Engineering Geology 260 (2019) 105238

Contents lists available at ScienceDirect

Engineering Geology
journal homepage: www.elsevier.com/locate/enggeo

Evaluating volume of coseismic landslide clusters by flow direction-based T

partitioning
⁎
R.L. Fan, L.M. Zhang , P. Shen
Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Hong Kong, China

A R T I C LE I N FO A B S T R A C T

Keywords: A strong earthquake can trigger thousands of co-seismic landslides. The evaluation of the volume of such co-
Landslide seismic landslides is essential for landslide risk assessment and long-term post-earthquake hazard evaluation.
Earthquake Empirical volume-area power laws are often applied to estimate the landslide volume. When several small
Mass wasting landslide deposits form a large cluster on the hillslope, blurry boundaries of the clustered co-seismic landslides
Slope stability
may lead to seemingly one large landslide and grossly overestimated landslide volume. In this paper, we propose
Risk assessment
Power law
a flow direction-based method to identify the boundaries of a cluster of landslides and to partition co-seismic
landslides. Three watersheds that were severely impacted by co-seismic landslides in the Wenchuan earthquake
region are chosen for case studies. Each co-seismic landslide cluster in these watersheds was interpreted from
images taken shortly after the earthquake, and partitioned into individual landslides in accordance with a flow
direction map that is processed based on a local digital terrain model. Finally, the volume of each individual
landslide is calculated using an empirical volume-area power law. The adopted power law scaling relationship is
selected from six widely used empirical relations and verified by 11 local landslide cases. The proposed flow-
direction based partitioning method is further verified by clear boundaries of landslides that were exposed after
storms, as well as supplementary field survey results. The proposed method results in similar landslide
boundaries and volumes as the field survey results. The uncertainty of this approach is below 19%. If no par-
titioning is conducted, the total volume of the landslide cluster can approach twice that from the field in-
vestigations.

1. Introduction The landslide volume may be estimated using field surveys, an

average thickness method, physically based modelling, multi-temporal
A strong earthquake can trigger numerous landslides. For instance, DEM analysis and empirical power laws. In the field survey method,
the number of co-seismic landslides induced by the 2008 Wenchuan direct survey or geophysical methods can be used to figure out the slip
earthquake ranged from around 56,000 (Dai et al., 2011) to 197,481 surface and the deposit depth of a single landslide deposit or numerous
(Xu et al., 2014) depending on how the landslide boundaries were landslides in a region (Cai et al., 2011; Chen et al., 2013; Jongmans and
identified; the number of co-seismic landslides triggered by the 2010 Garambois, 2007). The precise measurements of the area and volume of
Haiti earthquake exceeded 23,000 (Harp et al., 2016); and the 2018 an individual landslide also serve as a benchmark for calibrating em-
Hokkaido earthquake caused 7837 co-seismic landslides (Wang et al., pirical power law relationships. The method is however not practical
2019). These landslides led to numerous loose soil deposits on hillslopes when applied to a large area due to time and resource constraints. The
and in valleys (Fan et al., 2019), which provided abundant materials for average thickness method is often employed in shallow landslide areas
debris flows in rainy season (Fan et al., 2018b; Zhang et al., 2014). The (Fan et al., 2018a; Martin et al., 2002), or where elaborated borehole
volume of co-seismic landslide deposits is one of the essential input data are available (Zhang and Zhang, 2017). The accuracy of the
parameters for numerical simulations of post-earthquake debris flows average thickness method is limited by topography, embedded bedrock
(Chen and Zhang, 2015; Shen et al., 2018), and influences the risk as- and soil distributions. Physically based modelling approach estimates
sessment and engineering design, and the evolution pattern of post- the landslide volume by simulating the slope failure processes using soil
earthquake mass wasting (Zhang et al., 2012; Zhang et al., 2016; Zhang and rock mechanics. Many applications and scripts have been devel-
et al., 2017; Zhang and Zhang, 2017; Tang et al., 2019). oped to estimate the landside volume, e.g. Scoops3D by Reid et al.

⁎
Corresponding author.
E-mail addresses: [email protected] (R.L. Fan), [email protected] (L.M. Zhang), [email protected] (P. Shen).

https://doi.org/10.1016/j.enggeo.2019.105238
Received 10 March 2019; Received in revised form 15 July 2019; Accepted 25 July 2019
Available online 25 July 2019
0013-7952/ © 2019 Elsevier B.V. All rights reserved.
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

(2015), r.rotstab by Mergili et al. (2014) and the script by Marchesini introduced in Section 4.1. Of course, a landslide cluster cannot be
et al. (2009). However, this approach is limited when applied to a large simply partitioned equally in reality because the area of a landslide is
area as the parameter variations is hard to determine. Spatial DEM closely related to the location of the scar and the local topography.
analysis has been developed recently with the availability of high-re- However, how to partition a cluster of landslides into individual land-
solution digital topography data. It subtracts the pre- and post-event slides remains unsolved.
DEMs to find the aggradation and denudation of the topography (Tang The aim of this study is to present a flow direction-based method to
et al., 2019; Tsutsui et al., 2007). It performs well in areas with ela- partition a co-seismic landslide cluster into appropriate individual
borate and high-resolution DEMs. The empirical power law method landslides based on the flow directions of the topography, and to cal-
links the landslide volume with some indices that can be directly culate the total volume of the landslides using an empirical power law
measured, such as the relationship between angle of reach and landslide relationship. Observed clear boundaries that exposed after the erosion
volume by Corominas (1996) and the relationship between landslide of landslide dust, as well as field investigation results, are used as
area and landslide volume by Guzzetti et al. (2009). The empirical evidence to verify the proposed partitioning method. Finally, the vo-
method that correlates landslide volume with area is widely used lumes of the co-seismic landslides in three watersheds in the epicentral
(Guzzetti et al., 2009; Larsen et al., 2010; Li et al., 2014; Parker et al., area of the Wenchuan earthquake are assessed to further verify and
2011): illustrate the proposed method.

V = α·Aβ (1)
2. Proposed flow direction-based partitioning method
where V is the landslide volume; A is the landslide area; α and β are
coefficients. The development of a power law requires a dataset of In this section, the principle of the flow direction based method will
landslides with known areas and volumes. For example, Guzzetti et al. be introduced; how a landslide cluster is partitioned and how a suitable
(2009) derived an empirical relationship between landslide volume and empirical power law is selected to calculate the volume of sub-divided
landslide area from 677 global individual landslide cases; Larsen et al. landslides will be presented; and finally three special cases will be
(2010) compiled a landslide inventory of 4231 case to assess the vo- discussed. The workflow of this study is shown in Fig. 3.
lume-area scaling; Parker et al. (2011) applied the published scaling
parameters of Guzzetti et al. (2009) and Larsen et al. (2010) to estimate 2.1. Flow direction
the mass erosion in the Wenchuan earthquake area.
When applying an empirical volume-area relationship, the two Flow direction is one of the most important attributes of a mass flow
coefficients α and β should be carefully chosen as they vary significantly (Lemaître et al., 2009; Martha et al., 2010). It shows the direction of
in different geological conditions (Table 1) and triggers (earthquake flow out of each digital terrain cell on a discretized digital elevation
versus rainfall). The most sensitive variable is the landslide area. model (Jenson and Domingue, 1988; Tarboton, 1997). Flow direction
Table 1 summarises empirical power laws based on benchmark in- encoding is done in powers of two (20, 21, 22, 23 etc. representing eight
dividual landslide cases. flow directions from the central cell to eight surrounding cells) in
A landslide cluster is also called a landslide amalgamation (Marc Fig. 4(a), The binary encoding was adopted for its high computing ef-
and Hovius, 2015). It represents the landslide that has a closed outside ficiency (Jenson and Domingue, 1988). The flow direction of a cell is
boundary but has more than one source or deposition area that can be determined by the largest descent, i.e. the ratio of the elevation drop
observed from satellite images. When facing a large cluster of land- between two adjacent cells to the centre-to-centre distance. If the cell
slides, Tanyas (2019) proposed a successive-slope-failure hypothesis to size is 1, the distance between two orthogonal cells is also 1 and the
explain how the amalgamated landslides are developed and identified distance between two diagonal cells is 2 . If the maximum descent to
the failure process by interpreting multi-temporal satellite images. Li several neighbouring cells is the same, the neighbourhood should be
et al. (2014) noticed the overestimation of landslide volume in a cluster enlarged until the steepest descent is found. When the direction of the
of landslides by power laws if the landslide clusters are not subdivided. steepest descent is found, the output cell is coded with the value re-
A reduction trend of landslide volume can be found if a landslide cluster presenting that direction. If all neighbours are higher than the cell
is evenly partitioned into n smaller landslides (Fig. 1c and d). The ratio being processed, the cell is a sink and has an undefined flow direction.
of the landslide volumes before and after being partitioned into n equal This process can be performed with the Flow Direction function in the
sub-divided landslides can be calculated as (Li et al., 2014): ArcGIS platform (ESRI, 2011). Fig. 4b and c show a typical transfor-
V1/ Vn = α·(n·Ai ) β /(n·α·Ai β ) = n(β − 1) mation from a digital elevation model to a flow direction map.
(2)
The most common types of landslides triggered by the 2008
where V1 and Vn are the estimated volumes with no partitioning and Wenchuan earthquake were rock falls, rock avalanches, and rock-debris
n equal parts, respectively; Ai is the area of an averagely partitioned slides (Gorum et al., 2011). Rock avalanches and rock-debris slides have
landslide; n is the number of partitioned individual landslides in the been widely regarded as dry mass flows, whose motion is superimposed
landslide cluster. Fig. 2 shows the effect of number of partitions on the by the spreading of the granular mass that is controlled by the mass
estimated total volume: the volume will go down to approximately 76% body rheology (Campbell, 1990; Straub, 1997). Although the move-
if a landslide cluster is divided into two, and to 53% if divided into five. ments of rigid particles are far more irregular than a fluid when con-
The dots are partitioned landslides in our study which will be sidering interparticle collisions or particle-fluid interactions, the

Table 1
Parameters and application regions of six widely used empirical scaling relationships V = α·Aβ. V is the landslide volume, A is the landslide area, α and β are
coefficients.
No. logα β Region Reference

1 −0.836 ± 0.015 1.332 ± 0.005 Global landslides Larsen et al. (2010)

2 −0.73 ± 0.06 1.35 ± 0.01 Global bedrock landslides Larsen et al. (2010)
3 −0.59 ± 0.03 1.36 ± 0.01 Mixed Himalayan landslides Larsen et al. (2010)
4 −1.131 1.45 ± 0.009 Global landslides Guzzetti et al. (2009)
5 −0.974 ± 0.366 1.388 ± 0.087 Longmen Shan landslides Parker et al. (2011)
6 −0.995 ± 0.366 1.392 ± 0.087 Longmen Shan landslides Li et al. (2014)

2
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Fig. 1. (a) Landslide clusters in a Wenchuan earthquake catchment in May 2008 on a satellite image; (b) schematic of landslide volume calculation using an empirical
scaling relationship; (c) clear landslide deposit boundaries observed in 2010; (d) schematic of partitioned landslide volume calculation using an empirical scaling
relationship.

applicable environmental factor in deterministic modelling of land-

slides at large scales and in landslide runout modelling. Martha et al.
(2010) used flow direction as one of the key morphometric features to
recognize and classify landslides in the rugged Himalaya area.

2.2. Partitioning method

We first delineated the landslide cluster boundaries as polygons in

the ArcGIS platform (Figs. 5a), following the procedure proposed by
Harp et al. (2011). The landslides involved are called Landslide Cluster
(LC). These landslide clusters are then contrasted to the flow directions
at the same location (Fig. 5b). If one LC polygon involves more than one
flow direction, the landslide cluster is in two or more watersheds and
shall be divided further. The LC polygon is subsequently partitioned in
accordance with flow direction borders into several smaller landslides
that are called Partitioned Landslides by Flow Directions (PLFD)
Fig. 2. Effect of number of partitions on the estimated total volume of a (Fig. 5c). Landslide clusters smaller than four cells in the DEM with a
landslide cluster. The curve represents an idealized trend of volume ratio of resolution of 30 m, i.e. 3600 m2, are not partitioned further in the
equally partitioned landslides. The dots represent the partitioned landslides in study. If one half of a border cell is regarded uncertain, only the central
our study. V1 = volume without partitioning; Vn = volume with n partitions. area (one cell) is certain in a four-cell polygon.
Additionally, after the storms in 2009 and 2010, the boundaries of
movement of blocks would not affect the overall travel direction of the individual landslides became more clearly visible on satellite images
whole landslide. Therefore, the sliding orientation of the rock ava- (Fig. 1c). During the co-seismic landsliding process, a thin layer of dust
lanches and rock-debris slides can be represented by the flow direction covered the terrain and made the landslide deposit boundaries blurred.
that can be readily derived on a GIS platform. The motion of rock fall is The images of 2008 and 2010 were carefully compared. After the
much more complex; the fallen blocks slide with projection, long-dis- storms in 2009 and 2010, the dust layer eroded so that the boundaries
tance flying, rolling and particle crushing. Rock fall is taken as a special of the co-seismic landslide deposits were seen more clearly. Accord-
case which will be discussed in Section 2.4. ingly, the clustered landslides are partitioned using these clearer
Flow direction has been proven to be a reliable measure to evaluate boundaries and these landslides are called Partitioned Landslides by
the erosion and sediment process on a nature terrain. Jain and Kothyari Observed Boundaries (PLOB). In order to compare the influence of
(2000) identified sediment source areas and predicted the sediment clustering, the total areas of the landslide clusters (LC) and the parti-
yield from the Nagwa and Karso catchments in Bihar (India) by cal- tioned landslides by flow direction (PLFD) were kept unchanged during
culating the flow directions and the flow accumulation in their study interpretation and partitioning. The total area of the partitioned land-
area. van Westen et al. (2008) considered flow direction as a high-level slides by observed boundaries (PLOB) was slightly smaller due to the

3
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Fig. 3. The five-step workflow of this study.

erosion of the landslide dust. Totally four high resolution satellite deviate from the observed landslide volumes; while the results of power
images as well as a set of DEMs of the Wenchuan earthquake area with law 5 and power law 6 fit better as their varying ranges cover all the 11
30 m resolution are collected (Table 2 and Fig. 6). The satellite images landslide cases. These two power laws are generated using Wenchuan
were commercially available. These images and DEM are registered in earthquake landslide data while the first four are based on data from
the same spatial geo-system (GCS New Beijing) to delineate the LC, other regions. The Residual Sum of Squares (SSres) of power law 5 and
PLFD and PLOB on the ArcGIS platform. power law 6 are further compared and power law 5 by Parker et al.
(2011) is selected to calculate the landslide volume because it has a
smaller SSres.
2.3. Volume-area power laws

The total volumes of the Landslides Clusters (LC), Partitioned 2.4. Special cases
Landslides by Flow Directions (PLFD) and Partitioned Landslides by
Observed Boundaries (PLOB) are calculated using an empirical volume- Some special cases need to be clarified when applying the flow di-
area power law shown in Eq. (1). Six frequently used power law scaling rection based partitioning method. The first involves one large landslide
relationships are considered (Table 1), each of them differs in the occupying multiple flow directions. After a landslide occurs, the debris
coefficients and is more applicable to the region in which the data for moves along the failure surface and spreads on the steep terrain along
establishing the relationship is collected (Fan et al., 2018c; Li et al., the flow direction depending on the local topography. The DEM in use
2014; Parker et al., 2011). To calculate the volumes of the LC, PLFD and was a post-earthquake topography and captures the co-seismic land-
PLOB, the selected power law should be calibrated or modified using slide deposits well.
the Wenchuan earthquake area landslide data. Eleven landslide cases The second special case involves a large landslide running over the
with precise landslide area and volume information are collected from valley bottom and onto the opposite side of the valley. This problem
field investigations and the literature (Chen et al., 2013; Han et al., may cause significant underestimation of the landslide volume if we
2009; Li et al., 2009) (Table 3). These cases are separate landslides with simply partition the deposit into two parts. Hence it is necessary to
well-defined boundaries so they can be employed to calibrate the six emphasize that the flow direction-based partition method is only used
empirical power laws. Fig. 7 shows the suitability of the six relation- to partition landslide clusters with blurry boundaries. Before contrasted
ships. It can be found that the results of the first four relationships to the flow direction map, the landslides have been carefully delineated

Fig. 4. (a) Eight flow directions (E, SE, S, SW, W, NW, N and NE) of the center cell encoded by values (1, 2, 4, 8, 16, 32, 64 and 128); (b) a sample of a Digital
Elevation Model (DEM); (c) the flow direction map generated from the DEM.

4
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Fig. 5. Schematic of flow direction-based partition method: (a) the boundary of landslide cluster is delineated from satellite image; (b) the boundary of landslide
cluster is contrasted by the flow directions; (c) one landslide cluster is partitioned into three in accordance with the flow directions.

Table 2 3. Case studies and verification

Detailed information of Satellite images and Digital Elevation Model (DEM).
ID Image type Resolution (m) Shooting time Covering area
The seismic stability of a slope is strongly dependent upon the dis-
(km2) tance to the epicentre, co-seismic fault rupture, the magnitude, in-
tensity and peak ground acceleration (PGA) (Gorum et al., 2011;
1 Quick-bird 0.61 30 May 2008 180 Keefer, 1984; Xiao et al., 2016). The 2008 Wenchuan earthquake had a
2 Quick-bird 0.61 30 May 2008 100
magnitude of Mw 7.9 and a largest recorded PGA of 957.7 Gal (Li et al.,
5 Worldview-2 2.5 10 December 2010 460
6 Worldview-2 2.5 18 December 2010 135 2008). As mentioned earlier, > 56,000 co-seismic landslides were
7 ASTER GDEM 2 30 15 December 2011 1200 triggered along and around the Longmenshan Faults. Xu et al. (2014)
reported that nearly 1151 km2 (2.6%) of the Wenchuan earthquake area
was covered by landslides. Near the epicentre (Figs. 1a), almost one
from satellite images with 0.5–2.0 m resolution, in which a deposit that third of the terrain was covered by densely populated landslide deposits
covers both sides of a valley is identified as one landslide. and the boundaries of these co-seismic landslides overlap to a great
The third special case is rockfall. The flying, bouncing and projec- extent. In this section, Hongchun Watershed, Niujuan Watershed and
tion of rock blocks make it difficult to trace the rockfall path. Therefore, Yinchang Watershed, which are three severely damaged catchments
the identification of rockfall deposits in this paper still relies on image near the epicentre of the Wenchuan earthquake, are selected as study
interpretation. areas (Fig. 8). The number and volume of landslides are calculated
before and after partitioning based on flow direction. For verification,
post-earthquake images displaying clear landslide boundaries in 2010

Fig. 6. Covering areas of the collected satellite images and the DEM.

5
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Table 3
Detailed information of 11 cases of landslides in the Wenchuan earthquake area.
NO. Area (m2) Volume (m3) Location References

1 150 1500 Yinchang Watershed Field investigation by authors

2 2400 18,720 Gaojia Watershed Field investigation by authors
3 1125 11,250 Caopo town Field investigation by authors
4 5741 86,108 Wasi Watershed Field investigation by authors
5 1610 12,876 Qipan Watershed Field investigation by authors
6 128,878 165,000 Qipan Watershed Field investigation by authors
7 3000 11,200 Taoguan Watershed Field investigation by authors
8 85,000 2,000,000 Niujuan Watershed Li et al. (2009)
9 25,500 255,000 Caopo town Li et al. (2009)
10 1600 19,018 Caopo town Chen et al. (2013)
11 19,800 200,000 Liming Village Han et al. (2009)

are interpreted to calculate the landslide volume. Also, detailed field To calculate the volume of the co-seismic landslides, satellite images
investigation results in the literature are employed for comparison. and a DEM of Hongchun Watershed are processed following the five
steps in Section 2. Totally 65 landslides were detected with a total area
of 1.78 × 106 m2 (Fig. 9a). After applying the empirical scaling re-
3.1. Hongchun watershed lationship by (Parker et al., 2011), the calculated landslide volume is
13.7 × 106 m3. Subsequently, the landslide clusters are partitioned
Hongchun Watershed is located near Yingxiu Town, the epicentre of based on the local flow directions (Fig. 9b and c) and the observed
the 2008 Wenchuan earthquake. The watershed mouth (N31°03′56″, boundaries in 2010 (Fig. 9d). The 65 landslides are further divided into
E103°29′28″) is only 200 m away from the local residential area 103 PLFDs and 116 PLOBs using these two methods, resulting in re-
(Fig. 8c). The total area of the watershed is 5.87 km2, with elevations duced landslide volumes of 10.78 × 106 m3 and 9.74 × 106 m3, re-
ranging from 2144 to 878 m and an elevation difference of 1266 m. On spectively (Table 4), which are closer to the results of field investiga-
14 August 2010, a 162.1 mm rainstorm swept the vicinity, a severe tions reported by Domenech et al. (2019), i.e. 9.3 × 106 m3 (Fig. 10).
debris flow was triggered in the watershed, which blocked the Minjiang
River and affected the newly reconstructed Yingxiu town. In this event,
around 40 × 104 m3 of debris material flew into the Minjiang River, 3.2. Niujuan watershed
and a large debris dam of 15 × 104 m3 was formed. Totally
55 × 104 m3 of loose materials ran out of the valley (Xu et al., 2012). Niujuan Watershed is in the region where the Wenchuan earthquake

Fig. 7. Calibration of six empirical scaling relationships by 11 landslide cases with measured area and volume in the Wenchuan earthquake area. The red dots
represent the landslide cases with measured area and volume in the Wenchuan earthquake area. The circles represent the distribution of the calculated landslide
volumes of the 11 landslide cases given the same area for each landslide. The power law coefficients in varying ranges are listed in Table 1. SSres is the residual sum of
squares of the collected landslide volume and the corresponding volume by power law. Smaller SSres means better fitting. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)

6
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Fig. 8. Satellite images of the co-seismic landslides in three watersheds near the epicentre of the 2008 Wenchuan earthquake: (a) Yinchang Watershed; (b) locations
of the three watersheds; (c) Hongchun Watershed and Niujuan Watershed.

Fig. 9. Partitioning of landslides in Hongchun Watershed: (a) landslide clusters; (b) flow directions; (c) partitioned landslides by flow directions; (d) partitioned
landslides by observed boundaries in 2010.

7
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Table 4
Statistics of landslide clusters (LC), partitioned landslides by flow direction (PLFD) and partitioned landslides by observed boundaries (PLOB).
Watersheds Landslide area (106 m2) LC PLFD PLOB

6 3 6 3
Number of landslides Volume (10 m ) Number of landslides Volume (10 m ) Number of landslides Volume (106 m3)

Hongchun Watershed 1.78 65 13.8 103 10.86 116 9.83

Niujuan Watershed 4.32 116 38.43 444 20.96 451 20.01
Yinchang Watershed 3.61 22 50.38 172 21.91 177 22.60

Hongchun Watershed landslide volume becomes 22.60 × 106 m3 (Table 4 and Fig. 12d). The
partitioned landslide volume is similar to the detailed field measuring
Niujuan Watershed
result, i.e., 18.31 × 106 m3 (Fig. 10).
Yinchang Watershed
Landslide volume (106 m3)

60

50
4. Analysis of volume estimation error

40
In this section, the distribution patterns of the LC, PLOB and PLFD
30 are compared to examine the performance of the flow direction-based
20 partitioning method. The uncertainties of landslide area that arise from
the interpretation of satellite images and DEMs are analysed.
10

0
LC PLFD PLOB FS
4.1. Comparison of partitioning methods
Fig. 10. Statistics of the total volumes of landslide clusters (LC), partitioned
landslides by flow direction (PLFD), partitioned landslides by observed To compare the difference before and after partitioning based on the
boundaries (PLOB) in 2010 and field survey (FS) in Hongchun Watershed, flow direction, area-frequency curves of landslides are used to show the
Niujuan Watershed and Yinchang Watershed. distribution of landslide clusters and partitioned landslides. A landslide
area frequency relationship is a tool to represent the distribution of
epicentre (E 103°42′, N 31°01′) is located (Fig. 8c). The total area of the landslides of different scales (Dai and Lee, 2002). After the Wenchuan
watershed is 10.92 km2, with elevations between 2679 m and 853 m. earthquake, several co-seismic landslide inventories were generated by
Loose deposits were widely distributed in this watershed. Hence over manual or automatic processes, or combination of both (Fan et al.,
ten debris flows occurred in this watershed since the earthquake, 2019; Gorum et al., 2011; Huang and Li, 2009; Xu et al., 2014). Al-
among which the largest two were the 26 Sept. 2008 debris flow and though both covering range and smallest landslides were considered,
the 14 Aug. 2010 debris flow, with runout debris volumes of the boundary sheltering effect of landslide clusters was seldom con-
10.5 × 104 m3 and 8.27 × 104 m3, respectively (Liu et al., 2017). cerned, which is vital since landslide area is the most sensitive input to
After going through the procedure proposed in Section 2, a total of calculate the landslide volume from remote sensing images. Fig. 13
116 landslides were delineated in the watershed with a total area of shows the area-frequency relationships of the landslides in Hongchun
4.32 × 106 m2 (Fig. 11a). The calculated volume of landslide clusters Watershed, Niujuan Watershed and Yinchang Watershed. In Hongchun
reaches 38.28 × 106 m3 (Fig. 10). The landslide clusters are partitioned Watershed, 55.4% of the original 65 landslides are larger than
based on the local flow directions (Fig. 11b and c) and the observed 10,000 m2; after partitioned into 100 individual landslides based on the
boundaries in 2010 (Fig. 11d). The 116 landslides were thereafter se- flow direction (PLFDs) and 116 based on the observed boundaries
parated into 444 PLFDs and 451 PLOBs by these two methods, produ- (PLOBs), this proportion drops to 47.6% and 50.4%, respectively. In
cing total landslide volumes of 20.76 × 106 m3 and 19.81 × 106 m3, Niujuan Watershed, the 116 LC are separated into 444 PLFDs and 451
respectively (Table 4). PLOBs, respectively. The landslides larger than 10,000 m2 consist of
64.7% of the original clustered landslides but drop to 27.3% and 32.2%,
3.3. Yinchang watershed respectively, after partitioning by flow directions and observed
boundaries. In Yinchang Watershed, all the original 22 LCs are larger
Yinchang Watershed is a typical debris flow watershed in the than 10,000 m2. After separated into 172 PLFDs and 177 PLOBs, the
earthquake zone, which is 8.3 km from the Yingxiu Town on the left proportions of landslides larger than 10,000 m2 drop to 68.0% and
bank of the Yuzixi River (E 103°23′05″, N 31°03′50″) (Fig. 8a). The total 52.0%, respectively. In conclusion, before partitioning, the areas of
area of the watershed is 7.36 km2, with a local relief of 2483 m between landslide clusters are large, and the landslide number is relatively
(3701 m and 1218 m). Several debris flows of different scales occurred small. While after partitioning, most of the extremely large landslides
in this watershed since the Wenchuan earthquake. Among them the are partitioned into smaller ones. Hence the areas of the individual
largest debris flow occurred on 14 August 2010. During 12 August 2010 landslides decrease.
to 14 August 2010 a storm swept this area with a cumulative rainfall of The proposed flow direction based partitioning method performs
220 mm, and triggered channelized and hillslope debris flows which well for the three case study watersheds. The total number of PLFD is
subsequently buried the Province Road 303. The total run out volume almost the same as that of PLOB. The partitioned individual landslide
of this debris flow is estimated to be 2.11 × 106 m3 (Cai et al., 2011). volumes agree quite well with the results from both the observed
The same calculating procedure was applied to the Yinchang boundary method and the field surveys (Fig. 10). In terms of total
Watershed. Before partitioning, the total landslide cluster number, area landslide volume, the result of the flow direction method is close to that
and volume are 22, 3.61 × 106 m2 and 50.42 × 106 m3, respectively from the observed boundary method and the field survey result from
(Fig. 12a). After partitioning based on flow direction (Fig. 12b and c), the literature. It is worth mentioning that uniform partitioning always
the landslide number increases to 172 but the total volume decreases to underestimates the total landslide volume. As shown in Fig. 2, the total
21.91 × 106 m3. After partitioning by the observed boundaries revealed volume of the uniformly partitioned landslide deposits is smaller than
in 2010, the landslide number further increases to 177 and the total that from the flow direction method.

8
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Fig. 11. Partitioning of landslides in Niujuan Watershed: (a) landslide clusters; (b) flow directions; (c) partitioned landslides by flow directions; (d) partitioned
landslides by observed boundaries in 2010.

4.2. Uncertainty analysis 16 × 0.612 = 6.0 m2) on the 2008 satellite image. Therefore, in this
study, the interpretation results of the 2008 clustered landslides are
The resolution of the 2008 Quick-bird satellite images is 0.61 m, regarded as reliable. Similarly, the resolution of the 2010 satellite
which means each grid cell represents a smallest cell of 0.61 × 0.61 m2 image is 2.5 m, and the minimum TZU threshold is 100 m2 (i.e.
of the horizontal projection of the ground. As recommended by the IFFI 16 × 2.52 = 100 m2) hence the interpreted smallest landslide area, 617
Project (APAT and Trigila, 2007), Calvello et al. (2013) and Fell et al. m2, also satisfies the accuracy requirement.
(2008), the minimum dimension of a terrain zoning unit (TZU) is set to The main uncertainty comes from the flow directions which are
be 16 elementary pixels to identify the ground object, regardless of the derived from the DEM with a resolution of 30 m. Taking the outside
scale. In our study, the minimum interpreted landslide area is 617 m2, half-cell area of a landslide as an uncertain area, the uncertainty can be
which is much larger than the threshold of the 16 pixels area (i.e. calculated as:

Fig. 12. Partitioning of landslides in Yinchang Watershed: (a) landslide clusters; (b) flow directions; (c) partitioned landslides by flow directions; (d) partitioned
landslides by observed boundaries in 2010.

9
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

Fig. 14. Uncertainty of interpreted landslide area on a 30 m resolution DEM.

U = (n − 1)/ n2 (4)

Fig. 14 shows the decreasing area uncertainty with the increasing

landslide area. To further decrease the uncertainty, landslides smaller
than four grid cells, i.e. 3600 m2 will not be partitioned further by the
flow direction method, which controls the uncertainty of partitioned
landslides area below 25%. To verify whether the results are accep-
table, we calculated the total volume of the landslides smaller than
14,400 m2 (i.e. 16 × 302 m2), which is a terrain zoning unit (TZU) on
the DEM. According to Fig. 14, the uncertainty of landslides with an
area of 14,400 m2 is 18.75%. In the three gullies, landslides smaller
than 14,400 m2 only share 10.11%, 22.41% and 12.26% of the total
landslide volume, respectively, which means the landslides with area
uncertainties larger than 18.75% only account for a small portion of the
total landslide volume. The total area uncertainty can be controlled
well below 18.75%.

5. Conclusions

In this paper, we propose a flow direction approach to partition

clustered co-seismic landslides, and use 11 local landslide cases to select
an applicable empirical power law relationship to calculate the land-
slide volumes in three watersheds in the Wenchuan earthquake area.
For verification, satellite images showing clear landslide boundaries in
2010 are adopted to partition the landslide clusters in the three wa-
tersheds, and field survey results are used to further confirm the ana-
lysis results. The main conclusions can be summarized as follows:

(1) The volume of landslide clusters can be grossly overestimated by

empirical volume-area power laws if a cluster of landslides are in-
terpreted as one large landslide.
(2) As an alternative, the flow direction approach can be used to par-
tition landslide clusters soon after an earthquake. Post-earthquake
satellite images showing clear landslide boundaries, when avail-
Fig. 13. Relationship between landslide area and cumulative landslide number
able, can be used to verify the results from the proposed flow di-
in (a) Hongchun Watershed, (b) Niujuan Watershed and (c) Yinchang
rection approach.
Watershed. LC, PLFD and PLOB represent the landslide clusters, partitioned
landslides by flow direction and partitioned landslides by observed boundaries (3) Flow direction is confirmed to be a reliable parameter for parti-
in 2010, respectively. tioning clustered landslides. The volume of partitioned landslides
by the flow direction partition method is comparable to the results
from the observed boundaries and the field surveys.
U = (2n − 2)/ n2 (3)

where U is the uncertainty of interpreted landslide area and n is the Acknowledgments

number of grid columns and rows. However, the outside boundaries of
the landslide clusters have already been delineated from the 2008 sa- The authors would like to thank the Research Grants Council of the
tellite images, whose area uncertainty is negligible as discussed before. Hong Kong SAR (Nos. C6012-15G and 16206217), Sichuan Highway
The flow direction is only used to delineate the partitioning boundaries Planning, Survey, Design and Research Institute, and the colleagues
inside the landslide clusters. The uncertainty is hence halved: who jointed our annual field investigations in the past 10 years.

10
R.L. Fan, et al. Engineering Geology 260 (2019) 105238

References Li, X., Zhou, Z., Yu, H., Wen, R., Lu, D., Huang, M., Zhou, Y., Cu, J., 2008. Strong motion
observations and recordings from the great Wenchuan Earthquake. Earthquake Eng.
Eng. Vib. 7 (3), 235.
APAT, Trigila, A., 2007. Rapporto sulle frane in Italia: il Progetto IFFI: metodologia, ri- Li, X., Kong, J., Deng, H., Cui, Y., Tian, S.J., A, F.Y., 2009. Analysis on characteristics and
sultati e rapporti regionali. Report 78/2007. (ISBN: 978-88-448-0310-0). deformation failure mode of large-scale landslides induced by 5.12 Wenchuan
Cai, J., Liu, H., Xu, J., 2011. Yinchang Gully debris flow initiation mechanism analysis, earthquake. J. Sichuan Univ. 41 (3), 72–77.
prediction and evaluation study. Gansu Water Conservancy Hydropower Technol. 47 Li, G., West, A.J., Densmore, A.L., Jin, Z.D., Parker, R.N., Hilton, R.G., 2014. Seismic
(08), 26–28 (in Chinese). mountain building: landslides associated with the 2008 Wenchuan earthquake in the
Calvello, M., Cascini, L., Mastroianni, S., 2013. Landslide zoning over large areas from a context of a generalized model for earthquake volume balance. Geochem. Geophys.
sample inventory by means of scale-dependent terrain units. Geomorphology 182, Geosyst. 15 (4), 833–844. https://doi.org/10.1002/2013gc005067.
33–48. https://doi.org/10.1016/j.geomorph.2012.10.026. Liu, X., Tang, C., Fang, Q., Chang, M., 2017. Hazard assessment of dam-breaking debris
Campbell, C.S., 1990. Rapid granular flows. Annu. Rev. Fluid Mech. 22 (1), 57–90. flow based on numerical simulation. J. Sediment. Res. 42 (06), 28–34.
Chen, H., Zhang, L., 2015. EDDA 1.0: integrated simulation of debris flow erosion, de- Marc, O., Hovius, N., 2015. Amalgamation in landslide maps: effects and automatic de-
position and property changes. Geosci. Model. Dev. 8 (3), 829–844. tection. Nat. Hazards Earth Syst. Sci. 15 (4), 723–733.
Chen, Z., Lei, T., Yan, Q., Hu, H., Xiong, M., Li, Z., 2013. Measuring and calculation Marchesini, I., Cencetti, C., De Rosa, P., 2009. A preliminary method for the evaluation of
methods for landslide volume with 3-D laser scanner in Wenchuan earthquake area. the landslides volume at a regional scale. Geoinformatica 13 (3), 277–289.
Trans. Chin. Soc. Agric. Eng. 29 (8), 135–144. Martha, T.R., Kerle, N., Jetten, V., van Westen, C.J., Kumar, K.V., 2010. Characterising
Corominas, J., 1996. The angle of reach as a mobility index for small and large landslides. spectral, spatial and morphometric properties of landslides for semi-automatic de-
Can. Geotech. J. 33 (2), 260–271. tection using object-oriented methods. Geomorphology 116 (1–2), 24–36.
Dai, F., Lee, C., 2002. Landslide characteristics and slope instability modeling using GIS, Martin, Y., Rood, K., Schwab, J.W., Church, M., 2002. Sediment transfer by shallow
Lantau Island, Hong Kong. Geomorphology 42 (3–4), 213–228. landsliding in the Queen Charlotte Islands, British Columbia. Can. J. Earth Sci. 39 (2),
Dai, F., Xu, C., Yao, X., Xu, L., Tu, X., Gong, Q., 2011. Spatial distribution of landslides 189–205.
triggered by the 2008 Ms 8.0 Wenchuan earthquake, China. J. Asian. Earth. Sci. 40 Mergili, M., Marchesini, I., Rossi, M., Guzzetti, F., Fellin, W., 2014. Spatially distributed
(4), 883–895. three-dimensional slope stability modelling in a raster GIS. Geomorphology 206,
Domenech, G., Fan, X.M., Scaringi, G., van Asch, T.W.J., Xu, Q., Huang, R.Q., Hales, T.C., 178–195.
2019. Modelling the role of material depletion, grain coarsening and revegetation in Parker, R.N., Densmore, A.L., Rosser, N.J., de Michele, M., Li, Y., Huang, R.Q., Whadcoat,
debris flow occurrences after the 2008 Wenchuan earthquake. Eng. Geol. 250, 34–44. S., Petley, D.N., 2011. Mass wasting triggered by the 2008 Wenchuan earthquake is
https://doi.org/10.1016/j.enggeo.2019.01.010. greater than orogenic growth. Nat. Geosci. 4 (7), 449–452. https://doi.org/10.1038/
ESRI, 2011. ArcGIS Desktop: Release 10. Environmental Systems Research Institute, ngeo1154.
Redlands, CA. Reid, M.E., Christian, S.B., Brien, D.L., Henderson, S., 2015. Scoops3D–Software to
Fan, R.L., Zhang, L.M., Wang, H.J., Fan, X.M., 2018a. Evolution of debris flow activities in Analyze Three-Dimensional Slope Stability Throughout a Digital Landscape. Menlo
Gaojiagou Ravine during 2008–2016 after the Wenchuan earthquake. Eng. Geol. 235, Park, CA, USA, US Geological Survey Volcano Science Center.
1–10. Shen, P., Zhang, L., Chen, H., Fan, R., 2018. EDDA 2.0: integrated simulation of debris
Fan, X.M., Domènech, G., Scaringi, G., Huang, R., Xu, Q., Hales, T.C., Dai, L., Yang, Q., flow initiation and dynamics considering two initiation mechanisms. Geosci. Model.
Francis, O., 2018b. Spatio-temporal evolution of mass wasting after the 2008 M w 7.9 Dev. 11 (7), 2841–2856.
Wenchuan earthquake revealed by a detailed multi-temporal inventory. Landslides Straub, S., 1997. Predictability of long runout landslide motion: implications from
15 (12), 2325–2341. granular flow mechanics. Geol. Rundsch. 86 (2), 415–425.
Fan, X.M., Scaringi, G., Xu, Q., Zhan, W., Dai, L., Li, Y., Pei, X., Yang, Q., Huang, R., Tang, C., Tanyas, H., van Westen, C.J., Tang, C., Fan, X., Jetten, V.G., 2019. Analysing
2018c. Coseismic landslides triggered by the 8th August 2017 M s 7.0 Jiuzhaigou post-earthquake mass movement volume dynamics with multi-source DEMs. Eng.
earthquake (Sichuan, China): factors controlling their spatial distribution and im- Geol. 248, 89–101.
plications for the seismogenic blind fault identification. Landslide 15 (5), 967–983. Tanyas, H., 2019. Rapid Assessment of Earthquake Induced Landslides. University of
Fan, X.M., Scaringi, G., Domènech, G., Yang, F., Guo, X., Dai, L., He, C., Xu, Q., Huang, R., Twente.
2019. Two multi-temporal datasets that track the enhanced landsliding after the 2008 Tarboton, D.G., 1997. A new method for the determination of flow directions and upslope
Wenchuan earthquake. Earth Syst. Sci. Data 11 (1), 35–55. areas in grid digital elevation models. Water Resour. Res. 33 (2), 309–319.
Fell, R., Corominas, J., Bonnard, C., Cascini, L., Leroi, E., Savage, W.Z., 2008. Guidelines Tsutsui, K., Rokugawa, S., Nakagawa, H., Miyazaki, S., Cheng, C.-T., Shiraishi, T., Yang,
for landslide susceptibility, hazard and risk zoning for land-use planning. Eng. Geol. S.-D., 2007. Detection and volume estimation of large-scale landslides based on
102 (3–4), 99–111. elevation-change analysis using DEMs extracted from high-resolution satellite stereo
Gorum, T., Fan, X.M., van Westen, C.J., Huang, R.Q., Xu, Q., Tang, C., Wang, G.H., 2011. imagery. IEEE Trans. Geosci. Remote Sens. 45 (6), 1681–1696.
Distribution pattern of earthquake-induced landslides triggered by the 12 May 2008 van Westen, C.J., Castellanos, E., Kuriakose, S.L., 2008. Spatial data for landslide sus-
Wenchuan earthquake. Geomorphology 133 (3–4), 152–167. https://doi.org/10. ceptibility, hazard, and vulnerability assessment: an overview. Eng. Geol. 102 (3–4),
1016/j.geomorph.2010.12.030. 112–131. https://doi.org/10.1016/j.enggeo.2008.03.010.
Guzzetti, F., Ardizzone, F., Cardinali, M., Rossi, M., Valigi, D., 2009. Landslide volumes Wang, F., Fan, X., Yunus, A.P., Subramanian, S.S., Alonso-Rodriguez, A., Dai, L., Xu, Q.,
and landslide mobilization rates in Umbria, Central Italy. Earth Planet. Sci. Lett. 279 Huang, R., 2019. Coseismic landslides triggered by the 2018 Hokkaido, Japan (M w
(3–4), 222–229. 6.6), earthquake: spatial distribution, controlling factors, and possible failure me-
Han, J., Wu, S., He, S., Sun, W., Zhang, C., Wang, T., Yang, J., Si, J., 2009. Basal char- chanism. Earth Syst. Sci. Data 1–16.
acteristics and formation mechanisms of geological hazards triggered by the May 12, Xiao, J., Gong, W., Martin II, J.R., Shen, M., Luo, Z., 2016. Probabilistic seismic stability
2008 Wenchuan earthquake with a moment magnitude of 8.0. Earth Sci. Front. 16 analysis of slope at a given site in a specified exposure time. Eng. Geol. 212, 53–62.
(3), 306–326. Xu, Q., Zhang, S., Li, W.L., van Asch, T.W.J., 2012. The 13 August 2010 catastrophic
Harp, E.L., Keefer, D.K., Sato, H.P., Yagi, H., 2011. Landslide inventories: the essential debris flows after the 2008 Wenchuan earthquake. China. Nat. Hazards Earth Syst.
part of seismic landslide hazard analyses. Eng. Geol. 122 (1–2), 9–21. https://doi. Sci. 12 (1), 201–216. https://doi.org/10.5194/nhess-12-201-2012.
org/10.1016/j.enggeo.2010.06.013. Xu, C., Xu, X., Yao, X., Dai, F., 2014. Three (nearly) complete inventories of landslides
Harp, E.L., Jibson, R.W., Schmitt, R.G., 2016. Map of Landslides Triggered by the January triggered by the May 12, 2008 Wenchuan Mw 7.9 earthquake of China and their
12, 2010. Haiti earthquake. World Wide Web Address. http://pubs.er.usgs.gov/ spatial distribution statistical analysis. Landslides 11 (3), 441–461.
publication/sim3353. Zhang, S., Zhang, L.M., 2017. Impact of the 2008 Wenchuan earthquake in China on
Huang, R.Q., Li, W.L., 2009. Analysis of the geo-hazards triggered by the 12 May 2008 subsequent long-term debris flow activities in the epicentral area. Geomorphology
Wenchuan Earthquake. China. Bull. Eng. Geol. Environ. 68 (3), 363–371. https://doi. 276, 86–103. https://doi.org/10.1016/j.geomorph.2016.10.009.
org/10.1007/s10064-009-0207-0. Zhang, S., Zhang, L.M., Peng, M., Zhang, L.L., Zhao, H.F., Chen, H.X., 2012. Assessment of
Jain, M.K., Kothyari, U.C., 2000. Estimation of soil erosion and sediment yield using GIS. risks of loose landslide deposits formed by the 2008 Wenchuan earthquake. Nat.
Hydrol. Sci. J. 45 (5), 771–786. https://doi.org/10.1080/02626660009492376. Hazards Earth Syst. Sci. 12 (5), 1381–1392.
Jenson, S.K., Domingue, J.O., 1988. Extracting topographic structure from digital ele- Zhang, S., Zhang, L.M., Glade, T., 2014. Characteristics of earthquake- and rain-induced
vation data for geographic information system analysis. Photogramm. Eng. Remote. landslides near the epicenter of Wenchuan earthquake. Eng. Geol. 175, 58–73.
Sens. 54 (11), 1593–1600. https://doi.org/10.1016/j.enggeo.2014.03.012.
Jongmans, D., Garambois, S., 2007. Geophysical investigation of landslides: a review. Zhang, S., Zhang, L.M., Lacasse, S., Nadim, F., 2016. Evolution of mass movements near
Bull. Soc. Géol. Fr. 178 (2), 101–112. epicentre of Wenchuan Earthquake, the first eight years. Sci. Rep. 6, 36154.
Keefer, D.K., 1984. Landslides caused by earthquakes. Geol. Soc. Am. Bull. 95 (4), Zhang, L.M., Gao, L., Zhou, S.Y., Cheung, R.W., Lacasse, S., 2017. Stress testing frame-
406–421. https://doi.org/10.1130/0016-7606(1984)95<406:Lcbe>2.0.Co;2. work for managing landslide risks under extreme storms. In: Proc. 4th World
Larsen, I.J., Montgomery, D.R., Korup, O., 2010. Landslide erosion controlled by hillslope Landslide Forum (WLF4), Ljubljana, Slovenia, 29 May - 2 Jun 2017. Springer
material. Nat. Geosci. 3 (4), 247–251. https://doi.org/10.1038/ngeo776. International Publishing AG, pp. 17–32.
Lemaître, A., Roux, J.-N., Chevoir, F., 2009. What do dry granular flows tell us about
dense non-Brownian suspension rheology? Rheol. Acta 48 (8), 925–942.

11