Geomorphology 271 (2016) 40–54

Contents lists available at ScienceDirect

Geomorphology

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

Effects of soil depth and subsurface flow along the subsurface topography
on shallow landslide predictions at the site of a small granitic hillslope
Min Seok Kim a,⁎, Yuichi Onda b, Taro Uchida c, Jin Kwan Kim d
a
International Water Resources Institute, Chungnam National University, 99 Daehak-ro, Yuseong-gu, Daejeon 34134, Republic of Korea
b
Center for Research in Isotopes and Environmental Dynamics, University of Tsukuba, Tsukuba 305-8577, Japan
c
Research Center for Disaster Risk Management, National Institute for Land and Infrastructure Management, Tsukuba 305-0804, Japan
d
Department of Geography Education, Chonnam National University, Gwangju 500-757, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: Shallow landslides are affected by various conditions, including soil depth and subsurface flow via an increase in the
Received 7 May 2015 pore water pressure. In this study, we evaluate the effect of soil depth and subsurface flow on shallow landslide pre-
Received in revised form 13 July 2016 diction using the shallow landslide stability (SHALSTAB) model. Three detailed soil depth data—the average soil
Accepted 21 July 2016
depth, weathered soil depth, and bedrock soil depth—were collected using a knocking pole test at a small hillslope
Available online 25 July 2016
site composed of granite in the Republic of Korea. The SHALSTAB model was applied to a ground surface topographic
Keywords:
digital elevation model (DEM) using the three soil depths and upslope contributing area (SCA) assuming subsurface
Shallow landslides flow calculated from four DEMs: a ground surface topography (GSTO) DEM, weathered soil topography (WSTO)
Soil depth DEM, bedrock topography (BSTO) DEM, and low-level bedrock topography (EBSTO) DEM. The model performance
Subsurface flow was measured using a receiver operating characteristic (ROC) analysis. While evaluating the effect of the soil depth
Upslope contributing area with SCA using GSTO DEM, it was found that the bedrock soil depth had higher prediction accuracy compared to that
of the average soil depth or weathered soil depth. To evaluate the saturated subsurface flow between the soil and
bedrock, SCAs calculated using WSTO and BSTO DEMs were applied. From these simulations, we found that SCA
from BSTO DEM and the bedrock soil depth affect the shallow landslide prediction; however, these prediction effects
are not significantly increased by large differences in the elevation (between the lowest and highest elevation
values). Therefore, we considered the influence of the bedrock depression and SCA from EBSTO DEM. In applying
SCA from EBSTO, the prediction accuracy was significantly increased compared to the other predictions. Our results
demonstrate that the influence of the bedrock topography on the prediction of shallow landslides may be particu-
larly significant at the scale of a hillslope.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction centimeters to 3 m (Rogers and Selby, 1980; Gullà et al., 2004;

Shallow landslides are one of the most common geomorphological
processes, occurring over large areas and in different soils in various
climatic zones (Kirkby, 1987; Benda and Cundy, 1990; Selby, 1993;
Antronico et al., 2004; Borrelli et al., 2015a,b; Cascini et al., 2015).
They can cause environmental and economic damage in locations
worldwide depending on the intensity and duration of the rainfall
(Caine, 1980; Crozier, 2005; Glade et al., 2005; Guzzetti et al., 2007,
2008; Cascini et al., 2015). Shallow landslides have different morpho-
metric features depending on their localization along the slope and
have widths ranging from 3 to 15 m and lengths ranging from 10 to
100 m. The sliding surface can reach depths varying from a few
⁎ Corresponding author.
E-mail address: minseok_kim@cnu.ac.kr (M.S. Kim).
Crozier, 2005; Cascini et al., 2015).
The spatial distribution of the soil depth is controlled by complex
interactions of multiple factors, such as topography, parent material,
climate, and chemical and physical processes (Borrelli et al., 2007;
Pelletier and Rasmussen, 2009; Nicótina et al., 2011; Lanni et al.,
2012). Soil depth is a particularly important input parameter in hillslope
hydrology (Tromp-van Meerveld and McDonnell, 2006); however, its
estimation is often overlooked in landslide literature, where a soil of
uniform depth is often assumed to overlie an impermeable bedrock
(Lanni et al., 2012).
Recent hillslope hydrology studies considering subsurface flow pro-
cesses occurring during shallow landslides have shown that subsurface
topography has a strong impact on controlling the connectivity of satu-
rated areas at the soil–bedrock interface (e.g., Freer et al., 2002; Uchida
et al., 2011; Lanni et al., 2013). Because subsurface flow in steep forested
hillslopes plays an important role in stormflow generation, the landslide

http://dx.doi.org/10.1016/j.geomorph.2016.07.031
0169-555X/© 2016 Elsevier B.V. All rights reserved.
 M.S. Kim et al. / Geomorphology 271 (2016) 40–54
slip surface can also be strongly affected by its relationship with the pore
water pressure and the bedrock surface topography (e.g., Hewlett
and Hibbert, 1963, 1967; Anderson and Burt, 1978; Onda et al.,
2004; Uchida et al., 2005).
In addition, several studies have identified topography as being a
significant factor in subsurface flow (e.g., Anderson and Burt, 1978;
McDonnell, 1990; Onda et al., 2004; Uchida et al., 2011; Lanni et al.,
2013). Both field studies (e.g., Freer et al., 2002; Onda et al., 2004;
Tromp-van Meerveld and McDonnell, 2006) and numerical studies
(e.g., Hopp and McDonnell, 2009; Lanni et al., 2012) have shown that
subsurface topography has a strong impact on the connectivity of satu-
rated zones at the soil–bedrock interface and the timing and positioning
of shallow landslide initiation.
For shallow landslide predictions, increasingly complex shallow
landslide occurrence processes have been incorporated into physically
based models to predict the spatial patterns of the shallow landslide
susceptibility (e.g., Hiramatsu et al., 1990; Wu and Sidle, 1995; Rosso
et al., 2006; Talebi et al., 2008; Uchida et al., 2011), such as SHALSTAB
(Montgomery and Dietrich, 1994), SHETRAN (Ewen et al., 2000),
GEOtop FS + (Simoni et al., 2008), TRIGRS (Baum et al., 2010), and
H-slider (Uchida et al., 2011).
In these shallow landslide models, a topographic wetness index,
defined by the ratio between the specific upslope contributing area
and the local slope, is used as a surrogate for the lateral subsurface
flow processes. In general, most models use a digital elevation model
(DEM) of the ground surface to compute the steady-state wetness
index (Montgomery and Dietrich, 1994; Wu and Sidle, 1995; Pack
et al., 1998) or a “quasi-dynamic” wetness index (Barling et al., 1994;
Borga et al., 1998; Casadei et al., 2003), where it is assumed that the
specific upslope area derived from the surface topography is a surrogate
measure of the subsurface flow in response to a rainfall event of a spec-
ified duration. The subsurface flow paths (i.e., the drainage directions)
are then derived from the DEM analysis, and the land surface slope is
Fig. 1. Location of the study site in South Korea.
41
used as a substitute for the slope of the subsurface hydraulic gradients
(Lanni et al., 2013).
The slope stability component (i.e., the infinite slope stability model)
uses this topographic index to analyze the stability of each topographic
element. While hydrological models have been coupled to infinite slope
stability models (Montgomery and Dietrich, 1994; Wu and Sidle, 1995;
Pack et al., 1998; Borga et al., 2002; Casadei et al., 2003; Uchida et al.,
2011), nearly all such models assume that the soil–bedrock interface
is a simple topographic surface paralleling the soil surface.
However, saturated and unsaturated water movements on hillslopes
or catchments are affected by topography, soil depth, and hydraulic
properties in a complex manner. These properties serve as input data
for numerical simulations and have significant implications for the
simulation's accuracy; however, the effect of the flow path at different
soil depths on the slope stability is not clearly understood (Schmidt
et al., 2001). As many researchers have observed, the soil depth and sub-
surface flow are very important in shallow landslide predictions.
Therefore, the aim of this study was to investigate (1) the impact of
the three soil depths and (2) the effect of the subsurface flow at the
weathered soil layer and at the bedrock interface on shallow landslide
predictions. To determine the soil depth, we used a knocking pole test
in a small study area in Korea, and two high-resolution soil depth data
sets were collected.
2. Study area
The specific hillslope study area was located in the Jinbu-Myeon,
Pyeongchang-gun, Kangwon Prefecture in the Republic of Korea and
has a subtropical climate with year-round precipitation. The average an-
nual precipitation from 1978 to 2008 was 1400 mm. The rainfall occurs
primarily in the summer season (June–September) as a result of the
East Asian monsoon, during which time the territory of Korea is also
42 M.S. Kim et al. / Geomorphology 271 (2016) 40–54

impacted by the passage of severe tropical typhoons. The majority of the
heavy rainfall in Korea can be attributed to passing typhoons.
In particular, on July 16, 2006, shallow landslides occurred in the
study area (Fig. 1). The total amount of rainfall and the maximum
rainfall intensity of the triggering rainfall event were approximately
500 mm day− 1 and 45 mm h− 1, respectively (Korea Meteorological
Administration). The elevation of the study area ranges from 383 to
533 m above sea level, and the area and average slope angle in the
study area are approximately 1.2 ha and 40°, respectively. The surface
cover primarily consists of Korean red pine, Japanese larch, and acacia.
The shallow landslide scars range from a width of 16 to 30 m and a
length of 20 to 25 m (Fig. 2a, b), and the average soil depth on the
hillslopes in the study area is approximately 1 m as measured by the
knocking pole test (Fig. 2c).
The soil in the study area can be classified into six weathering classes
(Borrelli et al., 2007; Borrelli et al., 2015a,b): class I – fresh rock, class II –
slightly weathered rock, class III – moderately weathered rock, class IV –
highly weathered rock, class V – completely weathered rock, and class
VI – residual and colluvial soils and detrital weathered materials. On the
basis of the six weathering classes, the soil in the study area is classified
as detritical–colluvial soil over granite (Fig. 2d).
and Grant, 2000; Pack et al., 1998; Fernandes et al., 2004; Milledge
et al., 2012).
The calculation of the slope stability index, or the steady-state
critical rainfall (Rc), is based on the infinite slope form of the
Mohr–Coulomb failure law, which is expressed as the ratio of the sta-
bilizing force (shear strength) to the destabilizing force (shear
stress) on a failure plane parallel to the ground surface
(e.g., Montgomery and Dietrich, 1994; Dietrich et al., 1995; Rosso
et al., 2006; Zizioli et al., 2013):


 
b ρs ð sinβ−C Þ
Rc ¼ T sinβ 1− ð1Þ
a ρw ð cosβ tanϕÞ
where T is the saturated soil transmissivity (m 2 h− 1) defined as
khcosβ, k is the saturated hydraulic conductivity, h is the vertical
depth of the soil (m), β is the local slope angle (deg), ϕ is the internal
friction angle of the soil (deg), a is the upslope contributing area
(m2), b is the unit contour length (m), ρs is the wet soil bulk density
(g cm − 3 ), and ρw is the density of water (g cm− 3 ). Here, C is the
combined cohesion term, which is dimensionless relative to the ver-
tical soil depth, and is defined as

3. Materials and methods Cr þ Cs

C¼ ; ð2Þ
hρs g
3.1. SHALSTAB model for infinite slope analysis

where C r is the root cohesion (N m − 2), C s is the soil cohesion

For the shallow landslide hazard analysis, we used the SHALSTAB
(N m− 2), and g is the gravitational acceleration constant (9.81 m s−2).
model integrated with a simple steady-state hydrologic model and a
Given the assumptions and boundary conditions used in deriving
deterministic infinite slope stability model to delineate areas prone to
Eq. (1), it can be expressed using the conditions for the upper and
landslides due to surface topographic effects on the hydrologic
lower thresholds of elements that can possibly fail. Unconditionally stable
responses. This method was described by Montgomery and Dietrich
areas are predicted to be stable even when saturated and satisfy
(1994) and is based on earlier formulations proposed by O'Loughlin
(1986). The practicality and applicability of this methodology have

been demonstrated over the last decade, and previous studies have C ρ
tanθ≤ þ 1− s tanϕ ð3Þ
shown that its performance is satisfactory (Dietrich et al., 1995; cosθ ρw
Montgomery et al., 1998; Wu and Sidle, 1995; Borga et al., 1998; Duan

Fig. 2. Survey on the study area. a) and b) Pictures of shallow landslides at the upper and lower parts of the site, respectively. c) Picture of the soil depth measurements using a knocking
pole test. d) Soil profile.
 M.S. Kim et al. / Geomorphology 271 (2016) 40–54 43

Fig. 3. Distribution of soil depth and knocking pole data. (a) Locations of measurement points. (b) Distribution of the surface to bedrock soil depth. (c) Nd inside landslide scars obtained
from a knocking pole test. (d) Nd outside shallow landslides scars from a knocking pole test.

Unconditionally unstable elements, which in most cases are bedrock
outcrops, are unstable even when dry and satisfy


C a knocking pole test, we also recorded the locations of each using a real-
tanθN tanϕ þ
cosθ
Therefore, the “unconditionally unstable” prediction depends only
on the value assumed for the angle of internal friction and the cohesion,
which are subject to considerable uncertainty.
The predictive index of this model (i.e., the stability index) is
expressed in mm day−1 of critical rainfall and is of variable scale, with
lower values indicating a greater probability of instability and higher
values indicating a greater probability of stability. This scale also encom-
passes areas that are identified as unconditionally stable and uncondi-
tionally unstable based on the estimated rainfall value (Zizioli et al.,
2013). Here, unconditionally unstable cells are required to be unstable
because Tarolli et al. (2008) stated that the “unconditionally unstable”
prediction only depends on the value used for the angle of internal
friction and the cohesion, which are subject to considerable uncertainty.
Therefore, because an unconditionally unstable value in our study indi-
cates a very sensitive value in the shallow landslide prediction, we assume
that these values express a high potential for landslides.
3.2. Soil depth measurement
The soil depth in the study site was measured in the vertical gravita-
tional direction with measurements made at 120 points (10–15 m inter-
vals) along the slope (Fig. 3a). While the soil depth was measured using
ð4Þ
time kinematic global positioning system (RTK-GPS: GPT-7001L,
Topcon, Tokyo, Japan). Fig. 3b shows the distribution of the soil depth,
the detritical–colluvial soil layer (i.e., the weathered soil layer), and
the bedrock, which was classed as Nd (20), as shown in Fig. 3c, d. The
soil depth determined using the penetration tests was shallower in
the valley areas and deeper on the hillslopes, where some locations
had soil depths exceeding 200 cm.
Soil samples were collected at locations 1 m apart in the detritical–
colluvial soil layer and in the bedrock soil layer of the two failure scars.
Shallow landslides are triggered by elevated pore pressures that decrease
the effective normal stress (i.e., the normal load minus the pore pressure)
rather than by an increase in the shear stress. The physical soil properties,
such as the cohesion and the internal friction angle, were tested using the
triaxial compression tests. The hydraulic conductivity (Ks) was measured
in the laboratory on the basis of the “constant head” and “falling head” pa-
rameters using soil cores extracted from the field. The Ks for the study area
was calculated to be 5.0 × 10−2m s−1, which was 12 times the average
soil saturated conductivity measured in the undisturbed 100 cm3 soil
samples. Table 1 shows the soil properties.

To measure the soil depth, we used a dynamic cone penetrometer,

also known as a knocking pole (Yoshinaga and Ohnuki, 1995), which
consists of several 0.5 m flights of a 15 mm diameter stainless steel Table 1
rod, with a cone diameter of 25 mm and a weight of 5 kg with etched Soil parameters for shallow landslide prediction.
graduations every 10 cm. The penetration resistance value, Nd (drop/
Model input parameters Values
10 cm), was computed as the number of blows required to penetrate
10 cm. Uchida et al. (2009) compared the vertical Nd distributions Saturated soil weight (kN cm−3) 17.4
Dry density (kN cm−3) 14.9
between locations outside and inside shallow slope failures and found
Water density (kN cm−3) 10
that soil layers with Nd values ranging from 5 to 20 were not detected Hydraulic conductivity (m h−1) 0.05
at locations inside slope failures at their study site. They suggested Cohesion (kPa) 3.8
that soil depths with Nd ≤ 20 be defined as soil layers with the potential Internal friction angle (°) 36.5
to fail and soil depths with Nd ≥ 20 be defined as bedrock layers, which Slope degree (°) DEM
Average soil depth (m) 1m
do not fail.
44 M.S. Kim et al. / Geomorphology 271 (2016) 40–54

3.3. Construction of soil depth maps 8, Golden Software, Inc., Golden, CO, USA) was used:

The Kriging process is a useful geostatistical gridding method. The γðhÞ ¼ C 0 þ S  h; ð6Þ
ordinary Kriging method provides an estimate of the variable z at unob-
served locations based on a weighted average of the adjacent observed where C0 is the unknown nugget effect, S is the unknown slope, and h is
sites within a given area (Webster and Oliver, 1992). Multiple researchers the distance interval.
have used this method to estimate the distribution of unknown data To determine the accuracy of the interpolation, cross validation was
(e.g., Zhu and Lin, 2009; Huang et al., 2012). In the Kriging method, a conducted. The cross validation started by randomly eliminating some
variogram is used to describe the spatial structure of a variable, which available sample points, and then, the Kriging method was applied to
can be calculated as (Journel and Huijbregts, 1978; Isaaks and estimate the missing values on the basis of the remaining known points.
Srivastava, 1989) The errors between the interpolated and observed values at the missing
points were calculated using the mean absolute error (MAE) and the
mean squared error (MSE):
1 X
N ðhÞ
2
γðhÞ ¼ ½zðxi þ hÞ−zðxi Þ ; ð5Þ
2N ðhÞ i¼1
1X N
MAE ¼ ½jz0 ðxi Þ−zðxi Þj; ð7Þ
N i¼1
where γ(h) is the experimental semivariogram value at a distance inter-
val h, N(h) is the number of sample pairs within the distance interval h, 2
1X N
and z(xi) and z(xi + h) are sample values at two points separated by h. MAE ¼ ½jz0 ðxi Þ−zðxi Þj ; ð8Þ
N i¼1
All the pairs of points separated by a distance h were used to calculate
the experimental semivariogram.
The variogram model is usually a basic model or a linear combina- where z′(xi) and z(xi) are the predicted and measured values at a specific
tion of several basic models. Common theoretical models of variograms location, respectively.
can be linear, spherical, exponential, or Gaussian models. In this study, MAE is a measure of the sum of the residuals (e.g., the interpolated
to estimate the soil depth distribution, the ordinary Kriging method value minus the observed value), where a small value of MAE means
using a linear variogram available in a geostatistical analysis tool (Surfer there are few errors. However, it does not reveal the magnitude of the

Fig. 4. Soil depth maps using kriging interpolation. a) Average soil depth. b) Interpolated soil depth from the surface to weathered soil. c) Interpolated soil depth from the surface to
bedrock.
 M.S. Kim et al. / Geomorphology 271 (2016) 40–54 45

Table 2
Soil depth statistics and cross validation: a) comparison of measured soil depth distribu-
tion and interpolated soil distribution and b) cross validation results of interpolated soil.

Classification Measured IWSD MBSD IBSD

WSD

a) Soil depth statistics Minimum 0 0.013 0 0.013

(unit: m) Maximum 1.91 1.916 2.40 2.33
Mean 0.97 0.91 1.14 1.07
Standard 0.4467 0.4461 0.4941 0.4916
deviation
b) Cross validation MAE – 0.73 – 0.740
MSE – 3.51 – 4.23

MWSD: measured weathered soil depth (m).

IWSD: interpolated weathered soil depth (m).
MBSD: measured surface to bedrock soil depth (m).
IBSD: interpolated surface to bedrock soil depth (m).

error that might occur at any point. Therefore, MSE is needed to produce
an accurate estimation.
Fig. 4a shows a soil distribution map of the average soil depth in the
study area. Fig. 4b, c show the soil distribution maps produced using the
Kriging method on the weathered soil layer and bedrock layer, respec-
tively. Table 2 shows the soil depth statistics and the accuracy of the Fig. 6. The area under the receiver operating characteristics (ROC) curve (AUC) analysis
soil depth map according to the cross validation. (modified from Baum et al., 2010).

3.4. Preparation of topographic data and upslope contributing area

Unsaturated and saturated subsurface flow on a hillslope or in a
catchment is affected by the topography, the soil depth, and hydraulic
properties in a complex manner. These properties serve as input data
for numerical simulations and have significant implications for simula-
tion accuracy. Even though detailed surface topography data can usually
be readily obtained from DEMs, the soil depth and hydraulic properties
for an entire hillslope or catchment are often lacking.
We constructed a 1 m DEM using a digital elevation contour map
before the shallow landslides occurred as obtained from the National
Geographic Information Institute in Korea. Then, we used the d-infinite
algorithm of Tarboton (1997) to calculate the upslope contributing area
(SCA) for the input parameters of the SHALSTAB model, which is available
as part of the open source TauDEM software (http://www.engineering.
usu.edu/dtarb/taudem).
We assumed that the subsurface flow was concentrated at deep
depths and that the subsurface flow moves along the slope. Therefore,
we limited the four SCAs for the shallow landslide prediction by not
considering the sink effect.
First, we constructed the SCA from the ground surface topography
(GSTO DEM) in Fig. 5a. Second, to evaluate the effect of the subsurface
flow on the weathered soil layer, we constructed a DEM of the weath-
ered soil interface (WSTO DEM; the GSTO DEM minus the distribution
of the weathered soil depth map in Fig. 4b), as suggested by Freer
et al. (2002). Then, an SCA based on the WSTO DEM was calculated
using the d-infinite method (Fig. 5b). Third, to evaluate the subsurface
flow at the bedrock, a bedrock surface interface DEM (BSTO DEM) was
constructed using the same method, and an SCA was calculated on the
basis of this BSTO DEM (Fig. 5c).

Fig. 5. Calculated SCAs using the d-infinite method. (a) SCA from a GSTO DEM. (b) SCA from a WSTO DEM. (c) SCA from a BSTO DEM. (d) SCA from a EBSTO DEM.
46 M.S. Kim et al. / Geomorphology 271 (2016) 40–54

Fig. 7. The calculated distribution of steady state critical rainfall [mm day−1] for shallow landslide prediction using the SCA by GSTO and three soil depths.

Finally, Lanni et al. (2013) examined the response of the pore water
pressure in an experiment representing a bedrock interface topography
(width 1.7 m, length 3.9 m, soil depth 0–0.2 m, and three slope angles)
and compared their findings with observed experimental data and
modeling data using the Hydras 3D software. They concluded that topo-
graphic depressions at the bedrock layer induced localized zones of pore
pressure fields, which led to rapid landslide propagation in the experi-
ment and that a pore water pressure spatial distribution based on the
topography of the bedrock may significantly improve the predictive
ability of a landslide model. Therefore, we followed the methodology
of Lanni et al. (2013) to control the bedrock interface, and an SCA was
calculated using a low-level bedrock interface (EBSTO) DEM (Fig. 5d).
3.5. Simulation flow and analysis of the model performance
To test the effects of the soil depth and subsurface flow on shal-
low landslides, the distribution of the steady state critical rainfall
(mm day− 1) required to initiate shallow landslides, as predicted by
SHALSTAB, was compared to the known distribution of shallow landslides
mapped via fieldwork and the visual interpretation of high-resolution air
photos in July 2016. The slope instability that causes shallow landslides is
represented in terms of the critical rainfall (mm day−1), and the resulting
map of shallow landslide areas predicted by the SHALSTAB model is delin-
eated in terms of the steady-state rainfall intensity (mm day−1) for each
topographic element. A grid cell with a low value of steady-state critical

Table 3
Ratio of the distribution of steady-state critical rainfall [mm day−1] using three soil depths: average soil depth (ASD); weathered soil depth (WSD), and the surface to bedrock (BSD) and
the SCA from a GSTO DEM.

Simulated cells Stability classification (critical rainfall, mm day−1) Total predicted cells (%) Simulated cells in SLS (%) over total cells

AST WST BST AST WST BST

Unstable cells 0–50 0.70 1.40 0.51 0.00 0.28 0.13

50–100 0.53 0.35 0.23 0.00 0.05 0.04
100–200 1.49 1.37 0.88 0.00 0.23 0.13
200–400 6.31 6.80 5.19 0.39 0.48 0.56
400–500 5.35 4.23 4.32 0.45 0.40 0.39
Ratio of unstable cells 14.38 14.15 11.13 0.84 1.44 1.25
Stable cells 500b 50.96 51.67 45.58 2.19 1.98 2.14
Unconditionally stable 34.66 34.18 43.29 1.28 0.91 0.94
Ratio of stable cells 85.62 85.85 88.87 3.47 2.89 3.08

*Note: SLS is shallow landslide scar.

 M.S. Kim et al. / Geomorphology 271 (2016) 40–54
rainfall represents a high potential for landslides. Conversely, a high value
of steady-state critical rainfall represents a comparatively stable slope. All
grid cells with critical rainfall b 500 mm day−1 were considered to be un-
stable because, at our research site, shallow landslides were only observed
for events exceeding a total rainfall of 500 mm day−1. Furthermore, we
assumed that a calculated critical rainfall of 0 mm day−1 was uncondi-
tionally unstable because this model requires an input for the
hydrological parameter of the saturated hydraulic conductivity to indicate
that shallow landslides could occur due to rainfall.
We performed four sets of landslide predictions. In all the simulations,
the GSTO DEM, the soil physical parameters summarized in Table 1, and
the three different soil depths from Fig. 4, i.e., the average soil depth
(ASD) of 1 m, the weathered soil depth (WSD), and the surface to bedrock
depth (BSD), were kept constant. In the first simulation, the SCA from the
GSTO DEM (Fig. 5a) with an inputted subsurface flow was selected to
compare the relationships with the upslope contributing area on the
basis of the GSTO DEM and each of the three soil depths. In the second
and third simulations, the SCAs created using the WSTO DEM and the
BSTO DEM in Fig. 5b, c, respectively, with inputted wetness index param-
eters were selected to compare the effects of the subsurface flow. For the
fourth simulation, we constructed an SCA using the emphasized bedrock
interface DEM in Fig. 5d, which was also entered into the SHALSTAB
model. We conducted four runs per set, resulting in 12 different maps
of steady-state critical rainfall. These maps were used to evaluate the
effects of the soil depth and the subsurface flow.
Each of the simulated results was evaluated using a receiver operating
characteristic (ROC), defined by the false alarm rate (FPR) and the hit rate
(TPR) plotted on the x- and y-axes, respectively (Fig. 6). The prediction
Fig. 8. The calculated distribution of steady state critical rainfall [mm day−1] for shallow landslide prediction using the SCA by WSTO and three soil depths.
47
accuracy of regional landslide susceptibility models is typically evaluated
by comparing the locations of known landslides with simulation results
from the model (Swets, 1988; Montgomery et al., 1998, 2001; Fawcett,
2006; Van Den Eeckhaut et al., 2006; Godt et al., 2008). In the ROC
space, a point located in the upper left corner represents a perfect predic-
tion (TPR = 1 and FPR = 0), and points along the diagonal line for which
TPR = FPR represent random predictions. An acceptable prediction
requires that TPR/FPR N 1 (Fawcett, 2006; Rossi et al., 2010; Raia et al.,
2014). In addition, because an ROC curve is a two-dimensional depiction
of the model performance, a single scalar value representing the expected
performance is needed to compare the model performances. A common
method is to calculate the area under the ROC curve (AUC), and then,
the AUC is used as a quantitative measure of the performance of the clas-
sification. If AUC = 0.5, the classification is poor and indistinguishable
from a random classification, whereas a perfect classification has
AUC = 1 (Fawcett, 2006; Rossi et al., 2010).
4. Results and discussion
4.1. Three soil depths for landslide prediction
To determine the effect of the soil depth in shallow landslide predic-
tions, we used the three different soil depths in Fig. 4 and the SCA calcu-
lated from the GSTO DEM in Fig. 5a. Fig. 7 shows the results of the
shallow landslide prediction using the three different distributions of
soil depth [i.e., (a) ASD, (b) WSD, and (c) BSD], which is expressed by
the steady-state critical rainfall (mm day−1) using a safety factor.
Table 3 shows the result for the distribution ratio of the critical rainfall
48 M.S. Kim et al. / Geomorphology 271 (2016) 40–54

Fig. 9. The calculated distribution of steady state critical rainfall [mm day−1] for shallow landslide prediction using the SCA by BSTO and three soil depths.

Table 4
Ratio of the distribution of steady-state critical rainfall [mm day−1] using three soil depths: average soil depth (ASD); weathered soil depth (WSD), and the surface to bedrock (BSD) and
the SCA from a WSTO DEM.

Simulated cells Stability classification (critical rainfall, mm day−1) Total predicted cells (%) Simulated cells in SLS (%) over total cells

AST WST BST AST WST BST

Unstable cells 0–50 0.69 0.51 1.39 0.00 0.28 0.13

50–100 0.54 0.26 0.42 0.00 0.05 0.04
100–200 1.60 0.94 1.43 0.00 0.23 0.12
200–400 6.32 5.35 6.91 0.41 0.53 0.61
400–500 5.15 4.39 4.53 0.42 0.42 0.37
Ratio of unstable cells 14.31 11.46 14.67 0.83 1.51 1.27
Stable cells 500b 51.03 45.25 51.15 6.55 6.24 6.46
Unconditionally stable 34.66 43.29 34.18 1.28 0.91 0.94
Ratio of stable cells 85.69 88.54 85.33 7.83 7.15 7.40

*Note: SLS is shallow landslide scar.

Table 5
Ratio of the distribution of steady-state critical rainfall [mm day−1] using three soil depths: average soil depth (ASD); weathered soil depth (WSD), and the surface to bedrock (BSD) and
the SCAfrom a BSTO DEM.

Simulated cells Stability classification (critical rainfall, mm day−1) Total predicted cells (%) Simulated cells in SLS (%) over total cells

AST WST BST AST WST BST

Unstable cells 0–50 0.71 0.53 1.42 0.00 0.12 0.27

50–100 0.55 0.28 0.39 0.00 0.04 0.06
100–200 1.55 0.89 1.43 0.00 0.11 0.23
200–400 6.17 5.24 6.84 0.38 0.65 0.59
400–500 5.44 4.64 4.71 0.58 0.44 0.48
Ratio of unstable cells 14.41 11.58 14.80 0.96 1.36 1.63
Stable cells 500b 50.93 45.12 51.02 6.40 2.01 0.91
Unconditionally stable 34.66 43.29 34.18 1.28 0.94 1.78
Ratio of stable cells 85.59 88.42 85.20 7.68 2.95 2.69

*Note: SLS is shallow landslide scar.

 M.S. Kim et al. / Geomorphology 271 (2016) 40–54
Fig. 10. The calculated distribution of steady state critical rainfall [mm day−1] for shallow landslide prediction using the SCA by EBSTO and three soil depths.
values. In Fig. 7 and Table 3, a low steady-state critical rainfall value
resulted in a high potential for landslides. By contrast, a high steady-
state critical rainfall value indicates that a hillslope is comparatively sta-
ble. For comparison, all grid cells with critical rainfall b 500 mm day−1
were considered to be unstable (i.e., shallow landslides would occur) be-
cause shallow landslides occur when the total rainfall is approximately
500 mm day−1 at our research site.
Fig. 11a shows the accuracy of the shallow landslide prediction using
the ROC curve as a quantitative measure of the performance of the predic-
tion. Table 7a shows a summary of the ROC analysis. From these results,
we found that the bedrock soil depth influenced relatively shallow land-
slide predictions more than the weathered soil depth or the average soil
depth when the three soil depth data sets (Fig. 11a) were used as input
parameters. In particular, the average soil depth shows unreasonable re-
sults in these simulation cases. This is likely due to T (soil transmissivity)
used in Eq. (1), which is defined by the relationship between the soil
depth and the saturated hydraulic conductivity. Multiple researchers
(e.g., Montgomery and Dietrich, 1994; Montgomery et al., 1998; Godt
et al., 2008; Rosso et al., 2006; Chiang and Chang, 2009; Ho et al., 2012;
Lanni et al., 2012) have widely adopted the average soil depth or a uni-
form soil depth in slope instability analyses. However, the soil depth is a
sensitive parameter in shallow landslide models because the soil depth
is related to the saturated soil depth.
Claessens et al. (2005) created a sensitivity plot for the steady-state
critical rainfall Rc (mm day− 1) using Eq. (1) and found that the soil
depth and transmissivity had a positive correlation in shallow landslide
predictions. Uchida et al. (2011) determined that the use of the
49
measured soil depth resulted in higher prediction results than the use
of the average soil depth in their granitic study area using the H-slider
model, even though their model was different from SHALSTAB. Notably,
Ho et al. (2012) reported that shallow landslide predictions are more
sensitive to the uniform deep soil depth than the uniform shallow soil
depth (i.e., from 0.4 to 2 m) in the SHALSTAB model, even though
they did not consider the detailed soil depth distribution. Therefore,
the spatial variability in soil depth, as mentioned in other studies
(e.g., Keijsers et al., 2011; Lanni et al., 2012), is likely important in shal-
low landslide predictions on steep landscapes and knowing the soil
depth in detail might reduce the over-prediction of shallow landslides.
4.2. Shallow landslide prediction using different SCAs
Milledge et al. (2009) reported that the Q/T ratio for the probability
of a shallow landslide is changed by the topographic wetness index
when using SHALSTAB, even though they did not consider the SCA of
the subsurface topography and only focused on the ground surface.
Multiple researchers (e.g., Freer et al., 2002; Lanni et al., 2012) have
reported that shallow landslides occur between the soil and the bedrock
and that saturated subsurface flow on a hillslope is affected by the sub-
surface topography. Therefore, we performed simulations to determine
how the subsurface flow affected the shallow landslide prediction for
the SCAs calculated from the WSTO DEM (assuming the weathered
soil topography) and the BSTO DEM (assuming the bedrock topogra-
phy). Figs. 8 and 9 show the results of the shallow landslide predictions
using the SCAs from the WSTO and BSTO DEMs, respectively, with the
50 M.S. Kim et al. / Geomorphology 271 (2016) 40–54
three different distributions of soil depth [i.e., ASD, WSD, and BSD].
Tables 4 and 5 summarize the distribution ratios of the critical rainfall
values for the WSTO and BSTO DEMs, respectively. Fig. 11b,c shows
the accuracy of the shallow landslide predictions using the ROC curve
as a quantitative measure of the performance of the prediction.
Table 7b, c show a summary of the ROC analysis. All values indicating
model performance increased compared to those derived from the
GSTO DEM.
In particular, in the ROC analysis to determine the accuracy of the
shallow landslide predictions, the prediction of shallow landslides was
better compared to other run sets when we considered the bedrock sub-
surface flow. From this result, we found that the SCA derived from the
subsurface topography can affect the shallow landslide predictions.
However, the accuracy of the shallow landslide predictions did not sig-
nificantly increase according to the results of the ROC analysis
(Fig. 11a,b and c), even when the soil depth and subsurface flow were
considered (Figs. 8 and 9). This is because the flow path simulations
were not significantly improved when considering the subsurface inter-
face DEMs. From the simulation results, we found that the bedrock soil
depth and the SCA from BSTO affect the shallow landslide prediction,
which has a reasonable value, even though it did not increase greatly
as mentioned in Lanni et al. (2012). This is likely due to the large eleva-
tion range effect.
Freer et al. (2002) and Zhu and Lin (2009) studied hydrological
responses using the bedrock surface topography for the subsurface
flow mechanism. Freer et al. (2002) obtained reasonable results when
Fig. 11. The results of the area under the receiver operating characteristics (ROC) curve (AUC) analysis for prediction results using different SCAs.
comparing the monitored data and the simulated wetness index at
low elevation in their small study site. Conversely, Zhu and Lin (2009)
did not observe significantly increased subsurface simulation values at
their study site. They proposed that the topography-reflected bedrock
interface was not dominated by the variation in the ground surface ele-
vation due to the maximum difference in the surface elevation values
between the lowest point in the footslope and the highest point at the
ridge top. The difference in the elevation (between the lowest elevation
value and the highest elevation value) was approximately 150 m in our
study area. This means that the analysis of the SCA using the d-infinity
method of Tarboton (1997) may have affected the two SCA simulation
results from the WSTO and BSTO DEMs. We did not observe significantly
increased SCA values from the BSTO DEM compared to the WSTO DEM
(Fig. 5b and c). However, the prediction performances for the different
SCAs values increased. The results show that the use of an SCA, reflecting
the subsurface flow, affects the prediction results according to the ROC
analysis in Fig. 11b, c and Table 7b, c.
4.3. Influence of subsurface flow for low hillslope gradients
Specific catchment area values, which possibly control Rc, are usually
nullified because higher values tend to occur in the valley bottom
towards the outlet cell where the slopes are generally sufficiently low
and are classified as unconditionally stable. This was confirmed by com-
paring the descriptive statistics of the specific catchment areas with the
valid critical rainfall calculation and the unconditionally stable zones. In
 M.S. Kim et al. / Geomorphology 271 (2016) 40–54 51

Table 6
Ratio of the distribution of steady-state critical rainfall [mm day−1] using three soil depths: average soil depth (ASD); weathered soil depth (WSD), and the surface to bedrock (BSD) and
the SCA from an EBSTO.

Simulated cells Stability classification (critical rainfall, mm day−1) Total predicted cells (%) Simulated cells in SLS (%) over total
cells

AST WST BST AST WST BST

Unstable cells 0–50 0.69 0.92 1.12 0.02 0.18 0.23

50–100 0.72 0.77 0.79 0.06 0.09 0.08
100–200 1.97 2.15 2.07 0.19 0.39 0.33
200–400 5.30 4.99 4.83 0.77 0.74 0.75
400–500 2.72 2.35 2.61 0.22 0.29 0.29
Ratio of unstable cells 11.40 11.18 11.42 1.26 1.69 1.68
Stable cells 500b 53.91 45.50 52.49 1.77 1.69 1.70
Unconditionally stable 34.69 43.31 36.09 1.29 0.95 0.94
Ratio of stable cells 88.60 88.82 88.58 3.06 2.64 2.64

*Note: SLS is shallow landslide scar.

our study, the SCAs simulated from the GSTO DEM, the WSTO DEM, and its relation with the pore pressure and the soil depth. However, when
the BSTO DEM had different values; the simulated SCAs for the subsur- we considered a low slope inclination for the subsurface flow, the SCA
face flow of each DEM were nearly identical due to the difference in el- value assuming this subsurface flow substantially changed along the
evation shown in Fig. 5. For subsurface flow along a low slope bedrock bedrock topography. In addition, the SCA value has larger values in bed-
interface, Lanni et al. (2013) mentioned that despite the influence of rock depressions than at shallow depths, as mentioned by Lanni et al.
the slope angle the maximum pore pressure increases are localized in (2013). In other words, the subsurface flow along the bedrock topogra-
the bedrock depression, where the factor of safety is then most affected; phy is very important for shallow landslide predictions, and our main
however, the maximum value of the pressure head was not affected by finding is that the SCA for the subsurface flow particularly affects the
the slope inclination in the three experiments and the numerical shallow landslide prediction performance of the model.
modeling.
Therefore, to evaluate the influence of the subsurface in a low slope 4.4. Other factors and further studies
bedrock depression, we calculated the SCA (Fig. 5d) reflecting the low
slope bedrock interface given by Lanni et al. (2013) and applied it to Fig. 12 shows a comparison with the best results in shallow landslide
the SHALSTAB model to evaluate the subsurface flow related to the bed- predictions and Fig. 13 is the area under the receiver operating charac-
rock depression. Here, the SCA values from the EBSTO DEM increased teristics curve (AUC) analysis among four best results. In this study, we
along the bedrock depression, and a failure zone was localized in the employed the SHALSTAB model with three soil depths and four SCAs.
main bedrock depression where the maximum increase in the SCA However, our prediction of shallow landslides did not perfectly match
values were calculated (Fig. 5d). Fig. 10 shows a prediction of shallow the actual mapped shallow landslides (Fig. 12). This is because other
landslides using the SCA from the EBSTO DEM. Fig. 11d shows the parameters affect the shallow landslide mechanism. The limitations of
accuracy of the shallow landslide prediction using the ROC curve as a SHALSTAB, in particular, in predicting at a high specificity, can partly
quantitative measure of the performance of the prediction. Table 6 sum- be attributed to its “hard” classification of unconditionally unstable
marizes the distribution ratios of the critical rainfall values for EBSTO slopes, which does not provide a means to further continuously differ-
DEM. Table 7d shows a summary of the ROC analysis. From the best re- entiate between the numerous unconditionally unstable grid cells.
sults in our study, we found that the SCA from EBSTO and the bedrock These areas can be determined on the basis of the friction angle and
soil depth both affect the shallow landslide predictions. the cohesion, which were assumed to be spatially constant in our
Lanni et al. (2013) reported that topographic depressions at the bed- study and in most other published applications of SHALSTAB (Goetz
rock layer induce localized zones of increased pressure heads, which et al., 2011). Claessens et al. (2005) and De Sy et al. (2013) created
lead to rapid landslide propagation. Therefore, the mapped shallow sensitivity plots showing the relative importance of each variable versus
landslide prediction area was localized near high SCA values. In this the steady-state critical rainfall, Rc (mm day−1). Multiple researchers
study, as we only considered the SCA value, we did not exactly capture (e.g., Rosso et al., 2006; Minder et al., 2009; Milledge et al., 2012) have
determined that a safety factor needs to be achieved for stability when
Table 7 the cohesion has a high value. This suggests that the estimation of the
Summary of the simulated results using ROC analysis. cohesion is an important step in the prediction of shallow landslides.
DEM Soil depth TPR FPR TPR/FPR AUC ACC PRC Therefore, further studies of other factors affecting shallow landslide
prediction, including the effects of the distribution of soil cohesion, are
(a) GSTO ASD 0.19 0.104 1.83 0.6 0.86 0.07
WSD 0.28 0.07 4.00 0.7 0.87 0.12 necessary.
BSD 0.33 0.102 3.24 0.71 0.9 0.15
(b) WSTO ASD 0.19 0.09 2.11 0.6 0.87 0.08 5. Conclusions
WSD 0.29 0.06 4.83 0.7 0.88 0.146
BSD 0.34 0.09 3.78 0.72 0.9 0.176
Soil depth and subsurface flow play important roles in the shallow
(c) BSTO ASD 0.22 0.09 2.44 0.62 0.87 0.09
WSD 0.37 0.09 4.11 0.71 0.88 0.15 landslide mechanism. These properties serve as input data for numerical
BSD 0.31 0.06 5.17 0.73 0.9 0.18 simulations and have significant implications for simulation accuracy.
(d) EBSTO ASD 0.29 0.073 3.97 0.64 0.89 0.15 However, the effect of subsurface flow on slope stability at different
WSD 0.38 0.071 5.35 0.72 0.9 0.19
soil depths is not clearly understood. In this study, we collected data at
BSD 0.38 0.073 5.21 0.75 0.9 0.19
three soil depths, and SCAs assuming the subsurface flow were construct-
TPR: true positive rate. ed using four topographic DEMs. Each calculated SCA was entered into a
FPR: false positive rate.
ACC: accuracy.
physically based SHALSTAB model based on the ground surface DEMs
PRC: precision. and was used to evaluate the effect of the soil depth and the subsurface
AUC: area under the ROC curve. flow on shallow landslide predictions.
52 M.S. Kim et al. / Geomorphology 271 (2016) 40–54

Fig. 12. Comparison of the best results of shallow landslide predictions using SCA by (a) GSTO, (b) WSTO, (c) BSTO and (d) EBSTO, respectively.

Evaluating the soil depth using the bedrock soil depth produced
more frequent shallow landslide predictions than the use of the average
soil depth and the weathered soil depth. However, the accuracy of the
results did not ultimately improve, and the use of the average soil
depth was not considered reasonable. We found that the subsurface
flow had an effect on shallow landslide predictions, and the results in
combination with the bedrock soil depth and the flow path of the SCA
with the bedrock–soil interface topography produced good predictions.
However, the flow paths of the SCA, representing the subsurface flow,
were unsatisfactory in reflecting the topography due to the 150 m
elevation difference between the DEMs in our study area.
Therefore, we calculated the flow path of an SCA using an empha-
sized bedrock interface topographic DEM, which was applied to the
SHALSTAB model. The results indicate that the subsurface flow plays a
key role in shallow landslide predictions, and the accuracy was greatly
improved. Our results demonstrate that understanding the bedrock
topography is likely very important to make accurate shallow landslide
predictions.

Acknowledgments

Fig. 13. The area under the receiver operating characteristics curve (AUC) analysis among This research was supported by the Basic Research Project (Develop-
four best results. ment of an integrated early detection system of landslides based on
 M.S. Kim et al. / Geomorphology 271 (2016) 40–54
real-time monitoring) of the Korea Institute of Geoscience and Mineral
Resources (KIGAM) funded by the Ministry of Science, ICT and Future
Planning of Korea. We are very grateful to Editor Takashi Oguchi and
twoanonymous reviewers for the valuable criticisms and suggestions
during the review process.
References
Anderson, M.G., Burt, T.P., 1978. Experimental investigations concerning the topographic
control of soil water movement on hillslopes. Z. Geomorphol. 29, 52–63 (Supplement
band).
Antronico, L., Gullà, G., Borrelli, L., 2004. Shallow instabilities for sliding flow: regional in-
fluence and area affects. Proceedings of the IX International Symposium on Land-
slides, Rio de Janeiro, Brasile. Lacerda. Ehrlich, Fontoura & Sayao, London, UK,
pp. 1381–1388.
Barling, D.B., Moore, I.D., Grayson, R.B., 1994. A quasi-dynamic wetness index for
characterising the spatial distribution of zones of surface saturation and soil water
content. Water Resour. Res. 30, 1029–1044. http://dx.doi.org/10.1029/93WR03346.
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 infiltration.
J. Geophys. Res. 115. http://dx.doi.org/10.1029/2009JF001321.
Benda, L.E., Cundy, T.W., 1990. Predicting deposition of debris flows in mountain channels.
Can. Geotech. J. 27, 409–417.
Borga, M., Dalla Fontana, G., Gregoretti, C., Marchi, L., 1998. Shallow landslide hazard
assessment using a physically based model and digital elevation data. Environ. Geol.
35, 81–88.
Borga, M., Dalla Fontana, G., Cazorzi, F., 2002. Analysis of rainfall-triggered shallow
landsliding using a quasi-dynamic wetness index. J. Hydrol. 268, 56–71.
Borrelli, L., Greco, R., Gullà, G., 2007. Weathering grade of rock masses as a predisposing
factor to slope instabilities: reconnaissance and control procedures. Geomorphology
87, 158–175.
Borrelli, L., Coniglio, S., Critelli, S., Barbera, A.L., Gullà, G., 2015a. Weathering grade in gran-
itoid rocks: the San Giovanni in Fiore area (Calabria, Italy). J. Maps 12, 260–275.
http://dx.doi.org/10.1080/17445647.2015.1010742.
Borrelli, L., Cofone, G., Coscarelli, R., Gullà, G., 2015b. Shallow landslides triggered by con-
secutive rainfall events at Catanzaro Strait (Calabria–Southern Italy). J. Maps 11,
730–744. http://dx.doi.org/10.1080/17445647.2014.943814.
Caine, N., 1980. The rainfall intensity-duration control of shallow landslides and debris
flows. Geogr. Ann., Ser. A 62, 23–27.
Casadei, M., Dietrich, W.E., Miller, N.L., 2003. Testing a model for predicting the timing
and location of shallow landslide initiation in soil mantled landscapes. Earth Surf.
Process. Landf. 28, 925–950.
Cascini, L., Ciurleo, M., Nocera, S.D., Gullà, G., 2015. A new–old approach for shallow land-
slide analysis and susceptibility zoning in fine-grained weathered soils of southern
Italy. Geomorphology 241, 371–381.
Chiang, S.H., Chang, K.T., 2009. Application of radar data to modeling rainfall-
inducedlandslides. Geomorphology 103, 299–309.
Claessens, L., Heuvelink, G.B.M., Schoorl, J.M., Veldkamp, A., 2005. DEM resolution effects
on shallow landslide hazard and soil redistribution modeling. Earth Surf. Process.
Landf. 30, 461–477.
Crozier, M.J., 2005. Multiple-occurrence regional landslide events in New Zealand: hazard
management issues. Landslides 2, 247–256.
De Sy, V., Schoorl, J.M., Keesstra, S.D., Jones, K.E., Claessens, L., 2013. Landslide model per-
formance in a high resolution small-scale landscape. Geomorphology 190, 73–81.
Dietrich, W.E., Reiss, R., Hsu, M., Montgomery, D.R., 1995. A process-based model for col-
luvial soil depth and shallow landsliding using digital elevation data. Hydrol. Process.
9, 383–400.
Duan, J., Grant, G.E., 2000. Shallow landslide delineation for steep forest watersheds based
on topographic attributes and probability analysis. In: Wilson, J.P., Gallant, J.C. (Eds.),
Terrain Analysis - Principles and Applications. John Wiley & Sons, pp. 311–329.
Ewen, J., Parkin, G., O'Connell, P.E., 2000. SHETRAN: a coupled surface/subsurface model-
ling system for 3D water flow and sediment and solute transport in river basins.
J. Hydraul. Eng. ASCE 5, 250–258.
Fawcett, T., 2006. An introduction to ROC analysis. Pattern Recogn. Lett. 27, 861–874.
Fernandes, N.F., Guimarães, R.F., Gomes, R.A.T., Vieira, B.C., Montgomery, D.R., Greenberg,
M.H., 2004. Topographic controls of landslides in Rio de Janeiro: field evidence and
modeling. Catena 55, 163–181.
Freer, J., McDonnell, J.J., Beven, K.J., Peters, N.E., Burns, D.A., Hooper, R.P., Aulenbach, B.,
Kendall, C., 2002. The role of bedrock topography on subsurface storm flow. Water
Resour. Res. 38. http://dx.doi.org/10.1029/2001WR000872.
Glade, T., Anderson, M., Crozier, M.J., 2005. Landslide Hazard and Risk. John Wiley & Sons
Ltd., Chichester, UK.
Godt, J.W., Baum, R.L., Savage, W.Z., Salciarini, D., Schulz, W.H., Harp, E.L., 2008. Transient
deterministic shallow landslide modeling: requirements for susceptibility and hazard
assessments in a GIS framework. Eng. Geol. 102, 214–226.
Goetz, J.N., Guthrie, R.H., Brenning, A., 2011. Integrating physical and empirical landslide
susceptibility models using generalized additive models. Geomorphology 129,
376–386.
Gullà, G., Aceto, L., Niceforo, D., 2004. Geotechnical characterisation of fine-grained soils
affected by soil slips. Proc. of the 9th International Symposium on Landslides, Rio
de Janeiro, Brazil, pp. 663–668.
Guzzetti, F., Peruccacci, S., Rossi, M., Stark, C.P., 2007. Rainfall thresholds for the initiation
of landslides in central and southern Europe. Meteorog. Atmos. Phys. 98, 239–267.
53
Guzzetti, F., Peruccacci, S., Rossi, M., Stark, C.P., 2008. The rainfall intensity-duration control
of shallow landslides and debris flows: an update. Landslides 5, 3–17.
Hewlett, J.D., Hibbert, A.R., 1963. Moisture and energy conditions within a sloping mass
during drainage. J. Geophys. Res. 68, 1081–1087.
Hewlett, J.D., Hibbert, A.R., 1967. Factors affecting the response of small watersheds to
precipitation in humid areas. In: Sopper, W.E., Lull, H.W. (Eds.), Forest Hydrology.
Pergamon Press, New York, pp. 275–290.
Hiramatsu, S., Mizuyama, T., Ishikawa, Y., 1990. Study of a method for predicting hillside
landslides by analysis of transient flow of water in saturated and unsaturated soils.
J. Jpn. Soc. Erosion Control. Eng. 43, 5–15 (in Japanese).
Ho, J.Y., Lee, K.T., Chang, T.C., Wang, Z.Y., Liao, Y.H., 2012. Influence of spatial distribution
of soil thickness on shallow landslide prediction. Eng. Geol. 124, 38–46.
Hopp, L., McDonnell, J.J., 2009. Connectivity at the hillslope scale: identifying interactions
between storm size, bedrock permeability, slope angle and soil depth. J. Hydrol. 376
(3), 378–391.
Huang, Z., Wang, H., Zhang, R., 2012. An improved kriging interpolation technique based
on SVM and its recovery experiment in oceanic missing data. Am. J. Comput. Math. 2,
56–60.
Isaaks, E.H., Srivastava, R.M., 1989. An Introduction to Applied Geostatistics. Oxford Uni-
versity Press, New York, p. 561.
Journel, A.G., Huijbregts, C.J., 1978. Mining Geostatistics. 600. Academic Press, London.
Keijsers, J.G.S., Schoorl, J.M., Chang, K.T., Chiang, S.H., Claessens, L., Veldkamp, A., 2011.
Calibration and resolution effects on model performance for predicting shallow land-
slide locations in Taiwan. Geomorphology 133, 168–177.
Kirkby, M.J., 1987. General models of long-term slope evolution through mass movement.
In: Anderson, M.G., Richards, K.S. (Eds.), Slope Stability. John Wiley and Sons Ltd.,
Chichester, UK, pp. 359–379.
Lanni, C., Borga, M., Rigon, M., Tarolli, P., 2012. Modelling shallow landslide susceptibility
by means of a subsurface flow path connectivity index and estimates of soil depth
spatial distribution. Hydrol. Earth Syst. Sci. 16, 3959–3971. http://dx.doi.org/10.
5194/hess-16-3959-2012.
Lanni, C., McDonnell, J., Hopp, L., Rigon, R., 2013. Simulated effect of soil depth and bedrock
topography on near-surface hydrologic response and slope stability. Earth Surf. Process.
Landf. 38, 146–159. http://dx.doi.org/10.1002/esp.3267.
McDonnell, J.J., 1990. A rationale for old water discharge through macropores in a steep,
humid catchment. Water Resour. Res. 26 (11), 2821–2832.
Milledge, D.G., Lane, S.N., Warburton, J., 2009. The potential of digital filtering of generic
topographic data for geomorphological research. Earth Surf. Process. Landf. 34,
63–74.
Milledge, D.G., Griffiths, D.V., Lane, S.N., Warburton, J., 2012. Limits on the validity of infi-
nite length assumptions for modelling shallow landslides. Earth Surf. Process. Landf.
37, 1158–1166.
Minder, J.R., Roe, G.H., Montgomery, D.R., 2009. Spatial patterns of rainfall and shallow
landslide susceptibility. Water Resour. Res. 45, W04419. http://dx.doi.org/10.1029/
2008WR007027.
Montgomery, D.R., Dietrich, W.E., 1994. A physically based model for the topographic
control on shallow landsliding. Water Resour. Res. 30, 1153–1171. http://dx.doi.
org/10.1029/93WR02979.
Montgomery, D.R., Sullivan, K., Greenberg, M.H., 1998. Regional test of a model for shal-
low landsliding. Hydrol. Process. 12, 943–955.
Montgomery, D.R., Balco, G., Willett, S.D., 2001. Climate, tectonics, and the morphology of
the Andes. Geology 29, 579–582. http://dx.doi.org/10.1130/0091-
7613(2001)029b0579:CTATMON2.0.CO;2.
Nicótina, L., Tarboton, D.G., Tesfa, T.K., Rinaldo, A., 2011. Hydrologic controls on equilibri-
um soil depths. Water Resour. Res. 47, W04517. http://dx.doi.org/10.1029/
2010WR009538.
O'Loughlin, E.M., 1986. Prediction of surface saturation zones in natural catchments by to-
pographic analysis. Water Resour. Res. 22, 794–804.
Onda, Y., Tsujimura, M., Tabuchi, H., 2004. The role of subsurface water flow paths on hill-
slope hydrological processes, landslides and landform development in steep moun-
tains of Japan. Hydrol. Process. 18, 637–650.
Pack, R.T., Tarboton, D.G., Goodwin, C.N., 1998. The SINMAP approach to terrain stability
mapping. 8th Congress of the International Association of Engineering Geology.
Canada, Vancouver, British Columbia.
Pelletier, J.D., Rasmussen, C., 2009. Geomorphically based predictive mapping of soil
thickness in upland watersheds. Water Resour. Res. 45, W09417. http://dx.doi.org/
10.1029/2008WR007319.
Raia, S., Alvioli, M., Rossi, M., Baum, R.L., Godt, J.W., Guzzetti, F., 2014. Improving predic-
tive power of physically based rainfall-induced shallow landslide models: a probabi-
listic approach. Geosci. Model Dev. 7, 495–514.
Rogers, N.W., Selby, M.J., 1980. Mechanisms of shallow translational landsliding during
summer rainstorms: North Island, New Zealand. Geogr. Ann. 62 (A), 11–21.
Rossi, M., Witt, A., Guzzetti, F., Malamud, B.D., Peruccacci, S., 2010. Analysis of historical
landslide time series in the Emilia-Romagna region, northern Italy. Earth Surf. Process.
Landf. 35 (10), 1123–1137.
Rosso, R., Rulli, M.C., Vannucchi, G., 2006. A physically based model for the hydrologic
control on shallow landsliding. Water Resour. Res. 42, W06410. http://dx.doi.org/
10.1029/2005WR004369.
Schmidt, K.M., Roering, J.J., Stock, J.D., Dietrich, W.E., Montgomery, D.R., Shaub, T., 2001.
Root cohesion variability and shallow landslide susceptibility in the Oregon Coast
Range. Can. Geotech. J. 38, 995–1024.
Selby, M., 1993. Hillslope Materials and Processes. Oxford University Press, Oxford.
Simoni, S., Zanotti, F., Bertoldi, G., Rigon, R., 2008. Modeling the probability of occurrence
of shallow landslides and channelized debris flows using GEOtop-FS. Hydrol. Process.
22, 532–545.
Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science 240, 1285–1293.
54 M.S. Kim et al. / Geomorphology 271 (2016) 40–54
Talebi, A., Troch, P.A., Uijlenhoet, R., 2008. A steady-state analytical hillslope stability
model for complex hillslopes. Hydrol. Process. 22, 546–553.
Tarboton, D.G., 1997. A new method for the determination of flow directions and contrib-
uting areas in grid digital elevation models. Water Resour. Res. 33, 309–319.
Tarolli, P., Borga, M., Fontana, G.D., 2008. Analysing the influence of upslope bedrock out-
crops on shallow landsliding. Geomorphology 93, 186–200.
Tromp-van Meerveld, H.J., McDonnell, J.J., 2006. Threshold relations in subsurface
stormflow: 2. The fill and spill hypothesis. Water Resour. Res. W02411. http://dx.
doi.org/10.1029/2004WR003800.
Uchida, T., Tromp-van Meerveld, I., McDonnell, J.J., 2005. The role of lateral pipe flow in
hillslope runoff response: an intercomparison of non-linear hillslope response.
J. Hydrol. 311, 117–133.
Uchida, T., Mori, N., Tamura, K., Takiguchi, S., Kamee, K., 2009. The role of data preparation
on shallow landslide prediction. J. Jpn. Soc. Erosion Control. Eng. 62, 23–31 (in
Japanese).
Uchida, T., Tamura, K., Akiyama, K., 2011. The role of grid cell size, flow routing algorithm
and spatial variability of soil depth on shallow landslide prediction. Ital. J. Eng. Geol.
Environ. Book http://dx.doi.org/10.4408/IJEGE.2011-03.B-018.
Van Den Eeckhaut, M., Vanwalleghem, T., Poesen, J., Govers, G., Verstraeten, G.,
Vandekerckhove, L., 2006. Prediction of landslide susceptibility using rare events lo-
gistic regression: a case study in the Flemish Ardennes (Belgium). Geomorphology
76, 392–410. http://dx.doi.org/10.1016/j.geomorph.2005.12.003.
Webster, R., Oliver, M., 1992. Statistical Methods in Soil and Land Resource Survey. Oxford
Univ. Press, Oxford.
Wu, W., Sidle, R.C., 1995. A distributed slope stability model for steep forested basins.
Water Resour. Res. 31, 2097–2110.
Yoshinaga, S., Ohnuki, Y., 1995. Estimation of soil physical properties from a handy dy-
namic cone penetrometer test. J. Jpn. Soc. Erosion Control. Eng. 48, 200 (in Japanese).
Zhu, Q., Lin, H.S., 2009. Simulation and validation of concentrated subsurface lateral flow
paths in an agricultural landscape. Hydrol. Earth Syst. Sci. 13, 1503–1518.
Zizioli, D., Meisina, C., Valentino, R., Montrasio, L., 2013. Comparison between different
approaches to modeling shallow landslide susceptibility. Nat. Hazards Earth Syst.
Sci. 13, 559–573.