Remote Sensing Letters

ISSN: 2150-704X (Print) 2150-7058 (Online) Journal homepage: https://www.tandfonline.com/loi/trsl20

The reproducibility of SfM algorithms to produce

detailed Digital Surface Models: the example of
PhotoScan applied to a high-alpine rock glacier

Hanne Hendrickx, Sebastián Vivero, Laure De Cock, Bart De Wit, Philippe De

Maeyer, Christophe Lambiel, Reynald Delaloye, Jan Nyssen & Amaury Frankl

To cite this article: Hanne Hendrickx, Sebastián Vivero, Laure De Cock, Bart De Wit, Philippe
De Maeyer, Christophe Lambiel, Reynald Delaloye, Jan Nyssen & Amaury Frankl (2019) The
reproducibility of SfM algorithms to produce detailed Digital Surface Models: the example of
PhotoScan applied to a high-alpine rock glacier, Remote Sensing Letters, 10:1, 11-20, DOI:
10.1080/2150704X.2018.1519641

To link to this article: https://doi.org/10.1080/2150704X.2018.1519641

Published online: 24 Oct 2018.

Submit your article to this journal

Article views: 267

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=trsl20
REMOTE SENSING LETTERS
2019, VOL. 10, NO. 1, 11–20
https://doi.org/10.1080/2150704X.2018.1519641

The reproducibility of SfM algorithms to produce detailed

Digital Surface Models: the example of PhotoScan applied to
a high-alpine rock glacier
Hanne Hendrickxa, Sebastián Vivero b, Laure De Cocka, Bart De Wita,
Philippe De Maeyer a, Christophe Lambielb, Reynald Delaloyec, Jan Nyssena
and Amaury Frankla,d
a
Department of Geography, Ghent University, Ghent, Belgium; bInstitute of Earth Surface Dynamics,
University of Lausanne, Lausanne, Switzerland; cDepartment of Geosciences, University of Fribourg,
Fribourg, Switzerland; dPostdoctoral Fellow of the Research Foundation Flanders (FWO), Brussels, Belgium

ABSTRACT ARTICLE HISTORY

In geomorphology, PhotoScan is a software that is used to produce Received 1 March 2018
Digital Surface Models (DSMs). It constructs 3D environments from Accepted 31 August 2018
2D imagery (often taken by Unmanned Aerial Vehicles (UAV)) based
on Structure-from-Motion (SfM) and Multi-View Stereo (MVS) princi-
ples. However, unpublished computer-vision algorithms used, con-
tain random elements which can affect the accuracy of the outputs.
For this letter, ten model runs with identical inputs were performed
on UAV imagery of a rock glacier to analyse the magnitude of the
variation between the different model outputs. This variation was
quantified calculating the standard deviation of each cell value in the
respective DSMs and derivatives (curvature). Places with steep slope
gradients have considerably more DSM variation (up to 10 cm) but
stay within the range of the model’s accuracy (10 vertical cm) for 88 –
96% of the area. The edges of the model also show a larger variability
(0.10 – 3 m), related to a lower number of overlapping images. These
results should be accounted for when performing a geomorphologi-
cal research at centimetre scale using PhotoScan, especially in areas
with a complex relief. Using medium-quality runs, additional oblique
viewpoints and respecting a minimum of five overlapping images
can minimize the software’s variations.

1. Introduction
Structure-from-Motion (SfM) is a recent survey technique that merges novel advances in
computer vision with digital photogrammetry procedures. This technique is widely used to
construct 3D environments from 2D consumer grade images using the Multi-View Stereo
(MVS) principle (Fonstad et al. 2013; Smith, Carrivick, and Quincey 2015). It has proven to be a
successful technique in order to study forms and processes in geosciences (e.g., Niethammer
et al. 2012; Javernick, Brasington, and Caruso 2014; Frankl et al. 2015; Piermattei et al. 2016;
Dall’Asta et al. 2017) and yields accuracies similar to those of terrestrial laser scanning (Fonstad
et al. 2013; Smith, Carrivick, and Quincey 2015). At landform and hillslope scales, Digital Surface

CONTACT Hanne Hendrickx [email protected] Department of Geography, Ghent University, Ghent, Belgium
© 2018 Informa UK Limited, trading as Taylor & Francis Group
12 H. HENDRICKX ET AL.

Models (DSM) allow to investigate landforms at sub-centimetre accuracies (Harwin and Lucieer
2012; James, Robson, and Smith 2017), while processes (e.g., erosion and deposition) can be
quantified at centimetre to sub-decimetre accuracies (e.g., Niethammer et al. 2012; Lannoeye
et al. 2016; Piermattei et al. 2016; Dall’Asta et al. 2017). These accuracies highly depend on
image acquisition distance (flight height), accuracy of the Ground Control Points (GCPs),
camera properties (MegaPixels (MP), aperture, shutter speed, ISO and lens properties) and
the texture contrast of the surface. Both ground-based and aerial images can be used. Since
the recent innovation of Unmanned Aerial Vehicles (UAV), aerial imagery is becoming more
frequently used, providing a low-cost and effective way to conduct high-resolution topo-
graphic surveys (Westoby et al. 2012).
PhotoScan Professional is a commercial software developed by Agisoft (Agisoft 2018) and
is widely used in applying the SfM-MVS technique (e.g., Frankl et al. 2015; Uysal, Toprak, and
Polat 2015; Lannoeye et al. 2016; Kenner et al. 2017). This software package identifies
correspondences between features (key points) across overlapping images (Semyonov
2011), based on the Scale Invariant Feature Transform algorithm (SIFT, Lowe 2004).
Estimated camera model parameters are used for self-calibration and to optimize camera
position that then are refined after entering the GCPs and their coordinates, using a bundle-
adjustment algorithm. A dense point cloud, followed by a dense surface is then constructed.
Finally, texture mapping is conducted (Semyonov 2011). However, extensive technical
details about the algorithms used in Photoscan are unpublished.
To the best of our knowledge, the reproducibility of PhotoScan is not yet explicitly
addressed by earlier technical papers. However, the software processing algorithms directly
influence model accuracy (Uysal, Toprak, and Polat 2015; James et al. 2017a). Therefore, we
explored the reproducibility of a 3D model of a complex alpine environment from multiple
PhotoScan runs. This is done with exactly the same images, processing settings and GCPs as
these input parameters have been identified influencing accuracies by previous research
(Harwin and Lucieer 2012; James and Robson 2014; Uysal, Toprak, and Polat 2015; James,
Robson, and Smith 2017). Results are discussed here and recommendations to future users are
made.

2. Study area, survey method and data processing

2.1. Study area
The data was collected in October 2017 on the lower part of the Cliosses rock glacier,
primarily composed of different sizes of angular metamorphic rocks. The study site is
located on the right side of the Herens valley, Western Swiss Alps, at an elevation of
2400 – 2500 m a.s.l. (Figure 1). Surface velocities have been measured since 2006 and are
lower than 0.5 m per year (Delaloye, Lambiel, and Gärtner-Roer 2010).

2.2. UAV and camera specifications

The UAV survey was performed in 2017 using a 16MP Panasonic Lumix DMC-GM5
equipped with a 12–32 mm F3.5–5.6 lens. The focal length was fixed to 24 mm, a
shutter speed of 1/500 and ISO set on 400. The UAV was a custom-made Hexacopter DJI
F550 with a Pixhawk flight controller. In the first step of the flight planning, flight lines
 REMOTE SENSING LETTERS 13

Figure 1. Situation of the study area (a): Les Cliosses rock glacier (c), including slope gradient (b) and
a field photograph (d). Overall height of the rock glacier front is 25 m.

were drawn in a GIS environment (QGIS 2.18.2), parallel to the contour lines of the area.
A Python script was used to translate these flight lines into a waypoints-file. For every
line, the script adjusted the flight height of the UAV according to the topography. This
resulted in a constant flight height of approximately 90 m for every image. The speed of
the UAV was 4–6 m s−1 and the acquisition interval was set to 1 s to ensure abundant
image overlap. This resulted in 561 images. To make this experiment quickly reprodu-
cible and avoiding long processing time, 24 images were used from five sequential flight
lines, with a longitudinal overlap of 70% and a side overlap of 65–75%. This resulted in
an acquisition interval of 6 s (Figure 2). The ground pixel size and thus the Ground
Sample Density (GSD) was 1.41 cm/pixel and an area of 64 100 m2 was covered.

2.3. dGNSS measurements

Differential GNSS (Global Navigation Satellite System) measurements using Post-Processing
Kinematics (PPK) were performed to reference the model to a real-world system. Within the
small sample from the UAV survey used for this experiment, 8 ground control points (GCPs)
and 13 check points (CPs, subset of existing dGNSS points measured for monitoring the rock
glacier velocities) were used (Figure 2). The GCPs consisted of physical square targets
(0.5 × 0.5 m) with a high contrast and a clear centre (Figure 3(a)). The CPs are painted red
14 H. HENDRICKX ET AL.

Figure 2. Distribution of the Ground Control Points (GCPs), the check points (CPs) and images along
the flight lines, with the orthophoto mosaic as background.

circles with a drilled hole in the centre (Figure 3(b)). Differential post processing of the GNSS
data was conducted using Trimble Business Center (TBC) v4 surveying software, linking the
base station in the field with the permanent base station in Zermatt, 25 km away from the rock
glacier. x and y values were referred to the Revised Swiss Reference System (CH1903+LV95)
and elevation (z) was recorded with respect to the Swiss Geoid Model Version 2004
(ChGeo2004). Mean horizontal and vertical precision were around 0.014 and 0.02 m for the
GCPs and the CPs.

2.4. Preparation and processing of the data

Making the PhotoScan workflow (Figure 3) reproducible on ten model runs was done by
indicating GCPs and CPs on images prior to importing them into PhotoScan. The same pixel
was coloured with GNU Image Manipulation Program (GIMP). This way, GCPs and CPs could be
indicated manually at the same location (1 pixel positioning accuracy) for each software run.
First, the 24 images were imported in PhotoScan (Agisoft 2018), then the workflow in Figure 3
was followed. This workflow was repeated ten times, both in medium and high quality.
DSM accuracy was assessed through the RMSE of the CPs, that were introduced to verify
whether the error in z was in the same order of magnitude than that of the GCPs. This allows to
detect systematic DSM deformations (James and Robson 2014). The variation between the
 REMOTE SENSING LETTERS 15

Figure 3. Workflow in PhotoScan (method based on Agisoft 2018). To enhance reproducibility, GCPs
(a) and check points (b) were indicated on the images prior to the PhotoScan runs.

different model runs was assessed by calculating the standard deviation of both the z-value of
each cell of the DSM and the calculated curvature (Zevenbergen and Thorne 1987), as a
measure of the surface morphology, in ArcMap 10.3 (ESRI), using Cell Statistics (Spatial Analyst).

3. Results
The ten runs resulted in a 1.4 cm (=GSD) resolution DSM and a 3.14 cm orthophoto mosaic.
The overall Root Mean Square Errors (RMSE) of x, y and z for the GCPs are of the same order
of magnitude as for the CPs (Table 1). This suggests that the single model runs perform well.
Moreover, the RMSE in Z for the CPs shows a vertical model accuracy of about 10 cm without
DSM deformations. This single model accuracy is similar to accuracies reached in similar
environments and surface roughness (Dall’Asta et al. 2017; Kenner et al. 2017).
The overall variation of the RMSE for the ten model runs, expressed as standard
deviation (Table 1, Stdev), is considered small. The difference between the high-quality
run and the medium-quality run is also small, especially in z, which is of most interest for
volume calculations. This suggest that modelled coordinates differ little. Indeed, when
the x, y, and z coordinates of the different model runs are analysed, a standard deviation
16 H. HENDRICKX ET AL.

Table 1. Root Mean Square Errors (RMSE) of x, y and z to evaluate the overall model performance. The
average and the standard deviation of the ten model runs are presented for both medium and high quality
runs.
Medium High
Average (cm) Stdev (cm) Average (cm) Stdev (cm)
GCPs RMSE (x) 9.32 0.06 6.78 0.07
RMSE (y) 2.50 0.01 4.52 0.18
RMSE (z) 8.85 0.03 8.56 0.26
CPs RMSE (x) 11.84 0.07 10.95 0.08
RMSE (y) 3.88 0.03 5.05 0.07
RMSE (z) 9.96 0.20 9.27 0.19

below 5 × 10−10 m for both high- and medium-quality runs was found. Variation
between the coordinates of the different model runs is thus negligible.
To quantify the variation within the output products (DSMs z-value and curvature) of the
ten model runs, the standard deviation for each cell value was calculated (Figure 4). Both the
medium- and high-quality runs show more variation at the edges of the DSMs (10 cm – 3 m)
and curvature (3000 – 4000 within a range of [−80 000; +80 000]. This variation is due to less
GCPs and less overlapping images (Figure 5), resulting in a lower point density and thus a
higher model variation. The medium-quality run shows a clear pattern in variation with
slope and surface roughness (Figure 4(a), (c); Figure 6), while the high-quality run has a more
linear pattern of variation, which seems related to the flight lines and the overlap areas of
the images, for both DSM and curvature (Figure 4(b), (d); Figure 6). Comparing the point
clouds with the M3C2 algorithm in CloudCompare (Lague, Brodu, and Leroux 2013) gives
the same results, excluding the gridding process as cause of these model variations.

4. Discussion and conclusion

The variation between different PhotoScan model runs is relatively small, as shown by the
very small difference between the calculated coordinates and small RMSE (Table 1).
Nevertheless, different PhotoScan model runs can give different outcome products (in this
case point clouds, DSMs and calculated curvatures), although same input data and proces-
sing details are used. The variation between DSMs is small and stays under the model
accuracy for 88 to 96% of the study area (high- and medium-quality run respectively).
However, in some cases the DSM variation is larger than the accuracy of the model itself
(≥10 cm in z), whereas 3 to 6% of these variations lie at the edge of the model (image overlap
<5) and 1 to 6% lie within this boundary (high- and medium-quality run respectively).
DSM variation is clearly linked to slope gradient and surface roughness for the medium-
quality run (Figure 4(c), Figure 6(c)). This is to be expected, since inaccuracies in x and y
automatically resolve in inaccuracy in z, especially on steep slopes and areas with a high
surface roughness. Curvature variation, as a measure on how well the model can reproduce
the surface morphology, shows an even more pronounced relationship with slope gradient
(Figure 4(a), Figure 6(a)). Standard deviations and thus variation in the model result are largest
in areas with high slope gradient and surface roughness. This is especially relevant for
geomorphic studies since process magnitude is higher on steep slopes, and model inaccura-
cies on these location may mislead interpretations. It is, therefore, recommended to add
oblique viewpoint flights to get a higher point density and a better reproducibility by
 REMOTE SENSING LETTERS 17

Figure 4. Standard deviations of the curvature value of each cell for the ten model runs in medium- (a) and
high-quality run (b) and of the DSM value in z of each cell for the ten model runs in medium- (c) and high-
quality run (d).

PhotoScan in such areas. Moreover, oblique viewpoint shots have been proven to avoid
systematic DSM deformations caused by suboptimal camera calibration parameters (James
and Robson 2014) and loss of accuracy due to sparse GCPs (Harwin, Lucieer, and Osborn
2015). A rapid increase in error with slope classes was also observed by Piermattei et al. (2016).
This was due to areas with poor texture, like snow cover or shadow zones, which also result in
a lower point density and can thus cause more model variation. However, rock glacier surface
texture is primarily composed of different sizes of angular rocks (Ikeda and Matsuoka 2006)
that can have low to medium albedo, well defined structures and medium to high contrast.
Step-like differences in DSMs were also found by James et al. (2017b), corresponding to
changes in image overlaps. In order to find corresponding features, the software needs the
feature to be visible on at least two images. Since image texture can differ greatly from case to
case, it is difficult to advice a minimum number of overlapping images required for scene
construction. Westoby et al. (2012) stated that a minimum of three overlapping images should
be obtained. However, Figure 5 shows that variation in the modelled DSMs lowers significantly
18 H. HENDRICKX ET AL.

Figure 5. Variation of the model (standard deviation) as a function of the number of overlapping images.

Figure 6. The exponential relationship between slope gradient and the variation in curvature (a, b)
and the variation in DSM (z) (c, d) for both medium- and high-quality runs, where the number of
overlapping pictures is at least five. With R2 as the coefficient of determination.

starting from five overlapping images. Flight planning is, therefore, very important and is
usually done in multi-image blocks for an efficient aerial coverage (James and Robson 2014).
Flight lines should exceed the area of interest to get optimal results.
A good distribution of GCPs is also advisable and part of the reason of a higher variability
and a lower accuracy of the model at the edges. It will also minimize potential DSM deforma-
tions. Using CPs to validate the absolute model accuracy is also highly recommended (James
and Robson 2014).
The overall model performance from the high- and medium-quality run lie very close to each
other (Table 1, Figure 5). Especially the RMSE for z is very similar. Considering the difference in
computational effort calculation time (15 min vs. an hour), it is more efficient to use the
medium-quality run in this case. Model variability is also more predictable for the medium-
 REMOTE SENSING LETTERS 19

quality run, since it is more clearly linked to slope steepness. Moreover, both variations in DSM
(z) and curvature show a better performance using the medium-quality run (Figure 6).
From the above we can conclude that PhotoScan is reliable software to perform high
resolution processing of SfM-MVS data and that it can reproduce the same outcome with a
minimum of variation, even in a complex topography. However, the model varies more at the
model edges and in areas with pronounced/complex relief. Especially the latter may bias
geomorphic interpretations. These variations can be minimized by using medium-quality runs,
additional oblique viewpoints and realizing a minimum of five overlapping images. Therefore,
good planning is crucial, especially when surveying large and complex terrain.

Acknowledgments
Adeline Frossard (Msc student UNIL) is thanked for her help during the field work. Field logistics were
organized by IDYST-UNIL. The authors thank the anonymous reviewer for valuable comments.

External links
3D Model of this experiment: https://skfb.ly/6uM6w
Entire 3D model of the site: https://skfb.ly/6uNzZ

Disclosure statement
No potential conflict of interest was reported by the authors

Funding
H. Hendrickx’s field stay was funded by Research Foundations – Flanders under a travel grant
(V4.321.17N) for long research stay abroad.

ORCID
Sebastián Vivero http://orcid.org/0000-0002-1813-9575
Philippe De Maeyer http://orcid.org/0000-0001-8902-3855

References
Agisoft. 2018. “Agisoft PhotoScan User Manual.” Professional Edition, Version 1.4. http://www.
agisoft.com/downloads/user-manuals/
Dall’Asta, E., G. Forlani, R. Roncella, M. Santise, F. Diotri, and U. Morra Di Cella. 2017. “Unmanned
Aerial Systems and DSM Matching for Rock Glacier Monitoring.” ISPRS Journal of
Photogrammetry and Remote Sensing 127: 102–114. doi:10.1016/j.isprsjprs.2016.10.003.
Delaloye, R., C. Lambiel, and G.-R. Isabelle. 2010. “Overview of Rock Glacier Kinematics Research in
the Swiss Alps: Seasonal Rhythm, Interannual Variations and Trends over Several Decades.”
Geographica Helvetica 65 (2): 135–145. doi:10.5194/gh-65-135-2010.
Fonstad, M. A., J. T. Dietrich, B. C. Courville, J. L. Jensen, and P. E. Carbonneau. 2013. “Topographic
Structure from Motion: A New Development in Photogrammetric Measurement.” Earth Surface
Processes and Landforms. doi:10.1002/esp.3366.
Frankl, A., C. Stal, A. Abraha, J. Nyssen, D. Rieke-Zapp, A. De Wulf, and J. Poesen. 2015. “Detailed Recording
of Gully Morphology in 3D through Image-Based Modelling.” CATENA 127 (April): 92–101. doi:10.1016/
j.catena.2014.12.016.
20 H. HENDRICKX ET AL.

Harwin, S., and A. Lucieer. 2012. “Assessing the Accuracy of Georeferenced Point Clouds Produced via
Multi-View Stereopsis from Unmanned Aerial Vehicle (UAV) Imagery.” Remote Sensing 4 (6): 1573–
1599. doi:10.3390/rs4061573.
Harwin, S., A. Lucieer, and J. Osborn. 2015. “The Impact of the Calibration Method on the Accuracy
of Point Clouds Derived Using Unmanned Aerial Vehicle Multi-View Stereopsis.” Remote Sensing
7 (9): 11933–11953. doi:10.3390/rs70911933.
Ikeda, A., and N. Matsuoka. 2006. “Pebbly versus Bouldery Rock Glaciers: Morphology, Structure
and Processes.” Geomorphology 73 (3–4): 279–296. doi:10.1016/j.geomorph.2005.07.015.
James, M. R., and S. Robson. 2014. “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.
James, M. R., S. Stuart Robson, S. D’Oleire-Oltmanns, and U. Niethammer. 2017a. “Optimising UAV
Topographic Surveys Processed with Structure-from-Motion: Ground Control Quality, Quantity
and Bundle Adjustment.” Geomorphology 280: 51–66. doi:10.1016/j.geomorph.2016.11.021.
James, M. R., S. Robson, and M. W. Smith. 2017b. “3-D Uncertainty-Based Topographic Change Detection
with Structure-from-Motion Photogrammetry: Precision Maps for Ground Control and Directly
Georeferenced Surveys.” Earth Surface Processes and Landforms 42 (12): 1769–1788. doi:10.1002/
esp.4125.
Javernick, L., J. Brasington, and B. Caruso. 2014. “Modeling the Topography of Shallow Braided Rivers
Using Structure-from-Motion Photogrammetry.” Geomorphology 213: 166–182. doi:10.1016/j.
geomorph.2014.01.006.
Kenner, R., M. Phillips, C. Hauck, C. Hilbich, C. Mulsow, Y. Bühler, A. Stoffel, and M. Buchroithner. 2017.
“New Insights on Permafrost Genesis and Conservation in Talus Slopes Based on Observations at
Flüelapass, Eastern Switzerland.” Geomorphology 290: 101–113. doi:10.1016/j.geomorph.2017.04.011.
Lague, D., N. Brodu, and J. Leroux. 2013. “Accurate 3D Comparison of Complex Topography with
Terrestrial Laser Scanner: Application to the Rangitikei Canyon (N-Z).” ISPRS Journal of
Photogrammetry and Remote Sensing 82: 10–26. International Society for Photogrammetry and
Remote Sensing, Inc. (ISPRS). doi:10.1016/j.isprsjprs.2013.04.009.
Lannoeye, W., C. Stal, E. Guyassa, A. Zenebe, J. Nyssen, and A. Frankl. 2016. “The Use of SfM-
Photogrammetry to Quantify and Understand Gully Degradation at the Temporal Scale of
Rainfall Events: An Example from the Ethiopian Drylands.” Physical Geography 37 (6): 430–451.
doi:10.1080/02723646.2016.1234197.
Lowe, D. G. 2004. “Distinctive Image Features from Scale-Invariant Keypoints.” International Journal
of Computer Vision 60 (2): 91–110. doi:10.1023/B:VISI.0000029664.99615.94.
Niethammer, U., M. R. James, S. Rothmund, J. Travelletti, and M. Joswig. 2012. “UAV-Based Remote
Sensing of the Super-Sauze Landslide: Evaluation and Results.” Engineering Geology 128: 2–11.
doi:10.1016/j.enggeo.2011.03.012.
Piermattei, L., L. Carturan, F. De Blasi, P. Tarolli, G. D. Fontana, A. Vettore, and N. Pfeifer. 2016.
“Suitability of Ground-Based SfM-MVS for Monitoring Glacial and Periglacial Processes.” Earth
Surface Dynamics 4 (2): 425–443. doi:10.5194/esurf-4-425-2016.
Semyonov, D. 2011. “Algorithms Used in Photoscan [Msg 2].” www.agisoft.ru/forum/index.php?
topic=89.0
Smith, M. W., J. L. Carrivick, and D. J. Quincey. 2015. “Structure from Motion Photogrammetry in Physical
Geography.” Progress in Physical Geography 40 (2): 247–275. doi:10.1177/0309133315615805.
Uysal, M., A. S. Toprak, and N. Polat. 2015. “DEM Generation with UAV Photogrammetry and
Accuracy Analysis in Sahitler Hill.” Measurement: Journal of the International Measurement
Confederation 73: 539–543. doi:10.1016/j.measurement.2015.06.010.
Westoby, M. J., J. Brasington, N. F. Glasser, M. J. Hambrey, and J. M. Reynolds. 2012. “‘Structure-
From-Motion’ Photogrammetry: A Low-Cost, Effective Tool for Geoscience Applications.”
Geomorphology 179: 300–314. doi:10.1016/j.geomorph.2012.08.021.
Zevenbergen, L. W., and C. R. Thorne. 1987. “Quantitative Analysis of Land Surface Topography.”
Earth Surface Processes and Landforms 12 (1): 47–56. doi:10.1002/esp.3290120107.