Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: An accurate landslide-susceptibility assessment is fundamental for preventing landslides and minimizing
Received 12 May 2016 damage. In this study, a new time-variant slope-stability (TiVaSS) model for landslide prediction is
Received in revised form developed. A three-dimensional (3D) subsurface ow model is coupled with the innite slope-stability
9 August 2016
model to consider the effect of horizontal water movement in the subsurface. A 3D Richards' equation
Accepted 15 August 2016
is solved numerically for the subsurface ow. To overcome the massive computational requirements of
the 3D subsurface ow module, partially implicit temporal discretization and the simplication of rst-
order spatial discretization are proposed and applied in TiVaSS. A graphical user interface and two-
Keywords:
Landslide
dimensional data visualization are supported in TiVaSS. The model is applied to a 2011 Mt. Umyeon
Slope stability landslide in the Republic of Korea, and its overall performance is satisfactory.
3D subsurface unsaturated ow 2016 Elsevier Ltd. All rights reserved.
Numerical model
Mt. Umyeon
1. Introduction
Software/data availability
Landslides are common natural hazards in mountainous areas,
Name of software TiVaSS causing signicant damage and casualties worldwide every year. A
Developer's email address hyunuk@cnu.ac.kr high rainfall intensity or frequency can trigger landslides. An ac-
Tel 82-42-821-5797 curate landslide prediction is fundamental for preventing such di-
Fax 82-42-821-5791 sasters and minimizing casualties and property damage (Liao et al.,
Year rst available It will be open in public within this year 2011; Borga et al., 2002; PapathomaeKo hle et al., 2015; Pradhan
Software required windows OS, windows 7 or later and Lee, 2010; Ciabatta et al., 2016). Shallow landslide modeling
Program language C falls into three categories: empirical (Sirangelo et al., 2003; Aleotti,
Program size 16.3 MB zip le 2004; Guzzetti et al., 2007, 2008; Bezak et al., 2016), statistical
Download link https://sites.google.com/site/cnuaehelab/ (Dickson and Perry, 2016; Guzzetti et al., 1999, 2005, 2006; Pham
software/tivass/TiVaSS_release.zip? et al., 2016; Soeters and van Westen, 1996), and physically based
attredirects0&d1 User manual & tutorial models. Among these, physically based models are preferred
is now in preparation. because of their accuracy and ability to forecast the spatial and
temporal occurrence of landslides (Raia et al., 2014). They describe
the triggering processes of shallow landslides and provide spatially
variable slope-stability information, often as a safety factor (FS),
which is an index expressing the ratio between the local resisting
and driving force. The innite slope-stability model and simple
inltration-runoff process are commonly combined for computing
* Corresponding author. the FS in physically based slope-stability models. Those models are
E-mail address: jknoh3109@hanmail.net (J. Noh).
http://dx.doi.org/10.1016/j.envsoft.2016.08.009
1364-8152/ 2016 Elsevier Ltd. All rights reserved.
H. An et al. / Environmental Modelling & Software 85 (2016) 172e183 173
widely used and packaged as computer programs such as SINMAP 2. Model description
(Stability Index MAPping), SHALSTAB (Shallow Landsliding Stability
Model), TRIGRS (Transient Rainfall Inltration and Grid-Based 2.1. Innite slope-stability model
Regional Slope stability), and GEOtop-FS.
SINMAP (Pack et al., 1999) and SHALSTAB (Dietrich and Fig. 1 shows a schematic of an innite slope-stability model.
Montgomery, 1998; Montgomery and Dietrich, 1994) consider a Assuming an innite slope and failure parallel to the slope surface
simple steady-state hydrological process under constant rainfall. yields the following shear and normal stress:
This means that both models are useful for producing a spatially
distributed slope stability but are limited to the temporal predic- T W
t cos f sin f; (1)
tion of the slope stability because of the steady-state description of b=cos f b
hydrological uxes. On the other hand, TRIGRS (Baum et al., 2008;
Alvioli and Baum, 2016) computes the transient pore-water pres- P W
sure using the solution of the one-dimensional (1D) Richards s cos2 f; (2)
b=cos f b
equation with the assumption of a simple exponential soil-water
retention relationship (Gardner, 1958) and can produce a time- where W gsDb is the soil weight (kg/m), T is the shear force (kg/
variant distribution under time-variant rainfall conditions. Thus, m), P is the normal force (kg/m), D is the soil depth (m), b is the
this model can forecast the timing and distribution of shallow slope width (m), 4 is the slope angle (rad), and gs is the unit weight
landslides (Baum et al., 2010) and has been widely applied in the of the soil (kg/m3). According to MohreCoulomb theory, the shear
past decade (Baum et al., 2010; Liao et al., 2011; Liu and Wu, 2008; strength of an innite slope is
Luan et al., 2010; Montrasio et al., 2011; Saadatkhah et al., 2014;
Peres and Cancelliere, 2016). In recent years, the predictive accu- S c s0 tan 4; (3)
racy and computational efciency of TRIGRS have been improved
0
via the addition of a probabilistic approach (Raia et al., 2014) and where c is the cohesion (kg/m2), s is the effective stress (kg/m2),
0
the parallelization of the code (Alvioli and Baum, 2016). GEOtop-FS which was expressed as s su by Terzaghi (1943), u is the pore
2
(Simoni et al., 2008) solves the three-dimensional (3D) subsurface water pressure (kg/m ), and 4 is the angle of internal friction (rad).
ow numerically on the basis of the 3D Richards equation and Then, the FS is dened as
considers the physical hydrological process for estimating the slope
stability. GEOtop is also used for predicting soil erosion and depo- S
FS : (4)
sition (Zi et al., 2016). t
In the context of landslide early-warning systems, practical Substituting Eqs. (1)e(3) into Eq. (4) yields the following
physically based slope-stability models are essential for deter- equation:
mining the inuence of rainfall recharge on the subsurface hydro-
logical behavior and soil mechanics in triggering landslides. That is,
c gs D cos2 f u tan 4
advanced slope-stability tools should account for the time-varying FS : (5)
gs D sin f cos f
subsurface water ow during a landslide event and consequently
provide accurate spatio-temporal information about the slope Substituting u gwj into Eq. (5) gives
failure (Apip et al., 2010).
This study aims to develop a new time-variant slope stability c Dgs cos2 f gw j tan 4
FS ; (6)
(TiVaSS) model, which combines a 3D subsurface ow model and gs D sin f cos f
an innite slope stability model. An analytical solution is not
available for the 3D subsurface ow, and a numerical technique is where gw is the unit weight of water (kg/m3), and j is the pressure
used to solve the problem. A numerical solver establishes the head of the subsurface water (m). Eq. (6) is rewritten by separating
general soil-water retention relationships between the water con- the time-variant term and steady terms as (Iverson, 2000)
tent and the pore-water pressure such as van Genuchten model
(1980) or Brooks-Corey model (1964). 3D subsurface ow usually jgw tan 4 tan 4 c
FS ; (7)
requires a large amount of computational resources because fully gs D sin f cos f tan f gs D sin f cos f
implicit temporal discretization is essential for the numerical sta-
Eq. (7) is valid for saturated soil. However, Terzaghi's effective
bility owing to the high nonlinearity of the Richards equation. The
stress is unsatisfactory for unsaturated soil (Lu and Likos, 2006; Lu
large computational requirement can hinder the practical appli-
cation of landslide-occurrence assessment with the 3D subsurface
ow. To reduce this requirement, the 3D subsurface ow is solved
in a partially implicit manner in TiVaSS, with the assumption that
the horizontal ow is far slower than the vertical ow. The vertical
ow is solved implicitly, and the horizontal ow is solved explicitly
with a low-order discretization. Further, TiVaSS supports a graph-
ical user interface (GUI). The developed model is applied to the
2011 Mt. Umyeon landslides in Seoul, Korea for the evaluation of its
performance and functions.
This paper is organized as follows. The innite slope-stability
model and 3D subsurface ow model, including the numerical
scheme used in TiVaSS, are described in the next section. Details
about the Mt. Umyeon landslide, including the data set, are
provided in Section 3. The application results and discussion are
also presented in Section 3. We draw our conclusions in Section
4. Fig. 1. Schematic of the innite slope-stability model.
174 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183
et al., 2013). If the soil is highly dried, the large negative pressure
head causes an excessive suction force with a Terzaghi's effective
0
stress of s su. Lu and Likos (2006) generalized the effective
stress as follows:
s0 s ua ss ; (8)
ss ua uw Se : (9)
Here, uw is the water pressure, and Se is the effective saturation,
which is dened as
q qr
Se ; (10)
qs qr
where q is the volumetric moisture content, qs is the saturated
moisture content, and qr is the residual moisture content. Assuming
0
ua 0 and uw jgw yields an effective stress of s s jSegw.
Therefore, Eq. (7) is generalized for variably saturated soil as
Eq. (17) results in multiple tridiagonal matrices along the x- and discretization are applied for Eq. (22). The 1D inltration model is
y-directions, and each matrix system can be solved in separate only valid for uniform soil properties and the initial pore-water
manner. Therefore, the resulting system consumes computational pressure without a horizontal ow. If soil properties and the
resources on the order of N. initial pore-water pressure are uniform and the horizontal ow can
In this model, van Genuchten's [1980] soil-water retention curve be ignored, the pore-water pressure becomes uniform throughout
and Mualem's [1976] unsaturated hydraulic conductivity function the domain. This means that we are not required to compute the
are used, as follows: subsurface ow for all domain cells; the pore-water pressure
computed in one cell can be applied to all the study areas.
( )11=nv
q qr 1 TiVaSS does not account for surface ow routing and excess
Se ; (18) surface water, which is removed from the ow domain because the
qs qr 1 ajjjnv
downstream is hardly ever unsaturated while the upstream is
saturated. Only in this situation can the surface water affect the
2
1=2 n =n 1 11=nv subsurface ow and the FS.
K Ks Se 1 1 Se v v ; (19)
The current version of TiVaSS runs on a Windows operating
system. Fig. 3 shows the main windows of the program. Users can
where a and nv are van Genuchten parameters that depend on the create, open, save, or close a project using icons in the left toolbar.
soil properties, and Ks is the saturated hydraulic conductivity (m/s). Data are input on the left side of the main windows. An ESRI *.asc
The maximum rainfall intensity, rmax, is estimated using Darcy's raster format le is used for the DEM input. A constant value or an
law as ESRI *.asc raster format le can be used for the soil-depth setting.
After the input of the DEM and soil depth, a simulation domain is
j zsur j ztop jtop created by setting the number of vertical soil layers. The numbers of
rmax Ks Ks 1 : (20)
Dz=2 Dz=2 cells along the x- and y-directions are automatically set according
to the raster le. Unless all the input raster les have the same
Here, the subscript sur represents the surface of the ground,
domain, TiVaSS gives an error message. The other slope-stability
where it is assumed that the pore-water pressure is zero, and the
parameters, friction angle, soil cohesion, saturated soil weight,
subscript top represents the cell immediately below the surface. If
and hydraulic conductivity can be set using two options: a constant
the rainfall intensity is larger than rmax, the overow is eliminated
value or an ESRI *.asc raster format le. The soil-water retention
from the ow domain.
curve is uniform throughout the domain. The van Genuchten soil-
The discretized system of Eq. (17) is highly nonlinear, necessi-
water retention model is the only soil-water retention model
tating iterative calculations. For this, we employ the Newton iter-
available in the current version of TiVaSS. Input datadincluding
ation, which is expressed as
distributed 2D data, rainfall, and soil-water retention curvesdcan
be plotted at the top of the main windows. The initial and end times
J jn1;m Djn1;m F jn1;m ; (21) are set at the bottom of the main windows. The simulation results,
i.e., the FS, are plotted at the center of the main windows as the
where m denotes the iteration level, j represents the vector of j, simulation progresses. The results at different times can be plotted
J vF=vj is the Jacobian matrix, and in the result tab at the top of the main windows.
Djn1;m jn1;m1
jn1;m . The iteration process of Eq. (21)
n1;m 3. Model application
continues until Dj is less than the user-specied tolerance
for all cells (Djn1;m dj , where dj denotes the convergence 3.1. Study area and landslide event
tolerance). In this study, a tolerance of dj 104 m is used. Because
the system of Eq. (17) is implicit only for one dimension, the Jaco- As shown in Fig. 4, Mt. Umyeon, with a height of 293 m, is
bian matrix is a tridiagonal matrix and can be solved by the Thomas located in the southern part of Seoul, which is the capital city of
algorithm. Korea. The mountain consists of highly weathered banded gneiss
with subordinate granitic gneiss and has hill slopes with an average
angle of 15 . There are many manmade, eco-friendly facilities
2.3. Other features of TiVaSS
within the mountain, such as a natural ecological park and mineral
springs that local residents can easily access.
The combination of Eqs. (11) and (17) gives a spatiotemporal
Before the Mt. Umyeon landslides in 2011, torrential rainfall due
slope-stability distribution for time-variant rainfall. TiVaSS is coded
to Typhoon Kompasu in 2010 caused a few landslides, and some of
in C and supports a GUI using the Qt framework. An ESRI arc
these induced debris ows in the northern valleys of the mountain,
format le or a constant value is recognized as spatially distributed
but the damage was less serious than that caused by the 2011
input data such as a digital elevation model (DEM), the soil depth,
landslides. Despite the reinforcement of structural countermea-
friction angle, soil cohesion, root cohesion, saturated soil weight,
sures against both landslides and debris ows, the same areas
and saturated hydraulic conductivity without a zoning process.
damaged by the landslides in 2010 were damaged again by the
Two-dimensional (2D) visualization for simulation results,
severe landslides in 2011. The number of damaged sites was re-
including the FS, the pore-water pressure, and the input data, is
ported as 140 slope failures and 33 debris ows by the Mt. Umyeon
supported.
landslides Research Report of Seoul City (2014).
The 3D subsurface ow module and the spatially uniform 1D
Two automatic weather stations (Namhyeon and Seocho) are
inltration module can be optionally selected in TiVaSS. The 1D
operated by the Korea Meteorological Administration near Mt.
inltration module is based on the 1D version of Eq. (12), as follows
Umyeon. Fig. 5 shows the rainfall intensity from July 26th at 00:00
to the 28th at 24:00 at the Nahyeon and Seocho stations. According
vqj v vj z
Kj 0: (22) to Yune et al. (2013), the cumulative rainfall for two months before
vt vz vz
the landslide event was 1498.5 mm at Namhyeon (1105.0 mm at
The conventional nite volume and backward Euler Seocho), which is greater than the annual average rainfall in Seoul
176 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183
(1450.5 mm). Antecedent rainfall led to soil saturation on Mt. 3.2. Input data
Umyeon, and several sites had already become vulnerable to slope
failure. Field observation by Son et al. (2012) evidenced that the A 10 m 10 m resolution DEM of Mt. Umyeon is generated from
groundwater level at locations where debris ows were initiated a 1:5,000 digital topography map of the National Geographic In-
almost reached the surface. Additional torrential rainfall of 306 mm formation Institute in Korea. Tarolli and Tarboton (2006) consid-
for the 24-h period from 9 a.m. on July 26th to 9 a.m. on the 27th ered a 10-m DEM optimal for shallow landslide modeling, as it
produced surface runoff continuously and weakened the ground, overcomes the local slope noises detected with a higher-resolution
which is made up of gneiss, eventually resulting in the tremendous topography. Tarolli (2014) also suggested that a coarse DEM (10 m)
landslides. has a resolution more suitable for analyzing major hydro-
As shown in Fig. 6, most areas of the mountain collapsed. In geomorphic processes than a ne DEM (less than 10 m). The
particular, the residential areas near the mountain were damaged slope angle of each individual cell within the study domain was
by the debris ow: 16 people were killed, and basements were calculated using the same DEM data. The geologic parameters of
inundated by mud ows. The points (except for two green points, Eqs. (7) and (11), which are related to the soil properties, were
which represent the air force base at Mt. Umyeon) and yellow lines determined according to Table 1 following Park et al. (2013), who
indicate 140 landslide-initiation coordinates and 33 debris-ow investigated the same landslide event using TRIGRS. All available
paths, respectively. In this study, we generate time-variant slope data were obtained from the eld investigation for landslide hazard
stability maps for the three days from July 26th to 28th using restoration work conducted by the National Forestry Cooperative
TiVaSS and then validate the model accuracy with 140 landslide Federation, the Korean Society of Civil Engineers, and the Korean
observations. Geotechnical Society. After the landslides occurred on 27 July, a
H. An et al. / Environmental Modelling & Software 85 (2016) 172e183 177
Fig. 4. Topography of Mt. Umyeon and slope failure locations (blue dots). (For interpretation of the references to colour in this gure legend, the reader is referred to the web version
of this article.)
total of 58 geotechnical investigation boreholes were drilled for The results of TiVaSS are shown in Fig. 8. The distribution of the
collecting soil and hydrologic and geological information. Data pore-water pressure presented in Fig. 8 is obtained for the bottom
from 13 boreholes and 19 soil samples were used in our analysis. subsurface layer, and the pore-water pressure used in Eq. (11) is
With the large database, the average values were used. The van that at the soil bottom. Most of areas were in a stable condition
178 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183
Fig. 8. Safety-factor (FS) maps and pore-water pressure for the bottom subsurface layer obtained by TiVaSS on the 27th from 7 to 10 a.m.
180 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183
Fig. 9. Safety-factor (FS) maps obtained by TiVaSS with the 1D-inltration option on the 27th from 7 to 10 a.m.
Table 2
Slope-stability classes computed by TiVaSS on the 27th from 7 to 10 a.m. at 140 observed landslide locations.
27th 7 a.m. 27th 8 a.m. 27th 9 a.m. 27th 10 a.m. 27th 7 a.m. 27th 8 a.m. 27th 9 a.m. 27th 10 a.m.
FS < 1 (Unstable) 0 0 65 95 0 0 64 93
1 < FS < 1.25 (Quasi stable) 8 17 43 18 8 15 46 20
1.25 < FS < 1.5 (Moderately Stable) 48 52 14 12 48 54 13 13
FS > 1.5 (Stable) 84 71 18 15 84 71 18 14
stable areas estimated by TiVaSS. The input data or parameters such with point data contained within only one pixel for any grid reso-
as a constant soil depth, soil cohesion, soil friction angle, and initial lution. Therefore, in this study, the landslide ratio of each predicted
condition are considered as reasons for this disagreement. The FS class (LRclass) proposed by Park et al. (2013) is implemented to
slope-stability classes obtained by TiVaSS with the 1D-inltration estimate the model accuracy as follows:
option are listed in Table 2. The results are similar, but TiVaSS with
the 3D subsurface ow outperforms that with 1D inltration. % of observed slope failure locations in each class of FS
The accuracy of landslide-prediction models is typically evalu- LRclass :
% of predicted slope failure areas in each class of FS
ated by comparing observed landslides with the predicted results
(23)
obtained using the model. According to previous studies (Crosta
and Dal Negro, 2003; Salciarini et al., 2006; Vieira et al., 2010; Table 3 shows the results of the LRclass analysis. 19.257% of the
Raia et al., 2014), the agreement of cells between predicted and area is classied as unstable, and 67.86% of the actual slope failures
actual landslides can be used to evaluate the performance of are correctly predicted. The percentage of LRFS<1 is 72.54%, which
landslide-forecasting models. However, there are no precise land- means that if a landslide occurs, the predicted unstable area (FS < 1)
slide observation data for the 2011 Mt. Umyeon landslide event. The has a 72.54% chance of including the actual landslides.
only available observations are of the landslide-initiation locations Table 4 shows the results of the LRclass analysis with 1D inl-
shown in Fig. 4. The aforementioned method is difcult to apply tration. The performance of the 3D subsurface ow outperformed
with landslide observation data points. It is expected that the that with 1D inltration in this landslide simulation, although the
denser grid causes a lower accuracy with the cell-agreement difference is not critical. The Mt. Umyeon landslide was an extreme
method because the observed landslide areas become narrow event, and all of the mountainous area became saturated, limiting
H. An et al. / Environmental Modelling & Software 85 (2016) 172e183 181
Table 3
LRclass of results obtained with the 3D subsurface ow option at 10 a.m. on the 27th.
FS classes Slope failure location (a) % of location (c) a/b % of predicted area (d) LRclass (e) c/d % of LRclass e/f
Table 4
LRclass of results obtained with the 1D inltration ow option at 10 a.m. on the 27th.
FS classes Slope failure location (a) % of location (c) a/b % of predicted area (d) LRclass (e) c/d % of LRclass e/f
the effect of horizontal ow. It is expected that a partially saturated method. Nevertheless, if the local slope is mild and the simulation
area or a longer simulation period yields a more signicant differ- period is relatively short, the 1D-inltration method has an
ence between the 3D subsurface ow and the 1D inltration advantage with regard to the computational cost.
182 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183
model for study shallow landslide vulnerability assessment. A case study in Bavi predictive power of physically based rainfall-induced shallow landslide models:
area, Hanoi region, Vietnam. In: GeoInformatics for Spatial-infrastructure a probabilistic approach. Geosci. Model Dev. 7, 495e514.
Development in Earth and Allied Sciences (GIS-IDEAS) 2010. Hanoi University Ruhaak, W., Rath, V., Wolf, A., Clauser, C., 2008. 3D nite volume groundwater and
of Technology. Retrieved from. http://gisws.media.osaka-cu.ac.jp/gisideas10/ heat transport modeling with non-orthogonal grids, using a coordinate trans-
viewabstract.php?id369. formation method. Adv. Water Resour. 31, 513e524.
Lu, N., Likos, W.J., 2006. Suction stress characteristic curve for unsaturated soil. Saadatkhah, N., Kassim, A., Lee, M., 2014. Hulu Kelang, Malaysia regional mapping of
J. Geotech. Geoenvironmental Eng. 132 (2), 131e142. http://dx.doi.org/10.1061/ rainfall-induced landslides using TRIGRS model. Arabian J. Geosciences 1e12.
(ASCE)1090-0241(2006)132:2(131). http://dx.doi.org/10.1007/s12517-014-1410-2.
Lu, N., Wayllace, A., Oh, S., 2013. Inltration-induced seasonally reactivated insta- Salciarini, D., Godt, J.W., Savage, W.Z., Conversini, P., Baum, R.L., Michael, J.A., 2006.
bility of a highway embankment near the Eisenhower Tunnel, Colorado, USA. Modeling regional initiation of rainfallinduced shallow landslides in the eastern
Eng. Geol. 162, 22e32. http://dx.doi.org/10.1016/j.enggeo.2013.05.002. Umbria Region of central Italy. Landslides 3, 181e194.
Montgomery, D.R., Dietrich, W.E., 1994. A physically based model for the topo- Simoni, S., Zanotti, F., Bertoldi, G., Rigon, R., 2008. Modelling the probability of
graphic control on shallow landsliding. Water Resour. Res. 30 (4), 1153e1171. occurrence of shallow landslides and channelized debris ows using GEOtop-
http://dx.doi.org/10.1029/93WR02979. FS. Hydrol. Process. 22 (4), 532e545. http://dx.doi.org/10.1002/hyp.6886.
Montrasio, L., Valentino, R., Losi, G.L., 2011. Towards a real-time susceptibility Sirangelo, B., Versace, P., Capparelli, G., 2003. Forewarning model for landslides
assessment of rainfall-induced shallow landslides on a regional scale. Nat. triggered by rainfall based on the analysis of historical data le. In: Hydrology of
Hazards Earth Syst. Sci. 11 (7), 1927e1947. http://dx.doi.org/10.5194/nhess-11- the Mediterranean and Semiarid Regions, Proceedings International Sympo-
1927-2011. sium Held at Montpellier, 1Al IS Publ., 278, pp. 298e304.
Pack, R.T., Tarboton, D.G., Goodwin, C.N., 1999. SINMAP 2.0-A Stability Index Soeters, R., van Westen, C.J., 1996. Slope instability rec-ognition, analysis, and
Approach to Terrain Stability Hazard Mapping, User's Manual. Retrieved from. zonation. In: Turner, A.K., Schuster, R.L. (Eds.), Landslides Investigation and
http://works.bepress.com/robert_pack/8. Mitigation. National Academy Press. National Research Council, Washington,
Park, D.W., Nikhil, N.V., Lee, S.R., 2013. Landslide and debris ow susceptibility D.C., pp. 129e177. Transportation Research Board Special Report 247.
zonation using TRIGRS for the 2011 Seoul landslide event. Nat. Hazards Earth Son, S.H., Choi, B., Paik, J., 2012. Characteristics of rainfall and groundwater table in a
Syst. Sci. Discuss. 1 (3), 2547e2587. http://dx.doi.org/10.5194/nhessd-1-2547- small forested watershed of Mt. Umyeon. In: Proceedings of 38th Korea Society
2013. of Civil Engineering Conference and Civil Expo 2012, pp. 769e772 (In Korean).
Papathoma-Ko hle, M., Zischg, A., Fuchs, S., Glade, T., Keiler, M., 2015. Loss estimation Tarolli, P., 2014. High-resolution topography for understanding Earth surface pro-
for landslides in mountain areas e an integrated toolbox for vulnerability cesses: opportunities and challenges. Geomorphology 216, 295e312.
assessment and damage documentation. Environ. Model. Softw. 63, 156e169. Tarolli, P., Tarboton, D.G., 2006. A new method for determination of most likely
http://dx.doi.org/10.1016/j.envsoft.2014.10.003. landslide initiation points and the evaluation of digital terrain model scale in
Peres, D.J., Cancelliere, A., 2016. Estimating return period of landslide triggering by terrain stability mapping. Hydrol. Earth Syst. Sci. 10, 663e677.
Monte Carlo simulation. J. Hydrol. http://dx.doi.org/10.1016/j.jhy- Terzaghi, K., 1943. Theoretical Soil Mechanics. J. Wiley.
drol.2016.03.036 (in press). Vieira, B.C., Fernandes, N.F., Filho, O.A., 2010. Shallow landslide prediction in the
Pham, B.T., Pradhan, B., Tien Bui, D., Prakash, I., Dholakia, M.B., 2016. A comparative Serra do Mar, S~ ao Paulo, Brazil. Nat. Hazards Earth Syst. Sci. 10, 1829e1837.
study of different machine learning methods for landslide susceptibility http://dx.doi.org/10.5194/nhess-10-1829-2010.
assessment: a case study of Uttarakhand area (India). Environ. Model. Softw. 84, Yune, C.Y., Chae, Y.K., Paik, J., Kim, G., Lee, S.W., Seo, H.S., 2013. Debris ow in
240e250. metropolitan area d 2011 Seoul debris ow. J. Mt. Sci. 10 (2), 199e206. http://
Pradhan, B., Lee, S., 2010. Landslide susceptibility assessment and factor effect dx.doi.org/10.1007/s11629-013-2518-7.
analysis: backpropagation articial neural networks and their comparison with Zi, T., Kumar, M., Kiely, G., Lewis, C., Albertson, J., 2016. Simulating the spatio-
frequency ratio and bivariate logistic regression modelling. Environ. Model. temporal dynamics of soil erosion, deposition, and yield using a coupled
Softw. 25, 747e759. http://dx.doi.org/10.1016/j.envsoft.2009.10.016. sediment dynamics and 3D distributed hydrologic model. Environ. Model.
Raia, S., Alvioli, M., Rossi, M., Baum, R.L., Godt, J.W., Guzzetti, F., 2014. Improving Softw. 83, 310e325.