Environmental Modelling & Software 85 (2016) 172e183

Contents lists available at ScienceDirect

Environmental Modelling & Software

journal homepage: www.elsevier.com/locate/envsoft

Development of time-variant landslide-prediction software

considering three-dimensional subsurface unsaturated ow
Hyunuk An a, Tran The Viet b, Giha Lee b, Yeonsu Kim c, Minseok Kim c, Seongjin Noh d,
Jaekyoung Noh a, *
a
Dept. of Agricultural & Rural Eng., Chungnam National University, Daejeon, South Korea
b
Dept. of Construction & Disaster Prevention Eng., Kyungpook National University, Sangju, South Korea
c
International Water Resources Research Institute, Chungnam National University, Daejeon, South Korea
d
Department of Civil Engineering, University of Texas at Arlington, Arlington, USA

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)

where ua is the air pressure, and ss is the suction stress, which is a

characteristic function of the saturation or matric suction and is
expressed in a closed form as follows:

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

jSe gw tan 4 tan 4 c

FS  : (11) Fig. 2. Mesh used in TiVaSS: orthogonal horizontally and non-orthogonal vertically.
gs D sin f cos f tan f gs D sin f cos f
If the soil is saturated, the effective saturation becomes zero, and
Eq. (11) is equivalent to Eq. (7). The only time-variant value of Eq. second-order estimation of the gradient at the surface can be ob-
(11) is the pressure head because the effective saturation is usually tained by the coordinate transformation method (Jie et al., 2004;
given as the function of the pressure head. Ruhaak et al., 2008; An and Yu, 2012). Nevertheless, to reduce the
computational cost, a simple rst-order approximation is used in
TiVaSS, as follows:
2.2. 3D subsurface ow model
j zi1;j;k  j zi;k;k
The 3D subsurface ow system of partially saturated soil is Vj zr ;
Dx
formulated by the Richards equation as follows:
j zi;j1;k  j zi;j;k
Vj zf ; (15)
vqj Dy
 V$KjVj z  q 0; (12)
vt j zi;j;k1  j zi;j;k
Vj zt :
where j is the pressure head (m); q is the volumetric moisture
Dz
content; K is the hydraulic conductivity (m/s); t is time (s); z is the The gradients for the left, back, and bottom surfaces are esti-
vertical dimension, which is assumed to be positive in the upward mated in the same manner. Note that Dx(Dy) is far larger than Dz
direction; and q is the general source term (1/s), including rainfall. and that the vertical gradient is usually greater than the horizontal
Eq. (12) is discretized within a nite-volume framework. The gradient.
mesh used in TiVaSS is orthogonal horizontally and non-orthogonal Eq. (14) is usually discretized in a fully implicit manner as
vertically, as shown in Fig. 2. Integrating Eq. (12) over the control
volume V and applying the GausseGreen divergence theorem gives qn1 n
i;j;k  qi;j;k
X
Vi;j;k  n$Kpn1 Vj zn1 n1
p Sp  Vi;j;k qi;j;k 0;
(An and Yu, 2014) Dt p
Z Z Z
vq (16)
dV  n$KVj zdvV  qdV 0; (13)
vt
V vV V where the superscript n is the time index. This scheme requires
considerable computational resources because the large linear
where vV is the control-volume boundary. q and j are assumed to matrix should be solved iteratively, and matrix solvers have a
be the cell-averaged values according to the nite-volume computational cost of the order of Nm, where N is the number of
approximation. Thus, Eq. (13) can be rewritten as cells, and m is a value between 1 and 2 that depends on the linear
solver. To reduce the computational requirement, partially implicit
vqi;j;k X temporal discretization is implemented in TiVaSS, as follows:
Vi;j;k  n,Kp Vj zp Sp  Vi;j;k qi;j;k 0; (14)
vt p
qn1 n
i;j;k  qi;j;k
X
where the subscripts p {r,l,f,b,t,bot} denote the right, left, front, Vi;j;k  n$Kpn Vj znp Sp
Dt pr;l;f ;b
back, top, and bottom surfaces of the control volume (i,j,k), X
respectively, Sp is the surface area of the control volume, and n is a  n$Kpn1 Vj zn1 n1
p Sp  Vi;j;k qi;j;k
normal vector along the control-volume surface. The subscripts pt;bot
p {r,l,f,b,t,bot} can also be expressed as (i 1/2,j,k), (i  1/2,j,k), 0: (17)
(i,j 1/2,k), (i,j  1/2,k), (i,j,k 1/2), and (i,j,k  1/2), respectively. A
 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183 175

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
Fig. 3. Main windows of TiVaSS.
(1450.5 mm). Antecedent rainfall led to soil saturation on Mt.
Umyeon, and several sites had already become vulnerable to slope
failure. Field observation by Son et al. (2012) evidenced that the
groundwater level at locations where debris ows were initiated
almost reached the surface. Additional torrential rainfall of 306 mm
for the 24-h period from 9 a.m. on July 26th to 9 a.m. on the 27th
produced surface runoff continuously and weakened the ground,
which is made up of gneiss, eventually resulting in the tremendous
landslides.
As shown in Fig. 6, most areas of the mountain collapsed. In
particular, the residential areas near the mountain were damaged
by the debris ow: 16 people were killed, and basements were
inundated by mud ows. The points (except for two green points,
which represent the air force base at Mt. Umyeon) and yellow lines
indicate 140 landslide-initiation coordinates and 33 debris-ow
paths, respectively. In this study, we generate time-variant slope
stability maps for the three days from July 26th to 28th using
TiVaSS and then validate the model accuracy with 140 landslide
observations.
3.2. Input data
A 10 m  10 m resolution DEM of Mt. Umyeon is generated from
a 1:5,000 digital topography map of the National Geographic In-
formation Institute in Korea. Tarolli and Tarboton (2006) consid-
ered a 10-m DEM optimal for shallow landslide modeling, as it
overcomes the local slope noises detected with a higher-resolution
topography. Tarolli (2014) also suggested that a coarse DEM (10 m)
has a resolution more suitable for analyzing major hydro-
geomorphic processes than a ne DEM (less than 10 m). The
slope angle of each individual cell within the study domain was
calculated using the same DEM data. The geologic parameters of
Eqs. (7) and (11), which are related to the soil properties, were
determined according to Table 1 following Park et al. (2013), who
investigated the same landslide event using TRIGRS. All available
data were obtained from the eld investigation for landslide hazard
restoration work conducted by the National Forestry Cooperative
Federation, the Korean Society of Civil Engineers, and the Korean
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.)

Genuchten parameters a and nv were set as 20 (m1) and 1.4,

respectively, as shown in Fig. 7 following the retention curve in Park
et al. (2013). Rainfall data is generated by Thiessen polygon method
using data from Nahyeon and Seocho gauge stations. A uniform soil
depth of 2 m was considered according to site investigation reports
(Korean Society of Civil Engineering, 2011). The landslides mostly
occurred between depths of 1 and 3 m.
The setting of the initial condition is difcult because there is
usually no available soil-water observation to refer to, while the
initial pore-water condition signicantly affects the results of the
slope-stability model. Raia et al. (2014) found that the modeling of
the stability condition in low-gradient terrain is very sensitive to
the initial conditions, which are uncertain and difcult to deter-
mine spatially. Therefore, the initial conditions should be set
carefully. TiVaSS provides two options for the initial conditions:
spatially constant conditions and distributed pore-water condi-
tions. The latter can be input by an ESRI raster format le (*.asc).
Here, constant initial conditions (jinit 0.3 m) are calibrated
manually.
Fig. 5. Rainfall intensity and cumulative rainfall from 26th 00:00 to 28th 24:00 at the
Namhyeon and Seocho weather stations.
3.3. Results

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. 6. Landslide and debris-ow inventories.

Table 1 pressures in the bottom soil layer changed drastically, causing a

Parameter setting.
rapid change in the FS. At 9 a.m. on the 27th, the bottom soil layer
Ks (m/s) gs (kg/m3) 4 ( ) cs (kg/m2) cr (kg/m2) qs (m3/m3) qr (m3/m3) became saturated, and the unstable areas began to increase. At 10
1.3E-05 1874 29.63 1037 0 0.5 0.18 a.m. on the 27th, almost all the areas became saturated, and the
minimum FS values were estimated.
For comparison, TiVaSS with the 1D-inltration option was
performed using the same parameters setting and data. The results
are shown in Fig. 9. The pore-water pressure is not presented here,
because the 1D-inltration option assumes the spatially uniform
distribution of the pore water. The results of the 1D inltration are
very similar to those of the 3D subsurface ow. This is attributed to
the heavy rainfall during a relatively short time, which caused
saturation throughout the study area.
To precisely validate the model performance, we employed
slope-stability classes (unstable, FS < 1: stabilizing factors are
needed for stability; quasi-stable, 1 < FS < 1.25: minor destabilizing
factors can lead to instability; moderately stable, 1.25 < FS < 1.5:
moderate destabilizing factors lead to instability; stable, 1.5 < FS:
only major destabilizing factors lead to instability) at the 140
observed landslide initiations shown in Fig. 4, as listed in Table 2.
Before the heavy rainfall, 84 locations(points) were estimated as
stable, and 48 locations were estimated as moderately stable. After
the heavy rainfall, 95 locations were estimated as unstable and 18
Fig. 7. Soil-water characteristic curve for the study area (Park et al., 2013). locations were estimated as quasi-stable among the 140 locations.
We found that 15 locations were unconditionally stable regardless
of the pore-water pressure. Most of the areas were saturated and
until the heavy rainfall on the 27th from 7 to 10 a.m., and the pore- the maximum pore-water pressure, i.e., the minimum FS value, was
water from previous rainfall barely reached the bottom soil layer at obtained at 10 a.m. on the 27th, as shown in Fig. 8. However, the
7 a.m. on the 27th. After 7 a.m. on the 27th, the pore-water landslides were observed by the eld survey in unconditionally
 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183 179

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.

FS TiVaSS 3D subsurface ow TiVaSS 1D inltration

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

Sum 140 140 140 140 140 140 140 140

stable areas estimated by TiVaSS. The input data or parameters such
as a constant soil depth, soil cohesion, soil friction angle, and initial
condition are considered as reasons for this disagreement. The
slope-stability classes obtained by TiVaSS with the 1D-inltration
option are listed in Table 2. The results are similar, but TiVaSS with
the 3D subsurface ow outperforms that with 1D inltration.
The accuracy of landslide-prediction models is typically evalu-
ated by comparing observed landslides with the predicted results
obtained using the model. According to previous studies (Crosta
and Dal Negro, 2003; Salciarini et al., 2006; Vieira et al., 2010;
Raia et al., 2014), the agreement of cells between predicted and
actual landslides can be used to evaluate the performance of
landslide-forecasting models. However, there are no precise land-
slide observation data for the 2011 Mt. Umyeon landslide event. The
only available observations are of the landslide-initiation locations
shown in Fig. 4. The aforementioned method is difcult to apply
with landslide observation data points. It is expected that the
denser grid causes a lower accuracy with the cell-agreement
method because the observed landslide areas become narrow
with point data contained within only one pixel for any grid reso-
lution. Therefore, in this study, the landslide ratio of each predicted
FS class (LRclass) proposed by Park et al. (2013) is implemented to
estimate the model accuracy as follows:
% of observed slope failure locations in each class of FS
LRclass :
% of predicted slope failure areas in each class of FS
(23)
Table 3 shows the results of the LRclass analysis. 19.257% of the
area is classied as unstable, and 67.86% of the actual slope failures
are correctly predicted. The percentage of LRFS<1 is 72.54%, which
means that if a landslide occurs, the predicted unstable area (FS < 1)
has a 72.54% chance of including the actual landslides.
Table 4 shows the results of the LRclass analysis with 1D inl-
tration. The performance of the 3D subsurface ow outperformed
that with 1D inltration in this landslide simulation, although the
difference is not critical. The Mt. Umyeon landslide was an extreme
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

FS < 1 95 67.86 19.25 3.52 72.54

1 < FS < 1.25 18 12.86 20.81 0.61 12.72
1.25 < FS < 1.5 12 8.57 18.79 0.46 9.39
1.5 < FS 15 10.71 41.15 0.26 5.36

sum 140 (b) 100 100 4.86 (f) 100

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

FS < 1 93 66.43 19.26 3.45 70.81

1 < FS < 1.25 20 14.29 20.88 0.68 14.05
1.25 < FS < 1.5 13 9.29 18.81 0.49 10.14
1.5 < FS 14 10.0 40.06 0.24 5.00

sum 140 (b) 100 100 4.87 (f) 100

Fig. 10. TiVaSS plots of 1D and 2D data.

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
The computation times of 3D subsurface ow and 1D inltration
approaches are 931 s and 97sec, respectively, on a laptop computer
with intel core i5 2.20 GHz CPU.
As stated in Section 3.2, Park et al. (2013) investigated the same
landslide event using TRIGRS, which estimated 33.33% of landslide
locations and 70.3% of LRclass with similar simulation condition.
Although a direct comparison of two models is difcult because of
differences of model structures, initial condition, soil water reten-
tion model, and etc. it can be found that the performance of TiVaSS
is comparable to that of TRIGRS.
3.4. Visualization in TiVaSS
Visualization in TiVaSS is performed using the QCustomPlot li-
brary. The simulation results, i.e., the FS, are plotted at the center of
the main windows during the simulation, as shown in Fig. 3. The
results at a specic time can be plotted in the result tab at the top of
the main windows. Fig. 10 shows an example of plotting 1D and 2D
data using TiVaSS. Fig. 10(a) and (b) show the rainfall and soil-water
retention curves, respectively. Fig. 10(c) and (d) have the same
interface, which supports a 2D map with several color gradients.
Simple x and y point data and a polygon or polyline le (ESRI shape
format) can be plotted in a 2D map, as show in Fig. 10(d).
In addition, if the point data are loaded on an FS map, the FS
class and LR class, as shown in Tables 2 and 3, can be calculated in
TiVaSS. The 2D map supports the asc, png, and jpg formats for
saving les.
4. Conclusions
A new time-variant slope stability for landslide prediction is
developed. A 3D subsurface ow model is coupled with an innite
slope-stability model to consider the effect of horizontal water
movement in the subsurface. To overcome the large 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 GUI and 2D
data visualization are supported with the Qt framework. The widely
used ESRI arc le format is supported for the input and output les.
The performance of TiVaSS is validated against data from the
2011 Mt. Umyeon landslide. Despite many uncertainties, such as
the model structure, valid soil depth, and soil cohesion, the per-
formance of TiVaSS was satisfactory overall. Among 140 landslide
initiation locations, 78 locations were judged as unstable, and 34
locations were judged as quasi-stable, which means that minor
destabilizing factors can lead to instability. The time of occurrence
of the landslide was successfully reproduced by TiVaSS. The per-
formance of the 1D-inltration option was also investigated. The
3D subsurface ow module and the spatially uniform 1D inltra-
tion module can be optionally selected in TiVaSS. The 3D subsurface
ow showed a slightly better performance than the 1D inltration,
whereas the overall FS distributions were very similar. If the local
slope is mild or the simulation period is relatively short, 1D inl-
tration is an attractive option with regard to the computational
cost.
The main advantage of TiVaSS is the numerical modeling of the
3D subsurface ow, which allows more accurate and exible
simulation. Further applications for several landslide events are
ongoing, and additional functions such as the estimation of the
landslide-prediction accuracy and overland ow will be added,
which is not currently considered in TiVaSS. The overland ow or
debris ow could be included in the next version of TiVaSS. The
Monte Carlo approach for TiVaSS with a 1D inltration module
(Raia et al., 2014; Peres and Cancelliere, 2016) and the paralleliza-
tion of the model (Alvioli and Baum, 2016) are also in our future
research scope.
Acknowledgment
This study was supported by a grant (2014-2061-01) from
Chungnam National University in 2015.
References
An, H, Yu, S., 2012. Finite volume integrated surface-subsurface ow modeling on
non-orthogonal grids. Water Resour. Res. 50, 2312e2328. http://dx.doi.org/
10.1002/2013WR013828.
An, H., Yu, S., 2014. Finite volume integrated surface-subsurface ow modeling on
nonorthogonal grids. Water Resour. Res. 50 (3), 2312e2328. http://dx.doi.org/
10.1002/2013WR013828.
Aleotti, P., 2004. A warning system for rainfall-induced shallow failures. Eng. Geol.
73, 247e265.
Alvioli, M., Baum, R.L., 2016. Parallelization of the TRIGRS model for rainfall-induced
landslides using the message passing interface. Environ. Model. Softw. 81,
122e135. http://dx.doi.org/10.1016/j.envsoft.2016.04.002.
Apip, Takara K., Yamashiki, Y., Sassa, K., Agung, A.B., Fukuoka, H., 2010. A distributed
hydrologicalegeotechnical model using satellite-derived rainfall estimates for
shallow landslide prediction system at a catchment scale. Landslides 7 (3),
237e258. http://dx.doi.org/10.1007/s10346-010-0214-z.
Baum, R.L., Godt, J.W., Savage, W.Z., 2010. Estimating the timing and location of
shallow rainfall-induced landslides using a model for transient, unsaturated
inltration. J. Geophys. Res. Earth Surf. 115 (F3), F03013. http://dx.doi.org/
10.1029/2009JF001321.
Baum, R.L., Savage, W.Z., Godt, J.W., 2008. TRIGRS a Fortran Program for Transient
Rainfall Inltration and Grid-based Regional Slope-stability Analysis, Version
2.0. U.S. Geological Survey, Reston, Va. Retrieved from. http://pubs.usgs.gov/of/
2008/1159.

Bezak, N., Sraj, M., Mikos, M., 2016. Copula-based IDF curves and empirical rainfall
thresholds for ash oods and rainfall-induced landslides. J. Hydrol. http://
dx.doi.org/10.1016/j.jhydrol.2016.02.058 (in press).
Borga, M., Fontanaa, G.D., Cazorzi, F., 2002. Analysis of topographic and climatic
control on rainfall-triggered shallow landsliding using a quasi-dynamic
wetness index. J. Hydrol. 268, 56e71.
Ciabatta, L., Camici, S., Brocca, L., Ponziani, F., Stelluti, M., Berni, N., Moramarco, T.,
2016. Assessing the impact of climate-change scenarios on landslide occurrence
in Umbria Region, Italy. J. Hydrol. http://dx.doi.org/10.1016/j.jhy-
drol.2016.02.007 (in press).
Crosta, G.B., Dal Negro, P., 2003. Observations and modelling of soil slip-debris ow
initiation processes in pyroclastic deposits: the Sarno 1998 event. Nat. Hazards
Environ. Syst. Sci. 3, 53e69.
Dickson, M.E., Perry, G.L.W., 2016. Identifying the controls on coastal cliff landslides
using machine-learning approaches. Environ. Model. Softw. 76, 117e127.
Dietrich, W.E., Montgomery, D.R., 1998. SHALSTAB: a Digital Terrain Model for
Mapping Shallow Landslide Potential. NCASI (National Council of the Paper
Industry for Air and Stream Improvement) Technical Report, 1998.
Gardner, W., 1958. Some steady-state solutions of the unsaturated moisture ow
equation with application to evaporation from a water table. Soil Sci. 85,
228e232.
Guzzetti, F., Carrara, A., Cardinali, M., Reichenbach, P., 1999. Landslide hazard
evaluation: a review of current techniques and their application in a multi-scale
study, Central Italy. Geomorphology 31 (1e4), 181e216.
Guzzetti, F., Reichenbach, P., Cardinali, M., Galli, M., Ardizzone, F., 2005. Probabilistic
landslide hazard assessment at the basin scale. Geomorphology 72 (1e4),
272e299.
Guzzetti, F., Reichenbach, P., Ardizzone, F., Cardinali, M., Galli, M., 2006. Estimating
the quality of landslide susceptibility models. Geomorphology 81 (1e2),
166e184.
Guzzetti, F., Peruccacci, S., Rossi, M., Stark, C.P., 2007. Rainfall thresholds for the
initiation of landslides in central and southern Europe. Meteorol. Atmos. Phys.
98, 239e267.
Guzzetti, F., Peruccacci, S., Rossi, M., Stark, C.P., 2008. The rainfall intensityeduration
control of shallow landslides and debris ows: an update. Landslides 5, 3e17.
Iverson, R.M., 2000. Landslide triggering by rain inltration. Water Resour. Res. 36
(7), 1897e1910. http://dx.doi.org/10.1029/2000WR900090.
Jie, Y., Jie, G., Mao, G., Li, G., 2004. Seepage analysis based on boundary-tted co-
ordinate transformation method. Comput. Geotech. 31, 279e283.
Korean Society of Civil Engineering, 2011. Research Contract Report: Causes Survey
and Restoration Work of Mt. Woomyeon Landslide. Seoul.
Liao, Z., Hong, Y., Kirschbaum, D., Adler, Robert F., Gourley, Jonathan J., Wooten, R.,
2011. Evaluation of TRIGRS (transient rainfall inltration and grid-based
regional slope-stability analysis)s predictive skill for hurricane-triggered
landslides: a case study in Macon County, North Carolina. Nat. Hazards 58
(1), 325e339. http://dx.doi.org/10.1007/s11069-010-9670-y.
Liu, C.N., Wu, C.C., 2008. Mapping susceptibility of rainfall-triggered shallow
landslides using a probabilistic approach. Environ. Geol. 55 (4), 907e915. http://
dx.doi.org/10.1007/s00254-007-1042-x.
Luan, T., Long, N., Shibayama, M., Cannata, M., 2010. Area Informacitc and TRIGRS
 H. An et al. / Environmental Modelling & Software 85 (2016) 172e183
model for study shallow landslide vulnerability assessment. A case study in Bavi
area, Hanoi region, Vietnam. In: GeoInformatics for Spatial-infrastructure
Development in Earth and Allied Sciences (GIS-IDEAS) 2010. Hanoi University
of Technology. Retrieved from. http://gisws.media.osaka-cu.ac.jp/gisideas10/
viewabstract.php?id369.
Lu, N., Likos, W.J., 2006. Suction stress characteristic curve for unsaturated soil.
J. Geotech. Geoenvironmental Eng. 132 (2), 131e142. http://dx.doi.org/10.1061/
(ASCE)1090-0241(2006)132:2(131).
Lu, N., Wayllace, A., Oh, S., 2013. Inltration-induced seasonally reactivated insta-
bility of a highway embankment near the Eisenhower Tunnel, Colorado, USA.
Eng. Geol. 162, 22e32. http://dx.doi.org/10.1016/j.enggeo.2013.05.002.
Montgomery, D.R., Dietrich, W.E., 1994. A physically based model for the topo-
graphic control on shallow landsliding. Water Resour. Res. 30 (4), 1153e1171.
http://dx.doi.org/10.1029/93WR02979.
Montrasio, L., Valentino, R., Losi, G.L., 2011. Towards a real-time susceptibility
assessment of rainfall-induced shallow landslides on a regional scale. Nat.
Hazards Earth Syst. Sci. 11 (7), 1927e1947. http://dx.doi.org/10.5194/nhess-11-
1927-2011.
Pack, R.T., Tarboton, D.G., Goodwin, C.N., 1999. SINMAP 2.0-A Stability Index
Approach to Terrain Stability Hazard Mapping, User's Manual. Retrieved from.
http://works.bepress.com/robert_pack/8.
Park, D.W., Nikhil, N.V., Lee, S.R., 2013. Landslide and debris ow susceptibility
zonation using TRIGRS for the 2011 Seoul landslide event. Nat. Hazards Earth
Syst. Sci. Discuss. 1 (3), 2547e2587. http://dx.doi.org/10.5194/nhessd-1-2547-
2013.
Papathoma-Ko hle, M., Zischg, A., Fuchs, S., Glade, T., Keiler, M., 2015. Loss estimation
for landslides in mountain areas e an integrated toolbox for vulnerability
assessment and damage documentation. Environ. Model. Softw. 63, 156e169.
http://dx.doi.org/10.1016/j.envsoft.2014.10.003.
Peres, D.J., Cancelliere, A., 2016. Estimating return period of landslide triggering by
Monte Carlo simulation. J. Hydrol. http://dx.doi.org/10.1016/j.jhy-
drol.2016.03.036 (in press).
Pham, B.T., Pradhan, B., Tien Bui, D., Prakash, I., Dholakia, M.B., 2016. A comparative
study of different machine learning methods for landslide susceptibility
assessment: a case study of Uttarakhand area (India). Environ. Model. Softw. 84,
240e250.
Pradhan, B., Lee, S., 2010. Landslide susceptibility assessment and factor effect
analysis: backpropagation articial neural networks and their comparison with
frequency ratio and bivariate logistic regression modelling. Environ. Model.
Softw. 25, 747e759. http://dx.doi.org/10.1016/j.envsoft.2009.10.016.
Raia, S., Alvioli, M., Rossi, M., Baum, R.L., Godt, J.W., Guzzetti, F., 2014. Improving
183
predictive power of physically based rainfall-induced shallow landslide models:
a probabilistic approach. Geosci. Model Dev. 7, 495e514.
Ruhaak, W., Rath, V., Wolf, A., Clauser, C., 2008. 3D nite volume groundwater and
heat transport modeling with non-orthogonal grids, using a coordinate trans-
formation method. Adv. Water Resour. 31, 513e524.
Saadatkhah, N., Kassim, A., Lee, M., 2014. Hulu Kelang, Malaysia regional mapping of
rainfall-induced landslides using TRIGRS model. Arabian J. Geosciences 1e12.
http://dx.doi.org/10.1007/s12517-014-1410-2.
Salciarini, D., Godt, J.W., Savage, W.Z., Conversini, P., Baum, R.L., Michael, J.A., 2006.
Modeling regional initiation of rainfallinduced shallow landslides in the eastern
Umbria Region of central Italy. Landslides 3, 181e194.
Simoni, S., Zanotti, F., Bertoldi, G., Rigon, R., 2008. Modelling the probability of
occurrence of shallow landslides and channelized debris ows using GEOtop-
FS. Hydrol. Process. 22 (4), 532e545. http://dx.doi.org/10.1002/hyp.6886.
Sirangelo, B., Versace, P., Capparelli, G., 2003. Forewarning model for landslides
triggered by rainfall based on the analysis of historical data le. In: Hydrology of
the Mediterranean and Semiarid Regions, Proceedings International Sympo-
sium Held at Montpellier, 1Al IS Publ., 278, pp. 298e304.
Soeters, R., van Westen, C.J., 1996. Slope instability rec-ognition, analysis, and
zonation. In: Turner, A.K., Schuster, R.L. (Eds.), Landslides Investigation and
Mitigation. National Academy Press. National Research Council, Washington,
D.C., pp. 129e177. Transportation Research Board Special Report 247.
Son, S.H., Choi, B., Paik, J., 2012. Characteristics of rainfall and groundwater table in a
small forested watershed of Mt. Umyeon. In: Proceedings of 38th Korea Society
of Civil Engineering Conference and Civil Expo 2012, pp. 769e772 (In Korean).
Tarolli, P., 2014. High-resolution topography for understanding Earth surface pro-
cesses: opportunities and challenges. Geomorphology 216, 295e312.
Tarolli, P., Tarboton, D.G., 2006. A new method for determination of most likely
landslide initiation points and the evaluation of digital terrain model scale in
terrain stability mapping. Hydrol. Earth Syst. Sci. 10, 663e677.
Terzaghi, K., 1943. Theoretical Soil Mechanics. J. Wiley.
Vieira, B.C., Fernandes, N.F., Filho, O.A., 2010. Shallow landslide prediction in the
Serra do Mar, S~ ao Paulo, Brazil. Nat. Hazards Earth Syst. Sci. 10, 1829e1837.
http://dx.doi.org/10.5194/nhess-10-1829-2010.
Yune, C.Y., Chae, Y.K., Paik, J., Kim, G., Lee, S.W., Seo, H.S., 2013. Debris ow in
metropolitan area d 2011 Seoul debris ow. J. Mt. Sci. 10 (2), 199e206. http://
dx.doi.org/10.1007/s11629-013-2518-7.
Zi, T., Kumar, M., Kiely, G., Lewis, C., Albertson, J., 2016. Simulating the spatio-
temporal dynamics of soil erosion, deposition, and yield using a coupled
sediment dynamics and 3D distributed hydrologic model. Environ. Model.
Softw. 83, 310e325.