You are on page 1of 17

Earth Surface Processes and Landforms

Earth Surf. Process. Landforms 28, 231–247 (2003)


Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/esp.480

CLOSE RANGE DIGITAL PHOTOGRAMMETRIC ANALYSIS


OF EXPERIMENTAL DRAINAGE BASIN EVOLUTION
J. BRASINGTON1 * AND R. M. A. SMART2
1 Department of Geography, University of Cambridge, Cambridge, CB2 3EN, UK
2 Department of Geography, University of Hull, Hull, HU6 7RX, UK

Received 5 May 2001; Revised 16 December 2001; Accepted 18 December 2001

ABSTRACT
Despite the difficulties of establishing formal hydraulic and geometric similarity, small-scale models of drainage basins
have often been used to investigate the evolution and dynamics of larger-scale landforms. Historically, this analysis has
been restricted to planform basin characteristics and only in the last decade has the topographic similarity of experimental
landscapes been explored through explicitly three-dimensional parameters such as the distributions of cumulative drainage
area, area–slope and catchment elevation. The current emphasis on three-dimensional morphometry reflects a growing
awareness of the descriptive paucity of planform data and the need for more robust analysis of spatial scaling relationships.
This paradigm shift has been significantly facilitated by technological developments in topographic survey and digital
elevation modelling (DEM) which now present the opportunity to acquire and analyse high-resolution, distributed elevation
data. Few studies have, however, attempted to use topographic modelling to provide information on the changing pattern
and rate of sediment transport though an evolving landscape directly by using multitemporal DEM differencing techniques.
This paper reports a laboratory study in which digital photogrammetry was employed to derive high-resolution DEMs
of a simulated landscape in declining equilibrium at 15 minute frequency through a 240 minute simulation. Detailed
evaluation of the DEMs revealed a vertical precision of 1Ð2 mm and threshold level of change detection between surfaces
of š3 mm at the 95 per cent confidence level. This quality assurance set the limits for determining the volumetric
change between surfaces, which was used to recover the sediment budget through the experiment and to examine local-
and basin-scale rates of sediment transport. A comparison of directly observed and morphometric estimates of sediment
yield at the basin outlet was used to quantify the closure of the sediment budget over the simulation, and revealed an
encouragingly small 6Ð2 per cent error. The application of this dynamic morphological approach has the potential to offer
new insights into the controls on landform development, as demonstrated here by an analysis of the changing pattern of
the basin sediment delivery ratio during network growth. Copyright  2003 John Wiley & Sons, Ltd.
KEY WORDS: drainage basin evolution; digital photogrammetry; DEMs; sediment budgets

INTRODUCTION

The application of precisely controlled and intensively monitored small-scale (100 –102 m2 ) physical models
to study the processes involved in channel development is a well-established methodology in geomorphology.
This experimental approach has been extensively applied to investigate the threshold hydraulic conditions and
hydrological pathways controlling channel incision, particularly in the context of rill and gully development
in agricultural soils (e.g. Moss et al., 1982; Bryan and Poesen, 1989; Slattery and Bryan, 1992; Bryan
and Rockwell, 1998). Less commonly, experimental drainage basins have also been studied as small-scale
analogues of entire fluvially eroded catchment systems (see Schumm et al., 1987, for a review). While these
models are rarely formally scaled with reference to a larger (104 –1010 m2 ) prototype, experimental networks
and basins are often interpreted as appropriate analogues on the basis of ‘similarity of process and performance’
(Hooke, 1968). Historically, evidence for behavioural similarity has been predicated largely on the striking
visual and geometric correspondence of planform geometric parameters of simulated drainage networks.
More recent studies of basin topography, however, have tended to approach the study of basin similarity

* Correspondence to: J. Brasington, Department of Geography, University of Cambridge, Cambridge, CB2 3EN, UK.
E-mail: jb10016@cam.ac.uk

Copyright  2003 John Wiley & Sons, Ltd.


232 J. BRASINGTON AND R. M. A. SMART

and scaling through a more rigorous examination of explicitly three-dimensional parameters, including the
frequency distributions of cumulative drainage area, area–slope and catchment elevation, or hypsometry (e.g.
Willgoose, 1994; Hancock and Willgoose, 2001c). The scale independence of these descriptors in conjunction
with the visual correspondence of natural and experimental drainage networks is now increasingly interpreted
as signature of a ‘deep similarity’ and tendency towards self-organization in basin processes (Rodriguez-Iturbe
and Rinaldo, 1997).

Dynamic morphometry
Few experimental studies have pursued this morphometric similarity to consider the spatial and temporal
dynamics of sediment delivery within the basin that are ultimately responsible, through closely-coupled
feedback, for the evolving topographic form. Notable exceptions are the early laboratory studies of Parker
(1977) and, more recently, Brunton and Bryan (2000) who monitored channel network dynamics in a silt loam
soil under simulated rainfall. By incorporating water and sediment flux measurements within the evolving
tributary network Brunton and Bryan (2000) were able to establish the changing hydrologic and sediment
budgets for selected reaches of the experimental network. Significantly, their analysis highlighted the close
relationship between the intensity of headwater knickpoint incision and the dynamics of scour and fill in the
main truck channel, in line with the conceptual complex response model first hypothesized by Schumm and
Hadley (1957). While this approach is insightful and offers a new perspective for upscaling model results,
reliance upon direct observation of mass fluxes at specific locations within a drainage network constitutes
a major sampling problem and restricts a comprehensive interpretation of system dynamics. The limitations
of at-a-point sampling have been well established and an increasingly popular alternative is the inverse
estimation of sediment transport patterns and rates from analysis of morphological dynamics (Ashmore and
Church, 1998). Essentially this involves inter-comparison of topographic surfaces over time, an approach
which has been successfully used in fluvial geomorphology to reconstruct short-term bedload transport rates
(e.g. Goff and Ashmore, 1994; Lane et al., 1995) and channel sediment budgets over annual and longer
intervals (e.g. Brasington et al., 2000).
The recent surge of interest in this ‘morphological method’ stems largely from advances in survey tech-
nology and digital elevation modelling which enable detailed landform monitoring and quantitative analysis
at unprecedented spatial resolution and speed. A key development in this respect has been the automation of
stereo-matching in analytical photogrammetry through the development of digital pixel correlation methods
(Chandler, 1999). In principle, this enables largely automated construction of very dense digital elevation mod-
els (DEMs) at resolutions limited only by the scale and resolution of the imagery used. Since the widespread
availability of softcopy digital photogrammetric software, this approach has been successfully used to monitor
fluvial landforms over a range of scales: from individual gravel grains (e.g. Butler et al., 1998) through to
large, low relief braided river systems (e.g. Westaway et al., 2000; Brasington et al., 2001). The implications
of this new data acquisition method for the analysis of sediment transport dynamics are significant. Sequen-
tial comparison (differencing) of DEMs allows for robust estimation of space–time averaged transport rates
and can also be used to establish the distribution of erosion and deposition and thus the nature of spatial
interaction patterns. While photogrammetric monitoring has real potential in fluvial systems, the recovery of
information from the wetted zones of active channels, where the bed is partially or completely obscured,
remains problematic. Topographic survey of these areas often requires a ground-based approach, presenting
problems of variable data coverage and reliability. This problem is significantly diminished in laboratory
studies where a model experiment can be stalled, drained and photography acquired before rewetting (Stojic
et al., 1998). This enables the capture of a permanent record of the entire model simulation using consistent
survey methods with commensurate data precision and resolution.
This paper describes the application of repeat digital photogrammetric monitoring to investigate the evolving
morphology of a small-scale experimental drainage basin, and complements recent research by Hancock and
Willgoose (2001a,b,c). In a similar experimental vein, they used automated photogrammetric monitoring to
develop DEMs of simulated landscapes as part of a series of tests on the numerical landform evolution
model, SIBERIA (Willgoose et al., 1991). These tests focused upon the characteristic three-dimensional
topography generated by both laboratory and numerical models, quantified in terms of catchment hypsometry

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 233
and drainage flowpath analysis. This study, however, aims to extend this morphometric analysis by using
DEM differencing techniques to quantify the changing sediment budget of an experimental basin in the
early stages of its evolution. The two main research objectives were; (i) to consider the spatial and temporal
dynamics of sediment movement through the evolving drainage structure; and (ii) to examine the reliability
of morphologically derived sediment transport rates through a comparison with observed sediment flux at the
simulator outlet. Quantitative analysis of DEM quality and derived transport rate data demonstrate that the
remote survey method applied here is capable of delivering robust datasets which afford new perspectives on
spatial scaling relationships in geomorphology and provide a novel basis for numerical model validation.

METHODOLOGY
The experiment described below is the first of a set designed to explore small channel and network develop-
ment under different hydrologic, geometric and sedimentary conditions. No attempt has been made to scale
the experiment with respect to the rainfall drop-size or substrate grain-size of a given prototype, and the
experiment should not therefore be treated as a formal hydraulic model.

The laboratory simulator and experimental conditions


The experiment was conducted in a 2Ð1 ð 1 m experimental basin, with simulated rainfall delivered from
a system of three microsprinklers (Figure 1). The basin enclosure is constructed in steel and incorporates
an open-mesh base covered in a freely draining geotextile material which limits water table development.
The flume base is levelled upstream and downstream by adjustable screws which set the initial bedslope.
Baselevel is controlled by an adjustable weir at the outlet of the flume. This is connected to a delivery pipe
which allows flow to be tapped off to sample water and sediment discharge. All water and sediment discharges
are ultimately piped to storage for later analysis.
Rainfall was simulated by three pressurized sprinklers (Lechler axial-flow-cone nozzles) mounted 2Ð6 m
vertically above the flume. Flour tray experiments (Laws and Parsons, 1943) were conducted to determine
rain drop-size distributions for the 52 kPa supply used and revealed a D50 and D90 of 1Ð12 and 1Ð64 mm
respectively. Tests using a set of 16 manually recorded rain gauges across the flume found that while the
temporal pattern of rainfall was consistent, intensities varied spatially between 73Ð9 and 90Ð8 mm h1 , with
a maximum in the central area of the flume due to the overlap between spray cones. Large drips from the
most upstream nozzle were also observed as the flow was shut down between rain events and resulted in
significant soil splash directly within the drip zone.
The sediment used in this experiment was a 5 : 2 sand : kaolin mix chosen for its readiness to erode but
sufficient cohesiveness to maintain channel walls. The sediments were air-dried, machine mixed and then
placed in the flume in layers and compacted with a roller to provide a level surface with homogeneous
packing with a bulk density of 1Ð47 g cm3 . The flume slope was set to 12 per cent and an initial base level
perturbation of 0Ð08 m was simulated at the outlet by lowering the adjustable weir. Rainfall was then applied
to the surface for a total duration of 240 minutes, in 16 individual 15 minute events. At the end of each event
the flume was drained and photography of the surface acquired. Water and sediment samples were taken at the
flume outlet for a 15 second period every five minutes. These were later vacuum filtered through Whatman
54 papers and used to develop a time-trace of sediment yield through the simulation.
Some significant factors of this study should be noted when drawing comparisons with other recent exper-
imental studies of basin development. For example, despite the overall size similarity of Hancock and
Willgoose’s (2001a,b) simulator, their experiments employed a much finer calibre sediment (fly-ash) and
smaller rainfall drop size (D50 D 0Ð2 mm).

Photogrammetric data acquisition


The detection of morphological change through repeat survey is highly sensitive to the reproducibility and
errors inherent in DEM development. A critical stage in data acquisition is the accurate and precise definition
of a network of photo-control targets used to construct a robust photogrammetric net. Here, photo-control was
provided by 25–30 patterned targets approximately 10 mm in diameter placed on the model surface. These

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
234 J. BRASINGTON AND R. M. A. SMART

1000

sprinkler

3000

2000

adjustable
Adjustable weir
Frame height
to control
basin slope

To collection
drums

Figure 1. The experimental simulator. All dimensions are quoted in millimetres

were distributed in a grid-based pattern approximately 0Ð2–0Ð3 m apart. Each target was surveyed indepen-
dently from two observation stations using a TOPCON GTS4 total station with an angular resolution of 1 s.
Object space coordinates in x, y and z were then determined by three-dimensional intersection using software
provided by J. Chandler, Loughborough University, UK. Twenty-two resurveys of the initial control network
identified average root-mean-square precisions of 3Ð6 mm, 1Ð7 mm and 0Ð8 mm in x, y, and z respectively. In
subsequent experiments, positions have been determined by least-squares bundle adjustment using the CUBA
software package developed at City University, UK. This uses the ‘three-dimensional variation of coordinates’
approach outlined by Cooper (1987).
Following each rainfall event, the surface was allowed to drain but not to dry fully before digital imagery
of the basin surface was acquired using a non-metric Kodak DCS410 CCD system mounted on a Nikon
N90s with a Sigma 14 mm lens at f3Ð5 (Figure 2). This camera lacks fiducial marks, so the image space
coordinates were determined relative to image corners. Near-vertical photographs were taken from a height
of approximately 1Ð7 m as a compromise between photographic coverage and scale. From this height, with
an approximate focal length of 0Ð014 m and the 9 µm pixel dimensions of CCD imager, gives object space
pixel dimensions and thus an optimal vertical precision of 0Ð0011 m. Two strips of three photographs were
required to obtain overlapping pairs that covered the full flume surface. More recent experiments incorporating
additional oblique photographs taken from either end of the flume have been found to improve the recovery
of exterior orientation parameters (Stojic et al., 1998).
Photogrammetric processing and DEM extraction
Only a brief overview of the photogrammetric data processing is given here, and more comprehensive
accounts of the stages and theory involved are given in Lane et al. (1993) and Chandler (1999). Digital
photogrammetric construction of DEMs at each epoch is a three-stage process including photogrammetric

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 235

Figure 2. Close range stereo-pair photographs of downstream region of the eroded basin after 240 minutes

triangulation, automated DEM extraction and mosaicing of DEMs of the overlap areas. The initial stages of
processing are similar to those of standard analytical photogrammetry where camera parameters, basic image
information and image and object space photo-control coordinates are used for photogrammetric triangulation.
For non-metric cameras such as the DSC410, the interior orientation parameters of the camera have to be
estimated a priori. This was achieved prior to the main data acquisition stage, under the assumption that
the internal camera geometry is stable. While there are reasons to be sceptical about the stability of internal
camera geometry (Shortis et al., 1998), previous studies have shown that ‘on-the-fly’ calibration, in which the
camera parameters are estimated for each set of images, provides only marginal improvement in the resulting
photogrammetric solution (Chandler et al., 2001). Here therefore, the camera parameters were determined
only once, using the self-calibrating bundle adjustment software, CUBA, developed at City University, UK.
The parameters were defined through the measurement of a set of targets marked on a wall, whose positions
were found by intersection from angular measurements. For consistency, the camera lens focusing ring was
set to 1Ð7 m in line with the flying height used in the laboratory. Oblique images of the targets were then
acquired from seven camera positions, and photo-coordinates determined using the image corners to define
the reference system. Bundle adjustment was then used to find estimates of six interior-orientation parameters,
the lens focal length, the principle point offsets, xp , yp , and the radial distortion parameters K1 , K2 and K3 .
Photogrammetric triangulation and DEM extraction were performed using ERDAS IMAGINE OrthoMAX
software. Photogrammetric triangulation is based on a least-squares block bundle adjustment which estimates
the exterior orientation parameters of the camera which are used later with the enforced interior-orientation
parameters in DEM extraction. Ten to 15 control points and up to 15 tie points were used in the block
construction for each image pair. The quality of the photogrammetric solution can be assessed by examining
control- and photo-coordinate residuals. Average root-mean-square errors for the control points were close
to 1 mm in all dimensions and photo-coordinate residuals were 0Ð9 µm and 1Ð2 µm in x and y respectively.
The DEM extraction tool used in OrthoMAX is based on an automatic stereo-matching algorithm which
identifies corresponding points on pair images based on a hierarchical area-based search for pixels with similar
brightness numbers and spatial contrast (ERDAS, 1995). DEMs were collected at a resolution of 0Ð005 m
reflecting a compromise between image resolution, efficient data storage/later processing requirements and
surface representation. Automatic DEM collection statistics were derived which reveal the percentage of pixel
matches lying within categories defined relative to a user-specified threshold distance, with the cut-off set
here to 0Ð5 pixels. Pixels which are not successfully stereo-matched were estimated by bilinear interpolation.
Early experimentation with the DEM tool obtained relatively low pixel correlations with only 11–23 per cent
matches classified as good (0–0Ð17 pixels) and up to 60 per cent unmatched points. Later tests have improved
these matching statistics (up to 40 per cent good matches) by masking out areas outside of the flume during
DEM collection. This improvement reflects the exclusion of spurious movements in areas outside the flume
that were captured differentially on the image pairs.

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
236 J. BRASINGTON AND R. M. A. SMART

The last stage of DEM processing involved mosaicing the three overlapping DEMs (here taking the arith-
metic average to define the elevation in the overlap) and clipping the final product to the extent of active
flume surface.

RESULTS
Digital elevation models of the eroding surface were successfully developed for all but three of the sixteen
events (15–30 min; 105–120 min; 135–150 min), which were lost due to corruption of the image files.
Figure 3 shows greyscale DEMs of the experimental basin after 15, 90, 165 and 240 min which have been
normalized to account for the initial surface slope. This reveals the progressive dissection of basin topography
through the rapid development and extension of a major trunk channel aligned with the flume outlet giving
rise to a terrain characteristic of dissected badlands, with steep-sided channels and distinct arcuate headcuts.
Preliminary visual analysis suggests that the drainage network reached maximum extension (Glock, 1931)
between 135 and 180 minutes, with only minor later adjustments to the overall form and extent. However,
a notable exception is the capture, after 225 minutes, of a closed depression developed directly below the
upstream sprinkler nozzle which was created by splash erosion due to large drips formed as the simulator was
turned off after each event. The pattern of observed sediment flux at the basin outlet is shown in Figure 4.
While there is significant scatter in the record, the moving average trend shows a clear peak between 50
and 100 minutes which corresponds to a period of rapid headcut migration and channel degradation (see
discussion below).
The rapid approach to maximum extension and a steady-state form is corroborated by a timeseries of mean
basin elevation and absolute relief derived from the DEMs (Figure 5). These reveal a sharp decline in mean
elevation and general increase in relief (here defined simply as Zmax –Zmin ), up to 135–165 minutes. While

Figure 3. Slope normalized DEMs of the experimental basin after 15, 90, 165 and 240 minutes

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 237

2.5 Observed Flux

Sediment Yield at Outlet (gs-1) 5 pt Moving Average


2.0

1.5

1.0

0.5

0.0
0 50 100 150 200 250 300
Time (mins)

Figure 4. Observed sediment yield at the flume outlet

such coarse parameters are not particularly sensitive to the actual geometrical configuration of the basin, this
interpretation is reinforced by the trend in the standard deviation of elevations, which is a useful measure of
topographic complexity, which also shows a progressive increase before stabilizing after 135 minutes.
Significantly, the elevation timeseries highlights a sudden jump in average basin elevation between 90 and
105 min. Visual examination of the DEMs between these periods showed no unusual or significant landform
development. However, DEM differencing (subtraction) revealed a spatially consistent rise of 1–2 mm in
areas not directly subject to erosion. Various possible experimental procedures may account for this apparent
blunder. The most probable explanation is a shift in the survey datum caused either by movement of the flume
or the total station mountings. However, the effects of drying and subsequent rewetting causing sediment
shrinkage and expansion may also be a contributory factor. This particular event will have been highly
susceptible to this effect as the experiment was halted for 10 hours overnight between 90 and 105 min. While
the surface was kept covered during this time, desiccation cracks were evident on the surface that closed on
rewetting the following morning. Without precise identification of the cause, any general systematic correction
is difficult to identify and the results of later morphometric mass balancing over this period must be treated
very cautiously. More generally, the clear sensitivity of any morphometric interpretation to such blunders
highlights the need for external datum checks.

DEM QUALITY ASSESSMENT


The distributed morphometric approach to sediment budgeting is highly sensitive to DEM quality and an
assessment of model error is important in order to determine a methodological limit of detection to distinguish
significant morphological change between surfaces (Brasington et al., 2000). DEM validation is the subject
of growing interest within geomorphometry and recent research has drawn from approaches in engineering
surveying, with a focus upon identifying types of error, i.e. random, systematic and blunders (such as described
above) and separately quantifying the precision, accuracy and reliability of modelled surfaces (Chandler, 1999;
Lane et al., 2000). A standard approach to DEM quality assessment is through a comparison of modelled
elevations with a set of independently surveyed checkpoints. The accuracy of the modelled surface is controlled
by systematic errors in the survey and can be estimated from the mean difference between modelled and
observed checkpoint elevations. By contrast, statistical precision is a function of random errors and can be
quantified from the standard deviation of checkpoint errors.
To establish these error characteristics, 40 checkpoints, distributed in two perpendicular transects across the
model surface, were measured by intersection. It should be noted that while independent, these checkpoints
are not error free, and have similar positional and vertical errors to the photocontrol survey. DEM elevations

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
238 J. BRASINGTON AND R. M. A. SMART

(a) 0.898

0.896

Mean Elevation (m)


0.894

0.892

0.890

0.888

0.886
0 50 100 150 200 250 300
Time (mins)

(b) 0.28
0.26
Absolute Relief (m)

0.24
0.22
0.20
0.18
0.16
0.14
0.12
0.10
0 50 100 150 200 250 300
Time (mins)

(c) 0.050
Standard Deviation (m)

0.048

0.046

0.044

0.042

0.040

0.038

0.036
0 50 100 150 200 250 300
Time (mins)

Figure 5. Derived timeseries of (a) mean basin elevation, (b) basin relief, (c) standard deviation of elevations

at the control point coordinates were extracted using point-in-grid functions within the GIS ArcView and the
histogram of errors is shown in Figure 6 and the results summarized in Table I.
The histogram of differences is approximately normally distributed, with a relatively narrow spread and
high kurtosis. This result is confirmed by the low standard deviation of errors (SDE), 1Ð2 mm, and the
maximum absolute error, 5Ð6 mm. These results are encouraging as SDE compares well with the maxi-
mum theoretical vertical precision, 1Ð1 mm, afforded by the image resolution. This suggested precision also
compares favourably with the recent research by Hancock and Willgoose (2001b), who reported SDEs of
1Ð5–2 mm and maximum errors of 5–6 mm, measured with respect to the photo-control points used in
photogrammetric triangulation. Significantly, however, the distribution of errors is not centred about zero

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 239
30

25

20
Frequency (%)

15

10

0
-11.0 -5.5 0.0 5.5 11.0
Residual Error (mm)

Figure 6. Histogram of differences between DEM and checkpoint elevations

Table I. Summary of checkpoint and DEM elevations intercomparison

n Mean error Standard deviation of errors Max. absolute error


(m) (m) (m)

40 0Ð0015 0Ð0012 0Ð0056

and the mean error of 1Ð5 mm suggests that the photogrammetric elevations are systematically higher than
observed. Such a systematic offset is likely to have arisen from errors incorporated in the photogrammetric
block, but so long as the pattern of absolute error is spatially consistent (internal reliability), such systematic
offsets pose few concerns where the aim of the survey is for internal, relative measurement. However, in
applications where, as here, repeat survey and DEM intercomparison are used for budgeting and change
detection, such absolute errors may seriously confound the analysis if errors are not consistent between
surveys.
In order to assess the impact of such potential errors and to empirically define an appropriate statistical level
of detection, 10 DEMs of the initial flume surface were constructed using different images and photocontrol
measurements but the same camera model, to mimic the conditions involved in the actual sequential DEM
collection and analysis. Figure 7 shows the histogram of maximum elevation differences (range) between
DEM pixels (n D 69 682) measured for the set of 10 surfaces. Significantly, over 88 per cent of the replicate
measurements lie within 2 mm and less than 0Ð005 per cent of rogue measurements have maximum differences
of greater than 1 cm.
This analysis indicates a non-parametric 95 per cent confidence interval, and threshold for change detection
of approximately š3 mm, just 10 per cent of absolute relief as shown in Figure 5c. Despite this, the spatial
pattern of errors does exhibit some systematic variation, with differences most apparent in the overlap areas
between the mosaiced DEMs. This effect is clearly illustrated in Figure 8, which shows differences between
two overlapping DEMs increasing systematically away from the centre of the overlap. This distribution of
errors is likely to arise from imperfections in the polynomial lens calibration model, which decreases in
reliability with distance from the principal point.

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
240 J. BRASINGTON AND R. M. A. SMART

60
56.01

50

Frequency (%) 40.06


40

30

20

10
3.75
0.16 0.02 0.002 0.005
0
0.0010

0.0020

0.0030

0.0040

0.0050

0.0060

0.0070

0.0080

0.0090

0.0100

0.0110

0.0120
Range (m)

Figure 7. Histogram of maximum elevation differences measured in 10 repeat surveys of the initial (t D 0 min) flume surface

Overlap region

Extent of
downslope DEM

Extent of
upslope DEM

< -0.01 > 0.010

Figure 8. Map of differences in elevation in overlapping DEMs. Positive differences reflect high elevations in upslope DEM and
vice versa

NETWORK AND BASIN EVOLUTION


Recent laboratory research by Hancock and Willgoose (2001a) revealed a three-stage pattern of network
development similar to that identified experimentally by Parker (1977). This conceptual model involved:
(i) unchannelized erosion down to the level of initiator headcut in the region around the basin outlet;

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 241
(ii) development and elongation of the channel network by upstream headcut migration and bifurcation;
(iii) network at maximum extension with low rates of erosion by sheetflow and periodic knickpoint incision.
While both these experiments were initiated on surfaces created with a central valley to minimize the influ-
ence of local surface roughness, the DEM timeseries shown in Figure 5 appears to reflect a similar pattern
of development, despite initial microtopographical control. However, unlike the experiments of Hancock and
Willgoose (2001a), the network developed here occupied only approximately 75 per cent of the simulator
at maximum extension, reflecting the greater shear strength of the substrate and thus the higher threshold
drainage support area.
The pattern of evolution and the actual translocation of sediment through the basin can be better visualized
by DEMs of difference created from the timeseries of surfaces. Figure 9 shows five DEMs of difference
comprising changes between 0 and 15, 60 and 75, 105 and 135, 165 and 80, and 225 and 240 minutes,
where a high threshold of level of change of š3 mm has been used to distinguish areas of erosion (white)
and deposition (black) and no change (grey). This documents a sequential pattern of network development
dominated by headward erosion and tributary infilling similar to the conceptual model outlined by Glock
(1931).
Between 0 and 15 minutes two main areas of erosion developed: (i) a broad region around and on the
true right of the central outlet; and (ii) a small localized area 1Ð5 m from the outlet created by rainsplash
associated with a leaky sprinkler (see earlier). The major downstream incision was initiated by unchannel-
ized overland flow converging at the base of the flume in flowlines controlled by the microtopography.
At the flume boundary wall, and in particular around the outlet headcut, turbulent undercutting led to
instability and ultimately slumping with the failed material rapidly transported away. Topography–flow feed-
back increased the capture of drainage area and the headcut migrated upstream as a roughly semicircular
set of similar slab failures developing away from the outlet. The pattern of failure was partly controlled
by the lower flume boundary, but as erosion continued the capture of drainage above the central axis
became dominant and the rate of retreat grew asymmetrically, extending most rapidly upstream from the
outlet.

Figure 9. DEMs of difference showing areas of erosion and deposition in excess of a š3 mm threshold level of detection. Black lines
show approximate positions of significant channels

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
242 J. BRASINGTON AND R. M. A. SMART

By 75 minutes the central headcut had bifurcated and rapidly migrated 90 cm upstream and a central
channel 5–10 cm wide had developed. A smaller channel on the right side had also incised. Although
the central headcut was the primary focus of erosional activity, degradation of the main channel was also
significant and between 60 and 75 minutes, 2–4 cm of scour along the thalweg occurred as the baselevel
instability was transmitted upstream. Erosional activity peaked during this period, consistent with the observed
sediment yield trace (Figure 4), although the first signs of limited depositional activity near the outlet were
evident.
By 135 minutes, the main area of incision had migrated a further 25 cm upstream and divided again
creating four distinct headwater areas. While this zone remained the primary site of intensive erosion, a
tributary network began to develop off the left of the main channel, capturing drainage and incising headward.
Degradation of the main channel ceased as the wave of dissection penetrated upstream and the channel
regraded. Overall the basin was still in a phase of net erosion and any aggradation remained limited to
lower reaches of the main channel. However, by 165–180 minutes, deposition had become significantly more
widespread, with patches along most channel networks and close to the channel heads. The rate of headcut
incision declined reflecting a reduction in specific upslope area, and although the new left-bank tributary
continued to retreat headward, Figure 10 shows the long profiles of the central channel to have reached a
new equilibrium profile. Little modification of the basin is evident between 180 and 240 minutes, apart from
capture of the splash depression by the central channel and in this last event the basin was close to a state of
dynamic equilibrium with an approximate balance between scour and fill.
The dominance of headward erosion in basin development is consistent with Howard’s (1971) conceptual
model of landscape evolution. The driving force for headward retreat is topography–flow feedback, where
surface lowering enhances topographic convergence and increases drainage capture by the channel head,
which in turn enhances local bed shear stress. Given the free drainage of this model and limited water table
development, this feedback is driven here through the convergence of overland rather than subsurface flow
(Dietrich and Dunne, 1993). Despite the overall primacy of headwater areas, the pattern described above
also highlights the development of tributary networks branching off the main channel, once the core drainage
structure has become established. Such infilling is more characteristic of Glock’s (1931) model of network
development, where rapid elongation of the major drainage framework is followed by network elaboration,
through the addition of minor tributaries. Importantly any direct comparison with previous experimental work
must be approached carefully, given the differences in scale (Parker, 1977) and substrate calibre (Hancock
and Willgoose, 2001a).

Figure 10. Topographic profiles of the central valley of the main channel sampled at 5 mm intervals at 0, 15, 75, 180 and 240 minutes

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 243
MORPHOLOGICAL ASSESSMENT OF SEDIMENT TRANSPORT

Quantification of volumetric change through DEM differencing offers the potential for comprehensive sedi-
ment budgeting and the estimation of not only the basin-scale averaged sediment flux, but also the internal
temporal and spatial patterns of sediment transport. The basic principle underlying morphological flux esti-
mation is the continuity equation for sediment transport, which expressed in one-dimensional finite difference
terms is:
BD V
Qso D Qsi  1
t

where, Qso and Qsi are the sediment flux out from and into a given reach or area respectively, BD is bulk
density (1  ε), where  is the rock density and ε is the sediment porosity) and V is the volumetric
change in sediment storage.
For this experiment the high-frequency record of sediment yield at the flume outlet (Figure 4) provides a
robust independent test of the basin-scale morphometric estimates derived from the DEM timeseries. At this
scale, uncertainties in upstream sediment input, which complicate reach-scale studies (Ashmore and Church,
1998), are negated and the system can be treated as closed (Qsi D 0).
Figure 11 shows the volumetric rate of erosion and deposition determined from the DEM timeseries, where
changes less than the š3 mm level of detection used in the differencing experiments described above have
been excluded. The estimates associated with the 90 and 105 minute surveys respectively appear anomalous,
and these correspond to the experimental blunder caused either by a datum shift or the effects of wetting and
drying. Setting aside these rogue measurements, the pattern confirms the sequence of development discussed
above, with erosion rates peaking between 70 and 90 minutes and the deposition rate increasing late in the
experimental run.
The timeseries of basin sediment yield, computed assuming a uniform bulk density of 1Ð47 g cm3 , is
shown together with the time trace of observed flux monitored at the outlet in Figure 12. While the overall
trend is similar, the morphological flux determined over the 15 minute survey interval appears to overestimate
the direct transport rate, particularly at peak delivery. Given the episodic nature of erosion and deposition,
sediment budget-based rates are strongly dependent on the integration-time used in their calculation, with
longer-term average rates subsuming short-term fluctuations. The magnitude of such high frequency variation
in sediment discharge was highlighted in replicate 15 second samples taken during three separate five minute
periods during the simulation. These revealed a pattern of 10–40 per cent short-term variation in flux, with
variability increasing with sediment load. Such sampling dependence clearly complicates the inter-comparison
between direct and morphometric estimates of transport rate. However, for a closed system with no upslope
(upstream) sediment contributions, the volumetrically based rate should reflect the integrated effects of short-
duration fluctuations and the problem of negative bias due to topographic compensation observed in open-reach
fluvial systems is avoided (Lindsay and Ashmore, 2002).

0.00015
Volumetric Change ( m 3s-1)

0.00010
0.00005
0.00000
50 100 150 250 300
-0.00005 0 200
-0.00010
-0.00015
-0.00020 Erosion
Deposition
-0.00025
-0.00030 Time (mins)

Figure 11. Volumetric erosion (positive) and deposition (negative) transport rates through the simulation

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
244 J. BRASINGTON AND R. M. A. SMART

3.5 Observed Flux


15 Minute Morphological Flux
3.0 5 pt Moving Average

2.5
Sediment Yield (gs-1)

2.0

1.5

1.0

0.5

0.0
0 50 100 150 200 250 300
-0.5

-1.0
Time (mins)

Figure 12. Observed and morphometric estimates of sediment yield. Note that the anomalous readings at 90 and 105 minutes have been
omitted. Morphometric estimates plotted at mid-point of time interval

Table II. Observed and morphologically predicted final sediment budgets

Observed sediment loss Morphometric estimate Obs.–Pred. Difference


(kg) (kg) (kg) (%)

20Ð6 19Ð3 1Ð3 6Ð2

An alternative assessment of the morphometric approach is afforded by an analysis of the final closed budget,
comparing total volumetric change (0–240 minutes) and the total sediment lost during the experiment, which
was stored through the simulation for this purpose. This result is shown in Table II, which shows a much
more encouraging 6Ð2 per cent difference (underestimate) in the total budget. This better match reflects the
implicit inclusion of sediment flushes which occurred at the start of rainfall in each event, but which were not
adequately sampled at the basin outlet, as well as sediment losses in groundwater through the open base of the
flume. This result also implies that the morphometric rates shown in Figure 11 are in fact more representative
of the longer-term flux integrated over the 15 minute period. The final residual error may in part be accounted
for by changes in sediment density over time, but perhaps more significantly provides an indication of the
(high) topographic data quality.

SEDIMENT DELIVERY RATIO


In addition to direct flux budgeting, the separate calculation of volumes of erosion and deposition permits an
insight into the changing pattern of the sediment delivery ratio (SDR) over time, where:

Qso
SDR D 100 ð  2
E
A

where E is erosion (kg m2 t1 , and A the basin area. This is an indication of the ‘efficiency’ of basin
sediment transport and is an important geomorphic characteristic, often estimated to underpin the relationship

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 245
between riverine sediment load measurements and local erosion rates (Richards, 1993). The SDR can also be
estimated directly from the morphologically balance between erosion and deposition (D) where:

    
SDR D 100 ð 1  D E 3
A A

The dependence of the delivery ratio on the predicted volumes of erosion and deposition renders this parameter
susceptible to variations in DEM quality as discussed above. However, bivariate quotient expressions such as
Equation 3 are also sensitive to the relative magnitude and balance between their two constituent variables
(Taylor, 1982). In order to establish the robustness of morphometric delivery ratios, calculations based on
Equation 3 were repeated using volumes of erosion and deposition determined relative to three threshold
levels of detection, 0Ð003 (used in rate estimates discussed above), 0Ð005 and 0Ð007 m.
The changing pattern of the SDR though the simulation is shown in Figure 13. As would be expected when
transport rates are high (0–135 minutes), the SDR is comparatively insensitive to data errors (information
loss), but becomes increasing susceptible during the later stages of the experiment as rates of erosion and
deposition approach equilibrium, so that small variations in the estimate of either (interdependent) component,
propagate non-linearly into the SDR. Twice in the simulation at 90–105 and 180–195 minutes, negative values
were computed for all thresholds. Given the closed nature of the sediment budget, a negative SDR must be
an artefact due to data errors. This is clearly the case for the 90–105 minute anomaly which relates to the
blunder discussed above; however, the negative delivery ratio at 180–195 minutes reflects the more generic
sensitivity of the SDR to small DEM errors when predicted rates of erosion and deposition are similar.
Despite this sensitivity, the general pattern of the delivery ratio is consistent with the interpretation of basin
dynamics discussed above, where in the early stages of network development and incision up to 135 minutes,
the SDR is very high and near constant at 80–100 per cent. After this, the delivery ratio falls quickly to a
more moderate, but still high 40 per cent. This sudden decline reflects the increase in within-channel sediment
storage as a new lower gradient channel profile is established from the mouth upwards (Figure 10), and the
reduced rate of headward growth as the specific upslope area at channel heads declines as they approach
the divide. These data provide a unique insight into the changing pattern of sediment delivery and storage
and emphasize the strong coupling between headwater development and their downstream channel and valley
environments which control the movement of sediment through the basin.

100 + Threshold Level


+ of Detection (m)
90 + +
+ 0.003
Sediment Delivery Ratio (%)

80 0.005
0.007
70
60
+
50 +
40 + +

30
+
20
10
0
0 50 100 150 200 250
Time (mins)

Figure 13. Timeseries of morphometrically derived basin sediment delivery ratio computed for three threshold levels of change detection:
0Ð003, 0Ð005 and 0Ð007 m. Downward-pointing arrows indicate when computed SDR fell below zero

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
246 J. BRASINGTON AND R. M. A. SMART

CONCLUSIONS
The results of this research are particularly encouraging in terms of the application of digital photogrammetric
survey in experimental geomorphology. Specifically, it has shown that even using moderately cheap software
and a relatively low resolution non-metric digital camera, topographic models sufficient to characterize low
relief landforms and morphometric sediment budgets accurately can be readily derived by geomorphologists
with little direct photogrammetric training. Under the conditions applied here, tests on data quality suggest
that vertical precisions close to the theoretical optimum can be obtained relatively easily, although care must
be taken to assess and avoid datum errors and other blunders which may confound the analysis of DEMs of
difference. An important aspect of such surface inter-comparisons must be a sensitivity analysis to assess a
threshold level of detection that can be used to filter spurious changes which arise from errors differentially
incorporated in independent surveys. Here this was assessed through replicate surveys of a single surface,
although it should be recognized that this result might well be sensitive to the topographic complexity and
relief of the surface analysed. This approach can easily be extended to assess the external reliability (Chandler,
1999) of derived sediment transport or budget calculations as shown above, which in turn can be used to
identify a ‘geomorphological limit’ on the requisite data quality. This approach has been recently applied by
Brasington et al. (2001) within the context of sediment budgets in large braided rivers. In this case, their
analysis demonstrated the differential sensitivity with which patterns of scour (typically deep and localized)
and fill (shallow, extensive) can be identified in relation to data quality, and showed how morphometric
transport rates may become severely distorted as a result.
Some more general important methodological issues also arise. In particular, while this experiment shows
that it is possible to close the net sediment budget, inter-comparison of short-term sediment fluxes is less
straightforward, even for a closed system with no uncertainty in sediment inputs. These difficulties stem
from the incompatibility of temporal sampling intervals and while this does not challenge the fundamental
advantages of the morphometric approach, it does highlight the need to interpret time-integrated data dif-
ferently. Despite this, the research has outlined the significant potential of spatially distributed monitoring.
In particular, the ability to quantify volumes of erosion and deposition and determine the dynamics of the
sediment delivery ratio provides a significant new insight into basin processes and new data for numerical
model validation.
The evolution of the basin described here bears some similarities with other experimental studies (Schumm
et al., 1987; Hancock and Willgoose, 2001b), in which landscape development is dominated by headward
erosion of channel heads (Howard, 1971) driven by topography–flow feedback. More significantly, this
research has shown that this small-scale landscape displays not only qualitative morphological similarity
with larger-scale landforms, but also demonstrates similar internal patterns of response, in particular the
closely coupled behaviour of sediment transport in headwaters and trunk channels, which emphasizes the
need for a whole-basin perspective to understand local geomorphological dynamics. Current and on-going
experimentation is designed to extend this approach to evaluate the pattern of sediment translocation once an
established channel network has become established, and in particular to evaluate the response to baselevel
perturbations. A further significant area for future research lies in the use of sediment transport data derived
from this type of experimental design to validate and refine numerical models of landform development, which
complements the recent approach of Hancock and Willgoose (2001c). Ultimately such combined experimental
and numerical research offers the potential to test the recent theoretical concepts of self-organization and scale
invariance in basin dynamics (Rodriguez-Iturbe and Rinaldo, 1997).

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the support of Jim Chandler, Loughborough University, particularly for
his advice in the photogrammetric processing and survey data acquisition. Access to and help with the self-
calibrating bundle adjustment software, CUBA, was kindly provided by Stuart Robson. The experimental
research was undertaken with the support of the University of Hull, Department of Geography. Finally we
would like to thank Stuart Lane and one other anonymous referee whose constructive comments helped to
refine an earlier draft of this paper.

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)
PHOTOGRAMMETRIC ANALYSIS OF DRAINAGE BASIN EVOLUTION 247
REFERENCES
Ashmore PE, Church MA. 1998. Sediment transport and river morphology: a paradigm for study. In Gravel Bed-Rivers in the
Environment, Klingeman PC, Beschta RL, Komar PD, Bradley JB (eds). Water Resources Publications: Oregon; 115–139.
Brasington J, Rumsby BT, McVey RA. 2000. Monitoring and modelling morphological change in braided gravel-bed river using high
resolution GPS-based survey. Earth Surface Processes and Landforms 25: 973–990.
Brasington J, Langham JA, Rumsby BT. 2001. Evaluating the sensitivity of morphometric estimates of sediment transport rate in large
braided rivers. Proceedings of SPIE. Remote Sensing for Environmental Monitoring, GIS Applications, and Geology 4345: 43–55.
Brunton DA, Bryan RB. 2000. Rill network development and sediment budgets. Earth Surface Processes and Landforms 25: 783–800.
Bryan RB, Poesen JWA. 1989. Laboratory experiments on the influence of slope length on runoff, percolation and rill development.
Earth Surface Processes and Landforms 14: 211–231.
Bryan RB, Rockwell DL. 1998. Water table control on rill initiation and implications for erosional response. Geomorphology 23:
151–169.
Butler JB, Lane SN, Chandler JH. 1998. Assessment of DEM quality characterizing surface roughness using close range digital
photogrammetry. Photogrammetric Record 19: 271–291.
Chandler J. 1999. Effective application of automated digital photogrammetry for geomorphological research. Earth Surface Processes
and Landforms 24: 51–63.
Chandler JH, Shiono K, Rameshwaren P, Lane SN. 2001. Measuring flume surfaces for hydraulics research using a Kodak DCS460.
Photogrammetric Record 17: 39–61.
Cooper MAR. 1987. Control surveys in Civil Engineering. Nichols Publishing Company: New York.
Dietrich WE, Dunne T. 1993. The channel head. In Channel Network Hydrology, Beven KJ, Kirkby MJ (eds). Wiley: Chichester;
175–219.
ERDAS. 1995. Imagine OrthoMAX Users Guide, Version 8.2 . Vision International and ERDAS, Inc.: Atlanta, Georgia.
Glock WS. 1931. The development of drainage systems: a synoptic view. Geographical Review 21: 475–482.
Goff JR, Ashmore PE. 1994. Gravel transport and morphological change in braided Sunwapta River, Alberta, Canada. Earth Surface
Processes and Landforms 19: 195–212.
Hancock GR, Willgoose GR. 2001a. The interaction between hydrology and geomorphology in a landscape simulator experiment.
Hydrological Processes 15: 115–133.
Hancock GR, Willgoose GR. 2001b. The production of digital elevation models for experimental model landscapes. Earth Surface
Processes and Landforms 26: 475–490.
Hancock GR, Willgoose GR. 2001c. Use of a landscape simulator in the validation of the SIBERIA catchment evolution model: declining
equilibrium landforms. Water Resources Research 37: 1981–1992.
Hooke RL. 1968. Model geology. Prototype and laboratory streams: Discussion. Geological Society of America Bulletin 79: 391–394.
Howard AD. 1971. Simulation of stream networks by headward growth and branching. Geographical Analysis 3: 29–50.
Lane SN, Richards KS, Chandler JH. 1993. Developments in photogrammetry; the geomorphological potential. Progress in Physical
Geography 17: 306–328.
Lane SN, Chandler JH, Richards KS. 1995. Morphological estimation of the time-integrated bedload transport rate. Water Resources
Research 31: 761–772.
Lane SN, James TD, Crowell MD. 2000. Application of digital photogrammetry to complex topography for geomorphological research.
Photogrammetric Record 16: 793–821.
Laws JO, Parsons DA. 1943. The relation of raindrop-size to intensity. Transactions of the American Geophysical Union 24: 452–460.
Lindsay JB, Ashmore PE. 2002. The effects of survey frequency on estimates of scour and fill in a braided river model. Earth Surface
Processes and Landforms 27: 27–43.
Moss AJ, Green P, Hutka J. 1982. Small channels: their experimental formation, nature and significance. Earth Surface Processes and
Landforms 7: 401–416.
Parker RS. 1977. Experimental study of basin evolution and its hydrologic implications. PhD Thesis, Colorado State University, Fort
Collins.
Richards KS. 1993. Sediment delivery and the drainage network. In Channel Network Hydrology, Beven KJ, Kirkby MJ (eds). Wiley:
Chichester; 220–254.
Rodriguez-Iturbe RE, Rinaldo A. 1997. Fractal River Basins: Chance and Self-organization. Cambridge University Press: New York.
Schumm SA, Hadley RF. 1957. Arroyos and the semiarid cycle of erosion. American Journal of Science 255: 161–174.
Schumm SA, Mosley MP, Weaver WE. 1987. Experimental Fluvial Geomorphology. Wiley: New York.
Shortis MR, Robson S, Beyer HA. 1998. Principal point behaviour and calibration parameter models for Kodak DCS cameras.
Photogrammetric Record 16(92): 165–186.
Slattery MC, Bryan RB. 1991. Hydraulic conditions for rill incision under simulated rainfall: a laboratory experiment. Earth Surface
Processes and Landforms 16: 383–398.
Stojic M, Chandler J, Ashmore P, Luce J. 1998. The assessment of sediment transport rates by automated digital photogrammetry.
Photogrammetric Engineering and Remote Sensing 64: 387–395.
Taylor JR. 1982. An Introduction to Error Analysis. Oxford University Press: Oxford.
Westaway RM, Lane SN, Hicks DM. 2000. The development of an automated correction procedure for digital photogrammetry for the
study of wide, shallow, gravel-bed rivers. Earth Surface Processes and Landforms 25: 209–226.
Willgoose GR, Bras RL, Rodriguez-Iturbe I. 1991. A physically-based coupled network growth and hillslope evolution model: 1. Theory.
Water Resources Research 27: 1671–1684.
Willgoose G. 1994. A statistic for testing the elevation characteristics of landscape evolution models. Journal of Geomorphological
Research 99: 13 987–13 996.

Copyright  2003 John Wiley & Sons, Ltd. Earth Surf. Process. Landforms 28, 231–247 (2003)

You might also like