JOURNAL OF GEOPHYSICAL RESEARCH: EARTH SURFACE, VOL. 118, 1–12, doi:10.1002/jgrf.20128, 2013
Geomorphic signatures of deltaic processes and vegetation:
The Ganges-Brahmaputra-Jamuna case study
Paola Passalacqua,1 Stefano Lanzoni,2 Chris Paola,3 and Andrea Rinaldo2,4
Received 11 January 2013; revised 23 July 2013; accepted 9 August 2013.
[1] Deltas are complex ecogeomorphic systems where features such as channels and
interchannel islands are present over a wide range of spatial scales. A quantitative
description of the morphology of deltas is fundamental to address how they react to
changes in climate forcing and human pressure. In particular, it is interesting to ask how
the distributary patterns we observe in coastal areas around the world result from
processes and external forcing acting on deltas, and how such patterns might be related to
deltaic function, vulnerability, and resilience. Using the example of the
Ganges-Brahmaputra-Jamuna Delta, we show that the statistics of island size, shape
factor, aspect ratio, and nearest-edge distance show distinct spatial patterns. Comparison
between regions identified by our statistical analysis and a physiographic zonation of the
delta suggests that the planform extracted from satellite imagery carries the signature of
processes responsible for delta formation and evolution and of vegetation. The tidal
region is characterized by high channel density, small islands, and short nearest-edge
distance (shortest straight-line distance to the nearest water). The results suggest that
regions of the delta characterized by presence of vegetation and active transport of water
and sediment are statistically distinct from less active regions. Further, we perform a
weighted connectivity analysis of the channel patterns based on channel width. The
analysis suggests that channels connecting the upper portion of the delta to the coast do
not play a significant role in the transport of water and sediment.
Citation: Passalacqua, P., S. Lanzoni, C. Paola, and A. Rinaldo (2013), Geomorphic signatures of deltaic processes and
vegetation: The Ganges-Brahmaputra-Jamuna case study, J. Geophys. Res. Earth Surf., 118, doi:10.1002/jgrf.20128.
1. Introduction
sediment through the system. Although the channel network
distributes a range of materials (water, sediments, nutrients, organisms), sediment is critical in terms of creating
and maintaining the delta structure [Edmonds et al., 2011].
Deltaic sediments are also responsible for roughly 45% of
global carbon burial [Hedges and Keil, 1995].
[3] Deltas are threatened by several factors, including
anthropogenic disturbance (e.g., upstream sediment trapping
due to dam construction, sediment mining, navigation structures, accelerated subsidence due to oil or water extraction),
natural subsidence, and eustatic sea level rise [Ericson et al.,
2006]. The response of a deltaic system to these forcings
can be dramatic and result in loss of human lives, economic resources, and environmental services. Yet, under the
right conditions, deltas are resilient, capable of adapting to a
changing environment and recovering from damage caused
by extreme events, such as storms [Paola et al., 2011].
[4] Looking at delta distributary patterns, it is natural
to ask how the system’s morphological organization is the
result of the processes acting on the delta, and how the
distributary patterns might be related to deltaic function, vulnerability, and resilience. Does the spatial structure of the
deltaic network carry information about the dominant processes acting on it? Can we link connectivity of distributary
fluvial patterns, remotely acquired and objectively manipulated, to deltaic ecosystem services? Can a quantitative
[2] Delta distributary networks span a wide range of
spatial and temporal scales: channel widths, for example,
range in scale from hundreds to thousands of meters in the
main network, down to a few meters for drainage distributors within islands; channel migration and avulsions occur
on periods up to thousands of years, while the reworking of channel bed and banks due to flood events can
occur within a single year [e.g., Bristow, 1987; Slingerland
and Smith, 2004; Ashworth et al., 2007; Syvitski, 2008].
Channels are the conductors and distributors of water and
1
Department of Civil, Architectural and Environmental Engineering
and Center for Research in Water Resources, The University of Texas at
Austin, Austin, Texas, USA.
2
Dipartimento IMAGE and International Center for Hydrology ‘Dino
Tonini’, Università di Padova, Padua, Italy.
3
Department of Earth Sciences and St. Anthony Falls Laboratory,
University of Minnesota, Minneapolis, Minnesota, USA.
4
Faculté ENAC, École Polytechnique Fédérale, Lausanne, Switzerland.
Corresponding author: P. Passalacqua, Department of Civil, Architectural and Environmental Engineering and Center for Research in Water
Resources, The University of Texas at Austin, 301 E. Dean Keeton St.
STOP C1700, Austin, TX 78712-2100, USA. ([email protected])
©2013. American Geophysical Union. All Rights Reserved.
2169-9003/13/10.1002/jgrf.20128
1
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
et al., 2002; Soille and Grazzini, 2007; Grazzini et al., 2010]
by using mathematical morphology and other approaches
employed in the extraction of linear features from images
[e.g., Quackenbush, 2004]. However, these techniques usually require manual intervention and post-processing due to
the complexity and variability of channel patterns. This is
particularly true in coastal areas, where low gradients and the
presence of features such as sediment plumes make it challenging to separate channels from other water bodies within
the image. Recent efforts have focused on the extraction
of the shoreline [Shaw et al., 2008; Geleynse et al., 2012],
rather than the entire channel network.
[7] In this paper, we propose a statistically based analysis framework for deltaic networks extracted from satellite
imagery. Our goal is to identify key metrics and attributes
of the network (e.g., island geometry, channel width), analyze their statistical behavior, and explore their potential
linkages to processes acting on the delta. In particular, we
are interested in understanding whether the spatial structure
of the deltaic network carries any signature of the processes responsible for delta formation and evolution and
whether a statistical analysis of deltaic network metrics
may highlight scaling breaks and characteristic scales of
delta forming processes, vegetation type, and anthropogenic
modifications. Linkages between network statistics and processes have been previously explored in alluvial rivers
[Rodriguez-Iturbe and Rinaldo, 1997] and tidal channel networks [Fagherazzi et al., 1999; Marani et al., 2003]. We
apply our analysis framework to the Ganges-BrahmaputraJamuna (GBJ) Delta, one of the largest and most densely
populated deltas in the world.
[8] The paper is organized as follows: section 2 describes
the proposed analysis framework; in section 3, we describe
the GBJ Delta; the application of the analysis method is
presented in section 4; we discuss the analysis results in
section 5; and we present conclusions in section 6.
Figure 1. Landsat image of the GBJ Delta. Data specifications: Landsat GeoCover TM 1990 Edition Mosaics, tiles N–
45–20 and N–46–20; spatial resolution 28.5 m; spectral TM
bands: 7 (mid-infrared), 4 (near-infrared), 2 (visible green),
here represented in RGB false color. Source for this data set
is the Global Land Cover Facility, www.landcover.org. The
physiographic regions active, inactive, tidal are based on a
geological map of Bangladesh developed by the Geological
Survey of Bangladesh [Alam et al., 1990] and by the USGS
[Persits et al., 2001].
reading of the network help us understand and predict the
future of these fascinating and critical coastal systems?
[5] Among the early quantitative studies of delta networks
was that of Smart and Moruzzi [1972], who focused on
topology and proposed representing the deltaic network as
a directed graph and analyzing various functions of vertex
and link number. Among recent efforts, Syvitski [2005] and
Syvitski and Saito [2007] illustrated empirically the scaling
of the number of distributary channels with respect to river
length and delta gradient, and Edmonds et al. [2011] proposed five metrics (fractal dimension, distribution of island
size, nearest-edge distance, synthetic distribution of fluxes
at the shoreline, and nourishment area) for describing deltas
and comparing natural and experimental systems.
[6] The recent availability of satellite imagery over much
of the Earth (Figure 1 shows an example for the GangesBrahmaputra-Jamuna Delta) [Syvitski, 2005] has greatly
improved the quantitative analysis of geomorphic features.
Examples of features mapped from satellite imagery include:
number and size of distributary channels [Syvitski, 2005;
Syvitski and Saito, 2007], container valleys, floodplain
depressions, oxbow lakes [Syvitski et al., 2012], floodplain
channel network and morphology [Trigg et al., 2012], shoreline erosion/accretion patterns [Aly et al., 2012], and flood
area and volume [Rakwatin et al., 2013]. In terms of automatic feature extraction, the detection of roads and river networks has been addressed [e.g., Liu et al., 2001; Dillabaugh
2. Background: Statistical Analysis to Identify
and Interpret Delta-Forming Processes
[9] Our approach is based on the analysis of satellite imagery. We acknowledge the importance of the third
dimension (topography and bathymetry), as well as of the
fourth dimension (time). But at present, planform data are
far more readily available than high-resolution topography,
and so we wish to determine how much information can
be obtained from planform alone. Also, the generally low
relief of deltas means that suitably accurate elevation data,
e.g., submeter resolution lidar data sets, are typically available only for areas of small extent, rather than the scale of a
deltaic system such as the GBJ Delta.
[10] Likewise, we focus on a time snapshot of the GBJ
Delta. We acknowledge that channels are hydrodynamic and
the information extracted from satellite imagery changes due
to several factors including tides and seasonal changes in
vegetation cover. While ignoring temporal information prevents us from capturing short-term dynamical changes in the
system, we focus on the statistics of the network structure,
particularly of island geometry. Apart from extreme floods,
variations due to temporal fluctuations in water level would
not significantly change the statistical distribution of island
geometry over the delta.
2
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
bars, strongly worked by channel processes, while the latter are simply channel-bounded land masses. We lump them
together here for simplicity but consider this a point worth
further study.
[16] Metrics of interest here include island area, shape
factor, and aspect ratio. The island area reflects channel connectivity: areas of the delta characterized by small islands,
for example, represent highly connected parts of the network. The island shape factor ˇ , here defined as the ratio
pof
the wetted perimeter P to the square root of the area A,
quantifies the shape of the island and the degree of drainage
within the island. The wetted perimeter used to evaluate ˇ ,
in fact, includes also the overall length of the channels dissecting the island. As such, the shape factor ˇ is a measure
of relative boundary roughness [Wolinsky et al., 2010]. The
island aspect ratio is given by the ratio of the principal axes
(ratio of the major axis to the minor axis of the ellipse with
the same normalized second central moments), and quantifies the degree of elongation of the islands. If = 1, the
shape is equant, if >> 1 the shape is elongated.
2.2.2. Nearest-Edge Distance
[17] The nearest-edge distance L is defined as the shortest
straight-line distance from any land pixel to the nearest water
(channelized or unchannelized) [Edmonds et al., 2011]. As
such, the statistical analysis of nearest-edge distance quantifies spatial variations in distributary channel density. Areas
of the delta with relatively short nearest-edge distance, for
example, indicate the availability of nearby sources of water.
2.2.3. Channel Width
[18] Channel width can be computed along each extracted
channel within the deltaic network as the local shortest distance between the channel edges. Assuming the width has
a consistent scaling to landscape-forming discharge [e.g.,
Leopold et al., 1993], the analysis of mean channel width for
channel reach gives information about the water distribution
along the system.
[19] In fluvial basins the landscape forming discharge Q
(and, hence, the cross section geometry) is usually found
to be proportional to the total contributing area A, provided
the basin dimension does not exceed a scale characteristic of the heterogeneity of meaningful spatial patterns of
intense rainfall. This proportionality commonly holds for
several orders of magnitude. In tidal channels it is the tidal
prism that determines the cross section geometry, relating
the total water volume entering the channel watershed during a characteristic tidal cycle (i.e., spring tide) to the size of
a given cross section [Rinaldo et al., 1999b; D’Alpaos et al.,
2010]. Consequently, the width of fluvially dominated channels stays constant over length scales of hundreds to thousands of widths [Rodriguez-Iturbe and Rinaldo, 1997], while
tidally dominated channels show strong variations in space
[Leopold et al., 1993; Marani et al., 2003]. In systems like
the GBJ Delta, the scenario is much more complex, owing
to the different nature of dominant landscape forming processes likely active in the various delta regions (e.g., fluvial
discharge, tidal currents, wind waves, storm surges). These
differences should however be revealed by an analysis of the
channel width of the type pursued in Rinaldo et al. [1999b].
Indeed, the power law exponents that relate channel width,
flow depth, and mean velocity to the formative discharge are
different for terrestrial rivers and tidal estuaries [Myrick and
Leopold, 1963; Sassi et al., 2012].
[11] The first analysis step consists of extracting the channel network and validating it. We then define key metrics
and descriptors, taking into account the intrinsic distributary nature of deltaic networks and building on the previous work discussed above. Finally, we analyze the various
quantities statistically.
2.1. Channel Network Extraction
From Satellite Imagery
[12] In order to extract the channel network from satellite imagery, we use a relatively simple approach: taking
advantage of tools available in ArcGIS (as well as ENVI and
other software packages), satellite imagery can be exploited
through unsupervised and supervised classifications. With
an unsupervised classification, classes are identified based
on the spectral signature of the image such that each pixel
within the image can be assigned to a class, e.g., deep water,
vegetation cover, and soil. With a supervised classification,
classes can be clustered in two groups, water and land, such
that water pixels can be isolated from the rest of the image.
The channel network obtained from the unsupervised and
supervised classifications is thus a wet map of the submerged
pixels (map of the water surface). An additional product of
the binary classification of water and land pixels is the map
of interchannel islands, here defined as channel-bounded
land masses.
[13] The binary classification into water and land pixels can result in detection of spurious features and isolated
water bodies that are not necessarily part of the deltaic channel network, requiring further manual intervention. Once the
water pixels are identified, if desired, channel pixels can
be separated from ocean pixels using shoreline extraction
techniques [Shaw et al., 2008; Geleynse et al., 2012].
2.2. Metrics and Deltaic Channel Network Descriptors
[14] Once the channel network is extracted from the satellite image, key metrics representative of the deltaic network
can be identified and computed. A variety of metrics is available for the analysis of river networks [Rodriguez-Iturbe
and Rinaldo, 1997], some of which have been successfully extended to tidal networks [Fagherazzi et al., 1999;
Rinaldo et al., 1999a, 1999b; Marani et al., 2003], and
submarine tributary channel networks [Straub et al., 2007].
However, these metrics are not readily applicable to distributary networks since attributes such as upstream length and
elongation cannot be uniquely computed. In addition, while
the total planform fraction occupied by water (wet fraction) is vanishingly small in most tributary networks, this is
not the case for most deltas. Metrics specific to distributary
systems are thus needed [Edmonds et al., 2011].
2.2.1. Island Area, Shape Factor, and Aspect Ratio
[15] In many deltas, including the GBJ Delta, the channel network includes confluences and bifurcations, such
that nearly the whole deltaic landmass comprises islands
bounded by channels. Islands are among the best descriptors of a deltaic network, since they reflect the morphology
of the channels and the relationship among neighboring
channels. In process terms, we believe there is a difference between islands whose length scale is comparable to
or less than the width of the bounding channels, and islands
whose length scale is much greater than that of the bounding channels. The former might be called ‘true’ islands and
3
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
[20] Looking at the deltaic network as a graph, the network can be seen as composed of nodes (apex, junctions,
bifurcations, outlets) and links connecting the nodes. Thinking of water and sediment as a form of information transmitted between nodes, an estimate of mean channel width
per channel link quantifies link strength and allows more
substantive analysis of network connectivity. If the channel
carries a significant amount of water (and hence has a relatively large width), that connection should be present regardless of flow variations during the year. Conversely, relatively
narrow channels are less likely to transport water during
drought and low discharge periods. We can thus think about
channel width as a measure of the strength of connectivity among network nodes. Our use of the term “strength” is
analogous to the “weight” used in mathematical representations of networks as an adjacency matrix when a connection
between two nodes is not represented in binary fashion (1 if
two nodes are connected, 0 otherwise) but with a real number that represents how strong that connection is [Newman,
2010]. The strength of links within a network thus provides important information in network analysis: in the case
of deltas, water and sediment can be transferred between
two nodes as long as a connection (link) exists; however,
changes in external forcing (e.g., discharge, wind, tides)
and disturbances acting on the system (e.g., deposition) can
easily lead to the removal of weak network links, resulting in lack of information transfer among network nodes.
While here we focus on water and sediment as information,
and channel width as a measure of connectivity strength,
it is interesting to think about similar weights to quantify
connectivity strength in terms of, for example, dissolved
nutrients, or organisms. While we only have channel width
estimations available from satellite imagery, we note that
the link strength distribution could change depending on the
quantity analyzed.
2.2.4. Oxbow Density
[21] Channel avulsions and cutoffs result in abandoned
channels and oxbow lakes. The presence and spatial density
of these features may indicate the age of a given area within
the delta and/or the frequency of channel migration through
time. At present, without independent age control on the
oxbows, we cannot separate these effects, but we consider it
worth mapping oxbow density nonetheless, for example, to
investigate the possibility that oxbow density may be related
to vegetation and land-use patterns.
Figure 2. Map of the GBJ Delta. The map shows a
zonation of the delta into four main geomorphic regions:
active, tidally active, mature, moribund. The Sundarbans
(mangrove forest) is located within the tidally active
area. Map source: National Encyclopedia of Bangladesh
(www.banglapedia.org).
are scale free, i.e., the statistical characteristics of the system do not change across the range of scales over which the
power law applies. Lack of power law behavior, or breaks or
thresholds in the power law, highlight characteristic scales
of processes acting within the system. Here the presence or
absence of scale-free behavior in deltaic networks is relevant
as it could suggest ranges of scales over which processes
give rise to (internally) statistically similar morphology, such
as channels and islands.
3. Study Area
[24] The GBJ Delta, located in Bangladesh and eastern
India, includes the Ganges, Brahmaputra, Jamuna, Padma,
and Meghna Rivers [e.g., Coleman, 1969]. With an area of
100, 000 km2 , the GBJ Delta is one of the largest in the
world. The rivers discharge nearly 1.7 105 m3 /s to the Bay
of Bengal and transport about 106 tons/day of suspended
sediment during flood events [Coleman, 1969] for a total of
about 109 tons/year [Goodbred and Kuehl, 2000a]. Channel
width ranges from the scale of meters within islands, to the
scale of kilometers in the main channels (Figure 1). Along
the western coast of the Delta lie the Sundarbans, a tidal
halophytic mangrove forest (location shown in Figure 1 with
a white dotted boundary). The Sundarbans and most of the
delta experience significant tidal influence as the system is
subject to a macrotidal regime with a semidiurnal amplitude
of 4.7 m [Goodbred and Kuehl, 2000a]. The extent of the
tidally influenced region is shown in Figure 1 (white solid
boundary). In the same figure, the northern portion of the
delta is divided into active (yellow dotted boundary) and
2.3. Statistical Analysis
[22] The metrics just described should vary among deltas
and could also vary within the same system, especially if it
is large relative to the characteristic channel size. While statistical descriptors such as the mean and the variance may
help to summarize the salient characteristics of a system,
they are not sufficient to describe variables that vary over a
wide range of scales [Newman, 2005]. Analysis of the probability density function (PDF) or the cumulative distribution
function (CDF) is helpful in this respect, particularly in identifying statistical changes in the behavior of the system and
corresponding scales of change.
[23] In network theory, it is quite common to analyze the
distribution of network metrics and the possible presence
of power law behavior [e.g., Newman, 2005; Clauset et al.,
2009]. Power laws suggest that certain network properties
4
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
(a)
(b)
(c)
(d)
through a comparison with Google Earth imagery. Links
between oxbows and nearby channels were manually added
where Google Earth imagery showed a connection between
the oxbow and the channel (Figures 3c and 3d). Repeating
these operations on the whole delta converts the wet map to
the channel network shown in Figure 4.
[26] The extracted channel network shows how channel
morphology and channel density vary spatially within the
GBJ Delta. Channel density is particularly high near the
coast and within the mangrove forest (Figure 1). The northwestern region is characterized by a large number of oxbow
lakes. Through analysis of Google Earth imagery, we found
that 65% of the oxbow lakes (80 out of 120) in the northwestern portion of the delta are connected to channels. These are
likely relict fluvial channels whose lower reaches have been
tidally reworked, as suggested by Fagherazzi [2008]. The
number of oxbow lakes present in the northeastern portion
of the delta is much smaller (about 10).
Figure 3. Example of channel extraction from satellite
imagery. Several oxbow lakes are present in the upper portion of the GBJ Delta. (a) The majority of the oxbows
(about 80 out of 120) are connected to channels. An unsupervised classification followed by a supervised classification
performed in ArcGIS, allows the extraction of the channel network. Disrupted channels, spurious water pixels, and
isolated water sources are present in the (b) extracted network. By inspection of Google Earth imagery, (c) oxbows
connected to nearby channels and spurious pixels can be
identified. By manual intervention, (d) spurious pixels are
removed and links between oxbows and nearby channels
traced where Google Earth imagery shows a connection
between the oxbow and the channel.
4.2. Island Size and Geometry Analysis
[27] The map of interchannel islands of the GBJ Delta
(Figure 5) shows that islands of large area are located
within the upper portion of the delta, while smaller islands
are mainly located close to the coast, particularly within the
mangrove forest, and along the Ganges and Padma Rivers,
and the main branches of the Meghna River. The observed
CDF of island area (Figure 6) indicates a relatively well
identified regime where power law behavior (and log-log
linear distribution) with exponent ˛ = 1.9 for island area
88°E
inactive (red dotted boundary) regions. The physiographic
zonation of the GBJ Delta into three regions (tidal - near
the coast, including the Sundarbans, active - corresponding
to the area occupied by the major rivers, and inactive - the
upper western portion of the delta) is based on a geological map of Bangladesh developed by the Geological Survey
of Bangladesh [Alam et al., 1990] and the report and digital data released by the U.S. Geological Survey [Persits
et al., 2001]. The U.S. Geological Survey digital data covers only the Bangladesh portion of the GBJ Delta; we have
extended the region boundaries to the West Bengal (India)
portion based on an additional physiographic map of the
whole delta distributed by the National Encyclopedia of
Bangladesh (Figure 2).
89°
90°
91°
24°
N
23°
4. Analysis of the GBJ Delta Network
22°
4.1. Channel Network Extraction From Satellite
Imagery and Network Analysis
[25] For the present study, we use Orthorectified Landsat Thematic Mapper Mosaics available at the resolution
of 28.5 m (source: Global Land Cover Facility, www.
landcover.org), shown in Figures 1 and 3a. The images have
three spectral TM bands (7 mid-infrared, 4 near-infrared,
and 2 visible green), represented in RGB false color in
Figure 1. By applying unsupervised and supervised classification techniques, as described in section 2, we obtained
a noisy wet map, as shown in Figure 3b. The wet map
requires manual intervention due to the presence of disrupted channels, spurious water pixels, and isolated water
bodies. These spurious features were identified and removed
N
0
25 Km
100
Figure 4. Channel network of the GBJ Delta extracted
from the satellite image in Figure 1. The channel network
map shows how channel morphology and channel density
largely vary among different portions of the delta. Drainage
density is particularly large within the tidally dominated
region of the delta. The northwestern portion of the delta
(inactive) is characterized by a large number of oxbow lakes.
The number of oxbow lakes present in the northeastern portion of the delta is instead much smaller (about 10). The gray
boundaries represent the zonation shown in Figure 1.
5
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
88°E
89°
90°
Table 1. Estimated Power Law Parameters and p-Values
91°
log(A)
A
ˇ
L
3.39
24°
N
N
0
˛O
ntail
p
1.96 107 ˙ 4.2 106
13.7 ˙ 2.3
2.7 ˙ 0.6
1.054 103 ˙ 72.2
1.9 ˙ 0.05
5.2 ˙ 0.6
3.7 ˙ 0.36
2.96 ˙ 0.1
353 ˙ 50
138 ˙ 206
461 ˙ 138
403 ˙ 48
0.78
0.64
0.01
0.43
Clauset et al. [2009]. Table 1 shows the estimated power law
model parameters (lower bound xO min , exponent ˛O , and number of data points in the tail ntail ), their estimated uncertainty,
and the p-value. The total number of islands in the delta
(at the resolution of our satellite image) is 1291, of which
938 have area A < Amin and are thus located outside the
power law regime. These sub-power law islands are numerous, but quite small, covering only 7% of the exposed land
area. The map of the logarithm of island area normalized by
Amin (Figure 7) shows that smaller islands outside the power
law regime are predominantly located within the mangrove
forest and along the Ganges and Padma Rivers, and the main
branches of the Meghna River. This result suggests that these
portions of the GBJ Delta behave differently than the rest
of the system. The superimposed physiographic boundaries
indicate that the smallest islands are mainly located within
the tidal region, particularly within the mangrove forest. The
break in the power law emerging from the observed CDF of
island area (Figure 6) can then be interpreted as a distinction
between the part of the delta near the coast and the upper
part, particularly the inactive region.
[28] Islands vary not only in size, but also in shape
(Figure 5). This can be quantified by looking at the statistical distribution of the shape factor ˇ . It is worth noting
23°
22°
xO min
Metric
9.73
25 Km
100
Figure 5. Channel network of the GBJ Delta and islands
(colormap based on the logarithm of island area). Islands
of larger area are located within the northern portion of the
delta, while smaller islands are located near the coast and
along the main rivers (Ganges, Padma, Meghna).
A > Amin = 1.96 107 m2 holds. The goodness of fit
between the data and the model is given by the p value which
quantifies whether the difference between the data and the
hypothesized model can be attributed to statistical fluctuations (resulting in a p-value close to 1), or if the difference
is due to a lack of fit of the model to the data (resulting
in a small p-value) [Clauset et al., 2009]. The p-value of
the fitted power law of island area is 0.78, implying that a
power law distribution is not rejected at a conservative significance level (e.g., p > ˛ = 0.1 with 10% representing an
already conservative significance level). The analysis was
performed with the approach and tools made available by
88°E
89°
90°
91°
log(A/Amin)
2.44
-3.90
24°
N
100
23°
Pr(A ≥ a)
10-1
α = 1.9
10-2
22°
10-3
N
0
10-4 3
10
105
107
25 Km
100
109
a [m2]
Figure 7. Channel network of the GBJ Delta and map of
logarithm of island area normalized by Amin = 1.96 107 m2 .
The islands outside the power law regime shown in Figure 6
are mainly located near the coast and along the main
rivers (Ganges, Padma, Meghna). The superimposed physiographic boundaries (white) show that larger islands are
mainly located within the inactive portion of the delta.
Figure 6. Empirical distribution of island area and fitted
power law distribution. The power law model has exponent
˛ = 1.9 and lower bound (to the power law behavior) Amin =
1.96 107 m2 . The p-value for the fitted power law model
is 0.78.
6
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
100
100
α = 3.7
10-1
α = 5.2
Pr(γ ≥ g)
Pr(β ≥ b)
10-1
10-2
10-2
10-3
10-3
10-4 0
10
101
10-4 0
10
102
101
102
g [-]
b [-]
Figure 8. Empirical distribution of island shape factor
(defined as ˇ = P/A0.5 , where P is perimeter and A is area)
and fitted power law distribution. The power law model
has ˛ = 5.2 and lower bound (to the power law behavior)
ˇmin = 13.7. The p-value for the fitted power law model
is 0.63.
Figure 10. Empirical distribution of island aspect ratio
and fitted power law distribution. The fitted power law
model has ˛ = 3.7 and lower bound (to the power law behavior) min = 2.7. The p-value for the fitted power law model
is 0.01.
that the perimeter P is larger not only for islands of larger
size, but also for islands with a significant number of intraisland channels. A small but highly channelized island will
thus be characterized by a large shape factor. We note that
the extraction of the channel network depends on the image
resolution; thus, the metrics here analyzed are resolutiondependent. Part of these intraisland channels, characteristic
of the islands within the mangrove forest, could in fact be
through going, resulting in islands of even smaller size. The
resolution-dependency of this analysis is unavoidable and
the mangrove forest too thick for assessing the continuity of
the intraisland channels from Google Earth imagery, as we
did for assessing the connectivity of the oxbow lakes.
[29] The observed CDF of island shape factor ˇ (Figure 8)
shows that a power law distribution with exponent ˛ = 5.2
holds for ˇ > ˇmin = 13.7 (Table 1). The exponent is
quite large compared to the common range 2–3 [Clauset
et al., 2009]. The p-value of the fitted power law model
is 0.63, implying that the model is not rejected at a conservative significance level. The map of the logarithm of
88°E
89°
90°
91°
log(β/βmin)
88°E
89°
90°
91°
0.66
log(γ/γmin)
-0.50
0.81
24°
N
-0.42
24°
N
23°
23°
22°
N
22°
0
25 Km
N
100
0
Figure 9. Channel network of the GBJ Delta and map of
logarithm of island shape factor normalized by ˇmin = 13.7.
Islands with small shape factor are mainly located near the
coast and along the main rivers (Ganges, Padma, Meghna).
The superimposed physiographic boundaries show that
islands with large shape factor (rougher) are located within
the inactive portion of the delta.
25 Km
100
Figure 11. Channel network of the GBJ Delta and map of
island aspect ratio normalized by min = 2.7. The majority of
the elongated islands is located near the coast and along the
main rivers (Ganges, Padma, Meghna), but the distribution
of aspect ratio within the delta is heterogeneous.
7
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
88°E
89°
90°
91°
log(L/Lmin)
1.10
-1.57
24°
N
23°
22°
N
0
25 Km
100
Figure 12. Channel network of the GBJ Delta and map of
the logarithm of nearest-edge distance L. The nearest-edge
distance is calculated as the shortest straight-line distance
from the nearest source of water (channelized or unchannelized) from any land pixel within the delta. The smallest
values of nearest-edge distance are located near the coast, in
particular within the mangrove forest.
Figure 14. Channel network of the GBJ Delta and map of
the logarithm of nearest-edge distance normalized by Lmin =
1.054 103 m. The superimposed physiographic boundaries
show that islands with short nearest-edge distance are mainly
located within the tidal portion of the delta, particularly
within the mangrove forest.
island shape factor normalized by ˇmin (Figure 9) shows
that islands outside the power law regime are located near
the coast and along the Ganges and Padma Rivers, and the
main branches of the Meghna River, containing 1153 (out of
a total of 1291) relatively small islands. The superimposed
physiographic boundaries indicate that these regions roughly
coincide with the tidal portion, particularly the mangrove
forest, and the active portion of the delta. Islands within
these regions thus tend to have smaller ˇ values, resulting in
less rough shapes than the northwestern inactive portion of
the delta.
[30] The observed CDF of island aspect ratio (Figure 10) indicates a relatively well identified regime
where a power law distribution with exponent ˛ = 3.7 above
threshold equal to min = 2.7 (Table 1) may hold. The pvalue is 0.01; the power law model fit is thus not as robust
as for other metrics analyzed here (the model would not be
rejected only at significance level 1% or lower). The map of
the logarithm of the island aspect ratio normalized by min
(Figure 11) shows that although the majority of the elongated
islands are located near the coast and along the Ganges and
Padma Rivers, and the main branches of the Meghna River
(except for 6 islands, out of 461), the spatial distribution of
the island aspect ratio is more heterogeneous than island area
and island shape factor. In particular, the mangrove forest
appears to include both round and more elongated islands.
This is also supported by the fact that a sharp break is not
observed in the CDF of island aspect ratio (Figure 10).
100
α = 2.96
10-1
10-2
4.3. Nearest-Edge Distance
[31] The map of nearest-edge distance L (Figure 12)
shows that islands with the smallest values of L are located
near the coast, particularly within the Sundarbans, suggesting that the mangrove forest has the highest channel density,
smallest island area, and smallest nearest-edge distance. The
observed CDF of nearest-edge distance (maximum value
of L per island) (Figure 13) shows reasonable power law
behavior with exponent ˛ = 2.96 above threshold Lmin =
1.054 103 m (Table 1). The p-value of the model is 0.42,
implying that the power law fit is not rejected at a conservative significance level. The map of the logarithm of
10-3
10-4 1
10
102
103
104
105
Figure 13. Empirical distribution of nearest edge distance
(maximum value per island) and fitted power law distribution. The power law model has ˛ = 2.96 and lower bound (to
the power law behavior) Lmin = 1.054 103 m. The p-value
for the fitted power law model is 0.42.
8
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
88°E
89°
90°
91°
the relationships proposed by Wilkerson and Parker [2011]
for sand-bed rivers in equilibrium. Assuming a width to
depth ratio of 20, we estimate the bankfull Shields stress as
* 0.045, which is close to typical values of the critical
Shields stress *c , supporting the idea that sand transport in
the small channels is weak.
log(A)
10.57
4.40
24°
N
4.5. Oxbow Analysis
[33] The extracted channel network (Figure 4) is characterized by the presence of a large number of oxbow lakes
concentrated mainly in the northwestern part of the delta.
The spatial distribution of the number of oxbows per island
normalized by its maximum value (Figure 16) shows that
all the oxbow lakes are located far from the coast and are
concentrated in the northwestern region, while the northeastern region has few oxbows (< 6 oxbows per island), and
no oxbows are present near the coast and along the Ganges
and Padma Rivers, and the main branches of the Meghna
River. As seen from the superimposed physiographic boundaries, the area where oxbows are present coincides with the
inactive region of the delta. The greatest number of oxbow
lakes (40) is observed in the only area in the northwestern part of the delta that is disconnected from the main
channels, being drained only by the weak links previously
discussed. This suggests a connection between the lack of
channels able to carry significant amounts of water and sediment, and the presence of abandoned channels and oxbows.
These relict landforms presumably formed when this region
of the delta was more active. The likelihood of the oxbows
to be preserved should increase with distance from a chan-
23°
22°
N
0
25 Km
100
Figure 15. Channel network of the GBJ Delta with channels of width 57 m and islands (colormap based on the
logarithm of island area). The effect of eliminating the smallest channels of the network is that the upstream portion of
the delta behaves as a single large island. This is because the
channels connecting the coastline and the upstream portion
of the delta are predominantly narrow.
nearest-edge distance normalized by Lmin (Figure 14) shows
that the 888 islands outside the power law regime (out of
1291 total), are mostly located near the coast and along the
Ganges and Padma Rivers, and the main branches of the
Meghna River, confirming that these regions of the delta
exhibit the smallest values of L. The superimposed physiographic boundaries indicate that these regions coincide with
the tidal portion, particularly the mangrove forest, and part
of the active portion of the delta.
88°E
89°
90°
91°
O/Omax
1
0
24°
N
4.4. Width Analysis
[32] To investigate the relation between network structure and link strength in the GBJ Delta, we analyzed the
behavior of the system as the weakest links (i.e., the narrowest channels) are removed from the network. This is a
way of quantifying the sensitivity of network metrics to the
least important (and, in our case, least certain) links. Choosing 57 m (twice the pixel resolution) as the threshold width
below which channels are removed, we analyzed the resulting configuration of channels and islands (Figure 15). This
map shows that, as the weakest links within the network
are removed, the upstream portion of the delta behaves as
a single large island. This is because channels connecting
the northwestern part of the delta to the coastline are mostly
small, i.e., weak network links. Flow through the northwestern part of the delta occurs via weak connections, suggesting
lack of significant water and sediment transport. As seen
from the superimposed physiographic boundaries, the inactive region of the delta is included in this portion of weak
transport. These narrow channels may be active only during relatively large floods. A rough estimate of the bed shear
stress * in a 57 m wide channel can be obtained by using
23°
22°
N
0
25 Km
100
Figure 16. Channel network of the GBJ Delta and map
of number of oxbows per island normalized by its maximum value. All the oxbows are located away from the
coast. Based on the superimposed physiographic boundaries,
islands in the active part of the delta have a small number
of oxbows, while the majority of the oxbows are located in
the inactive portion. The island with the largest number of
oxbows is the only area in the upper part of the delta that is
disconnected from the main channels.
9
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
inactive channels. While less important for the overall morphology of the GBJ Delta [Goodbred and Kuehl, 2000a,
2000b; Goodbred et al., 2003], subgrid-scale features may
be needed for a complete quantification of human alterations
within the system.
[39] As noted earlier, the satellite image analyzed here
is only a snapshot of the water surface elevation of the
GBJ Delta. The channel network obtained from the unsupervised and supervised classifications is a wet map of the
submerged pixels (map of the water surface). We inferred the
channel-planform pattern from the wet map, but noting that
channels are hydrodynamic features evolving through time.
Monsoonal flooding, unusual tides, and storms, could substantially modify the wet map. Although the effect of water
level fluctuations on the results presented here could be
assessed only through analysis of multiple images through
time, we believe that the main findings of this study would
not be fundamentally altered by water level fluctuations over
the delta. We consider this an interesting topic for future
analysis once comprehensive image data for various wetting
conditions are available.
[40] One piece of information excluded from our analysis
is the direction that characterizes the delta network links and
thus directions of transport through the network. A method
based on an analogy with electric circuits was proposed by
Feola [2006] and could be used to analyze the behavior of
the system as a directed graph. Also, the correlation among
surface statistics, as the ones analyzed here, and subsurface
statistics related to the stratigraphic record [Goodbred and
Kuehl, 2000a, 2000b; Goodbred et al., 2003] could certainly
highlight other characteristic scales of the dynamical behavior of the system. Finally, it would be useful to compare
the GBJ results to a delta, such as the Po or the Danube,
that is smaller and more heavily anthropogenically modified. We believe that comparison could highlight thresholds
in terms of total sediment flux or morphology change rate
above which human modifications affect natural morphology. While in the GBJ Delta the natural processes are very
strong, and anthropogenic disturbance cannot be captured
at the resolution of the satellite imagery analyzed here, this
style of analysis may also indicate anthropogenic influences
in other deltas.
[41] While we focus here on the GBJ Delta, the analysis framework we propose is portable to any other deltaic
system, given the availability of satellite imagery. It is particularly relevant to large systems, such as the Niger or the
Mekong, where direct field observations are hard to make
over the entire delta. The metrics proposed here will not
always show power law behavior, particularly in smaller systems where network properties may not cover several orders
of magnitude. Power laws are one of many possible distributions. A network-scale statistical analysis of island and
channel properties, such as the one proposed here, provides a
useful quantitative description of delta morphology whether
the observed distributions are power law or not.
nel, and thus island area; the majority of the oxbows are
in fact located in the largest islands within the delta. A
combination of trimming (e.g., filling/opening of small
channels) and occupation and stabilization of relatively inactive landforms seems to be one important signature of human
effects in the GBJ Delta.
5. Discussion
[34] The statistical analysis of metrics and descriptors of
the channel network extracted from satellite imagery indicates that island area, shape factor, aspect ratio, and nearestedge distance show power law behavior above a threshold
value. The comparison with a physiographic zonation of
the GBJ Delta suggests that portions of the delta characterized by active flow and transport differ substantially
from the upper, less active, region, and are characterized
by high channel density, small and highly drained islands,
and short distances to the nearest water. This is observed in
particular within the Sundarbans, suggesting a link between
vegetation, especially mangroves, channel morphology, and
linkages within the network. Such a high channel density
is likely due to tidal drainage, as suggested by Fagherazzi
[2008]. We consider this a point worth further investigation
in the future.
[35] The uncertainty in the estimated power law lower
bounds does not affect the nature of the above results. This
was verified by creating maps of the network metrics for
a range of plausible values of Amin , ˇmin , min , and Lmin
(Table 1).
[36] Extraction of the whole network of the GBJ Delta
also allowed us to identify narrow channels connecting the
coast to the upstream northwestern part of the delta. The
high-density channel network characterizing a relatively narrow belt parallel to the coast is essentially dominated by tides
and by the mutual interplay of sediment fluxes and vegetation (mangrove) dynamics. The northwestern portion of the
delta is instead characterized by numerous oxbow lakes representing the current, low-activity stage of development fed
only by small, muddy, and sinuous ephemeral distributaries
of the Ganges. This surface morphology covers extensive
subsurface channel sands representing an earlier constructional stage of this area by the main stem Ganges [Goodbred
and Kuehl, 2000b].
[37] The weighted analysis of network connectivity, based
on the extracted channel width at each network link, highlights that many of the channels that extend from the coast
to the upper part of the delta are weak and do not seem
to play a significant role in terms of water and sediment
delivery, suggesting lack of significant transport in most
of the upper portion of the delta. In particular, the highest number of oxbow lakes was found within the only
island in the upper part of the delta not connected to the
main channels, showing the link between lack of strong
connectivity and the presence of abandoned channels and
oxbow lakes.
[38] While satellite imagery provides a unique, synoptic
way of analyzing a large system such as the GBJ Delta, it has
the limitation of not capturing features and thus processes
acting at scales smaller than the image resolution. This
is especially relevant to identifying human modifications,
which are likely to be most important for small, relatively
6. Conclusions
[42] In this work we have proposed a statistical analysis
framework of delta networks extracted from remotely sensed
data. We have identified several descriptors of the channel
network and analyzed them statistically. We have applied
10
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
the proposed statistical framework to the deltaic network
of the GBJ Delta. The main conclusions of this work are
the following:
[43] 1. Island area, shape factor, aspect ratio, and nearestedge distance vary spatially in different regions of the GBJ
Delta. In particular, the statistical behavior of these metrics
distinguishes the coastal region and the portion along the
main rivers (Ganges, Padma, Meghna) from the upper region
of the delta;
[44] 2. A comparison between our statistically based
zonation and preexisting qualitative physiographic maps of
the GBJ Delta shows good correspondence between the two
zonations. In particular, the region near the coast corresponds to the previously mapped tidally influenced portion
of the GBJ Delta, while the portion along the main rivers
is part of the previously mapped active region. These are
statistically different from the northwestern portion of the
delta previously classified as inactive and characterized by
the presence of numerous oxbow lakes;
[45] 3. The tidal region, particularly the mangrove forest, and the portion along the main rivers are characterized
by small islands, small shape factor, short nearest-edge distance, and absence of oxbows. Islands within this region tend
to have smooth boundaries compared to islands in the inactive portion of the delta. Some (about half) of the islands
within the Sundarbans are elongated;
[46] 4. Islands within the previously mapped inactive
region tend to have large area, large shape factor (resulting
in rough boundaries), and long nearest-edge distance. The
area is characterized by frequent oxbow lakes, suggesting a
link between the lack of strong connectivity to the rest of the
delta and the presence of abandoned channels and oxbow
lakes;
[47] 5. The active region of the delta presents characteristics between the inactive, and tidal-dominated regions.
The area along the main rivers (Ganges, Padma, Meghna)
behaves like the region near the coast, while the rest of
the active region is characterized by medium-sized islands,
moderate nearest-edge distance, and moderate shape factor,
indicating island shapes that are not as rough as in the inactive region. Oxbow lakes are present but in much smaller
number than in the inactive region;
[48] 6. While the shape factor has a clear spatial distribution within the delta, this is not the case for the island aspect
ratio, whose distribution is much more heterogeneous within
the delta and the power law fit less robust;
[49] 7. Large islands whose length scale is much greater
than that of the bounding channels (channel bounded land
masses) tend to have rougher boundaries than true islands
with length scale comparable to or less than the width of the
bounding channels, suggesting process differences between
the two island types;
[50] 8. A weighted connectivity analysis based on channel
width can identify weak channels within the delta network.
In particular, channels connecting the upper portion of the
delta to the coast do not seem to play a significant role in the
transport of water and sediment.
algorithm and, together with K. Caylor and I. Rodriguez-Iturbe, for contributing initial ideas related to this project. We thank A. Clauset and
C. R. Shalizi for developing and releasing the methods and tools for the
analysis of power law distributions in empirical data (http://tuvalu.santafe.
edu/~aaronc/powerlaws/). We are grateful for the comments received
from the Editor, Alexander Densmore, the Associate Editor, Andrew
Ashton, Matthew Wolinsky, and two anonymous reviewers who have
helped improve this paper.
References
Alam, M. K., A. K. M. S. Hasan, M. R. Khan, and J. W. Whitney,
(1990), Geological map of Bangladesh, Tech. Rep., Geological Survey of
Dhaka, Bangladesh.
Aly, M. H., J. R. Giardino, A. G. Klein, and H. A. Zebker (2012), InSAR
study of shoreline change along the Damietta Promontory, Egypt, J.
Coastal Res., 28(5), 1263–1269.
Ashworth, P., J. L. Best, M. H. Sarker, and J. E. Roden (2007), The
Brahmaputra-Jamuna River, Bangladesh, in Large Rivers: Geomorphology and Management, edited by A. Gupta, pp. 395–430, Wiley and Sons
Ltd, Chichester, U. K.
Bristow, C. S. (1987), Brahmaputra River: channel migration and deposition, in Recent Developments in Fluvial Sedimentology. Special Publication, vol. 39, edited by F. G. Ethridge, R. M. Flores, and M. D. Harvey,
pp. 63–74, Society of Economic Paleontologists and Mineralogists,
Tulsa, OK.
Clauset, A., C. R. Shalizi, and M. E. J. Newman (2009), Power-law
distributions in empirical data, SIAM Review, 51(4), 661–703.
Coleman, J. M. (1969), Brahmaputra river: Channel processes and
sedimentation, Sedim. Geol., 3(2–3), 129–239, doi:10.1016/00370738(69)90010-4.
D’Alpaos, A., S. Lanzoni, M. Marani, and A. Rinaldo (2010), On the
tidal prism—Channel area relations, J. Geophys. Res., 115, F01003,
doi:10.1029/2008JF001243.
Dillabaugh, C., K. Niemann, and D. Richardson (2002), Semi-automated
extraction of rivers from digital imagery, Geoinformatica, 6(3), 263–284.
Edmonds, D. A., C. Paola, D. C. J. D. Hoyal, and B. A. Sheets (2011),
Quantitative metrics that describe river deltas and their channel network,
J. Geophys. Res., 116, F04022, doi:10.1029/2010JF001955.
Ericson, J. P., C. J. Vorosmarty, S. L. Dingman, L. G. Ward, and M.
Meybeck (2006), Effective sea-level rise and deltas: Causes of change
and human dimension implications, Global Planet. Change, 50, 63–82,
doi:10.1016/j.gloplacha.2005.07.004.
Fagherazzi, S. (2008), Self-organization of tidal deltas, Proc. Natl. Acad.
Sci. U.S.A., 105(48), 18,692–18,695, doi:10.1073/pnas.0806668105.
Fagherazzi, S., A. Bartoluzzi, W. E. Dietrich, A. Adami, S. Lanzoni, M.
Marani, and A. Rinaldo (1999), Tidal networks: 1. Automatic network
extraction and preliminary scaling features from digital terrain maps,
Water Resour. Res., 35(12), 3891–3904, doi:10.1029/1999WR900236.
Feola, A. (2006), Hydrological and geomorphological studies in transition
environments, PhD thesis, University of Padova, Padova.
Geleynse, N., V. R. Voller, C. Paola, and V. Ganti (2012), Characterization of river delta shorelines, Geophys. Res. Lett., 39, L17402,
doi:10.1029/2012GL052845.
Goodbred, S. L. J., and S. A. Kuehl (2000a), The significance of
large sediment supply, active tectonism, and eustasy on margin
sequence development: Late Quaternary stratigraphy and evolution of
the Ganges-Brahmaputra delta, Sediment. Geol., 133(3–4), 227–248,
doi:10.1016/S0037-0738(00)00041-5.
Goodbred, S. L. J., and S. A. Kuehl (2000b), Enormous GangesBrahmaputra sediment discharge during strengthened early Holocene
monsoon, Geology, 28(12), 1083–1086, doi:10.1130/0091-7613.
Goodbred, S. L. J., S. A. Kuehl, M. S. Steckler, and M. H. Sarker (2003),
Controls on facies distribution and stratigraphic preservation in the
Ganges-Brahmaputra delta sequence, Sediment. Geol., 155(3–4), 301–
316, doi:10.1016/S0037-0738(02)00184-7.
Grazzini, J., S. Dillard, and P. Soille (2010), A new generic method for
the semi-automatic extraction of river and road networks in low and
mid-resolution satellite images, in Proc. of SPIE, vol. 7830, edited by
L. Bruzzone, 783007-1–783007-10, SPIE Europe, Cardiff.
Hedges, J. I., and R. G. Keil (1995), Sedimentary organic matter preservation: An assessment and speculative analysis, Mar. Chem., 49,
81–115.
Leopold, L. B., J. N. Collins, and L. M. Collins (1993), Hydrology of some
tidal channels in estuarine marshland near San Francisco, Catena, 20(5),
469–493, doi:10.1016/0341-8162(93)90043-O.
Liu, X., K. Chen, and D. Wang (2001), Extraction of hydrographic regions
from remote sensing images using an oscillator network with weight
adaptation, IEEE Trans. Geosci. Remote Sens., 39(1), 207–211.
[51] Acknowledgments. P.P. would like to acknowledge support from
the National Science Foundation (grants FESD/EAR-1135427, GSS/BCS1063228), and from the University of Texas at Austin. C.P. is grateful for
support from the National Science Foundation BanglaPIRE project (OISE
09-68354). We thank A. Feola for sharing her channel centerline extraction
11
PASSALACQUA ET AL.: GEOMORPHIC SIGNATURES IN DELTA NETWORKS
Shaw, J. B., M. A. Wolinsky, C. Paola, and V. R. Voller (2008), An imagebased method for shoreline mapping on complex coasts, Geophys. Res.
Lett., 35, l12405, doi:10.1029/2008GL033963.
Slingerland, R., and N. D. Smith (2004), River avulsions and their
deposits, Annu. Rev. Earth Planet. Sci., 32, 257–285, doi:10.1146/
annurev.earth.32.101802.120201.
Smart, J. S., and V. L. Moruzzi (1972), Quantitative properties of delta
channel networks, Z. Geomorphol., 16(3), 283–300.
Soille, P., and J. Grazzini (2007), Extraction of river networks from satellite
images by combining mathematical morphology and hydrology, in Proc.
CAIP, vol. LNCS 4673, edited by W. G. Kropatsch, M. Kampel, and A.
Hanbury, pp. 636–644, Springer-Verlag, Berlin, Heidelberg.
Straub, K. M., D. J. Jerolmack, D. Mohrig, and D. H. Rothman (2007),
Channel network scaling laws in submarine basins, Geophys. Res. Lett.,
34, L12613, doi:10.1029/2007GL030089.
Syvitski, J. P. M. (2005), The morphodynamics of deltas and their distributary channels, in River, Coastal, and Estuarine Morphodynamics:
RCEM 2005, edited by G. Parker and M. Garcia, pp. 143–150, Taylor
and Francis Group, London.
Syvitski, J. P. M. (2008), Deltas at risk, Sustainability Sci., 3, 23–32,
doi:10.1007/s11625-008-0043-3.
Syvitski, J. P. M., and Y. Saito (2007), Morphodynamics of deltas under the
influence of humans, Global Planet. Change, 57, 261–282.
Syvitski, J. P. M., I. Overeem, G. R. Brakenridge, and M. Hannon
(2012), Floods, floodplains, delta plains—A satellite imaging
approach, Sediment. Geol., 267–268, 1–14, doi:10.1016/j.segeo.2012.
05.014.
Trigg, M. A., P. D. Bates, M. D. Wilson, G. Schumann, and C. Baugh
(2012), Floodplain channel morphology and networks of the middle Amazon River, Water Resour. Res., 48, W10504, doi:10.1029/
2012WR011888.
Wilkerson, G. V., and G. Parker (2011), Physical basis for quasi-universal
relationships describing bankfull hydraulic geometry of sand bed rivers,
J. Hydr. Eng., 137(7), 739–753.
Wolinsky, M. A., D. A. Edmonds, J. Martin, and C. Paola (2010), Delta
allometry: Growth laws for river deltas, Geophys. Res. Lett., 37, L21403,
doi:10.1029/2010GL044592.
Marani, M., E. Belluco, A. D’Alpaos, A. Defina, S. Lanzoni, and A. Rinaldo
(2003), On the drainage density of tidal networks, Water Resour. Res.,
39(2), 1040, doi:10.1029/2001WR001051.
Myrick, R. M., and L. B. Leopold (1963), Hydraulic geometry of a small
tidal estuary, in USGS Numbered Series, 422-B, pp. B1–B18, U.S. Govt.
Print Off., Washington.
Newman, M. E. J. (2005), Power laws, Pareto distributions and Zipf’s law,
Contemp. Phys., 46, 323–351.
Newman, M. E. J. (2010), Networks An Introduction, Oxford Univ. Press
Inc., New York, N. Y.
Paola, C., R. R. Twilley, D. A. Edmonds, W. Kim, D. Mohrig, G. Parker,
E. Viparelli, and V. R. Voller (2011), Natural processes in delta restoration: Application to the Mississippi delta, Annu. Rev. Mar. Sci., 3, 67–91,
doi:10.1146/annurev-marine-120709-142856.
Persits, F. M., C. J. Wandrey, R. C. Milici, and A. Manwar, (2001), Digital
geologic and geophysical data of Bangladesh, U.S. Geol. Surv. Open File
Rep., 97-470H.
Quackenbush, L. J. (2004), A review of techniques for extracting linear features from imagery, Photogramm. Eng. Remote Sens., 70(12),
1383–1392.
Rakwatin, P., T. Sansena, N. Marjang, and A. Rungsipanich (2013), Using
multi-temporal remote-sensing data to estimate 2011 flood area and volume over Chao Phraya River basin, Thailand, Remote Sens. Lett., 4(3),
243–250, doi:10.1080/2150704X.2012.723833.
Rinaldo, A., S. Fagherazzi, S. Lanzoni, M. Marani, and W. E. Dietrich
(1999a), Tidal networks: 2. Watershed delineation and comparative network morphology, Water Resour. Res., 35(12), 3905–3917,
doi:10.1029/1999WR900237.
Rinaldo, A., S. Fagherazzi, S. Lanzoni, M. Marani, and W. E. Dietrich
(1999b), Tidal networks: 3. Landscape-forming discharges and studies in empirical geomorphic relationships, Water Resour. Res., 35(12),
3919–3929, doi:10.1029/1999WR900238.
Rodriguez-Iturbe, I., and A. Rinaldo (1997), Fractal River Basins. Chance
and Self-Organization, Cambridge Univ. Press, New York.
Sassi, M. G., A. J. F. Hoitink, B. de Brye, and E. Deleernnijder (2012),
Downstream hydraulic geometry of a tidally influenced river delta, J.
Geophys. Res., 117, F04022, doi: 10.1029/2012JF002448.
12