Available via license: CC BY 4.0
Content may be subject to copyright.
1
A low-cost technique to measure bank erosion processes along
middle-size river reaches
Gonzalo Duró
1
, Alessandra Crosato
1,2
, Maarten G. Kleinhans
3
, Wim S. J. Uijttewaal
1
1
Department of Hydraulic Engineering, Delft University of Technology, PO Box 5048, 2600 GA Delft, the Netherlands
2
Department of Water Engineering, IHE-Delft, PO Box 3015, 2601 DA Delft, the Netherlands 5
3
Department of Physical Geography, Utrecht University, PO Box 80115, 3508 TC Utrecht, the Netherlands
Correspondence to: Gonzalo Duró (G.Duro@tudelft.nl)
Diverse methods are currently available to measure river bank erosion at broad-ranging temporal and spatial scales. Yet, no
technique provides low-cost and high-resolution to survey small-scale bank processes along a river reach. We investigate the
capabilities of Structure-from-Motion photogrammetry applied with imagery from an Unmanned Aerial Vehicle (UAV) to 10
describe the evolution of riverbank profiles in middle-size rivers. The bank erosion cycle is used as a reference to assess the
applicability of different techniques. We surveyed 1.2 km of a restored bank of the Meuse River eight times within a year,
combining different photograph perspectives and overlaps to identify an efficient UAV flight to monitor banks. The
accuracy of the Digital Surface Models (DSMs) was evaluated compared with RTK GPS points and an Airborne Laser
Scanning (ALS) of the whole reach. An oblique perspective with eight photo overlaps was sufficient to achieve the highest 15
relative precision to observation distance of ~1:1400, with 10 cm error range. A complementary nadiral view increased
coverage behind bank toe vegetation. The DSM and ALS had comparable accuracies except on banks, where the latter
overestimates elevations. Sequential DSMs captured signatures of the erosion cycle such as mass failures, slump-block
deposition, and bank undermining. Although this technique requires low water levels and banks without dense vegetation, it
is a low-cost method to survey reach-scale riverbanks in sufficient resolution to quantify bank retreat and identify 20
morphological features of the bank failure and erosion processes.
Keywords: Riverbank erosion monitoring, erosion cycle, restoration, Unmanned Aerial Vehicles (UAV), Structure from
Motion (SfM).
1 Introduction
Bank erosion is a fundamental process in morphologically active river systems, and much research has been devoted to 25
understanding, quantifying and modelling it from disciplines such as engineering, geomorphology, geology and ecology.
River bank erosion involves interconnected physical, chemical and biological processes (e.g., Hooke, 1979; ASCE, 1998;
Rinaldi and Darby, 2008), resulting in a complex phenomenon that is difficult to thoroughly understand and predict (e.g.,
Siviglia and Crosato, 2016). Predicting and monitoring bank erosion is necessary for sound river management strategies and
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
2
also important for both socio-economic problems, such as preventing material losses (e.g., Nardi et al., 2013), and
environmental challenges, for instance, promoting habitat diversity through river restoration (e.g., Florsheim et al., 2008) and
improving water quality (e.g., Reneau et al., 2004).
Bank erosion can be monitored with different spatial resolutions, time frequencies and accuracies. The techniques
that identify the temporal change in vertical bank profiles detect and quantify the different phases of the erosion cycle 5
(Thorne and Tovey, 1981). This characteristic helps distinguishing the factors influencing bank erosion and their relative role
in the whole process (e.g., Henshaw et al., 2013). On the other hand, a simple record of sequential mass failure events (see
Fukuoka, 1994, for a graph of failure-driven retreat) is sufficient to track rates of local bankline retreat and estimate eroded
volumes, but does not provide further information on the role of single factors governing the bank erosion process. In
navigable rivers, for instance, it is important to differentiate the effects of vessel-induced waves from the effects of river 10
flow, as well as those of high flows and water level fluctuations. This requires high spatial resolution and relatively frequent
measurements that usually involve expensive equipment and field logistics when monitoring large extensions. Still no
proven low-cost technique is capable of measuring bank erosion processes along extensive distances.
We investigate whether the resolution, precision and frequency of acquisition of Structure-from-Motion (SfM)
photogrammetry applied with imagery from a low-cost multi-rotor Unmanned Aerial Vehicle (UAV) is capable of 15
monitoring banks at the process scale along a middle-size river reach. In order to do that, we compare the SfM-based Digital
Surface Model (DSM) with Real-Time Kinematic (RTK) GPS measurements and Airborne Laser Scanning (ALS), and
analyse erosion features in bank profiles considering the erosion cycle as a reference to distinguish approaches that measure
bank erosion. The study site is a 1.2 km reach of the Meuse River near the city of Gennep, the Netherlands, which has
recently undergone a large bank-restoration project. The Meuse is a heavily regulated river used as navigation route to 20
connect the eastern part of Belgium and the Netherlands to the industrial area in the West and the port of Rotterdam. The
restoration aims to re-naturalize the previously protected banks, which are now allowed to erode. Here we take advantage of
knowledge of the original bank and of a rare event of extremely low water level.
2 Framework of analysis
2.1 Bank erosion cycle 25
Bank erosion may consist of three phases (Thorne and Tovey, 1981): fluvial entrainment of near-bank river-bed and bank
material, mass failure, and disintegration and removal of slump blocks. These three phases are particularly important for
cohesive banks since their retreat is typically delayed by the protection offered by slump blocks at their toe (Thorne, 1982;
Lawler, 1992; Parker et al., 2011). The waste material settles at the bank toe where it remains for a time depending on its
resistance to fluvial erosion and on the flow capacity to transport the blocks. In contrast, loose waste material from non-30
cohesive banks is generally transported away relatively quicker by the river flow, leaving the bank sooner unprotected.
Entrainment of near-bank bed material and the intact bank face occurs once the bank toe is exposed again (e.g., see Clark
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
3
and Wynn, 2007), which continues until the collapse of the upper bank. Mass failure occurs due to geotechnical instability,
which can be triggered by different factors, such as fluvial bank-toe erosion (e.g., Darby et al., 2007) or a rapid drawdown of
the river stage (e.g., Thorne and Tovey, 1981; Rinaldi et al., 2004).
Not only the river flow triggers the bank erosion cycle but other drivers can contribute to it as well. For instance,
subaerial processes may weaken the bank and accelerate later fluvial erosion (Lawler, 1992; Kimiaghalam et al., 2015) or 5
also act as direct agent of erosion (Couper and Maddock, 2001). These effects are included in the entrainment phase for the
former example and in the mass failure phase for the latter, which respectively promote entrainment or deliver material to the
bank toe. Figure 1 illustrates the three phases of erosion in schematic cohesive banks. This representation shows a
homogeneous soil which undergoes a continuous cycle of erosion with varying water levels.
10
Figure 1: Schematic bank erosion phases: Slump-block removal (left), entrainment of bare bank (right) and incipient mass failure (centre)
Bank erosion can be analysed and measured at two different scales, i.e., the fluvial process and the river cross 15
section. The measurement at the process scale considers the bank face disintegration over time with evidence of erosion
phases (Fig. 1): the mechanisms of erosion develop and are captured at the vertical dimension of the bank. The measurement
of bank erosion at the cross-sectional scale, which can be referred to as bankline retreat, consists of tracking banklines over
time. In this case, the focus is on the planimetric changes of the bank edge and estimations of eroded volumes and sediment
yield. The former approach deals with processes and mechanisms (e.g., Rinaldi and Darby, 2008), whereas the latter with 20
landscape development at larger spatial and temporal scales. Bank-erosion studies determine the survey method based on
their aims and scales of interest, whereas in turn a given methodology constraints the scope of the findings (Massey, 2001;
Couper, 2004). Thus, it is important to identify capabilities and limitations of each survey technique in the context of river
banks, which are inherently steep features with small-scale irregularities independent of the scale of the river.
EROSION
CYCLE
Entrainment
Mass failure
Slump
-
block removal
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
4
2.2 Techniques to measure bank erosion
Measuring techniques have four essential characteristics: the extent, resolution, precision and frequency of
measurements. Extent refers to the area or distance along the river covered by each survey; resolution indicates the distance
between surveyed points; precision is the accuracy of position of each surveyed point; and frequency derives from the time
interval between consecutive surveys of the same spatial extent or point. The scale of interest may vary among disciplines 5
(e.g., geomorphology, engineering, ecology), so that a diversity of techniques is available with varying spatio-temporal
windows of inquiry (Lawler, 1993). Even though the methods currently adopted to measure river bank erosion range from
photo-electric erosion pins to terrestrial laser scanning, they have high resolution in either time or space (Couper, 2004;
Rinaldi and Darby, 2008).
The methods to determine bankline retreat and to estimate eroded volumes are typical of remote sensing, for 10
instance, ALS and aerial photography. The former technique has typical resolutions of 1 and 0.5 metres, and covers up to
hundreds of square kilometres per day. Bailly et al., (2012) indicate decimetre vertical precision, which depends on several
factors including beam footprint size, aircraft inertial measuring system, on-board GPS, vegetation cover and filtering
technique. ALS has been successfully applied to identify river morphological features, such as bar tops (Charlton et al.,
2003) and riffle–pool and step–pool sequences (Cavalli et al., 2008). In addition, sequential ALSs were used to quantify 15
volumes of eroded banks to subsequently estimate pollutant loads, achieving reasonable results for those aims (Thoma et al.,
2005). However, banks are particularly steep areas where this technique tends to increase the elevation uncertainty (Bangen
et al., 2014). Therefore banks are regions where lower ALS accuracies are expected compared to horizontal and flat areas.
Aerial photography has also been applied to measure bank migration, which is a useful source of information,
especially if historical imagery is available over extended periods of time. Yet, it provides only limited information on bank 20
heights. Thus, this planform survey technique requires other methods to estimate eroded volumes. For example,
photogrammetry can serve to quantify volumetric changes from overlapping photographs (Lane et al., 2010); or ALS may
provide recent topographic elevations to reconstruct past morphologies (Rhoades et al., 2009). Bank retreat can also be
estimated through other approaches, such as those described by Lawler (1993), that include planimetric resurveys for
intermediate timescales (years) and sedimentological and botanical evidence for long timescales (centuries to millennia). 25
Measuring bank erosion at the process scale involves measuring the evolution of the vertical bank profile over time
and several techniques are currently available to that end. Traditional methods include erosion pins and repeated cross-
profiling, which provide two-dimensional information with resolutions that respectively depend on the number of pins and
points across the profile (Lawler, 1993). Erosion pins are simple and effective, but their accuracy may be affected by several
factors, such as subaerial processes (Couper et al., 2002). More advanced versions are the photo-electric erosion pins that 30
automatically track the bank face during different erosion phases (Lawler, 2005). Cross-profiling can be done with GPS or
total stations with point accuracies of a few centimetres or millimetres, yet with spatial and temporal resolutions that may not
be sensitive to very localized or intermittent erosion (e.g., Brasington et al., 2000).
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
5
Bank geometries can currently be surveyed with their three-dimensional complexity through a number of
techniques whose geomorphic applications are broader than bank erosion studies: terrestrial photogrammetry, Terrestrial
Laser Scanning (TLS), boat-based laser scanning and SfM photogrammetry. Terrestrial photogrammetry has shown detailed
bank representations, with approximate resolutions of 2 cm and precision within 3 cm, covering up to 60 metres of banks
(Barker et al., 1997; Pyle et al., 1997). Yet, this method can be labour-intensive and requires an accessible bank (Bird et al., 5
2010), known camera positions and sensor characteristics, ground-control points, among other considerations (Lane, 2000).
TLS has shown detailed erosion patterns from sequential surveys, with millimetre resolutions, which in practice are usually
reduced to 2–5 centimetres, and approximate final accuracies of 2 cm (Resop and Hession, 2010; Leyland et al., 2015).
O’Neal and Pizzuto (2011) proved the advantages of 3D TLS in capturing patterns (e.g., overhanging blocks) and
quantifying eroded volumes over 2D cross-profiling. Even though TLS could cover thousands of meters, in practice the 10
extents are generally smaller due to accuracy decrease, large incidence angles, occlusion, etc. (Telling et al., 2017), so
several scans are necessary to measure long distances. For instance, Brasington et al. (2012) surveyed a 1 km river reach
scanning every 200 m along the channel. An alternative boat-based laser scanning can continually survey banks with
comparable resolutions and accuracies to those of TLS, with great time reduction but involving other field logistics,
resources and post-processing (Alho et al., 2009). 15
SfM photogrammetry has been applied to measure banks to show its potential use as survey technique with different
sensors and processing systems (Micheletti et al., 2015; Prosdocimi et al., 2015). Micheletti et al. (2015) indicated root mean
square errors (RMSE) within 7 cm, when combining a 5MP smartphone or a 16MP reflex camera with either PhotoModeler
or 123D Catch processing systems. Prosdocimi et al. (2015) identified eroded areas of a collapsed riverbank and computed
eroded and deposited volumes with a precision comparable to that of TLS. Bangen et al., (2014) matched the resolution and 20
practical extent of this technique to those of TLS, when SfM photogrammetry is used to survey river topography through
aerial platforms (e.g., Fonstad et al., 2013). The relatively recent and fast development of UAV technology to take airborne
photographs has greatly expanded the applications of SfM photogrammetry (Eltner et al., 2016). Recently, SfM has been
applied to quantify bank retreat at streams and small rivers with a fixed-wing UAV along several kilometres with 12 cm
resolution (Hamshaw et al., 2017). This study showed the UAV-SfM capabilities to produce extensive 2.5D DSM from a 25
100 m high nadiral view, which achieved 0.11 m mean error and 0.33 m RMSE compared to TLS. However, this work
generated DSMs similar to those of ALS, which allow for volume computations and bankline retreat, but did not use the full
3D capacities to investigate undermined banks or identify erosion processes.
Applications of this combined technology span in scale and complexity, covering glacial dynamics (Immerzeel et
al., 2014), landslides (Turner et al., 2015), agricultural watersheds (Ouédraogo et al., 2014), fluvial topography (Woodget et 30
al., 2015), etc. The accuracy achieved relative to the camera-object distance for the mentioned diverse settings was
approximately 1:1000, with distances ranging from 26 to 300 m and different cameras, lighting conditions and surface types.
Interestingly, this precision was also found for terrestrial SfM photogrammetry at different scales by James and Robson
(2012). However, other experiences showed lower accuracies, e.g., ~1:200 for moraine-mound topography (Tonkin et al.,
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
6
2014), and on the other hand higher ones, such as ~1:2100 for fluvial changes after a flood event (Tamminga et al., 2015).
Although it is not possible to generalize a precision for all settings, ~1:1000 seems an encouraging reference (RMSE of
10 cm for 100 m camera-object distance) to consider for unexplored conditions.
Every combination of field site, camera sensor, GCPs and SfM package in principle requires different photo
overlaps, resolutions and perspectives (image network geometry) to achieve certain model accuracy and resolution through 5
UAV-SfM (Elter et al., 2016). This is caused by different surface textures (Cook, 2017), lighting conditions (Gómez-
Gutierrez et al., 2014), camera characteristics (Prosdocimi et al., 2015), GCP characteristics (Harwin and Lucieer, 2012), and
SfM algorithms (Eltner and Schneider, 2015). Knowledge to improve the quality of SfM digital surface models keeps
expanding by investigating isolated variables, for example, assessing the influences of number and distribution of GCPs
(Clapuyt et al., 2016; James et al., 2017) or optimizing camera calibration procedures to manage without GCPs (Carbonneau 10
and Dietrich, 2017). The flexibility, range, high resolution and accuracy that UAV-SfM proved in other conditions shows
promising for analysing bank erosion processes throughout the scale of a middle-size river.
3 Methodology
We used the flexibility of a multi-rotor UAV platform to capture photographs from different perspectives of a
1200 m long riverbank and through SfM photogrammetry derived several DSMs over one year period. We describe the study 15
location in Sect. 3.1, the UAV paths for photo acquisition in Sect. 3.2, and the SfM imagery processing in Sect. 3.3. In order
to assess the capabilities of this survey methodology to measure bank erosion at the process scale we proceeded in three
steps. First, we verified the elevation precision against 129 RTK GPS points of several DSMs obtained with diverse number
of photographs and camera orientations. In this way, we identified an effective number of images to acquire the bank
topography with high accuracy. Second, we compared the chosen DSM with airborne LIDAR points to analyse elevation 20
precision over the whole river reach, differentiating between areas of bare ground, grassland and banks. Third, we searched
for bank features in SfM-based profiles and analogous ones from ALS, and for signatures of erosion processes along
sequential SfM surveys.
For the first step, the analysis of the minimum number of photographs needed to achieve the highest DSM
precision, we compared the DSMs with RTK GPS measurements to quantify vertical accuracy. We took 129 points across 25
eight profiles on 18-01-2017 (see Fig. 4) with a Leica GS14 RTK GPS, whose root mean square precisions according to the
manufacturer specifications are 8 mm + 0.5 ppm in horizontal and 15 mm + 0.5 ppm in vertical directions. On the same date,
we flew the UAV along the bank four times with different camera angles and perspectives. Eight photograph combinations
were considered to derive 8 DSMs. Then, the comparisons were done with the elevation differences between the GPS points
and the corresponding closest ones of the DSM point clouds (e.g., Westoby et al, 2012; Micheletti et al., 2015). We used 30
CloudCompare software (Girardeau-Montaut, 2017) for these computations.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
7
In the second step, we compared the selected DSM from the previous analysis with a reach-scale survey technique,
ALS, to analyse elevation differences over the whole river reach. The ALS was carried out on 17-01-2017 from an airplane
at 300 meters above the ground level. The laser scanner, a Riegl LMS-Q680i, measured a minimum of 10 points per square
metre with an effective pulse rate of 266 kHz and automatically generated a 0.5 m grid. We tested the ALS elevation
precision against the 129 RTK GPS points using a Delaunay triangulation of the ALS grid, due to the different resolutions 5
between both datasets. Then, we compared the elevation of the ALS grid points with the corresponding nearest ones of the
DSM point cloud. We did both computations with CloudCompare, distinguishing between surfaces of grassland, bare ground
and bank.
Third, we made profiles across six sections of dissimilar erosion rates to contrast the bank representations of i) the
SfM DSM, ii) the triangulated ALS grid, and iii) the RTK GPS points. The profiles were computed with MATLAB using i) 10
the Geometry Processing Toolbox (Jacobson et al., 2017) adapted to slice triangle meshes, ii) a linear interpolation across the
triangulated ALS grid, and iii) a projection of the RTK GPS points onto the exact cross-section locations. Then, we
identified and analysed a cross section over which sequential SfM-UAV surveys showed different stages of the erosion
cycle, since the bank erosion cycle was used as a reference to distinguish between techniques capable of measuring at either
the process or the cross-sectional scale. 15
3.1 Study site
The study site is a restored reach of the Meuse River, which used to be a single-thread freely meandering river. The river was
canalized to a straight reach of 120 m width, the banks were protected and the water levels regulated to improve navigability.
However, several kilometres of banks have been recently restored through the removal of revetments and groynes, following
the EU Water Framework Directive 2000/60/EC (http://data.europa.eu/eli/dir/2000/60/oj). This reactivated erosion processes 20
to improve the natural value of the river. Important questions have then arisen regarding bank retreat rates and the new
equilibrium of the river width. Monitoring bank evolution is necessary to answer these questions and to identify the need, if
so, to intervene and at which locations.
The study site is the left bank of a 1200 m long straight reach (Fig. 2) located between the Sambeek and Grave
weirs in southeast Netherlands. Seven years after restoration, this reach presents different bank retreat patterns, with sub-25
reaches of rather uniform erosion and others with embayments of different lengths. Grassy fields used for grazing cover the
riparian zone, followed by crop fields across the floodplain. In the near-bank area there are poplar trees every 100 m, some
of which have been dislodged during the erosion progression, which is possible to appreciate in Fig. 2 (left) considering the
~200 m embayment between the foreground tree and the next in the background.
30
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
8
Figure 2: Restored bank in the Meuse River, upstream view of first 300 m (left) and downstream view of middle 500 m (right). Bank
erosion has caused a series of bays in the 1,200 meter restored reach. Note eroded bank sediment in suspension.
The study site was surveyed eight times with a UAV in 2017. An extraordinary low water level in January provided
the opportunity to compare the SfM photogrammetry with ALS and RTK GPS not only for the banks and floodplain, but 5
also for the sub-aqueous terrace at the bank toe (see schematic cross sections in Fig. 2). This terrace was composed of bare
soil, without vegetation or obstructions, which adds an extra surface for the comparative analysis. This extraordinary
exposure was the consequence of a ship accident against the downstream weir of Grave (on 30 December 2016).
3.2 UAV flights for image acquisition
We used the low-cost UAV DJI Phantom 4 to take images of the banks. It has a built-in camera with a 1/2.3” 12 megapixels 10
sensor and a 94° horizontal angle of view. Prior to the image acquisition, a network of Ground-Control Points (GCPs) was
distributed every approximately 50 metres on the floodplain to georeference the DSMs (see Fig. 4). The GCPs were black
ceramic tiles fixed to the ground with a circular reflector (12 cm CD) at its centre for their fast recognition in the
photographs. We measured the GCP coordinates using the Leica GS14 RTK GPS unit, which was also deployed for the
cross-profiling. 15
An initial flight plan was designed to photograph the banks from four different perspectives, to later compare the
results of diverse combinations and find a convenient photo set to survey the target topography in subsequent campaigns.
The UAV flew four times in straight parallel lines along the banks (Fig. 3). The first track took oblique photos from above
the river at a height of 25 metres and an average (oblique) distance to the bank of 40 metres (~25 m from the least retreated
bankline). The second track had a top view from 40 metres above the floodplain level along the tree line (Fig. 2–4). The third 20
and fourth tracks followed the same path as the second one in respective upstream and downstream directions, but the
camera angle was 50 degrees forward inclined from the horizontal plane. These perspectives were thought to capture the
tortuous and complex bank surface (Fig. 2 and 3); including undermined upstream- and downstream-facing scarps, with an
average ground resolution of 1.7 cm per pixel.
We tested five specific combinations of photographs from the different UAV tracks. Test 1 corresponds to the photo 25
set of the first track only, which has the side view with the optimal coverage of the bank. Test 2 stands for the nadir view
Waterway
Terrace at
bank toe
Floodplain
Terrace at
bank toe
Floodplain
Waterway
sub-aqueous
cross section
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
9
alone, which is similar to the viewpoint of classic aerial photography. Test 3 is a combination of the previous two sets. Test 4
combines tracks 3 and 4, i.e. both paths from above the bank with the oblique forward perspectives in upstream and
downstream direction, which allows views on all parts of the irregular banks. Finally, test 5 utilizes the four tracks with all
photographs (Table 1).
We also used the first oblique track to evaluate the minimum longitudinal photo overlap to efficiently capture the 5
bank relief. The photo overlap along the river is a function of the UAV speed and distance to the bank, for a given maximum
photo sampling frequency, which in the case of the deployed UAV is one every 2 seconds. Then, flying at 2 m/s along
track 1 resulted in 20 photo overlaps for the most retreated areas and 16 for those zones with least bank retreat. Afterwards in
the processing phase, we successively selected a decreasing number of overlaps by twos that resulted in four DSMs. These
were test 1a when using all photos from track 1 (which is the same set as the aforementioned test 1), test 1b when using half 10
of them, and so forth for test 1c and test 1d (see Table 1).
Table 1. Number of photographs and overlaps for the tests
Test 1a Test 1b Test 1c Test 1d Test 2 Test 3 Test 4 Test 5
Track 1 293 147 73 37
147
293
Track 2
232 232
232
Track 3
232 232
Track 4 232 232
Min. overlaps 16 8 4 2 7 15 26 49
Max. overlaps 20 10 5 2 7 17 26 53
3.3 SfM imagery processing
The principles of SfM photogrammetry are similar to those of digital photogrammetry, but the former does not need 15
specifications on camera positions and lens characteristics to reconstruct 3D structures. The camera extrinsic and intrinsic
parameters are automatically estimated via tracking and matching pre-defined features in overlapping photos and an iterative
bundle adjustment procedure, which results in a sparse point cloud (Hartley and Zisserman, 2003; Snavely et al., 2008;
Westoby et al., 2012). Afterwards, the (dense) point matching is done at pixel scale to generate a detailed point cloud of the
scene that has the final survey resolution. The point cloud can then be georeferenced with GCPs, which is necessary when 20
monitoring bank erosion through sequential surveys. Alternatively, GCPs can be incorporated for the iterative bundle
adjustment as additional matched points, during which the georeferentiation takes place.
We used Agisoft PhotoScan software to process the imagery. For a successful photo alignment from different UAV
tracks (Table 1), the camera yaw, pitch and roll recorded during the UAV flight were necessary inputs. For this step we used
three GCPs along the reach, two at the extremes and one at the middle. These approximate orientations and a priori known 25
ground points helped obtaining a consistent sparse point cloud of the bank along the entire reach. The resulting camera
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
10
positions and orientations of the photo alignment are visible in Fig. 3, evidencing the UAV tracks. This figure also shows the
DSM textured with colours from the photographs, in which the green area on the left side with white patches corresponds to
the floodplain partially covered with snow and the right brownish area is the terrace at the bank toe, with snow remains as
well.
5
Figure 3: Camera positions and orientations in perspective view. The digital surface model shows the low-water condition during January
2017, which exposed a terrace at the bank toe.
After obtaining the sparse point cloud, we marked the remaining 15 GCPs (Fig. 4). Then, we refined the camera
parameters by minimizing the sum of GCP reprojection and misalignment errors. This camera optimization adjusts the
estimated point cloud by reducing non-linear deformations. Once the dense point cloud was computed, we removed the 10
points outside the area of interest, as well as those points at the water surface, tree canopies and individual bushes at the
floodplain. Finally, the point cloud was triangulated and interpolated to generate a triangle mesh. This mesh consisted of a
non-monotonic surface that was later processed in MATLAB to plot 2D cross sections.
Figure 4: Study reach of the Meuse River with GCPs, RTK GPS measurements and cross-section locations and numbers
15
Flow direction
in main channel
Track 4
Track 2
Track 3
Track 1
Floodplain
Terrace at
the bank toe
Bank
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
11
4 Results
4.1 DSM precision: identifying necessary photographs
The sequentially decreasing photo overlaps of Track 1 (Table 1) produced four DSMs, tests 1a–1d, whose elevation
differences with the 129 RTK GPS points are presented in the histograms of Fig. 5. The elevation errors mostly ranged
within 10 cm in all tests, but the mean and standard deviation (SD) presented some differences (Table 2 and dot with bar in 5
Fig. 5). Tests 1a, 1b and 1c presented mean values smaller than 1 cm and SD within 3–4 cm, while for test 1d these values
increased to 4 cm and 7 cm respectively (Table 2, rows 1–2). The mean errors on the bank area alone for test 1a, 1b and 1c
were lower than 1 cm (Table 2, row 4), but test 1c had a higher SD of 7 cm compared to 4 and 3 cm of tests 1a and 1b
respectively (Table 2, row 7). Then, tests 1a and 1b had the highest precisions and showed little error differences between
them: less than 1 cm for all values in Table 2. Consequently, test 1b with eight photo overlaps was as effective as test 1a 10
with 16 overlaps to achieve the highest DSM accuracy. In addition, test 1b fully covered the tortuous bank area in contrast to
test 1c, especially at the perpendicular stretches of embayments (Fig. 3–4), which assured the choice of 8 image overlaps
over 4, despite the general close performance of the latter in terms of accuracy (Table 2, all rows). Therefore, test 1b became
the reference for tests 1 and was used in combination with test 2 to generate test 3.
15
Figure 5: Elevation error distributions for SfM tests 1a, 1b, 1c, and 1d, assuming that the RTK points are correct and without error.
Indicated overlaps are the minimum.
Table 2. Mean and standard deviation of elevation differences between SfM DSMs and GPS points
Surface Error (m) Test 1a Test 1b Test 1c Test 1d Test 2 Test 3 Test 4 Test 5
All grounds Mean -0.01 0.00 0.00 0.04 0.03 -0.01 -0.05 0.00
Std. dev. 0.03 0.03 0.04 0.07 0.03 0.03 0.05 0.03
Grassland Mean 0.02 0.01 0.01 0.02 0.02 0.01 0.01 0.02
Bank
0.00 0.01 -0.01 -0.01 0.05 0.01 -0.03 0.01
Terrace
-0.02 -0.02 0.00 0.06 0.03 -0.02 -0.09 -0.01
Grassland Std. dev. 0.03 0.03 0.04 0.04 0.03 0.02 0.02 0.02
Bank
0.04 0.03 0.07 0.13 0.03 0.03 0.04 0.03
Terrace
0.02 0.03 0.03 0.06 0.03 0.02 0.03 0.02
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
12
Figure 6 shows the error distributions of the remaining four DSMs, i.e. tests 2–5, which also were mostly within
10 cm, except for Test 4. This test had evident higher errors than the rest, mostly concentrated at the terrace (Table 2, row 5).
Tests 3 and 5 had the lowest mean elevation errors, both lower than 1 cm, with the same SD at all surfaces that were lower
than 3 cm. Test 2 presented a similar SD, but the mean was biased 3 cm. This test in combination with test 1b slightly
reduced the SD errors of the latter (Table 2, rows 5 and 7), but without significant overall improvements. All in all, tests 1b, 5
3 and 5 had the best performances with average errors lower than 1 cm and standard deviations within 3 cm, however with
increasing number of photographs (Table 1). The most efficient one was then test 1b that used the lowest number of
photographs to achieve similar precision, especially on banks.
Figure 6: Elevation error distribution for tests 2, 3, 4, and 5. Ordinates indicate number of GPS points in each bin.
10
Interestingly, if we consider all tests, the elevation errors on grassland were similar to each other (Table 2, rows 3
and 6), means between 1 and 2 cm and SD between 2 and 4 cm, whereas the bank and terrace did not present this behaviour.
Furthermore, while the bank values (Table 2, rows 4 and 7) did not correlate with those of all grounds (Table 2, rows 1–2),
the terrace mean elevation differences (Table 2, row 5) linearly correlated with those of all grounds (Table 2, row 1) with R
2
= 0.97. Therefore, the error biases for all grounds throughout the tests were most likely due to the biases from the points over 15
the terrace.
To conclude, despite virtually doubling the number of images in comparison with test 1b, the test 3 setup with a
nadir track and a side-looking track was chosen for subsequent UAV surveys on the basis of two findings. First and most
important, growing vegetation at the bank toe occluded parts of the target surface from the oblique camera perspective.
Second, the GCPs on the floodplain laid almost horizontal, which made them easier to identify from the top-view during an 20
initial phase of GCP recognition in the photographs. Moreover, we found at later surveys that growing grass on the
floodplain was sometimes blocking GCP plaques from the angle of vision of UAV track 1, for which using the nadir view of
track 2 was advantageous to locate the plaque centres, preventing the otherwise disuse of some GCPs.
4.2 DSM precision over the reach: comparison with ALS
Compared to the ALS grid, test 3 point cloud showed a good agreement over most of the reach. This is observable 25
from Fig. 7, corresponding to the blue areas that indicate elevation differences lower than 5 cm. Yet, two regions surpassed
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
13
this difference, notably the bank and the extremes of the reach. The latter were zones beyond the GCPs, where higher errors
in the DSM are expected when using parallel image directions due to inaccurate correction of radial lens distortion (James
and Robson, 2014; Smith et al., 2014). Consequently, the results outside the GCP limits cannot be considered representative
of the whole domain and they were discarded for the subsequent statistical comparisons. Within the GCP bounds, the bank
area presented relatively high elevation differences, which makes the bankline visible in Fig. 7. 5
Figure 7: Absolute elevation differences between SfM and ALS along the reach. Banks and areas beyond GCPs presented the highest
differences.
Figure 8 presents the relative frequency distributions of the elevation differences divided into three regions: the
grassy floodplain, the steep bank, and the bare-ground terrace. Over the grassland, both SfM and ALS had rather similar 10
results (Fig. 8, centre left), with a zero mean difference and 3 cm of standard deviation (Table 3). On the contrary, the bank
had a bias between the techniques of 6 cm (Table 3) and a relatively high standard deviation of the same value. Finally, the
terrace showed similar results to those over the grassland regarding the deviation (Fig. 8 and Table 3) but with a bias of -
4 cm. The bank area together with the terrace dominate the overall bias in the elevation differences (Fig. 8, left). The former
with a small contribution to the total number of measurements and the latter with a greater number but a lower magnitude. 15
Figure 8: Comparison of elevation differences between SfM and ALS for distinct surface types
Table 3 also indicates the differences of SfM DSM and the ALS with the 129 RTK GPS points. Interestingly, the
ALS presented a constant bias of 1 cm across all surfaces, but the standard deviation did change significantly among them: 20
the bank had a standard deviation of 9 cm, which doubled the deviation of the terrace and tripled that of the grassland. While
the SFM DSM had comparable absolute biases than those of ALS, the standard deviations were all respectively lower.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
14
Particularly at banks, the standard deviation of the SfM DSM was only 3 cm in contrast to the 9 cm of the ALS, which
makes the former approach considerably more accurate than the latter. This could explain the relatively large elevation
differences between the two methods in the bank area (Fig. 8, centre right), occurring due to a lower precision of the ALS
and not vice versa.
5
Table 3. Mean and standard deviation of elevation differences between SfM, ALS and RTK GPS.
Subtraction All
grounds Grassland Bank Terrace
SfM - ALS (m) Mean -0.02 0.00 -0.06 -0.04
Std. dev. 0.04 0.03 0.06 0.03
ALS - GPS (m) Mean 0.01 0.01 0.01 0.01
Std. dev. 0.05 0.03 0.09 0.05
SfM - GPS (m) Mean -0.01 0.01 0.01 -0.02
Std. dev. 0.03 0.02 0.03 0.02
4.3 Bank erosion features and process identification
Six bank profiles were selected among those surveyed with GPS on January 2017 (see Fig. 4) to compare the bank
representation with the different survey techniques. Fig. 9 shows the bank profiles at Sections 1, 2, 4, 6, 7 and 8. These 10
sections presented distinct erosion magnitudes and features after seven years of restoration, for example, Section 8 (Fig. 4
and 9) appeared close to the original condition, with a mild slope and nearly no erosion, whereas Sections 6 and 7 had
vertical scarps. The SfM DSM profiles are represented by continuous lines, the ALS profiles with dashed lines, and GPS
points with circles. The SfM representation had better proximity to the GPS points than the ALS in almost all cases. What is
more, ALS generally overestimated the elevation corresponding to the GPS points, which confirms the bias observed in the 15
comparison of bank elevations shown in Fig. 7 and 8.
SfM profiles showed detailed bank features, such as a collapsed upper bank laying at the toe (Section 2), an
overhang at the bank top (Section 1), small-scale roughness on scarps (Sections 6 and 7), and slump-block deposits (Section
4). These features appeared as simple shapes in the profiles but they were confirmed with field observations. The ALS
depicted simpler profiles, smoothed by coarser resolution, which made it difficult to identify characteristic features of the 20
erosion cycle in them. Yet, ALS profiles had enough point spacing to capture gentle bank slopes with reasonable precision
(Section 8), but for steeper ones (Sections 1, 2 and 4) and specially at scarps (Sections 6 and 7), this technique provides
lower accuracies.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
15
Figure 9: Banks measured with SfM (continuous lines), ALS (dashed lines) and GPS (circles) on 17
–
18 January 2017. Cross-sections are
located from left to right, at river km. 153.4, 153.9, 153.5, 154.2, 154.1, and 154.3 (see Fig. 4 for locations).
The temporal development of Section 4 (Fig. 4 and 9) is illustrated in Fig. 10a by a sequence of SfM-UAV surveys. 5
The initial stage corresponds to the survey of Fig. 9 on 18 January 2017. The consecutive surveys showed the evolution of
the vertical bank profile, through which different processes can be inferred. The bank profile, initially characterized by a top
short scarp and slump blocks along the bank face, experienced a mass failure and a further removal of blocks between
January 18 and March 15 2017. Between March 15 and April 26, only toe erosion occurred. By June 21, another mass failure
happened, which left slump blocks along the lower half of the bank. On July 19, these blocks were removed, leaving a steep 10
bank face. Then, further toe erosion caused a small soil failure at the lower bank whose remains laid at the toe. On October
11, this wasted material was removed. Then, until the last survey on November 22, entrainment occurred at the lower half of
the bank profile, further steepening the bank. In light of the results, the methodology resolution and accuracy are high
enough to identify different phases of the erosion cycle, enabling the analysis of bank erosion processes in conjunction with
data on potential drivers. 15
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
16
Figure 10: Sequential surveys at cross section 4, Meuse River km. 153.9, over 2017. a) Bank profiles from DSMs; b) Eroded volume per
unit width between consecutive surveys; c) Cumulative erosion along surveys; d) Bankline locations at the top and toe of the bank.
In addition, the quantification of eroded volumes is possible computing the net area between sequential bank 5
profiles. For example, Fig. 10b shows eroded volumes per unit width between consecutive surveys plotted at the end of each
time interval, with an error bar based on the RMSE of test 3. Evidently, there were different erosion rates during the year and
the highest ones happened in the first part of it. Fig. 10c presents the respective cumulative eroded volumes per unit width,
where the two trends can be distinguished: a gentle slope towards the end and higher rates of sediment yield during the first
haft of the year. Given that the topographic measurements are limited to a single year, it is not possible to state whether this 10
behaviour in recurrent on a yearly basis. However, this case exemplifies the possibilities to quantify eroded volumes
throughout different phases of the erosion cycle.
The bankline retreat as a measure of bank erosion involves the identification over time of the bank top, but this
concept could be extended, for instance, to the bank toe. Fig. 10d shows the temporal progression of the bankline distance
from the river axis for both the bank top and the toe, which we arbitrarily defined for this case at 11.1 m and 8.1 m, 15
respectively. The top bankline showed a mayor jump between April and June and a smaller one between the first two
surveys, corresponding to mass failure events. The bank toe presented a more gradual retreat, with events of slumping and
temporal accretion that were timely captured along the surveys. This alternative bank retreat representation provides
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
17
evidence of the development of bank erosion at every survey. The contrast of bankline retreats at the top and toe of the bank
illustrates how different processes on their own represent dissimilar erosion evolutions, since they constitute different phases
of the erosion cycle, i.e. at the top mass failures and the toe slump block removal and entrainment. Finally, the average
bankline between bank toe and top would best represent the real retreat (dashed line in Fig. 10d), despite not necessarily
indicating an actual bank location for a specific elevation. This approach logically considers all erosion phases and follows a 5
similar trend as the cumulative erosion of Fig. 10c.
5 Discussion
5.1 UAV flight and SfM precision
In general, there were no large differences in accuracy between the DSMs derived with different photo perspectives and
overlaps. The accuracy of the tests, except for test 4, was approximately 10 cm and they all had sufficient resolution to 10
represent characteristic features of the erosion cycle, such as slump block deposited at the bank toe and mass failures. Yet,
other topographic features that were hidden from the nadir UAV perspective, such as undermining, were only captured from
oblique camera perspectives. For instance, the area below the top overhangs visible at cross sections 1 and 2 (Fig. 9) were
not captured in test 2, and were represented with a lower resolution in test 4. The UAV viewpoint of track 1 not only had the
largest bank area coverage compared to the other camera perspectives proposed in this work, but also achieved the highest 15
elevation precision without the need of other tracks. Yet, the nadir view of track 2 contributed to cover an additional bank
area behind trees and bushes growing at the bank toe along the first 200 m of the reach (Fig. 2, left), for which it was
complementarily used with track 1. Since vegetation can occlude the bank face, if denser and more abundant, it could
prevent the usage of the survey technique, in a similar way as high water levels do.
The results herein show that, in the absence of bank toe vegetation, a single oblique UAV track with eight photo 20
overlaps and visible GCPs appears effective to survey banks with the highest precision and coverage, for the given sensor
size and resolution, camera-object distance and lighting conditions. This number of photo overlaps agrees with the laboratory
experiment of Micheletti et al. (2015), who found that above eight the mean error was only slightly decreased, in contrast to
increasing overlaps within the range below eight. Nevertheless, they showed that overlaps higher than eight reduced the
number of outliers, a trend which in our case is evident for less overlaps: test 1c (4 overlaps) mainly differed from test 1b (8 25
overlaps) in a higher RMSE but not in the mean. This difference may arise from the distinct texture and complexity of each
surface, which presumably requires different number of images for a similar performance (James and Robson, 2012;
Westoby et al., 2012; Micheletti et al., 2015).
A RMSE of 3 cm (more precisely 2.8 cm) to measure a riverbank with the photo combination of test 3 results in a
relative precision with respect to the average camera-surface distance of 0.0007 or ~1:1400. This relative precision ratio is 30
somewhat higher than ~1:1000 achieved by James and Robson (2012) for steep irregular features at kilometre-scale in a
volcanic crater and decametre-scale in a coastal cliff, whereas our precision is somewhat lower than ~1:2000 those authors
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
18
proved at metre-scale. More precise results could be possible using a bigger and higher-resolution sensor, flying closer to the
bank, or even trying other oblique bank perspectives. However, this endeavour would only be reasonable if such data are
needed for the research purposes and if GCP positioning had also according higher precisions, since registration errors
translate into the DSM accuracy during camera parameter optimization and/or georeferentiation (Harwin and Lucieer, 2012;
Javernick et al., 2014; Smith et al., 2014). 5
A precision of 10 cm has implications for the representation of small-scale features at bank scarps. Despite the
presence of features in the order of decimetres, we could not assess their accuracy given the discrete GPS points and the
0.5 m ALS grid used to assess the DSM precision. For instance, Fig. 10a shows an upper bank scarp along the last four
surveys that, if assumed unchanged, it would indicate a maximum distance of 20 cm between surveys, which still would
remain within the ±10cm error estimated by the GPS comparison. Although these differences could have been caused by 10
weathering processes or growing grass on the bank face, potential sources of error at such scale could be given, for instance,
by registration errors or occlusions caused by the surface roughness (Lague et al., 2013). Then, further research is needed to
evaluate the precision at the roughness scale to, for example, analyse form drag at the bank face (Leyland et al., 2015).
The comparative analysis of the DSMs from different photo combinations showed that the ground surfaces
surveyed in the case study had different precisions. The grassland presented similar errors with a positive bias throughout all 15
tests. The positive elevation differences are typical of vegetated surfaces (Westoby et al., 2012; Micheletti et al., 2015),
whereas the similar performance of different photo combinations might be due to the presence of sufficient and well
distributed GCPs in this area (the floodplain). The terrace at the toe of the bank, in contrast, presented different error
skewness throughout the tests, which affected the error distribution for all grounds. The error skewness can be related to the
fact that the terrace was the most distant area from the GCPs and it was not surrounded by them, so that errors in lens 20
distortion corrections could have especially increased here (James and Robson, 2014; Javernick et al., 2014; Smith et al.,
2014). This effect was clear at the reach extremes (Fig. 7), where the elevation differences increased with respect to the ALS
survey further from the GCPs, for which it is called ‘dome’ effect.
This DSM distortion beyond the GCP surrounding area might affect the bank too when using the GCPs only on the
floodplain and not on the bank. In the case study, care was taken to sparsely place the GCPs across the floodplain and at 25
different elevations when it was possible, so that the control points were distributed over the three spatial dimensions as
much as possible to increase the georeferencing accuracy (Harwin and Lucieer, 2012). While James and Robson (2014)
showed that using different (convergent) camera angles is effective to mitigate the ‘dome’ effect, our results showed that the
DSM precision with eight photo overlaps along a single UAV track did not substantially improve by adding the extra
perspective of track 2. This may imply that the number of overlaps and used GCPs were sufficient to avoid such distortions 30
in the bank area, together with the fact the track had oblique and not nadiral perspective. Moreover, even though direct
georeferencing is a promising step forward to minimizing the SfM requirements through more accurate camera calibration
(Carbonneau and Dietrich, 2017), GCPs are currently necessary for highest precision by model optimization,
georeferentiation, and quality verification.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
19
5.2 Comparison of two reach-scale techniques: SfM and ALS
The elevation bias at the bank between the SfM-based DSM and the ALS grid (Fig. 8) was caused by the topographic
overestimation of ALS (Table 3 and Fig. 9). This ubiquitous error is ascribed to a known limitation of ALS systems related
to the laser beam divergence angle, which locates the closest feature within the laser footprint at the centre of the footprint.
This increases the ground elevation at high-slope areas (Bailly et al., 2012), which is the case for riverbanks. Still, the ALS 5
resolution and precision were enough to identify bank slopes, in accordance with other studies (e.g., Tarolli et al., 2012;
Ortuño et al., 2017). Furthermore, despite the ALS capability to estimate volume changes of eroded banks (Kessler et al.,
2013), the method omits information related to the phases of the erosion cycle by not surveying erosion features smaller than
its resolution (0.5 m), apparent in contrast with SfM profiles (Fig. 9).
The elevation differences between the methods observed for grassland (Fig. 8, centre left) were probably caused by 10
dissimilar ground resolutions, because a larger elevation scatter is expected in the SfM-based DSM when capturing grass
with 2 cm resolution, whereas ALS had samples spaced 50 cm from a comparable footprint size. Nonetheless, the mean
difference was zero (Table 3), so that both methods overestimate in the same way the real ground elevation due to grass
cover. The effect of this is visible, for instance, in the increasing surface elevations on the floodplain over a year (Fig. 10a).
The terrace at the bank toe presented a similar scatter as grassland, but had a small negative bias that could be explained by a 15
slight transverse ‘dome’ effect of the SfM DSM.
The distance covered by the SfM-UAV method depends on the flight autonomy. The deployed UAV had autonomy
of approximately 25 minutes, which limited the maximum bank survey extent to approximately 2 km for the tested UAV
height and speed, and camera resolution and shutter frequency. This practical limit will change with the progressive
development of UAVs, but the distance covered by a single flight is currently significantly smaller than the one covered by 20
ALS. Although a larger camera-object distance and speed than the used in this work would increase the surveyed area,
decreasing the ground resolution and the UAV stability may result in the loss of sufficient detail to capture erosion features,
and what is more, decrease the DTM precision that depends on the image scale (James and Robson, 2012; Micheletti et al.,
2015). Therefore, further investigations would be required to explore the practical limits of UAV-bank monitoring in views
of extending the survey coverage. 25
5.3 Surveying bank erosion
Sequential surveys allowed to capture different phases of the erosion cycle (Fig. 10a), which demonstrates that quantitative
detection of processes is feasible. Previous studies on bank erosion proved the capabilities of SfM for post-event analysis
(Prosdocimi et al., 2015), e.g. representing block deposition, or for 2.5D bank retreat quantification (Hamshaw et al., 2017),
whereas herein all erosion phases were sequentially captured, demonstrating the 3D potentialities over the complete process 30
of erosion. Of course, the ability to monitor banks at the process scale depends on the time interval with which the method
can re-survey the exposed part of banks and will only cover pre- and post-flood conditions. The survey frequency and the
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
20
duration of a full cycle of erosion determine the temporal resolution with which the development of processes is captured.
Then, the bank retreat rate of each case determines the necessary frequency of surveys to capture erosion processes within a
single cycle. Bank erosion rates naturally depend on each site, which in the presented study site varied enormously (Fig. 3).
Still, the performed eight surveys within a year successfully captured bank processes within a single erosion cycle in areas of
fast retreat such as Section 4. 5
The study site with a regulated water level and recently restored actively eroding banks was a perfect example for
the application of this technique, because banks were exposed and erosion rates were compatible with the proposed average
sampling frequency of six weeks. For other types of rivers, where erosion mainly occurs during floods when banks are not
exposed, this method would allow measuring pre- and post-event conditions only. Given the high resolution achieved, the
method is applicable to all river sizes. However, due to the accuracy obtained, the application is only advised in cases where 10
bank retreat is larger than approximately 30 cm between consecutive surveys.
Erosion processes happening at small spatial scales, such as weathering, would be hardly or not measurable with the
precision achieved in this investigation.. For this, other methods are already available, for instance TLS and boat-based laser
scanning, that provide higher precisions (mm before registration errors, e.g., O’Neal and Pizzuto, 2011) and comparable
resolutions (cm, e.g., Heritage and Hetherington, 2007). In addition, close-range terrestrial photogrammetry can also offer 15
the necessary precision for such endeavours, e.g., from a tripod (Leyland et al., 2015) or a pole on the near-bank area (Bird et
al., 2010), at the expense of covering shorter bank lengths. Another alternative are erosion pins, which may also provide
higher accuracies, yet with point resolution.
UAV-SfM appears a suitable survey method for both process identification and volume quantification in bank
erosion studies, given the decimetre precision range with 3 cm RMSE and the 3D high resolution achieved with a low-cost 20
UAV. As Resop and Hession (2010) suggested, high-resolution three-dimensional capabilities offer great possibilities when
spatial variability of retreat is critical compared to traditional cross-profiling methods. In addition, the reduced deployment
time of UAVs in the field is advantageous in relation to cross-profiling, while it also improves identification of complex
bank features (Figure 9) and volume computations as other 3D high-resolution techniques (O’Neal and Pizzuto, 2011).
Nonetheless, UAV-SfM require longer post-processing times at the office, which should not be underestimated (Westoby et 25
al., 2012; Passalacqua et al., 2015).
This technique remains low-cost compared to TLS or MLS, for which it is more convenient for cases where
roughness is beyond the scale of interest, and target bank lengths are smaller than 3000 m. This would approximately be the
longest distance for a single UAV flight in our case study. For longer reaches, MLS would then compete with UAV-SfM
from a practical perspective, since more than one survey/flight would be needed. However, all TLS, MLS and UAV-SfM 30
would have limitations to survey the bank surface in presence of dense bank vegetation (Hamshaw et al., 2017). In these
cases, ALS provides an alternative, albeit with significant lower resolution and higher costs (Slatton et al., 2007).
For large river extents, i.e., several kilometres, Grove et al. (2013) showed that process inference is possible
combining ALS with high-resolution aerial photography, two techniques that are typically applied for eroded volume
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
21
estimations and bank migration (Khan and Islam, 2003; Lane et al., 2010; De Rose and Basher, 2011; Spiekermann et al.,
2017). In that work, the scale of the river (banks higher than 6 m) allowed a spatial resolution of 1 m to capture features that
together with photo inspection provided information on mass failure type and fluvial entrainment. To date, UAV-SfM covers
smaller extents (Passalacqua et al., 2015), but provides much higher resolutions, allowing for process identification (such as
undermining) and more precise volume computations (see Figure 9 for profile differences between ALS and SfM). For a 5
similar (or higher) accuracy and resolution than those of UAV-SfM and large distances, boat-based laser scanning becomes
an attractive, yet more expensive, solution.
6 Conclusion
This work evaluated the capability of Structure-from-Motion photogrammetry applied with low-cost UAV imagery to
monitor bank erosion processes along a river reach. The technique’s precision was investigated by comparison with GPS 10
points and an airborne laser scanning. Vertical bank profiles were analysed to identify stages of erosion and infer processes.
We used a low-cost UAV with a 12 MP built-in camera flying 25 m from the least retreated bankline and 25 m above the
floodplain level, which produced a photograph set with an oblique perspective and at least eight image overlaps at each bank
point. Together with ground-control points, this photo set was enough to generate through SfM a digital surface model with
sufficient accuracy and resolution to recognize signatures of the different phases of the bank erosion cycle from the obtained 15
bank profiles.
The accuracy of the DSM constructed with the SfM technique did not significantly increase with more than eight
photo overlaps along a single oblique UAV track. The coverage of bank area behind bank toe vegetation, on the other hand,
was increased by adding a vertically oriented perspective, albeit without a significant accuracy increase. As a result, banks
were surveyed with 2 cm resolution and a 10 cm elevation precision, whose mean was 1 cm and standard deviation 3 cm 20
(~1:1400 relative to camera-object distance). This accuracy was confirmed along the river reach after comparison with the
airborne laser scanning. The SfM-based topography agreed well with ALS over horizontal areas, but over bank slopes the
latter overestimated elevations. Higher SfM errors were observed in areas beyond the extent of ground-control points,
showing that control points should also be placed outside the monitoring area.
The SfM resolution captured details of the bank face through which features of different phases of the erosion cycle 25
could be identified. A relative elevation precision with respect to the camera-object distance of ~1:1400 was obtained, in line
with previous SfM topographic applications. The technique’s accuracy, resolution and frequency enabled capturing erosion
processes over sequential surveys. The survey frequency depends on bank retreat rates, which depend on the river size,
hydraulic conditions, bank material, etc.
This investigation demonstrates the capabilities of a low-cost UAV to monitor banks at the process scale, while 30
covering a middle-size river reach of 1.2 km long in a single campaign. The combination of UAV and Structure-from-
Motion photogrammetry can provide relevant information of the spatial structure of bank erosion processes, and with
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
22
sufficient frequency of acquisition, represent the temporal evolution of morphological processes within the erosion cycle.
This method can also be used to compute eroded volumes throughout different phases of the cycle and analyse the
contribution of each mechanism to overall retreats.
The applied technique is most suitable when measuring bank lengths not exceeding the 3000 m. Its flexibility, fast
deployment and high resolution are especially convenient for surveying highly irregular banks. This method can survey the 5
full cycle of erosion, and not only pre- and post-event conditions. The main limitations are dense riparian vegetation and
high water levels, but the same applies to most survey techniques.
Data availability
The data utilized in this work will be publicly available at the 4TU repository after the publication of this manuscript.
Author contribution 10
The field campaigns, data processing, data analysis and discussion of results were carried out by the first author. The writing
of the manuscript was done by the first and second authors. The team authors provided expert opinion, refinement of paper
writing and figures, and ideas that improved the final quality of this work.
Acknowledgements
This study was part of the NCR RiverCare research programme, funded by NWO/STW (project number 13516). We would 15
like to thank Jaap van Duin and Ruben Kunz for their assistance during the field campaigns. We are grateful to Hans Bakker
from Rijkswaterstaat for timely sharing the ALS data.
References
Alho, P., Kukko, A., Hyyppä, H., Kaartinen, H., Hyyppä, J., and Jaakkola, A.: Application of boat‐based laser scanning for
river survey, Earth Surface Processes and Landforms, 34(13), 1831-1838, doi:10.1002/esp.1879, 2009. 20
ASCE Task Committee on Hydraulics, Bank Mechanics, and Modeling of River Width Adjustment. River width adjustment.
I: Processes and mechanisms, J. Hydraul. Eng. 124 (9), 881–902, doi:10.1061/(ASCE)0733-9429(1998)124:9(881), 1998.
Bailly, J. S., Kinzel, P. J., Allouis, T., Feurer, D., and Le Coarer, Y.: Airborne LiDAR methods applied to riverine
environments, Fluvial remote sensing for science and management, 141-161, doi: 10.1002/9781119940791.ch7, 2012.
Bangen, S. G., Wheaton, J. M., Bouwes, N., Bouwes, B., & Jordan, C. (2014). A methodological intercomparison of 25
topographic survey techniques for characterizing wadeable streams and rivers, Geomorphology, 206, 343-361,
doi:10.1016/j.geomorph.2013.10.010, 2014.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
23
Barker, R., Dixon, L., and Hooke, J.: Use of terrestrial photogrammetry for monitoring and measuring bank erosion, Earth
Surface Processes and Landforms, 22(13), 1217-1227, doi:10.1002/(SICI)1096-9837(199724)22:13<1217::AID-
ESP819>3.0.CO;2-U, 1997.
Bird, S., Hogan, D., & Schwab, J.: Photogrammetric monitoring of small streams under a riparian forest canopy, Earth
Surface Processes and Landforms, 35(8), 952-970, doi:10.1002/esp.2001, 2010. 5
Brasington, J., Rumsby, B. T., and McVey, R. A.: Monitoring and modelling morphological change in a braided gravel-bed
river using high resolution GPS-based survey, Earth Surface Processes and Landforms, 25(9), 973-990, doi: 10.1002/1096-
9837(200008)25:9<973::AID-ESP111>3.0.CO;2-Y, 2000.
Brasington, J., Vericat, D., and Rychkov, I.: Modeling river bed morphology, roughness, and surface sedimentology using
high resolution terrestrial laser scanning, Water Resources Research, 48(11), doi:10.1029/2012WR012223, 2012. 10
Carbonneau, P.E. and Dietrich, J.T.: Cost‐effective non‐metric photogrammetry from consumer‐grade sUAS: implications
for direct georeferencing of structure from motion photogrammetry, Earth Surface Processes and Landforms, 42(3), 473-486,
doi:10.1002/esp.4012, 2017.
Cavalli, M., Tarolli, P., Marchi, L., and Dalla Fontana, G.: The effectiveness of airborne LiDAR data in the recognition of
channel-bed morphology, Catena, 73(3), 249-260, doi:10.1016/j.catena.2007.11.001, 2008. 15
Charlton, M. E., Large, A. R., and Fuller, I. C.: Application of airborne LiDAR in river environments: the River Coquet,
Northumberland, UK, Earth surface processes and landforms, 28(3), 299-306, doi:10.1002/esp.482, 2003.
Clapuyt, F., Vanacker, V. and Van Oost, K.: Reproducibility of UAV-based earth topography reconstructions based on
Structure-from-Motion algorithms, Geomorphology, 260, 4-15, doi:10.1016/j.geomorph.2015.05.011, 2016.
Clark, L.A. and Wynn, T.M.: Methods for determining streambank critical shear stress and soil erodibility: Implications for 20
erosion rate predictions, Transactions of the ASABE, 50(1), 95-106, doi:10.13031%2F2013.22415, 2007.
Cook, K.L.: An evaluation of the effectiveness of low-cost UAVs and structure from motion for geomorphic change
detection, Geomorphology, 278, 195-208, doi:10.1016/j.geomorph.2016.11.009, 2017.
Couper, P. R.: Space and time in river bank erosion research: A review, Area, 36(4), 387-403, doi: 10.1111/j.0004-
0894.2004.00239.x, 2004. 25
Couper, P. R., and Maddock, I. P.: Subaerial river bank erosion processes and their interaction with other bank erosion
mechanisms on the River Arrow, Warwickshire, UK, Earth Surface Processes and Landforms, 26(6), 631-646,
doi:10.1002/esp.212, 2001.
Couper, P., Stott, T. I. M., and Maddock, I.: Insights into river bank erosion processes derived from analysis of negative
erosion‐pin recordings: observations from three recent UK studies, Earth Surface Processes and Landforms, 27(1), 59-79, 30
doi: 10.1002/esp.285, 2002.
Darby, S. E., Rinaldi, M., and Dapporto, S.: Coupled simulations of fluvial erosion and mass wasting for cohesive river
banks, Journal of Geophysical Research: Earth Surface (2003–2012), 112(F3), doi: 10.1029/2006JF000722, 2007.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
24
De Rose, R. C., and Basher, L. R.: Measurement of river bank and cliff erosion from sequential LIDAR and historical aerial
photography, Geomorphology, 126(1), 132-147, doi:10.1016/j.geomorph.2010.10.037, 2011.
Eltner, A., and Schneider, D.: Analysis of different methods for 3D reconstruction of natural surfaces from parallel‐axes
UAV images, The Photogrammetric Record, 30(151), 279-299, doi:10.1111/phor.12115, 2015.
Eltner, A., Kaiser, A., Castillo, C., Rock, G., Neugirg, F., and Abellán, A.: Image-based surface reconstruction in 5
geomorphometry–merits, limits and developments, Earth Surface Dynamics, 4(2), 359-389, doi:10.5194/esurf-4-359-2016,
2016.
Florsheim, J.L., Mount, J.F., and Chin, A.: Bank Erosion as a Desirable Attribute of Rivers, BioScience, 58(6), 519-529,
doi:10.1641/B580608, 2008.
Fonstad, M. A., Dietrich, J. T., Courville, B. C., Jensen, J. L., and Carbonneau, P. E.: Topographic structure from motion: a 10
new development in photogrammetric measurement, Earth Surface Processes and Landforms, 38(4), 421-430,
doi:10.1002/esp.3366, 2013.
Fukuoka, S.: Erosion processes of natural river bank, Proceedings of the 1st International Symposium on Hydraulic
Measurement, Beijing, Vol. 223229, 1994.
Girardeau-Montaut, D.: CloudCompare (version 2.9; GPL software), EDF RandD, Telecom ParisTech, available at: 15
http://www.cloudcompare.org/, last access: January 2018.
Gómez-Gutiérrez, Á., Schnabel, S., Berenguer-Sempere, F., Lavado-Contador, F. and Rubio-Delgado, J.: Using 3D photo-
reconstruction methods to estimate gully headcut erosion, Catena, 120, 91-101, doi:10.1016/j.catena.2014.04.004, 2014.
Grove, J. R., Croke J., Thompson, C.: Quantifying different riverbank erosion processes during an extreme flood
event, Earth Surface Processes and Landforms, 38(12), 1393-1406, doi:10.1002/esp.3386, 2013. 20
Hamshaw, S.D., Bryce, T., Rizzo, D.M., O'Neil‐Dunne, J., Frolik, J. and Dewoolkar, M.M.: Quantifying streambank
movement and topography using unmanned aircraft system photogrammetry with comparison to terrestrial laser scanning,
River Research and Applications, 33(8), 1354-1367, doi:10.1002/rra.3183, 2017.
Hartley, R., and Zisserman, A.: Multiple view geometry in computer vision, Cambridge university press, 2003.
Harwin, S., and Lucieer, A.: Assessing the accuracy of georeferenced point clouds produced via multi-view stereopsis from 25
unmanned aerial vehicle (UAV) imagery, Remote Sensing, 4(6), 1573-1599, doi:10.3390/rs4061573, 2012.
Henshaw, A. J., Thorne, C. R., and Clifford, N. J.: Identifying causes and controls of river bank erosion in a British upland
catchment, Catena, 100, 107-119, doi:10.1016/j.catena.2012.07.015, 2013.
Heritage, G., and Hetherington, D.: Towards a protocol for laser scanning in fluvial geomorphology, Earth Surface Processes
and Landforms, 32(1), 66-74, doi:10.1002/esp.1375, 2007. 30
Hooke, J.M.: An analysis of the processes of river bank erosion, Journal of Hydrology, 42(1-2):39-62, doi:10.1016/0022-
1694(79)90005-2, 1979.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
25
Immerzeel, W. W., Kraaijenbrink, A., Shea, J. M., Shrestha, A. B., Pellicciotti, F., Bierkens, M. F. P., and De Jong, S. M.:
Highresolution monitoring of Himalayan glacier dynamics using unmanned aerial vehicles, Remote Sens. Environ., 150, 93–
103, doi:10.1016/j.rse.2014.04.025, 2014.
Jacobson A. et al.: Geometry Processing Toolbox, available at: http://github.com/alecjacobson/gptoolbox, last access:
January 2018. 5
James, M. R., and Robson, S.: Straightforward reconstruction of 3D surfaces and topography with a camera: Accuracy and
geoscience application, Journal of Geophysical Research: Earth Surface, 117(F3), doi:10.1029/2011JF002289, 2012.
James, M. R., and Robson, S.: Mitigating systematic error in topographic models derived from UAV and ground‐based
image networks. Earth Surface Processes and Landforms, 39(10), 1413-1420, doi:10.1002/esp.3609, 2014.
James, M.R., Robson, S., d'Oleire-Oltmanns, S. and Niethammer, U.: Optimising UAV topographic surveys processed with 10
structure-from-motion: Ground control quality, quantity and bundle adjustment, Geomorphology, 280, 51-66,
doi:10.1016/j.geomorph.2016.11.021, 2017.
Javernick, L., Brasington, J., and Caruso, B.: Modeling the topography of shallow braided rivers using Structure-from-
Motion photogrammetry, Geomorphology, 213, 166–182, doi:10.1016/j.geomorph.2014.01.006, 2014.
Khan, N. I., and Islam, A.: Quantification of erosion patterns in the Brahmaputra–Jamuna River using geographical 15
information system and remote sensing techniques, Hydrological Processes, 17(5), 959-966, doi: 10.1002/hyp.1173, 2003.
Kessler, A. C., Gupta, S. C., and Brown, M. K.: Assessment of river bank erosion in Southern Minnesota rivers post
European settlement, Geomorphology, 201, 312-322, doi:10.1016/j.geomorph.2013.07.006, 2013.
Kimiaghalam, N., Goharrokhi, M., Clark, S. P., Ahmari, H.: A comprehensive fluvial geomorphology study of riverbank
erosion on the Red River in Winnipeg, Manitoba, Canada, Journal of Hydrology, 529, 1488-1498, 20
doi:10.1016/j.jhydrol.2015.08.033, 2015.
Lague, D., Brodu, N., and Leroux, J.: Accurate 3D comparison of complex topography with terrestrial laser scanner:
Application to the Rangitikei canyon (NZ), ISPRS Journal of Photogrammetry and Remote Sensing, 82, 10-26,
doi:10.1016/j.isprsjprs.2013.04.009, 2013.
Lane, S. N.: The measurement of river channel morphology using digital photogrammetry, The Photogrammetric Record, 25
16(96), 937-961, doi:10.1111/0031-868X.00159, 2000.
Lane, S.N., Widdison, P.E., Thomas, R.E., Ashworth, P.J., Best, J.L., Lunt, I.A., Sambrook Smith, G.H. and Simpson, C.J.:
Quantification of braided river channel change using archival digital image analysis, Earth Surface Processes and
Landforms, 35(8), 971-985, doi:10.1002/esp.2015, 2010.
Lawler, D. M.: Process dominance in bank erosion systems, Lowland floodplain rivers: Geomorphological perspectives, 30
117-43, 1992.
Lawler, D.M.: The measurement of river bank erosion and lateral channel change: a review, Earth surface processes and
landforms, 18(9), 777-821, doi:10.1002/esp.3290180905, 1993.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
26
Lawler, D.M.: The importance of high-resolution monitoring in erosion and deposition dynamics studies: examples from
estuarine and fluvial systems, Geomorphology, 64(1), 1-23, doi:10.1016/j.geomorph.2004.04.005, 2005.
Leyland, J., Darby, S. E., Teruggi, L., Rinaldi, M., and Ostuni, D.: A self-limiting bank erosion mechanism? inferring
temporal variations in bank form and skin drag from high resolution topographic data, Earth Surf. Process. Landforms, 40:
1600–1615, doi:10.1002/esp.3739, 2015. 5
Massey, D.: Talking of space-time, Transactions of the Institute of British Geographers, 26: 257–261, doi:10.1111/1475-
5661.00019, 2001.
Micheletti, N., Chandler, J. H., and Lane, S. N.: Investigating the geomorphological potential of freely available and
accessible structure‐from‐motion photogrammetry using a smartphone, Earth Surface Processes and Landforms, 40(4), 473-
486, doi:10.1002/esp.3648, 2015. 10
Nardi, L., Campo, L., and Rinaldi, M.: Quantification of riverbank erosion and application in risk analysis, Nat Hazards,
69(1), 869-887, doi:10.1007/s11069-013-0741-8, 2013.
O'neal, M. A., and Pizzuto, J. E.: The rates and spatial patterns of annual riverbank erosion revealed through terrestrial laser‐
scanner surveys of the South River, Virginia, Earth Surface Processes and Landforms, 36(5), 695-701, doi:10.1002/esp.2098,
2011. 15
Ortuño, M., Guinau, M., Calvet, J., Furdada, G., Bordonau, J., Ruiz, A., and Camafort, M.: Potential of airborne LiDAR data
analysis to detect subtle landforms of slope failure: Portainé, Central Pyrenees, Geomorphology, 295, 364-382,
doi:10.1016/j.geomorph.2017.07.015, 2017.
Ouédraogo, M. M., Degré, A., Debouche, C., and Lisein,J.: The evaluation of unmanned aerial system-based
photogrammetry and terrestrial laser scanning to generate DEMs of agricultural watersheds, Geomorphology, 214, 339–355, 20
doi:10.1016/j.geomorph.2014.02.016, 2014.
Parker, G., Shimizu, Y., Wilkerson, G. V., Eke, E. C., Abad, J. D., Lauer, J. W., Paola, C., Dietrich, W. E., Voller, V. R.: A
new framework for modeling the migration of meandering rivers, Earth Surface Processes and Landforms, 36(1), 70-86,
doi:10.1002/esp.2113, 2011.
Passalacqua, P., Belmont, P., Staley, D.M., Simley, J.D., Arrowsmith, J.R., Bode, C.A., Crosby, C., DeLong, S.B., Glenn, 25
N.F., Kelly, S.A. and Lague, D.: Analyzing high resolution topography for advancing the understanding of mass and energy
transfer through landscapes: A review, Earth-Science Reviews, 148, 174-193, doi:10.1016/j.earscirev.2015.05.012, 2015.
Pyle, C. J., Richards, K. S., and Chandler, J. H.: Digital photogrammetric monitoring of river bank erosion, The
Photogrammetric Record, 15(89), 753-764, doi:10.1111/0031-868X.00083, 1997.
Prosdocimi, M., Calligaro, S., Sofia, G., Dalla Fontana, G., and Tarolli, P.: Bank erosion in agricultural drainage networks: 30
new challenges from structure‐from‐motion photogrammetry for post‐event analysis, Earth Surface Processes and
Landforms, 40(14), 1891-1906, doi:10.1002/esp.3767, 2015.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
27
Reneau, S.L., Drakos, P.G., Katzman, D., Malmon, D.V., McDonald, E.V., Ryti, R.T.: Geomorphic controls on contaminant
distribution along an ephemeral stream. Earth Surface Processes and Landforms, 29(10):1209-23, DOI: 10.1002/esp.108,
2004.
Resop, J. P., and Hession, W. C.: Terrestrial laser scanning for monitoring streambank retreat: Comparison with traditional
surveying techniques, Journal of Hydraulic Engineering, 136(10), 794-798, doi:10.1061/(ASCE)HY.1943-7900.0000233, 5
2010.
Rhoades, E. L., O'Neal, M. A., and Pizzuto, J. E.: Quantifying bank erosion on the South River from 1937 to 2005, and its
importance in assessing Hg contamination, Applied Geography, 29(1), 125-134, doi:10.1016/j.apgeog.2008.08.005, 2009.
Rinaldi, M., and Darby, S. E.: Modelling river-bank-erosion processes and mass failure mechanisms: progress towards fully
coupled simulations, Developments in Earth Surface Processes, 11, 213-239. doi:10.1016/S0928-2025(07)11126-3, 2008. 10
Rinaldi, M., Casagli N., Dapporto S., and Gargini, A.: Monitoring and modeling of pore water pressure changes and
riverbank stability during flow events, Earth Surf. Processes Landforms, 29, 237–254, doi: 10.1002/esp.1042, 2004.
Siviglia, A., & Crosato, A.: Numerical modelling of river morphodynamics: Latest developments and remaining challenges.
Advances in Water Resources, 93(A), 1-3, doi:10.1016/j.advwatres.2016.01.005, 2016.
Slatton, K. C., Carter W. E., Shrestha R. L., and Dietrich W.: Airborne Laser Swath Mapping: Achieving the resolution and 15
accuracy required for geosurficial research, Geophys. Res. Lett., 34, L23S10, doi:10.1029/2007GL031939, 2007.
Smith, M. W., Carrivick, J. L., Hooke, J., and Kirkby, M. J.: Reconstructing flash flood magnitudes using “Structure-from-
Motion”: A rapid assessment tool, J. Hydrol., 519, 1914–1927, doi:10.1016/j.jhydrol.2014.09.078, 2014.
Snavely, N., Seitz, S. M., and Szeliski, R.: Modeling the world from internet photo collections, International Journal of
Computer Vision, 80(2), 189-210, doi:10.1007/s11263-007-0107-3, 2008. 20
Spiekermann, R., Betts, H., Dymond, J., and Basher, L.: Volumetric measurement of river bank erosion from sequential
historical aerial photography, Geomorphology, 296, 193-208, doi:10.1016/j.geomorph.2017.08.047, 2017.
Tamminga, A. D., Brett C. E., and Hugenholtz C. H.: UAS‐based remote sensing of fluvial change following an extreme
flood event, Earth Surface Processes and Landforms 40, no. 11, 1464-1476, doi: 10.1002/esp.3728 , 2015.
Tarolli, P., Sofi,a G., Dalla Fontana, G.: Geomorphic features extraction from high-resolution topography: landslide crowns 25
and bank erosion, Natural Hazards, 61(1), 65-83, doi:10.1007/s11069-010-9695-2, 2012.
Telling, J., Lyda, A., Hartzell, P., and Glennie, C.: Review of earth science research using terrestrial laser scanning, Earth-
Science Reviews, 169, 35-68, 10.1016/j.earscirev.2017.04.007, 2017.
Thoma, D. P., Gupta, S. C., Bauer, M. E., and Kirchoff, C. E.: Airborne laser scanning for riverbank erosion assessment,
Remote sensing of Environment, 95(4), 493-501, doi:10.1016/j.rse.2005.01.012, 2005. 30
Thorne, C. R.: Processes and mechanisms of river bank erosion, Gravel-bed rivers, 227-259, 1982.
Thorne, C.R. and Tovey N.K.: Stability of composite river banks, Earth Surface Processes and Landforms, 6(5), 469-484,
doi:10.1002/esp.3290060507, 1981.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
28
Tonkin, T. N., Midgley, N. G., Graham, D. J., and Labadz, J. C.: The potential of small unmanned aircraft systems and
structure-frommotion for topographic surveys: A test of emerging integrated approaches at Cwm Idwal, North Wales,
Geomorphology, 226, 35–43, doi:10.1016/j.geomorph.2014.07.021, 2014.
Turner, D., Lucieer, A., and de Jong, S.: Time Series Analysis of Landslide Dynamics Using an Unmanned Aerial Vehicle
(UAV), Remote Sensing, 7, 1736–1757, doi:10.3390/rs70201736, 2015. 5
Westoby, M. J., Brasington, J., Glasser, N. F., Hambrey, M. J., and Reynolds, J. M.: ‘Structure-from-Motion’
photogrammetry: A low-cost, effective tool for geoscience applications, Geomorphology, 179, 300-314,
doi:10.1016/j.geomorph.2012.08.021, 2012.
Woodget, A. S., Carbonneau, P. E., Visser, F., and Maddock, I. P.: Quantifying submerged fluvial topography using
hyperspatial resolution UAS imagery and structure from motion photogrammetry, Earth Surf. Proc. Land., 40, 47–64, 10
doi:10.1002/esp.3613, 2015.
Earth Surf. Dynam. Discuss., https://doi.org/10.5194/esurf-2018-3
Manuscript under review for journal Earth Surf. Dynam.
Discussion started: 16 March 2018
c
Author(s) 2018. CC BY 4.0 License.
