Geomorphology 202 (2013) 74–85

Contents lists available at ScienceDirect

Geomorphology
journal homepage: www.elsevier.com/locate/geomorph

An assessment of the degree to which Landsat TM data can support the

assessment of fluvial dynamics, as revealed by changes in vegetation
extent and channel position, along a large river
Alexander J. Henshaw a,⁎, Angela M. Gurnell a, Walter Bertoldi b, Nick A. Drake c
a
School of Geography, Queen Mary University of London, London E1 4NS, UK
b
Department of Civil, Environmental and Mechanical Engineering, University of Trento, Trento 38123, Italy
c
Department of Geography, King's College London, London WC2R 2LS, UK

a r t i c l e i n f o a b s t r a c t

Article history: Recent research has demonstrated the capacity of remotely sensed data to enhance our understanding of in-
Received 16 February 2012 teractions between fluvial processes and riparian vegetation. Multispectral Landsat TM satellite data covering
Received in revised form 6 October 2012 the last 30 years are now freely available and have the potential to support investigations at much broader
Accepted 7 January 2013
spatial and temporal scales than previously possible. This paper assesses the degree to which this data set
Available online 18 January 2013
can support the assessment of fluvial dynamics along a large gravel bed river, the Fiume Tagliamento, NE
Keywords:
Italy. Landsat TM scenes are used to determine spatio-temporal trends in vegetation cover and channel posi-
Fluvial processes tion at six sites of contrasting environmental conditions along the river. These trends are interpreted in com-
Vegetation dynamics bination with a number of additional hydrogeomorphological data sets before comparison with known
Remote sensing biogeomorphological characteristics of the study sites as revealed by previous field research and analyses
Landsat of higher resolution remotely sensed data sets. The research highlights a number of limitations with Landsat
Tagliamento TM data that arise as a result of its relatively low spatial resolution including geomorphic feature definition
and active tract delineation. Nevertheless, the analysis demonstrates that Landsat TM data can reveal a
wealth of information that could support further biogeomorphological investigations in other large rivers.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction
Interactions between fluvial processes and riparian vegetation are
attracting increasing research attention (see recent reviews by
Corenblit et al., 2007, 2009a; Osterkamp and Hupp, 2010; Collins et
al., 2012; Gurnell et al., 2012; Osterkamp et al., 2012; Stoffel and
Wilford, 2012). In particular, this work has emphasised the ways in
which some plant species can act as physical ecosystem engineers,
strongly influencing sediment retention and fluvial landform build-
ing, facilitating rapid colonisation and establishment of other plant
species, and inducing rapid adjustments in river channel size, position
and form. However, a fuller understanding of these interactions with-
in different river environmental settings and across spatial and tem-
poral scales is limited by the availability of appropriate data sets
describing vegetation dynamics.
Previous studies of spatio-temporal adjustments in the structure
and extent of riparian vegetation along rivers have typically utilised
data acquired in the field and drawn from discrete spatial units such
as transects across floodplains, quadrats, and individual landforms
such as islands and bars (e.g. Johnson, 2000; Gurnell et al., 2001;
⁎ Corresponding author. Tel.: +44 20 7882 5436; fax: +44 20 7882 7032.
E-mail address: a.henshaw@qmul.ac.uk (A.J. Henshaw).
Kalischuk et al., 2001; Corenblit et al., 2009b), or extracted from se-
quences of high resolution air photographs (e.g. Shafroth et al.,
2002; O'Connor et al., 2003; Zanoni et al., 2008; Hervouet et al.,
2011) or other airborne data sets (e.g. Whited et al., 2007; Bertoldi
et al., 2011a). Whilst photographs and other forms of aerial imagery
are able to provide detailed information over much larger areas
than ground observations, both approaches tend to be limited in
their temporal resolution. Recently, Bertoldi et al. (2011b) analysed
ASTER data to demonstrate the utility of multispectral satellite data
for investigating vegetation dynamics along large river corridors.
ASTER (Advanced Spaceborne Thermal Emission and Reflection Radi-
ometer), mounted on NASA's Terra satellite, provides repeat global
coverage with a spatial resolution of 15 m in the visible and near-
infrared (VNIR) bands, 30 m in shortwave infrared, and 90 m in ther-
mal infrared. However, one limitation of this satellite data source is
that it is quite new, only providing data since 1999 and thereby lim-
iting the extent of any historical analysis that can be achieved.
To extend the time frame for analysis, this paper explores the poten-
tial of Landsat Thematic Mapper (TM) data to quantify spatio-temporal
changes in riparian vegetation extent along a large, gravel bed river. De-
spite the relatively lower spatial resolution of data obtained by the
satellite-mounted, multispectral (7 bands) scanning radiometer (30 m
in the visible, near-infrared (NIR) and mid-infrared (MIR) bands, and

0169-555X/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.geomorph.2013.01.011
 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85 75

120 m in the thermal infrared band), Landsat TM sensors have captured
images of the Earth on a 16-day repeat cycle since the launch of NASA's
Landsat 4 satellite in July 1982. This three decade data source has been
exploited to provide information on a vast range of spatially- and
temporally-variable environmental phenomena, including assessment
of channel migration in large rivers (e.g. Salo et al., 1986; Philip et al.,
1989; Yang et al., 1999; Peixoto et al., 2009). The entire Landsat data ar-
chive is now freely available courtesy of the U.S. Geological Survey
(http://landsat.usgs.gov), providing an inexpensive and easily accessi-
ble source of multispectral data over a much longer time period than
is available from ASTER.
This paper assesses the degree to which Landsat data can support
the assessment of fluvial dynamics as revealed by changes in vegeta-
tion extent as well as channel position along a large river, the Fiume
Tagliamento, NE Italy. This is achieved through the analysis of a series of
Landsat TM images of six sites along the river that are distinguished by
changes in gradient, valley–channel confinement, surface–subsurface
water connectivity and channel planform type. The spatio-temporal dy-
namics extracted from the Landsat TM data are compared with known
characteristics of the six sites to assess the degree to which this three de-
cade data source can provide biogeomorphological insights.
2. Study area
The Fiume Tagliamento, located in north-eastern Italy (Fig. 1),
flows from its source in the Italian Alps to the Adriatic Sea, draining
an area of approximately 2580 km 2. In its upper and middle course,
the river follows a stepped profile of wide, gravel-filled basins sepa-
rated by narrow, bedrock-constrained sections, before flowing across
a wide plain between Villuzza and the coast (Fig. 1). Bed sediments

Fig. 1. Location map of the Tagliamento catchment illustrating the position of study sites A–F, reaches A1–F8 and hydrological gauging stations.
76 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85
fine downstream from boulder/cobble material in the headwaters to
gravel in the lower reaches (Petts et al., 2000). The main stem ex-
hibits a predominantly braided channel planform but in its lower
reaches, the active channel gradually narrows, adopting a transitional
to a meandering style (Gurnell et al., 2000).
The river has a flashy pluvio-nival flow regime with peak flows in
spring and autumn, although floods can occur at any time of the year
(Ward et al., 1999). The average discharge at Venzone (Fig. 1A) is
90 m 3 s −1 with 2, 5 and 10 year floods estimated to be 1100, 1600
and 2150 m3 s−1 (Maione and Machne, 1982). Downstream of
Venzone, water levels during low flows vary strongly along the river
as a result of surface–subsurface water exchange. Low flow discharges
increase in a downstream direction in the reach immediately to the
north of Villuzza due to the entry of two tributary streams and the influ-
ence of a bedrock outcrop that narrows the river to 130 m and induces
water upwelling. Downstream of Villuzza, the river loses water to a vast
alluvial aquifer, resulting in a substantial downstream reduction or ab-
sence of surface water during low flows until the river reaches the
“Linea delle risorgive” (Doering et al., 2007), where a band of silt and
clay forces groundwater to the surface (Fontana et al., 2008).
The Tagliamento retains a wooded riparian zone along most of its
length, characterised by distinct downstream changes in composition
and relative abundance of riparian tree species (Karrenburg et al.,
2003). Alnus incana is the dominant riparian tree species in the head-
waters whereas Populus nigra is dominant in the central and lower
reaches of the river. Five willow species (Salix alba, Salix daphnoides,
Salix elaeagnos, Salix purpurea, Salix triandra) also occur widely
along the river (Karrenburg et al., 2003). All of these species, but
particularly the Salicaceae, are capable of regenerating following
uprooting, transport and deposition by floods.
Wooded islands surrounded by water-filled channels or exposed
gravel are found throughout the river's course, although their spatial
extent varies greatly (Gurnell et al., 2000). Throughout most of the
river's length, these islands are formed by retention, aggradation
and stabilisation of finer sediments by riparian trees. The fine sedi-
ment is trapped by the developing tree canopies and then stabilised
by adventitious roots that extend into the aggrading sediment. How-
ever, islands can also be created by fluvial erosion and dissection of
already-vegetated areas including the floodplain, and where the
supply of fine sediment is limited, bed incision can combine with ag-
gradation to define topographic islands below vegetated patches
(Gurnell et al., 2001).
3. Materials and methods
Vegetation and channel dynamics were explored within six study
sites located in the headwaters and middle and downstream sections
of the Tagliamento main stem (Section 3.1). A set of Landsat TM scenes
were selected (Section 3.2), from which information on vegetation
(Section 3.3) and wetted channel dynamics (Section 3.4) was extracted.
These were used to explore the degree to which it was possible to iden-
tify firstly, properties of vegetation–fluvial interactions that are known
to exist at each site (1 to 6, Section 3.1); and secondly, broader trends
in vegetation–fluvial interactions along the river from the headwaters
to the lower reaches over almost three decades. The extracted informa-
tion was combined with other hydrogeomorphological information to
support the latter analysis (Section 3.5).
3.1. Study sites and reaches
In order to investigate the potential of Landsat TM data to provide
information on vegetation–fluvial interactions of river reaches, we se-
lected six alluvial study sites (A to F, Fig. 1, Table 1) of contrasting
biogeomorphological characteristics.
Site A is a relatively narrow and steep, boulder–cobble bed, braid-
ed headwater reach confined between steep mountain slopes. There
is close coupling between the valley side slopes and the river corridor,
with landslides delivering coarse sediment to the active channel and
so periodically disrupting the braid plain morphology and vegetation
distribution. This site provides the opportunity to investigate:
1. whether analysis of Landsat TM data can distinguish vegetation–
fluvial interactions within a relatively narrow, confined reach af-
fected by a combination of landslide activity, fluvial processes
and island construction, dissection and erosion;
2. whether it is possible to distinguish interactions on lower hillslopes
from those within the river's active tract.
Sites B, C and D form a near-continuous, cobble-bed, braided se-
quence in the middle reaches of the river where the active tract is rela-
tively wide. Site B is confined on the right bank by mountains. It is
separated from Site C by a section which contains a large rock cored is-
land and where the natural dynamics of the riparian forest on the left
bank were impacted by extensive bank engineering following a major
avulsion. Site C is confined by mountains on the left bank and by terraces
and mountains along the right bank, which converge into a gorge at
Villuzza. Although Site D flows over a wide plain downstream of Villuzza,
high terraces confine channel dynamics along the left bank of the braid
plain. Sites B, C and D form a downstream sequence of braided channels
that are distinguished by increasing low flow discharge and groundwater
upwelling through Sites B and C to Villuzza and then pronounced
downwelling of water into the channel bed downstream through Site
D. The riparian woodland and wooded islands throughout Sites B, C and
D are dominated by P. nigra, but its growth performance varies strongly
downstream in response to changing water availability (Gurnell and
Petts, 2006; Bertoldi et al., 2011b), leading to differing dynamics in the
spatial extent and development of vegetated patches (Bertoldi et al.,
2011a,b) and consequential changes in the morphology (bar surface ag-
gradation around vegetation) within the active tract (Bertoldi et al.,
2011a). These three sites offer the opportunity to investigate:
3. whether analysis of Landsat TM data can distinguish vegetation–
fluvial interactions within relatively wide reaches dominated by
the development, coalescence, floodplain attachment and dissec-
tion of islands within the active tract;
4. whether these analyses can distinguish differences in the extent
and dynamics of vegetation under different vegetation growth
conditions.
Sites E and F are located in the lower reaches of the river. They are
unconfined and, reflecting their location downstream of the “Linea
delle risorgive”, are subject to upwelling of groundwater and, thus,
a downstream increase in discharge during low flows. Accompanying
marked fining of bed (cobble to pebble to granule) and bank sedi-
ments, the braided channel of Site E transforms into a meandering
planform through Site F. This change in channel pattern is accompa-
nied by a change in vegetated landforms from wooded islands to
scroll bar and vegetated point bar formation as fluvial processes inter-
act with riparian vegetation to build vegetated landforms, and the
growth performance of riparian trees (assessed from P. nigra annual
growth rings) increases downstream (Gurnell and Petts, 2006).
These two sites offer the opportunity to investigate:
5. whether analysis of Landsat TM data can distinguish vegetation–
fluvial interactions where vegetated landforms become increas-
ingly elongated and located closer to the margins of a relatively
narrow active tract;
6. the degree to which such distinctions can be made as the active
tract narrows from a braided to a meandering planform.
In summary, the 6 study sites provide different topographic, soil
moisture and riparian tree growth contexts for fluvial dynamics. In
all 6 sites, vegetation dynamics should reflect geomorphological distur-
bances (Sites A to F — flood erosion, Site A — landslides) and, between dis-
turbances, the development of specific landforms (Site A — islands on
 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85 77

Table 1
River corridor characteristics at study Sites A to F.

Site A Site B Site C Site D Site E Site F

1 km reach numbers A1 to A6 B1 to B6 C1 to C6 D1 to D6 E1 to E8 F1 to F8
Distance from source (km) 9 to 15 69 to 75 77 to 83 84 to 89 111 to 119 119 to 127
Mean elevation 741 159 128 110 23 8
(metres a.s.l.)
Valley slope (m m−1) 0.019 0.004 0.003 0.003 0.002 0.001
Dominant riparian tree species A. incana P. nigra P. nigra P. nigra P. nigra P. nigra
Confinement Confined Semi-confined Confined Semi-confined Unconfined Unconfined
Surface–subsurface water connectivity No information No information Upwelling Downwelling Upwelling Upwelling
in a downstream directiona
Channel pattern Braided Braided Braided Braided Braided Transitional
to meandering
a
Information from Doering et al. (2007 and personal communication).

stabilised landslide surfaces, dissection islands; Sites B to E — aggrading
islands of fine sediment in response to differential riparian tree growth
performance; Site F — elongated islands, scroll bars and vegetated point
bars).
In order to provide a fixed spatial framework against which to assess
these changes across space and time, the six study sites were subdivided
longitudinally into forty reaches, each 1 km in length (6 reaches within
Sites A, B, C and D and 8 reaches within Sites E and F, Fig. 1).
3.2. Landsat TM data
Scenes for delimiting and comparing riparian vegetation distribu-
tion were selected from the USGS Landsat Archive using three criteria:
(i) a cloud-free river corridor; (ii) fully developed foliage on riparian
vegetation; and (iii) low river flow conditions. In total, 19 summer
(June to September) scenes were selected, covering a period from
1984 to 2011 with one scene each year for 1984, 1986–1994, 1999,
2001–2003, 2006, 2007 and 2009–2011. The data were supplied as
UTM-projected GeoTIFF files of individual Landsat TM bands that had
been processed to Level 1T (Standard Terrain Correction), providing sys-
tematic radiometric and geometric accuracy by incorporating ground
control points and information from a digital elevation model (DEM)
(USGS, 2011).
3.3. Vegetation identification from Landsat TM data
The spatial cover of vegetation in each scene was calculated in
ArcGIS 9.2 using the Normalized Difference Vegetation Index (NDVI)
(Rouse et al., 1974):
NIR−R
NDVI ¼ ; ð1Þ
NIR þ R
where NIR and R represent pixel values from Landsat TM bands 4
(near infrared, 0.76–0.90 μm) and 3 (red, 0.63–0.69 μm), respective-
ly. The NDVI is a sensitive indicator of both the presence and condi-
tion of green vegetation (Lillesand et al., 2004): vegetated areas
yield high NDVI values; features such as clouds, water and snow
yield negative NDVI values; and areas of bare soil, rock and gravel
tend to yield values close to zero. Although a variety of other vegeta-
tion indices exist (Bannari et al., 1995), NDVI was selected because it
has been shown to be sensitive to the types of low density vegetation
found in semi-arid areas (Makkeasorn et al., 2009) and active braided
river corridors (Bertoldi et al., 2011b).
Bertoldi et al. (2011b) discriminated both dense and sparse vege-
tation from gravel and water surfaces using ASTER data. As a result of
the lower spatial resolution of the TM data, a simple separation of
vegetated pixels from other cover (gravel, water) was employed in
the present study. Initially this was achieved using NDVI value ranges
reported in the literature (Nagler et al., 2005; Buhe et al., 2007; Bertoldi
et al., 2011b): gravel or water (NDVIb 0.2); vegetation (NDVI>0.2).
The outer boundary of the active tract in each of the forty study
reaches over the 1984–2011 study period was delimited by including
all pixels classified as, or surrounded by, gravel or water in any Landsat
TM scene within the area of river corridor. From this estimate of the ac-
tive corridor area, the percentage cover of gravel/water and vegetation
was calculated for all forty reaches from at least 18 of the 19 Landsat TM
scenes (atmospheric conditions restricted analysis to Sites A, E and F in
1999).
To check the accuracy of the NDVI-derived estimates, vegetation vis-
ible in 0.5 m resolution geo-referenced air photographs from 8 October
1991 was manually digitised and cover values were computed for study
reaches within the photographed area. These values were then com-
pared with NDVI-based estimates obtained from a Landsat TM scene
from 15 August 1991 (Fig. 2a & b). No large hydrological events oc-
curred between the two dates. There was a strong correlation (R2 =
0.90, p b 0.001) between the air photograph- and the NDVI-based esti-
mates of vegetation cover in the forty study reaches (Fig. 2c). When
the distribution and average of the photograph estimates of the per-
centage of vegetation cover in each 30 m Landsat TM pixel were com-
pared with TM-based pixel estimates of NDVI, further support for the
selection of 0.2 as the threshold NDVI value for vegetation identification
is provided (Fig. 2d). Although the estimates of percentage vegetation
cover in 30× 30 m Landsat TM pixels of a given NDVI value varied sub-
stantially, the mean vegetation cover in pixels with an NDVI value of 0.2
exceeded 50%, indicating that Landsat TM data and the proposed
NDVI-based vegetation classification scheme are appropriate for inves-
tigating vegetation dynamics at the study sites, where the width of the
geomorphologically-active river corridor is considerably larger than
30 m (Table 1).
Following application of the NDVI threshold of 0.2, the derived esti-
mates of vegetation extent were found not to conform to a normal dis-
tribution, although the degree of deviation from normal varied with
site, reach and date. Therefore, nonparametric Kruskal–Wallis tests
were used to assess whether the estimated vegetation cover was statis-
tically significantly different among study sites and reaches and be-
tween dates. Post-hoc pairwise comparisons between sites, reaches
and dates were conducted using Dunn's procedure with Bonferroni
correction to identify which subsets of the data were statistically signif-
icantly different from one another. Moreover, the median rather than
the mean was used throughout to characterise the central tendency of
the estimates of vegetation extent. These statistical analyses were
conducted using XLSTAT 2011.
3.4. Channel position identification from Landsat TM data
Channel positions were estimated from Landsat TM data in ArcGIS
9.2 using the Modified Normalized Difference Water Index (MNDWI)
(Xu, 2006):
G−SWIR
MNDWI ¼ ; ð2Þ
G þ SWIR
78 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85

Fig. 2. (a) NDVI distribution at study Sites B and C derived from the Landsat TM scene (08/10/91) and (b) air photograph composite of study Sites B and C (15/08/91). (c) Compar-
ison of 1 km reach estimates of the vegetation coverage at study Sites B and C derived from Landsat TM and high resolution air photographs. (d) Relationship between
Landsat-derived NDVI and mean vegetation coverage within pixels of a given NDVI estimated from air photographs (error bars represent ±one standard deviation).
Landsat TM data courtesy of the U.S. Geological Survey.

where G and SWIR represent pixel values from Landsat TM bands 2 main stem because of its large width and unstable braided bed. Further-
(green, 0.52–0.60 μm) and 5 (short wavelength infrared, 1.55–1.75 μm), more, many records only extend over relatively short periods of time.
respectively. The MNDWI exploits the strong absorption of energy at However, there are two narrow sections in the middle reaches at
near infrared and short wavelength infrared ranges by water, thereby en- Venzone and Villuzza (Fig. 1A) where records of stages are of good qual-
abling pixels that contain surface water (tend to have positive values) to ity and where long term records are available.
be distinguished from those that do not (tend to have negative values). Hourly observations of the stages are available from Villuzza for
A lack of high resolution aerial imagery for the exact times that the the whole of the 1984–2011 period encompassed by this study. This
Landsat TM scenes were captured prevented the identification of an record was used to provide a first indicator of the potential severity
MNDWI threshold value for surface water, but raw MNDWI values, repre- of flood disturbance of vegetation during the study period. Using
sented as a continuous colour gradation from white (MNDWI=−1) to
black (MNDWI=1), provided an indication of the position of the main
channels since darker areas are most likely to represent channels
containing surface water. Time series animations constructed by overlay-
ing the spatial distribution of riparian vegetation derived from each
Landsat TM scene on maps depicting the position of river channels
allowed temporal changes in channel planform configuration to be com-
pared with the changing spatial distribution of riparian vegetation. These
provided the main information source to underpin analysis of the poten-
tial of Landsat TM data to support investigation of vegetation–fluvial inter-
actions at Sites A to F, focussing on points 1 to 6 (Section 3.1). An overview
of the various stages of Landsat TM data processing is provided in Fig. 3.

3.5. Other hydrogeomorphological information

To support the analysis of broader trends in vegetation–fluvial in-

teractions along the river from its headwaters to lower reaches, four
additional hydromorphological indicators were introduced.
River flows are a fundamental influence on fluvial and vegetation
dynamics. Reliable discharge records are scarce along the Tagliamento Fig. 3. Overview of Landsat TM data processing stages.
 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85
photographs from fixed cameras, Bertoldi et al. (2009) identified a
200 cm stage at Villuzza as the level at which river flows start to in-
teract strongly with vegetation within Site C. Therefore, a river stage
in excess of 200 cm at Villuzza is used in this study as an indicator
of likely flood disturbance of vegetated patches within the river's ac-
tive tract. The stage record in excess of 200 cm is compared with es-
timates of changing vegetation cover through the study period to
identify the degree to which vegetation cover appears to respond to
flood disturbance, particularly at Sites B, C and D, which are located
close to Villuzza.
Stage records at Venzone extend back to the late 19th century, but
their temporal resolution is variable and generally lower than at Villuzza.
Researchers have attempted to develop a relationship between stage and
discharge at Venzone, with Maione and Machne (1982) estimating a
10 year flood discharge of 2150 m3 s−1 and Bertoldi et al. (2010)
suggesting a bankfull discharge of 1700 m3 s−1. These estimates were
used to provide some assessment of the validity of an analysis of historical
flow records collected in the 1930s and 1940s from six gauging stations
distributed across the Tagliamento catchment (Mosetti, 1983). The analy-
sis identified a relationship between the maximum recorded discharge
(Qmax in m3 s−1) and drainage area (A in km2) of Qmax =0.19A1.19
(r2 =0.96, Mosetti, 1983). A Qmax estimate at Venzone of 2000 m3 s−1,
derived from this relationship, compared favourably with the estimates
of high flow discharges at this site by Maione and Machne (1982) and
Bertoldi et al. (2010). Therefore, the relationship was used to provide a
second indicator representative of ‘channel-forming’ stream power
(Ωmax in W m−1) in the forty study reaches:
Ωmax ¼ ρgQ max S; ð3Þ
where ρ=fluid density (1000 kg m−3), g=gravitational acceleration
(9.81 m s −1), Qmax = representative discharge (m3 s −1); and S=
channel bed slope (m m−1). The S for each study reach was estimated
using elevation data extracted from a 30 m resolution DEM (ASTER
GDEM) and channel length values derived from recent aerial images
using ArcGIS 9.2.
Information on the calibre of bed material lag deposits was used to
provide two further indicators of river energy conditions. Channel lag
deposits tend to fine downstream in response to the combined pro-
cesses of selective fluvial entrainment and abrasion (Ferguson et al.,
1996), although lag deposits in mountainous headwater areas may
be anomalously large where they are controlled by non-fluvial pro-
cesses such as near-continuous recruitment of sediment from steep
unstable hillslopes (Rice and Church, 1996) or past glacial deposition
rather than contemporary extreme flood events. An archive of mea-
surements of the calibre of the coarsest channel lag deposits is avail-
able for 97 locations along the Tagliamento main stem. Each of these
were assembled from 5 or more replicate 0.5 m × 0.5 m quadrats
using the methodology described by Petts et al. (2000) to provide es-
timates of the median grain size (D50) and the b-axis of the largest in-
dividual clast (Dmax). Data from one or more sets of replicate quadrat
measurements were available within 19 of the forty study reaches.
Where more than one set of measurements had been made within a
single reach, the coarsest set of measurements was selected in an at-
tempt to define the coarsest lag deposits in the reach.
Of the above indicators, the last three (Ωmax, D50 and Dmax) are all in-
dicative of reach energy conditions and so their relationships with the
median and maximum reach vegetation cover over the 1984–2011
study period were investigated using linear regression analysis.
4. Results
4.1. Spatial variations in vegetation cover
Strong spatial variation in relative vegetation cover is evident
along the Tagliamento from the analysis of Landsat TM data (Fig. 4).
79
The statistical significance of these differences was tested using
Kruskal–Wallis tests (Table 2). Overall, headwater Site A (median
68%) showed significantly higher cover than the most downstream
Site F (45%), which in turn showed higher cover than Sites C (29%)
and D (24%) in the middle reaches and downstream Site E (34%),
with middle reach B (17%) showing the lowest cover overall (Table 2).
Within the six sites, Sites B (median B1= 10%, B6 = 27%), E (median
E1 = 28%, E8 = 49%) and F (median F1= 38%, F8 = 62%) showed an in-
crease in cover in a downstream direction, whereas reach A (median
A1 = 79%, A6 = 58%) showed a downstream decrease (Table 2). Sites
C (median C1= 25%, C4= 44% C6 = 27%) and D (median D1= 21%,
D3= 39% D6= 15%) showed an increase followed by a decrease in
cover in a downstream direction (Table 2).
4.2. Associations between reach energy conditions and vegetation cover
Associations between the median and the maximum vegetation
cover recorded in each reach and three indicators of river energy con-
ditions (Ωmax, D50 and Dmax) were explored to assess the degree to
which spatial variations in river energy appeared to influence vegeta-
tion extent.
A statistically significant negative linear regression relationship was
found between log-transformed median vegetation cover and log
−0.37
transformed reach Ωmax (Fig. 5a, 1511Ωmax , R 2 = 0.46, F1,38 = 32.7,
Pb 0.001). Whilst there is considerable scatter around this relationship,
reaches with values of Ωmax exceeding 20,000 W m−1 generally
supported much lower levels of vegetation cover, although there were
a number of exceptions to this rule, notably reaches E6 and E7 where
the median vegetation cover was 42% and 49%, respectively. The degree
of scatter was reduced slightly when the log transformed maximum
vegetation cover observed during the 1984–2011 period was plotted
(Fig. 5b, 2800Ωmax −0.44
, R2 = 0.49, F1,38 = 36.0, P b 0.001) rather than the
median.
Statistically significant linear relationships were also found between
the median and maximum reach vegetation cover and the grain size of
channel lag deposits (D50 and Dmax) (Fig. 6), but the strength of these
relationships, as indicated by the coefficient of determination, increased
greatly when information from Site A was excluded from the analysis.
Negative relationships with the highest coefficients of determination
were found between the maximum (R2 = 0.50, F1,12 = 12.9, P = 0.004)
and median (R2 =0.46, F1,12 = 10.2, P = 0.008) reach vegetation cover
and D50, with weaker relationships between the maximum (R2 = 0.32,
F1,12 = 5.6, P= 0.036) and median (R2 = 0.36, F1,12 = 6.2, P= 0.024)
vegetation cover and Dmax.
The relatively coarse nature of the bed material at Site A relative to
the other sites is further demonstrated when the values of the three en-
ergy indicators are compared on scatter plots relating D50 (Fig. 7a) and
Dmax (Fig. 7b) to Ωmax. The Site A data points are located as distinct out-
liers from Sites B–F, whilst the latter display a relatively strong relation-
ship between the energy indicators, particularly between Dmax and
Ωmax (Fig. 7b).
4.3. Temporal trends in vegetation cover
Temporal variations in median vegetation cover show some broad
similarities across the study sites through the study period. Fig. 8 il-
lustrates these changes in the context of river stage records that
exceeded 200 cm at Villuzza (the stage above which river levels in-
teract with vegetated patches within the active tract in reach C;
Bertoldi et al., 2009). Major flood periods at Villuzza, in terms of
both duration and magnitude occurred in the winter of 2000–2001,
and in the autumn–winter during late 1996 and 1984. Other events
that exceeded a 300 cm water level (approximately bankfull level
within reach C; Bertoldi et al., 2009) occurred late in 1982, 1984,
1990, 1993, 2002 and 2004. There were also two periods of extended
low flows between 1985 and 1990, and between 2005 and 2009. Sites
80 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85

Fig. 4. Box-and-whisker plots showing spatial variation in vegetation coverage among study sites and reaches during the 1984–2011 period.

B to D, which are closest to the Villuzza gauge, show a notable reduc- years 1987, 1988, 1989,1992 and 1993 as showing significantly higher
tion in vegetation cover between the 1994 and 2001 images that vegetation cover than other years at many sites.
bound the 1996 and 2000–2001 flood periods, and they also show a
broad expansion in vegetated area during the two low flow periods, 4.4. Vegetation dynamics interpreted from temporal animations at
particularly within Site C. Although Sites A, E and F are located in individual sites
the headwaters and lower reaches of the river and, thus, are some dis-
tance from the middle reach location of the Villuzza gauge, they show Temporal changes in the spatial distribution of riparian vegetation
similar temporal changes in vegetation extent, of which a reduction at Sites A to F estimated from Landsat TM data can be viewed in An-
between 1994 and 2001 is particularly marked at all sites. The statis- imations 1–6 (available in the web version of this paper). This section
tical significance of these changes was supported by Kruskal–Wallis describes the features that may be observed within the different sites.
tests, which compared reach vegetation coverage estimates across all At Site A (animation 1), the extensive vegetation cover within the
sites and within individual sites on the 19 analysed dates (Table 2). active tract tends to concentrate along the margins. Discrete islands of
These identified the years 2001, 2002 and 2003 as showing significantly vegetation are rare and large, and in many reaches, vegetation extends
lower vegetation cover than other years at virtually all sites and the across the entire river bed when vegetation cover is high (e.g. 1984 to
1993). Even when vegetation cover is relatively low (e.g. 1999 to
2006), it is difficult to identify any clear patterns in the grey scale pixels
Table 2 that might indicate the position of the main wetted channel(s).
Results of Kruskal–Wallis tests applied to the percentage of vegetation cover estimates At Site B (animation 2), vegetated areas are confined almost en-
according to their location and year of survey. tirely to the margins of the active tract, with a tendency to be more
Site Groups Kruskal Degrees of Probability Significant differences persistent and extensive in reaches B4 to B6. The grey scale pixels in-
K freedom between groups (P b 0.01)a dicate an extremely dynamic braided channel pattern across the
All sites Sites 341.0 5 b0.0001 A > F > C, D, E > B
E>D
A Reaches 16.8 5 0.005 A1 > A4, A6
B Reaches 31.3 5 b0.0001 B6 > B1, B2, B3
B5, B4 > B1
C Reaches 26.5 5 b0.0001 C4 > C1, C6
D Reaches 73.9 5 b0.0001 D2, D3, D4 > D1, D5, D6
E Reaches 60.8 7 b0.0001 E8, E7 > E6, E5, E4, E1
E8 > E2
E3 > E4
F Reaches 55.5 7 b0.0001 F8, F7 > F6, F3, F2, F1
F5 > F6
All sites Sites 86.4 18 b0.0001 89 > 84, 88, 01, 02, 03, 06, 09
87, 92, 93 > 02, 03
93 > 01
A Reaches 84.9 18 b0.0001 87 > 94, 99, 01, 02, 03, 06, 07
92 > 99, 02, 03, 06
88 > 99, 02
93 > 02
B Reaches 45.8 17 b0.001 NS
C Reaches 61.5 17 b0.0001 89, 93 >01, 02, 03
92 > 02
D Reaches 20.3 17 0.259 NS
E Reaches 38.7 18 b0.0001 89 > 99, 03
F Reaches 54.0 18 b0.0001 89 > 84, 88, 99, 01, 02, 03, 09
a
Pairwise comparisons undertaken using Dunn's procedure with Bonferroni Fig. 5. Relationships between (a) the median and (b) the maximum vegetation coverage and
correction. estimated reach Ωmax during the 1984–2011 period.
 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85 81

Fig. 6. Relationships between the reach median (a, b) and the maximum (c, d) vegetation coverage and bed material size (D50, Dmax, respectively) during the 1984–2011 period.
Trend lines fitted to data from Sites B–F only.

entire active tract that can be very dense (e.g. 1990 and 2002), and
can disappear completely throughout the site (e.g. 2003) but particu-
larly in reaches B1 to B3 (e.g. 1986, 1988, 1992 and 1993).
At Site C (animation 3), dynamic vegetated patches are present in
all reaches. In reaches C5 and C6, vegetated patches are relatively rare
and are confined to the active tract margins in the early part of the
study period (1984 to 1994). In reaches C1 and C2, numerous small,
dynamic vegetated patches can be observed across the active tract
throughout the study period, with some larger patches periodically
attaching and detaching to the left bank between 1984 and 1994. In
reaches C3 and C4, large, dynamic vegetated patches are present
across the active tract throughout the study period. The grey scale pixels
indicate a very complex and dynamic network of braid channels
extending across the entire braid plain throughout the study period, par-
ticularly in reaches C1 to C5. In C5 these channels become very closely
spaced and combine to form a predominantly single channel in reach C6.
At Site D (animation 4), vegetated patches form a very distinct pat-
tern. Towards the right bank of reaches D3 and D4 and to some extent
D2, there are clearly defined, large vegetated patches until 1987,
which almost disappear in 1988, and then expand, coalesce and attach
to the right bank from 1990 through to the end of the study period. In
the remaining part of the braid plain, small islands of vegetation occur

Fig. 7. Relationships between reach Ωmax and a) D50 and b) Dmax. The trend lines are based on data from Sites B–F only.
82 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85
Fig. 8. Temporal variation in the median vegetation coverage at study Sites A–F during the 1984–2011 period in relation to river levels in excess of 200 cm at Villuzza (dashed lines
indicate the approximate 300 cm bankfull level within Site C).
widely but are short-lived. The grey scale pattern highlights dynamic
braided channels that remain confined to the central and left bank
areas of the active tract and are often dominated by a single large chan-
nel in reaches D1 and D2.
At Site E (animation 5), very large dynamic vegetated patches persist
throughout the study period in reaches E6 to E8, whereas upstream in
reaches E1 to E5, vegetated patches are much smaller and rarer within
the central part of the braid plain. Instead, vegetated areas in reaches E1
to E5 tend to be confined to the margins of the active tract, usually at-
tached to the floodplain, and were present predominantly along the left
bank until 1999 and along the right bank from 2001 to the end of the
study period. The grey scale pixels indicate an extremely dynamic braided
channel pattern across the entire active tract, but there are only three im-
ages in which wetted channels are clearly identifiable throughout the site
(1990, 2002, and 2009). In other images, reaches with no apparent wet-
ted channels extend from E1 (1984, 1987, 1991, 2001, 2010, and 2011)
to include E2 and E3 (1988, 1992, 1994, 1999, 2006, and 2007), E4
(2003) and E5 (1986), and E6 (1993).
At Site F (animation 6), vegetated areas are predominantly confined
to the active tract margins, where they are attached to the floodplain
and are arranged in an alternating sequence (from the left to the right
bank) in a downstream direction. These alternating patches show some
downstream movement and lateral extension into the active tract
through the study period, particularly in reaches F4 to F8. Where discrete
islands of vegetation form in reaches F4 to F8, they tend be close to the
edge of the marginal vegetated patches, have a linear form, and eventu-
ally attach to the marginal patches. This is particularly apparent on the
left bank in reaches F4 to F5. Reaches F1 to F3 support larger islands of
vegetation that are also highly dynamic and, although mainly confined
to areas near the margins of the active tract, they appear occasionally
in more central locations. The grey scale pixels indicate a dynamic
multi-thread channel system, supporting two to three channels in
reaches F1 to F3, which gradually converge to define a single, sinuous
main channel through reaches E4 to E8.
5. Discussion
5.1. The utility of Landsat TM data and the impact of its spatial resolution
on the observed trends
The underlying aim of the present paper was to assess the degree
to which analysis of Landsat TM data can support the assessment of
 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85
fluvial dynamics, as revealed by changes in vegetation extent and
channel position, along a large river.
Analysis of the properties of 19 Landsat TM scenes within forty
1 km length reaches distributed across six sites, revealed statistically
significant contrasts in vegetation cover as well as dynamic, mainly
multi-thread wetted channels through time and along the river's
course. In most cases these patterns confirmed the understanding of
vegetation behaviour along the river's main stem, established by
ground surveys (e.g. Gurnell et al., 2000, 2001; Gurnell and Petts,
2006), and within Sites B to D, from the analysis of higher spatial res-
olution satellite data (Bertoldi et al., 2011b) and air photographs
(Zanoni et al., 2008). The major contribution of the present analysis
is to provide data to support previous interpretations for the entire
main stem of the river over three decades.
Nevertheless, the 30 m resolution of the Landsat TM data is con-
siderably coarser than many fluvial features. In particular, many
small vegetated features such as pioneer and small building islands
(Gurnell et al., 2001) can be absorbed within a single pixel as can a
mix of vegetated areas of different ages and densities. This means
that there are numerous mixed pixels containing several land cover
types (water, bare gravel, sparse vegetation and dense vegetation).
As a result, identification of discrete vegetated features was difficult,
explaining the large error margins illustrated in Fig. 2d.
Two very clear issues arise from this in the present analyses. Al-
though discrete groups of vegetated pixels were identified in all of
the study reaches, these are unlikely to correspond to individual
islands, but may incorporate several islands separated by small gaps
of exposed gravel or water. Additionally, the outer margin of the ac-
tive tract is defined in an equally fuzzy way, and, where the tract is
relatively narrow, such as at Sites A and F, these errors have a much
greater impact on the outcomes of analyses than in wider reaches. In-
deed, uncertainty relating to the distinction of islands from floodplain
woodland leads to the definition of a single rather than a temporally
dynamic boundary for the active tract, which defined the outer limit
of likely exposure to fluvial processes (c.f. Piégay et al., 2005) in the
present analyses.
The above issues are particularly marked in the analysis of the rel-
atively narrow active tract at Site A and may also be important within
the relatively narrow downstream section of Site F. In these sites, the
active tract identified by image overlay was between 2 and 10 pixels
wide at Site A, and was around 10 pixels in the narrowest sections of
Site F. One impact of this at Site A was that the estimated vegetated
area extended across the entire active tract in many images (anima-
tion 1), and achieved a site median vegetation cover close to 100%
for the most heavily vegetated image (Fig. 8, 1987). Such extensive
vegetation cover is not possible in this reach, which has supported a
well-defined braided channel at least since the 1940s (air photograph
evidence). Moreover, the remarkably few islands illustrated within an-
imation 1, suggests that when islands are present on the ground, they
become absorbed into the marginal vegetation in the image analysis.
This problem extends to large islands that are present throughout the
site (particularly in reaches A1, A3, and A5), which are formed when
trees grow on coarse, elevated landslide deposits within the active
tract. A second impact arises because the unvegetated areas are repre-
sented by very few pixels and so the grey scale variations yield little in-
formation concerning the position, number or relative size of wetted
braid channels. A third impact is that the heavily vegetated appearance
of the active tract, even in the least vegetated images, does not match
the largely vegetation-free actively braided appearance of the river,
which extends to the toes of adjacent mountain slopes throughout
most of this site. This is because the boundary of the active tract is de-
fined by pixels that were rarely vegetation free (often only in a single
image). These occasional exposures are probably mainly areas of
bare rock debris on the lower hillslopes, delivered by the numerous
landslides that feed into Site A and not part of the river's braid plain.
This illustrates a limitation of image spatial resolution and failure to
83
incorporate topographic data to identify the outer margin of the ac-
tive tract in a narrow, hillslope-confined site, where definition of the
riparian zone is in any case very difficult (e.g. Polvi et al., 2011).
Within the narrow downstream reaches of Site F, similar problems
of the aggregation of discrete vegetated areas may also be prevalent.
Whilst the 30 m spatial resolution of the analysed data will inevita-
bly result in aggregation of both vegetated patches and unvegetated
patches as mixed pixels are allocated to one or other class, this prob-
lem is particularly important in narrow reaches, where the active
tract width extends to only a few pixels.
5.2. Biogeomorphological interpretation
Despite the limitations imposed by the spatial resolution of the
data, statistically significant spatial and temporal variations in vegeta-
tion cover were identified from the analysis of Landsat TM data and,
in most cases, these variations matched trends previously identified
through ground survey or analysis of remotely-sensed data sets of
higher spatial resolution. Furthermore, changes in the wetted channel
network in relation to the extent and dynamics of vegetated islands
reflect previous analyses of 1:10,000 scale maps (Gurnell et al., 2000).
The spatial resolution of the data was not sufficient to produce
useful biogeomorphological insights for Site A and thus:
1. It was not possible to distinguish vegetation–fluvial interactions
affected by a combination of landslide activity, fluvial processes
and island construction, dissection and erosion at this site
2. Nor was it is possible to distinguish interactions on lower hillslopes
from those within the river's active tract.
However, the relative variations in vegetation cover through time
show similar features to other sites, demonstrating an aggregate re-
sponse to fluvial disturbances. Furthermore, analyses of additional
hydrogeomorphological data sets illustrate that the bed sediment is
relatively coarse at this site, indicating the potential importance of
landslides in delivering large sediment particles that are not readily
moved by the river and so leads to extremely coarse channel lag
deposits.
For Sites B to F, the analyses of Landsat TM data yield much useful
information relating to fluvial and vegetation dynamics. At these
sites, analysis of Landsat TM data was able to:
3. distinguish vegetation–fluvial interactions within relatively wide
reaches dominated by the development, coalescence, floodplain
attachment and dissection of islands within the active tract;
4. to distinguish differences in the extent and dynamics of vegetation
and to relate it to differences in the extent of wetted channels that
are indicative of moisture availability and thus may support differ-
ences in vegetation growth performance.
The contrasts in vegetated island and floodplain margin dynamics
that support statement 3 were fully described in Section 4.4. These
can be combined with two other pieces of information extracted
from the Landsat TM data to support statement 4 across all sites
from B to E: the width of the active tract, and the presence and extent
of wetted river channels. Gurnell and Petts (2006) attributed the
varying presence of vegetated islands along the Tagliamento to the
following two main controlling factors: (i) water availability that sup-
ports riparian tree growth, and (ii) a wide active tract where there is
space for floods to spread laterally, reducing unit stream power and
thus the degree of flood disturbance imposed by any particular flow
event.
Surface water availability can be inferred from the extent and con-
tinuity of the darkest grey scale pixels in the animations. Within Sites
B and E, several reaches show little evidence of surface water in at
least some of the 19 summer images. Near-surface groundwater is in-
dicated by flowing channels on the surface and is crucial to the pres-
ence and growth performance of riparian tree species (Xu et al., 2007,
84 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85
2009; Imada et al., 2008; Gonzalez et al., 2010; Orellana et al., 2012;
Singer et al., 2012). The number of images indicating negligible sur-
face water increase upstream within Sites B and E. All of reaches B1
to B6 are apparently free of surface water in one of the analysed im-
ages with all three upstream reaches, B1 to B3, indicating absence of
surface water on a further 4 occasions. Within Site E, reach E1 has
no evidence of surface water in 15 of the analysed images, with
reaches E2, E3, E4, E5 and E6, lacking surface water on 9, 9, 3, 2, and
1 occasion(s), respectively. Furthermore, the density of wetted chan-
nels increases downstream in most images for Sites B and E, and de-
creases downstream for most images for Site D, with reach D6
showing no significant surface water on one occasion. These patterns
are consistent with downstream changes in low flow discharge
through Sites B to F, reflecting groundwater upwelling and/or tribu-
tary water inputs (Gurnell and Petts, 2006; Doering et al., 2007).
Those reaches that show periods without significant surface water
show only a few small islands within the braid plain, whereas many
of the reaches where wetted channels are present show more exten-
sive island development, notably C3, C4, E7, E8, and F1 to F3. The
remaining reaches split into (i) wide reaches where water availability
is probably transitional between the dry and wet reaches (e.g. B4 to
B6, C1 to C3, D2 to D5) and islands within the braid plain are of rela-
tively small size and intermediate frequency, and (ii) narrowing to
narrow reaches, where unit stream power during high flows is elevat-
ed and so islands are not present (C5, C6, D1, F6 to F8) or are elongat-
ed and at the margins of the active tract (F4 to F5). All of these
detailed interpretations of the changing controls on vegetation vigour
(water availability) and the ability of vegetation to persist in narrow
reaches of relatively high stream power are reflected in the down-
stream changes in vegetation coverage that are shown in Fig. 4. Low
water availability also explains why the reaches in Site B have rela-
tively lower vegetation coverage than the other sites of similar
Ωmax, D50 and Dmax in Figs. 5 and 6.
The above observations also confirm that:
5. analysis of Landsat TM data can distinguish vegetation–fluvial in-
teractions where vegetated landforms become increasingly elon-
gated and located closer to the margins of a relatively narrow
active tract;
6. analysis of Landsat TM data can distinguish vegetation–fluvial in-
teractions where the active tract narrows from a braided to a
meandering planform.
The method of delimiting the boundary of the active tract has pro-
vided additional insights into how the floodplain edge is constructed
and, in some cases, how the active tract has migrated laterally. This is
particularly apparent in Site F, where a predominantly single thread
channel is bordered by distinct lateral and downstream migration in
the vegetated area, and in reaches F4 to F5 apparent attachment of
an elongated island (potentially representing a sequence of scroll
bars) to the floodplain woodland. Along reaches E1 to E5, islands at-
tach to the floodplain showing a trend of lateral movement of the
braid plain towards the left bank over the study period, whereas in
reaches D2 to D4, extensive large islands close to the right bank en-
large and coalesce inducing narrowing of the active tract over the
study period. In all of these reaches, the styles of channel migration
and evolution are illustrated, and the estimates of vegetation cover
are inflated by the extended boundary of the active tract that incorpo-
rates the entire channel movement that is achieved during the study
period, making some of the estimates of vegetation cover in Fig. 4
larger than that supported within the true active tract.
6. Conclusions
Analysis of Landsat TM data to identify variations in vegetated
area and wetted channels through time and across space has been
shown to have some limitations but it has also revealed a variety of
biogeomorphological information that is supported by previous field
observations and analysis of higher resolution remotely sensed data
sets.
The main limitation reflects the relatively low spatial resolution of
the data, which prevents more than aggregate analysis of narrow river
corridors (e.g. less than 250 m wide). Thus, analysis of river dynamics
using Landsat TM data is only appropriate for larger rivers. Moreover,
definition of the boundary of the active tract using Landsat TM data
alone results in the incorporation of dynamic sections of the riparian
zone at sites where the channel is narrowing or migrating, so providing
overestimates of vegetation cover within the active tract. This is partic-
ularly prominent at migrating meandering sites such as F. The definition
of the active tract boundary may also incorporate the lower sections of
hillslopes, where these are closely coupled to the active tract by land-
slide activity, and particularly where the river corridor is narrow rela-
tive to the spatial resolution of the data, again giving substantial
overestimates of the vegetated area.
Nevertheless, a combination of information on the locations and
extent of vegetated areas, the presence and extent of wetted chan-
nels, and their changes through time, as well as an active tract defined
to incorporate vegetated patch evidence of river margin migration, all
contribute to an integrated interpretation of the following three main
properties of the study river:
(i) variations in water availability and its relationship with island
frequency, size and dynamics,
(ii) the manner in which vegetated patches evolve along the active
tract margins and coalesce or are incised from the floodplain,
affecting migration, narrowing and/or enlargement of the ac-
tive tract in braided reaches
(iii) the association between elongated island (scroll bar) forma-
tion and meander migration in sinuous, single-thread reaches.
Supplementary data to this article can be found online at http://
dx.doi.org/10.1016/j.geomorph.2013.01.011.
Acknowledgements
The authors gratefully acknowledge funding from the Leverhulme
Trust (ref. no. F/07 040/AP), which supported the research reported in
this paper. Landsat TM data courtesy of the U.S. Geological Survey.
References
Bannari, A., Morin, D., Bonn, F., Huete, A.R., 1995. A review of vegetation indices. Re-
mote Sensing Reviews 13, 95–120.
Bertoldi, W., Gurnell, A.M., Surian, N., Tockner, K., Zanoni, L., Ziliani, L., Zolezzi, G., 2009.
Understanding reference processes: linkages between river flows, sediment dy-
namics and vegetated landforms along the Tagliamento River, Italy. River Research
and Applications 25, 501–516.
Bertoldi, W., Zanoni, L., Tubino, M., 2010. Assessment of morphological changes in-
duced by flow and flood pulses in a gravel bed braided river: the Tagliamento
River (Italy). Geomorphology 114, 348–360.
Bertoldi, W., Gurnell, A.M., Drake, N.A., 2011a. The topographic signature of vegetation
development along a braided river: results of a combined analysis of airborne lidar,
color air photographs, and ground measurements. Water Resources Research 47,
W06525.
Bertoldi, W., Drake, N., Gurnell, A.M., 2011b. Interactions between river flows and col-
onizing vegetation on a braided river: exploring spatial and temporal dynamics in
riparian vegetation cover using satellite data. Earth Surface Processes and Land-
forms 36, 1474–1486.
Buhe, A., Tsuchiya, K., Kaneko, M., Ohtaishi, N., Halik, M., 2007. Land cover of oases and
forest in Xinjiang, China, retrieved from ASTER data. Advances in Space Research
39, 39–45.
Collins, B.D., Montgomery, D.R., Fetherston, K.L., Abbe, T.B., 2012. The floodplain large-
wood cycle hypothesis: a mechanism for the physical and biotic structuring of
temperate forested alluvial valleys in the North Pacific coastal ecoregion. Geomor-
phology 139–140, 460–470.
Corenblit, D., Tabacchi, E., Steiger, J., Gurnell, A.M., 2007. Reciprocal interactions and ad-
justments between fluvial landforms and vegetation dynamics in river corridors: a
review of complementary approaches. Earth-Science Reviews 84, 56–86.
Corenblit, D., Steiger, J., Gurnell, A.M., Naiman, R.J., 2009a. Plants intertwine fluvial
landform dynamics with ecological succession and natural selection: a niche
 A.J. Henshaw et al. / Geomorphology 202 (2013) 74–85
construction perspective for riparian systems. Global Ecology and Biogeography
18, 507–520.
Corenblit, D., Steiger, J., Gurnell, A.M., Tabacchi, E., Roques, L., 2009b. Control of sedi-
ment dynamics by vegetation as a key function driving biogeomorphic succession
within fluvial corridors. Earth Surface Processes and Landforms 34, 1790–1810.
Doering, M., Uehlinger, U., Rotach, A., Schlaepfer, D.R., Tockner, K., 2007. Ecosystem
expansion and contraction dynamics along a large Alpine alluvial corridor (Tagliamento
River, Northeast Italy). Earth Surface Processes and Landforms 32, 1693–1704.
Ferguson, R., Hoey, T., Wathen, S., Werritty, A., 1996. Field evidence for rapid down-
stream fining of river gravels through selective transport. Geology 24, 179–182.
Fontana, A., Mozzi, P., Bondesan, A., 2008. Alluvial megafans in the Venetian–Friulian
Plain (north-eastern Italy): evidence of sedimentary and erosive phases during
Late Pleistocene and Holocene. Quaternary International 189, 71–90.
Gonzalez, E., Gonzalez-Sanchis, M.A., Comin, F.A., Muller, E., 2010. Hydrologic thresh-
olds for riparian forest conservation in a regulated large Mediterranean river.
River Research and Applications 28, 71–80.
Gurnell, A.M., Petts, G.E., 2006. Trees as riparian engineers: the Tagliamento River, Italy.
Earth Surface Processes and Landforms 31, 1558–1574.
Gurnell, A.M., Petts, G.E., Harris, N., Ward, J.V., Tockner, K., Edwards, P.J., Kollmann, J.,
2000. Large wood retention in river channels: the case of the Fiume Tagliamento,
Italy. Earth Surface Processes and Landforms 25, 255–275.
Gurnell, A.M., Petts, G.E., Hannah, D.M., Smith, B.P.G., Edwards, P.J., Kollmann, J., Ward,
J.V., Tockner, K., 2001. Riparian vegetation and island formation along the gravel-
bed Fiume Tagliamento, Italy. Earth Surface Processes and Landforms 26, 31–62.
Gurnell, A.M., Bertoldi, W., Corenblit, D., 2012. Changing river channels: the roles of hy-
drological processes, plants and pioneer landforms in humid, temperate, mixed
load gravel bed rivers. Earth-Science Reviews 111, 129–141.
Hervouet, A., Dunford, R., Piégay, H., Belletti, B., Trémélo, M.-L., 2011. Analysis of post
flood recruitment patterns in braided-channel rivers at multiple scales based on
an image series collected by unmanned aerial vehicles, ultra-light aerial vehicles,
and satellites. GIScience & Remote Sensing 48, 50–73.
Imada, S., Yamanaka, N., Tamai, S., 2008. Water table depth affects Populus alba fine
root growth and whole plant biomass. Functional Ecology 22, 1018–1026.
Johnson, W.C., 2000. Tree recruitment and survival in rivers: influence of hydrological
processes. Hydrological Processes 14, 3051–3074.
Kalischuk, A.R., Rood, S.B., Mahoney, J.M., 2001. Environmental influences on seedling
growth of cottonwood species following a major flood. Forest Ecology and Man-
agement 144, 75–89.
Karrenburg, S., Kollmann, J., Edwards, P.J., Gurnell, A.M., Petts, G.E., 2003. Patterns in
woody vegetation along the active zone of a near-natural Alpine river. Basic and
Applied Ecology 4, 157–166.
Lillesand, T.M., Kiefer, R.W., Chipman, J.W., 2004. Remote Sensing and Image Interpre-
tation (5th edition). John Wiley & Sons Ltd., New York (763 pp.).
Maione, U., Machne, G., 1982. Studio sulla formazione delle piene del Fiume
Tagliamento. ETACONSULT, Milan.
Makkeasorn, A., Chang, N.-B., Li, J., 2009. Seasonal change detection of riparian zones
with remote sensing images and genetic programming in semi-arid watershed.
Journal of Environmental Management 90, 1069–1080.
Mosetti, F., 1983. Sinisti sull'idrologica del Friuli–Venezia Giulia, Quaderni dell'Ente
Tutela Pesce del Friuli-Venezia Giulia. Rivista di Limnologia 6.
Nagler, P., Glenn, E.P., Hursh, K., Curtis, C., Huete, A., 2005. Vegetation mapping for
change detection on an arid-zone river. Environmental Monitoring and Assess-
ment 109, 255–274.
O'Connor, J.E., Jones, M.A., Haluska, T.L., 2003. Floodplain and channel dynamics of the
Quinault and Queets Rivers, Washington, USA. Geomorphology 116, 274–285.
Orellana, F., Verma, P., LoheideI, S.P., Daly, E., 2012. Monitoring and modelling water–
vegetation interactions in groundwater-dependent ecosystems. Reviews of Geo-
physics 50 (Paper 2011RG000383).
Osterkamp, W.R., Hupp, C.R., 2010. Fluvial processes and vegetation — glimpses of the
past, the present, and perhaps the future. Geomorphology 116, 274–285.
85
Osterkamp, W.R., Hupp, C.R., Stoffel, M., 2012. The interactions between vegetation and
erosion: new directions for research at the interface of ecology and geomorphology.
Earth Surface Processes and Landforms 37, 23–36.
Peixoto, J.M.A., Nelson, B.W., Wittmann, F., 2009. Spatial and temporal dynamics of river
channel migration and vegetation in central Amazonian white-water floodplains by
remote-sensing techniques. Remote Sensing of Environment 113, 2258–2266.
Petts, G.E., Gurnell, A.M., Gerrard, A.J., Hannah, D.M., Hansford, B., Morrissey, I.,
Edwards, P.J., Kollmann, J., Ward, J.V., Tockner, K., Smith, B.P.G., 2000. Longitudinal
variations in exposed riverine sediments: a context for the ecology of the Fiume
Tagliamento, Italy. Aquatic Conservation: Marine and Freshwater Ecosystems 10,
249–266.
Philip, G., Gupta, R.P., Bhattacharya, A., 1989. Channel migration studies in the middle
Ganga basin, India, using remote sensing data. International Journal of Remote
Sensing 10, 1141–1149.
Piégay, H., Darby, S.E., Mosselman, E., Surian, N., 2005. A review of techniques available
for delimiting the erodible river corridor: a sustainable approach to managing bank
erosion. River Research and Applications 21, 773–789.
Polvi, L.E., Wohl, E.E., Merritt, D.M., 2011. Geomorphic and process domain controls on
riparian zones in the Colorado Front Range. Geomorphology 125, 504–516.
Rice, S.P., Church, M., 1996. Bed material texture in low order streams on the Queen
Charlotte Islands, British Columbia. Earth Surface Processes and Landforms 21,
1–18.
Rouse, J.C., Hass, R.H., Schell, J.A., Deering, D.W., Harlan, J.C., 1974. Monitoring the Ver-
nal Advancement and Retrogradation (Greenwave Effect) of Natural Vegetation.
NASA/Goddard Flight Space Center, Greenbelt, MD (69 pp.).
Salo, J., Kalliola, R., Häkkinen, I., Mäkinen, Y., Niemelä, P., Puhakka, M., Coley, P.D., 1986.
River dynamics and the diversity of Amazon lowland forest. Nature 322, 254–258.
Shafroth, P.B., Stromberg, J.C., Patten, D.T., 2002. Riparian vegetation response to al-
tered disturbance and stress regimes. Ecological Applications 12, 107–123.
Singer, M.B., Stella, J.C., Dufour, S., Piégay, H., Wilson, R.J.S., Johnstone, L., 2012. Con-
trasting water-uptake and growth responses to drought in co-occurring riparian
tree species. Ecohydrology. http://dx.doi.org/10.1002/eco.1283.
Stoffel, M., Wilford, D.J., 2012. Hydrogeomorphic processes and vegetation: distur-
bance, process histories, dependencies and interactions. Earth Surface Processes
and Landforms 37, 9–22.
USGS, 2011. http://landsat.usgs.gov.
Ward, J.V., Tockner, K., Edwards, P.J., Kollmann, J., Bretschko, G., Gurnell, A.M., Petts, G.E.,
Rossaro, B., 1999. A reference river system for the Alps: the Fiume Tagliamento. Reg-
ulated Rivers: Research & Management 15, 63–75.
Whited, D.C., Lorang, M.S., Harner, M.J., Hauer, F.R., Kimball, J.S., Stanford, J.A., 2007. Cli-
mate, hydrologic disturbance, and succession: drivers of floodplain pattern. Ecolo-
gy 88, 940–953.
Xu, H., 2006. Modification of normalised difference water index (NDWI) to enhance
open water features in remotely sensing imagery. International Journal of Remote
Sensing 27, 3025–3033.
Xu, H.-l, Ye, M., Li, J.-m, 2007. Changes in groundwater levels and the response of nat-
ural vegetation to transfer of water to the lower reaches of the Tarim River. Journal
of Environmental Sciences 19, 1199–1207.
Xu, H.L., Ye, M., Li, J.M., 2009. The ecological characteristics of the riparian vegetation
affected by river overflowing disturbance in the lower Tarim River. Environmental
Geology 58, 1749–1755.
Yang, X., Damen, M.C.J., van Zuidam, R.A., 1999. Satellite remote sensing and GIS for the
analysis of channel migration changes in the active Yellow River Delta, China. In-
ternational Journal of Applied Earth Observation and Geoinformation 1, 146–157.
Zanoni, L., Gurnell, A.M., Drake, N., Surian, N., 2008. Island dynamics in a braided river
from analysis of historical maps and air photographs. River Research and Applica-
tions 24, 1141–1159.