A CHANGE-DETECTION ANALYSIS: USING REMOTELY SENSED DATA TO

ASSESS THE PROGRESSION OF DEVELOPMENT IN ESSEX COUNTY,

MASSACHUSETTS FROM 1990 TO 2001

Peter Sean Tardie

Russell G. Congalton
Department of Natural Resources
University of New Hampshire
Durham, NH 03824
ptardie@cisunix.unh.edu

ABSTRACT

Over the years, population and development of land in Essex County, Massachusetts have increased. As
surrounding urban centers grow, proliferation into neighboring communities has become apparent. New
residents create need for the development of land for housing, schools, ball-fields, parks, and place significant
strains on existing land cover and available natural resources. In order to monitor development in the county, a
change-detection analysis was performed to determine the extent of land cover change over time. Change-
detection is a technique used in remote sensing to determine the changes in a particular object of study between
two or more time periods. Change-detection is an important process for monitoring and managing natural
resources and urban development because it provides quantitative analysis of the spatial distribution in the area
of interest. In this study, three change-detection techniques were evaluated: (1) Multi-date visual composite, (2)
Image differencing, and (3) Post-classification. All were performed using Landsat Thematic Mapper (TM) data
to detect land cover changes that occured in Essex County, Massachusetts from 1990 to 2001. This study
provided an application of Landsat Thematic Mapper (TM) to detect land-use change and the methodology for
comparing change-detection techniques using standard accuracy assessment procedures.

INTRODUCTION

Since the 1970’s, Essex County has encountered major residential development and urban sprawl. As the
population and land development increased, a reduction in undeveloped land has become obvious. Increased
development pressures have kindled numerous concerns among existing residents: the fragmentation of
agricultural land, the disruption of wetlands, the infringement upon protected open-space, pollution, and the
possible loss of aesthetic charm and character of the existing communities. Without continuing community-
wide environmental education, misuse of land may be accepted as “just” the results of progress.
The Massachusetts economic boom of the late 1970’s and early 1980’s, had originally allowed for growth
to diffuse north of Boston, as many “high tech” research firms and defense contracting agencies were developed
(Harrison and Kluver, 1989). In 1988, more than 440,000 new jobs were created, and the value of residential
construction grew four times faster than the nation as a whole (Harrison and Kluver, 1989). King and Harris
1989, indicated that increased residential housing construction, traffic flow disturbances, and severe demands
made on infrastructure by a steady influx of new residents, can have numerous negative effects on ecologically
sensitive rural landscapes, and could lead to the systematic deterioration of rural character and depletion of
agricultural and forest resources.
In Essex County, farmland conversion for residential development and commercial expansion is rapid.
Agricultural farms have decreased 74% in six years, from 365 establishments in 1993, to 84 in 1999 (U.S.
Census, 2001).. Although the Commonwealth of Massachusetts has taken steps to preserve agricultural land
with the enactment of Chapter 61A of Massachusetts General Laws, referred to as the Preferential Assessment
of Agricultural and Horticultural Lands, high real estate values continue to downplay its intent (Levia, 1998).
Jacobs 1999, indicated that people desire land use to be rational, efficient, and equitable, and would like to
maintain rural character at all costs; they do not want land to be degraded. But, this is not always the case. In
order to control growth, deter large sprawling developments, and to keep a watchful eye on available natural
resources, many local officials have articulated and implemented planning strategies, adopted land use
zoning by-laws, subdivision regulations, and site plan reviews. These regulations are designed to help ease
development pressures and are not foolproof.
Changes to the environment can also reflect how the land has been managed, and using established change-
detection methodologies to monitor these changes can serve as an evaluation of these management practices
(Brothers and Fish, 1978). Since the introduction of satellite-based technology in the early 1970’s, the Landsat
program has made this process possible. With repetitive satellite coverage, the rapid evolution of computer
technology, and the integration of satellite and spatial data with geographic information systems (GIS),
development of environmental monitoring applications such as change-detection have become ubiquitous
(Singh, 1989; Jensen, 1996; Macleod and Congalton, 1998; Rynzar and Wagner, 2001). To detect land cover
change, a comparison of two or more satellite images acquired at different times, can be used to evaluate the
temporal or spectral reflectance differences that have occurred between them (Masry, Crawley, and Hilborn,
1975; Yuan and Elvidge, 1998;).
As change-detection studies advance, the development of procedures to determine accuracy of the different
techniques become increasingly important (Macleod and Congalton, 1998). The primary objective of this
research was to evaluate various change-detection techniques and determine which provided the highest
accuracy for Essex County, Massachusetts. The specific objectives were to (1) determine appropriate methods
to detect urban development, (2) perform various change-detection methods, and (3) assess the accuracy of the
change-detection methods using a change-detection error matrix.
Results of this research can be utilized as a temporal land-use change model for the county to quantify the
extent and nature of development change, foster learning about the surrounding environment, and aid planning
agencies in developing, sound, sustainable land-use practices.

METHODS

The methods section is divided into a description of the study area, satellite and reference data,
preprocessing, change-detection, and accuracy assessment.

Study Area
The study area (Figure 1) is Essex County, Massachusetts, which is located in the Merrimack River
watershed in the northeast corner of the state. The county borders the Atlantic Ocean to its east, New
Hampshire to its north, and Suffolk and Middlesex counties to its south and west. The county has an area of
501 square miles, which is approximately 1/15th of the size of the state, and contains thirty-four communities,
which are mostly rural in character. Essex County has five major interstate highways passing through it, routes:
1, 95, 93, 128, 495, and also contains three relatively large urban municipalities: Lawrence, Lynn, and Peabody.

Satellite and Reference Data

To insure the accurate detection of land cover change, and reduce effects of seasonal phenological
differences of vegetation (Jensen, 1996), two near-anniversary images Landsat TM images collected on 8
September 1990 and 29 August 2001, covering Essex County and the surrounding area (WRS 12/30) were used.
Reference data were obtained from MassGIS, and were used with a geographic information system (GIS).
These data consisted of 1-meter resolution 1:5000 scale black and white digital orthophoto images, which were
produced in 1995, and land-use (GIS) datalayers ranging from 1971 to 1999. These data would be later used for
comparison of Landsat TM image classifications. Local state, county, and township political boundaries (GIS)
datalayers were also acquired and were used for scene clipping and community landmark identification.

Preprocessing
In order to prepare two or more satellite images for an accurate change-detection comparison, it is
imperative to geometrically rectify the imagery (Townshend et al., 1992; Macleod and Congalton, 1998;
Kwarteng and Chavez, 1998). To lessen impact of misregistration on the change-detection results, geometric
registration was performed on a pixel-by-pixel basis. Erroneous land cover change results may result if any
misregistration greater than one pixel occurs (Lunetta and Elvidge, 1998). The accuracy of image registration is
usually conveyed in terms of root-mean-square (RMS) error. For Landsat TM imagery acceptable RMS error is
approximately 0.5 pixels (Townshend et al., 1992; Yuan and Elvidge, 1998; Lunetta and Elvidge, 1998). The
images from each date were subset using a single mask from a county boundary GIS layer. Once Essex County,
images were subset, the 1990 image was registered to the 2001 image, using 93 ground control points. The 1990
image was resampled using the nearest neighbor algorithm. Both images were reprojected into Massachusetts
State Plane Coordinates (NAD83).

Figure 1 – Location of Essex County, Massachusetts

Change-Detection
There are many change-detection techniques from visual comparison to detailed quantitative approaches
(Wickware and Howarth, 1981). In this project we evaluated three techniques: (1) Multi-date visual composite,
(2) Image differencing, and (3) Post-classification change-detection. In addition, a variety of data exploration
techniques were examined such as spectral pattern analysis, bi-spectral plots, derivative bands, and divergence
analysis, which were used as diagnostics to determine the best combination of spectral bands to use in the
change-detection process.

Multi-date Visual Composite Image Change-Detection. This technique provides a simple mechanism to
display changes between two dates of imagery quickly and efficiently. In our study, we chose to look at the
reduction in vegetation as an indicator of urban development. Therefore, band 4 (Landsat TM near infrared)
was selected. For the 1990 image, band 4 was displayed through the R (red) and B (blue) portions of the
monitor. Band 4 of the 2001 image was displayed through the G (green) portion of the monitor. The resulting
image is gray in areas where no change has occurred. Where the vegetation has decreased over time, the image
is magenta (blue and red) and where the vegetation has increased, it is green. This image can then show very
quickly the changes in vegetation and therefore urban development. This technique is not meant to be
quantitative but rather is used to begin to identify and explore the areas of change in Essex County.

Image Algebra (Differencing). Change-detection using geometrically corrected Landsat TM imagery can be
visually compared but, most change-detection studies require a more detailed quantitative approach to the
analysis (Wickware and Howarth, 1981). Image differencing is a technique that has been widely used for
change-detection (Singh, 1989), and provides lower change-detection errors when compared to other methods
(Jensen and Toll, 1982). In this process, the digital number (DN) value of one date for a given band is
subtracted from the DN value of the same pixel for the same band of another date (Jensen, 1981; Singh, 1989;
Jensen, 1996; Macleod and Congalton, 1998). In this study, image differencing was performed on the first four
raw bands (blue, green, red, and near infrared). Threshold values based on standard deviations from the mean
were utilized, in order to determine changed from unchanged pixels. Results of the image differencing
technique illustrate the pixels that have changed between the two time periods.

Post-Classification. In a variety of studies, the post-classification change-detection method was found to be the
most suitable for detecting land cover change (Weismiller et al., 1977; Wickware and Howarth, 1981). In the
post-classification technique, two images from different dates are independently classified (Jensen, 1981;
Jensen and Toll, 1982; Singh, 1989; Jensen, 1996; Yuan and Elvidge, 1998). Accurate classifications are
imperative to insure precise change-detection results (Foody, 2001). Preliminary classifications were performed
on the 1990 and 2001 images. The classifications of the 1990 and 2001 images with the highest overall
accuracy were used in the change-detection process. The classified images were combined to create a new
change image classification, which indicated the changes “from” and “to” that took place. However, image
classification and post-classification techniques are iterative and require further refinement in order to produce
more accurate change-detection results.

Accuracy Assessment
A preliminary accuracy assessment was performed on both the image differencing and post-classification
change detection techniques. For the post-classification approach, individual single date error matrices were
generated. In addition, a change-detection error matrix was created for the image differencing and the post-
classification techniques (Congalton, 1991; Macleod and Congalton, 1998). A change detection error matrix is
much more complicated than a single date error matrix as it records not only spatial differences, but temporal
differences as well. Therefore, sampling for assessing the accuracy of a change-detection, requires more time
and effort.

Preliminary Results

Changes in population density should be an indicator of development and vegetation change.

According to the U.S. Census Bureau in 2000, the population figures indicated a steady rise from 1990 to 2000.
In 1990, population in the county was estimated at 670,080 persons in 271,977 households, yielding a density of
2.46 persons per dwelling with approximately 1,337 people per square mile. In 2000, the population was
723,419 persons in 275,419 households, yielding a density of 2.62 persons per dwelling with 1,444 people per
square mile. On average, population increased 8.0% from 1990 to 2000.
Figure 2 represents the percent changes in population throughout the communities in Essex County
between 1990 and 2000 (U.S. Census, 2001). This figure and others like it generated from a variety GIS data
layers such as housing occupancy and tenure, tax assessor data, and zoning, should aid in exploring the change-
detection analysis.
Registration of the 1990 and 2001 images resulted in an overall root-mean-square (RMS) error of 11.8
meters, which was well within acceptable limits. Preliminary results from the multi-date visual change-
detection image (Figure 3) indicate that land cover change took place in Essex County between 1990 and 2001.
The initial interpretation of Figure 3, suggests that the magenta regions represent vegetation which has been
cleared for new development, and the green regions represent higher reflectance in the near infrared wavelength
due to the re-growth of vegetation. For example, the towns of North Andover, Middleton, and West Newbury,
located in Figure 3, exhibit large amounts of magenta and correspond with the population increase shown in
Figure 2. Conversely, Saugus, Peabody, and Lynnfield, exhibit large amounts of green and correspond with the
population decrease shown in Figure 2 as well.
 Figure 2– Population Change from 1990 - 2001

Figure 3– Multi-date Visual Composite Change-Detection Image

The preliminary result of the image differencing technique was the Highlight Change-Detection image (Figure
4). The results displayed suggest that the red pixel regions indicate vegetation which has increased or
reappeared, and the green pixel regions represent where vegetation has been cleared or decreased for new
development. Although this technique did not allow for “from-to” change classes to be distinguished, it was
effective in identifying the individual pixels that have changed in brightness value between 1990 and 2001.

Figure 4- Highlight Change Image Derived from Image Differencing

Discussion
This research compared three change-detection techniques for detecting urban development with Landsat
TM imagery. Performing the change-detection analysis on Essex County allowed for the monitoring of a
natural landscape over time. This change-detection study provides beneficial insight into the extent and nature
of change that has taken place in the county from 1990 to 2001, and lays the foundation for further research to
be conducted.
This research can also be used as a model for other regions encountering development and vegetation
change and illustrate the importance of satellite imagery.

.
 References

Brothers, G. L. and E. B. Fish. (1978). Image Enhancement for Vegetative Pattern Change Analysis.
Photogrammetric Engineering and Remote Sensing, 44(5):607-616.

Congalton, R. G. (1991). A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote
Sensing of Environment, 37:35-46.

Foody, G. M. (2001). Monitoring the Magnitude of Land-Cover Change around the Southern Limits of the Sahara.
Photogrammetric Engineering and Remote Sensing, 67(7):841-847.

Harrison, B. and J. Kluver. (1989). Reassessing the 'Massachusetts miracle': reindustrialization and balance growth
or convergence to 'Manhattanization', Environment and Planning A, 21:771-801.

Jacobs, H. M. (1999). Fighting Over Land: America’s legacy…America’s Future?. Journal of the American
Planning Association. 65(2):141-149.

Jensen, J. R. (1981). Urban Change Detection Mapping Using Landsat Digital Data. The American Cartographer,
8(2):127-147.

Jensen, J. R. (1996). Introductory Digital Image Processing: A Remote Sensing Perspective, Second Edition.
Prentice Hall, 316 p.

Jensen, J. R. and D. L. Toll. (1982). Detecting Residential Land-Use Development at the Urban Fringe.
Photogrammetric Engineering and Remote Sensing, 48(4):629-643.

King, L. and G. Harris. (1989). Local Responses to Rapid Rural Growth. Journal of the American Planning
Association, 55:181-191.

Kwarteng, A. Y. and P. S. Chavez Jr. (1998). Change detection study of Kuwait City and environs using multi-
temporal Landsat Thematic Mapper data. International Journal of Remote Sensing, 19(9):1651-1662.

Levia, D. F. (1998). Farmland Conversion and Residential Development in North Central Massachusetts. Land
Degradation and Development, 9:123-130.

Lunetta, R. S and C. D. Elvidge. (1998). Remote Sensing Change Detection: Environmental Monitoring Methods
and Applications, Ann Arbor Press, 318 p.

Macleod, R. D. and R. G. Congalton. (1998). A Quantitative Comparison of Change-Detection Algorithms for

Monitoring Eelgrass from Remotely Sensed Data. Photogrammetric Engineering and Remote Sensing,
64(3):207-216.

Masry, S. E., B. G. Crawley and W. H. Hilborn. (1975). Difference Detection. Photogrammetric Engineering and
Remote Sensing, 41(9):1145-1148.

Ryznar, R. M. and T. W. Wagner. (2001). Using Remotely Sensed Imagery to Detect Urban Change: Viewing
Detroit From Space. Journal of the American Planning Association, 67(3):327-336.

Singh, A. (1989). Digital change detection techniques using remotely-sensed data. International Journal of Remote
Sensing, 10(6):989-1003.

Townshend, J. R. G., C. O. Justice, C. Gurney, and J. McManus. (1992). The Impact of Misregistration on Change
Detection. IEEE Transactions on Geoscience and Remote Sensing, 30(5):1054-1060.
U.S. Census Bureau. (2001). Essex County Business Patterns Economic Profile Quickfacts.
http://quickfacts.census.gov/cgi-bin/cnty_QuickLinks?25009. (10 Jan. 2002).

Wesmiller, R. A., S. J. Kristof, D. K. Scholz, P. E. Anuta, and S. A. Momin. (1977). Change Detection in Coastal
Zone Environments*, Photogrammetric Engineering and Remote Sensing, 43(12):1533-1539.

Wickware, G. M. and Howarth P. J. (1981). Change Detection in the Peace-Athabasca Delta Using Digital Landsat
Data. Remote Sensing of Environment, 11:9-25.

Yuan, D. and C. Elvidge. (1998). NALC Land Cover Change Detection Pilot Study: Washington D.C Area
Experiments. Remote Sensing of Environment, 66:166-178.