You are on page 1of 16

Bulletin of Engineering Geology and the Environment (2020) 79:4717–4732

https://doi.org/10.1007/s10064-020-01870-3

ORIGINAL PAPER

A photogrammetric method for laboratory-scale investigation on 3D


landslide dam topography
Hao Wu 1 & De-feng Zheng 2 & Yan-jun Zhang 3 & Dong-yang Li 1 & Ting-kai Nian 1

Received: 10 January 2020 / Accepted: 28 May 2020 / Published online: 13 June 2020
# Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
Topography is a crucial characteristic reflecting the stability status of a landslide dam. The methods for measuring landslide dam
topography in a laboratory-scale test study are currently operator dependent, time-consuming, or only applicable to the measure-
ment of the two-dimensional section. In this paper, a laboratory-scale photogrammetric method based on the structure from motion
(SfM) technique was proposed to measure the three-dimensional (3D) topography of a landslide dam. The SfM technique, which is
a revolutionary, low-cost, user-friendly computer vision technique, was employed for reconstructing a landslide dam 3D computer
model. A scientific method to determine the topography parameters of a landslide dam was put forward. Meanwhile, two materials
with different particle sizes were used to simulate the formation process of the landslide dam in the laboratory-scale. Then, the
measurement results of the two materials were compared. Five parameters of a landslide dam topography with 100 parallel
measurement results for each parameter were obtained. The results show that the SfM technique could build a high-quality 3D
point cloud in a laboratory scale. The proposed method of determining the topography parameters of a landslide dam was useful and
has low measurement uncertainty. The material type affected the sparse density of the point cloud and then affected the measure-
ment uncertainty of landslide dam topography parameters. The measurement uncertainty of the gravel-type landslide dam was
significantly higher than that of the sand-type landslide dam. This research contributes to promoting the application of a photo-
grammetric method based on the SfM technique in geotechnical engineering laboratory-scale tests.

Keywords Photogrammetric method . Structure from motion technique . Landslide dam disaster . Three-dimensional
topography . Non-contact measurement

* De-feng Zheng Introduction


defengzheng@lnnu.edu.cn
* Ting-kai Nian Landslide, rockfall, and debris flow, caused by slope instabil-
tknian@dlut.edu.cn ity, are the most common mass movements in mountainous
Hao Wu areas throughout the world, threatening the safety of human
haowugeot@mail.dlut.edu.cn life and property (Nian et al. 2020; Sassa et al. 2015). The
mass movements may move into valleys and block rivers,
Yan-jun Zhang
yjzhang_cgs@163.com resulting in the formation of landslide dams (Casagli and
Ermini 1999; Costa and Schuster 1988; Shrestha and
Dong-yang Li
ldygeot@mail.dlut.edu.cn
Nakagawa 2016; Zhang et al. 2013). Landslide dams consist
of a mixture of soil and rock and may fail in a short time,
1
School of Civil Engineering and State Key Laboratory of Coastal and causing catastrophic disasters for defenseless downstream
Offshore Engineering, Dalian University of Technology, No. 2 areas (Costa and Schuster 1988; Cui et al. 2009; Dai et al.
Linggong Road, High-tech Zone, Dalian 116024, Liaoning, China 2005; Kidyaeva et al. 2017). Assessing the stability status of
2
School of Urban and Environmental Science, Liaoning Normal a landslide dam in a limited time is an important task.
University, No. 850, Huanghe Road, Shahekou Zone, However, the stability evaluation cannot be carried out be-
Dalian 116029, Liaoning, China
cause parameters, such as geotechnical physical and mechan-
3
Wuhan Center, China Geological Survey ical properties, are incapable of being obtained in a short time.
(Central South China Innovation Center for Geosciences),
Wuhan 430205, Hubei, China
Fortunately, previous studies showed that the stability status
4718 H. Wu et al.

of a landslide dam is associated with its three-dimensional of landscape evolution morphology quantification (Rossi
(3D) topography (Donnarumma et al. 2013; Ermini and et al. 2018; Saito et al. 2018). Recently, the photogrammet-
Casagli 2003; Nian et al. 2018; Korup 2005; Tacconi ric method based on the SfM technique has brought a rev-
Stefanelli et al. 2016). It is, therefore, very important to carry olutionary breakthrough for the experimental study.
out laboratory experimental studies on the 3D topography Wójcik et al. (2018) described how the photogrammetric
evolution of landslide dams. However, the primary problem method was applied to measure the repose angle of granu-
raised in experimental studies is how accurate measuring the lar materials. Masoodi et al. (2018) used Kinect which was
3D landslide dam topography. commonly used for human body tracing and indoor
At present, the measurement methods in geotechnical lab- mapping to monitor seepage erosion of riverbank by
oratory experimental studies are of two main classes: contact laboratory experiments. Galland et al. (2016) described in
measurement method and non-contact measurement method great detail the implementation of an open source SfM
(Li and Zhang 2018; White et al. 2001; White et al. 2003). For program (MicMac) for topography monitoring on labora-
example, in terms of the contact measurement method, grid tory experiments of magma emplacement in the crust and
lines are widely used as measuring instruments to measure the tectonic deformation. Schmiedel et al. (2017, 2019) inves-
landslide dam topography in laboratory flume experiments tigated the relationship between intrusion shape and host
(Awal et al. 2009; Jiang et al. 2017; Niu et al. 2012; Sharif material strength and deformation mode according to the
et al. 2015; Yang et al. 2015). Despite being a simple and time series of topographic maps measured by the photo-
effective approach, the measured results from this approach grammetric method, which was proposed by Galland et al.
are relatively inaccurate. Because landslide dam topography is (2016). Although this method appears promising and pro-
very complex, it is difficult to obtain 3D landslide dam topo- duces high-resolution results, it has surprisingly rarely
graphical characteristics through the methods mentioned been used in the laboratory-scale landslide dam topogra-
above. Moreover, the contact measurement may cost time, phy investigation. There has been no detailed investigation
and the target object status may be subjected to change during of measurement error analysis when the photogrammetric
measurement. method is applied in a laboratory-scale experimental study.
Non-contact measurement refers to the process of To implement the SfM technique in a laboratory-scale study
obtaining the surface parameter information of an object with- of landslide dam topography measurement, three issues must be
out touching the surface of the object through photoelectric, addressed. (a) According to our previous experiments on the 3D
electromagnetic, and other technologies (Son and Kim 2017; topography of landslide dams (Wu et al. 2020), it is difficult to
Tran et al. 2017). The non-contact measurement method is quantify the length, width, and height of a landslide dam due to
more rapid than the contact measurement method and reduces the irregular accumulation of granular materials. There is an ab-
operator dependence. At present, the non-contact measure- solute randomness in the measurement results. Thus, a system-
ment method based on point cloud reconstruction has been atic scientific method for measuring the acquired landslide dam
widely used in civil engineering site surveys (Liu et al. topography needs to be determined. (b) There has been no de-
2014). This method has achieved a revolutionary technologi- tailed investigation of measurement error analysis using the pho-
cal breakthrough from a single point measurement to full togrammetric method based on the SfM technique. Although the
three-dimensional surface measurements. For example, appli- SfM technique is widely used in regional site surveys, the accu-
cations in landslide deformation monitoring (Teza et al. 2007), racy of the measurement method still lacks evaluation when it is
tunnels excavation geometry inspection (Pejić 2013), and used in a laboratory-scale study. (c) Wójcik et al. (2018) found
slope stability and rockfall runout studies (Salvini et al. that material characteristics (e.g., texture and grain size) of the
2013) have been widely reported. However, photoelectricity object under measure affected the quality of point cloud recon-
devices used in non-contact measurements are extremely cost- struction of the SfM technique and further directly affect the
ly and are not economical to use in the indoor laboratory. accuracy of the measured results. Hence, the effects of geotech-
Moreover, due to the limitations of equipment size and the nical materials with different particle sizes on the measurement
experimental environment, these devices may not be applica- results should be recognized.
ble in the laboratory-scale study. This paper is aimed at proposing a photogrammetric meth-
Because of its low cost and high efficiency, the structure od for scientifically measuring landslide dam topography at
from motion (SfM) technique, which is a computer vision the laboratory scale. To achieve this goal, first, experimental
technique for high-resolution topographic reconstruction, equipment was built, and the layout of the equipment and the
has been rapidly developed for applications in geological photo acquisition approach were illustrated. Then, a scientific
engineering (Westoby et al. 2012), such as field investiga- method to determine the topographic parameters of a landslide
tions of rock slope characterization (Riquelme et al. 2016; dam was proposed. Third, the measurement method was test-
Török et al. 2018), tunnel surfaces monitoring (García- ed by materials with two different particle sizes (fine sand
Luna et al. 2019), and the quantification of the morphology group and the coarse gravel group). Finally, two indexes
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4719

(point cloud density and measurement uncertainty) were used overlapping photos; thus, it differs fundamentally from
to evaluate the accuracy of the measurement results of the conventional photogrammetry in that, without the need
proposed method. to specify a priori, a network of targets that have
known 3D positions is automatically obtained.
The schematic diagram of the SfM technique is shown in
Fig. 1. Assuming that the imaging point of the spatial point
Materials and methods CPi is CPi-n and the corresponding homogeneous coordinates
are CPi = (xw, yw, zw, 1)T and CPi-n = (u, v, 1)T, the following
General description of the structure from motion mathematical model was given by the previous study (Sturm
technique principle and Triggs 1996):
2 3
The structure from motion (SfM) technique is an imag- 2 3 2 32 3 xw
ing technique for estimating 3D structures from two- u f =d x 0 u0 r1 r2 r3 lx 6 7
yw 7
dimensional photo sequences that are used in the field λ4 v 5 ¼ 4 0 f =d y v0 5 4 r 4 r5 r6 l y 56
4 zw 5
1 0 0 1 r7 r8 r9 lz
of computer vision (Sturm and Triggs 1996). The 3D 1
structure (point locations) and motion (camera locations) 2 3 2 3
xw xw
can be recovered from the photo measurements. The 6 yw 7 6 yw 7
SfM technique simultaneously calculates the geometry ¼ K ðRt Þ6 7 6 7
4 zw 5 ¼ M 4 zw 5
of the scene and the positions and orientations of the
1 1
camera using a highly redundant, iterative bundle ad-
justment procedure based on a set of multiple ð1Þ

Fig. 1 Schematic diagram of the structure from motion technique. CPi is the number i spatial point on the experiment target. CPi-n is the number i
imaging point on the number n photo, which corresponds to the number i spatial point
4720 H. Wu et al.

λ is the non-zero scaling factor, which is called the projec- experiment, the same volume, 0.02 m3, of the two materials
tion depth; R and t are the external parameter matrixes, which was used to form landslide dams, so as to eliminate the influ-
represent the transformation relationship between the camera ence of the measured object volume on the evaluation of the
coordinate system and the world coordinate system. The rota- measured result.
tion matrix R is composed of r1~r9, and the translation vector t The layout of the experiment equipment is shown in Fig. 3.
is composed of lx, ly, and lz. K is the camera’s internal param- The equipment mainly included the landslide dam simulation
eter matrix, including the focal length f, the main point coor- system and the photo acquisition system.
dinates (u0, v0), and the pixel physical size dx, dy, and M = The landslide dam simulation system consisted of two
K(Rt) is the projection matrix. parts: a landslide simulation part and a landslide dam-
forming part. The landslide simulation part was mainly com-
Experimental scheme posed of an angle adjustment unit, a lifting motor, and a stor-
age tank that consists of a bottom plate, two side plates, a
The performance of the photogrammetric method applying in cover plate, and a baffle plate. These plates were able to move
the laboratory-scale tests was investigated by two materials freely and are assembled so that the position and size of the
with different particle sizes, which were the fine sand group storage tank can be changed. To simulate different slope slid-
and the coarse gravel group. The repose angles of the fine sand ing surface angles, continuous angular changes from 0° to 90°
group and the coarse gravel group were 35° and 45°, respec- could be achieved via the angle adjustment unit and the lifting
tively. Figure 2 shows the grain size distributions of the two motor. The storage tank was used to store the test materials.
different materials. The particle size range of the fine sand The landslide dam-forming part was 4.6 m long, 0.3 m wide,
group was 0.1–2 mm. The median diameter of the fine sand and 0.2 m high and is made of plexiglass. The landslide dam-
group was 0.60 mm. The specific gravity of the fine sand forming part was set perpendicular to the landslide simulation
group was 2.28. The non-uniform coefficient and curvature at the downstream end to receive the landslide mass
coefficient of the fine sand group were 2.19 and 0.96, respec- forming the landslide dam. The landslide dam-forming
tively. The particle size range of the gravel group was 2– part was also installed with an angle adjustment unit so
20 mm. The median diameter of the gravel group was that the angle can be changed from 0° to 12° to simu-
9.42 mm. The specific gravity of the gravel group was 2.50. late different valley bed inclinations.
The non-uniform coefficient and curvature coefficient of the The photo acquisition system in the experiments included a
gravel group were 1.66 and 1.03, respectively. In the small motion camera, an industrial personal computer, a
shooting path guide rail, and control points labels, as shown
in Fig. 3. The GoPro Hero 4 Black camera was utilized in the
tests. The nominal focal length of the camera was 3 mm. The
pixel size of the camera was 0.00155 mm (sensor size of 6.2 ×
4.65 mm). The GoPro Hero 4 Black camera had several photo
capture modes. In this study, the “Medium” mode was uti-
lized. The vertical and horizontal FOV of the “Medium” photo
capture mode were 94.4° and 122.6°, respectively. The photo
size was 7 MP (3000 × 2500 pixel). The computer configura-
tion used in this study was i7-6700K, 4.00 GHz, NVIDIA
GeForce GTX 970 2 GB. The shooting path guide rail was
composed of steel wire. The camera was attached to the shoot-
ing path guide rail and can move freely along the shooting
path guide rail. Marks, which were used as the camera shoot-
ing station, were made every 20 mm in the shooting path
guide rail. The control point labels, which were composed of
16-mm squares, were arranged on the landslide dam-forming
Fig. 2 Particle size distributions of two different materials used in the part (as shown in Fig. 3). A total of 27 control points with
tests. d10 means that the mass of soil particles smaller than this particle
size accounts for 10% of the total mass of soil particles and is also known
known coordinate were randomly distributed on the control
as effective particle size. d30 means that the mass of soil particles smaller point labels. The control points in this paper served two pur-
than this particle size accounts for 30% of the total mass of soil particles. poses. On the one hand, the control points were used to realign
d50 means that the mass of soil particles smaller than this particle size the acquired photos, which ensured accurate calculation of the
accounts for 50% of the total mass of soil particles and is also known as
median particle size. d60 means that the mass of soil particles smaller than
camera position in the custom coordinate system. On the other
this particle size accounts for 60% of the total mass of soil particles and is hand, the coordinates of the control points were used to verify
also known as the control particle size the accuracy of the measurement results.
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4721

Fig. 3 Layout of the experiment


equipment. a Material is put into
the storage tank before testing. b
A landslide dam forming

As shown in Fig. 3b, the camera was fixed on the shooting study, as the distance from the object is very small, which is
path guide rail with a 300-mm distance from the flume bot- approximately 300 mm (as shown in Fig. 3), the strong dis-
tom. A photo was taken approximately every 20 mm under a tortion effects are heightened. To ensure the accuracy of photo
fixed lighting condition. A total of 21 photos were taken dur- processing results, the inner orientation elements and optical
ing each entire measuring procedure. Based on the captured aberration of the GoPro Hero 4 Black camera should be
photos, a 3D point cloud of the landslide dam was reconstruct- checked for measurement purposes. The camera calibration
ed by the SfM technique. Even if the same object is measured work is mainly to obtain the following two groups of data:
in the same measurement method, the results of multiple mea- (1) interior orientation elements: the principal points coordi-
surements may be different. This is because an occasional nate (x0, y0) and the focal length of the camera (fx, fy); (2)
error cannot be avoided. In order to evaluate the accuracy of optical distortion: radial distortion (k1, k2, k3) and decentering
the measurement results by statistical analysis method, multi- distortion (p1, p2). The main camera correction methods in-
ple measurements of the same measurement object were re- clude direct linear transformation, bundle adjustment, Tsai
quired. Since the occasional errors may exist in the processes two-stage technique, and plane grid method (Gai et al.
of photo capture, point cloud generation, and parameter mea- 2018). In this paper, the plane grid method was used for cam-
surement, each process was repeated 10 times in this study. era calibration. The checkerboard grid was used as the cali-
Thus, a total of 10 landslide dam computer models expressed bration plane to shoot the checkerboard grid images from
by n = {1, …, 10} were formed after 10 times of photo capture different angles, following the steps of the literature (Balletti
processes and 3D point cloud generation processes. For each et al. 2014). The distances of taking photos should be evenly
landslide dam computer model, the parameter measuring pro- spaced and vary greatly in general, and a total of 14
cess was also repeated 10 times, according to an m = {1, …, images were taken. The program of camera calibration
10} repetition scheme. Thus, for each landslide dam parame- was compiled on MATLAB. The results of camera cal-
ter, 100 parallel measurement results were obtained. In the ibration are listed in Table 1.
“Data evaluation index” section, the obtained results were Taking a set of photographs as an example, the photo-
further statistically analyzed. graphs with strong distortion were converted to normal photos
In photogrammetric applications, a GoPro Hero 4 Black (as shown in Fig. 4).
camera has many advantages from a measurement perspec-
tive: it is compact, flexible, portable low-cost, and can be used
in complex conditions. Conversely, the disadvantage is that Photogrammetric procedure
the wide-angle lenses generate a distorting effect known as the
“Fish-Eye” effect (Balletti et al. 2014; Bolognesi et al. 2015). Figure 5 presents the procedure of the photogrammetric meth-
Because of loss of output accuracy, the use of the GoPro Hero od. As shown, the measurement method is based on the se-
4 Black camera in photogrammetry is not common. In this quential application of the following four main steps:
4722 H. Wu et al.

Table 1 The results of camera


calibration The camera parameters The calibration results/pixel

Focal length (fx, fy) (1.5607e3, 1.5607e3)


Coordinates of principal point (x0, y0) (− 63.6139, 5.3674)
Optical distortion (k1, k2, k3, p1, p2) (− 2.6815e−1, 1.1462e−1, − 3.6827e−2, 4.2210e−04, 9.0325)

Step 1: Preparatory work. To implement the photogrammet- Step 3: Building 3D computer model. The photographs ob-
ric method, a user-defined coordinate system had to tained from the second step may be distorted, which
be determined in advance. In this paper, the X- and Y- has an impact on 3D modeling. Thus, it is necessary
axes were on the plane of the flume, the Z-axis was to adjust the distortion of the photo, then determine
parallel to the normal vector of the flume, and the the control points in each photo artificially and input
coordinate origin was located downstream of the riv- the space coordinates of the control points. Next, the
erbed. Thus, a right-handed rectangular space coor- sparse point cloud and the dense point cloud of the
dinate system was established by taking the direction measured object are reconstructed using the com-
from the downstream to the upstream of the flume as mercial software Agisoft PhotoScan (Agisoft
the positive direction of the X-axis (as shown in Fig. 2016), which is a 3D scanning software package
3b). Enough control points need to be placed in the developed by the Agisoft company, Russia. For es-
measurement area, and the 3D coordinates of control tablishing the 3D model of the measured object ac-
points should be measured. Additionally, a fixed fo- curately, the points outside the study area and the
cus camera should be selected, and its lens should be noise points can be deleted in advance. Then, the
corrected before performing the measurement. Good spatial characteristics of the point cloud are derived,
lighting is also required to achieve better quality of and a best fit 3D geometry is generated by using
the results, parallel views with moving cameras over Kriging interpolation method, which configures the
a relatively flat and extensive object with respect of geometric parameters of a given solid so that its ge-
the area shot by the camera. ometry best fits a tested point cloud.
Step 2: Acquiring photos. As a landslide dam occupies a Step 4: Measuring topography parameters. The topography
relatively large flat space, the photos should be taken of the landslide dam is not regular, and different
as a parallel view with a motion camera on the results may be obtained from each measurement.
shooting path guide rail, which was described in Therefore, the measurement criteria, which were de-
detail in the “Experimental scheme” section. For scribed in detail in the “Parameter definition and
reconstructing the 3D model, two adjacent photos acquisition” section, need to be defined at first.
should overlap at least 60%, and each photo should Besides, in order to reduce the measurement error,
contain at least 4 control points. It is worth noting it is necessary to increase the measurement times.
that acquiring more photos can reconstruct a better Two indexes (point cloud density and measurement
3D model. Details such as textures and gaps can be uncertainty) were defined to evaluate the accuracy of
reconstructed clearly. the measurement results. The two indexes for

Fig. 4 Photograph distortion calibration: a photographs with strong distortion, b normal photographs after calibration
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4723

Fig. 5 Procedure of
photogrammetric method

evaluating the accuracy of the measurement results defined as Lt, and the bottom length is defined as Lb. θ is
were described in detail in the “Data evaluation in- defined as the valley bed slope, which is the angle between
dex” section. the plane where the riverbed lies and the horizontal plane.
There are two ways to describe the angles of the upstream
and downstream slopes of the landslide dam. One way is to
define the angle as the angle between the surface of the up-
Parameter definition and acquisition stream or downstream slope of the landslide dam and the
plane of the valley bed, which we call the inner angle.
As shown in Fig. 6a, the simplified topography of a landslide Another way is to define the angle as the angle between the
dam can be defined by the following eight parameters (Wu surface of the upstream or downstream slope of the landslide
et al. 2020). H is the height of the landslide dam and indicates dam and the horizontal plane, which we call the outer angle.
the distance between the top surface and the bottom surface. βd and βu are defined as the inner angles of the downstream
The top length of a longitudinal section of the landslide dam is slope and upstream slope, respectively, while βd’ and βu’ are

Fig. 6 Methods of parameter


definition and longitudinal
section simplification. a
Parameter definition method
based on simplified longitudinal
section of a landslide dam. b A
method for simplify the actual
longitudinal section of a landslide
dam
4724 H. Wu et al.

defined as the outer angles of the downstream slope and up-  


H
stream slope, respectively. The outer and inner angles are β d ¼ arctan
L
 1
geometrically related. Five parameters, therefore, can define ð2Þ
H
the topography of a landslide dam. β u ¼ arctan
However, the topography of a landslide dam cannot be as L2
regular as the simplified in Fig. 6a. For many granular mate-
rials (e.g., sands and gravels), there is a general tendency for
the accumulation body to curve at the top and have bulk ma-
Data evaluation index
terial accumulate at its base, as shown in Fig. 6b. The curved
top of the landslide dam model (as shown in Fig. 6b) differs
Digital photo processing, 3D modeling, and point cloud gen-
substantially from the simplified flat top (as shown in Fig. 6a).
eration entail certain errors that may affect the quality of
It is, therefore, difficult to define the top length (Lt) of the
photos and the accuracy of measurement results (Wójcik
landslide dam model. Furthermore, the loose material of the
et al. 2018). In this study, a total of 10 landslide dam computer
landslide dam toe lies a great distance from the central of the
models expressed by n = {1, …, 10} were formed. For each
landslide dam (as shown in Fig. 6b), which may, therefore,
landslide dam computer model, parameter measuring proce-
underestimate the inner angles of the downstream slope and
dure was repeated 10 times, according to an m = {1, …, 10}
upstream slope (βd and βu). For these various reasons, to sci-
repetition scheme. Thus, for each landslide dam parameter,
entifically measure the parameters of landslide dam topogra-
100 parallel measurement results were obtained.
phy, a method for simplifying the actual longitudinal section
Accordingly, the accuracy of the measurement results was
of a landslide dam was proposed. The following provisions
evaluated using two indexes, which were point cloud density
were made in this study (as shown in Fig. 6b):
and measurement uncertainty. The point cloud density was
used to compare the difference of measurement results be-
(1) A total of 10–15% of the bottom region of the landslide
tween sands and gravels. The measurement uncertainty was
dam model is omitted.
used to analyze the accuracy of the measurement results of
(2) A total of 15–20% of the top region of the landslide dam
five landslide dam parameters.
model is omitted.
(3) The remaining 65–75% of the point cloud is the actual
representation of the upstream and downstream surface.
Point cloud density
The best fitting line of the remaining part is used as the
upstream and downstream surface of the landslide dam
Point cloud density is an important attribute of SfM point
model.
cloud data, which reflects the characteristics of point space
(4) The angles between the two fitting lines and the valley
distribution, while the characteristics of point space distribu-
bed are defined as the inner angles of the downstream
tion directly reflect the spatial distribution and characteristics
slope and upstream slope (βd and βu).
of measurement objects (Clapuyt et al. 2016; Westoby et al.
(5) The valley bed intersects the two fitting lines at two
2012). The high-quality point cloud density is more conducive
points. The distance between the two points is the bottom
to reconstruct a high-quality measurement model. The point
length (Lb) of a landslide dam.
cloud density was defined in this study according to Formula
(6) A straight line parallel to the valley bed and across the
(3):
highest point of the landslide dam is made. This straight
line intersects the two fitting lines at two points. The N n;k
Dn;k ¼ ð3Þ
distance between the two points is the top length (Lt) of S n;k
a landslide dam.
(7) The dam height (H) is the vertical distance from the where Dn,k is the point cloud density of the landslide dams.
valley bed to the highest point. Nn,k is the number of points composing the 3D point cloud,
and Sn,k is the surface area of the landslide dam. When k = 1, it
The parameters of Lb, Lt, and H can be measured directly means sand; when k = 2, it means gravel. n = {1, …, 10}
on the landslide dam model. However, the parameters of βd indicates 10 landslide dam computer models.
and βu cannot be measured directly. According to the topog- From formula (3), the point cloud density (Dn,k) can be
raphy shown in Fig. 6b, the parameters of βd and βu can be understood as the number of points within the area of the
expressed by the following (Eq. (2)): target object unit. When Dn,k is larger, the point cloud quality
is higher, and the target object is well reconstructed.
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4725

Measurement uncertainty The parameters H, Lt, Lb, L1, and L2 were determined di-
rectly by measuring the landslide dam 3D computer model.
The measurement uncertainty is the evaluation of the magni- According to formula (4), the measurement uncertainties of
tude range that characterizes the actual value of a measure- these parameters can be expressed by the following (Eq. (5)):
ment. It is a scientific description of the uncertainty degree of sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1  2
measurement results affected by measurement error.
uðH Þn¼f1;…;10g ¼ ∑m H i −H
Specifically, measurement uncertainty quantification repre- mðm−1Þ i¼1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sents the comprehensive distribution range of random error
1  2
and systematic error. According to the standard uncertainty uðLt Þn¼f1;…;10g ¼ ∑m L t −L t
mðm−1Þ i¼1 i
determined by the error propagation formula (root-mean- sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
square error), the measurement uncertainty can be estimated 1  2
uðLb Þn¼f1;…;10g ¼ ∑m i¼1 Lbi −Lb
ð5Þ
as follows: mðm−1Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2
 2 1
1 uðL1 Þn¼f1;…;10g ¼ ∑m i¼1 L1i −L1
uðxÞn¼f1;…;10g ¼ ∑m x i −x ð4Þ mðm−1Þ
mðm−1Þ i¼1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1  2
uðL2 Þn¼f1;…;10g ¼ ∑m i¼1 L2i −L2
where u(x)n denotes the measurement uncertainty of a param- mðm−1Þ
eter. xi indicates measurement result of a parameter. x indi-
cates the average value of the 10 measurement results. n = {1, As βu and βd were determined indirectly by calculating Eq.
…, 10} indicates the number series of the 10 landslide dam (2), the measurement uncertainties of these parameters can be
computer models. m = {1, …, 10} indicates the number series expressed by the following (Eq. (6)):
of 10 parallel measurement results on each landslide dam 3D
computer model.

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u0 12 0 12
u
u H L
¼ t@ 2 A u2 ðL1n Þ @ A u2 ð H n Þ
n 1n
uðβd Þn¼f1;…;10g 2 n¼f1;…;10g þ 2 2 n¼f1;…;10g
H n þ L1n H n þ L1n
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð6Þ
u0 12 0 12
u
u H L
¼ t@ 2 A u2 ðL2n Þ @ A u2 ð H n Þ
n 2n
uðβu Þn¼f1;…;10g 2 n¼f1;…;10g þ 2 2 n¼f1;…;10g
H n þ L2n H n þ L2n

For a given n series, the final values of the five


sp ðxnm Þ
landslide dam topographic parameters with the measure- upn ¼f1;…;10g ¼ pffiffiffiffi ð9Þ
ment uncertainties are equal to x  uðxÞc. x is the aver- m
age value of the morphological parameters of a land- where Sp(xnm) indicates the standard deviation.
slide dam. uc is measurement uncertainty that can be
expressed in the formula below:

∑ni¼1 uðxÞn wn Results


uð x Þ c ¼ ð7Þ
∑ni¼1 wn
Validation
where wn indicates weighted mean, for a given series;
wn is expressed by the following (Eq. (8)): Ten of the 27 control points were selected to verify the preci-
sion of the SfM technique measurement results. Taking one of
1
wn¼f1;…;10g ¼ ð8Þ the 20 landslide dam 3D computer models as an example,
u2pn
Table 2 shows a comparison of the actual coordinates and
the measured coordinates of the 10 control points. The mea-
where upn indicates measurement uncertainty of the sured coordinates were found to be relatively close to the
measurement average in a given series, which is defined actual coordinates. The root-mean-square errors (RMSEs) of
by the following (Eq. (9)): the X, Y, and Z coordinates were 0.151 mm, 0.171 mm, and
4726 H. Wu et al.

Table 2 Comparison of the actual


coordinates and the measured Control Actual coordinates (mm) Measured coordinates (mm) Difference value (mm)
coordinates of the 10 control point
points number X Y Z X Y Z DX DY DZ

CP01 1357.2 303.0 200.0 1357.041 303.087 200.035 − 0.159 0.087 0.035
CP02 1308.7 303.0 200.0 1308.704 303.121 199.942 0.004 0.121 − 0.058
CP03 1244.7 303.0 200.0 1244.452 303.069 199.937 − 0.248 0.069 − 0.063
CP04 1196.2 303.0 200.0 1196.302 303.126 199.955 0.102 0.126 − 0.045
CP05 1373.5 − 8.0 200.0 1373.446 − 8.036 199.943 − 0.054 − 0.036 − 0.057
CP06 1293.5 − 8.0 200.0 1293.375 − 8.206 199.839 − 0.125 − 0.206 − 0.161
CP07 1197.0 − 8.0 200.0 1196.911 − 8.257 199.791 − 0.089 − 0.257 − 0.209
CP08 1116.2 303.0 200.0 1116.083 303.256 199.945 − 0.117 0.256 − 0.055
CP09 1132.5 − 8.0 200.0 1132.705 − 8.125 199.920 0.205 − 0.125 − 0.080
CP10 1067.7 303.0 200.0 1067.885 303.183 199.936 0.185 0.183 − 0.064
RMSE 0.151 0.171 0.066

0.066 mm, respectively, which were all less than 0.2 mm. It As shown in Fig. 7a–d, the sparse point cloud and the dense
can be concluded that the landslide dam 3D computer model point cloud can be generated by the SfM technique, following
reconstructed by the SfM technique has high precision. the steps described in the “Photogrammetric procedure” sec-
Additionally, as the CPLs were arranged on the landslide tion. It can be observed that different measuring materials
dam-forming part, the measuring materials had no effect on affect the number of points. Based on the 3D point cloud,
the measurement of the control point coordinates. the landslide dam 3D computer model can be reconstructed,

Fig. 7 The SfM technique is applied in a laboratory-scale test study: a of a gravel-type landslide, e reconstruction of the sand-type landslide dam
generation of a sparse point cloud of a sand-type landslide, b generation 3D computer model, and f reconstruction of the gravel-type landslide dam
of a sparse point cloud of a gravel-type landslide, c generation of a dense 3D computer model
point cloud of a sand-type landslide, d generation of a dense point cloud
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4727

as shown in Fig. 7e, f. The parameters of landslide dam to- Parameter analysis
pography can be measured based on the 3D computer model.
By using the computer modeling model (as shown in Fig. Two indexes, point cloud density and measurement uncertain-
7e, f), any position of interest can be extracted for specific ty, which are described in the “Data evaluation index” section,
analysis. However, it is almost difficult to measure the com- are analyzed in this section.
plete surface of a landslide dam by traditional contact mea- The surface areas of 20 landslide dam 3D computer models
surement methods. To verify the reliability of the measured and the densities of the sparse and dense point cloud of the
results of the computer modeling model, the height of the landslide dams are listed in Table 3. The average surface area
adjacent bank and the opposite bank of the landslide dam of the 10 gravel-type landslide dam 3D computer models was
was selected for comparison of the measured results between 1.4 times that of the sand type. The average density of the
the photogrammetric measurement on the computer modeling sparse point cloud of the sand-type landslide dam, 72.835 ×
model and direct measurement on the prototype model. To 10−3/mm2, was almost 3.5 times that of the gravel-type land-
measure on the prototype model directly, transparent grid pa- slide dam, 20.525 × 10−3/mm2. In contrast, the average densi-
pers were pasted on both sides of the plexiglass flume before ties of the dense point cloud of the two type landslide dams
an experiment. After the experiment, the longitudinal profiles almost coincide, which were 45.266/mm2 and 46.381/mm2,
of the adjacent bank and the opposite bank of a landslide dam respectively. The reason for the two densities of the dense
are shown in Fig. 8. The size of the adjacent bank and the point cloud was similar may be that the same SfM program
opposite bank of the landslide dam obtained from the test can was used in this study. The density of dense point clouds may
be measured through the transparent grid paper and steel ruler. be only related to the SfM algorithm. It can be concluded that
Figure 9 indicates the discrepancies between the two the density of the sparse point cloud was significantly corre-
measurement methods (measured on the prototype mod- lated with the particle size of the geotechnical materials; how-
el and measured on the computer modeling model). It ever, the density of the dense point cloud was independent of
can be observed from Fig. 9 that the results obtained by the particle size of geotechnical materials.
the two methods are in good agreement. Due to the Although the material type has no significant effect on the
limited data points measured on the prototype model, density of the dense point cloud, it may affect the measurement
the measured contour lines of the adjacent bank and uncertainty. According to Eqs. (4–9), 10 groups of the measure-
the opposite bank of the landslide dam were not smooth ment values with the uncertainty of the five landslide dam topo-
enough. However, the data points acquired by point graphic parameters are shown in Fig. 10. It can be seen from
cloud were abundant. As a consequence, the contour Fig. 10 that the measurement method proposed in this study had
lines of the adjacent bank and the opposite bank of good applicability in laboratory-scale tests. The measurement
the landslide dam extracted from the computer modeling uncertainty was quite small and can meet the requirement of
model correspond more to the actual condition. measuring accuracy. The measurement uncertainties of the top

Fig. 8 The longitudinal profiles


of the landslide dam obtained by
experiments: a the longitudinal
profiles of the adjacent bank of
landslide dam composed by fine
sand; b the longitudinal profiles
of the opposite bank of landslide
dam composed by fine sand; c the
longitudinal profiles of the
adjacent bank of landslide dam
composed by coarse gravel; d the
longitudinal profiles of the
opposite bank of landslide dam
composed by coarse gravel
4728 H. Wu et al.

Fig. 9 The discrepancies between the two measurement methods: a the measurement comparison results of fine sand; b the measurement comparison
results of coarse gravel

and bottom lengths of the gravel- and sand-type landslide dams uncertainties, the measurement uncertainty of the gravel-type
were both less than 4 mm. Among these measurement landslide dam top length was the largest, but it only accounted
for 1% of the top length. The measurement uncertainties of the
inner angle of the gravel and sand-type landslide dam were both
less than 0.5°, which only accounted for 0.5%–1.6% of the mea-
Table 3 The surface areas of 20 landslide dam 3D computer models
and the densities of the sparse and dense point clouds of the landslide dam
sured value (as shown in Fig. 10d, e).
It is a remarkable fact that the height measurement uncer-
Experimental Group Surface area Density of Density of tainties of the two landslide dam types were close and very small,
material number of landslide sparse point dense point which only accounted for 0.2% and 0.16% of the measured
dam (mm2) cloud (10−3/ cloud (/mm2)
mm2) results, respectively (as shown in Fig. 10a). The definition mode
of the landslide dam in this paper led to this result. Compared
Gravel No. 1 534,092 21.363 45.638 with the other geometric parameters, the definition mode of dam
No. 2 524,028 20.094 44.722 height dramatically reduced the measurement uncertainty.
No. 3 532,689 20.539 45.840 Besides, the particle size of granular materials had an effect
No. 4 537,923 20.492 44.427 on the dispersion of the measurement results of the geometric
No. 5 528,175 19.721 46.478 parameters. Except for the parameters of landslide dam height,
No. 6 527,854 20.214 46.017 the dispersion of the other geometric parameters of the gravel-
No. 7 536,720 20.467 45.381 type landslide dam was significantly higher than that of the
No. 8 532,498 21.114 44.885 geometric parameters of the sand-type landslide dam. For ex-
No. 9 538,245 20.823 45.034 ample, the uncertainty of the measured upstream inner angle
No. 10 536,890 20.414 44.241 of a gravel-type landslide dam was 2.7 times that of a sand-
Average 532,911 20.524 45.266 type landslide dam (as shown in Fig. 10d). It can be concluded
Sand No. 1 371,290 72.935 45.830 that materials with larger particle sizes, such as gravel, tending
No. 2 370,430 71.579 47.255 to accumulate in discrete patterns, led to higher uncertainty in
No. 3 363,875 74.658 46.192 measuring their geometric parameters.
No. 4 378,787 71.542 46.417
No. 5 370,234 73.219 46.197
No. 6 368,258 73.269 46.743 Discussion
No. 7 371,292 72.372 46.620
No. 8 369,907 73.256 46.572 The influence of measurement times on
No. 9 367,820 73.136 46.324 measurement uncertainty
No. 10 370,834 72.388 45.665
Average 370,273 72.835 46.381 As a novel measurement method, the measurement uncer-
tainties were investigated in detail through 100 sets of parallel
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4729

Fig. 10 Measurement values with uncertainty of five landslide dam dam, d measurement values with uncertainty of inner angle of landslide
topographic parameters: a measurement values with uncertainty of dam dam upstream, and e measurement values with uncertainty of inner angle
height, b measurement values with uncertainty of top length of landslide of landslide dam downstream
dam, c measurement values with uncertainty of bottom length of landslide

measurement results. However, it is unrealistic to analyze 100 measurement uncertainty. It can be concluded that a signifi-
sets of parallel measurement results for one parameter in the cant reduction in the measurement times may not reduce the
application. The measurement times need to be reduced. reliability of results. It is worth noting that further reduction of
Hence, the influence of measurement times on measurement measurement times will certainly reduce reliability of results.
uncertainty is discussed in this section. A balance between computational efficiency and reliability of
In order to investigate the possibility of reducing measure- results is required. In this study, we found that satisfactory
ment times, a total of 3 landslide dam computer models (n = 3) results could be obtained by 30 parallel measurements.
were selected randomly from all the datasets collected (n =
10). For each landslide dam computer model, parameter mea- Advantages of the proposed measurement method
suring was repeated 10 times, according to an m = {1, …, 10}
repetition scheme. Thus, for each landslide dam parameter, 30 The acquisition of the topographic parameters of landslide
parallel measurement results were obtained. Table 4 shows the dam is often challenging. For example, Awal et al. (2009)
selected measurement results. The results obtained for datasets pasted a grid line with a spacing of 10 mm on the sidewall of
of n = 3 were consistent with the results, shown in Fig. 10, a flume with a length of 0.5 m, a width of 0.3 m, and a depth
which were produced for the entire data collection within of 0.5 m. Niu et al. (2012) pasted a grid line with a spacing of

Table 4 The output quantities of the five landslide dam topographic parameters for n = 3 measurement times

Experimental material Measurement results Dam height Top length Bottom Inner angle Inner angle
H/mm Lt/mm length Lb/mm of upstream of downstream
βu/° βd/°

Gravel Average value 147.095 293.773 874.646 27.696 26.165


Measurement uncertainty (±) uc 0.242 3.229 3.274 0.392 0.395
Sand Average value 137.244 279.165 920.383 24.358 22.481
Measurement uncertainty (±) uc 0.142 1.195 2.209 0.191 0.293
4730 H. Wu et al.

0.5 m on the sidewall of a channel with a width of 0.4 m and Conclusions


a length of 12 m to mark the evolution characteristics of the
landslide dam topography. Sharif et al. (2015) studied the A measurement method based on the SfM technique was pro-
topography evolution characteristics of the erosion failure posed and applied in laboratory-scale experiments. Then, a
of a landslide dam in a flume with a length, width, and scientific method to determine the topography parameters of
height of 6.1 m, 0.45 m, and 0.25 m, respectively, using a a landslide dam was proposed. The measurement method was
grid line with a spacing of 50 mm. Jiang et al. (2017) studied tested for two materials with different particle sizes. Each
the process of dam failure by a grid line with an interval of material was subjected to the 3D computer model reconstruc-
50 mm in a flume with a length, width, and height of 6 m, tion for 10 times. For each landslide dam 3D computer model,
0.3 m, and 0.4 m, respectively. The above studies could only measuring procedure was repeated 10 times, which produced
measure the two-dimensional landslide dam topography 100 parallel output quantities for each parameter of the land-
evolution in the laboratory. To measure the 3D landslide slide dam topography. Two indexes were used to evaluate the
dam topography evolution process, Yang et al. (2015) accuracy of the measurement results of the proposed method.
installed a plane grid with a spacing of 50 mm in a flume The following conclusions could be drawn from this study:
with a width of 0.3 m and a length of 16 m. However, since
the plane grid is placed inside the flume, it will block the (1) The SfM technique could generate high-quality 3D point
movement path of the measured object. The previous exper- cloud data at the laboratory scale. The results show that
imental studies on landslide dam morphology could not the calculated coordinates of the control points are in
accurately obtain the 3D dam topography. For the first good agreement with their actual coordinates. The
time, the SfM technique was applied to a laboratory- RMSEs of the X, Y, and Z coordinates are 0.151 mm,
scale test to investigate landslide dam topography. In 0.171 mm, and 0.066 mm, respectively, which are all
comparison to the traditional measurement method, this less than 0.2 mm.
method may provide the 3D surface topography of the (2) The material type has an effect on the sparse density of
landslide dam. The above experiments clearly confirmed the point cloud and then affects the measurement uncer-
that the SfM based measurement method is able to re- tainty of landslide dam topography parameters. The mea-
construct the landslide dam 3D model with good accu- surement uncertainty of the gravel-type landslide dam is
racy. Therefore, this method may greatly promote labo- significantly higher than that of the sand-type landslide
ratory research on the characteristics of landslide dam dam.
topography. (3) A significant reduction in the measurement times may
The longitudinal section morphology of the landslide not reduce the reliability of results by comparing 30 par-
dam was usually defined as trapezoid and triangle allel measurement results with 100 parallel measurement
(Chen et al. 2015; Zhao et al. 2018). However, it is results.
found in this paper that the topography of the landslide
dam was irregular. Thus, the measurement of the topo- This paper promotes the application of the SfM technique-
graphic parameters was uncertain. The current work pro- based photogrammetric measurement method in geotechnical
vided a scientific method to measure the topographic engineering laboratory-scale tests. Future researches can focus
parameters of the landslide dam for the first time. on further improving measurement accuracy and measure-
According to the proposed measurement method, the ment automation.
measured topographic parameter dispersion was low.
This measurement method of the topographic parameters Acknowledgments The authors gratefully acknowledged the support
from the National Natural Science Foundation of China
of landslide dams can not only be used in laboratory-
scale model tests but also be extended to the landslide
Funding information This research was supported by the National
dam morphology field surveys. Natural Science Foundation of China (U1765107, 51879036, 51579032).
It may be mentioned here that some areas still need to
be further studied. The influence of the particle size of the
geotechnical material on the measurement results was in-
vestigated, but the internal reasons for such influence are References
not discussed in depth in this paper. The measurement
uncertainty results are primarily influenced by many fac- Agisoft LLC (2016) Agisoft PhotoScan Professional v. 1.2.4, available at
tors (Wójcik et al. 2018), such as photo alignment param- http://www.agisoft.com/
eters and SfM algorithms. From these aspects, the mea- Awal R, Nakagawa H, Kawaike K, Baba Y, Zhang H (2009) Three
dimensional transient seepage and slope stability analysis of land-
surement accuracy of the laboratory-scale test based on slide dam. Disaster Prevent Res Institute Ann B 52:689–696
the SfM technique can be further improved.
A photogrammetric method for laboratory-scale investigation on 3D landslide dam topography 4731

Balletti C, Guerra F, Tsioukas V, Vernier P (2014) Calibration of action Masoodi A, Noorzad A, Tabatabai MM, Samadi A (2018) Application of
cameras for photogrammetric purposes. Sensors 14(9):17471– short-range photogrammetry for monitoring seepage erosion of riv-
17490. https://doi.org/10.3390/s140917471 erbank by laboratory experiments. J Hydrol 558:380–391. https://
Bolognesi M, Furini A, Russo V, Pellegrinelli A, Russo P (2015) Testing doi.org/10.1016/j.jhydrol.2018.01.051
the low-cost RPAS potential in 3D cultural heritage reconstruction. Nian TK, Wu H, Chen GQ, Zheng DF, Zhang YJ, Li DY (2018) Research
Int Arch Photogramm Remote Sens Spat Inf Sci XL-5/W4:229–235 progress on stability evaluation method and disater chain effect of
Casagli N, Ermini L (1999) Geomorphic analysis of landslide dams in the landslide dam. Chin J Rock Mech Eng 37(8):1796–1812 (in
northern Apennine. Trans Jpn Geomorphol 20:219–249 Chinese)
Chen SC, Lin TW, Chen CY (2015) Modeling of natural dam failure Nian TK, Zhang YJ, Wu H, Chen GQ, Zheng L (2020) Runout simula-
modes and downstream riverbed morphological changes with dif- tion of seismic landslides using DDA with state-dependent shear
ferent dam materials in a flume test. Eng Geol 188:148–158. https:// strength model. Can Geotech J. https://doi.org/10.1139/cgj-2019-
doi.org/10.1016/j.enggeo.2015.01.016 0312
Clapuyt F, Vanacker V, Van Oost K (2016) Reproducibility of UAV- Niu ZP, Xu WL, Li NW, Xue Y, Chen HY (2012) Experimental inves-
based earth topography reconstructions based on Structure-from- tigation of the failure of cascade landslide dams. J Hydrodyn Ser B
Motion algorithms. Geomorphology 260(1):4–15. https://doi.org/ 24:430–441. https://doi.org/10.1016/S1001-6058(11)60264-3
10.1016/j.geomorph.2015.05.011 Pejić M (2013) Design and optimisation of laser scanning for tunnels
Costa JE, Schuster RL (1988) The formation and failure of natural dams. geometry inspection. Tunn Undergr Space Technol 37:199–206.
Geol Soc Am Bull 100:1054–1068 https://doi.org/10.1016/j.tust.2013.04.004
Cui P, Zhu YY, Han YS, Chen XQ, Zhuang JQ (2009) The 12 may Riquelme AJ, Tomás R, Abellán A (2016) Characterization of rock slopes
Wenchuan earthquake-induced landslide lakes: distribution and pre- through slope mass rating using 3D point clouds. Int J Rock Mech
liminary risk evaluation. Landslides 6(3):209–223. https://doi.org/ Min Sci 84:165–176. https://doi.org/10.1016/j.ijrmms.2015.12.008
10.1007/s10346-009-0160-9 Rossi G, Tanteri L, Tofani V, Vannocci P, Moretti S, Casagli N (2018)
Dai FC, Lee CF, Deng JH, Tham LG (2005) The 1786 earthquake- Multitemporal UAV surveys for landslide mapping and characteri-
triggered landslide dam and subsequent dam-break flood on the zation. Landslides 15(5):1045–1052. https://doi.org/10.1007/
Dadu river, southwestern China. Geomorphology 65:205–221. s10346-018-0978-0
https://doi.org/10.1016/j.geomorph.2004.08.011 Saito H, Uchiyama S, Hayakawa YS, Obanawa H (2018) Landslides
Donnarumma A, Revellino P, Grelle G, Guadagno FM (2013) Slope triggered by an earthquake and heavy rainfalls at Aso volcano,
angle as indicator parameter of landslide susceptibility in a geolog- Japan, detected by UAS and SfM-MVS photogrammetry. Prog
ically complex area. Landslide science and practice. Springer, pp Earth Planet Sci 5(15):1–10. https://doi.org/10.1186/s40645-018-
425–433. https://doi.org/10.1007/978-3-642-31325-7_56 0169-6
Ermini L, Casagli N (2003) Prediction of the behaviour of landslide dams Salvini R, Francioni M, Riccucci S, Bonciani F, Callegari I (2013)
using a geomorphological dimensionless index. Earth Surf Process Photogrammetry and laser scanning for analyzing slope stability
Landf 28:31–47. https://doi.org/10.1002/esp.424 and rock fall runout along the domodossola–Iselle railway, the
Gai S, Da F, Dai XQ (2018) A novel dual-camera calibration method for Italian Alps. Geomorphology 185:110–122. https://doi.org/10.
3D optical measurement. Opt Laser Technol 104:126–134. https:// 1016/j.geomorph.2012.12.020
doi.org/10.1016/j.optlaseng.2017.09.025 Sassa K, Tsuchiya S, Fukuoka H, Mikos M, Doan L (2015) Landslides:
Galland O, Bertelsen HS, Guldstrand F, Girod L, Johannessen RF, review of achievements in the second 5-year period (2009–2013).
Bjugger F, Burchardt S, Mair K (2016) Application of open- Landslides 12:213–223. https://doi.org/10.1007/s10346-015-0567-
source photogrammetric software micmac for monitoring surface 4
deformation in laboratory models. Earth Planet Sci Lett 121:2852– Schmiedel T, Galland O, Breitkreuz C (2017) Dynamics of sill and lac-
2872. https://doi.org/10.1002/2015JB012564 colith emplacement in the brittle crust: role of host rock strength and
García-Luna R, Senent S, Jurado-Piña R, Jimenez R (2019) Structure deformation mode. J Geophys Res Solid Earth 122:8625–9484.
from motion photogrammetry to characterize underground rock https://doi.org/10.1002/2017JB014468
masses: experiences from two real tunnels. Tunn Undergr Space Schmiedel T, Galland O, Haug Ø, Dumazer G, Breitkreuz C (2019)
Technol 83:262–273. https://doi.org/10.1016/j.tust.2018.09.026 Coulomb failure of earth's brittle crust controls growth, emplace-
Jiang XG, Cui P, Chen HY, Guo YY (2017) Formation conditions of ment and shapes of igneous sills, saucer-shaped sills and laccoliths.
outburst debris flow triggered by overtopped natural dam failure. Earth Planet Sci Lett 510:161–172. https://doi.org/10.1016/j.epsl.
Landslides 14:821–831. https://doi.org/10.1007/s10346-016-0751- 2019.01.011
1 Sharif YA, Elkholy M, Hanif Chaudhry M, Imran J (2015) Experimental
Kidyaeva V, Chernomorets S, Krylenko I, Wei F, Petrakov D, Su P, Yang study on the piping erosion process in earthen embankments. J
H, Xiong JN (2017) Modeling potential scenarios of the Hydraul Eng ASCE 141:04015012
Tangjiashan lake outburst and risk assessment in the downstream Shrestha BB, Nakagawa H (2016) Hazard assessment of the formation
valley. Front Earth Sci 11:579–591. https://doi.org/10.1007/s11707- and failure of the Sunkoshi landslide dam in Nepal. Nat Hazards 82:
017-0640-5 2029–2049. https://doi.org/10.1007/s11069-016-2283-3
Korup O (2005) Geomorphic hazard assessment of landslide dams in Son M, Kim M (2017) Estimation of the compressive strength of intact
South Westland, New Zealand: fundamental problems and ap- rock using non-destructive testing method based on total sound-
proaches. Geomorphology 66:167–188. https://doi.org/10.1016/j. signal energy. Geotech Test J 40(4):643–657. https://doi.org/10.
geomorph.2004.09.013 1520/GTJ20160164
Li L, Zhang X (2018) A new approach to measure soil shrinkage curve. Sturm P, Triggs B (1996) A factorization based algorithm for multi-image
Geotech Test J 42(1):1–18. https://doi.org/10.1520/GTJ20150237 projective structure and motion. In: Buxton B, Cipolla R (eds)
Liu P, Chen AY, Huang YN, Han JY, Lai JS, Kang SC, Wu TH, Wen Computer vision — ECCV '96. Springer, Berlin Heidelberg, pp
MC, Tsai M-H (2014) A review of rotorcraft unmanned aerial vehi- 709–720
cle (UAV) developments and applications in civil engineering. Tacconi Stefanelli C, Segoni S, Casagli N, Catani F (2016) Geomorphic
Smart Struct Syst 13(6):1065–1094. https://doi.org/10.12989/sss. indexing of landslide dams evolution. Eng Geol 208:1–10. https://
2014.13.6.1065 doi.org/10.1016/j.enggeo.2016.04.024
4732 H. Wu et al.

Teza G, Galgaro A, Zaltron N, Genevois R (2007) Terrestrial laser scan- Geotechnique 53:619–631. https://doi.org/10.1680/geot.2003.53.7.
ner to detect landslide displacement fields: a new approach. Int J 619
Remote Sens 28:3425–3446. https://doi.org/10.1080/ Wójcik A, Klapa P, Mitka B, Sładek J (2018) The use of the photogram-
01431160601024234 metric method for measurement of the repose angle of granular
Török Á, Barsi Á, Bögöly G, Lovas T, Somogyi Á, Görög P, Sciences ES materials. Measurement 115:19–26. https://doi.org/10.1016/j.
(2018) Slope stability and rockfall assessment of volcanic tuffs using measurement.2017.10.005
RPAS with 2-D FEM slope modelling. Nat Hazards Earth Syst Sci Wu H, Nian TK, Chen GQ, Zhao W, Li DY (2020) Laboratory-scale
18:583–597. https://doi.org/10.5194/nhess-18-583-2018 investigation of the 3-D geometry of landslide dams in a U-shaped
Tran T, Tucker-Kulesza S, Bernhardt-Barry M (2017) Determining sur- valley. Eng Geol 220:105428. https://doi.org/10.1016/j.enggeo.
face roughness in erosion testing using digital photogrammetry. 2019.105428
Geotech Test J 40:917–927. https://doi.org/10.1520/GTJ20160277 Yang Y, Cao SY, Yang KJ, Li WP (2015) Experimental study of breach
Westoby MJ, Brasington J, Glasser NF, Hambrey MJ, Reynolds JM process of landslide dams by overtopping and its initiation mecha-
(2012) ‘Structure-from-motion’ photogrammetry: a low-cost, effec- nisms. J Hydrodyn Ser B 27:872–883. https://doi.org/10.1016/
tive tool for geoscience applications. Geomorphology 179:300–314. S1001-6058(15)60550-9
https://doi.org/10.1016/j.geomorph.2012.08.021 Zhang Y, Chen G, Zheng L, Li Y, Wu J (2013) Effects of near-fault
White DJ, Take WA, Bolton MD and Munachen SE (2001) A deforma- seismic loadings on run-out of large-scale landslide: a case study.
tion measurement system for geotechnical testing based on digital Eng Geol 166:216–236. https://doi.org/10.1016/j.enggeo.2013.08.
imaging, close-range photogrammetry, and PIV image analysis. In: 002
Proceedings of the 15th international conference on soil mechanics
Zhao GW, Jiang YJ, Qiao JP, Yang ZJ, Ding PP (2018) Numerical and
and foundation engineering, Istanbul. Balkema, Rotterdam, 539–
experimental study on the formation mode of a landslide dam and its
542
influence on dam breaching. Bull Eng Geol Environ 78(4):2519–
White DJ, Take WA, Bolton MD (2003) Soil deformation measurement
2533. https://doi.org/10.1007/s10064-018-1255-0
using particle image velocimetry (PIV) and photogrammetry.

You might also like