introduction..................................................................................................................................... 399
Study Site........................................................................................................................................ 401
Methods...........................................................................................................................................402
Imagery Collection and Processing............................................................................................402
Field Data Collection............................................................................................................404
Image Classification and Accuracy Assessment....................................................................404
Morphological Characterization of Peatland Classes............................................................405
Results.............................................................................................................................................405
Classification Accuracies...........................................................................................................405
Morphological Analysis........................................................................................................408
Discussion.............................................................................................................................409
Conclusion...................................................................................................................................... 411
References....................................................................................................................................... 411
INTRODUCTION
Peatlands comprise 3% of Earth’s land surface (Harris and Bryant 2009) and 12% of Canada’s
(Tarnocai 2006), with most peatlands situated in remote, hard-to-access locations. In Canada, peat-
lands are frequently found in mineral-rich regions of the country and often present unique manage-
ment challenges for the extractive resource industries. The shallow water table of peatlands often
makes their exploration for mapping and landscape classification purposes very difficult. Remote
sensing enables the synoptic scale, passive and active data collection in peatlands, and is being used
increasingly in Canada for mapping and classifying land cover in remote lowland regions of the
country (e.g., Toyra and Pietroniro 2005; Grenier et al. 2008).
399
Multispectral satellite imagery is, arguably, the most common type of imagery used for peat-
land classification. Spatial patterns of vegetation are a strong indicator of peatland type within
heterogeneous peatland complexes and can be used as a basis for image classification. However,
there is often insufficient spectral information for accurate classification due to overlapping
spectral signatures of different vegetation communities. Moreover, transitional or ecotonal areas
between adjacent peatland classes (e.g., bog to fen) are particularly difficult to discriminate due
to spectral mixing and signature overlap (Belward et al. 1990; Russell et al. 1997; Ozesmi and
Bauer 2002).
The Hudson/James Bay Lowlands (JBL) in northern Ontario, Canada, is the world’s largest
contiguous expanse of peatlands. This mineral-rich region of Ontario is experiencing increased
pressure from extractive resource industries. The discovery of a diamondiferous kimberlite at a
location approximately 90 km west of the community of Attawapiskat resulted in the establishment
of Ontario’s first diamond mine (The De Beers’ Victor Diamond Mine Project). The surrounding
landscape, which is nearly 100% peatland complex, was mapped during initial baseline studies by
a project consultant through airphoto interpretation and ground truthing to produce a digitized map
used for landscape inventory (AMEC 2004). Subsequently, in 2005, a large collaborative research
project on peatland ecohydrology was initiated by researchers at the University of Toronto and
University of Waterloo, Ontario, Canada. The team instrumented a complex assortment of peatland
and nonpeatland landscapes at this site with the goal of monitoring hydrology and biogeochemistry
in anticipation of potentially large-scale changes to surface drainage conditions caused by mine
dewatering activities. To support this research effort, a detailed land cover classification was identi-
fied as an essential tool for understanding the hydrological linkages and ecosystem characteristics
across this heterogeneous landscape.
Although typically flat and devoid of large-scale topographical relief (Mitsch and Gosselink
2000), peatlands typically have characteristic geomorphologies that can be quantified with high-
resolution topographic data to improve ecosystem characterization and land cover classification
(Richardson et al. 2010; Anderson et al. 2010; Millard and Richardson 2013). Specifically, the fusion
of high-resolution (e.g., ≤5 m pixel resolution) topographical data such as that derived from LiDAR,
with standard spectral-based classifications, has been shown to improve the thematic distinction of
peatland types within a large peatland complex (Anderson et al. 2010). Fusion combines two inde-
pendent datasets, such as IKONOS and LiDAR, to derive more information than if they were used
individually (Pohl and Genderen 1998). Anderson et al. (2010) used IKONOS and LiDAR for land
cover classification of a 780 ha raised bog in Cumbria, United Kingdom. Fusion of LiDAR with
IKONOS imagery produced an increase in classification accuracy from 71.8% to 88.0%, respec-
tively. The fusion of LiDAR with multispectral satellite imagery such as Landsat has also been
shown to improve land cover classifications in other landscapes (Hudak et al. 2002; Barlow et al.
2006; Bork and Su 2007), often improving the separation of spectrally similar features like water
and marsh (Lee and Shan 2003). This recent trend of fusion of LiDAR with standard spectral-based
classification has therefore proven useful in providing more accurate and detailed landscape clas-
sifications (Bork and Su 2007).
The overarching goal of this study was to develop a landscape classification system for the JBL
peatland complexes in the vicinity of the De Beers’ Victor Diamond Mine to support ongoing
research and monitoring activities at this location. The specific objectives were to (1) combine
IKONOS imagery with LiDAR digital elevation derivatives to assess whether maximum likelihood
classification (MLC) performance can be improved over the use of multispectral imagery alone; (2)
apply the optimal classification scheme to map the distribution and arrangement of peatlands in the
North Granny Creek (NGC) watershed, an ~30 km2 first-order subwatershed in the Attawapiskat
River basin; and (3) compare characteristic morphologies of mapped peatland classes using LiDAR
digital elevation derivatives to evaluate topographic structure of different peatland landforms typi-
cal of the JBL ecoregion.
Fusion of LiDAR and Multispectral Imagery 401
STUDY SITE
The Victor Mine site is situated in the JBL, 90 km west of Attawapiskat in the Nayshkootayaow
River watershed, a tributary of the Attawapiskat River (Figure 18.1). The area experiences long
winters that typically last from October to late April and short summers. Annual precipitation is
approximately 680–720 mm/year (AMEC 2004). Regional soils consist of thick deposits of marine
clay and clay till that are overlain by peat deposits averaging approximately 2 m in thickness and
are situated upon a locally karstic Silurian limestone aquifer known as the Attawapiskat formation
(AMEC 2004). The groundwater table is at, near, or above the surface in most areas and is associated
with the development of a patterned peatland complex with an array of bogs and fens (Figure 18.2).
Minerotrophic fens (ribbed, riparian, ladder, etc.) are topographically low-lying and typically por-
tray directional seepage and/or convey water (NWWG 1997; Mitch and Gosselink 2000; Quinton
et al. 2003). Ombrotrophic bogs (domed, mound, and flat) are marginally raised in elevation above
the fens, thus receive precipitation as their sole source of water and act as important water storage
and release features (Sjörs 1959; NWWG 1997). Limestone bedrock outcrops (bioherms) exist spo-
radically around the landscape. Bioherms are ancient coral reef deposits that are round to irregular
domed features (treed or untreed) that can rise up to 5 m out of the muskeg (Cowell 1983). Palsas,
which are ice-cored mounds (Seppala 1986) similar in size, height, and sometimes in vegetation
cover to bioherms, also occur sporadically in the landscape. Bogs and fens occupy more than 90%
of the landscape (Tarnocai 1998).
Two bioherms straddle the eastern margin of the NGC subwatershed demarcating the start (south
bioherm) and the end (north bioherm) of a research transect bisecting an array of peatland types.
N North–North Granny Creek Watershed

North–North Granny Creek Atta
wap
iska
t Ri
ver
Research transect
Nayshkootayaow River
South–North Granny Creek

South–North Granny Creek Watershed Victor project
North Granny Creek Watershed
0 0.5 1 2 3 4
km
N
Ontario James
Bay
Attawapiskat Ri
ver
Victor project Quebec
0 125 250 500

km
FIGURE 18.1 Maps illustrating location of the De Beers’ Victor Diamond Mine in the Hudson/JBL of
northern Ontario (bottom) and the NGC watershed (yellow outline) and subwatersheds (red outline) just north-
west of the active mine site (top).
1 2 3 4 5
FIGURE 18.2 Photos depicting the complex mosaic of landscape features in the immediate vicinity of the
De Beers’ Victor Diamond Mine in the Hudson/JBL. Many of these landscape features are typical physio-
graphic features of the Hudson/JBL: (1) open water with floating fen mat, (2) Victor Mine site, (3) bedrock
outcrop islands of the Attawapiskat River, (4) bog and fen complex, (5) riparian transition preceded by treed
open bog, (6) northern ribbed fen with broad flark, (7) large northern ribbed fen with teardrop bogs, and
(8) teardrop bog surrounded by northern ribbed fen.
This transect, shown in Figure 18.1, is where detailed hydrological measurements are being made
as part of another study and where detailed ground truthing was undertaken for this study. The
center point of the transect is intersected by the easternmost edge of a domed bog. This domed
bog is the watershed divide between two upper reaches of NGC: North NGC (NNGC) and South
NGC (SNGC). NNGC and SNGC converge at Granny Creek, a small channel 1–2 m in width,
<1 m deep with an average flow rate of ~20,000 m3/day. Granny Creek meanders southeast (outside
the NGC subwatershed) into the Nayshkootayaow River (~1,000,000 m3/day), which flows into the
Attawapiskat River (~50,000,000 m3/day) and finally into James Bay. The NGC subwatershed is
situated between the Attawapiskat River to the north and the Nayshkootayaow River to the south.
The Victor Mine is located southeast of the NGC subwatershed, with the open pit mine for the proj-
ect located immediately to the south (Figure 18.1).
METHODS
Imagery Collection and Processing
IKONOS GeoEye data were collected over the study area in August 2008 and provided by the
vendor as a single 0.82 m panchromatic band, and four 3.2 m multispectral bands (IR/R/G/B), as
well as a 0.82 m multispectral pan-sharpened true color. Discrete-return airborne LiDAR data were
acquired in July 2007 by Terrapoint Canada Inc. Laser pulse returns were classified into bare-earth
and vegetation classes by the LiDAR contractor and delivered as tiled, xyz ASCII files. A 1 m pixel
resolution DEM was interpolated from the classified bare-earth returns using an inverse distance
weighted (IDW) interpolator with a low weighting exponent (0.5), using a maximum of four neigh-
boring points.
An accuracy assessment was conducted along the research transect using a Topcon HiPER GL
RTK GPS system. The root mean square error was determined to be 4.5 cm (vertical accuracy)
for surveyed versus LiDAR-derived elevations interpolated at 1 m resolution. The LiDAR data
were imported into SAGA GIS and clipped to the NGC watershed. The 1 m resolution DEM was
smoothed three times using a Gaussian filter to remove noise in the high-resolution DEM, which
produces smoother terrain derivatives that are more suitable for landscape classification (MacMillan
et al. 2003). It also minimizes potential issues associated with inaccuracies in the coregistration of
LiDAR and IKONOS images.
A suite of digital elevation derivatives were computed from the smoothed 1 m resolution LiDAR
DEM (Table 18.1). Some typical metrics such as slope, aspect, and curvature were included in this
list since they are very common terrain derivatives used in hydrological and soil mapping research.
The application of these common derivatives to peatland classification was expected to be limited,
however, because they represent highly localized surface conditions at 1 m pixel resolution. Three
additional derivatives (difference from mean elevation [DiME], deviation from mean elevation, and
percentile) were calculated at three different scales to provide important contextual information
related to local surface characteristics (Figure 18.3). Specifically, these derivatives describe the
TABLE 18.1
LiDAR DEM Terrain Derivatives Used in This Study
Derivatives (Abbreviation) Scale (m) Definition SAGA Method
Slope (SLP) 1 Slope measures the rate of change of elevation in Zevenbergen and
the direction of the steepest decent (Wilson and Thorne (1987)
Gallant 2000).
Aspect (ASP) 1 The steepest downslope direction from each cell to Zevenbergen and
its neighbors. Often thought of as slope direction Thorne (1987)
or the compass direction a hill faces.
Curvature (CUR) 1 Defined as a curvature tool that is a second Zevenbergen and
derivative of the surface—for example, the slope Thorne (1987)
of the slope. That is, curvature can be used to
describe the physical characteristics of a drainage
basin.
Difference from mean 1 DiME is the difference between the elevation at the SAGA “Residual
elevation (DiME) center of the window and the mean elevation in Analysis Function”
the window, which is a measure of relative Olaya and Conrad
topographic position of the central point (Wilson (2009)
and Gallant 2000).
Deviation from mean 15, 70, 250 Deviation from the mean is the difference from the SAGA “Residual
elevation (DME) mean divided by the standard deviation, providing Analysis Function”
a measure of the relative topographic position as a Olaya and Conrad
fraction of the local relief and is measured from (2009)
−1 to +1 (Wilson and Gallant 2000).
Percentile (PER) 15, 70, 250 Percentile is the ranking (i.e., nonparametric) of SAGA “Residual
the pixel at the center of the analysis window Analysis Function”
relative to all other pixel values in that window. Olaya and Conrad
It is calculated by counting the number of pixels (2009)
lower than the central pixel and returning this
value as a percentage (Wilson and Gallant 2000).
DiME 15 m scale DiME 70 m scale grid DiME 250 m scale grid
(a) (b) (c)
FIGURE 18.3 DiME LiDAR terrain derivative computed at three scales by varying the radius of the moving
window from 15 (a), 70 (b), and 250 m (c).
topographic position of the focal pixel relative to the pixels in the surrounding neighborhood. Three
different radii were used to define these neighborhoods (15, 70, and 250 m). These different neigh-
borhood sizes were used in order to determine the optimal analysis scale for characterizing peatland
morphologies and for augmenting IKONOS MLC of the Victor peatland complex.
Field Data Collection

A ground survey was conducted in October 2009 along the research transect (Figure 18.1).
Approximately 15 locations were visited, and vegetation communities were characterized with notes
and photographs. These ground observations were used to develop a larger training set through
manual interpretation of the IKONOS imagery, with the assistance of additional field staff who
have extensive knowledge of the land cover along the research transect. A total of approximately
20 polygons were digitized through interpretation of the IKONOS imagery for each map class listed
in Table 18.2. This dataset was randomly divided into two equal sized subsets for training and vali-
dation, respectively.
Image Classification and Accuracy Assessment

Using the training data subset described earlier, supervised MLCs were carried out in ArcGIS using
IKONOS R, G, and B image bands both with and without the IKONOS NIR band as each of the
LiDAR terrain derivatives shown in Table 18.1. Based on visual assessment of boxplots, all input
TABLE 18.2
List of Mapped Peatland Classes and Abbreviations
Landscape Unit Abbreviated Class Code
Mat around pools MAP
Bog-lichen BL
Bog-lichen/conifer BLC
Bog-dense conifer BDC
Fen-dense conifer FDC
Fen-riparian fen/sedges RFS
Fen-poor fen FPF
bands were found to have sufficiently normal distributions in each of the training data classes for
MLC assumptions, and a priori probability weightings were set to zero. For each imagery combi-
nation, the overall cross-validated classification accuracy was computed using the validation data
subset. Water was omitted from the accuracy assessment to prevent upward bias of the accuracy
assessment. LiDAR terrain derivatives that demonstrated an improvement in overall classification
accuracy over the use of the IKONOS bands alone were used in various combinations with the
spectral layers to assess potential for further improvements in classification accuracy resulting from
their inclusion. The final classification with the highest overall accuracy was used to produce a map
of the Victor peatland complex and for subsequent morphological analysis using the multiscale ter-
rain derivatives.
Morphological Characterization of Peatland Classes

Using the final classification scheme, the statistical distributions of different LiDAR terrain deriva-
tives were visually compared across the different peatland classes using boxplots. This was done
for each analysis scale (15, 70, and 250 m radii) and used to qualitatively evaluate and interpret the
morphological characteristics of the different landscape features throughout the Victor peatland
complex.
RESULTS
Classification Accuracies
A summary of cross-validated accuracy assessment results is provided in Table 18.3. The table
shows accuracies for RGB and RGB_IR classifications and for the three top-performing IKONOS–
LiDAR combinations. Confusion matrices for each of these classification trials are provided in
Tables 18.4 through 18.8. The overall accuracy is assessed by the sum of all the diagonals (top left
to bottom right) divided by the total sample size. The overall accuracy for the RGB-only trial was
62.9% (Tables 18.3 and 18.4). Landscape classes—mat around pools (MAP), fen-riparian fen/sedges
(RFS), and fen-poor fen (FPF)—were well separated and least confused among other classes as
revealed by the higher user’s and producer’s accuracies shown in Table 18.4. The remaining classes
of bog-lichen/conifer (BLC), bog-dense conifer (BDC), and fen-dense conifer (FDC) all exhibited
confusion, with user’s and producer’s accuracies lower than 50%. The addition of the IR band
increased the overall accuracy of the classification only marginally to 65.8% (Tables 18.3 and 18.5).
TABLE 18.3
Summary of Cross-Validated Accuracy Assessments for Classifications Based on
IKONOS Imagery Alone (1 and 2, Compared to Classifications Based on
IKONOS–LiDAR Fusion, 3–5)
Minimum Producer’s Minimum User’s
Band Combination Overall Accuracy (%) Accuracy (%) Accuracy (%)
IKONOS RGB 62.9 49.3 40.3
IKONOS RGB_IR 65.8 45.8 34.0
IKONOS RGB_IR + PER250m 71.8 48.6 51.6
IKONOS RGB_IR + DME250m 75.3 56.6 65.5
IKONOS RGB_IR + DiME250m 76.4 56.1 65.9
Note: For classifications including LiDAR derivatives, only the best three classifications are shown (based on
overall cross-validated accuracy).
TABLE 18.4
Confusion Matrix for IKONOS RGB Classification Based on Cross-Validated
Accuracy Assessment
Mat Bog- Bog- Fen Riparian
around Bog- Lichen/ Dense Conifer/ Fen/ Fen-Poor User’s
RGB Pools Lichen Conifer Conifer Sphagnum Sedges Fen Accuracy
Mat around pools 697 1 0 0 0 0 11 98.3%
Bog-lichen 0 525 261 30 15 0 38 60.4%
Bog-lichen/conifer 0 137 366 60 106 164 76 40.3%
Bog-dense conifer 0 15 33 316 280 25 0 47.2%
Fen conifer/ 0 8 3 237 262 65 1 45.5%
sphagnum
Riparian fen/ 0 6 16 16 35 264 6 77.0%
sedges
Fen-poor fen 30 55 64 1 0 1 613 80.3
User’s 95.9 70.3 49.3 47.9 37.5 50.9 82.3 62.9%
TABLE 18.5
Confusion Matrix for IKONOS RGB_IR Classification Based on Cross-Validated
Accuracy Assessment
Mat Bog- Bog- Fen Riparian
around Bog- Lichen/ Dense Conifer/ Fen/ Fen-Poor Producer’s
RGB_IR Pools Lichen Conifer Conifer Sphagnum Sedges Fen Accuracy
Mat around pools 705 0 0 0 0 0 20 97.2%
Bog-lichen 0 484 154 8 7 15 56 66.8%
Bog-lichen/conifer 0 193 494 46 101 180 64 45.8%
Bog-dense conifer 0 1 1 453 306 21 0 57.9%
Fen conifer/ 0 19 19 158 234 79 1 45.9%
sphagnum
Riparian fen/ 0 0 8 6 39 214 5 78.7%
sedges
Fen-poor fen 25 48 69 3 1 2 597 90.1
User’s 96.6% 65.0% 66.3 67.2 34.0% 41.9 80.3% 65.8%
As a result, the user’s accuracy for all landscape units increased, except for the MAP class where
the user’s accuracy decreased by only 1%. The producer’s accuracy for MAP, BLC, and BDC all
increased, while for BL, FDC, RFS, and FPF, there was a decrease in accuracy with the addition
of the IR band.
The landscape units for both supervised classifications with and without the IR band exhibited
similar confusion. This confusion occurred in the same landscape classes for both IR_RGB and the
RGB-only trials, as expressed by the relatively similar user’s and producer’s accuracies. There is,
however, a slight improvement in both the user’s and producer’s accuracy of BLC and BDC for the
IR_RGB classification, which is likely the result of the increased overall accuracy of the IR_RGB
classification. FDC (in both classifications) above all other classes yielded the poorest results with
confusion most among other classes with most confusion found in BL, BLC, and RSF. The MAP
TABLE 18.6
Confusion Matrix for IKONOS RGB_IR + LiDAR PER250m Derivative Classification Based
on Cross-Validated Accuracy Assessment
Mat Bog- Bog- Fen Riparian Producer’s
RGB_IR_ around Bog- Lichen/ Dense Conifer/ Fen/ Fen-Poor Accuracy
PER250m Pools Lichen Conifer Conifer Sphagnum Sedges Fen (%)
Mat around pools 728 2 0 0 1 0 22 97.2
Bog-lichen 0 387 194 8 7 0 31 96.7
Bog-lichen/conifer 0 261 500 48 140 0 80 61.7
Bog-dense conifer 0 1 12 468 135 1 0 48.6
Fen conifer/ 0 38 1 223 458 127 3 75.4
sphagnum
Riparian fen/ 1 3 0 3 5 614 0 53.9
sedges
Fen-poor fen 21 58 43 0 4 8 614 98.1
User’s 97.1% 51.6% 66.7 62.4 61.1% 81.9 81.9% 71.8
TABLE 18.7
Confusion Matrix for IKONOS RGB_IR + DME250m LiDAR Derivative Classification Based
on Cross-Validated Accuracy Assessment
DME250m Pools Lichen Conifer Conifer Sphagnum Sedges Fen (%)
Mat around pools 727 2 0 0 1 0 21 96.8
Bog-lichen 0 491 153 9 13 1 50 68.5
Fen conifer/ 0 26 2 243 534 136 3 56.6
sphagnum
Riparian fen/ 0 2 0 3 7 606 0 98.1
sedges
Fen-poor fen 23 57 44 0 4 7 603 81.7
User’s 96.9% 65.5% 72.5% 59.5% 71.2% 80.8 80.4% 75.3
class exhibited the least amount of confusion compared to all other classes with >96% user’s and
producer’s accuracy for both RGB and IR_RGB classifications. FPF also exhibited a high degree of
separation with >80% in both user’s and producer’s accuracy for both classifications. Overall, the
IR band resulted in only modest improvements in classification accuracy compared to the RGB-only
classification.
Of all the multiscale derivatives assessed, only the broadest scale (250 m radius) was found to
improve accuracy over the IKONOS-only classifications. The DiME LiDAR derivative computed
with a 250 m neighborhood (DiME250m) resulted in the best overall classification accuracy of
76.4%. It was found to be most effective without the addition of any other derivatives. BLC was still
confused with BL, FDC, and FPF, albeit to a much lesser extent then in the RGB and IR_RGB trials.
The addition of any other derivatives beyond the IR_RGB + DiME250m scenario resulted in equal
or reduced classification accuracy without improving separability of any class pairs.
TABLE 18.8
Confusion Matrix for IKONOS RGB_IR + DiME250m LiDAR Derivative Classification
Based on Cross-Validated Accuracy Assessment
DiME250m Pools Lichen Conifer Conifer Sphagnum Sedges Fen (%)
Mat around pools 695 2 0 0 0 0 21 96.8
Bog-lichen 0 487 136 6 13 4 48 70.2
Fen conifer/ 0 14 2 218 486 51 3 62.8
sphagnum
Riparian fen/ 1 1 0 1 12 454 0 96.8
sedges
Fen-poor fen 26 49 45 0 3 2 594 82.6
User’s 96.3% 65.9% 74.7% 59.8% 71.4% 88.8% 80.2% 76.4
Table 18.3 shows that the inclusion of the IR band of the IKONOS and the DiME250m deriva-
tive to the RGB bands of the IKONOS results in an increase from 65.8% (RGB_IR) to 76.4%
(RGB_IR + DiME250m). The map resulting from this trial is provided in Figure 18.4. Thus, a 10.6%
increase in landscape classification accuracy was achieved relative to the multispectral imagery
(IKONOS only) classification, through the inclusion of a single, easily computed LiDAR DEM
derivative. BLC and FDC classes exhibited the majority of the confusion in all classifications,
including the top performing trial (RGB_IR + DiME250m), as seen in Tables 18.4 through 18.8.
Morphological Analysis
The per-class pixel distributions of the DiME LiDAR terrain derivative using boxplots are shown
in Figure 18.5 at each of the three analysis scales tested. At the smallest window size (15 m radius),
1 water class
30 mat around pools
40 bog-lichen
50 bog-lichen/conifer
60 bog-dense conifer
70 fen-dense conifer
80 fen-riparian fen/sedge
90 fen-poor fen km
0 0.5 1 2 3
FIGURE 18.4 Final land cover classification for the NGC watershed near the De Beers’ Victor Diamond
Mine based on the RGB_IR + DiME250m trial with overall cross-validated accuracy of 76.4%.
4
2
0
–2
4 –4
Difference from mean elevation (m)
2
0
–2
4
3
2
1
0
–2 –1
Bog-lichen/conifer
Bog-dense conifer
Fen-dense conifer
Fen-poor fen
Mat around pools
Bog-lichen
Riparian fen/sedges
FIGURE 18.5 Pixel value distributions of DiME terrain derivative computed at 15 (right), 70 (middle), and
250 m (left) moving window radii for each mapped landscape class. The boxplots clearly show that class sepa-
rability is best using the largest window size (250 m). The relative positions of the class-specific distributions
also show that bog classes are generally raised above fen classes, as is expected of this landscape. Note the
relatively small range of relief on the y-axes (~6 m total).
there is substantial overlap among all classes, indicating poor contribution to class separability from
the variable. As window size increases, most notably at the 250 m radius window size, the sepa-
rability among classes improves substantially. The relative topographic positions of the different
peatland forms are also clearly illustrated at this scale. In general, bog landforms are raised above
the fen landforms, with some overlap between classes.
Discussion
The Canadian Wetland Classification System (CWCS) was created to help the science commu-
nity categorize and define the broad range of wetlands that exist across Canada (NWWG 1997).
Theoretically, it is based on hydrogeomorphic characteristics although practically, recognition of
vegetation forms is critical to their identification. GIS automation to partition the landscapes into
those identified within the NWWG is difficult because an optical sensor cannot identify the smaller
scale form and subform of the type of peatland that is included into a landscape classification as
outlined by the NWWG 1997. In the present study, the use of IKONOS multispectral imagery to
classify seven different subforms of the Victor peatland complex resulted in a relatively low overall
classification accuracy of 63.4% with considerable confusion among multiple class pairs (Table
18.4). These results illustrate a fundamental limitation of the use of multispectral imagery alone to
characterize peatland complexes using commonly applied pixel-based classifiers. Specifically, the
pixel-based classification of four-band multispectral imagery is inherently limited to discrimination
of broad vegetation classes with considerable spectral overlap. We did not test the potential to use
other classification methods or multispectral image derivatives such as object-based image analysis
or textural characterization.
Peatlands have very low relief with gradual transitions between peatland forms (Sjörs 1959;
Glaser et al. 2004). Not only is the topographic distinction very subtle, its role on vegetation com-
munity type changes gradually; thus, spectral confusion also occurs also in these areas of transition
(Ozesmi and Bauer 2002). The present study demonstrates that airborne LiDAR DEMs capture
subtle micro- to mesotopographic gradients in wetland environments such as the Hudson/JBL that
can be used to improve land cover classifications. Specifically, the DiME250m derivative provides
important contextual information on relative topographic position of peat landforms (Figure 18.5).
The incorporation of this nonspectral topographic information resulted in an ~10% improvement
in classification accuracy compared to results based on IKONOS R, G , B, NIR image bands alone.
It also resulted in substantial reduction in confusion between several bog versus fen subforms.
In the patterned peatlands, areas of dense conifer in bog and fen are spectrally similar, but their
contrasting topographic position can be used to discriminate them on the basis of morphological
differences.
Overall, these findings corroborate those of others who have also demonstrated successes in
the use of LiDAR for improved characterization of lowland environments. Several of these other
studies (Toyra et al. 2003; Millard and Richardson 2013) also incorporated the use of classified
LiDAR vegetation returns and associated derivatives, and similar approaches would likely fur-
ther improve land cover classification at the Victor peatland complex. These approaches were
not assessed here but would likely provide additional information to improve discrimination of
peatland subforms in the study region. For example, in our final classification, the majority of
error was the result of confusion between BLC and BL. This is not surprising given their strong
spectral similarity as well as the topographical characteristics they share. Both landscape units
are found predominantly at the higher elevations (nearer the dome) in bogs; thus distinguishing
between them proved difficult (Figure 18.5). This is an example of a class pair that might be fur-
ther discriminated through the use of other vegetation surface derivatives derived from aboveg-
round LiDAR returns.
Additional improvements in classification accuracy might also be expected through the use of
more sophisticated image classifiers rather than MLC-based approach used here. For example,
Millard and Richardson (2013) used Random Forest to evaluate the importance of different imag-
ery inputs for classifying land cover in another northern peatland complex and achieved acceptable
results using LiDAR derivatives alone. Others have successfully applied object-based image clas-
sifiers for classifying northern peatland and temperate wetland environments (Grenier et al. 2008;
Barker and King 2012, respectively).
Importantly, because of the multiscale variability in relief within peatland complexes, the analy-
sis scale used to extract terrain derivatives had an important influence on their suitability for land
cover classification. Among the derivatives tested in this study, only the broadest scale of analysis
provided useful contextual information into the classification (250 m moving window radius). This
scale sensitivity is clearly illustrated in Figure 18.5 and is probably reflective of the level of gener-
alization within the classification scheme. It is possible that other scales of relief (e.g., 15 or 70 m)
could help with the discrimination of other land cover types not considered here, such as lawns,
ridges, and other peatland microforms.
CONCLUSION
Overall, this study demonstrates how airborne LiDAR surveys can augment high-resolution optical
satellite imagery such as IKONOS to improve ecosystem classification and mapping in a heteroge-
neous, low-gradient, northern peatland complex. Specifically, a single LiDAR terrain derivative was
found to provide important contextual information about the relative topographic positions of dif-
ferent peatland subforms throughout the study site. This information contributed to a >10% increase
in classification accuracy over the use of IKONOS imagery alone. Use of other LiDAR derivatives,
particularly those based on aboveground vegetation returns and textural derivatives sensitive to sur-
face roughness, would likely provide further improvements to the separability of several spectrally
similar class pairs such as bog-lichen and BLC subforms.
In addition to improved land cover classification, the study results also demonstrate that LiDAR
derivatives can provide quantitative information on the morphological characteristics of different
landforms within the broader peatland complex. The average downslope gradient within the study
site is merely 0.1%. Nevertheless, the information content of a high-resolution LiDAR DEM is suf-
ficiently high to capture subtle geomorphic gradients that provide useful insight into the structure
and function of different ecosystem subtypes in this landscape. The LiDAR derivatives tested here
were highly scale sensitive, however, and it was found that only the broadest scale (250 m mov-
ing window radius) was effective for improving classification accuracy. Overall, the study demon-
strates that fusion of LiDAR derivatives with optical satellite imagery can be a powerful tool to help
hydrologists, ecologists, and environmental managers understand the microscale and macroscale
structural characteristics of northern peatland complexes. These findings are important in light of
the sustained pressure of extractive resource industries that is expected for many of Canada’s vast
northern lowland regions.
REFERENCES
AMEC Earth & Environmental Ltd. 2004. Victor Diamond project: Comprehensive study environmental
assessment.
Anderson, K., J.J. Bennie, E.J. Milton, P.D.M. Hughes, R. Lindsay, and R. Meade. 2010. Combining LiDAR
and IKONOS data for eco-hydrological classification. Journal of Environmental Quality 39:260–273.
Barker, R. and D.J. King. 2012. Blanding’s turtle (Emydoidea blandingii) potential habitat mapping using
aerial orthophotographic imagery and object based classification. Remote Sensing 4(1):194–219.
Barlow, J., S. Franklin, and Y. Martin. 2006. High spatial resolution satellite imagery, DEM derivatives, and
image segmentation for the detection of mass wasting processes. Photogrammetric Engineering and
Belward, A.S., J.C. Taylor, M.J. Stuttard, E. Bignal, J. Mathews, and D. Curtis. 1990. An unsupervised approach
to the classification of semi-natural vegetation from Landsat Thematic Mapper data. A pilot study on
Islay. International Journal of Remote Sensing 11:429–445.
Bork, E.W. and J.G. Su. 2007. Integrating LIDAR data and multispectral imagery for enhanced classification of
rangeland vegetation: A meta analysis. Remote Sensing of the Environment 111:11–24.
Cowell, D.W. 1983. Karst hydrogeology within a subarctic peatland: Attawapiskat River, Hudson Bay Lowland.
Journal of Hydrology 61:169–175.
ESRI (Environmental Systems Resource Institute). 2009. ArcMap 9.2. ESRI, Redlands, CA.
Glaser, P.H., B.C.S. Hansen, D.I. Siegel, A.S. Reeve, and P.J. Morin. 2004. Rates, pathways and drivers for
peatland development in the Hudson Bay Lowlands, northern Ontario, Canada. Journal of Ecology
92:1036–1053.
Grenier, M., S. Labrecque, M. Garneau, and A. Tremblay. 2008. Object-based classification of a SPOT-4 image
for mapping wetlands in the context of greenhouse gases emissions: The case of the Eastmain region,
Québec, Canada. Canadian Journal of Remote Sensing 34(S2):S398–S413.
Harris, A. and R.G. Bryant. 2009. A multi-scale remote sensing approach for monitoring northern peat-
land hydrology: Present possibilities and future challenges. Journal of Environmental Management
90:2178–2188.
Hudak, A.T., M.A. Lefsky, W.B. Cohen, and M. Berterretche. 2002. Integration of LiDAR and Landsat ETM +
data for estimating and mapping forest canopy height. Remote Sensing of the Environment 82:397–416.
Lee, D.S. and J. Shan. 2003. Combining Lidar elevation data and IKONOS multispectral imagery for coastal
classification mapping. Marine Geodesy 26:117–127.
MacMillan, R., T. Martin, T. Earle, and D. McNabb. 2003. Automated analysis and classification of land-
forms using high-resolution digital elevation data: Applications and issues. Canadian Journal of Remote
Sensing 29(5):592–606.
Millard, K. and M. Richardson. 2013. Wetland mapping with LiDAR derivatives, SAR polarimetric decompo-
sitions, and LiDAR–SAR fusion using a random forest classifier. Canadian Journal of Remote Sensing
39(04):290–307.
Mitsch, W.J. and J.G. Gosselink. 2000. The value of wetlands: Importance of scale and landscape. Ecological
Economics 35:25–33.
NWWG (National Wetlands Working Group). 1997. The Canadian Wetland Classification System. Environment
(2nd edn.). The National Wetlands Working Group, Ottawa, Ontario, Canada, 68pp.
Olaya, V. and O. Conrad. 2009. Geomorphometry in SAGA. In: Hengl, T., Reuter, H.I. (Eds.), Geomorphometry:
Concepts, Software, Applications. Elsevier, Amsterdam, the Netherlands, pp. 293–308.
10:381–402.
Pohl, C. and J.L. Van Genderen. 1998. Review article, multisensor image fusion in remote sensing: Concepts,
methods and applications. International Journal of Remote Sensing 19:823–854.
Price, J.C. 1994. How unique are spectral signatures? Remote Sensing of the Environment 49:181–186.
Quinton, W.L., M. Hayashi, A. Pietroniro, and F. Simpson. 2003. Connectivity and storage functions of channel
fens and flat bogs in northern basins. Hydrological Processes 17:3665–3684.
Richardson, M.C., C.P.J. Mitchell, B.A. Branfireun, and R.K. Kolka. 2010. Analysis of airborne LiDAR sur-
veys to quantify the characteristic morphologies of northern forested wetlands. Journal of Geophysical
Russell, G.D., C.P. Hawkins, and M.P. O’Neil. 1997. The role of GIS in selecting sites for riparian restoration
based on hydrology and land use. Restoration Ecology 5:56–68.
Scott, D.A. and T.A. Jones. 1995. Classification and inventory of wetlands: A global overview. Vegetatio
118:3–16.
Seppala, M. 1986. The origins of Palsas. Geografiska Annaler 64:141–147.
Sjörs, H. 1959. Bogs and fens in the Hudson Bay Lowlands. Arctic 12:2–9.
Tarnocai, C. 1998. The amount of organic carbon in various soil orders and ecological provinces in Canada, in
Soil Processes and the Carbon Cycle. Global and Planetary Change 53:222–232.
Tarnocai, C. 2006. The effect of climate change on carbon in Canadian peatlands. Global and Planetary
Change 53(4):222–232.
Thomas, V., P. Trietz, D. Jelinski, J. Miller, P. Lafleur, and H.J. McCaughey. 2003. Image classification of a
northern peatland complex using spectral and plant community data. Remote Sensing of the Environment
84:83–99.
Toyra, J. and A. Pietroniro. 2005. Towards operational monitoring of a northern wetland using geomatics-based
techniques. Remote Sensing of Environment 97:174–191.
Töyrä, J. et al. 2003. Assessment of airborne scanning laser altimetry (lidar) in a deltaic wetland environment.
Canadian Journal of Remote Sensing 29(6):718–728.
Wilson, J.P. and J.C. Gallant. 2000. Terrain Analysis: Principles and Applications. John Wiley & Sons,
New York, 479pp.
Zevenbergen, L.W. and C.R. Thorne. 1987. Quantitative analysis of land surface topography. Earth Surface
Processes and Landforms 12:47–56.
19 Airborne LiDAR-Based Wetland
and Permafrost-Feature
Mapping on an Arctic Coastal
Plain, North Slope, Alaska
Jeffrey G. Paine, John R. Andrews,
Kutalmis Saylam, and Thomas A. Tremblay
CONTENTS
Introduction..................................................................................................................................... 413
Methods........................................................................................................................................... 417
Comparison of LiDAR- and GPS-Derived Elevations.................................................................... 417
Mapping Permafrost Features: Polygons, Pingos, and Thermokarst Lakes................................... 420
Patterned Ground: Polygons and Interpolygon Ice Wedges....................................................... 422
Pingos and Frost Mounds........................................................................................................... 426
Thermokarst Lake and Stream Bathymetry............................................................................... 427
Lake Volume............................................................................................................................... 431
Conclusion...................................................................................................................................... 432
References....................................................................................................................................... 433
INTRODUCTION
We conducted a pilot study in the Prudhoe Bay area on the coastal plain of the Alaskan North
Slope in August 2012 using an airborne light detection and ranging (LiDAR) and imaging system
to rapidly map topographic and bathymetric features relevant to wetland distribution in this Arctic
tundra and permafrost-dominated landscape. The airborne system combines a high-pulse-rate,
near-infrared laser that is used to map topography and a green laser that penetrates water of reason-
able clarity to map bathymetry to depths beyond 10 m under ideal conditions. Using data acquired
during the 2012 survey, we are exploring how best to apply this multisensor instrument to environ-
mental mapping in permafrost-dominated terrain north of the Arctic Circle.
The study area is located on the western side of the Sagavanirktok River on the lower coastal
plain south of Prudhoe Bay along the Beaufort Sea (Figure 19.1). The broad, low-relief coastal plain
(Leffingwell 1919; Carson and Hussey 1962) is largely underlain by Quaternary gravels and silts
3–50 m thick deposited in laterally migrating ancestral streams and overlain by organic-rich silt to
fine sand that accumulated in lacustrine or lagoonal environments (Everett and Parkinson 1977).
Silts, sands, and gravels were brought to the area by rivers carrying sediment eroded from the Arctic
foothills and the northern piedmont of the Brooks Range to the south and redistributed by marine
processes during Quaternary marine transgressions and regressions, particularly at lower elevations
(Hinkel et al. 2007). The landscape has been modified by recent development of thaw-lake basins
413
Deadhorse
Sho
70.2° N, 148.5° W reli
ne
Prudhoe Bay
Beaufort Sea Beaufort Sea
2012
Airborne
survey
Nor
th Sl
ope co
Sagav
anirk astal plain
Arctic foothills
tok
R.
400
0 30 60 km
Elev. (m)
N 0 30 mi
<0
(a) (b)
FIGURE 19.1 (a) Landsat 5 TM image of the Prudhoe Bay and Deadhorse area, Alaskan North Slope, along
the Beaufort Sea showing the 2012 airborne LiDAR and imagery survey area. (b) DEM (30 m cell size) of
the Deadhorse area constructed from TM data. Landsat 5 image and topographic data acquired in June 2009.
(Landsat data from the USGS.)
and lateral stream migration (Carson and Hussey 1962). Eolian deposits are common on the western
sides of the rivers; along the Sagavanirktok River, silt dunes occur in a belt several hundred meters
wide (Leffingwell 1919). Distinctive features include braided stream channels and deltas of rivers
such as the Sagavanirktok that discharge into the Beaufort Sea (Figures 19.1 and 19.2), polygonal
patterned ground associated with ice-wedge processes, hundreds of aligned and elongated lake
basins (drained or undrained) associated with thermokarst processes, and pingos and smaller frost
mounds mostly found within shallow basins. Arctic and subarctic vegetation includes a mixture of
short grasses, sedges, shrubs, mosses, and lichens, with low willows on the banks of larger streams
(Black and Barksdale 1949). Airborne topographic and bathymetric LiDAR can provide data that
aid the study and morphometric characterization of these features and can provide precise baseline
information that will allow future change to be monitored in this area that has undergone rapid
change since the end of the last glaciation.
Estimates of the depth to the base of permanently frozen ground on the North Slope coastal plain
reached depths of 180–350 m based on temperatures measured in an 11 m deep shaft near Point
Barrow (Leffingwell 1919), from greater than 180 borehole measurements 100 km east of Point
Barrow (Cabot 1947), and about 600 borehole geophysical measurements in the study area south
and west of Prudhoe Bay (Osterkamp et al. 1985). In the Prudhoe Bay area, permafrost thickness
decreases from 600 m near the coast to about 400 m near the southern boundary of the LiDAR
survey area (Osterkamp and Payne 1981; Jorgenson et al. 2008b). There is a thin seasonal thaw
layer (active layer) at the surface that is generally less than 1 m thick (Leffingwell 1919; Carson and
Hussey 1962) and has a mean depth of 0.43 m by mid-August (Everett and Parkinson 1977).
The wetland system in the airborne survey area, mapped on recent imagery at a scale of
1:10,000 based on 2009–2010 SPOT satellite imagery supplemented by 2012 airborne survey data
is dominated by the mixed unit of emergent/scrub–shrub covering 15,547 ha (Figure 19.3) (Paine
et al. 2013b). Scrub–shrub species consist of dwarf to low shrubs, primarily willows (Salix spp.).
Airborne LiDAR-Based Wetland and Permafrost-Feature Mapping 415
2013
Seismic
survey
Alcor #1 Alcor #1
Polygon block
Merak #1 Merak #1
2012
R.
Seismic
k
Sagavanirkto
survey
Test well
303
Elev.
(m)
18
0 10 km
2012 LiDAR survey 2012 LiDAR survey
N 0 5 mi
FIGURE 19.2 Comparison of (left) the USGS topographic map (1:63,360 scale) and (right) DEM constructed
from nearly 13 billion topographic LiDAR data points over the North Slope survey area (Figure 19.1). Point
density at full resolution is about 15–20 points/m2. Dark areas on the DEM are lake and river surfaces with
no near-infrared laser return. Also shown are Alcor #1 and Merak #1 stratigraphic test well locations, 2012
and 2013 seismic survey boundaries, and the location of a block with abundant ice-wedge polygons shown in
detail in Figures 19.8 and 19.9.
Emergents are composed of perennial herbs and forbs such as Langsdorf’s lousewort (Pedicularis
langsdorfii). The next most common habitat with a total area of 8164 ha is freshwater emer-
gent. Lake area, including ecologically related palustrine (<8 ha) and lacustrine (>8 ha) habi-
tats (Cowardin et al. 1979), encompasses 2304 ha. The mixed unit scrub–shrub/emergent covers
1405 ha. In this area of the coastal plain, scrub–shrub assemblages are assigned the least wet water
regime and cover 286 ha. Most wetland habitats, other than scrub–shrub, are assigned a midlevel
moisture regime. Wetland water regimes in this part of the North Slope coastal plain tend toward
the less wet water regimes, as compared to the least common and wettest regime. Relatively speak-
ing, wetlands in the study area are of medium to lower ground moisture regimes. Compared to
National Wetlands Inventory (NWI) maps, prepared using 1970s and 1980s imagery following
the approach of Cowardin et al. (1979), we mapped fewer emergent wetlands and more emergent/
scrub–shrub, indicating a shift toward drier ground moisture conditions that favor scrub–shrub
habitat. This trend is also reflected in our increased mapping of scrub–shrub. Combined scrub–
shrub/emergent and scrub–shrub totals for both time periods are similar, although we mapped
more scrub–shrub.
Environmental objectives addressed with the 2012 airborne survey include (1) mapping and mor-
phologically characterizing distinctive permafrost landscape features on the North Slope Arctic
Coastal Plain, including patterned ground, frost mounds and pingos, and thermokarst lakes and
NWI BEG
Habitat
Lake
Pond
Emergent wetland
Emergent/scrub-shrub
Scrub-shrub wetland
Scrub-shrub/emergent
Upland
0 10
km
mi
0 5
N
FIGURE 19.3 Comparison of wetland maps of the airborne survey area from (left) NWI based on aerial
photographs acquired in the 1970s and 1980s and (right) from SPOT imagery acquired in 2009–2010 supple-
mented with surface elevation data and imagery acquired during the August 2012 airborne survey.
basins, (2) discriminating local uplands and wetlands on a microtopographical scale on the low-
relief coastal plain using high-resolution elevation models, and (3) determining depths and volumes
of shallow Arctic lakes to predict potential water availability. We are applying topographic LiDAR
to help identify local areas where soil is likely to be water saturated (a key factor in mapping Arctic
wetlands) and bathymetric LiDAR to rapidly survey lake depths and determine water volumes. In a
remote area such as the North Slope where climate is harsh, few roads exist, and off-road access is
restricted, permafrost landscape and wetlands mapping efforts and lake surveys are labor intensive
and logistically difficult. Satellite-based remote sensing provides insufficient spatial resolution to
identify small and subtle topographic features such as ice-wedge polygons that strongly influence
surface-water flow, soil moisture, and wetland distribution. Information acquired using airborne
LiDAR and imagery promises to increase our ability to rapidly map, characterize, and monitor per-
mafrost features, measure lake depths, calculate water volumes, and focus ground-based wetlands
investigations on those areas most likely to have water-saturated soil, a fundamental discriminator
in wetland mapping.
METHODS
Space-, airborne-, and ground-based remote sensing methods are being increasingly used to
characterize and monitor change in permafrost features, terrain, and processes (e.g., Frohn
et al. 2005; Kääb 2008; Hubbard et al. 2013; Jones et al. 2013; Paine et al. 2013a,b). Airborne
LiDAR in particular offers the benefits of rapid, accurate, and absolute determination of ground-
surface morphology at spatial resolutions of a few to a few tens of centimeters laterally and
vertically, which is adequate to characterize the most important permafrost features and monitor
their change over time with repeat surveys. The airborne LiDAR system (Chiroptera, Airborne
Hydrography AB) used for our North Slope survey consists of a near-infrared (1.0 µm wave-
length) laser system pulsing at selectable rates as high as 400 kHz, a green (0.5 µm) laser system
pulsing at 35 kHz, and a 50 megapixel, three-channel digital camera that acquired color-infrared
(CIR) images to aid permafrost feature and wetlands delineation. The near-infrared laser system
acquired topographic data at single-pass, ground-point densities greater than 20 points/m 2 at a
nominal flight height of 400 m. These data were used to create high-resolution topographic models
of the 490 km 2 survey area south of Deadhorse, Alaska (Figure 19.1). Topographic data acquired
with the near-infrared laser can be displayed as point clouds or gridded to produce digital eleva-
tion models (DEMs). The water-penetrating green laser produced bathymetric point densities
of 1–2/m 2 at the 400 m survey flight height. The bathymetric system recorded waveforms from
which reflections from the water column, including the water surface and the water bottom, were
identified (Figure 19.4). The speed of light in water, when combined with time delays between
identified arrivals in the waveform, determines water depths for individual pulses. Information
derived from the detected surfaces, such as total water volume for lakes and incremental water
volume greater than threshold depths that govern the presence of year-round fish populations,
were calculated by differencing detected upper and lower water surfaces. Spatial accuracy of the
topographic and bathymetric LiDAR data, which are dictated by the accuracy of GPS- and iner-
tial management unit–derived aircraft trajectories that are determined during data processing, is
commonly 10 cm or less.
COMPARISON OF LiDAR- AND GPS-DERIVED ELEVATIONS

Three-dimensional seismic data were acquired in parts of the study area (Figure 19.5) for an oil
company in the winter months of 2012 and 2013, a few months before and a few months after the
August 2012 airborne survey. The seismic survey contractor (CGG) provided kinematic GPS loca-
tions and elevations for seismic source and receiver locations acquired during the two surveys to
allow comparisons of GPS- and topographic LiDAR-derived elevations in areas where the surveys
overlapped. The GPS elevations, acquired using fixed base stations and roving receivers, should pro-
vide locations and elevations accurate to a few centimeters but were acquired during winter when
the ground was frozen and snow and ice were present. The airborne LiDAR survey was acquired in
late summer when the land surface was partly thawed and free of snow and ice.
For the 2012 seismic survey (Figure 19.5), we compared 23,438 source or receiver locations
with topographic LiDAR-derived elevations extracted from the 1 m resolution DEM at the source
or receiver location. The LiDAR-derived elevations were adjusted to match the vertical datum
(NAVD88, Geoid99) used for the seismic surveys. On average, the GPS-derived elevations are
0.45 m higher than LiDAR-derived elevations (Table 19.1). The standard deviation of the elevation
difference is 0.3 m. There were 7817 source or receiver locations available for comparison from the
2013 seismic survey. For these points, GPS-derived elevations were an average of 0.69 m higher
than LiDAR-derived elevations (Table 19.1). Standard deviation for this set of points was 0.2 m.
Combining the data from the two seismic surveys yields 31,255 points for comparison. On average,
GPS-derived elevations are 0.51 m higher than LiDAR-derived elevations, with a standard deviation
of 0.3 m (Table 19.1).
(a)
(b)
(c)
FIGURE 19.4 Bathymetric LiDAR data acquired across a water body in the Deadhorse survey area, August
5, 2012. (a) Map view showing bathymetric LiDAR returns along a single flight-line segment across a shallow
water body. (b) Bathymetric LiDAR waveform showing large-amplitude return (marked by a white line) at
the water surface and smaller-amplitude return (marked by red line) from the water bottom. (c) Cross sec-
tion showing bathymetric LiDAR returns from land surface, water surface, and water bottom across a shal-
low water body located at the white line on (a). The white square in (c) marks the location of the waveform
displayed in (b).
2013 Seismic survey
2012 Seismic survey
2.4 GPS-LiDAR
elevation
difference
–0.5 (m)
0 10 km
2012 LiDAR survey 2012 LiDAR survey 0 5 mi
N
FIGURE 19.5 Difference between GPS- and airborne LiDAR-derived elevation over the 2012 (left) and
2013 (right) seismic survey areas. Positive values denote GPS-derived elevations higher than LiDAR-derived
elevations. GPS-derived elevations are those of source and receiver locations provided by CGG.
TABLE 19.1
Comparison of GPS-Derived Seismic Source and Receiver Elevations with
Airborne LiDAR-Derived Elevations
2012 Seismic Survey 2013 Seismic Survey Combined
Number of points 23,438 7,817 31,255
Average difference (m) 0.45 0.69 0.51
Standard deviation (m) 0.30 0.20 0.30
Notes: GPS-derived elevations were provided by CGG for seismic surveys completed in winter 2012 and winter
2013. Positive differences indicate GPS-derived elevations higher than LiDAR-derived elevations.
Maps prepared to depict the difference in elevation between the LiDAR- and GPS-derived eleva-
tions show higher GPS-derived elevations over most of the survey area (Figure 19.5). East–west
banding evident in both seismic surveys is strongly correlated with the orientation of seismic source
and receiver lines. The banding pattern is attributable to systematic elevation differences in adjacent
seismic source and receiver lines, not the LiDAR-derived elevation data. The average vertical offset
between GPS- and LiDAR-derived elevations of 0.45–0.69 m is likely attributable to true elevation
differences between winter conditions (presence of frozen ground, ice, and snow) during the 2012
and 2013 seismic surveys and partly thawed ground with no ice or snow during the 2012 airborne
survey. This difference is comparable to the mean April snow depth of 0.40 m in the Prudhoe Bay
area (Everett and Parkinson 1977).
The only significant area where GPS elevations are lower than LiDAR-derived elevations is
along the western margin of the Sagavanirktok River (predominantly blue areas in Figure 19.5).
Because this is an active part of the riverbed, it is possible that the apparent increase in elevation
between the time the GPS-based elevations were measured for the first seismic survey (acquired
in winter 2012) and the airborne LiDAR survey (late summer 2012) is attributable to sedimenta-
tion during the spring and summer thaw. The same general area appears to have lost elevation
between the August 2012 airborne survey and the time the GPS-based elevations were acquired for
the winter 2013 seismic survey, possibly caused by stream erosion of the riverbed. The airborne
LiDAR survey thus provides a baseline surface-elevation datum that can be compared to future
ground or airborne elevation surveys to provide site-specific estimates of river sedimentation or
erosion and snow and ice thickness, likely accurate to within a decimeter, anywhere within the
survey footprint.
MAPPING PERMAFROST FEATURES: POLYGONS,

PINGOS, AND THERMOKARST LAKES
Airborne and ground-based laser systems have been used since the 1990s to produce high-resolution
DEMs at bin sizes of 1 m or smaller at a primary elevation accuracy of a few to a few tens of cm. The
Alaskan North Slope in the Prudhoe Bay region is an area of generally low relief having important
periglacial and permafrost features such as patterned, polygonated ground and pingos (roughly con-
ical ice-cored structures mantled with soil that can be a few to tens of meters high) (Pihlainen et al.
1956; Ferrians 1988; van Everdingen 1998). Public-domain topographic data over this region within
the Arctic Circle consist of small-scale (1:63,360) U.S. Geological Survey (USGS) topographic
maps contoured at 25 ft (~8 m) intervals (Figure 19.2) and DEMs created from Landsat 5 data at
30 m spatial resolution. These data have insufficient lateral and vertical resolution to map critical
permafrost features or the elevation-influenced distribution of wetlands. Higher-resolution topo-
graphic maps of the area have been produced for the oil industry using photogrammetric methods
at a scale of 1:6000 and 1.5 m contour intervals (Walker et al. 1985), but these too have insufficient
resolution to accurately map critical permafrost features such as polygonal soils and frost mounds.
Airborne interferometric synthetic aperture radar (InSAR) has been employed on the North Slope
coastal plain near and west of the Colville River to assess pingo distribution and shape characteris-
tics through the production of digital surface models at 5 m horizontal resolution and 0.1 m vertical
resolution (Jones et al. 2012). Although adequate for larger features such as pingos, this resolution is
insufficient to identify individual polygons characteristic of patterned ground common in the survey
area. High-frequency topographic LiDAR systems can produce high-resolution elevation models at
single-pass point densities greater than 20/m2 and vertical accuracy of a few centimeter, about an
order of magnitude greater than that achieved with InSAR. At the survey-area scale (Figure 19.2),
these LiDAR-derived maps of the North Slope depict the general elevation decrease on the coastal
plain from the foothills of the Brooks Range to the south of the Sagavanirktok River delta and
associated deposits adjacent to the Beaufort Sea on the north. Lack of backscattered near-infrared
laser reflection from specular surfaces such as lakes and rivers leaves gaps in the LiDAR data where
water stands at the surface. These gaps can be used to delineate river channel and thermokarst-
related lake boundaries.
Full-resolution (400 kHz) laser returns reveal spatial detail that cannot be readily achieved with
available satellite, aerial imagery, and ground-based surveying methods. We constructed a detailed
DEM (0.25 m resolution) from full-resolution LiDAR data at a well pad along the Sagavanirktok
River and the adjacent highway and compared that to the 30 m resolution DEM constructed from
Landsat 5 Thematic Mapper (TM) data (Figure 19.6). Permafrost-related features such as soil and
Permafrost polygons
y
wa
igh
nH
lto
Da
Stream channel
Merak well pad
(a) (b)
Permafrost polygons
y
wa
igh
nH
lto
Da
Stream channel
Merak well pad
66 60
Elev. Elev.
(m) (m)
55 55 100 m N
(c) (d)
FIGURE 19.6 Comparison of resolution achieved in a 400 × 400 m area on the west bank of the Sagavanirktok
River near the Merak #1 well pad (Figure 19.2) using (a) 2009–2010 SPOT commercial imagery (2.5 m resolu-
tion), (b) 2012 airborne survey CIR imagery (0.1 m resolution), (c) a DEM (30 m resolution) constructed from
Landsat 5 TM data, and (d) a DEM (0.25 m resolution) constructed from 2012 airborne survey topographic
LiDAR data. Visible features from the 2012 survey data include the Dalton Highway, ice-wedge polygons,
individual channels of the Sagavanirktok River, and the well pad. (Landsat 5 data from the USGS.)
FIGURE 19.7 Perspective view of 0.25 m from 400 kHz data DEM (Figure 19.6d) toward the northwest of
the Merak 1 well pad (foreground), the Dalton Highway, and permafrost-related soil and ice polygons in the
North Slope survey area. Image constructed from full-resolution topographic LiDAR data before vegetation
and feature removal.
ice polygons that are a few to a few tens of meters across are not detectable using lower-resolution
data, but are well captured and readily mappable at topographic LiDAR resolution (Figure 19.6d).
DEMs such as these allow discrimination of individual soil polygons and the lower-elevation areas
between them, while high-resolution CIR imagery can distinguish textural patterns associated with
vegetation assemblages and help discriminate among them. Elevation profiles across individual
polygons can be used to classify them as low or high centered. This information can then be used
to better predict water movement and soil moisture patterns in wetland and upland environments.
LiDAR data are also readily converted to 3D surfaces (Figure 19.7) that can be viewed from advan-
tageous perspectives to highlight surface features and aid analysis and interpretation of permafrost
terrain.
Patterned Ground: Polygons and Interpolygon Ice Wedges

Ground polygons that can be a few to a few tens of meters across (Leffingwell 1919; Cabot 1947;
Carson and Hussey 1962; Washburn 1980; Kanevskiy et al. 2013) are a distinctive type of pat-
terned ground that is common on the North Slope and in the survey area (Washburn 1956).
Polygonal ground patterning is principally attributed to thermal contraction in continuous per-
mafrost zones. Thermal contraction polygons commonly develop ice wedges within the cracks
at polygon boundaries (Leffingwell 1919; Carson and Hussey 1962; Lachenbruch 1962, 1966;
Washburn 1980) that are typically 3–4 m deep and narrow downward from maximum widths of
1–2 m at the surface (Carson and Hussey 1962). Topography associated with patterned ground
(low- and high-centered polygons and the trenches between them) can strongly influence local
drainage patterns and distribution of soil moisture, both of which are critical factors in wetland
development and classification.
Both the imaging and LiDAR systems on the aircraft acquired data at sufficient resolution to
map the boundaries (using imagery and LiDAR) and detailed morphology (using LiDAR only) of
individual ice-wedge polygons in the Prudhoe Bay area (Figure 19.6). Perspective views of DEMs
constructed from LiDAR data at 0.25 m resolution adjacent to the Sagavanirktok River (Figure 19.7)
clearly show prominent patterned ground boundaries and other significant landscape features
(stream channels, river bars, possible eolian landforms adjacent to the river, and infrastructure such
as roads and well-pad structures) that influence wetland distribution and landscape drainage.
In a 390 × 340 m test site located within a large area of patterned ground in the central part of the
Prudhoe Bay survey area (Figure 19.2), a 0.5 m resolution DEM and a 0.1 m resolution CIR image
were created to illustrate ice-wedge polygon mapping capabilities (Figure 19.8). Polygon boundar-
ies are readily mappable on both the CIR imagery (Figure 19.8a) and the LiDAR-derived DEM
(Figure 19.8b). Total relief within this polygon field is about 1 m. The appearance of these distinc-
tive geomorphic features is enhanced through common visualization methods including perspective
views of CIR imagery (Figure 19.8c), topography only with vertical exaggeration (Figure 19.8e), and
vertically exaggerated topography draped with semitransparent CIR imagery (Figure 19.8d). This
approach combines the vegetation-interpretation advantages of CIR imagery with detailed landform
morphology obtained from DEMs.
Enlargements of smaller areas of patterned ground (Figure 19.9) can be used to map individual
polygons; classify the land surface into intrapolygon and interpolygon, ice-wedge environments;
Enlarged area
(c)
(a)
(d)
Enlarged area
(e)
53.5
Elev.
(m NGVD88)
N 52.8 100 m
(b)
FIGURE 19.8 Ice-wedge polygon field in a 390 × 340 m area within the North Slope survey area near
Prudhoe Bay (Figure 19.2). Patterned ground is depicted (a) on high-resolution (0.1 m pixel width) CIR imag-
ery, (b) in a LiDAR-derived DEM (0.5 m cell size), and on perspective views of (c) CIR imagery only, (d) a
DEM draped with a minor imagery component, and (e) a DEM only. Vertical exaggeration on the perspective
DEM views is 3×. The enlarged area indicated on (a) and (b) is shown in greater detail in Figure 19.9.
Vehicle C
tracks
D
(c)
(a)
Vehicle C (d)
tracks
(e)
B
53.1
Elev.
(m NGVD88)
N 20 m
(b) 52.9
FIGURE 19.9 Enlargement of a 75 × 70 m area within an ice-wedge polygon field in the North Slope survey
area near Prudhoe Bay (Figures 19.2 and 19.8). Low-centered polygons and interpolygon ice-wedge trenches
as depicted (a) on high-resolution (0.1 m pixel width) CIR imagery, (b) in a LiDAR-derived DEM (0.5 m cell
size), and on perspective views of (c) CIR imagery only, (d) a DEM draped with a minor imagery component,
and (e) a DEM only. Vertical exaggeration on the perspective DEM views is 4×. Locations of topographic
transects A, B, C, and D (Figure 19.10) across individual polygons are indicated on (a) and (b).
and determine morphological characteristics of the polygons. High-resolution CIR images (Figure
19.9a), a LiDAR-derived DEM (Figure 19.9b), and various perspective views (Figure 19.9c through e)
of an area measuring 75 × 70 m in an ice-wedge polygon field captured 19 complete and 11 partial
polygons. Calculated polygon density in this area is greater than 4700 polygons/km2. Average poly-
gon width is about 15 m. Total elevation range across the imaged area is slight; high ground on
polygons is generally no more than a fraction of a meter higher than surfaces within the interpoly-
gon ice-wedge space, further emphasizing the need for highly accurate elevation data to adequately
characterize polygon surfaces.
Elevation transects across three polygons (A, B, and C, Figures 19.9 and 19.10) that range from
15 to 22 m in width depict low-centered polygons with boundary ridges that are 10–20 cm above the
53.1
Elevation (m NGVD88)
53.0
52.9
52.8
Transect A
52.7
0 5 10 15 20 25
(a) Distance (m)
53.2
53.1
53.0
52.9
Transect B
52.8
0 5 10 15 20 25
(b) Distance (m)
53.2
53.1
53.0
52.9
Transect C
52.8
0 5 10 15 20 25
(c) Distance (m)
53.2
53.1 Vehicle tracks on polygon surface
53.0
52.9
Transect D
52.8
0 5 10 15 20 25
(d) Distance (m)
FIGURE 19.10 Elevation transects A, B, C, and D (Figure 19.9) across low-centered ice-wedge polygons
in the North Slope survey area. Transect D crosses relict north–south vehicle tracks evident on CIR imagery
(Figure 19.9a) and the LiDAR-derived DEM (Figure 19.9b).
centers. Maximum relief is found between the polygon boundary ridges and the interpolygon ice-
wedge trenches. For these three polygons, the elevation difference between the ridges and trenches
is 17–18 cm along transects B (Figure 19.10b) and C (Figure 19.10c) and 23 cm along transect A
(Figure 19.10a). Subtle elevation changes on the polygons and orientation, connection, and slope
of the interpolygon trenches are critical inputs to models of soil moisture distribution in the active
layer and rain and meltwater flow across the landscape. Landscape disturbance, such as a prominent
series of north–south vehicle tracks across the tundra (Figure 19.9a and b), can be detected and
associated minor relief can be quantified (a few centimeters along transect D) (Figure 19.10d) on
high-resolution imagery and DEMs.
Pingos and Frost Mounds

Pingos are distinctive and dynamic ice-cored, soil-mantled permafrost features (Porsild 1938;
Pihlainen et al. 1956; Ryckborst 1975; Flemal 1976; Washburn 1980; Mackay 1998) found on the
Alaskan North Slope coastal plain (gravel mounds of Leffingwell 1919; Ferrians 1988) and in
other Arctic, subarctic, and Antarctic regions. It is estimated that there are more than 5000 pingos
worldwide, including significant populations in North America, Russia, Fennoscandia, Spitsbergen,
Greenland, Mongolia, and the Tibetan Plateau (Mackay 1998). More than 1500 have been estimated
to exist in Alaska alone (Galloway and Carter 1978; Carter and Galloway 1979; Hamilton and Obi
1982; Ferrians 1988; Mackay 1998) and perhaps more than 3000 (Jorgenson et al. 2008a). Recent
airborne InSAR-based studies indicate this number may be conservative. Jones et al. (2012) counted
1247 pingos between 2 and 21 m high within a 40,000 km2 area on the western Arctic Coastal
Plain, the majority (98%) of which occur within drained lake basins. More than 97% of the mapped
Alaska Arctic Coastal Plain occurs in areas underlain by sandy unconsolidated deposits (Ferrians
1988), allowing basal infiltration of water to supply pingo growth. Pingo-like features have also
been documented that rise from the floor of the Beaufort Sea, although the origin process may dif-
fer (Washburn 1980; Paull et al. 2007). In the Prudhoe Bay region of the North Slope, soil-mantled,
frost-mound structures form steep-sided, roughly conical shapes that emerge from lake basins and
can dominate the low-relief landscape, as well as broader, lower structures on older surfaces out-
side lake basins (Walker et al. 1985). Through soil analysis, Walker et al. (1996) determined that
the steep-sided types are found on landscape surfaces of all ages but exclusively within thaw-lake
basins. Broad-based types with gentler slopes are found exclusively outside lake basins. Regression
analysis of soil profile development yielded minimum age estimates for steep-sided pingos of 5 kyr
and broad-based pingos of 14–22 kyr (Walker et al. 1996).
Vertical growth rates of basal ice lenses that support pingo growth reach 50–300 mm/year in
pingos of the nearby Tuktoyaktuk Peninsula, Northwest Territories, Canada (Ryckborst 1975).
Repeat surveys at Ibyuk Pingo, one of the tallest in the world at 49 m, documented summit growth
rates averaging 23 mm/year (Mackay 1986). Electrical freezing potentials monitored in a borehole
through another 5 m high Tuktoyaktuk pingo recorded basal freezing-front advancement at aver-
age rates of 15 mm/year (Parameswaran and Mackay 1996). Pingo height is roughly correlated
to age; Ryckborst (1975) estimates that 14 m high pingos can form in about 75 years, while the
highest of pingos (60–70 m relief) require a millennium to develop. Topographic LiDAR data and
high-resolution imagery greatly assist in characterizing and monitoring the growth or decay of
these features over time. LiDAR data (DEMs and profiles) acquired across a pingo in the northern
part of the survey area (Figure 19.11), for example, reveal that it is a 15 m high and 160 m wide
conical structure that occupies a basin that is about 3 m below the surrounding coastal plain. High-
resolution (0.05 m pixel size) imagery overlain on the DEM (1 m cell size) enhances the appearance
of the summit crater, the central ice core, mantling soil, and flank crevices (Figure 19.11).
Analysis of LiDAR data over the 490 km 2 airborne survey area identified 19 frost-mound
features (Figure 19.12, Table 19.2) within and outside lake basins, including the one (feature 12)
previously described. The features range from about 1 to more than 15 m in relief above the sur-
rounding land surface. Large frost-mound areal density is about 0.04/km 2, similar to the 0.03/
km 2 density calculated from the much larger InSAR-based study on the western Arctic Coastal
Plain (Jones et al. 2012). Ten of these features are 3 m high or higher and would certainly be clas-
sified as pingos, although the relief distinction between large (pingos) and smaller frost mounds
is subjective. The width of these features, also measured from LiDAR data, ranges from about
Topo LiDAR DEM

Topo LiDAR
(1 m resolution)
elevation profiles
Topo LiDAR
elevation perspective
RGB imagery
(~ 0.05 m resolution)
RGB imagery draped

on LiDAR-derived DEM
FIGURE 19.11 Topographic models, elevation profiles, perspective views, and RGB imagery of a large
pingo (15 m high and 160 m wide) in the northern part of the North Slope survey area.
30 to more than 700 m (Table 19.2). They are typically elongate, averaging 130 m for the long axis
and 89 m for the short axis. Most frost-mound features, including the larger pingos, are ellipti-
cal to nearly circular in plan view. One highly linear, ridgelike feature with 5 m of relief above
the surrounding lake floor was identified (feature 19, Figure 19.12). All but the smallest feature
(13, Figure 19.12, Table 19.2) are located within elongate basins as much as 2.8 km across. Some
of the basins had no appreciable surface water, some were partly flooded, and one was mostly
flooded at the time of the survey. Because many frost mounds are located in basins that are partly
water filled, measurements of pingo relief and basal dimensions were facilitated by creating a
DEM that merged dry-land elevations from topographic LiDAR returns with lake-bottom eleva-
tions from bathymetric LiDAR returns (Figure 19.13). In this way, basin depths, water depths, and
total frost-mound relief can be determined from a single, merged DEM. The two largest pingos,
feature 12 with more than 15 m relief (Figure 19.11) and feature 18 with more than 13 m relief
(Figure 19.13), are both found within lake basins that were partly filled with water at the time of
the survey late in the Arctic summer. Important associated features relevant to wetland distri-
bution, such as drainage channels between adjacent lakes at differing elevations and polygonal
ground adjacent to (and commonly beneath) lakes, are also readily identifiable on DEMs created
from merged topographic and bathymetric LiDAR returns.
Thermokarst Lake and Stream Bathymetry

There are thousands of freshwater lakes on the Alaskan North Slope coastal plain (Cabot 1947; Black
and Barksdale 1949; Carson and Hussey 1962; Sellman et al. 1975; Hinkel et al. 2007; Jorgenson
et al. 2008a) that are tens to thousands of meters across, typically as much as a few meters deep,
commonly aligned on a northerly trend, elongate, and found within basins that can be a few meters
below the relatively flat and featureless coastal plain. Lake basins are commonly subelliptical to
2
1
3
5
7
8
9
10 11
12
13
14
15
16
Pingo or frost mound
303
17 Elev.
18 (m)
19 18
20
0 10 km
0 5 mi
N
FIGURE 19.12 Locations of pingos and other frost-mound features (Table 19.2) identified on land-surface
DEM constructed from airborne LiDAR data.
subrectangular; larger basins tend to taper to the north (Carson and Hussey 1962). These lakes are
considered to be the dominant geomorphic element of the coastal plain in the Prudhoe Bay area
(Everett and Parkinson 1977). Similar shallow lakes are found in other permafrost areas of Alaska,
including on the Seward Peninsula and in eastern Alaska, and are commonly interpreted as ther-
mokarstic features that formed from subsidence caused by local thawing (and associated volume
loss) of perennially frozen ground (Hopkins 1949; Washburn 1980) and can expand by wave ero-
sion (Wallace 1948). On the North Slope, thaw basins commonly originate in low-center polygonal
areas and at the junction of ice wedges where water can accumulate (Carson and Hussey 1962).
Developmental stages of thaw lakes and surrounding basins in northern Alaska have been recently
reevaluated by Jorgenson and Shur (2007), who propose six stages of development that began in
the early Holocene and include initial flooding of surface depressions, lateral expansion and sedi-
ment redistribution, lake drainage with stream network expansion, differential ice aggradation in
the basin, secondary development of lakes within basins, and finally basin stabilization. Detailed
morphological data derived from topographic and bathymetric LiDAR surveys have obvious utility
in evaluating thaw-lake basin formation and evolution over extensive areas.
Lakes at higher elevations on the coastal plain tend to be smaller and occur at lower density than
lakes found at lower elevations (Sellman et al. 1975; Hinkel et al. 2007). Most lakes are shallower
TABLE 19.2
Physical Dimensions and Context of Deadhorse Area Pingos and Other Frost-Mound
Features Identified Using Airborne Topographic and Bathymetric LiDAR Data Acquired
in August 2012
Elongation
Surface Long-Axis Short-Axis Orientation (Degrees
Feature Relief (m) Length (m) Length (m) cw from N) Notes
1 7.5 90 82 75 In an elongate basin 2.8 km across;
basin partly water filled
4 2.8 43 37 90 In an elongate basin 650 m across;
6 1.9 50 34 10 In an elongate, water-filled basin
1.75 km across
7 3.5 75 50 80 In a mostly emergent, rectangular
basin 1.5 km across
basin 1.5 km across
basin 1.75 km across
12 1.1 67 64 320 Outside any well-defined basin; small
feature
13 1.9 73 61 80 In a mostly water-filled, elongate
basin 650 m across
14 4.0 115 91 350 In a mostly emergent, elongate basin
2.5 km across
15 1.8 83 60 0 In a water-filled, elongate basin 1 km
across
1.6 km across
17 13.3 140 110 80 In a half-emergent, nearly circular
1.3 km across
19 3.1 90 54 315 In a partly water-filled, elongate
basin 1.2 km across
Note: Locations shown in Figure 19.12.

19 (linear mound)
Channel N
Lake basin
Lake basin Channel

18 (pingo)
Lake basin
Polygon field
V.E. = 10×
102
Elev.
(m)
86
FIGURE 19.13 Perspective view (to the south–southwest) of two large frost mounds (18 and 19, Figure 19.12)
located within three shallow, partly drained North Slope lake basins adjacent to a permafrost polygon field.
Darkest blue areas within the basins indicate lake area in August 2012. At the time of the survey, LiDAR data
indicated that the channel between the southern basin and the middle basin was dry. The channel between
the middle and northern basin contained water. This hybrid elevation image was constructed from merged
topographic (ground surface) and bathymetric (lake bottom) LiDAR data.
than 2 m and freeze completely during the winter, but some deeper lakes do not freeze to the bottom
(Sellman et al. 1975). Preferred northerly elongation and orientation has been attributed to develop-
ment of wave-eroded sublittoral shelves on the eastern and western margins of the lakes that reduce
subsequent wind-driven wave erosion and sublake thawing during prevailing east–northeasterly and
west–southwesterly winds (Black and Barksdale 1949; Carson and Hussey 1962). Rates of lateral
retreat range from 0.1 to 2 m/year, depending on ice content, soil texture, thickness of the organic
mat, and wave action (Jorgenson et al. 2008a). Lakes in the study area are relatively clear and shal-
low and can be important sources for water in winter and habitat for overwintering fish populations
if the water depths are greater than about 2.1 m and do not completely freeze in winter (Sellman
et al. 1975). The lakes are also potential water sources for ice-road construction and hydrocarbon-
development activities in the Prudhoe Bay area. Lake depth determines the feasibility of using the
lakes for winter water supply and strongly influences the biological resources of the lakes (Sellman
et al. 1975).
In 2011, lake volume and fish population studies were completed by a consulting firm in 26 lakes
along the Dalton Highway using helicopters to transport field crews and fathometer surveys from
small boats (ASRC Energy Services 2012). The lakes surveyed by boat fathometer in 2011 had sur-
face areas between 4 and 191 ha, measured depths as deep as 3.5 m, and water clarity (a key factor
controlling bathymetric laser penetration depths) ranging from 0.7 to 4.6 nephelometric turbidity
units (NTU) (ASRC Energy Services 2012). The 2012 airborne LiDAR survey covered an area that
included nearly all the lakes surveyed by boat in 2011, providing an opportunity to compare results
from the different surveying methods and analyze airborne bathymetric LiDAR data from the hun-
dreds of lakes within the survey area.
Amplitudes of returns from the bathymetric laser are recorded as a waveform (Figure 19.4)
from which interpreted features such as the water surface and water bottom can be interpreted
and extracted. DEMs can be created from each return type. Survey area–wide DEMs constructed
from bathymetric LiDAR returns classified as the upper surface of the water and the lake or
stream bottom define the location and extent of water bodies as well as the elevation of the water
bottom.
Lake Volume
Bathymetric LiDAR has been shown to penetrate to the water bottom over most of the Deadhorse
area lakes and streams. Surfaces that define the water surface and water bottom can be combined
to determine water depths (Figure 19.14) and total or incremental water volumes below chosen
threshold depths. We developed a semiautomated, custom-software-based procedure that combines
information derived from survey-wide topographic and bathymetric LiDAR data to produce water
surface, water bottom, and water depth layers that can be manipulated to generate total and incre-
mental water-volume estimates for arbitrary lake area thresholds and depth ranges.
Analysis of airborne LiDAR data indicates that there are 283 lakes within the survey area that
have a surface area larger than 0.8 ha (Figure 19.14). The largest has a surface area of 234 ha. The
average lake has a surface area of 10.5 ha. In all lakes combined, the total volume of water is cal-
culated to be 20.3 × 106 m3, although the estimated volume of individual lakes varies greatly from
less than about 100 m3 to as much as 1.63 × 106 m3. Average lake volume for lakes with surface area
greater than 0.8 ha is 72.1 × 103 m3.
Most of the lake volume occupies shallow water. Of the 283 lakes analyzed, only 84 (30%) are
deeper than 1.5 m and only 39 (14%) are deeper than 2.1 m. Just over 35% of the total lake water
LiDAR-surveyed
lake
0
Depth
(m)
8
0 10 km
N 0 5 mi
FIGURE 19.14 Map of 283 LiDAR-surveyed North Slope lakes and water depth calculated by subtracting
LiDAR-derived, water-surface elevations from water-bottom elevations. Bathymetric LiDAR point density is
1–2/m2.
volume (7.14 × 106 m3) is found in water depths of 0.3 m or less. Almost 4% (725 × 103 m3) of the
total lake water volume is at depths greater than 1.5 m in 84 of the lakes, and less than 1% (116.8 ×
103 m3) of the total water volume is at depths greater than 2.1 m.
Comparisons of water volumes calculated using the August 2012 LiDAR data (this study) with
volumes calculated using boat-based fathometer data acquired in August 2011 (ASRC Energy
Services 2012) show that, for unobscured lakes, the LiDAR-calculated total volumes are greater
by an average of about 9%. Examination of aerial photographs taken during the airborne survey
suggests that, over parts of some lakes, green laser penetration to the lake bottom may be obscured
by the presence of floating matter or suspended sediment or is absorbed by a dark lake bottom.
Analysis of photographs over the entire survey area indicates that bottom obscuration (and thus
possible underestimated total volume determined from LiDAR data) may be present in as many
as 21 lakes, less than 10% of the total number of lakes surveyed. Aside from the potentially large
differences arising from lake-bottom obscuration in a small number of lakes, nonobscured lakes
have minor differences in volumes determined using independent airborne LiDAR and boat-based
fathometer methods. These small differences in calculated volumes could represent real volumet-
ric differences resulting from water-level differences between the two survey dates, or they could
be artifacts of one or both methods. For example, LiDAR sampling of the water bottom is more
complete than fathometer data would be, particularly in very shallow water along lake margins not
accessible by boat. Alternatively, misclassification of water-bottom returns in LiDAR data could
lead to erroneous water-depth calculations that would cause errors in volumetric calculations, as
would errors in water-surface elevations.
CONCLUSION
Airborne topographic and bathymetric LiDAR systems permit rapid surveying of upland, wetland,
lacustrine, and riverine environments in periglacial and permafrost terrain such as those on the
Alaskan North Slope. Topographic detail achieved using a near-infrared laser pulsing at rates as
high as 400 kHz allows applications that include mapping large and small geomorphic features
(such as frost mounds and ice-wedge polygons), better defining the role of microtopography in wet-
land distribution and surface-water flow, monitoring topographic change over time that can accom-
pany permafrost change at a vertical scale of a few centimeter, detecting and quantifying effects
of vehicular activity on the landscape, and perhaps aiding the design of infrastructure develop-
ment and seismic surveys to minimize environmental impact of human activities. High-resolution
DEMs of the land surface in remote regions such as the North Slope, where detailed topographic
information has not been widely available, allow better delineation of drainage networks and pre-
diction of soil moisture that are critical to wetland classification. Bathymetric LiDAR has been
shown to penetrate reasonably clear North Slope lakes and streams to depths as great as 6 m at
single-pass densities of about 1/m2, allowing precise calculations of total and incremental water
volumes that are important parameters for determining which lakes are likely to support fish popu-
lations. Detailed maps of lake-bottom morphology provide a unique and comprehensive dataset
to support studies of the formation, sedimentation, evolution, and future change of these common
Arctic landscape features.
ACKNOWLEDGMENTS
Funding for this project was provided by Great Bear Petroleum Operating LLC under contract no.
UTA12-000752. Ed Duncan and Karen Duncan (Great Bear Petroleum) provided logistical support
for the airborne survey and field investigations. Jeffrey G. Paine and Michael H. Young (Bureau of
Economic Geology) served as principal investigators. Bureau of Economic Geology staff Tiffany
L. Caudle arranged airborne survey logistics and Aaron R. Averett participated in data acquisi-
tion. Petter Kullenberg, Daniel Andersson, and Torbjörn Karlsson (Airborne Hydrography AB)
participated in data acquisition, preliminary data processing, and algorithm development to identify
water-surface and water-bottom laser returns. Bradley Busch and Marcos Rico (Aspen Helicopters,
Inc.) flew and maintained the aircraft. Joseph Christopher (ASRC Energy Services, Inc.) provided
fathometer-based bathymetric data from select North Slope lakes and guidance on wetland map-
ping. Publication authorized by the director of the Bureau of Economic Geology, the University of
Texas at Austin.
REFERENCES
ASRC Energy Services. 2012. 2011 Great Bear Lake Studies Report, North Slope, Alaska. ASRC Energy
Services, Anchorage, AK.
Black, R.F. and W.L. Barksdale. 1949. Oriented lakes of Northern Alaska. Journal of Geology 57:105–118.
Cabot, E.C. 1947. The Northern Alaskan coastal plain interpreted from aerial photographs. Geographical
Review 37:639–648.
Carson, C.E. and K.M. Hussey. 1962. The oriented lakes of Arctic Alaska. Journal of Geology 70:417–439.
Carter, L.D. and J.P. Galloway. 1979. Arctic coastal plain pingos in the National Petroleum Reserve in
Alaska. In Johnson, K.M. and J.R. Williams (eds.), The United States Geological Survey in Alaska:
Accomplishments During 1978, pp. B33–B35. U.S. Geological Survey Circular 804-B, Reston, VA.
Habitats of the United States. U.S. Department of the Interior, Fish and Wildlife Service, Washington,
DC. FWS/OBS-79/31.
Everett, K.R. and R.J. Parkinson. 1977. Soil and landform associations, Prudhoe Bay area, Alaska. Arctic and
Alpine Research 9:1–19.
Ferrians, O.J., Jr. 1988. Pingos in Alaska: A review. In Senneset, K. (ed.), Proceedings of Fifth International
Conference on Permafrost, pp. 734–739. Trondheim, Norway.
Flemal, R.C. 1976. Pingos and pingo scars: Their characteristics, distribution, and utility in reconstructing
former permafrost environments. Quaternary Research 6:37–53.
Frohn, R.C., K.M. Hinkel, and W.R. Eisner. 2005. Satellite remote sensing classification of thaw lakes and
drained thaw lake basins on the North Slope of Alaska. Remote Sensing of the Environment 97:116–126.
Galloway, J.P. and L.D. Carter. 1978. Preliminary map of pingos in the National Petroleum Reserve in Alaska.
U.S. Geological Survey, Open-File Report, pp. 78–795.
Hamilton, T.D. and C.M. Obi. 1982. Pingos in the Brooks Range, Northern Alaska, U.S.A. Arctic and Alpine
Hinkel, K.M., B.M. Jones, W.R. Eisner, C.J. Cuomo, R.A. Beck, and R. Frohn. 2007. Methods to assess natural
and anthropogenic thaw lake drainage on the western Arctic coastal plain of northern Alaska. Journal of
Geophysical Research 112:F02S16, 9. doi:10.1029/2006JF000584.
Hopkins, D.M. 1949. Thaw lakes and thaw sinks in the Imuruk Lake area, Seward Peninsula, Alaska. Journal
of Geology 57:119–131.
Hubbard, S.S., C. Gangodagamage, B. Dafflon, H. Wainwright, J. Peterson, A. Gusmeroli, C. Ulrich et al.
2013. Quantifying and relating land-surface and subsurface variability in permafrost environments using
LiDAR and surface geophysical datasets. Hydrogeology Journal 21:149–169.
Jones, B.M., G. Grosse, K.M. Hinkel, C.D. Arp, S. Walker, R.A. Beck, and J.P. Galloway. 2012. Assessment
of pingo distribution and morphometry using an IfSAR derived digital surface model, western Arctic
Coastal Plain, Northern Alaska. Geomorphology 138:1–14.
Jones, B.M., J.M. Stoker, A.E. Gibbs, G. Grosse, V.E. Romanovsky, T.A. Douglas, N.E.M. Kinsman, and
B.M. Richmond. 2013. Quantifying landscape change in an arctic coastal lowland using repeat airborne
LiDAR. Environmental Research Letters 8:045025, 10.
Jorgenson, M.T. and Y. Shur. 2007. Evolution of lakes and basins in Northern Alaska and discussion of the thaw
lake cycle. Journal of Geophysical Research 112:F02S17, 12. doi:10.1029/2006JF000531.
Jorgenson, M.T., Y.L. Shur, and T.E. Osterkamp. 2008a. Thermokarst in Alaska. In Kane, D.L. and K.M. Hinkel
(eds.), Extended Abstracts. Ninth International Conference on Permafrost, pp. 869–876, Fairbanks, AK.
Jorgenson, M.T., K. Yoshikawa, M. Kanevskiy, Y. Shur, V. Romanovsky, S. Marchenko, G. Grosse, J. Brown,
and B. Jones. 2008b. Permafrost characteristics of Alaska. Institute of Northern Engineering, University
of Alaska Fairbanks, map scale 1:7,200,000.
Kääb, A. 2008. Remote sensing of permafrost-related problems and hazards. Permafrost and Periglacial
Processes 19:107–136.
Kanevskiy, M., Y. Shur, M.T. Jorgenson, C.-L. Ping, G.J. Michaelson, D. Fortier, E. Stephani, M. Dillon, and
V. Tumskoy. 2013. Ground ice in the upper permafrost of the Beaufort Sea coast of Alaska. Cold Regions
Science and Technology 85:56–70.
Lachenbruch, A.H. 1962. Mechanics of thermal contraction cracks and ice-wedge polygons in permafrost. U.S.
Geological Survey, special paper 70.
Lachenbruch, A.H. 1966. Contraction theory of ice-wedge polygons: A qualitative discussion. In Proceedings
of Permafrost International Conference, pp. 63–71, Lafayette, IN, November 1963. U.S. National
Academy of Sciences, Washington, DC, publication 1287.
Leffingwell, E.K. 1919. The Canning River region, Northern Alaska. U.S. Geological Survey Professional
Paper 109.
Mackay, J.R. 1986. Growth of Ibyuk Pingo, Western Arctic Coast, Canada, and some implications for environ-
mental reconstructions. Quaternary Research 26:68–80.
Mackay, J.R. 1998. Pingo growth and collapse, Tuktoyaktuk Peninsula Area, Western Arctic Coast,
Canada: A long-term field study. Geographie Physique et Quaternaire 52:271–323.
Osterkamp, T.E. and M.W. Payne. 1981. Estimates of permafrost thickness from well logs in northern Alaska.
Cold Regions Science and Technology 5:13–27.
Osterkamp, T.E., J.K. Peterson, and T.S. Collet. 1985. Permafrost thicknesses in the Oliktok Point, Prudhoe
Bay and Mikkelsen Bay areas of Alaska. Cold Regions Science and Technology 11:99–105.
Paine, J.G., J.R. Andrews, K. Saylam, T.A. Tremblay, A.R. Averett, T.L. Caudle, T. Meyer, and M.H. Young.
2013a. Airborne LiDAR on the Alaskan North Slope: Wetlands mapping, lake volumes, and permafrost
features. The Leading Edge 32:798–805.
Paine, J.G., J.R. Andrews, K. Saylam, T.A. Tremblay, M.H. Young, C. Abolt, B. Bradford, T. Caudle, T. Meyer,
and A. Neuenschwander. 2013b. Determining Wetlands Distribution, Lake Depths, and Topography
using Airborne LiDAR and Imagery on the North Slope, Deadhorse Area, Alaska. Bureau of Economic
Geology, The University of Texas at Austin. Final Technical Report prepared for Great Bear Petroleum
Operating LLC under Sponsored Research Agreement UTA12-0000752.
Parameswaran, V.R. and J.R. Mackay. 1996. Electrical freezing potentials measured in a pingo growing in the
western Canadian Arctic. Cold Regions Science and Technology 24:191–203.
Paull, C.K., W. Ussler III, S.R. Dallimore, S.M. Blasco, T.D. Lorensen, H. Melling, B.E. Medioli, F.M. Nixon,
and F.A. McLaughlin. 2007. Origin of pingo-like features on the Beaufort Sea shelf and their possi-
ble relationship to decomposing methane gas hydrates. Geophysical Research Letters 34: L01603, 5.
doi:10.1029/2006GL027977.
Pihlainen, J.A., R.J.E. Brown, and R.F. Legget. 1956. Pingo in the Mackenzie Delta, N.W.T. Geological Society
of America Bulletin 67:1119–1122.
Porsild, A.E. 1938. Earth mounds in unglaciated Arctic northwestern America. Geographical Review 28:46–58.
Ryckborst, H. 1975. On the origin of pingos. Journal of Hydrology 26:303–314.
Sellmann, P.V., J. Brown, R.I. Lewellen, H. McKim, and C. Merry. 1975. The classification and geomorphic
implications of Thaw Lakes on the Arctic coastal plain, Alaska. U.S. Army, C.R.R.E.L., Hanover, NH.
Research Report 344.
van Everdingen, R., (ed.). 1998 (revised May 2005). Multi-language Glossary of Permafrost and Related
Ground-Ice Terms. National Snow and Ice Data Center, Boulder, CO.
Walker, D.A., M.D. Walker, K.R. Everett, and P.J. Weber. 1985. Pingos of the Prudhoe Bay region, Alaska.
Arctic and Alpine Research 17:321–336.
Walker, M.D., K.R. Everett, D.A. Walker, and P.W. Birkeland. 1996. Soil development as an indicator of rela-
tive pingo age, Northern Alaska, U.S.A. Arctic and Alpine Research 28:352–362.
Wallace, R.F. 1948. Cave-in lakes in the Nabesna, Chisana and Tanana River valleys, Eastern Alaska. Journal
of Geology 56:171–181.
Washburn, A.L. 1956. Classification of patterned ground and review of suggested origins. Bulletin of the
Geological Society of America 67:823–865.
Washburn, A.L. 1980. Permafrost features as evidence of climate change. Earth-Science Reviews 15:327–402.
20 Hybrid Mapping of Pantropical
Wetlands from Optical
Satellite Images, Hydrology,
and Geomorphology
Thomas Gumbricht
CONTENTS
Introduction..................................................................................................................................... 435
Data and Methods........................................................................................................................... 438
Study Area.................................................................................................................................. 438
Datasets...................................................................................................................................... 438
Satellite Data......................................................................................................................... 438
Climate Data.......................................................................................................................... 438
Topographic Data.................................................................................................................. 438
Note on Visualization............................................................................................................ 438
Geomorphology and Wetland Classification.............................................................................. 438
Deriving Surface Wetness from Optical Images........................................................................ 439
Surface Wetness Phenology.......................................................................................................440
Compounded Soil Surface Wetness Indices............................................................................... 441
Topographic Wetness, Water Balances, and Wetlands............................................................... 442
Global Runoff and Flooding.................................................................................................. 443
Defining Topographic Wetness Indices for Wetlands............................................................444
Wetland Topographic Convergence Indices..........................................................................444
Hybrid Mapping of Tropical and Subtropical Surface Wetness and Wetlands..........................446
Wetland Scoring....................................................................................................................446
Wetland Classification........................................................................................................... 447
Discussion.......................................................................................................................................448
Conclusion...................................................................................................................................... 452
References....................................................................................................................................... 452
INTRODUCTION
Wet land has distinctly different properties compared to dry land, including how they interact
with different wavelengths in the electromagnetic spectrum. In the visible and near-infrared
(NIR) wavelengths, wet objects absorb more strongly than dry objects and appear darker (Lobell
and Asner). In the infrared to microwave wavelengths, water absorbs strongly at some wave-
lengths and scatters at others. These characteristics of wet surfaces and water are used for map-
ping wetlands from satellite images using a variety of approaches. The most direct method is
435
to identify darker areas in optical satellite images (visible to NIR wavelengths region). Optical
sensors, however, have drawbacks for surface wetness mapping, including problems with cloud
and cloud shadows, atmospheric attenuation, vegetation perturbation, shallow penetration depth,
and poor relations found between different wetness indices and actual soil surface wetness (see
Chapter 5 for further discussion). Microwave sensors indirectly measure the surface water content
through the dielectric constant of the top soil (Nghiem et al. 2012). Passive microwave sensors,
measuring the brightness temperature (BT) of the natural background microwave emissivity, have
coarse resolutions (tens of kilometers) and wider swaths. Dependent on the design wavelengths,
atmospheric water is either transparent or the actual target for the measurements. For example,
the Tropical Rainfall Measurement Mission (TRMM) dual microwave sensors target precipita-
tion (Masunaga et al. 2002). Otherwise, most active microwave sensors, including radio detec-
tion and ranging (RADAR) sensors, are designed to avoid wavelengths that water absorbs. Most
active synthetic aperture RADAR (SAR) sensors use a side-looking antenna, and open water
bodies appear dark as the beam is not returned (backscattered) to the sensor. If, however, a wet
surface is covered by vegetation, the beam bounces between the vegetation and wet surface, and
the backscatter return signal will be stronger than if the surface would have been dry (Hess et al.
1990). Interpreting either the microwave BT or backscatter is hence nontrivial (Njoku et al. 2003;
Nghiem et al. 2012).
Blending optical data, passive microwave emissivity, and active microwave backscatter, the
Global Inundation Extent from Multi-Satellite (GIEMS) dataset is one of the most comprehen-
sive dataset of surface wetness (Prigent et al. 2007; Papa et al. 2010) (see Chapter 24 in this
book). GIEMS is a coarse-scale dataset at about 25 km spatial resolution with monthly time steps
covering almost two decades. Recently, both high-resolution SAR data (Aires et al. 2013a) and
optical images from the moderate-resolution imaging spectroradiometer (MODIS) sensor (Aires
et al. 2013b) have been used for downscaling GIEMS. Alternatively, both optical images and SAR
images can be used for mapping surface elevation: optical images by using stereo image pairs and
SAR images by timing the duration between beam transmission and return signal recording. The
surface elevation data can then be used to identify troughs with large drainage areas and indi-
rectly used for mapping surface wetness and wetlands. The overwhelming majority of land cover
maps derived from satellite imagery use statistical approaches, either adopting training datasets
(supervised classifications) or statistical clumping (unsupervised classifications) for assigning class
memberships to either single pixels or a group of pixels forming an object (Blaschke 2010). A sta-
tistical mapping approach has several advantages, including well-developed methods and relaxed
need for a physical interpretation of the (image) data. Microwave data can, for instance, be used
without any need for detailed interpretations. Any ratio or interval scale data can be included,
allowing mixing of image data from RADAR and optical sensors, and even geomorphological
and hydrological data (e.g., Bwangoy et al. 2010), for example. The main constraints of a statisti-
cal approach are the demand for an extensive set of training data or postclassification of statisti-
cal clumps. Statistical models derived for a particular scene or region usually cannot be used for
directly classifying other scenes or regions. The latter is exacerbated in tropical and subtropical
regions with strong spatial and temporal variations in both climate seasonality and vegetation phe-
nology. The intertropical convergence zone creates a pronounced gradient of precipitation follow-
ing the low-pressure zone girdling the solar zenith equatorial oscillation. This causes a mirrored
climate seasonality across the equator, with two precipitation seasons at the equator and a single
precipitation season at the tropics.
As wetlands are not randomly distributed in the landscape, a knowledge-based approach has dis-
tinct advantages over statistical approaches. But a knowledge- or expert-based classification of land
cover (e.g., wetlands) requires that the input data are physically consistent and intelligible. Wetlands
in general only develop where water input exceeds the atmospheric water demand (potential evapo-
transpiration), the geomorphology allows surface water accumulation, and the surface is actually
Hybrid Mapping of Pantropical Wetlands 437
wet or inundated for prolonged periods. A knowledge-based mapping of wetlands hence requires
data on the regional and local water balances, geomorphology, and surface wetness phenology.
Accessing such datasets disentangles the need for extensive training datasets or postclassifications
when mapping wetlands. Additionally, biophysically intelligible field data derived at ratio or inter-
val scale are excellent candidates for improving statistical classifications.
This chapter presents an attempt to map global tropical and subtropical wetlands at moderate-
scale resolution (250–500 m) using a knowledge-based approach. The hybrid wetland classification
system combines two different approaches for mapping surface wetness by (1) analyzing annual
time series of satellite-derived estimates of soil moisture and (2) modeling the topographic wetness
from a static water balance model with an annual step. These independent estimates of surface
wetness are then combined with geomorphology to derive maps of wetlands. The detailed methods
developed for acquiring the necessary background data are schematically shown in Figure 20.1.
This chapter summarizes the various background datasets and illustrates how the derived indi-
ces of surface wetness can be used together with geomorphological data for delineating wetlands
in different tropical and subtropical environments. Since the methods used are fairly complex,
they will only be briefly described here; papers that provide details are in preparation (listed as
Gumbricht 2014).
Climate data
Satellite data Topographic data
(precipitation, Input dataset
(MCD4 3A4) (SRTM)
evapotranspiration)
Eigenvector Summation of monthly Local slope and

transformation net precipitation profile curvature
Transformed
wetness
index (TWI) Cell runoff
Cell
evapotranspiration Flow routing:
Time-series
smoothing and Runoff
phenology analysis
Evapotranspiration
Topographic
Compound convergence General and
Geomorphology
TWI indexes (TCI) thematic
indexes and
classes
General and thematic

wetland maps
FIGURE 20.1 Schematic illustration of the hybrid wetland classification system. The illustration does not
show all steps and intermediate layers.
DATA AND METHODS

Study Area
The study area includes the entire global tropical and subtropical regions, from approximately
38° N to 56° S. Remote islands in the Atlantic and Pacific Oceans are not included.
Datasets
Satellite Data
The 16-day interval bidirectional reflectance distribution function (BRDF)–corrected MODIS
(MCD43A4) images were used for mapping the duration of wet and inundated soil conditions for
2011. Data from 2010 and 2012 were used to fill data gaps from the beginning and end of 2011. For
tropical regions (MODIS vertical tiles 08 and 09) with heavy cloud cover, the complete data for
these latter years were used.
Climate Data
Monthly precipitation was taken from the WorldClim global dataset (Hijmans et al. 2005), represent-
ing the average precipitation situation for approximately the period 1950–2000. Evapotranspiration
was taken from the FAO-adopted monthly reference evapotranspiration, produced by the Climate
Research Unit, University of East Anglia (New et al. 2002). The climate data were resampled to the
MODIS sinusoidal grid in 236 m spatial resolution using a bilinear interpolation.
Topographic Data
Version 4 of the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) pre-
pared by International Center for Tropical Agriculture (CIAT) and available for download in
approximately 250 m resolution was used. The absolute error of the SRTM dataset is estimated to
be 11.25 m, with relative height error for adjacent pixels estimated to be 1.6–3.3 m (Brown et al.
2005). Over open water surface, the SRTM DEM is less accurate due to the weak return signal of
the side-looking SAR antenna (Alsdorf et al. 2007). The SRTM data were resampled to the MODIS
sinusoidal grid in 236 m spatial resolution using a bilinear interpolation.
Note on Visualization
For illustration purposes, the global maps presented in this chapter have been resampled to a global
cylindrical projection at coarser scale (approximately 3000 m). While the general pattern of each
map is preserved, the maps presented here are smeared out and do not accurately represent the
original resolution maps, especially in regions with smaller-scale varying conditions and along
major rivers.
Geomorphology and Wetland Classification

Geomorphological data can assist in both mapping wetlands and interpreting wetland attributes,
including wetland class and depth. Typically ombrotrophic bogs appear as dome-shaped land-
forms; fens are bound to groundwater discharge areas; and floodplains are most commonly found
in and along wet valley floors, marshes in valleys and plains, and wet meadows as transition zones
between wetter areas and the surrounding drylands, sometimes on open slopes. Large floodout
wetlands associated with alluvial deposits in seismically active regions form the world’s larg-
est tropical wetlands. They evolved where a river with heavy bed load floods out and deposits
sediments over vast plains. If the river carries enough water, wetlands form on top of the alluvial
deposits. These wetlands resemble floodplains, but with much wider expansions. Wetlands of this
type include wetlands in the (Bajo) Gran Chaco Plains and Pantanal in South America; the Niger
N
IG
N
S
P O
GC
Hills and mountains M
Plains and valleys
0 3000 km
Water surfaces
FIGURE 20.2 Generalized global tropical and subtropical landform map identifying plains and valleys. The
boxes indicate the positions of the regional maps in Figures 20.7 through 20.11. Some of the major alluvial
deposits harboring floodout wetlands are indicated as letter codes: GC, Gran Chaco; IG, Indo-Gangetic Plains;
M, Macquarie; N, Niger Inland Delta; O, Okavango; P, Pantanal; S, Sudd.
Inland Delta, the Sudd on the White Nile, and the Okavango Delta in Africa; wetlands in the Indo-
Gangetic (or North Indian River) Plain in Asia; and the Macquarie Marshes in eastern Australia
(Figure 20.2).
One problem with geomorphological data is that landforms are usually defined for local or
regional conditions, including lithological and vegetation classes (e.g., Ballantine et al. 2005). This
is difficult to generalize globally, so a more generic approach is needed for defining global geo-
morphology in support of mapping wetlands. Elsewhere I will explain the development of a series
of global landform maps derived from the SRTM DEM and hydrological data (Gumbricht 2014a).
Existing landform maps have made use of both multiscale approaches and combinations of field-
specific and region-specific variables (Shary et al. 2002). The aim of the landform mapping was to
create scale and region neutral maps, relevant for mapping hydrological conditions in general and
wetlands in particular. Eight different types of landform maps were developed, for (1) hydrological
terrain relief (three versions), (2) channel valleys (valleys with flowing channels), (3) valleys (two
versions, for smaller and larger valleys regardless of the existence of flowing channels), (4) gen-
eral geomorphology (Weiss 2001), (5) morphology of plains and planar surfaces, (6) dome-shaped
features, (7) valley-shaped features, and (8) a more generalized map of global plains combining
maps 2–7. The hydrological terrain relief maps estimated the absolute elevation difference between
actively flowing channels (see Section, Global Runoff and Flooding) of various sizes and the sur-
rounding terrain. The maps of dome-shaped features, valley-shaped features, and morphology of
plains and planar surfaces were derived by analyzing multiscale profile and cross curvatures (Wood
1996) and used as a supplement for mapping various thematic wetland classes. The most important
map for general wetland mapping is the generalized map of global plains (Figure 20.2). In principle,
the majority of tropical and subtropical wetlands are bound to areas defined as plains. The main
exceptions are wet meadows on steeper slopes, steeper sides of peat domes, and depressional wet-
lands in mountains.
Deriving Surface Wetness from Optical Images

With recent strong developments in mapping both surface wetness and wetlands from microwave BT
and backscatter (Chapter 5), less attention has been paid to using optical satellite images. However,
optical data have the advantage of higher spatial and temporal resolutions and a more developed
library of processing routines. They are also more freely available and easier to interpret.
The transformed wetness index (TWI) is a generic surface wetness index, developed for mul-
tispectral optical image data (Gumbricht 2014b). TWI uses an eigenvector transformation of the
original image bands and then adopts a nonlinear normalized difference approach using two of
the transformed vectors for mapping surface wetness. The eigenvector transformation resembles a
principal component analysis (PCA) but forces the eigenvectors to represent biophysical features,
or spectral end-members. The spectral end-members for mapping wetlands include dark soil, light
soil, photosynthetic vegetation (PV), nonphotosynthetic vegetation (NPV), and water. These can
either be drawn from a spectral library or from the image(s) under study. For mapping global wet-
lands, end-members were identified using an automated algorithm extracting spectral signals from
annual time series of MCD43A4, except for the spectral end-member for NPV that was drawn from
a spectral library. From the ensemble of spectral signals derived from each tropical MODIS tile,
an optimized set of spectral end-members for mapping wetlands were extracted. The optimization
was done as a Monte Carlo simulation evaluating the separability of wetlands and nonwetlands and
allowing the spectral end-members to vary within one standard deviation of the global tropical
ensemble mean.
The eigenvector conversion of the spectral end-members to a biophysical feature space closely
resembles the tasseled-cap transformation first suggested for Landsat MSS data (Kauth and
Thomas 1976) and then developed for Landsat TM data (Crist and Cicone 1984; Huang et al.
2002) and other sensors (Yarbrough et al. 2005; Lobser and Cohen 2007). Compared to the
traditional tasseled-cap transformations, the transformation for mapping wetlands uses the dark
soil as an offset, letting the first tasseled-cap component (brightness) represent the soil line (Baret
et al. 1993).
Reflection from bare soil in the visible to NIR wavelength region depends on mineral compo-
sition, soil structure, and organic content, but with the water content being the dominant factor
(Muller and Décamps 2000; Lobell and Asner 2002). Some of the information on surface wetness is
hence retained in the brightness (i.e., soil line) component and some in the wetness (water) compo-
nent. Using the transformed soil-adjusted vegetation index (TSAVI) (Baret et al. 1989) as a starting
point, TWI is defined using brightness and wetness as input bands:
⎡⎣β1 ∗ ( Wetness − β1 ∗ Brightness − β0 ) ⎤⎦

TWI =
⎡⎣ Brightness + β1 ∗ Wetness − β1 ∗ β0 + C (1 + β1 ∗ β1 ) ⎤⎦

where
β 0 and β1 define the intercept and slope of the iso-wetness line for the average soil reflectance,
with β 0 set to −2080 and β1 set to 1.6
C is a noise reduction parameter set to 0.7 through optimization
The TWI results presented here are derived from a rotation of the TWI algorithm to fit the spectral
end-members and are rescaled to values between −3000 and 3000 (see Gumbricht 2014b for details).
A value of 3000 represents dark open water. But due to floating debris, algal blooms, waves,
shallow depth, and other factors, in practice, a TWI value of about 2000 or more indicates an open
water surface. Negative values represent drier conditions, with a value of −3000 indicating desert
conditions. Figure 20.3 illustrates TWI mean surface wetness for 2011, calculated from annual time
series of 16-day interval MODIS BRDF composites (MCD43A4).
Surface Wetness Phenology

The mean annual surface wetness is an indicator of the existence of wetlands, but the phenology
of surface wetness and in particular the duration of wet and inundated conditions are potentially
stronger indicators. The phenological analysis of TWI first fills data gaps using temporal inverse
distance weighting. The reliability of each TWI recording is then evaluated using a local regres-
sion comparing individual values with adjacent dates. Filled data points are given lower weights in
Open water
1000 (inundated)
0 (intermediate)
0 3000 km
–3000 (dry)
FIGURE 20.3 Global tropical and subtropical average TWI for 2011. The index is rescaled to values between
−3000 (dry) and 3000 (deep open water), with a value of 1000 theoretically marking the boundary between
very wet soil conditions and inundation. A neutral value (TWI = 0) indicates medium wet conditions.
the local regression. Nonconforming values are omitted, and the phenology is calculated from the
regression smoothed data (see Gumbricht 2014c for details). Apart from primary statistical mea-
sures, the onsets and offsets and lengths of periods of wet soil conditions (TWI > 0) and surface
inundation (TWI > 1000) are extracted.
Compounded Soil Surface Wetness Indices

For a globally generic mapping effort, the most important surface wetness phenology components
include
• TWI minimum (pTWImin): driest annual conditions

• TWI maximum (pTWImax): wettest annual conditions
• TWI mean (pTWImean): annual mean wetness conditions
• TWI length of inundated season(s) (pTWIlfs): number of days with inundated conditions
• TWI length of wet season(s) (pTWIlws): number of days with wet conditions
These phenological indices are originally in 472 m spatial resolution, with some data gaps in wet
regions with heavy cloud cover. The TWI-derived phenology for the year 2011 was resampled to a
spatial resolution of 236 m to fit the other data used in this study. Cells with no recorded phenology
over land were filled by interpolation from adjacent cells with valid recordings.
The details for the extraction of compound indices for the soil surface wetness (cTWI) for vari-
ous thematic wetland classes will be presented elsewhere (Gumbricht 2014d). The compound index
assigns weights to different components of the wetness phenology, with, for example, floodout wet-
lands (cTWIfloodout) having higher weight for the duration of inundation. For soil humidity and gen-
eral wetland mapping, the inundation period is not used; instead the length of the period(s) with wet
soil conditions is given a higher weight:
cTWI gen = 128 + 15∗ ∑ ( pTWI ∗ W )

i i

where
cTWIgen is the general surface wetness compound index
pTWIi is the phenological component i
Wi is the weights of the phenological component i
Open water
125
0 3000 km
0 (dry)
FIGURE 20.4 Global tropical and subtropical cTWI for 2011. cTWI represents the annual wetness consid-
ering annual minimum, maximum, and mean wetness conditions and the length of the period with wet soil
conditions (TWI > 0).
The components and weights used for defining cTWIgen include pTWIlws * 0.00825, pTWImin * 0.0005,
pTWImax * 0.001, pTWImean * 0.001, and pTWIrange * 0.001, where range is defined as pTWImin −
pTWImean and hence a negative number.
The distribution of surface wetness in the tropics and subtropics derived using the compound
index is shown in Figure 20.4.
Topographic Wetness, Water Balances, and Wetlands

In the hydrological sciences as well as in many land surface models (LSMs), wetlands are usually
not mapped from satellite images but from topography and the water balance and estimated depth
to the groundwater table. The most popular approach has been to adopt the topographic index or
topographic convergence index (TCI) (Beven and Kirkby 1979; Quinn and Beven 1993; Quinn
et al. 1995):
⎛ A /b ⎞
I = ln ⎜ ⎟
⎝ tan β ⎠
where
A is the upstream catchment area
b is the contour length
β is the local slope steepness in degrees
TCI was initially developed for a dynamic rainfall–runoff model, TOPMODEL (Beven and Kirkby
1979). Originally, TOPMODEL and TCI are based on at least the following assumptions: (1) Rainfall
is equal over the drainage area, (2) the slope equals the hydraulic gradient of the groundwater table,
(3) the water transmissivity increases exponentially toward the soil surface and is equal over the
drainage area, (4) soil wetness can be represented as a series of steady-state conditions, and (5) there
is no unsaturated lateral flow. In tropical and subtropical climates, assumptions (1) and (3) are usu-
ally less accurate. Over terrain with large topographic relief and basins affected by the intertropi-
cal convergence zone, the precipitation variation can be very large, and in more deeply weathered
soils, the transmissivity feedback is not pronounced as in temperate soils (formed by glacial pro-
cesses). To overcome the problem with variations in precipitation, Merot et al. (2003) exchanged the
upstream catchment area with water volume (total water flow or runoff). Other studies suggest that
exchanging the local slope for profile curvature or slope angle toward the nearest drainage improves
TCI estimates of surface wetness (e.g., Hjerdt 2004).
Many studies have used various versions of the TCI for directly mapping wetlands in primar-
ily temperate climates (e.g., Rodhe and Seibert 1999; Merot et al. 2003; Curie et al. 2007). Both
TOPMODEL and variants of the topographic wetness index are used for mapping wetlands in both
LSMs and global hydrological models (GHMs). In general, LSMs are more physically based and
focus on ecosystems (vegetation), carbon, and vertical energy balance at the land surface, whereas
GHMs use more empirical formulations for the energy balance and focus on lateral water flow.
Haddeland et al. (2011) compared 6 LSMs and 5 GHMs and found that model predictions of the
global terrestrial water cycle are highly variable and a major cause of uncertainty. Ringeval et al.
(2012) used TOPMODEL for downscaling 1° resolution LSM data to 1 km, using the HYDRO1K
dataset. They compared their results to gauged runoff and the GIEMS dataset and concluded that
the prediction of saturated areas was difficult at the detailed scale. Wania et al. (2012) summarized
10 global models and compared their prediction of wetlands and wetland methane balances. Also
their results showed large intermodel variations.
The shortcomings of the global efforts in mapping runoff and wetlands from LSM and GHM
can be attributed to several factors. The coarse climate datasets used do not capture the real spa-
tial variability, the DEM (usually derivatives of the GTOPO30 dataset [Gesch et al. 1999]) is not
very accurate in its details, and some of the assumptions underlying both TOPMODEL and TCI
are not valid for tropical and subtropical environments. Additionally, bogs fed mainly by direct
precipitation are not dependent on horizontal water inflow. Predicting the distribution of bogs
must be based on the local climatic water balance and not on topographically induced lateral flow
(cf. Kirkby et al. 1995).
Global Runoff and Flooding

A prerequisite for adopting a hydrological approach for mapping tropical and subtropical wetlands
includes global consistent datasets on both runoff and flooding. As noted earlier, existing global
water balance models estimate runoff at coarse scales (see Haddeland et al. 2011). Runoff estimates
at moderate scales can be achieved by either downscaling coarser models (Ringeval et al. 2012)
or defining a water balance model that can directly use higher spatial resolution data. The water
balance model that uses the SRTM DEM at 250 m spatial resolution for modeling global tropi-
cal and subtropical water balances with an annual time step will be described in detail elsewhere
(Gumbricht 2014e). The model’s basic concept is that surface wetness, and thus cell evapotranspira-
tion, can be estimated from the cell vertical water balance (precipitation minus evapotranspiration),
water inflow from upstream areas, and cell local slope steepness and profile curvature. After esti-
mating the summed monthly vertical water balance at each cell (Rcellyr), runoff and evapotranspira-
tion are separately routed over the DEM. Remaining evaporative power is filled up dependent on
runoff estimates and local topography. The actual runoff (Raccyr) is then calculated as the difference
between the unabated runoff and the actual evapotranspiration for each cell. Evaporative power
remaining after the flow routing is calculated (ETRcellyr) and later used for determining whether
the water balance can support wetlands. Compared to existing GHMs and LSMs, the water bal-
ance is directly calculated at 250 m spatial resolution using a higher-quality DEM and a multiple
flow direction algorithm that forces flow out of pits (Ehlschlaeger 1989). There is no transmissivity
feedback function, and the profile curvature is used in addition to the slope steepness for estimating
surface wetness.
The water flow routing over the DEM confines flow to channels while flooded areas, includ-
ing wetlands adjacent to a defined channel, will be recorded as nonwet areas. To also identify
flood-prone valleys and plains, the hydrological terrain relief of cells adjacent to channels with
sustained water flow (as identified from the water balance model) is analyzed (Gumbricht 2014f).
The flooding model for valley-confined wetlands (e.g., floodplains and riverine/lacustrine wet-
lands) is different compared to the flooding model for extended floodout wetlands on alluvial
deposits. In the former geomorphological setting, a distance decay function more rapidly exhausts
the flood volumes (Faccyr). This generates a more realistic estimate of annual flooding volumes,
100,000
1,000
1,000
100
10
1 m3/s
0.1
0.01
0 3000 km 0.001
0.0001
0
FIGURE 20.5 Global tropical and subtropical water flow (m3/s) using monthly precipitation and evapotrans-
piration data compiled to an annual time step and routing over the SRTM DEM.
but floodout rivers spilling out over vast alluvial plains are not captured. To also capture these
wetlands, an annual floodout volume estimation (FOaccyr) with a very weak distance decay func-
tion is used. The latter model, however, also floods many regions that are not actually flooded. To
separate alluvial deposits hosting floodout wetlands from other alluvial deposits, the estimated
duration of inundation derived from the phenological analysis of TWI was used. Figure 20.5
illustrates the combined runoff and valley flooding map (RFaccyr = Raccyr + Faccyr) for the global
tropics and subtropics.
Defining Topographic Wetness Indices for Wetlands

Fens are wetlands mainly fed by (often nutrient-rich) groundwater and restricted to hydrological
discharge areas. To separate fens from other types of wetlands and correctly identify their wetness
conditions, a simplified estimate of groundwater flow was developed, separating total water flow
(RFaccyr) into surface (RFSaccyr) and groundwater (GRWaccyr) components by using the extended
profile curvature (see Gumbricht 2014e). Also many other wetland types are bound to discharge
areas in lower slope positions or valley bottoms and need to consider both groundwater flow and
surface runoff. For riverine or lacustrine wetlands, it is the flow vector of the adjacent water chan-
nel that mainly feeds the wetland. Marshes and wet meadows can develop over extended plains
and open slopes and are supported by precipitation, surface flow, and sometimes groundwater flow.
Floodout wetlands are surface water driven, with groundwater being of negligible importance.
Ombrotrophic bogs are fed solely by precipitation and bound to regions where precipitation largely
exceeds evapotranspiration.
To account for the large variation in tropical and subtropical wetlands, several modifications
of the original TCI were implemented for mapping wetlands: (1) inclusion of local (cell) vertical
water balances, (2) upstream water flow estimations derived from a spatially resolved water bal-
ance model, (3) identification of flood-prone areas and estimation of flood volumes, (4) separation
of lateral groundwater flow and surface water flow, (5) definition of local topographic condi-
tions including profile curvatures, and (6) consideration of full atmospheric evapotranspiration
potential.
Wetland Topographic Convergence Indices

Using the aforementioned modifications of the original TCI, a set of novel TCIs were derived.
Rather than using the cell contour length for normalizing inflow, a set of scale-dependent constants
were used for weighting the relative importance of the inflow components, with surface flow hav-
ing the lowest weight (regarded as a flow vector), local runoff generation an intermediate weight
(regarded as a field), and groundwater flow the highest weight. The general wetland topographic
convergence index (WTCI) (WTCIgen) estimates soil moisture after fully satisfying the atmospheric
water demand (ETRcellyr):
⎡ ( RFSaccyr /12 ) + ( Rcellyr / 2 ) + ( GRWaccyr /1) + ( Tcellyr / 3 ) − ETRcellyr

( ) ⎥⎤
WTCI gen = ln ⎢
⎢
⎣
( tan* Curvfac ) ⎥
⎦

where
RFSaccyr, Rcellyr, GRWaccyr are inflow components described
Tcellyr is the annual net transpiration (water used up by vegetation and estimated as 50% of the
actual evapotranspiration)
ETRcellyr is the water balance net remaining evapotranspiration potential (from the water
balance model) and Curvfac the profile curvature factor, calculated separately for concave
terrain (negative profile curvature)
Slope
Curvfac =

(1 − Profsum )
and for convex terrain (positive profile curvature)
Curvfac = Slope ∗ ( Profsum + 1)

with Profsum defined as the sum of profile curvatures (Wood 1996) at scales representing 250, 500,
750, and 1500 m (i):
Profsum = Σ (profci)
WTCIgen generates valid values only in positions with a positive water balance (total water inflow
exceeding atmospheric water demands) and is a general index for the occurrence of wetlands
(Figure 20.6). By omitting inflowing water balance components from WTCIgen, different WTCI
can easily be constructed for various wetland classes. By, for instance, omitting surface water
Open water
125
0 3000 km
0
FIGURE 20.6 Combined WTCI for general wetlands and floodout wetlands (WTCIwetflood) portraying the
annual surface wetness from runoff and flooding data combined with local slope steepness and topographic
profile curvatures.
inflow (RFSaccyr), the wetness balance for fens can be calculated. Elsewhere I will present thematic
WTCIs for fens, riverine wetlands, floodplains, floodout wetlands, and bogs (Gumbricht 2014g).
The WTCI for floodout wetlands differs from the other thematic WTCI. To account for the more
widespread inundation from floodout rivers, it uses the estimated floodout volume over plains (see
previous text):
⎡ ( FOaccyr /18 ) + ( Rcellyr /1) − ETRcellyr

( ) ⎥⎤
WTCI floddout = ln ⎢
⎢
⎣
( tan10 ∗ Cuurvfac ) ⎥
⎦

The WTCIfloodout was calibrated to fit WTCIgen at overlapping sites with known floodout wetlands.
The calibration was done by changing the weighting of the flood volume (divisor set to 18) and a
faster runoff by forcing the local slope curvature to 10°. The latter can be seen as an adjustment for
overland flow versus groundwater flow.
A combined WTCI (WTCIwetflood), valid for both valley-confined wetlands and floodout wetlands,
was then constructed by merging WTCIgen and WTCIfloodout, using the recorded length of the wet
seasons (pTWIlws) for determining which index to apply at any particular position:
WTCIwetflood = max[WTCIgen, WTCIfloodout]; if pTWIlws ≥ 90
WTCIwetflood = WTCIgen; if pTWIlws < 90
Hybrid Mapping of Tropical and Subtropical Surface Wetness and Wetlands

The compounded transformed wetness index (cTWI) derived from the time series of optical satel-
lite images and the WTCI derived from hydrological modeling both capture the wetness of the land
surface, the former by estimating per-pixel wetness using annual time series of the satellite-based
TWI and the latter by regional and pixel-wise analyses of water balances and local topography.
WTCI and cTWI are more complementary than redundant. TWI, for instance, has high values in
humid tropical forests on steep slopes, whereas TCI will portray the landscape wetness as unaf-
fected by human management (e.g., drained areas in both fields and urban areas, and excavated
wetlands will be estimated to be wet). Combining cTWI and WTCI eliminates some of the com-
mission errors compared to using them singularly. Gumbricht (2014d) uses wetland thematic indi-
ces of cTWI and TWCI for mapping different wetland classes. The following text presents a general
(global) classification aiming at capturing wetlands by combining cTWIgen and WTCIwetflood with
geomorphological data.
Wetland Scoring
The hybrid scoring for wetland mapping (Wetlandscore) is formulated as
WTCI wetflood + cTWI gen − 200

Wetlandscore =
2
cTWI values above 200 indicate open water, and WTCI values above 200 only appear in the largest
of the world’s river (e.g., the Amazon River). Only open water surfaces hence receive scores above
100, simultaneously allowing the capturing of marshes in the lower ranges of the score values. Score
values above approximately 50 indicate inundated wetlands. Score values of 0 are regarded as areas
with no wetlands and include all areas with a negative water balance (WTCI = 0).
Wetland Classification
Wetlands can be delineated by simply thresholding the Wetlandscore, but geomorphological con-
straints enhance the accuracy. Thresholds and constraints can be set globally, but regional and/or
local adjustments improve the classification (see Gumbricht 2014d).
The results of the hybrid classification are shown in a series of maps: one for the worldwide distri-
bution of wetlands in plains and valleys of the tropics and subtropics (Figure 20.7) and others for major
wetland regions in this zone—Amazon region (Figure 20.8), Central Africa (Figure 20.9), Southern
Asia, Indochina and China (Figure 20.10), and Southeast Asia (Figure 20.11). Figures 20.12 through
20.14 present more detailed maps over selected areas in the Amazon and Central Africa (i.e., the areas
within the boxes in Figures 20.8 and 20.9). The detailed maps for the Amazon and the Congo (Figures
20.12 and 20.13) include overlays showing the results from earlier studies of wetland extent (Hess et al.
2012 for the Amazon and Bwangoy et al. 2010 for the Congo). Note that all these maps restrict wetland
100
50
0 3000 km
0
FIGURE 20.7 Global wetland scores derived from the hybrid wetland classification system for tropical and
subtropical plains and valleys. (Note: Other global landforms were not evaluated.) The scores are derived by
combining satellite-derived estimates of surface wetness (cTWI) and runoff and flooding estimates (WTCI).
Areas with a negative water balance are scored as no wetland.
100
50
0 1000 km
FIGURE 20.8 Wetland scores derived from the hybrid wetland classification system for the Amazon region.
The box indicates the location of the detailed map presented in Figure 20.12.
100
50
0 1000 km
FIGURE 20.9 Wetland scores derived from the wetland hybrid classification system for the Central Africa
region. The boxes indicate the positions of the detailed maps presented in Figures 20.13 and 20.14.
100
50
0 1000 km 0
FIGURE 20.10 Wetland scores derived from the wetland hybrid classification system for parts of mainland
Asia.
scoring to areas identified as geomorphological plains and valleys (Figure 20.2) as analysis of other
landforms was not attempted. Consequent wetlands are not shown for other landforms.
DISCUSSION
Moderate-scale wetland mapping at a global scale is hindered by a lack of suitable and consistent
datasets. The hybrid wetland classification system is an attempt to bridge some of the shortcomings,
but is hampered by quality problems in individual source datasets.
100
50
0
0 1000 km
FIGURE 20.11 Wetland scores derived from the wetland hybrid classification system for Southeast Asia.
The TWI overcomes some of the problems that have hitherto prevented the adoption of optical
satellite images for mapping surface wetness. The MODIS MCD43A4 data product is cleaned for
atmospheric attenuation as well as for clouds and cloud shadows. Remaining quality problems are
partly solved by a time-series analysis using a local regression for identifying and removing non-
consistent data recordings.
The TWI fails to capture the wet surface covered by the relatively drier biomass over very densely
vegetated wetlands (e.g., papyrus swamps). For tropical regions with heavy cloud cover, the number
of scenes with accurate data is too low for estimating of the surface wetness phenology. This was
tackled by combining 3 years of data (2010–2012). But the actual wetness cycle in those years might
have been different. If so, the data cleaning procedure will find fewer relevant data points to use. But
a multiyear approach has other advantages, not the least of which is better capture of the average
situation. It is more likely that wet (cloudy) season data are missing in the time series, causing an
underestimate of surface wetness.
The climate datasets used for estimating runoff are not consistent over some regions, and
local and regional (e.g., topographic) calibration of either the climate data or the model formula-
tions for mapping wetlands is needed for improved wetland delineation. The SRTM DEM has
apparent errors, particularly for water surfaces of wider rivers, but also for other regions. This
causes errors both in the hydrological modeling and in the geomorphological mapping. These
errors are propagated and affect the hybrid classification of wetlands. This is clearly illustrated
in Figure 20.14, where the subtle variation in topography is not properly captured by the SRTM
DEM. Water flow routing alone can thus not be used for identifying surface wetness and forced
a separate estimation of floodout volumes. Whether or not to include the floodout estimates is a
question of regional calibration and demands knowledge of geomorphology and genesis of plains
and valleys in different regions. Similarly, moderate-scale DEMs do not capture anthropogenic
drainage systems well, causing overestimates of flooded areas in, for example, agricultural and
urban areas.
The hybrid classification system is at its core a deterministic approach grounded in biophysical
characterizations of surface wetness, hydrology, and geomorphology. The deterministic approach
can easily be translated to different spatial resolutions and be used for estimating both historical
and future situations. The system is not restricted to using MODIS-derived surface wetness or to
100
50
0
0 100 km
FIGURE 20.12 Wetland scores derived from the wetland hybrid classification system for the Pastaza-
Marañón Basin in the Peruvian Amazon, wetland mask in black derived from Hess et al. (2012). The location
of this area is indicated in Figure 20.8.
statistical estimates of climate variables. Other data sources, including soil moisture information
derived from microwave BT and/or backscatter data, can be used instead of the MODIS data. The
TWI is potentially an attractive substitute for global downscaling (e.g., GIEMS) for more detailed
tropical and subtropical wetland classifications. In the future microwave data at spatial and tempo-
ral resolutions, closer matching of the MODIS products will be a more ideal data source. The Soil
Moisture Active Passive (SMAP, smap.jpl.nasa.gov) and the Surface Water and Ocean Topography
(SWOT, swot.jpl.nasa.gov) missions will close this data gap within the next years.
The only alteration needed for directly adopting other optical sensors for deriving TWI is a
reformulation of the spectral end-members and the eigenvector transformations. Adopting (more
sparse) time series of higher-resolution satellite images, including SAR imagery and images from
the Landsat and SPOT programs, and Advanced Spaceborne Thermal Emission and Reflection
Radiometer (ASTER) for downscaling full annual time series of MODIS-derived surface wetness
estimates to 30 m spatial resolution is possible. Combined with higher-resolution SRTM data, or
DEM data derived from ASTER stereo pairs, this could potentially allow a hybrid classification of
global wetlands at 30 m spatial resolutions.
Exchanging the statistical climate data for actual precipitation data derived, for example, from
the TRMM, would allow annually adjusted estimates of also the WTCI. It is also possible to use
FIGURE 20.13 Wetland scores derived from the wetland hybrid classification system for the Central Congo
Basin, wetland mask in black derived from Bwangoy et al. (2010). The location of this area is indicated in
Figure 20.9.
N N
100 100
50 50
0 100 km 0 100 km
0 0
(a) (b)
FIGURE 20.14 Wetland scores derived from the wetland hybrid classification system for the Okavango
Delta in Botswana, wetland mask in black derived from McCarthy et al. (2003). The wetland mask is divided
into four regions: (1) the tectonically bound entry channel, (2) the central permanent wetlands, (3) the season-
ally inundated wetlands, and (4) the most distal areas only temporally flooded and lacking permanent wet-
lands. The panels show the wetland scoring (a) regarding floodout water and (b) disregarding floodout. The
location of this area is indicated in Figure 20.9.
scenarios for predicted future climates and hence derive estimates of future surface wetness condi-
tions from the WTCI.
The humidity and wetland maps are derived from more than 30 different input layers, all at a
spatial resolution of 250–500 m. Almost all of these layers are physically intelligible, with scales
allowing thresholding and statistical evaluations. The relative importance of the different layers
for mapping wetlands has not been validated, but is likely to vary. The input layers are potential
candidates for improved global mapping not only of wetlands, but also for soil classifications, map-
ping of soil organic content, and vegetation mapping.
CONCLUSION
The hybrid approach for mapping global tropical and subtropical wetlands combines two different
methods to estimate surface wetness at a moderate spatial resolution and then uses geomorphologi-
cal constraints for identifying wetlands. The hybrid model does not depend on reference data, but is
rather an expert approach. Results are consistent but depend on the quality of the input data. Local
and regional calibration relating to both climatic and geomorphological conditions can be used to
improve the classification accuracy. Compared to earlier attempts in mapping global wetlands, the
spatial resolution is higher. The TWI improves the capturing of relevant information on surface wet-
ness from optical image data. TWI can be adopted for any optical sensor, but the use of the MODIS
MCD43A4 product reduces problems with atmospheric attenuation, clouds, and cloud shadows.
The accessibility of MODIS BRDF data and the adoption of well-documented satellite image pro-
cessing algorithms were prerequisites for the generation of global tropical and subtropical maps of
surface wetness. Many of the presented maps portray biophysical properties of the land surfaces
at ratio or interval scales and are prime candidates to use both for downscaling coarser but more
accurate estimates of, for example, surface wetness derived from microwave data and for direct use
as independent variables in statistical mapping approaches of various land surface characteristics,
including wetlands. Finally, some of the inaccuracies in the models of surface wetness are related to
quality problems propagated from the original source data.
ACKNOWLEDGMENTS
The wetland mapping was partly done in cooperation with the Center for International Forestry
Research (CIFOR), with support from the U.S. Agency for International Development (USAID).
The development of global models for geomorphology and hydrology was partly done by support
from the World Agroforestry Centre (ICRAF).
REFERENCES
Aires, F., F. Papa, and C. Prigent. 2013a. A long-term, high-resolution wetland dataset over the Amazon
Basin, downscaled from a multiwavelength retrieval using SAR Data. Journal of Hydrometeorology
14(2):594–607.
Aires, F., F. Papa, C. Prigent, J.-F. Crétaux, and M. Berge-Nguyen. 2013b. Characterization and space/time
downscaling of the inundation extent over the Inner Niger Delta using GIEMS and MODIS data. Journal
of Hydrometeorology 15:171–192.
Alsdorf, D.E., E. Rodríguez, and D.P. Lettenmaier. 2007. Measuring surface water from space. Reviews of
Geophysics 45(2):RG2002.
Ballantine, J., G. Okin, D. Prentiss, and D. Roberts. 2005. Mapping North African landforms using continental
scale unmixing of MODIS imagery. Remote Sensing of the Environment 97(4):470–483.
Baret, F., G. Goyot, and D. Major. 1989. TSAVI: A vegetation index which minimizes soil brightness effects
on LAI and APAR estimation. In: 12th Canadian Symposium on Remote Sensing and IGARSS’90, Vol. 4,
Vancouver, British Columbia, Canada: Geoscience and Remote Sensing Society of Institute of Electrical
and Electronics Engineers, IEEE, Piscataway, NJ, pp. 1355–1359.
Baret, F., S. Jacquemoud, and J.F. Hanocq. 1993. The soil line concept in remote sensing. Remote Sensing
Reviews 7(1):65–82.
Hydrological Sciences Bulletin 24(1):43–69.
Blaschke, T. 2010. Object based image analysis for remote sensing. ISPRS Journal of Photogrammetry and
Brown, C.G., K. Sarabandi, and L.E. Pierce. 2005. Validation of the Shuttle Radar Topography Mission height
data. IEEE Transactions on Geoscience and Remote Sensing 43(8):1707–1715.
Bwangoy, J.-R.B., M.C. Hansen, D.P. Roy, G. De Grandi, and C.O. Justice. 2010. Wetland mapping in the
Crist, E.P. and R.C. Cicone. 1984. A physically-based transformation of thematic mapper data - the TM Tasseled
Cap. IEEE Transactions on Geoscience and Remote Sensing GE-22:256–263.
Curie, F., S. Gaillard, A. Ducharne, and H. Bendjoudi. 2007. Geomorphological methods to characterise wet-
lands at the scale of the Seine Watershed. The Science of the Total Environment 375(1–3):59–68.
Ehlschlaeger, C. 1989. Using the AT search algorithm to develop hydrologic models from digital elevation data.
In: Proceedings of International Geographic Information Systems (IGIS) Symposium’89, Baltimore,
MD, pp. 275–281.
Gesch, D.B., K.L. Verdin, and S.K. Greenlee. 1999. New land surface digital elevation model covers the Earth.
Eos, Transactions American Geophysical Union 80(6):69–70.
Gumbricht, T. 2014a. Hydromorphological maps of global tropical and sub-tropical landforms. Manuscript.
Gumbricht, T. 2014b. Retrieving Soil Surface Wetness from Optical Satellite Images: Part 1, Model Formulation
and Validation. Submitted to Remote Sensing.
Gumbricht, T. 2014c. Retrieving Soil Surface Wetness from Optical Satellite Images: Part 3, Extracting Wetness
Phenology. Submitted to Remote Sensing.
Gumbricht, T. 2014d. Knowledge based thematic mapping of global tropical and sub-tropical wetlands and
peatlands from surface wetness phenology and hydromorphology. In preparation.
Gumbricht, T. 2014e. A moderate scale global tropical and sub-tropical geometric water balance model.
Manuscript.
Gumbricht, T. 2014f. Estimation of moderate scale global tropical and sub-tropical flooding from runoff and
digital elevation models. Manuscript.
Gumbricht, T. 2014g. Topographic wetness indexes for tropical and sub-tropical ecosystems and wetlands.
Manuscript.
Haddeland, I., D.B. Clark, W. Franssen, F. Ludwig, F. Voß, N.W. Arnell, N. Bertrand et al. 2011. Multimodel
estimate of the global terrestrial water balance: Setup and first results. Journal of Hydrometeorology
12(5):869–884.
Hess, L.L., J.M. Melack, E.M.L. Novo, C.C.F. Barbosa, and M. Gastil. 2012. LBA-ECO LC-07 JERS-1
SAR Wetlands Masks and Land Cover, Amazon Basin: 1995–1996. Oak Ridge, TN. doi:http://dx.doi.
org/10.3334/ORNLDAAC/1079.
Hess, L.L., J.M. Melack, and D.S. Simonett. 1990. Radar detection of flooding beneath the forest canopy: A
review. International Journal of Remote Sensing 11(7):1313–1325.
Hijmans, R.J., S.E. Cameron, J.L. Parra, P.G. Jones, and A. Jarvis. 2005. Very high resolution interpolated cli-
mate surfaces for global land areas. International Journal of Climatology 25(15):1965–1978.
Hjerdt, K.N. 2004. A new topographic index to quantify downslope controls on local drainage. Water Resources
Research 40(5):1–6.
Huang, C., B. Wylie, L. Yang, C. Homer, and G. Zylstra. 2002. Derivation of a Tasseled Cap Transformation
Based on Landsat 7 at-Satellite Reflectance. Raytheon ITSS, USGS EROS Data Center, Sioux Falls, SD.
Kauth, R.J. and G.S. Thomas. 1976. The Tasseled Cap—A graphic description of the spectral-temporal develop-
ment of agricultural crops as seen by Landsat. In: Proceedings of the Symposium on Machine Processing
of Remotely Sensed Data, LARS, Purdue University, West Lafayette, IN, pp. 4B41–4B51.
Kirkby, M.J., P.E. Kneale, S.L. Lewis, and R.T. Smith. 1995. Modelling the form and distribution of peat mires.
In: A.L. Hugues and M.R., Heathwaite (eds.). Hydrology and hydrochemistry of British Wetlands, Wiley,
New York, pp. 83–93.
Lobell, D.B. and G.P. Asner. 2002. Moisture effects on soil reflectance. Soil Science Society of America Journal
66(3):722–727.
Lobser, S.E. and W. Cohen. 2007. MODIS Tasselled Cap: Land cover characteristics expressed through trans-
formed MODIS data. International Journal of Remote Sensing 28(21–22):5079–5101.
Masunaga, H., T. Iguchi, R. Oki, and M. Kachi. 2002. Comparison of rainfall products derived from TRMM
microwave imager and precipitation radar. Journal of Applied Meteorology 41(8):849–862.
McCarthy, J.M., T. Gumbricht, T. McCarthy, P. Frost, K. Wessels, and F. Seidel. 2003. Flooding patterns of the
Okavango Wetland in Botswana between 1972 and 2000. AMBIO 32(7):453–457.
Merot, P., H. Squividant, P. Aurousseau, M. Hefting, T. Burt, V. Maitre, M. Kruk, A. Butturini, C. Thenail,
and V. Viaud. 2003. Testing a climato-topographic index for predicting wetlands distribution along a
European climate gradient. Ecological Modelling 163(1–2):51–71.
Muller, E. and H. Décamps. 2000. Modeling soil moisture-reflectance. Remote Sensing of the Environment
76:173–180.
New, M., D. Lister, M. Hulme, and I. Makin. 2002. A high-resolution data set of surface climate over global
land areas. Climate Research 21:1–25.
Nghiem, S.V., B.D. Wardlow, D. Allured, M.D. Svoboda, D. LeComte, M. Rosencrans, S.K. Chan, and
G. Neumann. 2012. Microwave remote sensing of soil moisture, science and applications. In: B.D.
Wardlow, M.C. Anderson, and J.P. Verdin (eds.). Remote Sensing of Drought. CRC Press, Taylor &
Francis, Boca Raton, FL, pp. 197–226.
Njoku, E.G., T.J. Jackson, V. Lakshmi, T.K. Chan, and S.V. Nghiem. 2003. Soil moisture retrieval from
AMSR-E. IEEE Transactions on Geoscience and Remote Sensing 41(2):215–229.
Papa, F., C. Prigent, F. Aires, C. Jimenez, W.B. Rossow, and E. Matthews. 2010. Interannual variability of
surface water extent at the global scale, 1993–2004. Journal of Geophysical Research 115(D12):D12111.
Prigent, C., F. Papa, F. Aires, W.B. Rossow, and E. Matthews. 2007. Global inundation dynamics inferred from
multiple satellite observations, 1993–2000. Journal of Geophysical Research 112(D12):D12107.
Quinn, P.F. and K.J. Beven. 1993. Spatial and temporal predictions of soil moisture dynamics, runoff, variable
source areas and evapotranspiration for Plynlimon, Mid-Wales. Hydrological Processes 7(4):425–448.
Quinn, P.F., K.J. Beven, and R. Lamb. 1995. The In(a/tan/β) Index: How to calculate it and how to use it within
the Topmodel Framework. Hydrological Processes 9(2):161–182.
Ringeval, B., B. Decharme, S.L. Piao, P. Ciais, F. Papa, N. de Noblet-Ducoudré, C. Prigent et al. 2012.
Modelling sub-grid wetland in the ORCHIDEE Global Land Surface Model: Evaluation against river
discharges and remotely sensed data. Geoscientific Model Development 5(4):941–962.
Rodhe, A. and J. Seibert. 1999. Wetland occurrence in relation to topography: A test of topographic indices as
moisture indicators. Agricultural and Forest Meteorology 98–99:325–340.
Shary, P.A., L.S. Sharaya, and A.V. Mitusov. 2002. Fundamental quantitative methods of land surface analysis.
Geoderma 107(1–2):1–32.
Wania, R., J.R. Melton, E.L. Hodson, B. Poulter, B. Ringeval, R. Spahni, T. Bohn et al. 2012. Present state of
global wetland extent and wetland methane modelling: Methodology of a model intercomparison project
(WETCHIMP). Geoscientific Model Development Discussions 5(4):4071–4136.
Weiss, A.D. 2001. Topographic positions and landforms analysis (poster). In: ESRI International User
Conference, San Diego, CA.
Wood, J. 1996. The geomorphological characterisation of digital elevation models. PhD thesis. University of
Leicester, Leicester, U.K.
Yarbrough, L.D., G. Easson, and J.S. Kuszmaul. 2005. Quickbird 2 Tasseled Cap transform coefficients: A
comparison of derivation methods. In: Global Priorities in Land Remote Sensing, Sioux Falls, SD.
21 Capturing the Dynamics of
Amazonian Wetlands Using
Synthetic Aperture Radar
Lessons Learned and
Future Directions
Thiago Sanna Freire Silva, John Melack, Annia Susin Streher,
Jefferson Ferreira-Ferreira, and Luiz Felipe de Almeida Furtado
CONTENTS
Introduction..................................................................................................................................... 455
How Do SAR Systems See the Amazon Wetlands: The Role of Acquisition Parameters.............. 456
Frequency/Wavelength............................................................................................................... 456
Incidence Angles........................................................................................................................ 458
Polarization................................................................................................................................ 459
Temporal Changes...................................................................................................................... 459
Overview of SAR Studies in the Amazon Floodplain.................................................................... 461
Current Limitations and Future Applications.................................................................................466
References.......................................................................................................................................469
INTRODUCTION
The Amazon basin comprises the largest continuous expanse of tropical forests in the world, cov-
ering about 78% of the basin’s 7 million km2 (Goulding et al. 2003). Despite the predominance of
upland (“terra firme”) forests, the Amazon is recognized by its large network of large rivers, fueled
by precipitation that can average up to 3500 mm/year, with annual extremes of 4500 mm or more
(Fisch et al. 1998; Satyamurty et al. 2009). Open water surfaces can reach up to 3% of the entire area
for the central Amazon main stem (Arraut et al. 2013), and the entire basin has an average discharge
of 209,000 m3/s from its more than 6 million km2 of drained area (Latrubesse et al. 2005).
The fairly flat topography of the basin results in the occurrence of vast floodplains along the
drainage network, covered by seasonally flooded forests, woodlands, shrublands, and grasslands
(Junk 1997, Junk et al. 2012). The spatial distribution and phenological responses of these plant
communities are intimately tied to the “flood pulse”—the annual inundation cycle observed for
these areas (Junk et al. 1989; Junk and Wantzen 2004). The predictability of the flood pulse has
important evolutionary implications for the flora and fauna living on the floodplain, and the charac-
teristic zonation of plant communities reflects the interaction between flood dynamics, topography,
and geomorphology (Mertes et al. 1995; Parolin et al. 2004).
In addition to the floodplains, interfluvial wetlands occur throughout the basin, having a hydro-
period largely dependent on local rainfall and with variations in vegetation cover and ecosystem
455
function (Belger et al. 2010; Cintra et al. 2013). Finally, the Amazon coast supports over 7500 km2
of mangrove forests, one of the largest continuous belts in the world (Souza-Filho 2005).
The sheer geographical scale of the Amazon wetland habitats has dictated the use of remote
sensing methods as a means to capture and understand their nature and complexity. Since the advent
of land observing satellites, much attention has been given to the task of mapping wetland environ-
ments in the Amazon basin and monitoring its hydrological and ecological processes. The humid
nature of tropical environments results in frequent cloud cover, hindering the application of optical
remote sensing over large geographical areas or continuously over time.
Synthetic aperture radar (SAR) systems have been successfully employed to study the Amazon
environment (Melack 2004). Considering the well-known capacity of SAR systems to identify inun-
dation under vegetation canopies (Hess et al. 1990), its capacity to penetrate cloud cover (Henderson
and Lewis 2008), and the correlation between radar backscattering and vegetation structure and
biomass (Lu 2006; Mitchard et al. 2012), it is no surprise that SAR has dominated the spectrum of
remote sensing applications to the study of Amazon wetlands. In this chapter, we review the history
and diversity of these applications, summarize what we have learned over the past 35 years of SAR
remote sensing in the Amazon, and offer some advice for future applications and research directions.
HOW DO SAR SYSTEMS SEE THE AMAZON WETLANDS:

THE ROLE OF ACQUISITION PARAMETERS
The Amazon wetland environments comprise multiple habitats (Junk et al. 2011), forming a mosaic
of flooded forests, woodlands, and grasslands, whose distribution is explained by the interactions
among river hydrology, local topography, and geomorphological dynamics (Junk and Piedade 1997;
Wittmann et al. 2010). Most remote sensing studies have focused on mapping environments along
the floodplains of the main stem Amazon River and major tributaries, comprised várzea (nutrient-
rich) and igapó (nutrient-poor) habitats—the dominant forms of wetland in the basin (Junk et al.
2011, 2012). The former are characterized by a mosaic of species-rich, high-biomass elevated forests
(high várzea), dense and species-poor low-lying tree and shrub formations (low várzea and chavas-
cal), and stands of aquatic herbaceous vegetation (macrophytes) (Wittmann et al. 2002; Junk and
Piedade 2010). During the low water season, vast expanses of herbaceous vegetation colonize most
of the exposed floodplain soil, while bare and recently exposed soil and clear sandbanks also occur
(Peixoto et al. 2009; Silva et al. 2010). Large open water surfaces represented by the main river
channels and floodplain lakes occur at all times.
Each of these land cover types interacts in different ways with the microwave radiation emitted
by SAR sensors, strongly depending on the acquisition parameters and on the timing of the observa-
tions in relation to the hydrological year. Most SAR systems are rather limited in terms of spectral
resolution, imaging within a single band/frequency of the electromagnetic spectrum. This limita-
tion is offset by configurability in other acquisition parameters (e.g., spatial resolution, incidence
angle, and polarization). The relative insensitivity of SAR sensors to cloud cover allows consistent
multitemporal acquisition, which adds significant flexibility to SAR applications. The effects of
these imaging parameters on the backscattering response of Amazon wetland environments are
reviewed in the following sections.
Frequency/Wavelength
The choice of SAR frequency/wavelength will largely determine the backscattering response from
different land covers. Orbital systems have operated mostly at L- (15–30 cm, 1–2 GHz), C- (3.7–
7.5 cm, 4–8 GHz), and X- (2.5–3.7 cm, 8–12 GHz) bands, which offer progressively less capacity
of penetration into vegetation canopies and which are increasingly sensitive to surface roughness.
Airborne systems have been operated at longer (e.g., P-band—1 m/300 MHz) or shorter (e.g.,
Ku—1.7–2.5 cm, 12–18 GHz) bands.
Capturing the Dynamics of Amazonian Wetlands Using Synthetic Aperture Radar 457
The existing historical record of SAR data at different wavelengths for the Amazon region
results from the acquisition strategies and lifetimes of each sensor platform. C-band data have
been acquired by ENVISAT Advanced Synthetic Aperture Radar (ASAR), RADARSAT-1, and
RADARSAT-2 only upon request, except for the regular ASAR Global Monitoring (ScanSAR,
1 km pixel spacing) acquisition schedule (Miranda et al. 2013). X-band data for the Amazon wet-
lands have been acquired mostly by TerraSAR-X, as part of their global DEM initiative, in addi-
tion to user acquisition requests. L-band data from both the Japanese Earth Resources Satellite
(JERS)-1 and Advanced Land Observing Satellite (ALOS)/Phased Array type L-band SAR
(PALSAR) have been the major source of SAR information for the region, since both programs
included a semiregular acquisition schedule for the basin, ensuring the existence of a large catalog
of archival imagery.
Figure 21.1 shows a portion of the Lago Grande de Curuai várzea, near the towns of Óbidos
and Santarém (lower Amazon, Pará, Brazil), which includes several land cover types: forests,
(a) (b)
Óbidos
(c) (d)
km
0 2 4 8 12 16
FIGURE 21.1 Four views of the northern portion of the Curuai floodplain, across from the town of Óbidos
(Pará, Brazil). (a) L-band (ALOS/PALSAR) (Courtesy of Japanese Aerospace Exploration Agency [JAXA/
EORC].), (b) C-band (RADARSAT-2) (Courtesy of MacDonald, Dettwiler and Associates Ltd.), and (c)
X-band (TerraSAR-X) (Courtesy of German Aerospace Center [DLR]). (d) Color composite of the three wave-
lengths, with R = X, G = L, and B = C. The large channel dominating the upper portion of the scene is the
Amazon River. The lower portion of the scene shows the Curuai floodplain region separated from the river
by levees, while the northeastern portion shows an area of darker upland (terra firme), a small brighter stretch
of floodplain, and a large water body (black). Deforestation patches and herbaceous areas are visible on the
L-band imagery (dark) and color composite (pink), on the left, and the effect of Bragg scattering over the open
water areas is visible on all bands. The enhanced backscattering from double bounce is visible on all three
bands, particularly on the northeastern part of the scene.
woodlands, shrublands, herbaceous vegetation, and water bodies, as seen by ALOS/PALSAR

(L-band), RADARSAT-2 (C-band), and TerraSAR-X (X-band), during the high water season. The
effect of inundation is clearly visible on all three images, with floodplain areas being noticeably
brighter than the surrounding “terra firme” (upland) due to the intensification of double-bounce
backscattering. Within the different vegetation types, the increasing attenuation of the backscat-
tered signal by the canopy as wavelength decreases is visible.
The response of different wavelengths to the relative surface roughness is more visible on floating
macrophyte and open water regions. The low emergent biomass and low density of the former are
hardly detected by the longer L wavelength, causing little to no backscattering to occur and appear-
ing as black surfaces. As frequency increases, the vegetation elements start to be registered, offering
further discrimination between emergent vegetation and actual open water surfaces. Relative wave
roughness plays a role on the effect of Bragg scattering on the signal returned by large water bod-
ies. Wind action on these areas will cause ripples and waves that, upon reaching a critical size and
inclination, will reflect large amounts of radiation back to the sensor, giving a very bright appear-
ance to these features. This effect is more pronounced at shorter wavelengths (and steep incidence
angles, see Section, Incidence Angles) and can be a major source of confusion in the discrimination
between open water and vegetated surfaces (Silva et al. 2009).
Incidence Angles
As a ranging device, radar-based systems are side-looking platforms, allowing the spatial separa-
tion of incoming signals by splitting the backscattered returns according to return time. The angle
formed between the travel direction of the beam and nadir will determine the incidence angle
(i.e., small values mean steeper incidence angles). Most orbital systems can image between 20°
and 40° incidence angles, with steeper or shallower options being available according to platform
(Table 21.1).
The effect of varying incidence angles can be seen for inundated vegetation types in the Amazon
floodplain wetlands (Figure 21.2). Shallower incidence angles increase the path of the radiation
through the canopy, increasing beam attenuation and volumetric scattering and reducing double-
bounce scattering. The responses decrease approximately linearly up to 34°, after which shrubs
do not cause significant reduction, while grass and forest stands cause decreases to continue. This
results in lower separability among vegetation classes for intermediate incidence angles. For low-
biomass emergent vegetation, backscattering is reduced until 39° and rises again at 46°, as only very
shallow angles will interact with the short vertical profile of this type of vegetation.
TABLE 21.1
Range of Incidence Angles for Major Orbital SAR Systems
Sensor Minimum Angle Maximum Angle
ALOS PALSAR (L-band, 2006–2011) 8° 60°
ENVISAT ASAR (C-band, 2002–2012) 15° 45.2°
RADARSAT-1 (C-band, 1995–2013) 10° 59°
RADARSAT-2 (C-band, 2007–present) 10° 60°
TerraSAR-X (X-band, 2008–present) 20° 55°
COSMO-SkyMed (X-band, 2007–present) 20° 60°
Note: Maximum range across all imaging modes is given, but not all angles are
available for all imaging modes.
Class Floating herbaceous Flooded forest Flooded grasses Flooded shrubs

0
–5
Radar backscatter—C-band (dB)
–10
–15
–20
–25
23.5 27 34 39 46
Incidence angle (degrees)
FIGURE 21.2 Changes in radar backscattering for different floodplain vegetation classes as a function of
incidence angle. Data obtained from a set of RADARSAT-2 standard-dual and standard-quad polarization
data.
Polarization
SAR systems are capable of emitting and recording the electromagnetic energy at specific polariza-
tions. Although early systems offered a single combination of send and receive polarizations (often
in HH configuration, meaning sending and receiving in horizontal polarization), current systems
often allow for dual polarimetric imaging, with HH and HV or HH and VV as the common choices.
As the wave interacts with distributed targets (such as vegetation), the radiation becomes partially
depolarized, and thus the proportion of like- and cross-polarized returns carry information about
the biophysical structure of the target.
Recent orbital systems have offered the possibility of acquiring fully polarimetric image data
(PolSAR). By enabling the antenna to send and receive at both horizontal and vertical polariza-
tions, the scattering matrix of the target can be reconstructed, and any intermediate polarization
status can be calculated from the data. Figure 21.3 shows the response of different land covers in the
Amazon floodplain to three different polarization configurations: HH, HV, and VV. The response of
the like-polarized data is similar for woody and grass classes and also for water surfaces. Floating
herbaceous vegetation, however, has a significantly reduced range of variation in the VV signa-
ture, likely due to its small vertical profile. Cross-polarized responses are overall lower than like-
polarized and more sensitive to the structural heterogeneity of herbaceous targets, and also avoid
most of the open water signal variability caused by surface waves.
Temporal Changes
The preceding examples show how all three components (frequency, polarization, and incidence
angle) can contribute to separating different vegetation types in wetland environments. Some
Class Floating herbaceous Flooded forest Flooded grasses Flooded shrubs Open water
0
Radar backscatter—C-band (dB)
–20
–40
–60
HH HV VV
SAR polarization
FIGURE 21.3 Response of floodplain land cover types to different polarization configurations, on the lower
Amazon floodplain, derived from a polarimetric RADARSAT-2 image acquired during high water season.
limitations prevent the full use of this information to achieve such separation: at a given time, a
satellite SAR sensor will be configured to acquire at a specific frequency, with a specific incidence
angle. Images acquired at different incident angles can be obtained from successive orbits, but
multifrequency data require the simultaneous acquisition by different orbital platforms, increasing
acquisition costs and incurring technical difficulties, such as coregistration between the scenes.
Airborne SAR systems, however, are able to image at multiple frequencies simultaneously. Dual-
polarized imagery can be usually acquired with a single pass of most modern sensors, but for
the complex and spatially heterogeneous land cover types observed in the Amazon floodplain, the
amount of extra information provided by dual polarizations is limited.
Time-series data add another data dimension and have been applied with success in Amazon
studies. Because of the flood pulse, Amazon floodplain ecosystems have a marked seasonality in
phenology and hydrology. These conditions cause stronger variations in the SAR backscattering
response than differences in polarization or incidence angle, thus providing relevant information
about the surface. While multitemporal optical remote sensing is of limited application in the humid
tropics due to cloud cover, regular temporal acquisitions are achievable using SAR systems.
The potential of multitemporal SAR can be visualized in Figure 21.4. TerraSAR-X images were
acquired during both low- and high-water seasons, at three different acquisition modes: ScanSAR
(18.5 m pixel size), spotlight (1.5 m pixel size), and spotlight high resolution (0.5 m pixel size).
The color compositions show the high water level image in red, the low water image in green, and
the ratio between low- and high-water images in blue. The level of discrimination among vegeta-
tion types is remarkable. Red hues represent the emergent aquatic grasses, which attain maximum
biomass during the peak of the flooding season. Green hues show forest and shrub vegetation, and
dark green hues correspond to floating herbaceous communities. Blue areas correspond to lake
bottom areas that are only exposed during the dry season (and are often colonized by opportunistic
10
km
1 0.1
km km
FIGURE 21.4 Multitemporal composition of SAR images acquired by TerraSAR-X. The top right inset
shows a scene acquired in ScanSAR mode (~18.5 m pixel spacing), the left image corresponds to the spotlight
acquisition mode (1.5 m pixel spacing), and the lower right inset shows a close-up view of a spotlight high-
resolution image (0.5 m pixel spacing). Images acquired during the high water season are shown in red, low
water images are shown in green, and the difference between the two dates is shown in blue (Courtesy of
German Aerospace Center [DLR]).
terrestrial herbaceous species), and black areas show the permanent water surface areas. Interesting
aspects of the fluvial geomorphology are also apparent, revealing the floodplain channels connect-
ing the lake basins, which have a key role in maintaining ecological and hydrological connectivity.
No combination of polarization, frequency, and incidence angle can provide such level of visual and
radiometric discrimination.
OVERVIEW OF SAR STUDIES IN THE AMAZON FLOODPLAIN

The Amazon floodplain has a rich history of SAR-based studies, and research in this geographical
region has contributed significantly to shape the current knowledge on the application of micro-
wave remote sensing to the study of wetlands. One of the earliest radar studies of the Amazon envi-
ronment was the “Radar na Amazônia” (“Radar in the Amazon”) Mission, organized in the early
1970s, by the Departamento Nacional da Produção Mineral (“National Department for Mineral
Production”). This initiative was later expanded into the RADAMBRASIL Project, mapping the
entire Brazilian territory using an X-band side-looking airborne radar (SLAR) system, aboard
a Caravelle aircraft. At the time, the SLAR system represented a major technological advance,
allowing image acquisition at any time of day, under any cloud cover condition. Data acquired
from the mission were manually interpreted, resulting in a large collection of 1:250,000 maps
of mineral resources, vegetation, and geomorphology, among others (RADAMBRASIL 1983).

The Amazon had a second large airborne mission in 1992, with the South American Radar
Experiment (SAREX-92), led by the European Space Agency (ESA), in support of the Earth
Resources Satellite (ERS)-1 imagery interpretation. A year later, the National Aeronautics and
Space Administration flew its AIRSAR sensor over South America (Hoekman et al. 1994).
The first spaceborne SAR system was launched aboard the SEASAT platform in 1978. The
instrument was an L-band SAR, designed primarily to monitor sea ice conditions, but it was rec-
ognized as having potential for land studies (Bryan and Clark 1984). Unfortunately, SEASAT only
operated for 106 days. The next orbital SAR system was launched in 1991 aboard the first European
Remote Sensing Satellite (ERS-1) from ESA, a C-HH instrument. With the main purpose of moni-
toring ocean surface waves, the spatial resolution of 30 m and 100 km swath offered by the system
was well matched with optical satellite data for the study of land surface processes.
Orbital SAR technology became established in the 1990s, with the deployment of several systems.
ERS-1 was followed by ERS-2 in 1995. The Japanese National Space Development Agency (currently
known as Japan Aerospace Exploration Agency [JAXA]) launched the JERS-1 in 1992, carrying an
L-band SAR instrument, as well as two optical sensors. In 1994, the Spaceborne Imaging Radar
(SIR-C/X-SAR) mission was flown twice using the space shuttle Endeavour. The SIR-C instrument
acquired images at both C- and L-bands and was a pioneer in the recording of different polariza-
tions, as well as establishing the first use of the wide-swath burst-mode SAR acquisition, currently
known as ScanSAR, and the use of radar interferometry to determine inundation height (Alsdorf
et al. 2001a). Also in 1995, RADARSAT-1 and ERS-2 were launched, both C-band instruments with
~30 m spatial resolution, focused on ocean and ice monitoring as well as land applications.
The simultaneous availability of multiple SAR datasets for many regions of the globe ushered
an increase in SAR applications to the study of land ecosystems, including a focus on wetlands.
For the Amazon region, SAR applications were spearheaded by the Global Rain Forest Mapping
(GRFM) initiative, which aimed to provide continental-wide mosaics of spatially and temporally
continuous JERS-1 images over large tropical forest areas. The entire Amazon basin was imaged
in September–December 1995 and again in May–August 1996, to coincide with the low and high
water seasons of the floodplain (Chapman et al. 2002).
The first SAR studies focusing specifically on the Amazon wetlands appeared in the peer-
reviewed literature at this time. L-band JERS-1 data were established as an efficient tool to map
both land cover and inundation status (Hess et al. 1995; Wang et al. 1995), which could be related to
carbon biogeochemistry (Rosenqvist et al. 1998). This led to the first and, to date, the only complete
map of wetland cover type and extent for the central Amazon basin (Hess et al. 2003). The study
was possible because of the availability of the two region-wide JERS-1 mosaics, corresponding to
the low water season of 1995 and the high water season of 1996, and produced under the JERS-1
Amazon Multi-Season Mapping Study within the GRFM initiative (Chapman et al. 2002). The
study by Hess et al. (2003) also applied large-scale use of image segmentation to delineate land-
scape units, followed by an unsupervised classification and extensive manual editing, to generate the
wetland mask based on the high water image mosaic. Five vegetation cover classes were identified
within the delineated wetland areas, as well as its inundation status on each mosaic. This pixel-
based classification was based on a rectangular partition of the 2D space formed by the backscat-
tering signatures of each class during low and high water seasons. Validation was based on a large
database of high-resolution aerial videography (Hess et al. 2002), and accuracies ranged between
55% and 94%, per class. Recently, these results have been extended to the whole lowland Amazon
basin (Melack and Hess 2010). The geometric accuracy of the original wetland mask has been
rectified in relation to the Shuttle Radar Topography Mission global elevation dataset, using local
warping algorithms (Rennó et al. 2013). Efforts improving the fine-scale accuracy of the dataset by
extensive manual editing are also ongoing (Ferreira et al. 2013; L. Hess, personal communication).
The results of Hess et al. (2003) established the significant contribution of multitemporal SAR
imaging for identifying land cover classes in seasonally inundated wetlands. This was also explored
by Costa et al. (2002) and Costa (2004) to map the spatial distribution of vegetation in a portion
of the Amazon floodplain, while also investigating the potential of using multiple frequencies to
improve the discrimination among classes. The authors analyzed the backscattering coefficients
from a multitemporal RADARSAT-1 and JERS-1 (C- and L-bands, respectively) dataset, for dif-
ferent land cover classes in the Monte Alegre Lake floodplain (lower Amazon, Pará, Brazil) and
surrounding upland regions. The results suggested that a multitemporal and multifrequency region-
based classification approach was an optimal choice for mapping floodplain habitats, with sea-
sonal map accuracies higher than 95% for vegetated areas. Regions subjected to seasonal flooding
had larger variations in radar backscattering (for both C- and L-bands), with lower values during
low water levels and rising with inundation. The study demonstrated how the sensitivity of C- and
L-bands to the scattering elements in the ground differs according to the relative contribution of
each scattering mechanism.
Silva et al. (2010), also working at Monte Alegre Lake, proposed a hybrid approach, combining
multisource and multitemporal optical Moderate Resolution Imaging Spectroradiometer (MODIS)
and C-band RADARSAT-1 imagery using a hierarchical, object-oriented classification method. The
multitemporal information allowed the discrimination between temporally stable land cover classes
(trees, shrubs, and permanent open water) and “transient” classes, such as aquatic macrophytes,
bare soil (during low water levels), and areas that are completely covered by water only during the
higher levels of the flood pulse. Within the stable vegetation classes, flooded shrubs composed of
Montrichardia arborescens stands or early succession open forests (Pseudobombax munguba and
Cecropia spp.) had the strongest double-bounce backscattering on C-band (also shown by Hess et al.
[1995] and Costa et al. [2002]). Flooded forests had variable backscattering signals resulting from
variations in biomass and leaf area index (LAI), with variable mixtures of double-bounce and volu-
metric scattering (Silva et al. 2010). Once the transient areas were properly separated, the confu-
sion between the highly variable macrophyte areas and other vegetation areas was minimized, and
additional optical information improved the separation between macrophytes and either wet bare
soil or windy water surfaces, both of which can be as bright as herbaceous stands on single-date
C-band imagery. This study offered the first temporal quantification of macrophyte cover variability
within a single flooding season in the Amazon floodplain, and also demonstrated the susceptibility
of macrophyte cover to changes in flooding levels between years.
The studies by Hess et al. (1995), Costa et al. (2002), and Silva et al. (2010) emphasize how
Amazonian macrophyte stands can show the entire range of radar scattering responses, from smooth
surfaces to complex combinations of double-bounce and volumetric scatterings and usually within a
single growth season. This ample range of variation is induced by changes in species diversity and
the occurrence of different species associations and by changes in biomass along the season.
Assemblages of highly productive grasses (Hymenachne amplexicaulis, Echinochloa polys-
tachya, Paspalum spp., Panicum spp., and Oryza spp.) occur on exposed littoral areas and in regions
of high nutrient input, such as floodplain channels communicating directly with the main river (iga-
rapés and furos). Alternatively, associations of floating species (Nymphaeaceae spp., Echinochloa
spp., Neptunia spp., and Ludwigia spp.) can be found on still, backwater areas. These communities
have distinct SAR backscattering responses, due to their biomass levels and biophysical configura-
tion (Figure 21.5). During early growth and senescence, plant densities and biomass are low, reduc-
ing volumetric and double-bounce responses, with overall low values of backscattering coefficients,
especially on L-band. As macrophyte stands accumulate biomass, volumetric backscattering starts
to be the dominant mode of interaction, resulting in an increase of the observed signal at high water,
complemented by double-bounce interactions. Amazonian aquatic grasses are among the most pro-
ductive plants in the world, with net primary productivity rates of 4–10 kg/m2/year (dry weight) (Silva
et al. 2009), providing a strong backscattering response at peak biomass levels. In L-band, macro-
phyte backscattering values for aquatic grasses remain lower than for flooded forests and shrubs at all
times, while for C-band, the backscatterings of forests and grass stands are comparable during high
water season, both slightly lower than shrubs (Hess et al. 1995; Costa et al. 2002; Silva et al. 2010).
(a) (b) (c)
Grasses
Floating
km
0 1 2
FIGURE 21.5 Different backscattering responses for floating macrophytes (low-biomass and short verti-
cal profile) and emergent grasses (high-biomass and tall profile) on (a) L-band ALOS/PALSAR (Courtesy of
ALOS ScanSAR image), (b) C-band RADARSAT-2 (Courtesy of MacDonald, Dettwiler and Associates Ltd.),
and (c) X-band TerraSAR-X (Courtesy of German Aerospace Center [DLR]).
Because SAR data are sensitive to the underlying water in vegetation canopies, especially at
longer wavelengths (L-band and above), another major application of radar remote sensing has been
the mapping and monitoring of flood extent and timing in Amazon wetlands. Multitemporal SAR
inundation mapping was first performed by Rosenqvist et al. (2002) to characterize the duration and
extent of flooding in the Jaú basin, using a band threshold algorithm. Martinez and Le Toan (2007)
used a time series of 21 images acquired by JERS-1, from 1993 until 1997, to map the temporal
dynamics and spatial distribution of flooding and vegetation in the Curuai Lake floodplain, Pará
State, Brazil. The study developed a classification algorithm based on the mean flooding duration of
each pixel, using the mean seasonal backscattering coefficient to provide a classification of general
vegetation types and an absolute change index that yielded an estimate of flood dynamics, catego-
rized in three flood conditions (never flooded, occasionally flooded, and permanently flooded). The
accuracy of the flood extent mapping had a “kappa” value of 0.82. The classifier was run iteratively
on the occasionally flooded pixels to monitor intermediate flood stages, yielding good results with a
“kappa” value of 0.96. Discrimination between land cover classes was harder for classes with simi-
lar roughness (bare soil and open water), which resulted in low mapping accuracy (approximately
30%). The confusion between smoother surfaces was attributed to the lower sensitivity of L-band
data (Martinez and Le Toan 2007).
Arnesen et al. (2013) introduced a method for monitoring changes in flood extent, using a multi-
temporal and multisensor approach derived from Silva et al. (2010). Twelve L-band ScanSAR/ALOS
images were acquired over the Curuai Lake floodplain (Lower Amazon, Pará, Brazil) between
2006 and 2010 and combined with optical imagery (TM/Landsat-5 and Terra and Aqua MODIS).
A data mining algorithm (J.48, WEKA Data Mining Software, Hall et al. [2009]) allowed the fitting
of a decision tree classifier, implemented as a hierarchical object-based algorithm. Flooding status
was mapped with accuracies varying between 80% and 90% and showed close agreement to results
obtained by dynamic hydraulic modeling in the same region (Rudorff et al. 2014). A logistic model
was fitted between estimated flooded area and water levels observed at the Curuai gauge station,
with a pseudo-R2 of 0.97 and a Root Mean Squared Error (RMSE) of 71 km2 (3% of maximum
flooded area). The authors concluded that for the less forested landscape of the lower Amazon
floodplain, a prior characterization of general land cover classes is paramount to achieve accurate
flood mapping, as the different vegetation types will have distinct sometimes opposite changes in
backscattering when subject to flooding. Arnesen et al. (2013) also observed a strong dependency
of backscattering on incidence angle, a characteristic of the wide swath of the ScanSAR mode (also
shown by Lang et al. [2008] and O’Grady et al. [2013]), which had a noticeable effect on classifica-
tion accuracy and had to be considered during algorithm development.
The hydrological dynamics of the floodplain have also been studied using interferometric SAR
methods by Alsdorf et al. (2001a,b, 2007). Although the radar beam is specularly reflected by open
water surfaces, the authors have shown that the presence of tree cover induces double-bounce scat-
tering, greatly improving the coherence for different baselines even if only a few trees are present.
Accuracies of approximately 5 cm were obtained using multiple passes from both orbital imagery
(JERS-1) and SIR-C shuttle data. The authors then combined interferometric data with radar altim-
etry profiles to estimate temporal changes in water volume (Alsdorf et al. 2001b). Furthermore,
passive microwave remote sensing has been successfully applied to monitor changes in flooding for
the Amazon floodplain and other large tropical wetlands (Sippel et al. 1998; Hamilton et al. 2002).
The availability of local and regional maps of change in wetland extent and vegetation cover
has contributed significantly to improve the knowledge on carbon biogeochemistry in Amazonian
wetlands. Using the results from Hess et al. (2003), Richey et al. (2002) estimated the atmospheric
CO2 outgassing for the entire region, and Melack et al. (2004) performed a similar analysis for CH4
emissions. Both studies established the important role of the Amazonian wetlands in the regional
carbon budget. In a recent study, Abril et al. (2014) demonstrated how a large portion of the carbon
dioxide export and outgassing from the Amazon River is fuelled by the associated floodplains,
deriving vegetation cover and flooding patterns from ALOS/PALSAR data using the algorithm
proposed by Martinez and Le Toan (2007).
SAR studies have allowed the characterization of the contribution of herbaceous vegetation
to the floodplain carbon cycle. Costa (2005) estimated standing biomass using a combination of
RADARSAT-1 and JERS-1 imageries and empirical models based on field observations. Silva et al.
(2010) modeled total annual production by randomly associating field-measured monthly biomass
values to RADARSAT-1-derived estimations of monthly macrophyte cover and bootstrapping the
results. The method was later expanded by Silva et al. (2013) who used empirical models to relate
both macrophyte cover extent and macrophyte seasonal growth to observed flooding levels and then
used these relationships to simulate the responses of macrophyte annual productivity to interannual
variations in the flood pulse for a period of 40 years. These studies have highlighted the strong
dependency and rapid response of macrophyte distribution and growth to changes in flooding pat-
terns, with important implications under future climate change scenarios.
For forested areas, Hawes et al. (2012) have used flooding maps derived from ALOS/PALSAR
data to extrapolate standing biomass over the extent of the Juruá and Uacari reserves in the central
Amazon. SAR has been used to regionalize estimates of CO2 and CH4 emissions from chavascal
interfluvial areas in the upper Negro River basin (Belger et al. 2010). These areas are character-
ized by a combination of herbaceous plants, shrubs, and palms and are subjected to nonperiodic
flooding. A series of 24 RADARSAT-1 images were used to derive vegetation cover and inundation
frequency and distribution (illustrated in Melack et al., 2009).
The extensive mangrove belts of the Amazon coast, subject to both tidal cycles and river flood-
ing, near the mouth of the Amazon River, have also been studied using SAR. Souza-Filho et al.
(2009) and Rodrigues and Souza-Filho (2011) investigated the extent to which coastal land cover
and geomorphology could be characterized based on the digital integration of Landsat-5/TM and
RADARSAT-1 wide beam imagery. The data were acquired over the northeastern coast of Pará
State, in the easternmost border of the Brazilian Amazon. A multisensor data fusion process, called
selective principal component-SAR, was used, combining a principal component analysis and an
intensity–hue–saturation transformation. According to the authors, the Landsat-5/TM images con-
tributed to enhance vegetation and sedimentary environments based on the optical response, while
RADARSAT-1 data enhanced differences between superficial roughness and moisture. Based on
the results, Souza-Filho et al. (2009) concluded that the Amazon shoreline might be remarkably
sensitive to oil spills.
Souza-Filho et al. (2011) determined the best polarization configurations (HH, VV, or HV) to
discriminate different types of coastal Amazon wetlands. A maximum-likelihood supervised clas-
sification was applied to both single polarization and a multipolarized image composition, showing
that accuracy was slightly improved when combining polarizations, with an overall accuracy of
83% versus single-pol accuracies between 78% and 81%. According to the authors, VV was the best
polarization for the recognition of shallow-water morphology in the intertidal zone, while for man-
groves and marshes, HH polarization worked best, followed by VV. The results also showed that
HH and VV polarizations were better for delineating transitional zones between different coastal
environments, while the HV polarization was best for mapping abrupt boundaries in the coastal
zone.
Nascimento et al. (2013) used JERS-1 SAR and ALOS PALSAR data, acquired in 1996 and
2008, respectively, to map changes in the extent of mangroves along the entire northern Brazilian
belt, from east of the Amazon River mouth, Pará State, to São José Bay, Maranhão. The authors used
object-orientated image analysis to identify major land covers (mangrove, secondary vegetation,
gallery and swamp forest, open water, intermittent lakes, and bare areas) for multiple years, with
accuracies above 80%. The results show that the Amazon macrotidal mangrove coast (AMMC) is
the largest continuous mangrove area in the world, 1500 km² larger than the extensive mangroves of
India and Bangladesh, and, more importantly, that the AMMC is also the only mangrove region in
the world that is increasing in extent, on the order of 10% (~718 km²) between 1996 and 2008. This
increase may be attributed to lateral and landward migration within highest-elevation areas, sug-
gesting that salinization of the upper courses of the estuaries produces propitious areas for upstream
mangrove development.
CURRENT LIMITATIONS AND FUTURE APPLICATIONS

Although largely successful in the study of Amazon wetlands, the use of SAR imagery in the
region is still hindered by some limitations inherent to the technology and/or the strategic planning
of most orbital missions. One prime limitation is the lack of a consistent historical record of SAR
imagery, when compared with optical systems such as the Landsat series of sensors (1972–present),
the Advanced Very High Resolution Radiometer series (AVHRR, 1978–present), and the pair of
Terra and Aqua MODIS sensors (2000–present). This is because SAR sensors have been operated
on a per-request acquisition basis, without a continuous acquisition program, and because of long
hiatuses between the launchings of sensor systems.
For the Amazon, the only continuous time series of SAR imagery are available from the ALOS/
PALSAR mission, which will be followed by PALSAR-2/ALOS-2 this year. For both missions,
JAXA adopted a “Basic Observation Scenario” (BOS), developed to produce spatially and tempo-
rally consistent coverage of the planet on a repetitive basis at fine- and medium-resolution scales
(Rosenqvist et al. 2007, in press). The scenario was established in collaboration with the Kyoto and
Carbon Initiative Science Advisory Panel (Rosenqvist et al. 2010). The guiding principles of the
BOS are to maximize spatial and temporal consistency as well as seasonal homogeneity to meet
the requirements for its key applications: semiannual repetition for forestry/land cover and every-
cycle repetition over inundated wetlands (Rosenqvist et al. 2007). The data were set to be acquired
during the same time period every year, to minimize temporal bias. PALSAR ScanSAR mode
was designed to provide data for seasonal phenomena studies, such as flood dynamics and extent
monitoring, and ScanSAR observations were undertaken over 27 wetland environments of global
significance, including the catchment basins of the Amazon, Paraná, Paraguay, Congo, Mekong,
Irrawaddy, Ob, and Yenisey rivers, as well as all of Southeast Asia. ScanSAR observations were
performed in a 46-day cycle and scheduled to capture a full flood cycle from one dry season to the
next. Fine-beam mode observations were scheduled to capture seasonal patterns as much as pos-
sible, according to each geographical region, such as summer/winter in the temperate and boreal
regions and dry/wet seasons in the tropics and subtropics.
The ALOS successor mission—ALOS-2—is scheduled for launch in May 2014. The mission
objectives, as defined by JAXA, include monitoring of disasters, global forests, land use change,
agriculture (rice), and wetlands (Rosenqvist et al. in press). ALOS-2 carries a single imaging
instrument—the PALSAR-2—which has improved spatial resolution and radiometric sensitivity
in relation to the original PALSAR and will have a greater range of beam modes. ALOS-2 will
continue to build long-term time-series archives of L-band SAR data for monitoring of the environ-
ment at local, regional, and global scales, especially over the tropics (Rosenqvist et al. in press). The
fine-beam mode of PALSAR-2 will be set to provide gap-free coverage at the equator, while the
ScanSAR mode will have a shorter revisit over wetland areas.
Other limitations of SAR in wetlands are geophysical. Despite the relative insensitivity of micro-
wave beams to nebulosity, the dense cloud cover often observed in the Amazon season can have an
effect on X-band and even C-band (Figure 21.6).
A second large limitation is the necessity of proper scene acquisition timing for some methods to
be applied. For example, elevation or biomass estimation using interferometric methods needs to be
carried on images acquired during the lowest possible water level to ensure that the signal returned
corresponds to the entire vegetation canopy and/or to the bare ground surface, without being influ-
enced by inundation. This also limits the applicability of archival images to Amazon wetland stud-
ies. Interferometric analysis is also hindered by the loss of coherence between consecutive dates
caused by changes in inundation levels or soil moisture.
The current status of SAR research in the Amazon reflects the state of the remote sensing field.
Multitemporal approaches have proved multiple times as the best strategy to characterize such a
dynamic environment, turning the usual limitation of strong seasonality into an additional and pow-
erful source of information. In addition, the combination of optical and SAR remote sensings, either
through data fusion or through multisource methods, has also been shown to be a good strategy for
overcoming the limitations of current SAR systems for single-date information.
The recent availability of orbital sensors with full polarimetric capability can provide addi-
tional physically based information about land cover without the usual shortcomings of optical data
(a) (b)
FIGURE 21.6 Cloud effect on a RADARSAT-1 C-HH ST2 Mode (C-band) frame of Curuai Lake, eastern
Amazon floodplain. (a) Image acquired on September 2, 2004; (b) image acquired on September 26, 2004.
acquisition in the humid tropics. Polarimetric studies in the Amazon wetlands are in their infancy
but show promising results. Sartori et al. (2011) have used polarimetric ALOS/PALSAR imagery
to discriminate among macrophyte species in Monte Alegre Lake (lower Amazon, Pará, Brazil),
showing that structural differences between the dominant plant species in different forest stands can
be successfully identified through the use of polarimetric decomposition.
Furtado (2014) studied the applicability of polarimetric RADARSAT-2 data to map vegeta-
tion distribution and estimate forest biophysical parameters (Figure 21.7). The author compared
the performance of different polarimetric decompositions to discriminate between main floodplain
vegetation types, identifying the Cloude–Pottier and the van Zyl decompositions as providing the
best class separability. The author also tested the effect of incidence angle on the computation of
the polarimetric attributes, without finding a substantial effect. The use of polarimetric informa-
tion resulted in small gains in classification accuracy, with a kappa value of 0.75, compared to the
accuracy obtained by using amplitude data alone (kappa ≈ 0.7). The addition of a second image,
acquired during the opposite period of the seasonal cycle, was able to increase accuracy to kappa =
0.87 using only amplitude data, again demonstrating the potential of multitemporal imagery. For
determining forest structure, polarimetric information was shown to be effective, even at C-band.
Good fits between polarimetric attributes and forest biophysical variables were found using simple
linear regression models, even with small sample sizes (n = 17). For example, LAI was predicted by
the volumetric scattering component of the van Zyl decomposition with an R2 of 0.68 and RMSE of
0.61; and the mean diameter at breast height was predicted by the double-bounce component of the
same decomposition with R2 of 0.51 and RMSE of 2.91 cm. Considering that C-band lacks canopy
Class Open water Flooded shrubs Flooded forest Upland forest Flooded grasses Floating herbaceous Bare soil
Cloude–Pottier, alpha angle Cloude–Pottier, anisotropy Cloude–Pottier, entropy

0.8
50
0.6 0.8
40
0.4
Radar response (see caption for units)
0.6
30 0.2
0.4
van Zyl, double-bounce van Zyl, odd van Zyl, volumetric
–5 –7.5
–5
–10.0
–10
–10 –12.5
–15.0
–15 –15
–17.5
–20 –20 –20.0
Land cover classes
FIGURE 21.7 Polarimetric responses of land covers in Curuai Lake, eastern Amazon floodplain, Brazil.
Note the variable y scales. Cloude–Pottier components are α-angle (degrees), anisotropy (unitless), and
entropy (unitless). Van Zyl components are double bounce (dB), surface/odd (dB), and volumetric (dB). (From
Furtado, L.F.A., Dados polarimétricos de banda C para o mapeamento e modelagem da estrutura da vegetação
na várzea amazônica—Lago Grande de Curuai, PA. Master thesis in Remote Sensing, Instituto Nacional de
Pesquisas Espaciais [INPE], São Paulo, Brazil, 2014.)
penetration and the small sample size of the study, these results indicate a real potential for forest
structure mapping using polarimetric information.
The future of SAR applications for the study of Amazon wetlands seems bright. Several orbital
SAR sensors are now concurrently in orbit, with RADARSAT-2, ALOS-2/PALSAR-2, Sentinel-1,
TerraSAR-X, and COSMO-SkyMed covering the three primary SAR frequencies. In the foreseeable
future, more missions are expected to become operational, such as the low-spatial-resolution Soil
Moisture Active Passive mission, which will provide L-HH, L-HV, and L-VV images at a 1 km grid
with 1000 km swath, allowing high-frequency monitoring of surface patterns at regional scales, or the
SAOCOM (Satélite Argentino de Observacíon con Microondas) mission, carrying an L-band fully
polarimetric sensor. These modern sensor systems include a wide variety of acquisition modes, rang-
ing from wide-swath, moderate-resolution (ScanSAR) to very fine-resolution data (i.e., the new star-
ing spotlight mode of the TerraSAR-X mission, offering a final pixel size of approximately 0.25 m).
Some convergence can be observed between the different acquisition modes. While ScanSAR modes
have been considered as a trade-off between coverage and spatial resolutions, acquisition modes such
as RADARSAT-2 wide modes offer large swathes paired with fine spatial resolution.
Most of the current and planned systems are expected to operate as constellations providing
higher frequencies of acquisition and increased interferometric capabilities. TerraSAR-X is cur-
rently paired with the twin TanDEM-X satellite, and the RADARSAT system is planned to include
three satellites by 2018. The recently launched Sentinel-1A is expected to be followed by a twin
satellite, and SAOCOM is planned as a set of two satellites, which combined with four planned
COSMO-SkyMed satellites will comprise the Italian-Argentine System of Satellites for Emergency
Management (SISAGE). This unprecedented availability of interferometrically enabled data will
likely increase and popularize the use of radar interferometry for environmental applications in
the Amazon basin. In fact, interferometric studies appear to be the most lacking in Amazon wet-
lands, considering the current literature. Recent developments in the field include several methods
to derive 3D information from volume scatterers (i.e., vegetation), such as polarization coherence
tomography (Cloude 2006), repeat-pass interferometry (Askne et al. 1997; Treuhaft and Siqueira
2000), and polarimetric interferometry (PolInSAR) (Neumann et al. 2010), but no applications have
been reported to date for Amazon wetlands.
REFERENCES
Abril, G., Martinez, J.-M., Artigas, L.F., Moreira-Turcq, P., Benedetti, M.F., Vidal, L., Meziane, T. et al. 2014.
Amazon River carbon dioxide outgassing fuelled by wetlands. Nature 505:395–398.
Alsdorf, D., Bates, P., Melack, J.M., Wilson, M., and Dunne, T. 2007. Spatial and temporal complexity of the
Amazon flood measured from space. Geophysical Research Letters 34(8):L08402.
Alsdorf, D.E., Birkett, C., Dunne, T., Melack, J.M., and Hess, L.L. 2001a. Water level changes in a large
Amazon lake measured with spaceborne radar interferometry and altimetry. Geophysical Research
Letters 28(14):2671–2674.
Alsdorf, D.E., Smith, L.C., and Melack, J.M. 2001b. Amazon floodplain water level changes measured with
interferometric SIR-C radar. IEEE Transactions on Geoscience and Remote Sensing 39(2):423–431.
Arnesen, A.S., Silva, T.S.F., Hess, L.L., Novo, E.M.L.M., Rudorff, C.M., Chapman, B.D., and McDonald, K.C.
2013. Monitoring flood extent in the lower Amazon River floodplain using ALOS/PALSAR ScanSAR
images. Remote Sensing of the Environment 130:51–61.
Arraut, E.D.M., Silva, T.S.F., and Novo, E.M.L.M. 2013. Secas extremas na planície de inundação amazônica:
Alguns impactos sobre a ecologia e biodiversidade. In: L.S. Borma and C.A. Nobre (eds.). Secas na
Amazônia: Causas e Consequências, pp. 243–267. São Paulo, Brazil: Oficina de Textos.
Askne, J.I.H., Dammert, P.B.G., Ulander, L.M.H., and Smith, G. 1997. C-band repeat-pass interferometric
SAR observations of the forest. IEEE Transactions on Geoscience and Remote Sensing 35(1):25–35.
Belger, L., Forsberg, B.R., and Melack, J.M. 2010. Carbon dioxide and methane emissions from interfluvial
wetlands in the upper Negro River basin, Brazil. Biogeochemistry 105(1–3):71–183.
Bryan, M.L. and Clark, J. 1984. Potentials for change detection using seasat synthetic aperture radar data.
Chapman, B., Siqueira, P., and Freeman, A. 2002. The JERS Amazon multi-season mapping study (JAMMS):
Observation strategies and data characteristics. International Journal of Remote Sensing 23(7):1427–1446.
Cintra, B.B.L., Schietti, J., Emillio, T., Martins, D., Moulatlet, G., Souza, P., Levis, C., Quesada, C.A., and
Schöngart, J. 2013. Soil physical restrictions and hydrology regulate stand age and wood biomass turn-
over rates of Purus–Madeira interfluvial wetlands in Amazonia. Biogeosciences 10(11):7759–7774.
Cloude, S.R. 2006. Polarization coherence tomography. Radio Science 41(4):RS4017.
Costa, M.P.F. 2004. Use of SAR satellites for mapping zonation of vegetation communities in the Amazon
floodplain. International Journal of Remote Sensing 25(10):1817–1835.
Costa, M.P.F. 2005. Estimate of net primary productivity of aquatic vegetation of the Amazon floodplain using
Radarsat and JERS-1. International Journal of Remote Sensing 26(20):4527–4536.
Costa, M.P.F., Niemann, O., Novo, E.M.L.M., and Ahern, F. 2002. Biophysical properties and mapping of
aquatic vegetation during the hydrological cycle of the Amazon floodplain using JERS-1 and Radarsat.
Ferreira, R.D., Leão, J.A.D., Silva, T.S.F., Rennó, C.D., Novo, E.M.L.M., and Barbosa, C.C.F. 2013. Atualização
e correção do delineamento de áreas alagáveis da bacia Amazônica. In: Anais XVI Simpósio Brasileiro de
Sensoriamento Remoto, pp. 5864–5871. Foz do Iguaçu, Brazil: Instituto Nacional de Pesquisas Espaciais.
Fisch, G., Marengo, J.A., and Nobre, C.A. 1998. Uma Revisão Geral sobre o Clima da Amazônia. Acta
Amazonica 28(2):101–126.
Furtado, L.F.A. 2014. Dados polarimétricos de banda C para o mapeamento e modelagem da estrutura da veg-
etação na várzea amazônica—Lago Grande de Curuai, PA. Master thesis in Remote Sensing. Instituto
Nacional de Pesquisas Espaciais (INPE), Brazil.
Goulding, M., Barthem, R., and Ferreira, E.J.G. 2003. Smithsonian Atlas of the Amazon. London, U.K.:
Smithsonian Books.
Hall, M., Frank, E., Holmes, G., Pfarhringer, B., Reutemann, P., and Witten, I. 2009. The WEKA data mining
software: An update. SIGKDD Explorations 11(1):10–18.
Hamilton, S.K., Sippel, S.J., and Melack, J.M. 2002. Comparison of inundation patterns among major South
American floodplains. Journal of Geophysical Research 107(D20):1–14.
Hawes, J.E., Peres, C.A., Riley, L.B., and Hess, L.L. 2012. Landscape-scale variation in structure and biomass
of Amazonian seasonally flooded and unflooded forests. Forest Ecology and Management 281:163–176.
Henderson, F.M. and Lewis, A.J. 2008. Radar detection of wetland ecosystems: A review. International Journal
Hess, L.L., Melack, J.M., and Filoso, S. 1995. Delineation of inundated area and vegetation along the Amazon
floodplain with the SIR-C synthetic aperture radar. IEEE Transactions on Geoscience and Remote
Sensing 33(4):896–904.
Hess, L.L., Melack, J.M., Novo, E.M.L.M., Barbosa, C.C.F., and Gastil, M. 2003. Dual-season mapping of
87:404–428.
Hess, L.L., Melack, J.M., and Simonett, D.S. 1990. Radar detection of flooding beneath the forest canopy:
Hess, L.L., Novo, E.M.L.M., Slaymaker, D.M., Holt, J., Steffen, C., Valeriano, D. M., Mertes, L.A.K. et al. 2002.
Geocoded digital videography for validation of land cover mapping in the Amazon basin. International
Hoekman, D., van der Sanden, J.J., and Bijker, W. 1994. The AIRSAR-93 and SAREX-92 campaigns in
Guyana and Colombia. In: Proceedings of IGARSS’94—1994 IEEE International Geoscience and
Remote Sensing Symposium, pp. 1051–1053. Piscataway, NJ: IEEE.
Junk, W.J. 1997. The Central Amazon Floodplain: Ecology of a Pulsing System, 1st edn. New York: Springer.
Junk, W.J., Bayley, P.B., and Sparks, R.E. 1989. The flood pulse concept in river-floodplain systems. Canadian
Special Publications in Fisheries and Aquatic Sciences 106:110–127.
Junk, W.J. and Piedade, M.T.F. 1997. Plant life in the floodplain with special reference to herbaceous plants.
In: W.J. Junk (ed.). The Central Amazon Floodplain: Ecology of a Pulsing System, pp. 147–186. Berlin,
Germany: Springer-Verlag.
Junk, W.J. and Piedade, M.T.F. 2010. An introduction to South American wetland forests: Distribution, defini-
tions and general characterization. In: W.J. Junk, M.T.F. Piedade, F. Wittmann, J. Schöngart, and P. Parolin
(eds.). Amazonian Floodplain Forests: Ecophysiology, Biodiversity and Sustainable Management,
pp. 3–25. Dordrecht, the Netherlands: Springer.
Junk, W.J., Piedade, M.T.F., Schöngart, J., Cohn-Haft, M., Adeney, J.M., and Wittmann, F. 2011. A classifica-
tion of major naturally-occurring Amazonian lowland wetlands. Wetlands 31(4):623–640.
Junk, W.J., Piedade, M.T.F., Schöngart, J., and Wittmann, F. 2012. A classification of major natural habitats of
Amazonian white-water river floodplains (várzeas). Wetlands Ecology and Management 20(6):461–475.
Junk, W.J. and Wantzen, K.M. 2004. The flood pulse concept: New aspects, approaches and applications—An
update. In: Proceedings of the Second International Symposium on the Management of Large Rivers for
Fisheries, pp. 117–149. Phnom Penh, Cambodia: Food and Agriculture Organization and Mekong River
Commission, FAO Regional Office for Asia and the Pacific.
Lang, M., Townsend, P.A., and Kasischke, E.S. 2008. Influence of incidence angle on detecting flooded forests
using C-HH synthetic aperture radar data. Remote Sensing of the Environment 112:3898–3907.
Latrubesse, E.M., Stevaux, J.C., and Sinha, R. 2005. Tropical rivers. Geomorphology 70(3–4):187–206.
Lu, D. 2006. The potential and challenge of remote sensing—Based biomass estimation. International Journal
Martinez, J.M. and Le Toan, T. 2007. Mapping of flood dynamics and spatial distribution of vegetation in the
Amazon floodplain using multitemporal SAR data. Remote Sensing of the Environment 108:209–223.
Melack, J.M. 2004. Remote sensing of tropical wetlands. In: S. Ustin (ed.) Manual of Remote Sensing, 3rd edn.
Vol 4. Remote Sensing for Natural Resources Management and Environmental Monitoring. John Wiley &
Sons, New York, pp. 319–343.
Melack, J.M. and L.L. Hess. 2010. Remote sensing of the distribution and extent of wetlands in the Amazon
basin. In: W.J. Junk, M. Piedade, F. Wittmann, J. Schöngart, and P. Parolin. Amazonian Floodplain
Forests: Ecophysiology, Ecology, Biodiversity and Sustainable Management. Ecological Studies,
Springer, pp. 43–59.
Melack, J.M., Hess, L.L., Gastil, M., Forsberg, B.R., Hamilton, S.K., Lima, I.B.T., and Novo, E.M.L.M. 2004.
Regionalization of methane emissions in the Amazon Basin with microwave remote sensing. Global
Change Biology 10(5):530–544.
Melack, J.M., Novo, E.M.L.M., Forsberg, B.R., Piedade, M.T.F., and Maurice, L. 2009. Floodplain ecosystem
processes. In: M. Keller, M. Bustamante, J. Gash, and P.S. Dias (eds.). Amazonia and Global Change,
p. 576. Washington, DC: American Geophysical Union.
Mertes, L.A.K., Daniel, D., Melack, J.M., and Nelson, B.W. 1995. Spatial patterns of hydrology, geomorphol-
ogy, and vegetation on the floodplain of the Amazon River in Brazil from a remote sensing perspective.
Geomorphology 13(1–4):215–232.
Miranda, N., Rosich, B., Meadows, P., Haria, K., Small, D., Schubert, A., Lavalle, M. et al. 2013. The ENVISAT
ASAR mission: A look back at 10 years of operation. In: Proceedings of ESA Living Planet Symposium
2013, pp. 1–17. Edinburgh, U.K.: European Space Agency.
Mitchard, E.T.A., Saatchi, S.S., White, L.J.T., Abernethy, K.A., Jeffery, K.J., Lewis, S.L., Collins, M. et al.
2012. Mapping tropical forest biomass with radar and spaceborne LiDAR in Lopé National Park, Gabon:
Overcoming problems of high biomass and persistent cloud. Biogeosciences 9(1):179–191.
Nascimento, W.R., Souza-Filho, P.W.M., Proisy, C., Lucas, R.M., and Rosenqvist, A. 2013. Mapping changes
in the largest continuous Amazonian mangrove belt using object-based classification of multisensor sat-
ellite imagery. Estuarine, Coastal and Shelf Science 117:83–93.
Neumann, M., Ferro-Famil, L., and Reigber, A. 2010. Estimation of forest structure, ground, and canopy
layer characteristics from multibaseline polarimetric interferometric SAR data. IEEE Transactions on
O’Grady, D., Leblanc, M., and Gillieson, D. 2013. Relationship of local incidence angle with satellite radar
backscatter for different surface conditions. International Journal of Applied Earth Observation and
Geoinformation 24:42–53.
Parolin, P., Ferreira, L.V., Albernaz, A.L.K.M., and Almeida, S.S. 2004. Tree species distribution in Várzea
forests of Brazilian Amazonia. Folia Geobotanica 39(4):371–383.
Peixoto, J.M.A., Nelson, B.W., and Wittmann, F. 2009. Spatial and temporal dynamics of river channel migra-
tion and vegetation in central Amazonian white-water floodplains by remote-sensing techniques. Remote
RADAMBRASIL. 1983. Levantamento de recursos naturais. Rio de Janeiro, RJ, Brazil.
Rennó, C.D., Novo, E.M.L.M., and Banon, L.C. 2013. Correção geométrica da máscara de áreas alagáveis da
bacia amazônica. In: Anais do XVI Simpósio Brasileiro de Sensoriamento Remoto—SBSR, pp. 5507–
5514. Foz do Iguaçu, Brazil: Instituto Nacional de Pesquisas Espaciais.
Richey, J.E., Melack, J.M., Aufdenkampe, A.K., Ballester, M.V.R., and Hess, L.L. 2002. Outgassing from
Amazonian rivers and wetlands as a large tropical source of atmospheric CO2. Nature 416:617–620.
Rodrigues, S.W.P. and Souza-Filho, P.W.M. 2011. Use of multi-sensor data to identify and map tropical coastal
wetlands in the Amazon of northern Brazil. Wetlands 31(1):11–23.
Rosenqvist, A., Forsberg, B.R., Pimentel, T., Rauste, Y.A., and Richey, J.E. 2002. The use of spaceborne radar
data to model inundation patterns and trace gas emissions in the central Amazon floodplain. International
Rosenqvist, A., Forsberg, B.R., Pimentel, T., and Richey, J.E. 1998. Using JERS-1 L-band SAR to estimate
methane emissions from the Jaú river floodplain (Amazon/Brazil). In: IGARSS’98. Sensing and Managing
the Environment. 1998 IEEE International Geoscience and Remote Sensing. Symposium Proceedings,
Vol. 3, pp. 1623–1625. Piscataway, NJ: IEEE.
Rosenqvist, A., Shimada, M., Ito, N., and Watanabe, M. 2007. ALOS PALSAR: A pathfinder mission for
global-scale monitoring of the environment. IEEE Transactions on Geoscience and Remote Sensing
45(11):3307–3316.
Rosenqvist, A., Shimada, M., Lucas, R., Chapman, B.D., Paillou, P., Hess, L.L., and Lowry, J. 2010. The Kyoto
and carbon initiative—A brief summary. IEEE Journal of Selected Topics in Applied Earth Observations
Rosenqvist, A., Shimada, M., Suzuki, S., Ohgushi, F., Tadono, T., Watanabe, M., Tsuzuku, K., Watanabe, T.,
Kamijo, S., and Aoki, E. in press. Operational performance of the ALOS global systematic acquisition
strategy and observation plans for ALOS-2 PALSAR-2. Remote Sensing of the Environment.
Rudorff, C.M., Melack, J.M., and Bates, P.D. 2014. Flooding dynamics on the lower Amazon floodplain: 1.
Hydraulic controls on water elevation, inundation extent, and river-floodplain discharge. Water Resources
Research 50(1):619–634.
Sartori, L.R., Imai, N.N., Mura, J.C., Novo, E.M.L.M., and Silva, T.S.F. 2011. Mapping macrophyte species
in the Amazon floodplain wetlands using fully polarimetric ALOS/PALSAR data. IEEE Transactions on
Satyamurty, P., Castro, A.A., Tota, J., Silva Gularte, L.E., and Manzi, A.O. 2009. Rainfall trends in the Brazilian
Amazon Basin in the past eight decades. Theoretical and Applied Climatology 99(1–2):139–148.
Silva, T.S.F., Costa, M.P.F., and Melack, J.M. 2009. Annual net primary production of macrophytes in the east-
ern Amazon floodplain. Wetlands 29(2):747–758.
Silva, T.S.F., Costa, M.P.F., and Melack, J.M. 2010. Spatial and temporal variability of macrophyte cover
and productivity in the eastern Amazon floodplain: A remote sensing approach. Remote Sensing of the
Silva, T.S.F., Melack, J.M., and Novo, E.M.L.M. 2013. Responses of aquatic macrophyte cover and productiv-
ity to flooding variability on the Amazon floodplain. Global Change Biology 19(11):3379–3389.
Sippel, S.J., Hamilton, S.K., Melack, J.M., and Novo, E.M.L.M. 1998. Passive microwave observations of
inundation area and the area/stage relation in the Amazon River floodplain. International Journal of
Souza-Filho, P.W.M. 2005. Costa de manguezais de macromaré da Amazônia: Cenários morfológicos,
mapeamento e quantificação de áreas usando dados de sensores remotos. Revista Brasileira de Geofísica
23(4):427–435.
Souza-Filho, P.W.M., Gonçalves, F.D., Rodrigues, S.W.P., Costa, F.R., and Miranda, F.P. 2009. Multi-sensor
data fusion for geomorphological and environmental sensitivity index mapping in the Amazonian man-
grove coast, Brazil. Journal of Coastal Research (56):1592–1596.
Souza-Filho, P.W.M., Paradella, W.R., Rodrigues, S.W.P., Costa, F.R., Mura, J.C., and Gonçalves, F.D. 2011.
Discrimination of coastal wetland environments in the Amazon region based on multi-polarized L-band
airborne synthetic aperture radar imagery. Estuarine, Coastal and Shelf Science 95(1):88–98.
Treuhaft, R.N. and Siqueira, P.R. 2000. Vertical structure of vegetated land surfaces from interferometric and
polarimetric radar. Radio Science 35(1):141–177.
Wang, Y., Hess, L.L., Melack, J.M., and Filoso, S. 1995. Understanding the radar backscattering from flooded
and nonflooded Amazonian forests: Results from canopy backscatter modeling. Remote Sensing of the
Wittmann, F., Anhuf, D., and Junk, W.J. 2002. Tree species distribution and community structure of central
Amazonian várzea forests by remote-sensing techniques. Journal of Tropical Ecology 18(6):805–820.
Wittmann, F., Schöngart, J., and Junk, W.J. 2010. Phytogeography, species diversity, community structure
and dynamics of central Amazonian floodplain forests. In: W.J. Junk, M.T.F. Piedade, F. Wittmann,
J. Schöngart, and P. Parolin (eds.). Amazonian Floodplain Forests: Ecophysiology, Biodiversity and
Sustainable Management, pp. 61–102. New York: Springer Verlag.
22 Mapping China’s Wetlands
and Recent Changes with
Remotely Sensed Data
Zhenguo Niu
CONTENTS
Introduction..................................................................................................................................... 473
Definition and Classification for Mapping Wetlands in China....................................................... 474
Wetland Definition..................................................................................................................... 474
Wetland Classification................................................................................................................ 474
Classification System for Mapping China’s Wetlands.......................................................... 475
Satellite Imagery Data Sources and Mapping Methods.................................................................. 476
Data Collection.......................................................................................................................... 477
Data Preprocessing..................................................................................................................... 477
Geometric Correction................................................................................................................. 478
Image Interpretation................................................................................................................... 478
Map Validation and Revision.....................................................................................................480
Wetland Change Methods..........................................................................................................480
Results............................................................................................................................................. 482
Geographical Distribution of Wetlands in China (2008)............................................................ 482
Wetland Changes between 1978 and 2008.................................................................................484
Comparison with Another Nationwide Inventory...................................................................... 488
Perspective on Visual Interpretation........................................................................................... 488
Need for Updated Wetland Mapping......................................................................................... 488
References....................................................................................................................................... 489
INTRODUCTION
As the nation with the third largest land area in the world, China contains many wetlands. The total
wetland area of China is more than 38 million ha, including about 36 million ha natural wetlands
and 2 million ha of constructed wetlands (e.g., rice fields and reservoirs) (Lei and Zhang 2005).
Wetlands cover about 3.8% of China’s land area, yet they provide 54.9% of ecosystem services
produced by the country’s ecological systems (Chen and Zhang 2000). All of the 26 natural and
9 human-made wetland types defined in the Ramsar international wetland classification can be
found in China (An et al. 2007; Lei and Zhang 2005). Of the total natural wetland area, 38.1%
(13.8 million ha) are marshes and swamps, 23.2% (8.4 million ha) are lakes, 22.7% (8.2 million ha)
are rivers, and 16.3% (5.9 million ha) are marine and coastal wetlands (SFA 2005).
473
China’s natural wetlands suffered tremendous loss over the past 50 years: 23.0% of freshwater
swamps, 16.1% of lakes, 15.3% of rivers, and 51.2% of coastal wetlands were lost (Wang 1998). Such
large-scale wetland destruction may have caused an estimated annual loss of ecosystem services
of $1.57 billion (U.S.$) (An et al. 2007). Wetland loss and degradation mainly resulted from land
reclamation, water diversion and dam construction, pollution, resource overuse, biological invasion,
and desertification and climate change (An et al. 2007; He and Zhang 2001).
Until recently, the Chinese government did not appreciate the importance of the full range of
ecosystem services provided by wetlands, especially the role of healthy wetland ecosystems in con-
tributing to sustainable economic development (An 2003; An et al. 2007). In 1972, the government
passed the first environmental law and started taking a series of actions to address environmental
degradation and wetland loss (An 2003). Since 2000, the Chinese government has implemented the
National Wetland Conservation Program—one of the largest of its kind in the world—with ambi-
tious goals, massive investments, and potentially enormous beneficial impacts (Wang et al. 2012). In
China, natural wetlands are protected under a three-class system: wetland reserves, wetland parks,
and scenic parks.
The outbreak of avian flu in 2003 raised concerns about the possible relationship between migra-
tory water bird habitat and the spread of avian flu, but Chinese scientists attempting to study this
relationship were stymied by the lack of any publicly available wetland map for China. Therefore,
in 2007, the State Key Laboratory of Remote Sensing Science (SKLRSS) at Chinese Academy of
Sciences commenced work to produce a set of comprehensive wetland maps for the country.
The first national wetland map was generated from satellite imagery for the year 2000. This was
the first year for which Landsat Enhanced Thematic Mapper Plus (ETM+) satellite imagery became
freely available in China. Upon completion of that map, a 1990 map was created to identify changes
that had occurred over the previous decade. The success of these two maps prompted an update of
the national map using 2008 imagery. Finally, a retrospective map was created for 1978, a year in
which far-reaching economic reforms altered land use practices in the country.
Due to changes in the availability of satellite imagery over these four dates, four different image
sources were used for the 1978, 1990, 2000, and 2008 maps: Landsat Multispectral Scanner (MSS),
Landsat Thematic Mapper (TM), Landsat ETM+, and China–Brazil Earth Resource Satellite
(CBERS-02B CCD), respectively. This chapter does not seek to evaluate the relative merits of the
different image sources because their choice was dictated by availability. Rather, the purpose of this
chapter is to discuss the operational procedures used to generate the series of four SKLRSS wetland
maps for 1978, 1990, 2000, and 2008 and to illustrate uses of the mapped data.
DEFINITION AND CLASSIFICATION FOR MAPPING WETLANDS IN CHINA

Wetland Definition
Given that there is no universally accepted definition of wetland, it is essential to clearly define
the term “wetland” for mapping purposes at the project’s outset (Scott and Jones 1995). Various
definitions of wetlands exist among the different domains and disciplines (Liu 1995). In view of the
fact that the Ramsar Convention has 168 contracting parties worldwide (The Ramsar Convention
on Wetlands 2014) and its definition of wetland was widely accepted in the world in many fields
(Scott and Jones 1995), the Ramsar definition of wetland was adopted by the SKLRSS for mapping
China’s wetlands.
Wetland Classification
Many countries have undertaken studies on wetland classifications (Cowardin and Golet 1995;
Cowardin et al. 1979; Gopal and Sah 1995; Kwi-Gon et al. 2006; National Wetlands Working
Group 1997; Pressey and Adam 1995; Scott and Jones 1995; Semeniuk and Semeniuk 1995;
Mapping China’s Wetlands and Recent Changes with Remotely Sensed Data 475
see Chapter 2). One of the most comprehensive and widely applauded wetland classification
systems is that developed for the United States by Cowardin et al. (1979). Under the auspices
of the Ramsar Convention, an internationally accepted classification system was developed by
Scott (1989), which was loosely based on the U.S. system (Scott and Jones 1995). Twelve types
of coastal wetlands, twenty types of inland wetlands, and ten types of artificial wetlands were
included in this classification frame (see Chapter 2) (Ramsar Convention Secretariat 2010; Scott
and Jones 1995).
Wetland classification research in China can be dated back to the 1960s. Various classification
systems were established for specific places such as Sanjiang Plains and Zoige Plateau (Chai and
Jin 1963; Yang 1988). Traditionally, peatlands have received the most attention (Long et al. 1983;
Niu and Ma 1985), and these have been classified into three categories: eutrophic mire, mesotrophic
mire, and oligotrophic mire (Long et al. 1983). An Asian-wide wetlands inventory conducted from
1986 to 1989 utilized a classification system with 22 categories of wetlands (Scott 1989). Lu (1995)
proposed a national classification system that divided Chinese natural wetlands into three catego-
ries, while a more recent classification included 5 major categories and 28 types of wetlands, all in
accordance with the “Ramsar Convention” system (Chen and Huang 1995; Liu and Ma 2006; Tang
and Huang 2003).
Given the existence of all these classification systems, it is likely that no single classification
will meet all needs of different wetland inventories. Rather, it is recommended that a classification
suited to the purposes of a particular inventory should be chosen or developed. Three principles
were considered for developing a national wetland classification system for the China wetland map-
ping initiative:
1.
Generalization. A national classification should be sufficiently generalized to include all
wetland categories, but not directly define any specific wetland types.
2.
Compatibility. The classification should be comparable with international wetland clas-
sification at first level (i.e., Ramsar system) and should also be dividable into more detailed
categories that are comparable with local wetland classifications.
3.
Operability. The classification should be applicable to wetland mapping by remote sens-
ing, and not merely a perfect theoretical classification that cannot be applied directly. For
example, “shallow water below six meters at low tides” is encompassed in the Ramsar
Convention definition, but cannot be directly mapped by optical remote sensing methods,
and was not included in the classification. Rice field is another wetland type that was not
included in the classification, because its locations usually change rapidly from year to
year, especially in northern China.
Classification System for Mapping China’s Wetlands

The classification system adopted for the SKLRSS mapping effort contained three major catego-
ries: coastal wetland, inland wetland, and artificial wetland. They encompass 15 wetland types, of
which 13 were applied (Table 22.1). Two wetland types, shallow coastal waters and paddy fields,
were not applied: the former because of the difficulty of identifying the seaward limit of shallow
coastal waters and the latter because agricultural maps already provided reliable depictions of
paddy fields. This system aims to not only be compatible with the international convention but
can also be conveniently converted to local classification systems through rational combination of
categories.
The first level of this classification, which includes three wetland types (coastal wetland, inland
wetland, and artificial wetland), is identical to main categories of the Ramsar Convention to facili-
tate comparison of wetland distribution among different countries. The second level is more specific,
to make the scheme more operational during national wetland mapping using remote sensing. The
second level was an abridged edition of classification system adopted by the Ramsar Convention and
the State Forestry Administration (SFA).
TABLE 22.1
Classification of Wetlands in China
Code Class Code Subcategory Description
1 Coastal 11 Tidal zone/shallow Marine beach areas with <30% vegetation cover including
wetland beach/beach marine subtidal aquatic beds, rocky marine shores, sand,
shingle or pebble shores, intertidal mud, and sand or salt flats.
12 Coastal marshes/ Vegetation cover exceeds 30% including mangroves, giant
swamps reeds, salt marshes, meadows, and swamps.
13 Estuarine water From zero tidal difference river to freshwater fronts in the
coastal water—permanent water of estuaries.
14 Estuarine deltas/ Estuarine systems of deltas, sandy beaches, and islands
sandy islands (including under water), and vegetation cover <30%.
15 Lagoons Brackish to saline lagoons with at least one relatively narrow
connection to the sea.
2 Inland 21 Rivers Permanent or intermittent running water bodies (>90 m in
wetland width, >5 km in length).
22 Flooded wetland Seasonally inundated grassland (including natural wet
meadows), shrublands, woodlands and forests, lakehead river
deltas, foothill flood fan, and river beaches.
23 Lakes Water-covered areas including shallow water mud flats.
24 Inland marshes Permanent and seasonal marshes including moss, grass, shrubs
and forest wetlands, saline marshes and green island
wetlands, and spring wetlands (water coverage <30%).
3 Artificial 31 Reservoirs/ponds Artificial wetland for water storage and hydropower generation
wetland and ponds for freshwater fish farming (<8 ha).
32 Artificial river Water irrigation and transportation channels.
channels
33 Seawater fish farm Artificial wetland for fish farming, salt production ponds.
and salt flats
34 Rice and paddy Rice fields or water-covered crop fields or crop fields with
fieldsa winter water storage.
35 Landscaping and Artificial wetland for urban landscaping, water recreation.
recreational
water bodies
36 Other Water bodies caused by mining or other engineering projects,
wastewater processing fields.
a Not included: (1) shallow water in coastal wetlands as specified in subtype 11 and (2) rice and paddy fields as specified
in subtype 34.
SATELLITE IMAGERY DATA SOURCES AND MAPPING METHODS

Because wetlands are usually located in remote areas and are inaccessible by traditional inves-
tigation methods, remote sensing techniques and Geography Information System (GIS) have
been proven to be effective and have been widely employed around the world to map wetlands
(Kuenzer et al. 2011; Ozesmi and Bauer 2002; Premalatha et al. 2010; Rebelo et al. 2009). Various
remote sensing datasets have been used for mapping wetlands at global, regional, national, and
local levels (Ramsar Convention Secretariat 2010). The selection of datasets generally depends
on the following factors: (1) program objectives and requirements, (2) sensor characteristics, and
(3) the program’s budget and cost of datasets.
Data Collection
As mentioned, different sources of satellite imagery were used to produce the national wetland
maps for years 1978, 1990, 2000, and 2008 maps: MSS, Landsat TM, ETM+, and CBERS-02B
CCD, respectively. All Landsat image data used for the SKLRSS wetland map (MSS for 1978,
TM for 1990, and ETM+ for 2000) were obtained from the University of Maryland (ESDI 2014)
(Table 22.2) (Niu et al. 2012). Landsat 7 ETM+ imagery could not be used for producing the 2008
map because the system’s scan line corrector failed in 2003 resulting in an estimated 22% loss of
data for any given scene (SLC-off Products 2014). Consequently, CBERS-02B CCD image data
derived from a sensor having similar spatial and spectral resolution to Landsat TM (Tables 22.3 and
22.4) were used to develop a 2008 wetland map of China (China Centre for Resources Satellite Data
and Application 2014). However, more than 1440 scenes of CBERS-02B images were required to
cover China due to its small swath width. Besides the image datasets, other auxiliary datasets such
as topographical data—Shuttle Radar Topography Mission (SRTM)—soil data, vegetation data,
land use/cover data, and the national marshland dataset were also collected to assist in interpretation
of images and validate the resulting map.
Data Preprocessing
Atmospheric correction of images was undertaken based on the Fast Line-of-sight Atmospheric
Analysis of Spectral Hypercubes (FLAASH) method with ENVI (The Environment for Visualizing
Images) software (Exelis Visual Information Solutions, Inc.). The ZYC_AUTOPREPRO program
TABLE 22.2
Collections of Satellite Images Used for Mapping Wetlands in China
Correction
Date (Reference Year) Data Sources Data Volume (Scenes) Geometric Error Interpretation Method
1978 Landsat MSS 546 <1 pixel Manual
1990 Landsat TM 542 <1 pixel Manual
2000 Landsat ETM+ 597 — Manual
2008 CBERS-02B 1442 <2 pixels Manual/semiautomatic
TABLE 22.3
Technical Specification of Payloads of CBERS-02B
Spectral Spatial Swath Side- Repetition Data
Band Range Resolution Width Looking Cycle Transmission
Satellite Payload No. (µm) (m) (km) Ability (Days) Rate (Mbps)
CBERS-02B CCD B01 0.45–0.52 20 113 ±32° 26 2 × 53
camera
B02 0.52–0.59 20
B03 0.63–0.69 20
B04 0.77–0.89 20
B05 0.51–0.73 20
HR camera B06 0.5–0.8 2.36 27 104 60
Wide field B07 0.63–0.69 258 890 5 1.1
imager B08 0.77–0.89 258
(WFI)
TABLE 22.4
Orbital Characteristics of CBERS-02B
Satellite CBERS-02B
Orbit Sun synchronous recurrent frozen orbit
Altitude 778 km
Inclination 98.5°
Repetition cycle 26 days
Descending node (local time) 10:30 AM
On-board capacity CCD: ≥47.7 GB HR: ≥54 GB
was developed based on ENVI and IDL functions by incorporating data uncompressing, band
stacking, format conversion, image radiometric calibration, revised FLAASH algorithm, and topo-
graphic calibration, to automatically calibrate images for atmospheric correction (Gong et al. 2013;
Tang et al. 2014).
Geometric Correction
Landsat TM images around 1990 served as reference images, and the other period images (1978,
2000, and 2008) were corrected by using the image-to-image method with ENVI software. The
georegistration error was constrained to within two pixels for CBERS-02B images and less than one
pixel for ETM and MSS as measured by root-mean-square error (Table 22.2).
The original projection of each scene of image was a Universal Transverse Mercator (UTM)
projection, and there were 10 zones across China. The images were first mosaicked within the
same zone based on geometric images, and then their projection was transformed from UTM to
Albers equivalent conical projection for each mosaic zone image. Finally, all transformed images
were mosaicked together into one huge image for China. The purpose of mosaicking the whole
images was to avoid the duplication of each original scene of images, which would lead to incon-
sistent, even conflicting interpretation results when based on original images with different acqui-
sition dates. Images obtained within the wet season were preferred over images obtained within
the dry season. The national image was divided into four parts and then smaller slices to be
interpreted visually based on natural environment characteristics and the processing capability of
individual computers.
Image Interpretation
While SPOT satellite data have been successfully used to identify 12 wetland classes with 80.9%
accuracy in Florida’s Everglades (Rutchey and Vilcheck 1994), given the heterogeneity of China’s
landscape, manual interpretation was chosen over automation to map wetland vegetation. This
allowed interpreters lacking expertise in remote sensing image processing to analyze the imagery
and produce the maps (Johnston and Barson 1993). For this project, most of the classification was
done by visual interpretation using ENVI4.3 and ArcMap9.1.
Because water absorbs most energy in the near-infrared spectrum, near-infrared bands are
usually employed to distinguish wetland from upland. The most important Landsat TM band for
wetland identification is band 5 because of its ability to discriminate vegetation and soil mois-
ture levels. Landsat TM bands 5, 3, and 4 are usually the best combination for wetland detection
(Ozesmi and Bauer 2002). Band 1 should be added if water quality is of interest (FGDC 1992).
Landsat TM bands 5, 4, and 3 plus CBERS-02B CCD bands 4, 3, and 2 were used during the visual
interpretation.
The spatial resolution of Landsat TM and ETM+ is 30 m, MSS is 80 m, and CBERS-02B is 20 m.

A 3 pixels width was established as the threshold for river width and 10 pixels × 10 pixels (9 ha) as
the minimum mapping unit for lakes, which controls the final mapping precision. The following
steps outline the interpretation process (Niu et al. 2009):
1. Identified image signatures associated with different wetland types for each geographic
region of China (Figure 22.1). The recommended band combination was adjusted as neces-
sary during on-screen digitizing to maximize wetland recognition. The range of map scale
was restricted so that the operators could not zoom in or out unlimitedly.
2. Divided the work among eight working groups, with each group responsible for a
geographic area. Groups exchanged their results periodically (weekly initially and
later monthly) for quality checking and to answer interpretation questions. While six
bands were recommended, the interpreters could change this combination according to
their working area. (Note: There are six bands to be utilized for Landsat TM and four
bands for CBERS-02B CCD.) All visual interpretation was completed on-screen using
Environmental Systems Research Institute (ESRI) ArcGIS 9.1, yet some other tools were
developed to facilitate interpretation, such as customization of interface of ArcMap,
dynamically linking ArcMap with Google Earth. Feature Analyst was also used to
(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
FIGURE 22.1 Examples of satellite images (Landsat ETM + bands 4, 3, and 2) used for visual interpreta-
tion of different wetland types. (a) Coastal wetland, (b) river in North China plain, (c) river in mountains,
(d) wetland in desert, (e) park in city, (f) lake in mountains, (g) marshes in Northeast China, (h) lake in North
China, and (i) peatland in Zoige plateau.
facilitate extraction of water from CBERS CCD images. Feature Analyst (Overwatch
System) is the software plug-in for ESRI ArcGIS® geospatial analysis software, provid-
ing accelerated feature extraction of all image features. A semiautomated method for
extraction of water from CBERS-02B CCD images was developed to increase efficiency,
in contrast to completely visual interpretation for Landsat images. Wetland water bodies
of every type were first classified using the Feature Analyst software and then identified
one by one according to the image and compared with the previous map. In the second
step, the other wetland types were also interpreted manually based on the 1990 wetland
map. Finally, after validation, all wetland types were merged.
3. Performed quality assurance by examining the interpretations and making necessary
corrections.
4. Combined the results from all the groups using ArcGIS 9.1. All interpretation results (vec-
tor format) were merged by using merge module in ArcGIS and edge-matched and adjusted
as necessary to guarantee seamless integration.
5. Trained image analysts then compared the wetland map to relevant geographic data such
as elevation and slope data and the marshland database and made necessary adjustments.
The masks of slope data layer (less than 3°) and marshland data layer were overlaid with
the images and wetland map. Google Earth provided high-resolution (HR) images for spe-
cific locations that aided image interpretation and validation. The interpretation results
were converted to KMZ files that were overlaid on Google Earth for performing additional
quality control.
6. Constructed the China wetlands database using tools in ArcMap (e.g., built the topology,
corrected errors in map units, calculated areas, and assigned attribute values). All maps
were produced in vector format and stored as a shape file.
Map Validation and Revision

Wetland reconnaissance for the 2008 map was undertaken collaboratively by 11 teams from differ-
ent universities and institutes from July 2009 to September 2009. Six representative regions, includ-
ing the Qinghai–Tibet Plateau, the Northeast Plain, the Yellow River Delta in Shandong, Dongting
and Poyang lakes, coastal wetland in Jiangsu, and southeast mangrove wetlands in Guangdong,
were chosen for field investigation (Figure 22.2). More than 10,000 field photos were taken and 459
qualified samples were collected to validate the 2008 wetland map (Niu et al. 2012).
The overall accuracy of wetland and nonwetland was 0.98. There were only seven wetland sam-
ples along rivers that were mistakenly interpreted as nonwetland. However, the overall accuracy of
wetland types was 0.70 and kappa coefficient is 0.63 (Tables 22.5 and 22.6). The average producer
accuracy was 0.58 for nine wetland types. The seawater fish farm/salt flat type had the highest
producer accuracy with 0.95, followed by inland marshes with producer accuracy of 0.83. The
flooded wetland had the lowest producer accuracy of 0.08. The main reason of this low accuracy
was confusion between flooded wetlands and rivers, lakes, and marshlands that could be caused by
the wetland seasonal fluctuation and discrepancy between image acquisition date and investigation
date. Another reason was related to the difficulty of identification of flooded wetlands based on
visual interpretation of image, where the vegetation and water content were hard to identify. It is
also reasonable that reservoirs/ponds, inland marshes, and lakes had the high user accuracy (0.93,
0.83, and 0.81) because they were all easily identified by visual identification on-screen, even when
the images were not atmospherically corrected.
Wetland Change Methods

After validation of the wetland map in 2008, the other three maps (circa 2000, 1990, and 1978) were
intersected, and a cross-table of change between every two-period wetland type was constructed.
B C
A
1:1,043,878 1:7,682,969 1:6,367,550
C
E
B
D
A
1:2,307,415 1:31,435,701
FIGURE 22.2 Reconnaissance of wetlands sites in 2009 showing field investigation sites and routes under-
taken in 2009. A, Leizhou Peninsula; B, Tibet Plateau; C, Great Khingan in Inner Mongolia; D, Yellow River
Delta in Shandong and coastal wetlands in Jiangsu Province; E, Sanjiang Plain in Heilongjiang Province;
F, Poyang Lake and Dongting Lake in Hubei and Jiangxi provinces.
According to the change model of wetland types between different periods, combinations were
identified as impossible wetland transformation, such as from coastal wetlands to inland wetlands,
and were rechecked on the images, and the corresponding map was revised accordingly. Auxiliary
materials and literature related to wetland distribution were collected to validate regions we could
not confirm during the visual interpretation. The Google Earth information was fully utilized for
the visual interpretation and validation processes.
The variable spatial resolution of the different satellite images used for the map of each period
inevitably introduced new errors during result comparisons. To avoid these problems, window
filtering based on results was employed. Interpretation results were first rasterized according to
the corresponding spatial resolution of the satellite image. Then, different window size filtration
was undertaken, and resultant raster maps at 240 m resolution were produced. In each filtration
TABLE 22.5
Contingency Table of Validation for Visual Interpretation of Wetland in 2008 Showing
In-Field Classification (Ground Reference) versus Interpreted Category (Classification)
Ground Reference
Wetland Type (code) w21 w22 w23 w24 w31 w33 w11 w12 w13 Total
Classification
Unclassified 0 4 0 3 0 0 0 0 0 7
River (21) 58 17 4 10 0 0 0 0 0 89
Flooded wetland (22) 11 3 2 12 0 0 0 0 0 28
Lake (23) 1 1 59 0 11 0 0 1 0 73
Inland marshes (24) 8 10 6 119 0 0 0 1 0 144
Reservoirs/ponds (31) 1 0 1 0 43 0 0 1 0 46
Seawater fish farm/salt flats (33) 0 0 0 0 16 19 1 2 0 38
Tide zone/shallow beach (11) 0 1 1 0 2 1 12 3 1 21
Coastal marshes/swamps (12) 0 0 0 0 0 0 3 7 0 10
Estuarine water (13) 2 0 1 0 0 0 0 0 0 3
Total 81 36 74 144 72 20 16 15 1 459
TABLE 22.6
Accuracy of 2008 Wetland Map
Wetland Type Producer Accuracy Omission User Accuracy Commission
River 0.72 0.28 0.65 0.35
Flooded wetland 0.08 0.92 0.11 0.89
Lake 0.80 0.20 0.81 0.19
Inland marshes 0.83 0.17 0.83 0.17
Reservoirs/ponds 0.60 0.40 0.93 0.07
Seawater fish farm/salt flats 0.95 0.05 0.50 0.50
Tide zone/shallow beach 0.75 0.25 0.57 0.43
Coastal marshes/swamps 0.47 0.53 0.70 0.30
Estuarine water 0 1.00 0.00 1.00
window, only the wetland type whose area equaled that of the window can be retained. The win-
dow moved one by one by window size (not by pixels) during the recalculation, and the last row
and line were discarded if they cannot fill the window. The wetland area of the resultant map
would obviously be less than the area of the original after these filtration calculations, because
substantial minor wetland patches were inevitably ignored.
RESULTS
Geographical Distribution of Wetlands in China (2008)
The total national wetland area in 2008 was about 324,097 km2 (Table 22.7). Not surprisingly given
the country’s huge land area, inland wetlands (including rivers and lakes) predominated, occupying
about 85% of China’s wetland area (Figure 22.3). Inland marshes/swamps were the most common
type, comprising an area of about 11,385 km2 or 35% of national wetland area. Artificial wetland
TABLE 22.7
Extent of China Wetlands by Type in 2008 Based on Detailed
Analysis of CBERS-02B Imagery and Field Verification
Wetland Type Area (km2)
Tidal zone/shallow beach 2,478
Coastal marshes and swamps 1,958
Estuarine water 5,148
Estuarine delta 337
Lagoons 114
Coastal wetland total 10,036
Rivers 48,553
Flooded wetland (floodplain) 30,292
Lakes 82,708
Inland marshes (including swamps) 113,852
Inland wetland total 275,404
Reservoirs/ponds 23,697
Artificial river channels (including canals) 3,137
Seawater farm and salt flats 11,773
Landscaping and recreational water bodies 41
Others 9
Artificial wetland total 38,657
All wetlands (excluding rice and paddy fields) 324,097
Note: Sums are computer generated and may not sum exactly due to round-off procedures.
Distribution of China’s wetlands by major type
12%
Inland marshes/swamps
3% 35%
9% Lakes
15% Rivers
26% Floodplains
Coastal wetlands
Artificial wetlands
FIGURE 22.3 Wetland distribution by type for China in 2008. Note: For purposes of this inventory, rivers
and lakes are considered inland wetlands. (Redrawn from Niu, Z. et al., Chin. Sci. Bull., 57(22), 2813, 2012.)
area represented about 12%. Reservoirs/pools comprised about 23,697 km2 in area and were the
most widely distributed type among artificial wetlands. However, coastal wetland area, which
excluded shallow sea water area to 6 m depth at low tide, only accounted for about 10,035 km2 or
only 3% of China’s wetlands.
Most wetlands occurred in Heilongjiang, Inner Mongolia, Qinghai, and Tibet provinces,
with their combined wetland area totaling about 177,800 km2, or about 55% of national wet-
land area (Figure 22.4). Xinjiang, Jiangsu, and Jilin were next, with total wetland area of 16%.
70°E 80°E 90°E 100°E 110°E 120°E 130°E

50°N
50°N
Wetland map in 2008
40°N
40°N
30°N
30°N
Tidal zone/beaches
Coastal marshes
Estuarine water
Estuarine delta
Lagoons
River
Flooded wetland
Lake
20°N Inland marshes
Reservoir/pool
20°N
Artificial channel
Coastal fish farmland/salt field
0 410 820 km
Landscape water
Others 1:83,202,173
90°E 100°E 110°E 120°E
FIGURE 22.4 China wetland map in 2008. (From Niu, Z. et al., Chin. Sci. Bull., 57(22), 2813, 2012.)
Hubei province—known as the “province of a thousand lakes”—had about 8030 km2 of wetland,

only 2.48% of the national area. Because of agricultural development for thousands of years, espe-
cially over the last 30 years, there is almost no natural wetland in the plains of eastern China.
All coastal wetlands are located in eastern China, including the following provinces: Liaoning,
Hebei, Shandong, Tianjin, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, Guangxi, Hainan,
Taiwan, Hong Kong, and Macao. The wetland area in northern Hangzhou Bay was about 5863 km2,
representing 60% of China’s coastal wetlands. Most coastal wetlands were in Liaoning (18%),
Shandong (30%), Jiangsu (16%), and Fujian (15%) provinces. Among the coastal wetlands, estua-
rine water, tidal zone/beach, and coastal marshes were the most common types, occupying 51%,
28%, and 20%, respectively. (Note: Estuarine water was very difficult to distinguish from river
and shallow water in Landsat TM and CBERS CCD images, so the area of estuarine water was
much more uncertain, and, according to our estimations, its area also had not changed as quickly
in recent years.)
Wetland Changes between 1978 and 2008

China’s total wetland area decreased by 101,399 km2 from 1978 to 2008 based on the wetland map
comparison (Figure 22.5; Table 22.8). The rate of wetland loss slowed greatly over time: 5523 km2/
year from 1978 to 1990, 2847 km2/year from 1990 to 2000, and 831 km2/year from 2000 to 2008
Changes in China’s wetlands

1978–2008
350,000
300,000
250,000
200,000 Inland marshes/swamps

km2
Rivers/floodplains
150,000
Artificial wetlands
100,000
Lakes
50,000
Coastal wetlands
0
(a) 1978 1990 2000 2008
Changes in wetland types

1978–2008
160,000
140,000
120,000 1978
100,000
km2
80,000 1990
60,000 2000
40,000
20,000 2008
0
s
s
s
ps
s
in
nd
ke
nd
am
la
La
la
la
dp
et
et
w
/s
w
oo
lw
es
al
/fl
ta
sh
ci
rs
as
ifi
ar
ve
Co
rt
m
Ri
A
nd
(b)
la
In
FIGURE 22.5 Wetland changes from 1978 to 2008 in China: (a) overall changes graphed through time and
(b) highlighting changes in individual types. Note: declines in all types except artificial wetlands.
(Figure 22.5a). This could be attributed to increasingly protective efforts of the Chinese government
(An 2003; Wang et al. 2012). Another possible explanation is that there is no more available wetland
to be exploited in developed eastern China, and hostile environments in sparsely populated western
China impede wetland exploitation. The highlights of this analysis are largely presented in tables;
for details including a discussion of the causes, see Niu et al. (2012).
Inland marshes/swamps experienced trends of change that were similar to that of total wetlands
(Figure 22.5b). Losses of inland marshes/swamps decreased tremendously, by 20% and 26% during
the periods 1978–1990 and 1990–2000, respectively. However, loss rates stabilized between 2000
and 2008 (Table 22.8).
Seasonal fluctuation of water level lead to increased uncertainties in floodplain identification,
especially for distinguishing floodplains from rivers. Thus, floodplains and rivers were com-
bined as river wetlands during the wetland change analysis. River wetlands and lakes decreased
significantly from 1978 to 1990 and continued to decrease considerably from 1990 to 2008
(Table 22.8).
Natural wetland area continually and strongly decreased over the past 30 years, from 299,504 km2
in 1978 to 186,152 km2 in 2008 (Table 22.8; Figure 22.6). In contrast, there was a continual increase
in artificial wetland, from 9,793 km2 (in 1978) to 21,745 km2 (in 2008). Nearly all lost natural
wetlands (98%) were converted to nonwetlands before 1990. The conversion ratio then gradually
TABLE 22.8
Summary of Wetland Area across Different Periods in China (Area in km2)
Wetland Changes
Wetland
Type Subtype Area in 1978 1978–1990 1990–2000 2000–2008
Coastal wetland Tide zone/shallow beach 3,399 1,014 −1,202 −1,368
Marine marshes/mangrove 4,363 −2,297 −810 254
Estuarine water 4,987 −517 −328 21
Estuarine delta 140 178 37 −66
Lagoons 215 −19 −51 −60
Sum 13,104 −1,641 −2,355 −1,218
Inland wetland River wetlands a 65,293 −27,936 −1,955 −3,779
Lakes 82,903 −12,463 −2,609 −3,436
Inland marshes/swamp 138,203 −26,894 −28,995 −70
Sum 286,400 −67,293 −33,559 −7,285
Artificial wetland Reservoirs/ponds 5,157 212 5,338 1,780
Artificial channel 616 −521 140 125
Seawater fish farms/salt 3,984 2,976 1,966 −42
flats
Other wetlandsa 35 −6 −5 −12
Sum 9,793 2,661 7,439 1,851
Total area 309,297 −66,273 −28,475 −6,652
Negative numbers indicate a loss of area. The “window filtering” procedure resulted in locating less wetland area in
2008 than the recorded on the original 2008 wetland map (Table 22.7) that was more detailed.
a We combined floodplains with rivers as river wetlands and landscaping and recreational water bodies and others
as other wetlands.
30
2%
Area (× 10,000 km2)
Natural wetlands
20
14%
98%
10 23%
86% 77% Conversion to
nonwetlands
Artificial wetlands Conversion to
artificial wetlands
0
1975 1980 1985 1990 1995 2000 2005 2010
FIGURE 22.6 Wetland-type conversions in China showing wetland loss and proportions of conversion from
wetlands to nonwetlands. (From Niu, Z. et al., Chin. Sci. Bull., 57(22), 2813, 2012.)
dropped to 86% from 1990 to 2000 and to 77% after 2000. This likely indicates that agricultural
exploitation chiefly contributed to the lost wetlands during early periods, and more artificial wet-
lands were converted from natural wetlands in subsequent periods.
Wetland changes between provinces varied greatly from 1978 to 2008, accompanied by rapid
social and economic development (Table 22.9). The majority of the wetland changes were in prov-
inces with the most wetlands, such as Heilongjiang, Inner Mongolia, and Tibet. In addition, the
directions of wetland change were also not identical over the different periods since 1978. Wetland
changes in direction and amount provide a comprehensive overview that may reflect different char-
acteristics of natural environment and social economy in those provinces.
TABLE 22.9
Wetland Changes Based on Provinces from 1978 to 2008 (Area in km2)
Wetland Changes
Province 1978–1990 1990–2000 2000–2008
1. Heilongjiang −9,808 −13,766 −1,325
Inner Mongolia −11,347 −7,771 −2,479
2. Xinjiang −8,609 564 −160
Qinghai −9,601 337 −1,855
Tibet −6,380 1m674 4,387
3. Hebei −242 −755 129
Gansu −1,436 −585 −97
Beijing 23 −30 −70
Shanxi −329 −50 −38
Tianjin 104 104 −275
Zhengjiang −943 5 63
Shaanxi −588 −306 −28
Jiangsu −619 28 −334
Shandong 1,239 −740 682
Anhui −980 −93 −514
Henan −425 −236 101
Ningxia −340 −273 50
Hunan −2,094 401 −1,416
Sichuan −2,203 −65 −755
Hubei −883 −189 −1,832
Jiangxi −206 −1,260 −813
Jilin −3,701 −5,166 866
Liaoning −1,978 −393 −220
Chongqing −351 −49 98
Shanghai −169 −278 −36
Yunnan −565 76 −146
Guizhou −202 25 4
Fujian −775 170 −420
Taiwan −483 2 −102
Guangxi −1,112 140 −230
Guangdong −895 −60 −67
Hong Kong −32 2 0
Macao 8 −1 2
Hainan −350 61 −122

Comparison with Another Nationwide Inventory
Since wetland mapping in this study initially was designed to identify wild bird habitats across
China, the inventory used a different classification system from the one utilized by China SFA
(SFA 2000). This created differences in wetland and water areas across China (Niu et al. 2009).
For example, coastal wetland area from the SFA report was 59,400 km2 compared to 17,610 km2
of coastal wetland area from our wetland map in 2000. The SFA survey had included three sub-
categories (shallow marine water, marine subtidal aquatic beds, and coral reefs) of coastal wetland
that are not considered in our mapping. At the same time, our inventory reported 25,377 km2 area
of artificial wetlands; that was also less than mapped by SFA survey. But our inventory detected
36,765 km2 more in inland wetlands than the SFA result.
In the National Wetland Resources Survey and Monitoring Technique Regulations established
by the SFA (SFA 2000), the river wetland investigation was limited to main rivers whose stream
order was larger than four level, with their average river bed (dry channels) width greater than 10 m
and area greater than 100 ha. If reservoirs located within the river were smaller than 100 ha, they
were counted as rivers, but the minimal area for lakes and ponds was not specified. Due to the
spatial resolution constraint of remotely sensed data used in our study, rivers narrower than 90 m in
width and wetlands smaller than 9 ha in area were not included.
Despite these differences, wetland geographic distribution and wetland change analysis based on
these maps could still be effective and helpful to understand the wetland situation in China and for
wetland management and decision making.
Perspective on Visual Interpretation

Visual interpretation allows the interpreters to use their knowledge of both wetland ecosystem and
images to produce a reliable wetland map and has a long history of usage for mapping both at
large scale and for complex regions. With advanced remote sensing technologies, visual interpre-
tation is an often-overlooked method. Although it is more labor and time intensive, it usually is
more efficient than automated classification, especially since automated classification of images
often requires repeated processing to achieve good performance for a specific region and cannot be
applied to other regions in many cases without major modification.
There are other limitations to visual interpretation besides cost and labor. First, updating maps
is equally time-consuming and expensive, and it is virtually impossible to replicate the knowledge
of interpreters to an automated system. Next, the boundary lines drawn by individual interpreters
usually lead to differences on precision of map units that is especially noted during edge match-
ing of work by different individuals. While visual interpretation on the computer screen is easier
to operate zooming in or zooming out than working with aerial photos, operating in different map
scales can lead to different precision of map unit even where a minimum map unit is defined before-
hand. Fortunately, this situation can be controlled using GIS tools, such as extent modules in ESRI
ArcGIS 9.1 (ESRI). Consistency among interpreters is another consideration when using visual
interpretation techniques. Training and development of standard methods are vital to producing a
consistent product as are quality control procedures.
Need for Updated Wetland Mapping

Given the central government’s interest and investment in protecting wetlands, China’s wetland
maps will need to be updated (Wang et al. 2012). Local government and various stakeholders
are focused on developing the economy, often at the expense of natural resources including wet-
lands. Consequently, wetland monitoring will be needed. Recognizing the limitations of visual
interpretation, an automated/semiautomated classification method will be necessary for updating

the existing maps. Improved technologies for image processing and data management and acces-
sibility of tremendous amount of satellite images and auxiliary datasets will facilitate application of
remote sensing technology for these updates.
ACKNOWLEDGMENTS
I am deeply grateful to Professor Ralph Tiner, who not only provided me with the opportunity to
introduce research on mapping China’s wetlands but also gave many constructive suggestions and
edited the chapter carefully. I owe my deepest gratitude to Professor Carol Johnston, who helped me
revise the draft manuscript and provided generous support during my visiting at South Dakota State
University. I would also like to give thanks to Professor Peng Gong, who suggested this project and
gave many important suggestions and support.
REFERENCES
An, S. 2003. Ecological Engineering of Wetlands. Chemical Industry Press, Beijing, China (in Chinese).
An, S., H. Li et al. 2007. China’s natural wetlands: Past problems, current status, and future challenges. AMBIO
36:335–342.
Chai, X. and S. Jin. 1963. Peatland types and its evolution in Zoige Plateau. Acta Geographica Sinica
29(3):219–240.
Chen, J.W. and G.L. Huang. 1995. Research of wetland classification system and indexes in China (in Chinese).
Management of Forestry Research 1995:65–71.
Chen, Z. and X. Zhang. 2000. The value of the ecosystem services of China. Chinese Science Bulletin 45:17–22.
China Centre for Resources Satellite Data and Application. 2014. Technical specification of payloads of
CBERS-01/02. http://www.cresda.com/n16/n92006/n92066/n98590/index.html. Accessed January 17,
2014.
Habitats of the United States. FWS/OBS 79/31. U.S. Department of the Interior, Fish and Wildlife
Service, Washington, DC.
Cowardin, L.M. and F.C. Golet. 1995. US-Fish-and-Wildlife-Service 1979 wetland classification: A review.
Plant Ecology 118(1–2):139–152.
ESDI (Earth Science Data Interface at Global Land Cover Facility). 2014. http://glcfapp.glcf.umd.edu:8080/
esdi/index.jsp. Accessed January 17, 2014.
Federal Geographic Data Committee. 1992. Application of satellite data for mapping and monitoring wet-
lands. Technical Report 1, Wetlands Subcommittee. Federal Geographic Data Committee (FGDC),
Washington, DC.
Gong, P., J. Wang et al. 2013. Finer resolution observation and monitoring of global land cover: First mapping
results with Landsat TM and ETM+ data. International Journal of Remote Sensing 34(7):2607–2654.
Gopal, B. and M. Sah. 1995. Inventory and classification of wetlands in India. Vegetatio 118:39–48.
He, Y. and M.X. Zhang. 2001. Study on wetland loss and its reasons in China. Chinese Geographical Science
11(3):241–245.
Johnston, R.M. and M.M. Barson. 1993. Remote sensing of Australian wetlands: An evaluation of Landsat
TM data for inventory and classification. Australian Journal of Marine and Freshwater Resources
44:235–252.
Kim, K.-G., M.-Y. Park, and H.-S. Choi. 2006. Developing a wetland-type classification system in the Republic
of Korea. Landscape and Ecological Engineering 2:93–110.
Kuenzer, C., A. Bluemel et al. 2011. Remote sensing of mangrove ecosystems: A review. Remote Sensing
3(5):878–928.
Lei, K. and M.X. Zhang. 2005. The resource and conservation suggestion of wetlands in China (in Chinese).
Wetland Science 3:81–86.
Liu, H.T. 1995. Definition and types of wetland (in Chinese). Chinese Journal of Ecology 14:73–77.
Liu, Z.G. and X.H. Ma. 2006. Classification of wetland (in Chinese). Wetland Science Management 2:60–63.
Long, H., W. Zhu, and S. Jin. 1983. Chinese Mires (in Chinese). Shandong Science and Technology Press,
Jinan, China. 269pp.
Lu, J. 1995. Ecological significance and classification of Chinese wetlands. Vegetatio 118:49–56.
National Wetlands Working Group. 1997. The Canadian Wetland Classification System, 2nd edn. B.G. Warner
and C.D.A. Rubec (eds.). University of Waterloo, Wetlands Research Centre, Waterloo, Ontario, Canada.
Niu, H. and X. Ma. 1985. Peatlands in Our Country [M]. Commercial Press, Beijing, China. p. 180 (in Chinese).
Niu, Z., P. Gong et al. 2009. Geographical characteristics of China’s wetlands derived from remotely sensed
data. Science in China Series D: Earth Sciences 52(6):723–738.
Niu, Z., H. Zhang et al. 2012. Mapping wetland changes in China between 1978 and 2008. Chinese Science
Bulletin 57(22):2813–2823.
10(5):381–402.
Premalatha, M., T. Abbasi et al. 2010. Applications of GIS in wetland management: An overview. Research
Journal of Chemistry and Environment 14(4):87–102.
Pressey, R.L. and R. Adam. 1995. Review of wetland inventory and classification in Australia. Vegetatio
118:81–101.
Ramsar Convention Secretariat. 2010. Wetland inventory: A Ramsar framework for wetland inventory and
ecological character description. Ramsar Handbooks for the Wise Use of Wetlands, 4th edn., vol. 15.
D. Pritchard (ed.). Ramsar Convention Secretariat, Gland, Switzerland.
Rebelo, L.M., C.M. Finlayson et al. 2009. Remote sensing and GIS for wetland inventory, mapping and change
analysis. Journal of Environmental Management 90(7):2144–2153.
Rutchey, K. and L. Vilcheck. 1994. Development of an Everglades vegetation map using a SPOT image and the
Global Positioning System. Photogrammetric Engineering and Remote Sensing 60:767–775.
Scott, D.A. 1989. A Directory of Asian Wetlands. IUCN, ICBP and IWRB, Cambridge University Press,
Cambridge, U.K. pp. 129–294.
Scott, D.A. and T.A. Jones. 1995. Classification and inventory of wetlands: A global overview. Vegetatio
118:3–16.
Semeniuk, C.A. and V. Semeniuk. 1995. A geomorphic approach to global classification for inland wetlands.
SFA (State Forestry Administration). 2000. China National Wetlands Conservation Action Plan [M]. China
Forestry Publishing House, Beijing, China. pp. 1–25 (in Chinese).
SFA (State Forestry Administration). 2005. China Forestry Yearbook 2005. China Forestry Press, Beijing,
China (in Chinese).
SLC-off Products: Background. http://landsat.usgs.gov/products_slcoffbackground.php. Accessed January 7,
2014.
Tang, P., H.W. Zhang et al. 2014. Practice and thoughts of the automatic processing of multispectral images
with 30 m spatial resolution. Journal of Remote Sensing 18(2):232–253.
Tang, X.P. and G.L. Huang. 2003. Study on classification system for wetland types in China (in Chinese).
Forest Research 16:531–539.
The Ramsar Convention on Wetlands. 2014. The Ramsar Convention And Its Mission. http://www.ramsar.org/
about/the-ramsar-convention-and-its-mission. Accessed October 7, 2014.
Wang, Y. 1998. Wetlands in China are confronting threats of degradation and loss [N]. China Green Times,
1998.2.9 (in Chinese).
Wang, Z., J. Wu, M. Madden, and D. Mao. 2012. China’s wetlands: Conservation plans and policy impacts.
AMBIO 41:782–786.
Yang, Y. 1988. Ecological classification of mire in the Sanjiang Plain. Geographical Research 7(1):27–35 (in
Chinese).
23 Mapping Invasive
Wetland Plants
Carol A. Johnston
CONTENTS
Introduction..................................................................................................................................... 491
Mapping Invasive Species from Herbarium Records..................................................................... 494
Using Image Time Series to Document the Spread of Invasive Plants........................................... 494
Timing Imagery to Capture Phenological Events........................................................................... 497
Identifying Invasive Species from Multispectral Imagery.............................................................. 498
Landsat....................................................................................................................................... 498
Single-Date IKONOS and QuickBird........................................................................................ 498
Multidate QuickBird.................................................................................................................. 501
Identifying Invasive Species from Hyperspectral Imagery............................................................. 505
Identifying Invasive Species Using Imaging Radar........................................................................ 505
Modeling Invasive Plant Environments to Predict Potential Location...........................................506
Conclusion...................................................................................................................................... 507
References....................................................................................................................................... 507
INTRODUCTION
Invasive plants are species or genotypes that can thrive and spread aggressively outside their natural
range. They cause problems in wetlands and other habitats worldwide by displacing native flora,
reducing biodiversity, impeding boat traffic, and degrading wetland habitat values (Jakubauskas
et al. 2002; Madden 2004). Some plants are intentionally introduced and become invasive in the
absence of the natural pests that controlled their spread in their natal range. Other invasive plants
are unintentionally introduced and escape and spread by competitive dispersal mechanisms, such
as the copious seed production of purple loosestrife (Lythrum salicaria) (Galatowitsch et al. 1999).
Submersed aquatic invasives including common watermilfoil (Myriophyllum spicatum) can be
spread by fragments that cling to boat trailers, taking advantage of human mobility (Johnson et al.
2001). Wetland invaders generally have high productivity under nutrient enrichment that allows
them to take advantage of disturbed conditions (Zedler and Kercher 2004).
The terms “noxious weeds” and “nonnative plants” are related to invasive plants (Table 23.1).
“Noxious weeds” are a special category of invasive plants that are subject to regulation or quar-
antine because of their adverse impacts, particularly to agriculture (NRCS 2013). “Nonnative”
(i.e., nonindigenous, exotic, alien, or introduced) refers to a plant that was introduced from another
part of the world. Nonnative status does not necessarily mean that a plant is invasive. Interestingly,
most introduced plant species never become invasive. It is also a fact that not all invasive plants are
nonnative. For example, reed canary grass (Phalaris arundinacea) is native to the United States and
Canada, but is considered invasive because it takes over wetlands, forms monotypes, and resists
control efforts (Healy and Zedler 2010). Some species, such as common reed (Phragmites aus-
tralis), have U.S. native genotypes that are not invasive but Eurasian genotypes that are invasive
(Saltonstall 2002). The cattail family (Typhaceae) is particularly complex because hybridization
491
TABLE 23.1
Invasive Species Discussed in This Chapter and States in Which They Have
Been Declared Noxious
Scientific Name Common Name State Noxious Status
Arundo donax Giant reed grass TX
Colubrina asiatica Lather leaf FL
Eichhornia crassipes Water hyacinth AL, AZ, CA, CT, SC, TX
Hydrocharis morsus-ranae European frogbit CA, ME, VT, WA
Lepidium latifolium Perennial pepperweed AK, CA, CO, CT, HI, ID, MA, MT, NV, NM, OR, SD, UT,
WA, WY
Lythrum salicaria Purple loosestrife AL, AZ, AR, CA, CO, CT, FL, ID, IN, IA, MA, MI, MN, MO,
MT, NC, NE, NV, NM, ND, OH, OR, PA, SC, SD, TN, TX,
UT, VT, VA, WA, WI, WY
Nelumbo lutea American lotus CT
Phalaris arundinacea Reed canary grass CT, MA, WA
Phragmites australis Common reed AL, CT, MA, SC, SD, VT, WA
Spartina alterniflora Smooth cordgrass OR, WA
Trapa natans European water chestnut AL, AZ, CT, ME, MA, OR, SC, VT, WA
Typha angustifolia, Narrow-leaved cattail, —
Typha x glauca hybrid cattail
Source: NRCS, Introduced, invasive, and noxious plants, Natural Resource Conservation Service, U.S. Department of
Agriculture, Washington, DC, http://plants.usda.gov/java/noxComposite, accessed March 2013, 2013.
between native, noninvasive broad-leaved cattail (Typha latifolia), and introduced narrow-leaved
cattail (Typha angustifolia) produces the invasive blue cattail (Typha x glauca) (Olson et al. 2009;
Snow et al. 2010; Kirk et al. 2011). By increasing vigor and broadening ecological tolerances, this
hybridization may serve as a primary driver of invasiveness (Travis et al. 2010). The southern
cattail (Typha domingensis) is native and not normally considered invasive, but its spread in the
Everglades in response to nutrient enrichment is displacing indigenous saw grass (Cladium jamai-
cense) marshes.
Plant invasiveness varies with location. For example, American lotus (Nelumbo lutea; Figure 23.1)
is on the noxious weed list for Connecticut because it is potentially invasive there, yet the same spe-
cies is considered threatened or endangered in the states of Michigan, New Jersey, and Pennsylvania.
Smooth cordgrass (Spartina alterniflora), a keystone species in U.S. Atlantic coast salt marshes, is
an invader of salt marshes and tidal flats on the U.S. Pacific coast and elsewhere in the world (Civille
et al. 2005; Cheng et al. 2006; Rosso et al. 2006). Subtropical wetlands such as those in Florida are
particularly prone to species invasion because of their mild climate.
Mapping of invasive wetland plants helps scientists and managers document their distribution
and monitor their migration. Many invasive grasses reproduce vegetatively by belowground rhi-
zomes that allow them to spread rapidly once established (Travis et al. 2011). Control efforts are
more effective when new invasions are detected early. Mapping of invasive wetland plants can also
help identify corridors that facilitate their spread, such as roadside ditches (Maheu-Giroux and
Blois 2005; Jodoin et al. 2008). The presence of invasive species is often the first readily detectable
sign of declining health in wetlands (Hatch and Bernthal 2008).
Several websites provide excellent information on invasive plant biology and ecology. The USDA
National Invasive Species Information Center serves as a gateway to invasive species informa-
tion, particularly in the United States (NISIC 2013). The Invasive Species Specialist Group of the
International Union for Conservation of Nature maintains the Global Invasive Species Database
Mapping Invasive Wetland Plants 493
FIGURE 23.1 Flower of Nelumbo lutea. (Photo courtesy of Mirela Tulbure.)
(GISD 2013). Information and literature about invasives and other aquatic plants are available from
the Aquatic Plant Information Retrieval System (APIRS 2013).
This chapter focuses on mapping invasive emergent plants in northern U.S. freshwater wetlands
using case studies from my own research and examples from the literature. Multiple mapping meth-
ods and sensors are discussed (Table 23.2). For a more complete review of the invasive plant remote
sensing literature, see “Applications of Remote Sensing to Alien Invasive Plant Studies” (Huang
and Asner 2009).
TABLE 23.2
Characteristics of Imagery Discussed in This Chapter
Imagery Platform Sensor Type Pixel Resolution (m) Acronym
Aerial photos Aircraft Digital camera Variable —
ALOS PALSAR Satellite Imaging radar 10–20 Advanced Land Observing Satellite
Phased Array Type L-Band Synthetic
Aperture Radar
HyMap Aircraft Hyperspectral 2–10 —
Hyperion Satellite Hyperspectral 30 —
IKONOS Satellite Multispectral 4 —
Landsat ETM+ Satellite Multispectral 30 Enhanced Thematic Mapper Plus
Landsat MSS Satellite Multispectral ~60 Multispectral Scanner
Landsat TM Satellite Multispectral 30 Thematic Mapper
LiDAR Aircraft Laser Variable —
MODIS Satellite Multispectral 250, 500, 1000 Moderate Resolution Imaging
Spectroradiometer
PROBE-1 Aircraft Hyperspectral 1–10 —
QuickBird Satellite Multispectral 2.4 and 2.8 —
SPOT HRV Satellite Multispectral 10 and 20 Satellite Pour l’Observation de la
Terre, High-Resolution Visible
MAPPING INVASIVE SPECIES FROM HERBARIUM RECORDS

Herbarium records can provide important information about the range and spread of invasive plants.
When entered into a geographic information system (GIS), georeferenced herbarium specimens can
be used to plot the current and past distribution of invasive species and uncover plant–environment
relationships that might not otherwise be observed. An advantage of this approach is that it can
provide a historical perspective of invasive species spread extending back more than a century,
preceding the first aerial photographs. Several studies have demonstrated the utility of natural his-
tory collections for determining the distribution of invading plant species (Pyšek and Prach 1993;
Delisle et al. 2003; Wu et al. 2005; Lavoie et al. 2009). Delisle et al. (2003) identified 1963–1984
as a period of P. australis expansion in southern Quebec based on 533 specimens stored in seven
herbaria of Quebec and the Canadian Government. Herbarium specimens can also be valuable
for identifying environmental preferences of invasive plant species, particularly if the collection
location is digitally georeferenced in a GIS (Funk et al. 1999; Boylen et al. 2006; Lambert and
Casagrande 2006; Elith and Leathwick 2007).
The lack of adequate georeferencing is a barrier to the use of herbarium specimens for mapping.
Collectors of herbarium specimens were often students or “citizen scientists,” who may not have
followed a particular protocol for identifying the place of collection. In the midwest and western
United States, the Public Land Survey System (PLSS) was often used to identify location to the
nearest square mile section, but that system does not exist for eastern U.S. collections. Location
was often described relative to natural features in the landscape, using local place names that may
lack specificity or no longer be in use. The creation of global positioning system (GPS) technology
has revolutionized the ability to identify plant specimen location since the 1990s (Johnston 1998).
Ground locations could be accurately documented via handheld GPS instruments. Georeferenced
imagery such as Google Earth (Google 2014) has been used to retroactively georeference herbar-
ium specimens of P. australis in South Dakota wetlands (Johnston and Miller 2011). To determine
the point locations of herbarium specimens identified by township and range, a tool from Earth
Point (Earth Point 2014) was used to view a township, range, and section grid in Google Earth.
Within the section, a more precise location was found by reviewing the field location notes and
observing the Google Earth image. Point locations were assigned using the placemark tool in
Google Earth, which generated latitude and longitude coordinates in decimal degrees. For loca-
tions identified only by place names (e.g., lake or creek), place names were zoomed to and exam-
ined on Google Earth imagery to determine the precise point location of the specimen. Locations
in decimal degrees to three decimal places were imported into ArcMap 9.3 (ESRI, Redlands,
CA), where they could be joined with collection information typed from the specimen label. This
approach allowed display of specimen locations by date of collection and associated water body
type (Figure 23.2).
USING IMAGE TIME SERIES TO DOCUMENT THE SPREAD

OF INVASIVE PLANTS
Some invasive wetland species take advantage of exposed mudflats, quickly establishing when
low water levels provide a fertile substrate for germination (Tulbure and Johnston 2010). This is
particularly problematic in the Great Lakes, where water levels fluctuate by 0.3–0.5 m annually
due to seasonal hydrologic changes and by a meter or more over longer time periods due to cli-
matic variation (see Chapter 15, Figure 23.3). Aerial images taken during successive years can be
used to document rapid vegetation changes that could signal the establishment of invasive plants.
For example, the Point au Sable wetland on the east shore of Green Bay, Wisconsin, was an open
water lagoon in 1999 (Figure 23.3a), but by 2005 it was a jungle of invasive Typha x glauca and
Phragmites (Figure 23.3b). Field work verified that invasive plants filled the former lagoon (Figure
23.3c) (Tulbure et al. 2007). The lake level had fallen by a meter between 1999 and 2003, exposing
Lake
Wetland
Creek
Ditch
Stock dam
Other
0 25 50 100 150 200
(a) km
M
iss
ou
ri
Ri
ve
r
Collection date
Prior to 1980
1980s
1990s
2000s
0 25 50 100 150 200
km
(b)
FIGURE 23.2 Herbarium specimen locations of Phragmites australis in South Dakota, classified by (a)
associated water body and (b) collection date. (Reprinted from Johnston, C.A. and Miller, K.I. Prairie Nat.,
43, 38, 2011. With permission.)
the lagoon’s bottom sediments to invasion. Other researchers have used time series of aerial pho-
tos or satellite imagery to document the expansion of Phragmites in Great Lakes coastal wetlands
(Arzandeh and Wang 2003; Wilcox et al. 2003).
Fluctuating Great Lakes water levels naturally cause expansion and contraction of emergent
vegetation in coastal wetlands, so changes in the extent of emergent vegetation are not necessarily
invasive species. Shifting of plant zones with more aquatic beds during wetter years and more emer-
gent wetlands during drier years is the norm. Field inspection or additional analysis may be needed
to distinguish invasion from natural variation. For example, a time series of five aerial images taken
between 1974 and 1992 was used to map the migration of wet meadow versus cattail marsh in three
Green Bay wetlands, and these data were used to derive highly significant linear regressions relating
meadow and marsh area to water level (Frieswyk and Zedler 2007). However, an anomalous pattern
emerged in 2000, when the meadow area was smaller than predicted by the linear equation despite
low water levels. The authors determined that in 2000, the meadow area was being squeezed out
by the landward advance of marsh habitat as the invasive hybrid Typha x glauca displaced native
T. latifolia.
Researchers working for the South Florida Water Management District used time series of aer-
ial photos and satellite imagery to document the spread of T. domingensis in saw grass marshes of
0 0.1 0.2 km 0 0.1 0.2 km
(a) (b)
(c)
FIGURE 23.3 Rapid invasion of Point au Sable wetland by Phragmites: (a) The lagoon in the middle of Point
au Sable was open water as of 1999 (USGS orthophoto). (b) Open water was replaced by Phragmites, visible
as senescent detritus in this June 2005 image (USDA NAIP photo). (c) The author verifies robust Phragmites
stand in the field. (Photo courtesy of Mirela Tulbure.)
the Everglades due to phosphorus enrichment. They initially mapped changes within the Florida
Everglades Water Conservation Area 2A using Landsat Multispectral Scanner (MSS) (1973,
1976, and 1982) and Satellite Pour l’Observation de la Terre, High-Resolution Visible (SPOT
HRV) multispectral data (1987 and 1991) (Jensen et al. 1995). A later analysis delineated cat-
tail stands using conventional air photo interpretation of 1995 color-infrared aerial photography,
comparing the results with the digital image processing of the 1991 SPOT image, and found that
overall accuracies were 95.2% and 83.4%, respectively (Rutchey and Vilchek 1999). The 1991
SPOT classified image overestimated cattail coverage due to interacting confounding mecha-
nisms such as floating periphyton masses that formed around the stems of emergents and varying
growth morphology of emergent stands (e.g., cattail can occur in stands that are sparse to dense,
short to tall, and clumped to monotypic).
TIMING IMAGERY TO CAPTURE PHENOLOGICAL EVENTS

Some invasive wetland plants exhibit periodic life-cycle (phenological) events that distinguish them
from other species. For example, the colorful flower spike of L. salicaria enables identification from
a distance, particularly when it grows in nearly monotypic stands (Figure 23.4). One study found
that vertical small format (35 mm) aerial photography taken from a small plane could be success-
fully applied to discriminate flowering loosestrife plants from other background wetland vegetation
(Frazier and Moore 1993).
Invasive species that initiate growth earlier in spring and continue to grow late in fall effec-
tively extend their growing season, allowing them to outgrow natives by photosynthesizing at a
similar rate but for a longer time (Zedler and Kercher 2004). This trait is well documented for
P. arundinacea—mid-October is best for distinguishing this species from other wetland vegetation
because of its late senescence (Bernthal and Willis 2004).
Nonpersistent emergents exhibit the phenological trait of having their aboveground biomass fall
to the surface of the substrate or below the surface of the water at the end of the growing season so
that, at certain seasons of the year, there is no obvious sign of emergent vegetation (Cowardin et al.
1979). This trait was used to identify beds of N. lutea from QuickBird images taken in summer and
FIGURE 23.4 Flower spikes of Lythrum salicaria. (Photo courtesy of Mirela Tulbure.)
spring (Ghioca-Robrecht et al. 2008). It was accurately mapped on multiseason imagery because
these beds were fully developed in September but had not yet emerged above water in early April
(Figure 23.5).
IDENTIFYING INVASIVE SPECIES FROM MULTISPECTRAL IMAGERY

Landsat
Multispectral satellite imagery from the Landsat Thematic Mapper (TM) and Enhanced Thematic
Mapper Plus (ETM+) sensors (30 m pixels) has been available since the early 1980s and has been
tested for a number of wetland remote sensing applications (Ozesmi and Bauer 2002). Landsat is a
joint initiative between the U.S. Geological Survey (USGS) and the National Aeronautic and Space
Administration (NASA) and is the world’s longest continuously acquired collection of space-based
moderate-resolution land remote sensing data. The successful deployment of Landsat 8, which
began normal operations on May 30, 2013, will continue this program into the future.
In Wisconsin, Landsat imagery was shown to be effective for identifying P. arundinacea, the
state’s most widespread wetland invader (Bernthal and Willis 2004). Bands 1–5 and 7 were analyzed
using an unsupervised classification algorithm known as Iterative Self-Organizing Data Analysis
Technique (ISODATA) in ERDAS 8.4 image processing software to produce the initial classifica-
tion. Band 6, the thermal band, was left out due to its lower resolution. From the continuous spectra
of all pixels in the scene, clusters were assigned to preliminary classes. These classes formed the
basis for further analysis in a “maximum-likelihood” supervised classification to refine the output.
The classification produced a GIS raster coverage of wetlands mapped to a 0.5 acre minimum map-
ping unit with a three-category Phalaris cover classification: dominant (80%–100% cover), codomi-
nant (50%–79% cover), and absent to subdominant (0%–49% cover). Ground-truthing fieldwork
was conducted at 249 plots and showed the classification to have an overall accuracy of 71%. For
the cover classes heavily dominant, codominant, and subdominant, respectively, the user’s accura-
cies were 86%, 41%, and 69%. Because the accuracy was high for both the heavily dominant and
subdominant classes, the authors rated this classification a success (Bernthal and Willis 2004).
Building upon this earlier success, the Wisconsin Department of Natural Resources (WIDNR)
went on to apply the technique to the entire state (Hatch and Bernthal 2008). Landsat imagery
was acquired to meet the following conditions: (1) as close to mid-October as possible, (2) 10%
or less of cloud cover, and (3) taken within 1999–2003. Landsat-7 ETM+ imagery was planned
for use across the entire state, but its scan line corrector malfunctioned on May 31, 2003 (USGS
2014), necessitating the use of Landsat-5 TM imagery for the 2003 images. Twelve scenes from
the Landsat 7 (ETM+) and Landsat 5 (TM) satellites (all taken during October of 1999–2003)
were selected and used to analyze 96% of Wisconsin’s wetlands (20,500 km2). Small portions of
the state had to be masked out due to cloud cover. Following the suggestion from the pilot study,
a simplified two-category classification was used for the statewide classification: (1) areas with
<50% cover of Phalaris and (2) areas with ≥50% cover. The resulting raster GIS map identified
2016 km2 of wetlands that were dominated by Phalaris (Figure 23.6). The accuracy assessment
revealed a range of 61%–83% overall accuracy and 72%–92% user’s accuracy for the reliability of
areas mapped as reed canary grass–dominated wetlands in the individual Landsat scenes. WIDNR
reported many uses for the map, such as wetland assessment, wetland restoration, watershed plan-
ning, research into control of Phalaris, and outreach programs on invasive species (Hatch and
Bernthal 2008).
Single-Date IKONOS and QuickBird

Although Landsat’s 30 m resolution works well for species that occupy large contiguous expanses
of wetland, finer resolution imagery is desirable to classify wetland plants that occur in smaller
Legend Sept. NIR Sept. NDVI

Water
Lotus
Nonpersistent
0 100 200 300 400

Meters
(a) (b)
April NIR
(c)
FIGURE 23.5 Nelumbo lutea beds near Gard Island in Erie Marsh, showing appearance on selected
QuickBird image layers and field determination of reference point classes: (a) Nelumbo beds are clearly visible
on September NIR image. (b) Bright photosynthesizing vegetation contrasts with dark water on September
NDVI image, and tree shadows are minimized. (c) Nelumbo beds are undetectable on April image, prior to
emergence. (Reprinted from Ghioca-Robrecht, D.M. et al., Wetlands, 28, 1028, 2008. With permission.)
16
21 22
Legend
Value
<50% RCG
≥50% RCG
Meters
0 200 400 600 800
FIGURE 23.6 Reed canary grass abundance determined by the WIDNR for a portion of Lebanon Township,
superimposed on a USGS digital raster graphic background.
patches. IKONOS and QuickBird are commercial earth observation satellites that provide this finer
resolution: 4 m for IKONOS and 2.4 or 2.8 m for QuickBird (DigitalGlobe, Longmont, CO). In
contrast to Landsat, IKONOS and QuickBird supply only four multispectral bands: blue, green, red,
and near infrared (NIR).
Researchers in Minnesota used IKONOS imagery to map Typha and five other wetland plant
types at Swan Lake (Sawaya et al. 2003). Typha was classified with 100% producer’s accuracy (i.e.,
the probability that field sites known to be Typha were correctly classified as Typha) and 88.9%
user’s accuracy (i.e., the probability that pixels mapped as Typha were actually Typha in the field).
Overall accuracy (i.e., field sites correctly classified/total number of field sites) for the six cover
types was 79.5%. IKONOS was also useful for mapping water hyacinth (Eichhornia crassipes)
that was blocking navigation on the Rio Grande river in Texas, but no error analysis was provided
(Jakubauskas et al. 2002).
European frogbit (Hydrocharis morsus-ranae) is an invasive, floating-leaved species of concern
in Ontario, Quebec, and New York state. QuickBird multispectral imagery for a 6 km section of
the South Nation River in Ontario was used to determine if the spectral signature of frogbit was
separable from other wetland vegetation (Proctor et al. 2012). A three-stage supervised fuzzy clas-
sification methodology based on fuzzy segmentation, feature analysis, and defuzzification was used
to conduct a land cover classification that included a species-level class for frogbit (see Benz et al.
2004 for details on fuzzy analysis methods). Classification results indicated this methodology pro-
duced a good approximation of the frogbit spatial distribution, with user’s and producer’s accuracies
of 81.0% and 77.9%, respectively.
A study of diverse tidal wetlands of the Hudson River National Estuarine Research Reserve
applied a maximum-likelihood classification on QuickBird to estimate the presence of three
invasive species: L. salicaria, P. australis, and water chestnut (Trapa natans) (Laba et al. 2008).
QuickBird was a relatively reliable data source for mapping invasive wetland plants (accuracy ≥
65%). QuickBird data also proved highly accurate (accuracy ≥ 86%) for delineating invasive giant
reed (Arundo donax) along a riparian zone in Texas (Everitt et al. 2005).
Despite the overall benefits of fine-scale imagery, spectral confusion can cause misclassification
between open water and shadows cast by trees (Sawaya et al. 2003). Since tree shadows are subpixel
features in coarser-scale imagery, they present less of a classification problem on imagery such as
Landsat.
Multidate QuickBird
The use of multidate QuickBird imagery combines the benefits of capturing phenological events
with those of fine-scale multispectral imaging. A diked wetland at the western end of Lake Erie was
mapped using QuickBird images acquired during peak biomass in early September 2002 and the pre-
growing season in early April 2003 (Figure 23.7). The imagery was evaluated for mapping emergent
wetland vegetation communities, particularly invasive species such as Phragmites and Typha spp.
(Ghioca-Robrecht et al. 2008). An eight-layer stack of all four bands from both dates was initially
used for image analysis, but there was confusion of dark tree shadows with water during test runs.
Reference
point
Lake
Erie
North Maumee Bay
Gard Island
0 450 900 1350 1800
Meters
FIGURE 23.7 False-color infrared QuickBird image of Erie Marsh study site taken in September 2002,
showing locations of reference points used to verify classification accuracy. Red color denotes photosynthesiz-
ing vegetation.
To reduce this confusion, a normalized difference vegetation index (NDVI) layer was computed for
the September image (Figure 23.5b) and added as a ninth layer to the image stack. After masking
out uplands and the open water of the North Maumee Bay and Lake Erie (Figure 23.7), unsuper-
vised classification (ISODATA) with 100 classes was performed in ERDAS IMAGINE™ 8.7 (Leica
Geosystems, Atlanta, GA). The 100 ISODATA classes were grouped into eight classes using a clas-
sification scheme developed from ground reconnaissance and quantitative vegetation cover data.
The classification distinguished three monodominant genera (Phragmites, Typha, and Nelumbo),
three multigenera plant communities (wet meadow, nonpersistent emergents, and woody vegeta-
tion), and two unvegetated cover types (water and soil) (Figure 23.8).
The spectral characteristics of the eight wetland classes were very different in September versus
April due to the overwinter change from photosynthesizing vegetation to dead plant litter or water
(Figure 23.9). In the NIR band, the lotus and nonpersistent emergent classes changed the most, from
photosynthesizing vegetation in September to open water areas with very low NIR values in April
(see Figure 23.5). The soil/cultivated class exhibited a similarly large NIR change with planted
crops present in September and plowed soil in April. Stands confirmed in the field to be Phragmites
and Typha had distinct September NIR digital number (DN) values, with the average DN values
for Phragmites being 69% greater than those of Typha. Both Phragmites and Typha had similar
NIR DN values in April. Woody vegetation had September NIR DN values intermediate between
those of Phragmites and Typha and lower September DN values in the visible light bands than other
vegetation types. The wet meadow class dominated by Phalaris differed from the other vegetation
classes in April due to its high DN values in the visible light bands. This difference was attributed
to the lack of subcanopy water and the reflectivity of prostrate grass remains from the previous
Legend
Water
Soil/cult.
Woody
Lotus
Nonpersistent
Phragmites
Typha
Meadow
Meters
0 500 1000 1500 2000
FIGURE 23.8 Vegetation map of Erie Marsh generated using late summer and spring QuickBird images.
growing season. The wet meadow class also had higher NIR DN values in April than all other
classes, which may be due to earlier green-up of vegetation in these drier portions of the wetland.
Predictably, water had the lowest NIR DN values of any class on both dates (Figure 23.9).
An accuracy assessment of the vegetation map was done by collecting ground reference data at
196 randomly selected points within the marsh (Figure 23.7). The random locations were generated
by computer and uploaded into a handheld GPS unit (Geo XH 2005 Series, Trimble Navigation
Limited, Sunnyvale, CA), which was used to navigate to the accuracy assessment points in the field
(Johnston et al. 2009). Data collection was done without knowledge of the mapped class that might
bias the assignment. An error matrix of the image analysis yielded an overall accuracy of 64%, but
450
n=5
Green band April
400 September
n = 10
350
n = 16
Average digital number (DN)
n = 19
300
n = 39 n = 11
n = 17 n=4
250
200
150
100
50
0
t
a
lt.
w
dy
s
er
en
tu
ite
ph
do
cu
oo
at
Lo
ist
gm
Ty
ea
il/
W
rs
ra
M
So
pe
Ph
on
(a)
N
400
Red band n=5
April
350 September
300
250 n = 10
n = 16 n = 39 n = 11
200 n = 17
n = 19
150 n=4
100
50
0
lt.
a
er
w
y
s
nt
tu
ite
d
cu
ph
do
at
te
oo
Lo
gm
il/
Ty
W
ea
is
W
So
rs
ra
M
pe
Ph
on
(b)
N
FIGURE 23.9 Digital numbers (mean + SE) for (a) green band, (b) red band. (Continued )
900
NIR band n=4 April
n = 16
800 September
n = 39
700 n=5
600 n = 10
n = 17
500
n = 11
400
300
n = 19
200
100
a
er
lt.
w
dy
s
t
tu
ite
en
ph
do
cu
at
oo
Lo
gm
ist
Ty
W
ea
il/
rs
ra
M
So
pe
Ph
on
(c)
N
FIGURE 23.9 (Continued) Digital numbers (mean + SE) for (c) NIR band at points confirmed to be cor-
rectly mapped within the eight wetland classes. (Reprinted from Ghioca-Robrecht, D.M. et al., Wetlands, 28,
1028, 2008. With permission.)
TABLE 23.3
Confusion Matrix Resulting from an Unsupervised Classification with Eight Final Classes
Based on QuickBird High-Resolution Imagery of Erie Marsh North Maumee Bay, Michigan
Reference Classification
User’s
Unsupervised Classification 1 2 3 4 5 6 7 8 Row Total Accuracy (%)
1—Water 19 1 1 21 90
2—Soil 17 1 1 1 1 21 81
3—Woody 17 1 2 20 85
4—Nelumbo 13 2 5 20 65
5—Other nonpersistents 3 1 1 4 11 3 1 24 17
6—Phragmites 1 2 39 7 2 51 76
7—Typha 1 8 12 21 57
8—Meadow 1 1 8 3 5 18 28
Column Total 22 19 22 14 8 74 28 9 196
Producer’s accuracy (%) 86 90 77 93 50 53 43 56 Overall accuracy 64
Source: Springer and the original publisher (With kind permission from Springer Science + Business Media: Wetlands,
Assessing the use of multiseason QuickBird imagery for mapping invasive species in a Lake Erie coastal marsh,
28, 2008, 1036, Ghioca-Robrecht, D.M., Johnston, C.A., and Tulbure, M.G., Table 3 © 2008.)
map accuracy varied substantially among the eight classes mapped (Table 23.3). The biggest error
was caused by confusion of vegetation management with phenology: large areas of Phragmites
had been intentionally burned between the September and April image dates and were incorrectly
classified as nonpersistent emergent species because the dark, burned soil surface had a DN simi-
lar to that of water. As a result, the nonpersistent emergent class had a very low user’s accuracy.
The method demonstrated moderate success at distinguishing the three graminoid invasives (i.e.,
Phragmites, Typha, and Phalaris).
IDENTIFYING INVASIVE SPECIES FROM HYPERSPECTRAL IMAGERY

In contrast to broadband multispectral imagery (e.g., Landsat TM and QuickBird), hyperspectral
imagers such as HyMap and PROBE-1 contain hundreds of narrow spectral bands usually located
in the visible to mid-infrared (0.4–2.5 μm) portion of the electromagnetic spectrum (Adam et al.
2010). Airborne hyperspectral sensors have been evaluated for mapping coastal wetland vegeta-
tion, which often includes invasive Phragmites (Bachmann et al. 2002; de Lange et al. 2004; Lopez
et al. 2004; Artigas and Yang 2005; Becker et al. 2007). Bachmann et al. (2002) distinguished
Phragmites from other wetland species with 68% accuracy using the HyMap airborne sensor with
a spatial resolution of 4.5 m. Lopez et al. (2004) were able to distinguish between monodominant
stands of Phragmites, Typha, and mixed wetland vegetation with 91% accuracy using the PROBE-1
airborne hyperspectral sensor with 5 m resolution. Other invasive wetland plant species that have
been experimentally mapped using hyperspectral sensors include Asiatic nakedwood (Colubrina
asiatica), E. crassipes, and broad-leaved pepperweed (Lepidium latifolium) (Hirano et al. 2003;
Hestir et al. 2008). However, the high cost and limited coverage of airborne sensors limit their use
for wider-scale management and study.
The Hyperion sensor aboard NASA’s EO-1 satellite is an experimental sensor that merges the
spectral resolution of airborne hyperspectral instruments with the practicality of satellite remote
sensing (USGS 2011). The applicability of Hyperion (30 m pixels) for mapping monodominant
stands of Phragmites in Great Lakes coastal wetlands was tested to determine if a space platform
hyperspectral sensor could identify this invasive species (Pengra et al. 2007). The Hyperion image
covered a 7.5 km swath extending along a 30 km section of the Green Bay, Wisconsin, shoreline cen-
tered on the Pensaukee and Oconto rivers. Hyperion imagery must be ordered in advance, and image
acquisition was requested to occur between September 1 and October 15, 2004, to take advantage
of more pronounced spectral differences between Phragmites and other wetland species in mid to
late autumn (Bachmann et al. 2002). Image analysis was performed with the ERDAS IMAGINE 8.7
spectral analysis module (Leica Geosystems, Atlanta, GA) using a target detection mode of opera-
tion, which searched the input image for Phragmites target material, suspected to be present in very
low concentration. A Phragmites reference spectrum for use in target detection was extracted from
the Hyperion image using spectral datasets centered on three pure Phragmites stands identified dur-
ing field reconnaissance. The target detection operation was then run with the Spectral Correlation
Mapper (SCM) spectral analysis method, a refinement of the Spectral Angle Mapper (SAM) algo-
rithm. The SCM created a continuous value raster with values ranging from 0 to 1, where 0 repre-
sented the greatest similarity between the reference spectrum and the image spectrum and 1 the
least similarity. Both SCM and SAM are designed for hyperspectral image analysis and determine
the spectral similarity between two spectra by calculating the angle between them, treating them as
vectors in a space with dimensionality equal to the number of bands (Kruse et al. 1993).
The final classification predicted monodominant Phragmites covering 3.4% of the study area,
mostly concentrated in long linear features parallel to the Green Bay shoreline. Phragmites was dens-
est within the strip of land between the 1998 and 2004 water levels: 52% of the predicted Phragmites
occurred on this newly exposed mudflat that represented less than 15% of the study area (Figure
23.10). An error matrix using 196 field validation points showed good overall accuracy—81.4%.
IDENTIFYING INVASIVE SPECIES USING IMAGING RADAR

Imaging radar is a promising method for detection of invasive plants because of its ability to dis-
tinguish differences in structure and biomass among plant species. Bourgeau-Chavez et al. (2013)
classified Phragmites using multiseason L-band (23 cm wavelength) data from Advanced Land
Predicted Phragmites
Newly exposed areas 1998 to 2004
Study outline
0 0.3 0.6 1.2 km
FIGURE 23.10 Location of sites mapped as Phragmites in relation to the newly exposed mudflat between
the 1998 and 2004 shorelines. (Reprinted from Pengra, B.W. et al., Remote Sens. Environ., 108, 74, 2007. With
permission.)
Observing Satellite Phased Array-Type L-Band Synthetic Aperture Radar (ALOS PALSAR),
a satellite imaging radar sensor that is sensitive to differences in plant biomass and inundation
patterns (see Chapter 15 for review of remote sensing Great Lakes coastal wetlands). The use of
spring, summer, and fall datasets improved discrimination of Phragmites by taking advantage of
seasonal changes in vegetation and inundation. They then applied this technique to map “potential
Phragmites” at minimum mapping unit of 0.2 ha along the entire U.S. Great Lakes coast (Bourgeau-
Chavez et al. 2013). The designation as “potential” recognizes that there may be some confusion
with other types (e.g., high biomass Typha spp.) or omission due to ongoing control efforts and rapid
spread of Phragmites. Overall basin-wide map accuracy was 87%, with 86% producer’s accuracy
and 43% user’s accuracy for invasive Phragmites.
MODELING INVASIVE PLANT ENVIRONMENTS TO PREDICT

POTENTIAL LOCATION
Although remote sensing is typically used to detect the actual objects of interest, an alternative
approach is to use remote sensing to detect the environmental conditions that are conducive to
invasive plant species and predict species location by modeling those habitat suitability variables
(Johnston 1993). For example, researchers in Kansas used a time series of six 16-day compos-
ite NDVI images from Moderate Resolution Imaging Spectroradiometer (MODIS) (250 m pixels)
along with known locations of L. salicaria to develop a model using the genetic algorithm for rule
set prediction (GARP) (Anderson et al. 2006). GARP used these data to produce a model of the
plant’s niche requirements, which was validated using an independent dataset and then projected
into environmental space to generate a map of the potential distribution of L. salicaria in the study
region. MODIS imagery is far too coarse to detect individual wetland plant species; however, the
approach may be useful for assessing the regional colonization potential of new detected invasive
species before site-specific studies can be undertaken (Anderson et al. 2006).
A habitat suitability modeling approach was also used by Andrew and Ustin (2009), who mapped
the potential distribution of the invasive species L. latifolium in a California brackish marsh using
both predictor and response variables derived from remote sensing. They extracted presence/absence
data of the plant’s current distribution from hyperspectral image data (Hestir et al. 2008), used a
high-resolution LiDAR digital elevation model to derive predictor environmental variables (distance
to channel, distance to upland, elevation, slope, aspect, and convexity), and applied aggregate deci-
sion tree models to predict the potential distribution of the species (Andrew and Ustin 2009).
Walker (2007) developed and tested a GIS-based model to define areas at risk of Phragmites
colonization in the North Inlet–Winyah Bay estuary of South Carolina. This integrated approach
combined in situ sampling, remote sensing, geographic information processing, predictive mod-
eling, and statistical analyses. As a result of this work, the author recommended an integrated
approach utilizing a suite of dynamic, geospatial techniques in concert with more traditional prac-
tices for managing Phragmites.
CONCLUSION
Most wetland vegetation mapping classifies plants by life-form (e.g., the National Wetlands
Inventory) or community, but the mapping of invasive plants requires identification of individual
species, which has long been a challenge in remote sensing. The graminoid structure of many
invasive species makes them difficult to distinguish from other graminoids. Fortunately for remote
sensing (but not for the invaded wetland community), invasive species typically form monotypic
stands that aid their detection.
Knowledge of the phenological variation of invasive plant species is essential to separating them
using any remote sensing technique. In some cases, plant species may be spectrally distinctive dur-
ing a specific window of time, as was the case for P. arundinacea. Distinguishing other invasive
plant species may require two or more dates of imagery to take advantage of phenological differ-
ences. Remote sensing analysts should be aware of the timing of plant development and senescence
relative to the timing of the images they interpret.
This chapter gave a range of examples illustrating how remote sensing and GIS can be used to
map invasive plants in wetlands. Although some are experimental, others are not. The examples of
Landsat use for mapping reed canary grass in Wisconsin and PALSAR use for mapping Phragmites
in the Great Lakes demonstrate the operational capability of remote sensing for monitoring these
invasive species. The WIDNR found classification of Landsat satellite imagery to be “the most
cost-effective method to broadly capture the impact reed canary grass has on native wetland plant
communities.” The Great Lakes Phragmites maps are being used to identify major environmental
drivers of this invader’s distribution, to assess areas vulnerable to new invasion, and to provide
information to regional stakeholders through a decision support tool. Remote sensing cannot halt
the environmental degradation caused by invasive plant species, but it can play an important part in
efforts to reduce their impacts and spread.
REFERENCES
and mapping of wetland vegetation: A review. Wetlands Ecology and Management 18:281–296.
Anderson, R.P., A. Townsend Peterson, S.L. Egbert, and C. Lauver. 2006. Vegetation-index models pre-
dict areas vulnerable to purple loosestrife (Lythrum salicaria) invasion in Kansas. The Southwestern
Naturalist 51:471–480.
Andrew, M.E. and S.L. Ustin. 2009. Habitat suitability modelling of an invasive plant with advanced remote
sensing data. Diversity and Distributions 15:627–640.
APIRS. 2013. Aquatic Plant Information Retrieval System. Center for Aquatic and Invasive Plants, University
of Florida, Gainesville, FL. http://plants.ifas.ufl.edu/ (Accessed March 2013).
Artigas, F.J. and J.S. Yang. 2005. Hyperspectral remote sensing of marsh species and plant vigour gradient in
the New Jersey Meadowlands. International Journal of Remote Sensing 26:5209–5220.
Arzandeh, S. and J. Wang. 2003. Monitoring the change of Phragmites distribution using satellite data.
Canadian Journal of Remote Sensing 29:24–35.
Bachmann, C.M., T.F. Donato, G.M. Lamela, W.J. Rhea, M.H. Bettenhausen, R.A. Fusina, K.R. Du Bois,
J.H. Porter, and B.R. Truitt. 2002. Automatic classification of land cover on Smith Island, VA, using
HyMAP imagery. IEEE Transactions on Geoscience and Remote Sensing 40:2313–2330.
Becker, B.L., D.P. Lusch, and J. Qi. 2007. A classification-based assessment of the optimal spectral and spatial
resolutions for Great Lakes coastal wetland imagery. Remote Sensing of the Environment 108:111–120.
Benz, U.C., P. Hofmann, G. Willhauck, I. Lingenfelder, and M. Heynen. 2004. Multi-resolution, object-oriented
fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry and
Bernthal, T.W. and K.G. Willis. 2004. Using Landsat 7 imagery to map invasive reed canary grass (Phalaris
arundinacea): A landscape level wetland monitoring methodology. Wisconsin Department of Natural
Resources, Madison, WI.
Bourgeau-Chavez, L.L., K.P. Kowalski, M.L. Carlson Mazur, K.A. Scarbrough, R.B. Powell, C.N. Brooks,
B. Huberty et al. 2013. Mapping invasive Phragmites australis in the coastal Great Lakes with ALOS
PALSAR satellite imagery for decision support. Journal of Great Lakes Research 39(Suppl. 1):65–77.
Boylen, C., L. Eichler, J. Bartkowski, and S. Shaver. 2006. Use of geographic information systems to moni-
tor and predict non-native aquatic plant dispersal through north-eastern North America. Hydrobiologia
570:243–248.
Cheng, X., Y. Luo, J. Chen, G. Lin, J. Chen, and B. Li. 2006. Short-term C4 plant Spartina alterniflora inva-
sions change the soil carbon in C3 plant-dominated tidal wetlands on a growing estuarine island. Soil
Biology and Biochemistry 38:3380–3386.
Civille, J.C., K. Sayce, S.D. Smith, and D.R. Strong. 2005. Reconstructing a century of Spartina alterniflora
invasion with historical records and contemporary remote sensing. Ecoscience 12:330–338.
Habitats of the United States. U.S. Fish and Wildlife Service, Washington, DC.
de Lange, R., M. van Til, and S. Dury. 2004. The use of hyperspectral data in coastal zone vegetation monitor-
ing. EARSeL eProceedings 3:143–153.
Delisle, F., C. Lavoie, M. Jean, and D. Lachance. 2003. Reconstructing the spread of invasive plants: Taking
into account biases associated with herbarium specimens. Journal of Biogeography 30:1033–1042.
Earth Point. 2014. Tools for Google. http://earthpoint.us/ (Accessed November 2014).
Elith, J. and J. Leathwick. 2007. Predicting species distributions from museum and herbarium records using
multiresponse models fitted with multivariate adaptive regression splines. Diversity and Distributions
13:265–275.
Everitt, J., C. Yang, and C. Deloach. 2005. Remote sensing of giant reed with QuickBird satellite imagery.
Journal of Aquatic Plant Management 43:81–85.
Frazier, B. and B. Moore. 1993. Some tests of film types for remote sensing of purple loosestrife, Lythrum
salicaria, at low densities. Wetlands 13:145–152.
Frieswyk, C.B. and J.B. Zedler. 2007. Vegetation change in Great Lakes coastal wetlands: Deviation from the
historical cycle. Journal of Great Lakes Research 33:366–380.
Funk, V.A., M.F. Zermoglio, and N. Nasir. 1999. Testing the use of specimen collection data and GIS in biodiver-
sity exploration and conservation decision making in Guyana. Biodiversity and Conservation 8:727–751.
Galatowitsch, S., N. Anderson, and P. Ascher. 1999. Invasiveness in wetland plants in temperate North America.
Ghioca-Robrecht, D.M., C.A. Johnston, and M.G. Tulbure. 2008. Assessing the use of multiseason QuickBird
imagery for mapping invasive species in a Lake Erie coastal marsh. Wetlands 28:1028–1039.
GISD. 2013. Global invasive species database. Invasive Species Specialist Group, International Union for
Conservation of Nature, New Zealand. http://www.issg.org/database/welcome/ (Accessed March 2013).
Google. 2014. Google Earth. http://www.google.com/earth/ (Accessed November 2014).
Hatch, B.K. and T.W. Bernthal. 2008. Mapping Wisconsin wetlands dominated by reed canary grass,
Phalaris arundinacea L.: A landscape level assessment. Wisconsin Department of Natural Resources,
Madison, WI.
Healy, M.T. and J.B. Zedler. 2010. Set-backs in replacing phalaris arundinacea monotypes with sedge meadow
vegetation. Restoration Ecology 18:155–164.
Hestir, E.L., S. Khanna, M.E. Andrew, M.J. Santos, J.H. Viers, J.A. Greenberg, S.S. Rajapakse, and S.L. Ustin.
2008. Identification of invasive vegetation using hyperspectral remote sensing in the California Delta
ecosystem. Remote Sensing of the Environment 112:4034–4047.
Huang, C.-y. and G. Asner. 2009. Applications of remote sensing to alien invasive plant studies. Sensors
9:4869–4889.
Jakubauskas, M.E., D.L. Peterson, S.W. Campbell, F. deNoyelles Jr., S.D. Campbell, and D. Penny. 2002.
Mapping and monitoring invasive aquatic plant obstructions in navigable waterways using satellite mul-
tispectral imagery. Integrating Remote Sensing at the Global, Regional and Local Scale. Pecora 15/Land
Satellite Information IV Conference. Denver, CO.
Water Conservation Area 2A using a time series of normalized remotely sensed data. Photogrammetric
Jodoin, Y., C. Lavoie, P. Villeneuve, M. Theriault, J. Beaulieu, and F. Belzile. 2008. Highways as corridors
and habitats for the invasive common reed Phragmites australis in Quebec, Canada. Journal of Applied
Ecology 45:459–466.
Johnson, L.E., A. Ricciardi, and J.T. Carlton. 2001. Overland dispersal of aquatic invasive species: A risk
assessment of transient recreational boating. Ecological Applications 11:1789–1799.
Johnston, C.A. 1993. Introduction to quantitative methods and modeling in community, population, and land-
scape ecology. In: M.R. Goodchild, B.O. Parks, and L.T. Steyaert (eds.), Environmental Modeling with
GIS. Oxford University Press, New York, pp. 276–283.
Johnston, C.A. 1998. Geographic Information Systems in Ecology. Blackwell Science, Oxford, U.K.
Johnston, C.A., T. Brown, T. Hollenhorst, P. Wolter, N. Danz, and G. Niemi. 2009. GIS in support of ecologi-
cal indicator development. In: M. Madden (ed.), Manual of Geographic Information Systems. American
Society for Photogrammetry and Remote Sensing, Bethesda, MD, pp. 1095–1113.
Johnston, C.A. and K.I. Miller. 2011. Phragmites australis in South Dakota: Historical distribution and envi-
ronment. Prairie Naturalist 43:38–44.
Kirk, H., C. Connolly, and J.R. Freeland. 2011. Molecular genetic data reveal hybridization between Typha
angustifolia and Typha latifolia across a broad spatial scale in eastern North America. Aquatic Botany
95:189–193.
Kruse, F.A., A.B. Lefkoff, J.W. Boardman, K.B. Heidebrecht, A.T. Shapiro, P.J. Barloon, and A.F.H. Goetz.
1993. The spectral image processing system (SIPS)—Interactive visualization and analysis of imaging
spectrometer data. Remote Sensing of the Environment 44:145–163.
Laba, M., R. Downs, S. Smith, S. Welsh, C. Neider, S. White, M. Richmond, W. Philpot, and P. Baveye.
2008. Mapping invasive wetland plants in the Hudson River National Estuarine Research Reserve using
QuickBird satellite imagery. Remote Sensing of the Environment 112:286–300.
Lambert, A.M. and R.A. Casagrande. 2006. Distribution of native and exotic Phragmites australis in Rhode
Island. Northeastern Naturalist 13:551–560.
Lavoie, C., C. Dufresne, and F. Delisle. 2009. The spread of reed canary grass (Phalaris arundinacea) in
Québec: A spatio-temporal perspective. Ecoscience 12:366–375.
Lopez, R.D., C.M. Edmonds, A.C. Neale, T. Slonecker, K.B. Jones, D.T. Heggem, J.G. Lyon, E. Jaworski,
D. Garofalo, and D. Williams. 2004. Accuracy assessments of airborne hyperspectral data for mapping
opportunistic plant species in freshwater coastal wetlands. In: R.S. Lunetta and J.G. Lyon (eds.), Remote
Sensing and GIS Accuracy Assessment. CRC Press, New York, pp. 253–267.
Madden, M. 2004. Remote sensing and geographic information system operations for vegetation mapping of
invasive exotics. Weed Technology 18:1457–1463.
Maheu-Giroux, M. and S. D. Blois. 2005. Mapping the invasive species Phragmites australis in linear wetland
corridors. Aquatic Botany 83:310–320.
NISIC. 2013. National Invasive Species Information Center. U.S. Department of Agriculture, Washington, DC.
http://www.invasivespeciesinfo.gov/index.shtml (Accessed March 2013).
NRCS. 2013. Introduced, invasive, and noxious plants. Natural Resource Conservation Service, U.S.
Department of Agriculture, Washington, DC. http://plants.usda.gov/java/noxComposite (Accessed
March 2013).
Olson, A., J. Paul, and J.R. Freeland. 2009. Habitat preferences of cattail species and hybrids (Typha spp.) in
eastern Canada. Aquatic Botany 91:67–70.
10:381–402.
Pengra, B.W., C.A. Johnston, and T.R. Loveland. 2007. Mapping an invasive plant, Phragmites australis, in
coastal wetlands using the EO-1 Hyperion hyperspectral sensor. Remote Sensing of the Environment
108:74–81.
Proctor, C., V. Robinson, and Y. He. 2012. Multispectral detection of European frog-bit in the South Nation
River using QuickBird imagery. Canadian Journal of Remote Sensing 38:476–486.
Pyšek, P. and K. Prach. 1993. Plant invasions and the role of riparian habitats: A comparison of four species
alien to central Europe. Journal of Biogeography 20:413–420.
Rosso, P.H., S.L. Ustin, and A. Hastings. 2006. Use of lidar to study changes associated with Spartina invasion
in San Francisco Bay marshes. Remote Sensing of the Environment 100:295–306.
Rutchey, K. and L. Vilchek. 1999. Air photointerpretation and satellite imagery analysis techniques for map-
ping cattail coverage in a northern Everglades impoundment. Photogrammetric Engineering and Remote
Sensing 65:185–191.
Saltonstall, K. 2002. Cryptic invasion by a non-native genotype of the common reed, Phragmites australis, into
North America. Population Biology 99:2445–2449.
Sawaya, K.E., L.G. Olmanson, N.J. Heinert, P.L. Brezonik, and M.E. Bauer. 2003. Extending satellite remote
sensing to local scales: Land and water resource monitoring using high-resolution imagery. Remote
Snow, A.A., S.E. Travis, R. Wildová, T. Fér, P.M. Sweeney, J.E. Marburger, S. Windels, B. Kubátová,
D.E. Goldberg, and E. Mutegi. 2010. Species-specific SSR alleles for studies of hybrid cattails (Typha
latifolia × T. angustifolia; Typhaceae) in North America. American Journal of Botany 97:2061–2067.
Travis, S.E., J.E. Marburger, S. Windels, and B. Kubátová. 2010. Hybridization dynamics of invasive cattail
(Typhaceae) stands in the Western Great Lakes Region of North America: A molecular analysis. Journal
of Ecology 98:7–16.
Travis, S.E., J.E. Marburger, S.K. Windels, and B. Kubátová. 2011. Clonal structure of invasive cattail
(Typhaceae) stands in the upper midwest region of the US. Wetlands 31:221–228.
Tulbure, M.G. and C.A. Johnston. 2010. Environmental conditions promoting non-native Phragmites australis
expansion in Great Lakes coastal wetlands. Wetlands 30:577–587.
Tulbure, M.G., C.A. Johnston, and D.L. Auger. 2007. Rapid invasion of a Great Lakes coastal wetland by non-
native Phragmites australis and Typha. Journal of Great Lakes Research 33(sp3):269–279.
USGS. 2011. Earth observing 1 (EO-1) sensors—hyperion. U.S. Geological Survey. http://eo1.usgs.gov/
sensors/hyperion (Accessed March 2013).
USGS. 2014. SLC-off Products: Background. http://landsat.usgs.gov/products_slcoffbackground.php (Accessed
November 2014).
Walker, S.P. 2007. An integrated modeling approach for monitoring and predicting common reed (Phragmites
australis) colonization in a managed South Carolina estuary. Department of Environmental Health
Sciences, University of South Carolina, Columbia, SC.
Wilcox, K.L., S.A. Petrie, L.A. Maynard, and S.W. Meyer. 2003. Historical distribution and abundance of
Phragmites australis at long point, Lake Erie, Ontario. Journal of Great Lakes Research 29:664–680.
Wu, S.-H., M. Rejmánek, E. Grotkopp, and J.M. DiTomaso. 2005. Herbarium records, actual distribution, and
critical attributes of invasive plants: Genus Crotalaria in Taiwan. Taxon 54:133–138.
Zedler, J.B. and S.M. Kercher. 2004. Causes and consequences of invasive plants in wetlands: Opportunities,
opportunists, and outcomes. Critical Reviews in Plant Sciences 23:431–452.
24 Multisatellite Remote
Sensing of Global Wetland
Extent and Dynamics
Catherine Prigent and Fabrice Papa
CONTENTS
Introduction..................................................................................................................................... 511
Existing Static Global Wetland Inventories............................................................................... 512
Conventional Satellite Remote Sensing Methodologies............................................................ 513
Suggested Solution..................................................................................................................... 513
Methodology to Derive GIEMS...................................................................................................... 513
Preprocessing of Three Sources of Satellite Observations......................................................... 514
Merging of Satellite Observations and Detection of Inundation................................................ 515
Quantification of the Inundated Fraction within Each Inundated Pixel..................................... 515
Adaptation of the Methodology to Discontinuities in Satellite Observations............................ 515
Results from GIEMS....................................................................................................................... 515
Evaluation of GIEMS Results......................................................................................................... 516
Direct Comparisons with Available Estimates........................................................................... 517
Comparison with Global Static Estimates............................................................................. 517
Comparison with Regional Estimates................................................................................... 517
Indirect Comparison with Related Hydrologic Variables.......................................................... 518
GIEMS Applications....................................................................................................................... 523
Hydrological Applications......................................................................................................... 523
Evaluation of Global Hydrological Modeling....................................................................... 523
Estimation of Surface Water Volume Change....................................................................... 523
Applications to CH4 Modeling................................................................................................... 525
Estimation of CH4 Emission from Wetland Extent............................................................... 525
Evaluation of Inverse CH4 Modeling.................................................................................... 525
Conclusion and Future Perspective................................................................................................. 525
References....................................................................................................................................... 526
INTRODUCTION
There is now a widespread recognition of the need for better observations and understanding of
surface water distribution and variations globally (e.g., United Nations 2004; Marburger and Bolten
2005). However, with approximately 60% of the world’s floodplains and wetlands inundated during
some portions of the year, our current knowledge of the spatial extent and variability of the land
surface water cycle at the regional-to-global scales is still incomplete (Finlayson et al. 1999).
Definitions of wetlands and/or inundation vary according to research foci and, consequently,
no overall consensus on the subject exists (see Chapter 1). In this chapter, we will concentrate on
flooded wetlands. Inundation occurs as a result of flooding associated with estuaries, rivers and
511
lakes, rises in local water tables, snowmelt above permafrost, and irrigation. Identifying and charac-
terizing wetlands globally is further complicated by their distribution throughout tropical to boreal
environments encompassing a wide variety of vegetation covers, hydrological regimes, natural sea-
sonality, and land-use impacts (see Chapter 3).
Surface waters, including wetlands, irrigated lands, and episodically inundated areas, play a
crucial role in the global biochemical and hydrological cycles (Alsdorf et al. 2007b) with significant
influence on the climate variability. From a biogeochemical point of view, wetlands and inunda-
tion have a major impact on methane production and carbon storage/release. They are the world’s
dominant natural source of methane (CH4) and the only one dominated by climate variations.
Approximately 20%–40% of the world’s total annual methane emission to the atmosphere comes
from natural wetlands and irrigated rice fields (Houweling et al. 1999; Gedney et al. 2004); hence,
changes in wetlands and their inundation are key contributors to interannual variability in methane
surface emissions (Bousquet et al. 2006; Ringeval et al. 2010). Since preindustrial times, methane
contribution to the global warming is equal to roughly one-half that of CO2 (Shindell et al. 2004).
From a hydrological perspective, analysis of the flow, spatial distribution, and storage of freshwater
on land are key issues for understanding the global water and energy cycles. Floodplains and peri-
odically inundated wetlands are also critical for understanding hydrological processes (Decharme
et al. 2012) as they help regulate river hydrology, modulate atmospheric temperatures, and contrib-
ute to increased evaporation. Finally, surface freshwaters are also important for water resources
management (Vörösmarty et al. 2000) and biodiversity conservation (Ramsar Convention 1971;
IPCC 2007).
Existing Static Global Wetland Inventories

Existing global or regional surveys represent various components of wetland and open water dis-
tributions comprising inundated and noninundated wetlands, lakes, and rivers (Matthews and Fung
1987; Loveland et al. 2000; Cogley 2003; Lehner and Doll 2004). These static datasets reflect what
is considered to be the climatological maximum extent of active wetlands and inundation, but they
lack information on temporal and spatial dynamics at seasonal or interannual time scales. For
instance, the global lakes and wetlands dataset (GLDW) from Lehner and Doll (2004) is a com-
prehensive dataset of global surface water area, including small and large lakes, reservoirs, smaller
water bodies, and rivers and a good representation of the maximum global wetland extent with a
~1 km spatial resolution (Figure 24.1). These datasets are complemented by global inventories of
irrigated crops (Matthews et al. 1991; Portman et al. 2010).
The global dataset of monthly irrigated and rain-fed crop areas (MIRCA) around the year
2000 (Portman et al. 2010) is such a dataset providing monthly areas of 26 crops and their related
calendars, with a 9 km × 9 km spatial resolution at the equator. Some national inventories exist but
Lake
Reservoir
River
Freshwater marsh, floodplain
Swamp forest, flooded forest
Coastal wetland
Pan, brackish/saline wetland
Bog, fen, mire
Intermittent wetland/lake
50%–100% Wetland
25%–50% Wetland
Wetland complex
(0%–25% wetland)
FIGURE 24.1 GLDW distribution from Lehner and Doll (2004).

Multisatellite Remote Sensing of Global Wetland Extent and Dynamics 513
are usually carried out for resource management or conservation purposes and consequently iden-
tify only those wetlands of particular interest for a specific application. For instance, the Convention
on Wetlands signed in Ramsar, Iran (Ramsar Convention 1971), is a framework for national and
international actions for the preservation and wise use of wetlands with an emphasis on wetlands
important for water birds.
Conventional Satellite Remote Sensing Methodologies

Surface water extents and their variations have been measured with a variety of remote sensing
techniques employing visible, infrared, or microwave observations (Alsdorf et al. 2007b; Prigent
et al. 2007). The following discussion lists techniques that offer varying degrees of success.
Estimations of surface water extent using visible or infrared measurements provide relatively high
spatial resolution, but they have limitations for detecting surface water beneath clouds or dense
vegetation. The potential for using MODerate resolution Imaging Spectroradiometer (MODIS) to
monitor temporal changes in flooding is however demonstrated over specific regions such as the
Mekong River (Sakamoto et al. 2007) or arid regions such as the Inner Niger Delta (Berge-Nguyen
et al. 2008). The Dartmouth Flood Observatory produces digital maps of Earth’s changing surface
water, using visible and infrared observations from MODIS and VIIRS (Brakenridge et al. 2005;
http://floodobservatory.colorado.edu/). Studies in the Amazon basin (Hess et al. 2003; Alsdorf et al.
2007a) or in the Congo prove the ability of synthetic aperture radars (SARs) to accurately delineate
surface water in tropical forests, with high spatial sampling intervals (∼100 m). However, the large
data volumes have limited these studies to a few samples of a small number of basins, preventing
systematic, long-term assessments of inundation dynamics. Other active microwave instruments,
such as the dual-frequency Topex–Poseidon radar altimeter, are also valuable for detecting inun-
dation, but because of their narrow-surface track, their estimation of water body extent is limited
to polar regions where the satellite’s polar orbit yields better spatial coverage (Papa et al. 2006b).
Passive microwave observations have long been shown to detect surface water extent (Gidding
and Choudhury 1989). Most studies have been dedicated to specific areas like the Amazon basin
(Sippel et al. 1998) or the Arctic regions (Mialon et al. 2005). Until recently, the available satel-
lite microwave observations from 6 to 190 GHz could be contaminated by the atmosphere (water
vapor), clouds, and rain or altered by the presence of vegetation in mixed open water/vegetation
scenes. These factors could distort time series estimates of surface water variations.
Suggested Solution
Recognizing the limitations of the existing remote sensing techniques to detect open water regard-
less of the environment led to the development of a multisatellite methodology to retrieve surface
water extent and dynamics at the global scale. Combining observations from different sensors made
it possible to capitalize on their complementary strengths, to extract maximum information about
inundation characteristics, and to minimize problems related to analysis of measurements by one
instrument only. In this chapter, this methodology—Global Inundation Extent from Multi-Satellite
(GIEMS)—is described, and its evaluation and applications are presented.
METHODOLOGY TO DERIVE GIEMS

The GIEMS methodology to detect and quantify the surface water extent at global scale has been
described in Prigent et al. (2001b, 2007, 2012) and Papa et al. (2010b). The selected satellite data
have different sensitivities to surface properties (e.g., vegetation, topography, and soil properties),
making it possible to separate the contributions of the various factors to the observations. The fol-
lowing satellite observations are used: (1) Advanced Very High Resolution Radiometer (AVHRR)
SSM/I ERS AVHRR

observations observations observations
Preprocessing
Emissivity Angular NDVI
calculation interpolation calculation
(PathFinder)
Microwave Backscattering
emissivities NDVI
coefficients
Classification
Fractional inundation
calculation
Monthly-mean
inundation extent
FIGURE 24.2 Schematic description of the methodology to obtain GIEMS. (From Prigent, C. et al.,
J. Geophys. Res., 112, D12107, 2007.)
visible (0.58–0.68 μm) and near-infrared (0.73–1.1 μm) reflectances and the derived normalized dif-
ference vegetation index (NDVI); (2) passive microwave special sensor microwave/imager (SSM/I)
measurements between 19 and 85 GHz (1.58–0.35 cm in wavelength); and (3) active microwave
backscattering coefficients at 5.25 GHz (wavelength = 5.71 cm) from the scatterometer on board the
European remote sensing (ERS) satellite. Detection of inundation relies primarily on the passive
microwave land surface signal in which inundated regions are characterized by lower microwave
emissivities and higher emissivity polarization differences than their surroundings, even under a
dense vegetation canopy (Prigent et al. 2001b). The vegetation contribution is estimated with infor-
mation from the coincident ERS active microwave and NDVI observations. Figure 24.2 presents a
schematic description of the GIEMS methodology.
Preprocessing of Three Sources of Satellite Observations

Detection and extent estimation of the surface water involve several steps, the first of which is satel-
lite data preprocessing. This is especially important for the passive microwave observations. The
microwave emissivities of land surfaces are estimated from SSM/I passive microwave observations
by removing the contribution from the atmosphere (including clouds and rain) and the modula-
tion by the surface temperature, using ancillary data from visible and infrared satellite observa-
tions from the International Satellite Cloud Climatology Project (ISCCP) (Rossow and Schiffer
1999) and the National Centers for Environmental Prediction (NCEP) reanalyses (Kalnay et al.
1996) (see Prigent et al. 2006 for details). In contrast to the direct use of the microwave bright-
ness temperatures for surface characterization, these calculated emissivities are related to the sur-
face properties themselves without confusion with atmospheric or surface temperature variations.
The active microwave observations are provided by the scatterometer at 5.3 GHz on board the
ERS satellite. Multiple measurements of a single location with different viewing angles are lin-
early interpolated at 40° incidence angle (Prigent et al. 2001a). The AVHRR NDVI is derived
from the 8 km monthly composite from NASA-NOAA Earth Observing System Pathfinder project.
All satellite observations are projected onto an equal area grid of 0.25° × 0.25° resolution at the
equator (773 km2), and monthly mean values are calculated.
Merging of Satellite Observations and Detection of Inundation

An unsupervised clustering algorithm using Kohonen topological maps (Kohonen 1984) is used
to merge and classify the three sources of satellite observations. The distinguishing feature of this
algorithm, also called self-organizing topological maps, is the neighborhood requirement imposed
on the prototypes. When the algorithm converges, prototypes corresponding to nearby points on the
map grid have nearby locations in the data space. Sensitivity analyses determined the optimal subset
of observations for input to the clustering scheme. Thirty classes are identified for all snow- and
ice-free continental areas. Examination of the maps resulting from the multisatellite clustering pro-
cedure carried out for each month enables the selection of six classes characterized by the presence
of surface water (Prigent et al. 2001a).
Quantification of the Inundated Fraction within Each Inundated Pixel

Because the response to inundation of microwave emissivity is modulated by vegetation, a method
that accounts for the contribution of vegetation to the emissivity is needed to quantify the flooded
area within a pixel classified as inundated. The emissivity polarization difference increases with
inundation fraction but declines with increasing vegetation density. Backscattering is not very
sensitive to the presence of inundation and increases with vegetation density. A linear mixture
model is created from the statistical relationship between the emissivity polarization difference
at 37 GHz and backscattering (Prigent et al. 2001b). There is no tuning to specific wetland type,
so this mixture model is used to calculate flooded fractions for inundated pixels all over the
globe and for all seasons. Since the microwave measurements are sensitive to snow cover, snow
masks are used to edit the results and avoid any confusion with snow-covered pixels. The weekly
Northern Hemisphere and Southern Hemisphere snow mask from the National Snow and Ice Data
Center (NSIDC, University of Colorado) is used and averaged on a monthly basis (Armstrong and
Brodzik 2005).
Adaptation of the Methodology to Discontinuities in Satellite Observations

The ERS scatterometer encountered serious technical problems after 2000, and the consistency of
an NDVI dataset over a long time period was also in doubt. This required adaptation of the process-
ing scheme. Several solutions were explored and carefully tested, including the use of alternative
satellite observations (Papa et al. 2010b). A satisfactory solution has been found, using a monthly
mean climatology of the ERS and AVHRR observations. Instead of using the ERS and AVHRR
data for a given month in a given year, the monthly mean climatology calculated over the 1993–
2000 is adopted. The mean difference between the two versions of the dataset over the 1993–2000
period is 0% with a standard deviation less than 10%. The observed trends for 1993–2000 are not
changed. A correction was also applied to the extended SSM/I 1993–2004 emissivity dataset after
a problem in the time series end of 2001 was detected in the ancillary data used in the emissivity
calculation (Papa et al. 2010b).
RESULTS FROM GIEMS

Monthly mean multisatellite inundation estimates have been produced at a global scale for
1993–2007. Note that lacking additional external information, the technique captures, but does
not discriminate among inundated wetlands, rivers, small lakes, irrigated agriculture, or even
ocean-contaminated coastal pixels. For each given 773 km 2 pixel on the globe, all satellite
60
400
350
30
300
Latitude
250
km2
200 0
150
100
–30
50
0
–60
–180 –150 –120 –90 –60 –30 0 30 60 90 120 150 180 0 50 100 150
(a) Longitude (b) Water extent (103 km2)
FIGURE 24.3 Global map of the annual mean surface water extent derived from the multisatellite method
with a spatial resolution of 773 km2 (a), along with the latitudinal distribution of the water extend per 1° (b).
(From Prigent, C. et al., Geophys. Res. Lett., 39, L08403, 2012.)
observations falling within this box are considered. This does not necessarily restrict the full
satellite field of view to fall within the pixel, so a significant part of the radiation received by the
satellite might actually come from neighboring pixels. In the results, coastal pixels have been
partly filtered as well as large inland water bodies using the International Geosphere-Biosphere
Program (IGBP) 1 min land ecosystem classification map (Loveland et al. 2000). Figure 24.3
represents the global map of the annual mean surface water extent derived from GIEMS, with a
spatial resolution of 773 km 2.
Major inundated wetlands are well delineated in all latitudes and environments. The Pantanal
and riverine wetlands along the Amazon are clearly captured in South America. The Inner
Niger Delta (Mali) and wetlands associated with Lake Chad in semiarid Africa are realistically
represented. The inundated region around the Ob River in northern Russia is clearly observed.
The extensive features in Canada represent short-duration complexes of pools and inundated
tundra. Twenty percent of the total mean inundation extent over the year is located above 50°N.
Asian rice paddies have a strong signature, especially in India, Bangladesh, and China, where
rice dominates agricultural landscapes. Checks indicate that the inundation duration is con-
sistent with the number and duration of rice crops. South Asia (the region between 0° 40°N
and 60°E 140°E) amounts to 30% of the global mean inundation extent over the year. The
latitudinal distribution (Figure 24.3b) emphasizes the contributions of the tropical and boreal
regions to the total surface water extent. The month-to-month variations also show a realistic
seasonal cycle and modest interannual variability (Figure 24.4), with a mean annual maximum
extent over the record of 5.660 × 10 6 km 2 (StD = 286,000 km 2) for the globe, 2.598 × 10 6 km 2
(StD = 199,600 km 2) for the tropics (30°S–30°N), and 660,000 km 2 (StD = 750,000 km 2) for the
boreal region (>50°N).
EVALUATION OF GIEMS RESULTS

Because GIEMS is the only global surface water dataset available with dynamic information, its
exhaustive evaluation is not straightforward. As a consequence, its validation is limited to com-
parison with global static inventories and to checks of the consistency of its variability with related
hydrological variables such as river height (from in situ or satellite altimeter data), river discharge,
or satellite estimates of water volume change.
Global
6
Tropics
Extent (106 km2)
Boreal
0
1993 1995 1997 1999 2001 2003 2005 2007
FIGURE 24.4 Monthly mean surface water extent for 1993–2007 in black, for (top) the globe, (middle) the
tropics (30°S–30°N), and (bottom) the boreal region (50°N–90°N); red and green lines for the 15-year mean
maximum and minimum, respectively.
Direct Comparisons with Available Estimates

Comparison with Global Static Estimates
The results have been compared with different independent static datasets, including the wetlands
distribution of Matthews and Fung (1987) or Lehner and Doll (2004), the rice paddy distribution
of Matthews et al. (1991) and Portmann et al. (2010), and the inland permanent water from IGBP
(Loveland et al. 2000). Given that the multisatellite method does not discriminate among inun-
dated wetlands, coastal wetlands, rivers, small lakes, irrigated agriculture, and floodplains, each
of these datasets represented a subset of the satellite-derived inundation extent. Figure 24.5 shows
the comparison between the latitudinal distribution of surface water derived from the multisatellite
technique and the estimates from the global lakes and wetlands database (GLWD).
Adam et al. (2010) performed a detailed comparison of GIEMS with the GLDW (Lehner and
Doll 2004, at ~1 km resolution) and the MIRCA database (Portmann et al. 2010, at 9 km resolution).
For most areas the GIEMS mean annual maximum agrees well with the static open water extent. In
16% of the area, GIEMS estimates larger water extents than GLDW/MIRCA. This could be due to
confusion between very wet soils and inundated areas with GIEMS.
Comparison with Regional Estimates

High spatial resolution surveys of wetland extent over large regions are very scarce. SAR imag-
ery provides estimates with good spatial resolution but suffers from a lack of temporal coverage.
Hess et al. (2003) studied the flooding in the Amazon basin for both low-water and high-water
60
30
Latitude
0
–30
–60
0 300 600 900
Water extent (103 km2)
FIGURE 24.5 The latitudinal distributions of global surface water for three estimates: the annual maxi-
mum surface water extent averaged over 1993–2007 (as shown on the map, solid black line), the maximum
surface water extent reached for each pixel over 1993–2007 (dashed black line), and the surface water extent
from the global lakes and wetlands database (GLWD, level 3, red line) of Lehner and Doll (2004). Surface
water extent values are aggregated in steps of 3°. (Adapted from Papa, F. et al., J. Geophys. Res., 115, D12111,
2010b.)
conditions with a 100 m resolution using L-band SAR observations from the Japanese Earth
Resources Satellite-1 (JERS-1). Figure 24.6 compares our estimates with the Hess et al. (2003)
results for the same time periods. The SAR estimates are projected onto the 0.25° × 0.25° equal
area grid (773 km 2) for comparison purposes. The spatial structures of the inundation are very
similar. The total SAR-derived flooded area for low-water to high-water stage is 118,000–
243,000 km 2 compared to our results of 105,000–171,000 km 2. The difference in total flooded
area is larger during high-water stage. With its much better spatial resolution, the SAR can esti-
mate more accurately small areas that are flooded under generally dry conditions or small dry
areas in generally flooded conditions, whereas our lower resolution observations may miss such
small fractional coverages. The accuracy of our wetland extent estimate is estimated of 10%
(<80 km 2) (Prigent et al. 2007).
Indirect Comparison with Related Hydrologic Variables

Radar altimetry entails vertical range measurements between the satellite and Earth surface, and
the water levels are given by the difference between the satellite orbit information and the range
(or altimetric height) (Fu and Cazenave 2001). Radar altimeter water level series have been shown
to be precise enough for continental water studies (Birkett 1998; Birkett et al. 2002; Cretaux
et al. 2006) and evaluated intensively against in situ gauge stations providing water level heights
(Calmant et al. 2008).
Figure 24.7 displays the 9-year time series (1993–2001) comparing GIEMS with river/inun-
dation heights derived from the Topex–Poseidon altimeter at three locations representative of a
variety of rivers and other environments. The surface water extent estimates are averaged over
the region around the altimeter measurement locations. For the three locations, there is a strong
agreement between the two datasets for the seasonal cycle and the interannual variability over
the 10 years, even in environments of complex variability. Over southern Parana, for instance
Low-water stage (September–October 1995) High-water stage (May–June 1996)

0 0
–2 –2
–4 –4
SAR SAR
–6 estimate –6 estimate
–8 –8
–72 –70 –68 –66 –64 –62 –60 –58 –56 –54 –72 –70 –68 –66 –64 –62 –60 –58 –56 –54
0 0
90
–2 80
–2
70
60
Latitude
–4 –4 50
Multisatellite Multisatellite 40
30
–6 estimate –6 estimate 20
10
–8 –8
–72 –70 –68 –66 –64 –62 –60 –58 –56 –54 –72 –70 –68 –66 –64 –62 –60 –58 –56 –54
Longitude
FIGURE 24.6 Comparison of multisatellite inundation (GIEMS) estimates (bottom) with the SAR estimates
from Hess et al. (2003) (top) for the same periods of time, in terms of fractional inundation over the equal
area pixels of 773 km2, showing (left) the low-water stage during September–October 1995 and (right) the
high-water stage during May–June 1996. (From Prigent, C. et al., Bull. Am. Meteorol. Soc., 87, 1573, 2006.)
(Figure 24.7a and b), very good agreement is observed including during the exceptional 1997–
1998 El Niño event, with a correlation R of 0.82 between the satellite-derived water extent and
the altimetric river height over 1993–2001 and a correlation of 0.80 between their deseasonal-
ized anomalies. Over the Ganges Delta, both datasets exhibit very similar variations (R = 0.78)
with both extent and water height showing a slight increase (Figure 24.7c and d) with a simi-
lar proportional rate of increase. For the Paraguay River (Figure 24.7e and f), the high peaks
observed in the anomaly in 1995 and 1997 are well reproduced, although the magnitude observed
in the water extent in 1997 is not as strong as in the altimeter signal. The correlation between the
satellite-derived water extent and the altimetric river height over 1993–2001 is 0.81 with a lag of
1 month (the extent preceding the height) and a correlation of 0.80 with 1-month lag between their
deseasonalized anomalies.
In situ observations of stage and discharge overlapping in time with the GIEMS estimates are
only available for a few river systems worldwide. Figure 24.8 compares the surface water extent,
in situ river discharge, and height and altimeter-derived river height over the Amazon basin,
along with the basin-averaged monthly mean precipitation estimated from Global Precipitation
Climatology Centre (GPCC) (Rudolf and Schneider 2005). The Amazon River discharge is pro-
vided by the Brazilian Water National Agency and the HyBAM project (http://www.ore-hybam.
org). Its discharge comprises 20% of world’s total continental runoff (Richey et al. 1989). The
Amazon basin shows strong seasonal and interannual variations in precipitation, surface water
extent, and discharge. Time series of the Amazon River discharge is closely linked to the total
amount of satellite-derived surface water extent in the whole basin (Figure 24.8a), with a maxi-
mum lagged correlation of 0.90 (144 months are used to calculate the linear correlation coef-
ficient, giving a p-value < 0.01, with R > 0.21) with the extent preceding by 1 month the in
situ discharge obtained from the Brazilian Water National Agency (Figure 24.8a). The lagged
correlation between the deseasonalized anomalies of these two variables is 0.59 (p < 0.01). The
temporal pattern indicates alternatively wet and dry events associated with the El Niño/La Niña
4 × 104 16 2 × 104 3
Surface water extent anomaly (km2)

South Parana River
(29°S–31°S; 59°W–61°W) 2
Surface water extent (km2)
3 × 104 1 × 104
Height anomaly (m)

14 1
Height (m)
2 × 104 0 0
12 –1
1 × 104 –1 × 104
–2
4 –3
0 10 –2 × 10
1993 1995 1997 1999 2001 1994 1996 1998 2000 2002
(a) (b)
2.0 × 104 6 3000 1.0

Ganges Delta
(22°N–23°N;88°E–89°E) 2000
1.5× 104 0.5
Height anomaly (m)

4 1000
Height (m)
1.0 × 104 0 0.0
2 –1000
5.0 × 103 –0.5
–2000
0 0 –3000 –1.0
1993 1995 1997 1999 2001 1994 1996 1998 2000 2002
(c) (d)
4 6
5 × 104 86 3 × 10
Paraguay River
(22°N–23°N;88°E–89°E)
4 × 104
84 2 × 104 4
Height anomaly (m)

82 1 × 104 2
3 × 104
Height (m)
80 0 0
2 × 104
78 – 1 × 104 –2
1 × 104 76 –2 × 10 4 –4
0 74 –3 × 10 4 –6
1993 1995 1997 1999 2001 1994 1996 1998 2000 2002
(e) (f)
FIGURE 24.7 Correspondence between the monthly mean surface water extent and the altimeter-derived
water height for the southern Parana River (a and b), the Ganges Delta (c and d), and the Paraguay River
(e and f). On the left, the monthly satellite-derived surface water extent (solid line, left axis) and water level
derived from the Topex–Poseidon altimeter (dashed line, right axis) for 1993–2001. On the right, the deseason-
alized anomalies (obtained by subtracting the 12-year mean monthly value from individual months). A mov-
ing average of ±1 month is applied to all curves. (From Papa, F. et al., J. Geophys. Res., 115, D12111, 2010b.)
phenomena (Prigent et al. 2007). For the entire Amazon basin, the lagged correlation between
the surface water extent and the basin-averaged precipitation reaches 0.86, with rain preceding
the water extent by 2 months. Part of the inundation change is due mainly to precipitation events
in upstream locations that do not flood, but contribute to the water discharged from the entire
basin (Papa et al. 2006a, 2007; Prigent et al. 2007). For other locations such as the Obidos region
(800 km upstream from the river mouth in the State of Pará), variations in surface water extent
are similarly consistent with river height measured in situ and by the Topex–Poseidon altimeter
(Figure 24.8b).
The times series of GIEMS and in situ river discharge are compared for three other large river
basins representing environments from tropical to boreal river basins (Figure 24.9). The Mississippi
River and MacKenzie River discharges come from in situ observations provided by the Global
2.0
1.5 Amazon basin
Normalized anomalies 1.0
0.5
0.0
–0.5
–1.0
–1.5
1993 1995 1997 1999 2001 2003 2005
(a)
1.5
Obidos region
(0.50°S–4.50°S; 53.51°W–5 7.51°W)
1.0
Normalized anomalies
0.5
0.0
–0.5
–1.0
–1.5
1993 1995 1997 1999 2001 2003 2005
(b)
FIGURE 24.8 Correspondence among the satellite-derived monthly mean surface water extent, in situ river
discharge and height, and rain gauge measurements in the Amazon. Normalized deseasonalized anomalies
(obtained by subtracting the 12‐year mean monthly value from individual months and by dividing the stan-
dard deviations of the raw time series). (a) For the entire Amazon basin, the black line is the satellite-derived
surface water extent, the red line is the in situ river discharge from the Brazilian Water National Agency, and
the green line is the basin mean precipitation from the GPCC. (b) For Obidos, Pará, Brazil, located 800 km
upstream from the mouth of the Amazon, the black line is the satellite‐derived surface water extent averaged
for a 4° × 4° region centered at Obidos, the red line is the in situ river height measured at Obidos, and the green
line is the altimeter-derived river water height near Obidos (2.51°S, 56.50°W) from 1993 to 2002. A moving
average of ±1 month is applied to all curves. The shading indicates El Niño events. (From Papa, F. et al.,
J. Geophys. Res., 115, D12111, 2010b.)
Runoff Data Center (GRDC) archives (GRDC 2009). For the Ganges–Brahmaputra, the discharge
products from Papa et al. (2010a) that combine in situ and satellite altimetry-derived observations
were used. For all three rivers, the plots in Figure 24.9 show a good agreement between the surface
water extent and the river discharge (both time series are normalized deseasonalized anomalies).
Over 15 years, the correlation coefficients between the two monthly variables are in general really
Ganges–Brahmaputra
9 3 1
6 2
0
3 1
0 0 –1
Discharge (105 m3/s)
Mississippi
3 1.0 3
Extent (105 km2)
2
2
0.5 1
1 0
–1
0 0.0
Mackenzie
3 0.6 2
2 0.4 1
1 0.2 0
–1
0 0.0
1993 1995 1997 1999 2001 2003 2005 2007 1993 1995 1997 1999 2001 2003 2005 2007
(a) (b)
FIGURE 24.9 (a) Variability of (black) the monthly mean wetland extent between 1993 and 2007 over three
river basins and (red) corresponding in situ river discharge variations (when available). (b) The corresponding
normalized deseasonalized anomalies are plotted on the right panel. (From Papa, F. et al., J. Geophys. Res.,
115, D12111, 2010b.)
high. For example, over the common period of availability of the data over the Ganges–Brahmaputra
basin, the linear correlation between the water extent and in situ discharge is 0.90 (p < 0.001) for the
raw data and 0.66 (p < 0.001) for the deseasonalized results. For all rivers, the positive and nega-
tive peaks in the in situ discharge during the 15 years are usually well captured by the inundation
estimates.
Since 2002, the Gravity Recovery and Climate Experiment (GRACE) gravity mission offers, for
the first time, direct estimates of the spatiotemporal variations of total terrestrial water storage (the
sum of groundwater, soil water, surface water, and snow pack) at time scales from months to several
years (Tapley et al. 2004). Figure 24.10 compares the monthly time series of precipitation (as esti-
mated from the Global Precipitation Climatology Project [GPCP]) (Adler et al. 2003), in situ river
discharge, the surface water extent, and GRACE-based total water storage over 2003–2007 for four
large river basins (the data are aggregated to basin averages) that represent different environments
(e.g., tropical, monsoon dominated, mid-latitudes, and boreal regions). Regardless of the environ-
ment, this figure shows that the seasonal and interannual patterns of the time series are generally
in good agreement with some highly significant linear correlations (>0.80 with 0 or ±1 month lag)
between the surface water extent and the other variables. Time lags illustrate that, in large basins
such as the Amazon River or the Mississippi River, stream flows in upstream regions contribute
with a delay in time to large downstream flooding due to the long concentration times in the large
hydrographic network. For the Ganges–Brahmaputra basin, which receives intense local rainfall
during the annual monsoon, the precipitation and the surface water extent are highly correlated
with no lag in time (R = 0.93). In the boreal region, the Mackenzie basin exhibits less agreement
between the different variables, especially regarding the amplitude of the signal. Indeed, the basin
hydrology is complicated by the flood dynamics partially driven by spring snowmelt and ice jam
(Papa et al. 2007, 2008b). Phase differences and the lagged temporal correlations between precipi-
tation, water surface extent, discharge, and GRACE total water storage reveal the different effect
of water transport processes within large river basins and help to better understand the large water
travel, accumulation times, and floodplain storage and groundwater recharges in large river basins
(Frappart et al. 2008, 2010, 2011, 2012; Papa et al. 2008a,b).
Amazon
2
1
0
–1
Ganges–Brahmaputra
1
–1
Mississippi
2
1
0
–1
Mackenzie
2
1
0
–1
2003 2005 2007
FIGURE 24.10 Variability of (black) the wetland extent between 2003 and 2007 (deseasonalized and nor-
malized anomaly) over four basins and (red) corresponding river discharge, (blue) basin-averaged precipita-
tion, and (green) GRACE-derived total water volume change.
GIEMS APPLICATIONS
Hydrological Applications
Evaluation of Global Hydrological Modeling
In general circulation models (GCMs), wetlands and surface waters are generally not parameter-
ized. Efforts have been made recently to develop a simple flooding river scheme in GCMs. GIEMS
has been used to evaluate this modeling results. At Météo-France, the interaction soil–biosphere–
atmosphere (ISBA) land surface model and a river routing model including a prognostic flood res-
ervoir have been coupled to simulate the flood dynamics. This reservoir fills when the river height
exceeds the critical river bank full height (elevation) and vice versa. The flood interacts with the soil
hydrology through infiltration and with the overlying atmosphere through precipitation interception
and free water surface evaporation. The results show reasonable agreement between the simulated
flooded areas and satellite-derived inundation estimates, and they show an improved monthly dis-
charges for most basins (Decharme et al. 2008, 2012). Figure 24.11 presents a comparison over
South America. Similar efforts have been performed for other hydrological models, for instance,
in the Organising Carbon and Hydrology in Dynamic Ecosystems (ORCHIDEE) model (Ringeval
et al. 2012), and the results have been compared with GIEMS as well.
Estimation of Surface Water Volume Change

Spatiotemporal variations of water volume over inundated areas have been determined using com-
bined observations from our inundation extent dataset and water height from satellite altimetry.
Obs Flood
15N 15N
EQ EQ
15S 15S
30S 30S
45S 45S
60S 60S
80W 60W 40W 80W 60W 40W
1 2.5 5 10 15 20 1 2.5 5 10 15 20
(a) (b)
FIGURE 24.11 (a) Simulated flood fraction with the ISBA model (in percentage of 1° × 1°) and (b) spa-
tial comparisons between GIEMS flood fraction (in percentage of 1° × 1° (b). (From Decharme, B. et al.,
J. Geophys. Res., 113, D11110, 2008.)
Maps of monthly surface water volume change can be generated. The results show the high poten-
tial for GIEMS to provide valuable information to improve our understanding of large river basin
hydrologic processes. The basin of the Negro River, the largest tributary in terms of discharge to
the Amazon River, was selected as a test site in Frappart et al. (2008). These surface water changes
have been further used to estimate water stored in aquifer. Changes in water stored in the aquifer are
isolated from the total water storage measured by GRACE by removing contributions of both the
surface volume change, derived from GIEMS and altimetry, and the root zone reservoir simulated
by hydrological models. The groundwater anomalies show a realistic spatial pattern compared with
the hydrogeological map of the basin and similar temporal variations to local in situ groundwater
observations (Frappart et al. 2011).
More recently, continuous multisatellite observations of inundation extent and water levels
between 2003 and 2007 have been used to monitor monthly variations of surface water storage for
the whole Amazon River basin (Frappart et al. 2012). During the 2005 drought, it has been shown
that the amount of water stored in the river and floodplains of the Amazon basin was ∼130 km3
(∼70%) below its 2003–2007 average.
Over the entire Amazon basin, GIEMS was also combined with topographic data from the global
digital elevation model (GDEM) from Advanced Spaceborne Thermal Emission and Reflection
Radiometer (ASTER) in order to derive surface freshwater water volume variations using a hypso-
graphic curve approach (Papa et al. 2013). Monthly surface water storage variations for 1993–2007
show a basin-scale mean annual amplitude of ~1200 km3 associated with large interannual vari-
ability and contribute to about half of GRACE total water storage variations. For the first time, the
surface water volume anomalies were mapped for the extreme droughts of 1997 and 2005. Such
database of water volume change will be soon available at a global scale.
Applications to CH4 Modeling

Estimation of CH4 Emission from Wetland Extent
Methane (CH4) is an important greenhouse gas, and its atmospheric concentration has nearly tri-
pled since preindustrial times (e.g., Shindell et al. 2004). The growth rate of atmospheric methane
is determined by the balance between surface emissions and photochemical destruction by the
hydroxyl radical, the major atmospheric oxidant. Since natural wetlands are the world’s largest
methane source, the GIEMS dataset provides a unique resource for reducing uncertainties in the
role of natural wetlands in the interannual variability of the growth rate of atmospheric methane.
First, GIEMS has been used directly to model the CH4 emissions from natural wetlands and to
assess the impact of varying wetland area on the seasonal and interannual variability of CH4 wet-
land emissions. In Ringeval et al. (2010), wetland area is optimized using GIEMS as a first estimate
and further adjusted to match the seasonal cycle of CH4 fluxes retrieved from a global atmospheric
inversion. CH4 emissions are calculated by coupling the ORCHIDEE global vegetation model with
a process-based wetland CH4 emission model. For bogs north of 50°N, variations in wetland area
contributed about 30% to the annual flux. For temperate and tropical wetlands, the variations in area
have almost no influence on the annual CH4 emissions but contribute significantly to the seasonal
behavior. In contrast, the interannual variability of wetland area appears to be the dominant cause
of interannual variations in regional CH4 emissions from wetlands at all latitudes. This study points
to the necessity to predict the variations in wetland extent, in order to obtain reliable simulations of
changing methane emissions perturbed by climate (Ringeval et al. 2010). The study of the boreal
region by Petrescu et al. (2010) demonstrates the feasibility of estimating interannual variations
in CH4 emissions by coupling hydrological and CH4 emission process models and highlights the
importance of an adequate understanding of hydrology in quantifying the total emissions from
northern wetlands. Hodson et al. (2011) explain a large fraction of the global and tropical vari-
ability in wetland CH4 emissions through correlation with the ENSO index. The Wetland CH4
Inter-comparison of Models Project (WETCHIMP) is a European initiative that investigates the
ability to simulate large-scale wetland characteristics and corresponding CH4 emissions. To ensure
intercomparability, most models use GIEMS dataset to estimate the wetland location, extent, and
seasonality (Melton et al. 2013).
Evaluation of Inverse CH4 Modeling

GIEMS has been used to evaluate results from global atmospheric inversion. Bousquet et al. (2006)
quantified the processes that control variations in methane emissions, using an inversion model of
atmospheric transport and chemistry. Atmospheric CH4 observations, from air samples from 68
sites, were collected and were inverted. The results indicated that fluctuations in wetland emissions
are the dominant contribution to interannual variability in surface emission, explaining 70% of the
global emission anomalies over the last two decades, as compared with only 15% contributed by
biomass burning. This top-down estimate of changes in wetland emission is in good agreement with
our multisatellite estimate of wetland extent.
CONCLUSION AND FUTURE PERSPECTIVE

For climate-related applications, there is a need for global estimates of surface water extent and
dynamics over a long time periods. Monthly mean estimates of surface water extent have been
derived from a combination of satellite observations with a spatial resolution of 25 km. Global
inundated area varies from a maximum of 5.9 × 106 km2 to a mean minimum of 2.1 × 106 km2.
GIEMS has been thoroughly evaluated by comparison with existing independent static inventories
(including regional estimates at higher spatial resolution) and with hydrological variables such as
river height from altimetry or river runoff. It has been adopted in several studies for hydrological
modeling or for the estimation of CH4 emission. The GIEMS dataset, covering the time period from
1993 to 2007 on a monthly basis, is available upon request. It should be available soon for the same
period of time, on a 10-day average.
In the early 2000s, when the multisatellite methodology was being developed to estimate
global wetland extent, the use of multisatellite observations was dictated by the limitation of
each observation alone for mapping the wetlands in all types of environments. Since then, the
Soil Moisture and Ocean Salinity (SMOS) mission (Kerr et al. 2001) and Aquarius have been
launched with observations at 1.4 GHz (L-band), with increased sensitivity to soil moisture and
standing water, limited impact from the vegetation, and no effect from the atmosphere. Note that
these two missions only provide limited spatial resolution (50 km for SMOS and even worse for
Aquarius) compared with 25 km resolution from GIEMS. Developments are underway to develop
a retrieval method to estimate surface water extent from SMOS. Two other satellite missions are
planned and should have potential for surface water mapping. The NASA Soil Moisture Active
Passive (SMAP) will be equipped with both passive and active instruments between 1.2 and 1.4
GHz. The active instrument (a SAR) will provide a spatial resolution of 3 km and will help down-
scale the passive observations that will be available with a broader swath. The Surface Water
Ocean Topography (SWOT) NASA-CNES joint satellite mission, to be launched around 2020,
will be equipped with wide-swath altimetry and will completely cover the world’s oceans and
freshwater bodies with repeated high-resolution elevation measurements. Its main objectives for
continental hydrology include the provision of a global inventory of all terrestrial surface water
bodies whose surface area exceeds 250 m × 250 m (lakes, reservoirs, wetlands) and rivers whose
width exceeds 100 m.
REFERENCES
Adam, L., P. Doll, C. Prigent, and F. Papa. 2010. Global-scale analysis of satellite-derived time series of natu-
rally inundated areas as a basis for floodplain modeling. Advances in Geosciences 27:45–50, doi:10.5194/
adgeo-27-45-2010, 2010.
Adler, R.F., G.J. Huffman, A. Chang, R. Ferraro, P. Xie, J. Janowiak, B. Rudolf et al. 2003. The Version 2
Global Precipitation Climatology Project (GPCP) monthly precipitation analysis 1979-present. Journal
of Hydrometeorology 4:1147–1167.
Alsdorf, D., P. Bates, J. Melack, M. Wilson, and T. Dunne. 2007a. Spatial and temporal complexity of the Amazon
flood measured from space. Geophysical Research Letters 34:L08402, doi:10.1029/2007GL029447.
Alsdorf, D.E., E. Rodrı́guez, and D.P. Lettenmaier. 2007b. Measuring surface water from space. Reviews of
Geophysics 45:RG2002, doi:10.1029/2006RG000197.
Armstrong, R.L. and M.J. Brodzik. 2005. Northern Hemisphere EASEGrid Weekly Snow Cover and Sea Ice
Extent Version 3. National Snow and Ice Data Center, Boulder, CO.
Berge-Nguyen, M., J.F. Crétaux, and S. Calmant. November 2008. Combining of radar altimetry and MODIS
for the monitoring of flood events: Application to the Inner Niger delta. Observing and forecasting the
Ocean. OSTST Meeting, pp. 1–13. Nice, France.
Birkett, C.M. 1998. Contribution of the TOPEX NASA radar altimeter to the global monitoring of large rivers
and wetlands. Water Resources Research 34(5):1223–1239.
Birkett, C.M., L.A.K. Mertes, T. Dunne, M.H. Costa, and M.J. Jasinski. 2002. Surface water dynamics in the
Amazon basin: Application of satellite radar altimetry. Journal of Geophysical Research 107(D20):8059,
doi:10.1029/2001JD000609.
Bousquet, P., P. Ciais, J.B. Miller, E.J. Dlugokencky, D.A. Hauglustaine, C. Prigent, G.R. Van der Werf et al.
2006. Contribution of anthropogenic and natural sources to atmospheric methane variability. Nature
443:439–443, doi:10.1038/nature05132.
Brakenridge, G.R., E. Anderson, S.V. Nghiem, and S. Chien. 2005. Space based measurement of river runoff.
EOS Transactions of the American Geophysical Union 86:185–188.
Calmant, S., F. Seyler, and J.F. Cretaux. 2008. Monitoring continental surface waters by satellite altimetry.
Surveys in Geophysics 29(4–5):247–269, doi:10.1007/s10712-008-9051-1.
Cogley, J. 2003. GGHYDRO-Global Hydrographic Data, release 2.3. Technical Note 2003-1. Department of
Geography, Trent University, Peterborough, Ontario, Canada.
Crétaux, J.-F. and C. Birkett. Lake studies from satellite altimetry. Comptes Rendus Geoscience 338:14–15,
1098–1112, doi: 10.1016/J.cre.2006.08.002, 2006.
Decharme, B., R. Alkama, F. Papa, S. Faroux, H. Douville, and C. Prigent. 2012. Global off-line evaluation of
the ISBA-TRIP flood model. Climate Dynamics 1389–1412, doi:10.1007/s00382-011-1054-9.
Decharme, B., H. Douville, C. Prigent, F. Papa, and F. Aires. 2008. A new river flooding scheme for global
climate applications: Off-line evaluation over South America. Journal of Geophysical Research
113:D11110, doi:10.1029/2007JD009376.
Finlayson, C.M., N.C. Davidson, A.G. Spiers, and N.J. Stevenso. 1999. Global wetland inventory and current
status and future priorities. Marine and Freshwater Research 50:717–727.
Frappart, F., F. Papa, J.S. Famiglietti, C. Prigent, W.B. Rossow, and F. Seyler. 2008. Interannual variations
of river water storage from a multiple satellite approach: A case study for the Rio Negro River Basin.
Journal of Geophysical Research 113:D21104, doi:10.1029/2007JD009438.
Frappart, F., F. Papa, A. Güntner, S. Werth, G. Ramilien, C. Prigent, W.B. Rossow, and M.-P. Bonnet. 2010.
Interannual variations of the terrestrial water storage in the Lower Ob’ Basin from a multisatellite
approach. Hydrology and Earth System Sciences 14:2443–2453.
Frappart, F., F. Papa, A. Güntner, S. Werth, J. Santos da Silva, J. Tomasella, F. Seyler et al. 2011. Satellite-based
estimates of groundwater storage variations in large drainage basins with extensive floodplains. Remote
Sensing of the Environment 115:588–1594, doi:10.1016/j.rse.2011.02.003.
Frappart, F., F. Papa, J. Santos da Silva, G. Ramillien, C. Prigent, F. Seyler, and S. Calmant. 2012. Surface fresh-
water storage and dynamics in the Amazon basin during the 2005 exceptional drought. Environmental
Research Letters 7:044010, doi:10.1088/1748−9326/7/4/044010.
Fu, L.L. and A. Cazenave. 2001. Satellite Altimetry and Earth Sciences. A Handbook of Techniques and
Applications. International Geophysics Series, 69. Academic Press, San Diego, CA.
Gedney, N., P.M. Cox, and C. Huntingford. 2004. Climate feedback from wetland methane emission.
Geophysical Research Letters 31:L20503, doi:10.1029/2004GL020919.
Giddings, L. and B.J. Choudhury.1989. Observation of hydrological feature with Nimbus-7 37 GHz data
applied to South America. International Journal of Remote Sensing 10:1673–1686.
Global Runoff Data Centre (GRDC). 2009. Long-term mean monthly discharges and annual characteristics
of GRDC Stations, Koblenz, Germany. Available at http://www.bafg.de/GRDC/EN/ Home/homepage_
node.html, Fed. Inst. of Hydrol.
Hess, L.L., J.M. Melack, E.M.L.M. Novo, C.C.F. Barbosac, and M. Gastil. 2003. Dual-season mapping of wetland
inundation and vegetation for the central Amazon basin. Remote Sensing of the Environment 87:404–428.
Hodson, E.L., B. Poulter, N.E. Zimmermann, C. Prigent, and J.O. Kaplan. 2011. The El Niño–Southern
Oscillation and wetland methane interannual variability. Geophysical Research Letters 38:L08810,
doi:10.1029/2011GL046861.
Houweling, S., T. Kamisnki, F. Dentefer, J. Lelieveld, and M. Heinmann. 1999. Inverse modeling of meth-
ane sources and sinks using the adjoint of a global transport model. Journal of Geophysical Research
104:26137–26160.
Intergovernmental Panel on Climate Change (IPCC). 2007. Climate Change. 2007. The Physical Science Basis.
Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on
Climate Change (Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor, and
H.L. Miller (eds.)). Cambridge University Press, Cambridge, U.K.
Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. lredell et al. 1996. The NCEP/
NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society 77:437–470.
Kerr, Y.H., P. Waldteufel, J.P. Wigneron, J.M. Martinuzzi, J. Font, and M. Berger. 2001. Soil moisture retrieval
from space: The Soil Moisture and Ocean Salinity (SMOS) mission. IEEE Transactions on Geoscience
and Remote Sensing 39:1729–1735.
Kohonen, T. 1984. Self-Organization and Associative Memory. Springer, New York.
Lehner, B. and P. Doll. 2004. Development and validation of a global database of lakes, reservoirs and wet-
lands. Journal of Hydrology 296:1–22.
Loveland, T.R., B.C. Reed, J.F. Brown, D.O. Ohlen, J. Zhu, L. Yang, and J.W. Merchant. 2000. Development of
a global land cover characteristics database and IGBP DISCover from 1-km AVHRR data. International
Marburger, J.H. and J.B. Bolten. 2005. FY 2007 administration research and development budget priorities.
Memo. M-05-18. Executive Office of the U.S. President, Washington, DC. (http://www.ostp.gov/html/
budget/2007/ostp_omb_guidancememo_FY07.pdf.)
Matthews, E. and I. Fung. 1987. Methane emission from natural wetlands: Global distribution, area, and envi-
ronmental characteristics of sources. Global Biogeochemical Cycles 1:61–86.
Matthews, E., I. Fung, and J. Lerner. 1991. Methane emission from rice cultivation: Geographic and seasonal
distribution of cultivated areas and emissions. Global Biogeochemical Cycles 5:3–24.
Melton, J.R., R. Wania, E.L. Hodson, B. Poulter, B. Ringeval, R. Spahni, T. Bohn et al. 2013. Present state
of global wetland extent and wetland methane modelling: Conclusions from a model inter-comparison
project (WETCHIMP). Biogeosciences 10:753–788, doi:10.5194/bg-10-753-2013.
Mialon, A., A. Royer, and M. Fily. 2005. Wetland seasonal dynamics and interannual variability over northern
high latitudes derived from microwave satellite data. Journal of Geophysical Research 110:D17102,
doi:10.1029/2004JD005697.
Papa, F., F. Durand, W.B. Rossow, A. Rahman, and S.K. Bala. 2010a. Seasonal and interannual variations of
the Ganges-Brahmaputra river discharge, 1993–2008 from satellite altimeters. Journal of Geophysical
Research 115:C12013, doi:10.1029/2009JC006075.
Papa, F., F. Frappart, A. Guntner, C. Prigent, F. Aires, A.C.V. Getirana, and R. Maurer. 2013. Surface freshwater
storage and variability in the Amazon basin from multi-satellite observations, 1993–2007. Journal of
Geophysical Research 118:11,951–11,965, doi:10.1002/2013JD020500.
Papa, F., A. Güntner, F. Frappart, C. Prigent, and W.B. Rossow. 2008a. Variations of surface water extent and
water storage in large river basins: A comparison of different global data sources. Geophysical Research
Letters 35:L11401, doi:10.1029/2008GL033857.
Papa, F., C. Prigent, F. Durand, and W.B. Rossow. 2006a. Wetland dynamics using a suite of satellite observa-
tions: A case study of application and evaluation for the Indian Subcontinent. Geophysical Research
Letters 33:L08401, doi:10.1029/2006GL025767.
Papa, F., C. Prigent, C. Jimenez, F. Aires, and W.B. Rossow. 2010b. Interannual variability of surface water extent
at global scale, 1993–2004. Journal of Geophysical Research 115:D12111, doi:10.1029/2009JD012674.
Papa, F., C. Prigent, and W.B. Rossow. 2007. Ob’ River flood inundations from satellite observations: A rela-
tionship with winter snow parameters and river runoff. Journal of Geophysical Research 112:D18103,
doi:10.1029/2007JD008451.
Papa, F., C. Prigent, and W.B. Rossow. 2008b. Monitoring flood and discharge variations in the large
Siberian Rivers from a multi-satellite technique. Surveys in Geophysics 29:297–317, doi:10.1007/
s10712-008-9036-0.
Papa, F., C. Prigent, W.B. Rossow, B. Legresy, and F. Remy. 2006b. Inundated wetland dynamics over boreal
regions from remote sensing: The use of Topex-Poseidon dual-frequency radar altimeter observations.
International Journal of Remote Sensing 27:4847–4866, doi:10.1080/01431160600675887.
Petrescu, A.M.R., R.L.P.H. van Beek, J. Van Huissteden, C. Prigent, T. Sachs, C.A.R. Corradi, F.W. Parmentier,
and A.J. Dolman. 2010. Modeling regional to global CH4 emissions of boreal and arctic wetlands. Global
Biogeochemical Cycles 24:GB4009, 10.1029/2009GB003610.
Portmann, F.T., S. Siebert, and P. Döll. 2010. MIRCA2000—Global monthly irrigated and rainfed crop areas
around the year 2000: A new high-resolution data set for agricultural and hydrological modeling. Global
Biogeochemical Cycles 24:GB 1011, doi:10.1029/2008GB003435.
Prigent, C., F. Aires, and W.B. Rossow. 2006. Land surface microwave emissivities over the globe for a decade.
Bulletin of the American Meteorological Society 87:1573–1584, doi:10.1175/BAMS-87-11-1573.
Prigent, C., F. Aires, W.B. Rossow, and E. Matthews. 2001a. Joint characterization of vegetation by satel-
lite observations from visible to microwave wavelength: A sensitivity analysis. Journal of Geophysical
Research 106:20665–20685.
Prigent, C., E. Matthews, F. Aires, and W.B. Rossow. 2001b. Remote sensing of global wetland dynamics with
multiple satellite data sets. Geophysical Research Letters 28:4631–4634.
Prigent, C., F. Papa, F. Aires, C. Jimenez, W.B. Rossow, and E. Matthews. 2012. Changes in land surface water
dynamics since the 1990s and relation to population pressure. Geophysical Research Letters 39:L08403,
doi:10.1029/2012GL051276.
Prigent, C., F. Papa, F. Aires, W.B. Rossow, and E. Matthews. 2007. Global inundation dynamics inferred
from multiple satellite observations, 1993–2000. Journal of Geophysical Research 112:D12107,
doi:10.1029/2006JD007847.
Ramsar Convention. 1971. The Convention on Wetlands of International Importance Especially as Waterfowl
Habitat. UNESCO, Ramsar, Iran.
Richey, J.E., C. Nobre, and C. Deser. 1989. Amazon River discharge and climate variability: 1903 to 1985.
Science 6:101–103, doi:10.1126/science.246.4926.101.
Ringeval, B., B. Decharme, S.L. Piao, P. Ciais, F. Papa, N. de Noblet-Ducoudré, C. Prigent et al. 2012. Modelling
sub-grid wetland in the ORCHIDEE global land surface model: Evaluation against river discharges and
remotely sensed data. Geoscientific Model Development Discussion 5:941–962.
Ringeval, B., N. de Noblet-Ducoudre, P. Ciais, P. Bousquet, C. Prigent, F. Papa, and W.B. Rossow. 2010. An
attempt to quantify the impact of changes in wetland extent on methane emissions at the seasonal and
interannual time scales. Global Biogeochemical Cycles 24:GB2003, doi:10/1029/2008GB003354.
Rossow, W.B. and R.A. Schiffer. 1999. Advances in understanding clouds from ISCCP. Bulletin of the American
Meteorological Society 80:2261–2287.
Rudolf, B. and U. Schneider. 2005. Calculation of gridded precipitation data for the global land-surface using
in-situ gauge observations. Proceedings Second Workshop of the International Precipitation Working
Group IPWG. Monterey, CA. October 2004, pp. 231–247, EUMETSAT, Darmstadt, Germany.
Sakamoto, T., N.Van Nguyen, A. Kotera, H. Ohno, N. Ishitsuka, and M. Yokozawa. 2007. Detecting tempo-
ral changes in the extent of annual flooding within the Cambodia and the Vietnamese Mekong Delta
from MODIS time series imagery. Remote Sensing of the Environment 109:295–313, doi:10.1016/
j.rse.2007.01.011.
Shindell, D.T., G. Falugevi, N. Bell, and G.A. Schmidt. 2004. An emission-based view of climate forcing by
methane and tropospheric ozone. Geophysical Research Letters 32:L04803, doi:10.1029/2004GL021900.
Sippel, S.J., S.K. Hamilton, J.M. Melack, and E.M.M. Novo. 1998. Passive microwave observations of inunda-
tion area and the area/stage relation in the Amazon River floodplain. International Journal of Remote
Sensing 19:3055–3074.
Tapley, B.D., S. Bettadpur, J.C. Ries, P.F. Thompson, and M. Watkins. 2004. GRACE measurements of mass
variability in the Earth system. Science 305:503–505.
United Nations. 2004. International Decade for Action. “Water for Life” 2005–2015. General Assembly
Resolution. A/RES/58/217, New York. Available at www.un.org/Depts/dhl/resguide/r58.htm.
Vörösmarty, C.J., P. Green, J. Salisbury, and R.B. Lammers. 2000. Global water resources: Vulnerability from
climate changes and population growth. Science 289:284–288.
Section IV
Promising Developments
and Future Challenges
25 Promising Developments
and Future Challenges for
Remote Sensing of Wetlands
Megan W. Lang, Sam Purkis, Victor V. Klemas,
and Ralph W. Tiner
CONTENTS
Introduction..................................................................................................................................... 533
LiDAR............................................................................................................................................. 534
Synthetic Aperture Radar................................................................................................................ 534
Landsat Time Series........................................................................................................................ 535
Data Fusion..................................................................................................................................... 535
Advanced Classification Techniques............................................................................................... 536
Faster, Lower-Cost Data Dissemination......................................................................................... 536
Other Advances............................................................................................................................... 537
UAS............................................................................................................................................ 537
Advances in Temporal, Spatial, and/or Spectral Resolution...................................................... 537
Hyperspectral Imagery............................................................................................................... 538
Future Challenges for Wetland Mapping and Monitoring.............................................................. 538
Predicting Wetland Function...................................................................................................... 538
Detection of Wetland Water Regimes........................................................................................ 539
Plant Species and Properties......................................................................................................540
Submerged Wetlands and Aquatic Habitats............................................................................... 541
International Wetland Definition and Classification Standard........................................................ 541
Conclusion...................................................................................................................................... 543
References....................................................................................................................................... 543
INTRODUCTION
Wetlands are challenging targets to map and monitor because of their spatial and spectral com-
plexity, as has been detailed throughout this book (see Chapter 3 for a review of major issues). The
diversity of remotely sensed data and the techniques available to process these data have increased
rapidly since the 1970s, when the United States and other countries began to systematically map
national wetland resources. Recently developed remote sensing technologies described throughout
this book have demonstrated their potential to further improve the detail and reliability of wetland
maps. Contributors to this text have a wide variety of wetland mapping, monitoring, and classifica-
tion expertise and have examined a diverse array of focus areas in terms of habitat type, climate,
spatial scale, and topics of interest (e.g., water level and invasive species). These varying focus areas
require different datasets and techniques, but when considered broadly, some trends become evi-
dent. What follows is a brief description of some of the most promising new developments in remote
sensing data and techniques highlighted in this text and implications for future wetland mapping and
533
monitoring. The systems and techniques that are highlighted are still very active within the research
domain but have also begun to contribute to operational wetland mapping. Topics requiring more
research before broader application operationally are discussed in the section—Other Advances.
LiDAR
The wetland management and science communities stand to benefit significantly from the collection
of light detection and ranging (LiDAR) data. Elevation data collected using more traditional means,
such as stereo-interpretation of aerial photographs, are often unable to detect the relatively small
variations in elevation that lead to the formation of wetlands, especially on coastal and glaciolacus-
trine plains. Furthermore, stereo-interpretation of aerial photographs in support of wetland mapping
(e.g., U.S. National Wetlands Inventory [NWI]) has been largely replaced with on-screen monoscopic
interpretation of digital imagery, which provides no information on elevation unless integrated with
other sources (e.g., digital elevation data). In contrast, LiDAR data provide highly accurate, fine-
resolution digital information on topography. This technique works well in many forested wetlands,
which are among the most difficult types of wetlands to map. LiDAR data can be incorporated into
the mapping process via multiple formats, including LiDAR intensity data, LiDAR-based digital ele-
vation models (DEMs), and derived topographic metrics. Map makers can take advantage of these
flexible, information-rich datasets through techniques ranging from visual interpretation of LiDAR
base maps to inclusion within machine learning and other advanced algorithms. These data can be
used both to guide fine spatial scale wetland mapping (e.g., NWI) and to enhance moderate spatial
scale wetland mapping efforts (e.g., NOAA Coastal Change Analysis Program [C-CAP]) through
the fusion of multispectral satellite data and LiDAR-derived topographic metrics.
Although LiDAR data are not available worldwide, the current use of LiDAR data provides a
foundation upon which to develop even more advanced applications so that the global wetland map-
ping community can take full advantage of LiDAR’s significant benefits when these data become
available. LiDAR data are used to supplement aerial photography to create NWI maps, and some
NWI map creators consider LiDAR data to be their most important ancillary dataset (Snyder and
Lang 2012) (see Chapter 9). In addition to greater data availability, increased application and proto-
col development and quantification of benefits derived from enhanced elevation data would greatly
benefit the successful operational use of these data. Many chapters in this book describe LiDAR
applications (Chapters 5, 9 through 15, 16, 18, and 19).
SYNTHETIC APERTURE RADAR

The integral role that synthetic aperture radar (SAR) plays in wetland mapping projects described
in this book highlights the importance of this dataset. The utility of SAR for wetland mapping is
primarily based on its sensitivity to water—both as soil moisture and inundation—and its ability
to collect data regardless of cloud cover, solar illumination, and most rain events and even through
many types of plant canopies. These features make SAR well adapted not only for wetland map-
ping but also for monitoring hydroperiod and estimating wetland function (e.g., carbon and nutrient
regulation). SAR-based information on water height and plant structure is also vital to improved
characterization of wetlands at the landscape scale.
A variety of different types of SAR sensors are currently operational or planned, each better
suited for specific applications (see Chapter 5). The historic record of SAR data, although not as
lengthy as that of Landsat, is helpful for detecting change. Finer spatial resolution data are now
being offered with a variety of different polarizations and wavelengths. Polarimetric SAR (PolSAR)
is available when additional information is needed concerning the source and character of backscat-
tered energy; this can be compared to the use of multiple bands in multispectral remote sensing.
Interferometric SAR (InSAR) data are also readily available, when the height of objects, including
the water surface, is important to the mapping project.
Promising Developments and Future Challenges for Remote Sensing of Wetlands 535
Recognizing the value of SAR data, natural resource managers are becoming increasingly reli-
ant on this data source for wetland mapping and monitoring, particularly when wetland boundaries
are obscured by a plant canopy or optical data are not available due to cloud cover or the lack of
solar illumination. Although the application of SAR for wetland mapping is still evolving, the use
of this dataset in support of the Canadian Wetland Inventory demonstrates its operational potential
(Chapters 5, 6, and 15), while case studies in several other chapters (Chapters 7 through 9, 15, 17,
20, 21, 23, and 24) show its use in cutting edge research studies.
LANDSAT TIME SERIES

The dynamic nature of wetlands, in terms of their response to changes in phenology, weather, and
hydrology, can make wetlands more difficult to differentiate from the surrounding landscape at
certain times (e.g., growing season and drought), and it often makes change detection more com-
plicated. These dynamics can, however, provide an important mechanism for improved wetland
detection (Chapters 5, 9, 12 through 14, 16, 22, and 23). A time series of images reveals changes in
spectral signatures through time that are often unique and can facilitate separating wetlands from
surrounding landscapes. These temporal signatures also offer important information for detecting
wetland change, both in terms of land cover or land use change and change in response to weather
and other ecological drivers. For these reasons the extensive record of Landsat historic data is
invaluable and unique among moderate to fine spatial scale sensors.
The Landsat satellite series has been routinely collecting global earth observations in the vis-
ible and near-infrared bands since 1972 (15–60 m spatial resolution), the thermal band since 1978
(60–120 m spatial resolution), and the mid-infrared band since 1982 (30 m spatial resolution). The
resulting dataset is the only long-term moderate-resolution civilian archive of satellite imagery.
The Landsat Data Continuity Mission (LDCM; Landsat-8), a joint USGS and National Aeronautics
and Space Administration (NASA) mission, was launched in 2013 and is designed to extend the
Landsat historic record into the future. In addition to the advantages provided by the long-term
historic record, the enhanced radiometric resolution of Landsat 8 improves detection capacity for
wet or submerged targets that are often dark and therefore difficult to resolve. Although operational
wetland mapping efforts do not take full advantage of the Landsat historic record, the fact that
operational wetland mapping efforts (e.g., C-CAP) already incorporate multiple Landsat images
bodes well for the future inclusion of a more robust temporal series. Challenges to the expanded use
of Landsat data for wetland mapping and monitoring include its moderate spatial resolution when
mapping smaller wetlands or wetland features and the limited ability of Landsat to detect variations
in water regime, especially below the vegetative canopy. An additional challenge includes radiomet-
ric variability between images due to the atmosphere, sensor characteristics, and other factors that
complicate interimage comparison.
DATA FUSION
As outlined in Chapter 5, the advancement of wetland mapping and monitoring demands the use
of multiple types of remotely sensed data, each capable of detecting different components of the
wetland signature (e.g., water regime and vegetation type). These unique datasets can be brought
together to improve the characterization of the wetland landscape through data fusion where mul-
tiple types of complementary data are used in unison (e.g., simultaneous consideration of optical
and radar data [Chapters 8, 9, 15, 20, and 24], optical and LiDAR data [Chapters 9 and 18]). The
simultaneous use of multiple data types is often necessary when attempting to map wetland types
or characteristics that are difficult to distinguish using one type of data due to spectral, elevation,
or other similarities with adjacent classes. Equally important is the capability to fuse data acquired
in the same portion of the electromagnetic spectrum but with dissimilar spatial or temporal resolu-
tions. Most common in this realm is the fusion of a multispectral coarse resolution image with a
finer spatial resolution image (e.g., panchromatic band). Here the goal is to obtain a high-resolution
multispectral image that combines the spectral characteristic of the low-resolution data with the
spatial resolution of the panchromatic image.
We predict that data fusion will greatly assist in mapping wetland classes and characteristics
that have previously been difficult to detect. The current use of LiDAR and aerial photography to
produce NWI maps, and the use of LiDAR with moderate-resolution multispectral images, aerial
photo-based ancillary products (e.g., NWI and Soil Survey Geographic Database), and elevation
data to produce C-CAP maps, provides evidence of the advantages of these composite datasets
for operational mapping. Future advances will enhance fusion techniques through more rigorous
preservation of the spectral information within each pixel for tasks such as unmixing and other
enhanced classification techniques. Improved algorithms to fully automate the registration of data
captured with several sensors at dissimilar resolutions and using different acquisition modes would
also greatly advance the application of fusion techniques. Additional background information on
data fusion is provided in Chapter 8.
ADVANCED CLASSIFICATION TECHNIQUES

Many advances in image classification techniques have taken place during the last two decades
(Chapter 5), and further refinement of classification algorithms will undoubtedly occur in the
future. While the parametric classification routines that have been the mainstay for mapping wet-
lands using satellite imagery over the last 30 years and will continue to be employed, a greater
degree of automation is expected to lessen the time required to convert a raw image into useful
geographic knowledge. Increased reliance on textural, spatial, and context-based metrics is likely,
as is the further incorporation of ancillary data (e.g., topographic models, image time series, and
vector data) into the classification process through decision tree and other classification techniques
(e.g., Chapters 9 and 20). Advances also include further development of image segmentation rou-
tines, pattern recognition, and object-oriented classification (e.g., Chapters 9 and 13), which is par-
ticularly helpful when observing wetlands with finer spatial resolution datasets. These technologies
have enjoyed a resurgence with the advent of powerful processors and neural-network analysis.
Machine learning algorithms (e.g., Random Forests; Breiman 2001) have been particularly popu-
lar for moderate-resolution wetland mapping and often take advantage of an ensemble of deci-
sion trees through enhanced training (e.g., bootstrap) and/or weighting (e.g., boosting) techniques.
C-CAP currently uses machine learning algorithms (See5 and Cubist software; Rulequest) as well
as spatial recognition software (Ecognition; Trimble). Chapters 9, 12, 15 through 17, and 20 high-
light improved classification techniques leading to advances in wetland mapping and monitoring.
Continued rapid development of classification routines is expected.
FASTER, LOWER-COST DATA DISSEMINATION

The Internet has opened the door for wide application of remotely sensed data and wetland geo-
spatial data that are now available with the click of a computer mouse. In just the last decade, the
distribution of raw data and processed geospatial information has undergone a paradigm shift with
the release of Google Earth (Google 2014) and NASA World Wind (U.S. National Aeronautics
and Space Administration, 2014). Other data portals include the international Group on Earth
Observations’ Global Earth Observation Systems of Systems (GEOSS) Portal, the U.S. Geological
Survey’s Global Visualization Viewer (GloVis) and EarthExplorer, and NASA’s Distributed Active
Archive Centers. Web-based mapping tools have virtually replaced the need for published maps.
In the United States, NWI, NWI+, C-CAP, and USDA soil survey data can all be easily accessed
online. These tools provide users with the ability to print custom “maps” for their specific area
of interest. Further developments in the Internet’s capability as a means to archive and access
remotely sensed and geospatial data will occur.
The availability of information online regarding remote sensing fundamentals and wetland map-
ping specifics is also quite helpful. Websites that can assist with the development of remote sensing
applications and techniques are too numerous to mention. However, the increasing availability of
online training courses, such as the U.S. Fish and Wildlife Service’s Wetland Mapping Training
website (U.S. Fish and Wildlife Service 2014) and online instructional presentations, often based on
college-level courses, assists greatly in the dissemination of techniques. Groups like the Association
of State Wetland Managers’ Wetland Mapping Consortium (WMC), which is based in the United
States but open to members from other countries, greatly enhance communication among wetland
map makers. The WMC facilitates the transfer of techniques and generally supports the wetland
mapping community through online resources and regular web-based presentations on a wide vari-
ety of mapping topics (U.S. Association of State Wetland Managers 2014).
OTHER ADVANCES
Other significant advances in wetland mapping and monitoring include the use of unmanned aerial
systems (UASs), simultaneous improvement in temporal and spatial and/or spectral resolution, and
hyperspectral imagery. Although these advances are not currently used for operational wetland
mapping and monitoring, they represent significant resources for wetland scientists and managers
and their use will likely increase in the future.
UAS
While UASs have been employed routinely by the military, civilian use of the technology has been
hampered by tough restrictions on the use of drones by aviation administrations. This situation,
however, is rapidly changing. Beyond its low cost, the use of an UAS as a platform for remote sens-
ing data collection has further advantages over the use of satellite or conventional aircraft. First,
a drone can be deployed quickly and repeatedly to gather data. Second, since they typically fly
at low altitude, a UAS is well poised to gather very-high-resolution remotely sensed data. Third,
the device can be employed to reach inaccessible areas to collect needed data when conditions
are ideal for wetland detection and classification. The ability to collect very-fine-resolution data
at key times makes UASs well positioned for supplying calibration/validation data in support of
broader-scale wetland mapping and monitoring efforts and for monitoring natural disasters (see
Chapter 10).
Advances in Temporal, Spatial, and/or Spectral Resolution

The simultaneous advancement of temporal and spatial and/or spectral resolution is a laudable
achievement. In the past the spatial or temporal resolution of many sensors improved indepen-
dently. It is only recently that both have become possible through advanced technologies, such
as the deployment of multiple microsatellites, multiple view angle technology, or simply the
deployment of more fine-resolution sensors. Currently, multispectral satellite data are available
at a spatial resolution similar to aerial photography (e.g., GeoEye and WorldView), thus allowing
the detection of relatively small features and more timely standardized digital collections that
support the automated, repeatable detection of wetland signatures. These features are especially
important for disaster monitoring. Microsatellites were used during the aftermath of Hurricane
Katrina, as well as to track the oil slick emanating from the seabed of the Gulf of Mexico after
British Petroleum’s Deepwater Horizon oil rig exploded in 2010. In the past there was also a more
pronounced trade-off between spectral and spatial resolution, with finer spatial resolution sensors
having fewer bands (e.g., IKONOS and Quickbird) and sensors with more bands having coarser
spatial resolutions (e.g., moderate-resolution imaging spectroradiometer [MODIS]). However, this
trade-off has lessened in recent years with the development of sensors like WorldView-2 that
provides nine data bands with a spatial resolution between 0.5 and 1.8 m. More information on
trade-offs when mapping wetlands using datasets with different spatial, temporal, and spectral
resolutions can be found in Chapter 5.
Hyperspectral Imagery
Airborne hyperspectral sensors, like the airborne visible/infrared imaging spectrometer (AVIRIS;
224 0.01 μm bands from 0.4 to 2.45 μm), have been available for decades, and satellite-based
hyperspectral data are also available. The EO-1 satellite-borne Hyperion system provided imag-
ery with 220 spectral bands at a spatial resolution of 30 m. Designed for only a limited life span,
Hyperion was primarily used for technology demonstration and was somewhat limited for work
in wetland environments by its rather large 30 m2 pixel size. Results to date have shown that dis-
crimination between multiple wetland species is possible (Blasco et al. 2005; Vaiphasa et al. 2005;
Heumann 2011). Therefore, hyperspectral data may be capable of global scale species discrimina-
tion in the future.
Substantial challenges to using hyperspectral data have restricted widespread application. These
challenges include those related to large data volume and complexity of image processing resulting
in long analysis times and requiring the use of specialized software. For these reasons and others,
hyperspectral data are not used operationally for wetland mapping. They are, however, well suited
for solving some particularly difficult problems, such as the discrimination of vegetation species
(e.g., types of mangroves), detection of biophysical properties (e.g., water content), and identification
of chemical constituents (e.g., nitrogen) in plant tissue and water bodies. The potential of hyper-
spectral data for coastal and submerged wetland mapping may be particularly strong due to the
unique challenges posed by this environment, which include the rapid attenuation of electromag-
netic energy by water and suspended materials (see later). The question of how differing levels of
tidal inundation affect the reflectance characteristics of emergent marsh vegetation still needs to be
better documented. Chapter 5 presents more detailed discussion of the applications and restrictions
of hyperspectral data, while Chapters 11 and 12 offer a review of the use of hyperspectral imagery
for mapping mangroves and some other coastal wetlands.
FUTURE CHALLENGES FOR WETLAND MAPPING AND MONITORING

Despite the many advances in the field of wetland mapping and monitoring, considerable challenges
remain. While not exhaustive, the following discussion is intended to highlight those that, if solved,
have the greatest potential for improving wetland mapping, monitoring, and classification in the
next decade, both in terms of technological potential and resource availability.
Predicting Wetland Function

Wetland functional classification systems have existed for decades, including the hydrogeomor-
phic (HGM) classification (Brinson 1993) and Tiner’s LLWW descriptors for landscape position,
landform, water flow path, and water body type (Tiner 1997, 2011) (Chapter 2). Their applica-
tion relies heavily on manual interpretation of aerial imagery or on-the-ground observations.
Automating this type of classification effort would greatly expand their application for predict-
ing wetland condition and function. However, doing so may influence the detail of characteriza-
tion that can be achieved. For this reason, a two-step classification (automated then manual) has
been recently implemented for LLWW classification. Continued search for ways to streamline the
process would be desirable to maximize its application. One of the stumbling blocks in achiev-
ing more automation is the lack of a comprehensive stream layer for the nation. Although the
National Hydrography Dataset (NHD) provides useful information for this type of analysis, NHD
data do not always match up well spatially with the NWI data. This significantly limits its use
for assigning headwater designations and water flow paths and linking wetlands to watercourses
depicted as linear features (i.e., narrow streams).
Most existing wetland functional analyses represent static conditions—an assessment in the
field that day or on the date of acquisition of the imagery analyzed. Remote sensing techniques
can be used to provide information on dynamic functional drivers and their effects on wetlands.
This is especially important for wetlands where drastic changes in hydrology causing significant
changes in vegetation are the norm (e.g., Great Lakes coastal wetlands or emergent wetlands in
arid and semiarid regions) and where human-induced or natural processes affect wetland charac-
teristics and accompanying functions. In other words, remote sensing provides an effective moni-
toring tool for wetlands as described in many chapters of this book. Analyses of time series data
will also aid researchers studying the relationship between dynamic wetland properties or events
(e.g., biomass, fire, saturation, and inundation) and wetland functions. Maps could be based on
direct measurement of key functional drivers, such as hydroperiod, or on potential indicators of
drivers. For example, depressions have a higher potential for supporting longer hydroperiods than
convex landforms. In this example, SAR could support the direct measurement of hydroperiod,
whereas a LiDAR-based DEM could be used to indicate potential hydroperiod. Relationships
are bound to vary between landscapes that have different edaphic, topographic, and climate
conditions, and these landscape types should be considered before results are applied to other
locations.
As is true for all wetland inventories more field validation is desirable, with increased attention
given to wetland function instead of simply validating the wetland class. It is possible that the
use of UAS could expand the number of field checks, especially for sites that are inaccessible on
foot or by boat. Such data will also aid researchers in evaluating relationships between wetland
characteristics, including spatial relationships between and among wetlands and other landscape
features, and ecological and environmental functions. These fundamental relationships are the
basis of most functional classifications and would benefit greatly from improved calibration and
validation.
The resource management impact of information on dynamic functional drivers can be
enhanced through the inclusion of these data in a variety of different models, including those
that focus on water quality, habitat provision, carbon sequestration, and flood mitigation. Remote
sensing–derived metrics and relationships can be used to support modeling in a variety of ways
including (1) the creation of conceptual models, (2) the development of algorithms in support of
process-based modeling, (3) the parameterization of models, and (4) the calibration/validation of
models. The ability to provide robust estimates of wetland function through modeling or empirical
relationships should help support predictions of ecosystem services and the development of envi-
ronmental markets, as well as providing a decision-support tool for resource planners, managers,
and policy analysts.
Detection of Wetland Water Regimes

Although numerous factors interact to influence wetlands, hydrology is the single most important
force in the formation and functioning of a wetland as it influences vegetation patterns, soil proper-
ties, and wildlife and human use. Wetlands are “wet lands” because they are in landscape positions
or other areas that remain wet for long periods (Chapter 1). Prolonged wetness affects soil develop-
ment, plant colonization, animal usage, and the environmental services delivered to humankind.
While many wetlands possess relatively unique biota and soil properties, other types are difficult
to separate from neighboring uplands (Chapter 3). Information derived from remote sensors can be
used to properly characterize wetland hydrology as well as to aid in interpreting wetland functions
as described in the preceding section.
Due to the interference of the forest canopy with viewing the soil surface and the spectral simi-
larity between upland and wetland trees, it is often difficult to use optical data, such as aerial
photographs and multispectral images, to map certain forested wetlands, especially those that are
temporarily flooded or dominated by evergreen species. In drier wetlands, saturation or the pres-
ence of high organic carbon soils, an indicator of wetness, is often hidden by leaf litter or last sea-
son’s standing dry biomass. Since species composition alone cannot be used to distinguish some
forested wetlands from adjacent upland forests and soils are hidden beneath evergreen trees and
fully leafed-out deciduous trees, wetland map makers often rely on observations of hydrology (i.e.,
inundation or high levels of soil moisture) during the wet season or indicators of potential hydrology
(e.g., landscape position) to map forested wetlands. However, these indicators are often not discern-
ible using more commonly available optical images that are acquired during the growing season.
Research scientists have developed new methods to detect forested wetlands using other types of
remotely sensed datasets (Chapter 5 among others). While our ability to map forested wetlands has
improved in large part due to advancements including the use of SAR, LiDAR, and multitemporal
data and enhanced classification techniques (e.g., machine learning algorithms), more research is
needed to better understand the degree to which these techniques can be applied to other landscapes
and incorporated into operational frameworks. New tools and techniques are needed to improve
mapping of seasonally saturated and temporarily flooded wetlands, forested types that dominate the
world’s temperate and tropical zones.
Many non-forested evergreen wetlands are equally and perhaps more challenging, particularly
those in the tropics where cloud cover during the wet season precludes acquisition of optical imag-
ery. Higher leaf area index (LAI) levels in these forests can reduce transmission of energy from
active sensors, such as SAR and LiDAR. Greater availability of longer wavelength SAR sensors
may help to better map this wetland type. The collection of multiple point return LiDAR data with
higher point densities should help increase bare earth point density in evergreen areas, but this
approach may not be successful in evergreen forests with very high LAI. While full waveform
LiDAR is currently more difficult to obtain, its availability is increasing and therefore it may also
assist in improving evergreen wetland mapping capabilities. Additional research is needed to better
understand the advantages and limitations of these approaches.
Plant Species and Properties

Despite some problematic situations noted earlier, differences in vegetation type can be used to
distinguish many wetlands from uplands. Readily available optical data are able to detect these
differences. For example, aerial photography can be used to distinguish marsh from adjacent for-
est, albeit this distinction might be easier to recognize at some times of the year and more accurate
boundaries could be achieved by using finer spatial resolution imagery. However, it is often difficult
to distinguish more subtle differences between plant species or other vegetative qualities that may
be important in determining wetland function. These differences include (1) the identification of
invasive species, including Phragmites australis (common reed), Lythrum salicaria (purple looset-
rife), Phalaris arundinacea (reed canary grass), and Typha angustifolia (narrow-leaved cattail), (2)
the characterization of LAI, and (3) separating wetland herbs from those on adjacent upland fields.
Although multiple sensor types have been used to improve our ability to detect invasive species
(e.g., Chapters 15 and 23), hyperspectral data are most commonly used to map invasives, with SAR
and LiDAR providing key information on differences in plant structure. However, as mentioned ear-
lier, hyperspectral data can be challenging to use and more research is needed to test and enhance
their usability for this application. Although LAI is commonly mapped using optical data, optical
data saturate at high levels of LAI. Current efforts are focused on the use of LiDAR and SAR to
improve the accuracy of LAI estimations and related parameters, such as biomass. Where natural
cover is dominated by grasses and other herbaceous species, it is often difficult to differentiate wet
meadows from upland fields. Two technologies stand out as having significant potential to meet this
challenge. Again hyperspectral data may be helpful, while SAR data could be used to detect varying
hydroperiods between these wetland and upland areas.
Submerged Wetlands and Aquatic Habitats

Because water rapidly attenuates electromagnetic energy and suspended materials (e.g., sediment
and dissolved carbon) further affect the scattering and absorption of light, submerged wetlands and
aquatic habitats pose particular challenges for accurate detection. The spectral signal of submerged
aquatic vegetation (SAV) must be isolated not only from the water column but also from the under-
lying substrate and the atmosphere.
Recent enhancements in satellite technology and water column optical models have advanced
the use of remote sensing for detecting SAV and other submerged wetlands. For example, these
models have been used to correct water and bottom effects by including bathymetric information
as one of the variables in order to relate change in SAV to water depth. In addition, several recently
launched satellites have been equipped with so-called “coastal” bands, which use short-wavelength
visible blue light with wavelengths between 400 and 450 nm. By virtue of their high frequency, this
portion of the spectrum penetrates water well (~50 m) when turbidity is low and is therefore quite
effective for detecting submerged targets.
Coral reef ecosystems usually exist in clear water and can be classified to show different forms
of coral reef, dead coral, coral rubble, algal cover, sand, lagoons, different densities of sea grasses,
and more. In addition to being spatially complex, SAV often grows in somewhat turbid waters and
thus is more difficult to map. Aerial hyperspectral scanners and high-resolution multispectral satel-
lite imagers, such as IKONOS and QuickBird, have been used to map SAV with accuracies of about
75% for general classes, including high-density sea grass, low-density sea grass, and unvegetated
bottom (Dierssen et al. 2003; Wolter et al. 2005; Mishra et al. 2006; Akins et al. 2010) (Chapter 11).
Major advances in airborne hyperspectral sensors offer the spatial and spectral capabilities to dis-
cern subtle classes of SAV beds that can be used as indicators of health. SAV has been mapped with
high accuracies using airborne hyperspectral imagers and regression models, binary decision trees
incorporating spectral mixture analysis, spectral angle mapping, and band indexes (Underwood
et al. 2006; Hestir et al. 2008; Peneva et al. 2008). Also, environmental variables describing the
boundary conditions around reefs and SAV beds can be related to processes occurring on the reefs
and beds themselves. The majority of airborne hyperspectral radiometers are flexible in that they
can be “tuned” to the demands of a specific project, such as mapping SAV or coral reefs. Airborne
LiDARs have also been used with multispectral or hyperspectral imagers to map coral reefs and
SAV. Protocols have been developed for hyperspectral remote sensing of SAV in shallow waters
(Bostater and Santoleri 2004).
Challenges to the mapping and monitoring of submerged habitats, however, remain. The main
challenge for remote sensing of submerged aquatic plants is to isolate the plant signal from the inter-
ference of sunglint, the water column, the bottom, and the atmosphere. In addition to atmospheric
effects and bottom reflectance, optically active materials, such as plankton, suspended sediment,
and dissolved organics, affect the scattering and absorption of radiation. Shorter “coastal band”
wavelengths are helpful, but they are more susceptible to atmospheric absorption and scattering
and therefore pose a challenge for atmospheric correction. Although hyperspectral data can pro-
vide enhanced detection of SAV, the volume and complexity of processing this type of data can be
daunting.
INTERNATIONAL WETLAND DEFINITION AND CLASSIFICATION STANDARD

The influence of wetlands on global cycles of water, carbon, and methane is very significant, but
the lack of a globally accepted wetland definition and classification standard plus variations in the
quality and spatial scale of existing data complicates the use of existing wetland data in global
models and the development of informed management strategies. While wetlands in some regions
have been the subject of major mapping projects, existing wetland inventories are incomplete and
inconsistent and they employ different wetland definitions, classification systems, and inventory
methods (Tiner 2009). Even the United States, which has dedicated millions of dollars to produc-
ing what is the most detailed national wetland inventory in the world, has an inventory that is
(1) incomplete due to the challenges of detecting wetlands through remote sensing (Chapter 3),
(2) somewhat inconsistent due to manual interpretation and changes in inventory procedures and
wetland recognition practices over time, and (3) outdated for much of the country (i.e., most areas
have 1980s coverage). Consequently, this inventory has limited utility for environmental modeling
across regions. The alternative is to use satellite imagery for the wetlands inventory. Wetland data
generated from satellite-based remote sensors will be, by necessity, more generalized than data
produced by manual interpretation of aerial imagery, but the use of satellite data offers a more
consistent approach that can be frequently updated to monitor the changing status of wetlands. This
type of data is crucial for modeling climate change, carbon sequestration, and methane emission.
A comparison of the probability-based wetland trends assessment approach currently used by the
FWS with a full-scale satellite-based assessment using the latest technology would provide interest-
ing results for wetland managers to evaluate. Mapping of global wetlands can only be accomplished
by analysis of satellite-based data with the end product being a generalized depiction of the world’s
wetland resource. The same holds true for large geographic regions. To date in large part due to
limitations of the type of sensors used, satellite-based inventories have emphasized “inundated
wetlands” (wetlands with standing water) with significant underestimates of “saturated wetlands”
(Krankina et al. 2008; Papa et al. 2010). In addition, the handling of “artificial wetlands” (e.g., rice
paddies) has been inconsistent.
Underlying any technical issues is the fact that agreement must be reached on what is a wetland.
Many definitions have been used for mapping and conserving wetlands (Chapter 1). International
scientists should try to reach consensus on what is a wetland as that is critical for any inventory, espe-
cially for producing a global satellite-based wetland inventory. The following definition attempts
to combine concepts found in numerous classification systems reviewed in preparing this book and
represents a starting point for discussion:
Wetlands are lands covered by shallow water or saturated at or near the surface for significant periods
(permanently, seasonally, or intermittently) sufficient enough to support plants and animals dependent
on or tolerant of prolonged inundation or saturation and to create hydric soils or substrates. Wetlands
include: (1) habitats dominated by hydrophytic vegetation and hydric soils (e.g., marshes, swamps, bogs,
and fens), (2) habitats characterized by aquatic vegetation and hydric substrates (i.e., aquatic beds and
seaweed-covered rocky shores), (3) nonvegetated habitats subject to frequent or prolonged inundation or
saturation (e.g., mudflats, rocky shores, beaches, and salinas), (4) coral reefs, and (5) the shallow water
zone (<2.5 m at mean low water) of water bodies.
This definition excludes deeper water bodies that are included in the Ramsar definition but includes
coral reefs and SAV regardless of water depth. While there should be widespread agreement on the
central concept of a wetland, key questions remain on the upper and lower limits. For the upper
boundary, what is the minimum frequency and duration of wetness necessary for an area to be
defined as wetland? This may need to be addressed regionally to account for “episodic wetlands” in
arid regions. For the lower limit, the question is: For permanent water bodies, what is the maximum
depth of a wetland?
The next step would address wetland classification. While the classification system used to
describe wetlands will continue to be determined by a given project’s objective, a hierarchical
approach offers the most flexibility and may work best for application with different sensors. The
approach used by Cowardin et al. (1979) has proved a useful model and one that can be read-
ily expanded to provide more detail when desirable (e.g., Tiner 2011). This approach has served
as a standard for mapping wetlands in the United States for nearly 40 years (FGDC Wetlands
Subcommittee 2009) and has been used in various countries. Since the Ramsar system has wide-
spread international usage, perhaps integrating Ramsar terminology into this hierarchy could pro-
vide a system that could gain international support. Regardless of the classification that is selected,
it must be well suited or adapted to work well with an automated mapping process that is dependent
on a globally available data source (see Chapters 20 and 24).
CONCLUSION
The future of remote sensing applications for wetland mapping is bright from a technological stand-
point. Perhaps the greatest challenge will be securing sufficient funds to use the techniques operation-
ally so that the results can provide a decision-support tool to achieve improved wetland conservation
and management around the world. The incorporation of these developments into operational pro-
grams will require not only acquisition of new datasets and possibly hardware and software, and
trained analysts, but also government leadership to fully support and finance these inventories. The
additional expenditures could be offset by savings (in terms of staff time due to increased automa-
tion), but more importantly, they should be viewed as investments that will provide better informa-
tion for making decisions about the health and well-being of wetlands now and in the future.
REFERENCES
Akins, E.R., Y. Wang, and Y. Zhou. 2010. EO-1 advanced land imager data in submerged aquatic vegetation
mapping. In: Wang, J. (ed.) Remote Sensing of Coastal Environments. Boca Raton, FL: CRC Press.
Blasco, F., M. Aizpuru, and D. Ndongo. 2005. Mangroves remote sensing. In: Schwartz, M.L. (ed.) Encyclopedia
of Coastal Science. Dordrecht, the Netherlands: Springer.
Bostater, C.R. Jr. and R. Santoleri. 2004. Hyperspectral remote sensing protocol development for submerged
aquatic vegetation in shallow waters. In: Proceedings of the SPIE 5233, Barcelona, September 9,
pp. 199–215. Bellingham, WA: Society of Photo-Optical Instrumentation Engineers.
Breiman, L. 2001. Random Forests. Machine Learning 45:5–32.
Brinson, M.M. 1993. A Hydrogeomorphic Classification for Wetlands. U.S. Army Engineers Waterways
Experiment Station, Vicksburg, MI. Wetland Research Program Technical Report WRP-DE-4. http://
www.dtic.mil/dtic/tr/fulltext/u2/a270053.pdf. Accessed December 4, 2014.
Cowardin, L., V. Carter, F. Golet, and E. LaRoe. 1979. Classification of Wetlands and Deepwater Habitats of
the United States. Washington, DC: U.S. Department of the Interior, Fish and Wildlife Service, Office of
Biological Services. FWS/OBS-79/31.
Dierssen, H.M., R.C. Zimmermann, R.A. Leathers, V. Downes, and C.O. Davis. 2003. Ocean color remote sens-
ing of seagrass and bathymetry in the Bahamas banks by high resolution airborne imagery. Limnology
and Oceanography 48:444–455.
FGDC Wetlands Subcommittee. 2009. Wetlands Mapping Standard. Federal Geographic Data Committee
(FGDC). FGDC-STD-015-2009. http://www.fgdc.gov/standards/projects/FGDC-standards-projects/
wetlands-mapping/2009-08%20FGDC%20Wetlands%20Mapping%20Standard_final.pdf.
Google. 2014. Google earth. earth.google.com. Accessed April 12, 2014.
Hestir, E.L., S. Khanna, M.E. Andrew, M.J. Santos, J.H. Viers, J.A. Greenberg, S.S. Rajapakse, and S. Ustin.
2008. Identification of invasive vegetation using hyperspectral remote sensing in the California Delta
ecosystem. Remote Sensing of the Environment 112:4034–4047.
Krankina, O.N., D. Pflugmacher, M. Friedl, W.B. Cohen, P. Nelson, and A. Baccini. 2008. Meeting the chal-
lenge of mapping peatlands with remotely sensed data. Biogeosciences 5:1809–1820.
Mishra, D., S. Narumalani, D. Rundquist, and M. Lawson. 2006. Benthic habitat mapping in tropical marine
environments using QuickBird multispectral data. Photogrammetric Engineering and Remote Sensing
72:1037–1048.
Papa, F., C. Prigent, F. Aires, C. Jimenez, W.B. Rossow, and E. Matthews. 2010. Interannual variability of sur-
face water extent at the global scale, 1993–2004. Journal of Geophysical Research 115:1–17.
Peneva, E.I., J.A. Griffith, and G.A. Carter. 2008. Seagrass mapping in the Northern Gulf of Mexico using
airborne hyperspectral imagery: A comparison of classification methods. Journal of Coastal Research
24:850–856.
Snyder, D. and M. Lang. 2012. Significance of the 3D Elevation Program to wetlands mapping. National
Wetlands Newsletter 34(5):11–15.
Tiner, R.W. 1997. Keys to Landscape Position and Landform Descriptors for U.S. Wetlands. Hadley, MA:
Operational Draft. U.S. Fish and Wildlife Service, National Wetlands Inventory Program, Northeast
Region.
Tiner, R.W. 2009. Global distribution of wetlands. In: Likens, G.E. (ed.) Encyclopedia of Inland Waters. Vol. 3.
Elsevier Publishers, Ltd., Oxford, U.K.
Tiner, R.W. 2011. Dichotomous Keys and Mapping Codes for Wetland Landscape Position, Landform, Water
Flow Path, and Waterbody Type Descriptors. Hadley, MA: Version 2.0. U.S. Fish and Wildlife Service,
National Wetlands Inventory Program, Northeast Region.
Underwood, E.C., M.J. Mulitsch, J.A. Greenberg, M.L. Whiting, S.L. Ustin, and S.C. Kefauver. 2006.
Mapping invasive aquatic vegetation in the Sacramento-San Joaquin Delta using hyperspectral imagery.
Environmental Monitoring and Assessment 121:47–64.
U.S. Association of State Wetland Managers. Wetland mapping consortium (WMC). Accessed April 12, 2014.
U.S. Fish and Wildlife Service. Wetland mapping training. Accessed April 12, 2014.
U.S. National Aeronautics and Space Administration. NASA World Wind. worldwind.arc.nasa.gov. Accessed
April 12, 2014.
Vaiphasa, C.S. Ongsomwang, T. Vaiphasa, and A.K. Skidmore. 2005. Tropical mangrove species discrimina-
tion using hyperspectral data: A laboratory study. Estuarine, Coastal and Shelf Science 65:371–379.
Wolter, P.T., C.A. Johnston, and G.J. Niemi. 2005. Mapping submerged aquatic vegetation in the US Great
Lakes using Quickbird satellite data. International Journal of Remote Sensing 26:5255–5274.
WATER SCIENCE, TECHNOLOGY AND ENGINEERING
Remote Sensing of Wetlands

Effectively Manage Wetland Resources Using the Best Available Remote Sensing Techniques
Utilizing top scientists in the wetland classification and mapping field, Remote Sensing of Wetlands:
Applications and Advances covers the rapidly changing landscape of wetlands and describes the latest
advances in remote sensing that have taken place over the past 30 years for use in mapping wetlands.
Factoring in the impact of climate change, as well as a growing demand on wetlands for agriculture,
aquaculture, forestry, and development, this text considers the challenges that wetlands pose for remote
sensing and provides a thorough introduction on the use of remotely sensed data for wetland detection.
Taking advantage of the experiences of more than 50 contributing authors, the book describes a variety
of techniques for mapping and classifying wetlands in a multitude of environments ranging from
tropical to arctic wetlands including coral reefs and submerged aquatic vegetation. They discuss the
advantages and disadvantages of using different remote sensing techniques for wetland detection under
varied conditions and circumstances. They also analyze commonly available data, reveal cost-effective
methods, and offer useful insights into future trends.
Comprised of 25 chapters, this text:
• Presents methods readily applicable to real-world challenges
• Contains advanced, new techniques communicated by top scientists in the field
• Covers a diverse set of landscapes and technologies
• Reviews many of the datasets and techniques that are responsible for advances in this discipline
and their application for wetland mapping
• Addresses the need to effectively manage this environmental resource
Remote Sensing of Wetlands: Applications and Advances uses a variety of contributors, touching on
pertinent topics, to help you gain a greater understanding of the latest technologies, strengths, and
limitations surrounding this emerging field.
K23165
6000 Broken Sound Parkway, NW

Suite 300, Boca Raton, FL 33487
711 Third Avenue
New York, NY 10017
an informa business
2 Park Square, Milton Park
w w w. c r c p r e s s . c o m Abingdon, Oxon OX14 4RN, UK w w w. c r c p r e s s . c o m

Remote Sensing of Wetlands - Applications and Advances-Crc Press (2014)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Remote Sensing of Wetlands - Applications and Advances-Crc Press (2014)

Uploaded by

Copyright:

Available Formats

Edited by

Boca Raton London New York

CRC Press is an imprint of the

No claim to original U.S. Government works

International Standard Book Number-13: 978-1-4822-3738-2 (eBook - PDF)

Chapter 1 Wetlands: An Overview................................................................................................ 3

Chapter 2 Classification of Wetland Types for Mapping and Large-Scale Inventories............... 19

Chapter 3 Introduction to Wetland Mapping and Its Challenges................................................ 43

Chapter 4 Early Applications of Remote Sensing for Mapping Wetlands................................... 67

Chapter 7 Wetland InSAR: A Review of the Technique and Applications............................... 137

Chapter 9 Theory and Applications of Object-Based Image Analysis and Emerging

Chapter 10 Unmanned Aerial Systems and Structure from Motion Revolutionize

Chapter 12 Remote Sensing of Mangroves.................................................................................. 243

Chapter 14 Using Moderate-Resolution Satellite Sensors for Monitoring the Biophysical

Chapter 15 Great Lakes Coastal Wetland Mapping.................................................................... 315

Chapter 18 Fusion of Multispectral Imagery and LiDAR Digital Terrain Derivatives

Chapter 19 Airborne LiDAR-Based Wetland and Permafrost-Feature Mapping on an

Chapter 20 Hybrid Mapping of Pantropical Wetlands from Optical Satellite Images,

Chapter 21 Capturing the Dynamics of Amazonian Wetlands Using Synthetic Aperture

Chapter 23 Mapping Invasive Wetland Plants............................................................................. 491

Section IV Promising Developments and Future Challenges

Chapter 25 Promising Developments and Future Challenges for Remote

Megan W. Lang is a research associate professor at the Department of Geographical Sciences,

Michael J. Battaglia Antonio Difebo

Laura L. Bourgeau-Chavez Jefferson Ferreira-Ferreira

Sang-Hoon Hong Zachary M. Laubach

Carol A. Johnston Kyle McDonald

Megan W. Lang David M. Mushet

Zhenguo Niu Sam Purkis

Nancy O’Hare Lian P. Rampi

Alessandro Pasqua Chris Roelfsema

Jonathan Price Kutalmis Saylam

Catherine Prigent Thiago Sanna Freire Silva

Annia Susin Streher Shimon Wdowinski

Thomas A. Tremblay Jane Whitcomb

DEFINITIONS AND CONCEPT

FACTORS AFFECTING WETLAND FORMATION

–30 Wet meadow

U.S. Dept. of Agriculture

Inland water bodies km

No wetlands (or too small to display)

Country boundaries are not authoritative. Washington D.C. 1996

GENERAL WETLAND TYPES

GLOBAL DISTRIBUTION OF WETLANDS

THREATS TO THE WORLD’S WETLANDS

TABLE 1.6 (Continued)

CHARACTERISTICS USED FOR WETLAND CLASSIFICATION

TYPES OF WETLAND CLASSIFICATION SYSTEMS

TABLE 2.1 (Continued )

MAJOR WETLAND CLASSIFICATION SYSTEMS

Ramsar International Wetland Classification System

TABLE 2.2 (Continued )

Cowardin et al. Classification System

Brinson’s Hydrogeomorphic Classification System

OTHER PREEXISTING CLASSIFICATION SYSTEMS

RECENT DEVELOPMENTS IN WETLAND CLASSIFICATION

Modification of the Cowardin et al. System

Adapting Cowardin et al. for Mediterranean Wetlands

Adapting the HGM Concept for Wetland Mapping and

Sodic and/or saline water

Key C. Water flow path

TABLE 2.12 (Continued )

(lotic stream)—at a scale of 1:24,000.