Engineering Geology 301 (2022) 106616

Contents lists available at ScienceDirect

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

A novel approach to probabilistic seismic landslide hazard mapping using

Monte Carlo simulations
Chao Li a, Gongmao Wang b, Jianjian He a, Yubing Wang a, *
a
Center for Hypergravity Experimental and Interdisciplinary Research, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou 310058, China
b
MOE Key Laboratory of Soft Soils and Geoenvironmental Engineering, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou 310058, China

A R T I C L E I N F O A B S T R A C T

Keywords: As a potential destructive secondary hazard initiated by earthquakes, earthquake-triggered landslides can cause
Probabilistic analysis significant losses. Seismic landslide hazard maps are important for land use planning and landslide hazard
Seismic landslide reduction in tectonically active areas. A comprehensive probabilistic seismic landslide hazard mapping approach
Hazard map
incorporating the uncertainties of slope properties and displacement prediction is required to promote the
Monte Carlo simulation
Critical slope angle
application of hazard maps. Based on the definition of the critical slope angle, this paper proposes an alternative
seismic landslide hazard category considering the uncertainty of geotechnical properties of soil using Monte
Carlo simulations and incorporating the ground motion variability based on the displacement hazard curve. The
proposed new hazard category is applied to the region of Anchorage, Alaska, for the seismic landslide hazard
mapping process. Subsequently, parameter analyses are performed to explore the sensitivity among the factors
that contributed to the seismic landslide hazard map. The results of seismic landslide hazard maps show that the
area of hazardous zone of most geologic units in present study is smaller than that in the deterministic analysis.
Comparing with the probabilistic method using logic tree analysis, the proposed approach using Monte Carlo
simulations predicts different landslide hazardous zone due to the probabilistic distribution of geotechnical
properties of soil. The predicted hazardous zone is also sensitive to the resolution of the digital elevation model
and different coefficient of variation combinations.

1. Introduction Vahedifard, 2013; Zhang et al., 2019). Therefore, many studies on

seismic landslide hazard mapping based on the concept of Newmark
As a destructive secondary hazard triggered by earthquakes, seismic displacement have been conducted in recent decades. The rigid sliding
landslides can cause significant damage to infrastructure and result in block method has been adopted by the United States Geological Survey
the loss of human lives. Seismic landslide zonation is commonly (USGS) and the California Geological Survey for mapping seismic
required for land use planning, recreational area management, and landslide hazards (e.g., Jibson et al., 2000; McCrink, 2001).
development of major infrastructure, such as highways and railways. As an early application, Jibson et al. (1998, 2000) proposed a
Many approaches, including logic regression and fuzzy logic system method to generate probabilistic seismic landslide hazard maps, in
analysis, have been introduced to produce seismic landslide hazard which the probability of failure was produced by comparing the deter­
maps (e.g., Lee et al., 2010; Miles and Keefer, 2009). In general, three ministic Newmark displacement with landslide inventories. Subse­
methods have been broadly adopted to evaluate seismic slope stability quently, similar procedures were applied to case studies with
with explicit physical concepts, including the pseudo-static analysis modifications to empirical displacement models (e.g. Bojadjieva et al.,
based on the static limit equilibrium theory, stress deformation analysis 2017; Gallen et al., 2017; Liu et al., 2017; Niño et al., 2014; Tao et al.,
based on numerical simulations, and permanent displacement analysis 2013; Wang and Lin, 2009; Zhao and Song, 2012) or were combined
extended from Newmark’s method (Jibson, 2011). Among them, the with logistic regression to consider the regional features (Wang et al.,
permanent displacement is regarded as an excellent index for the seismic 2016). To overcome the limitations of the infinite slope assumption
performance of a slope, and a series of empirical displacement models adopted in previous studies, many attempts have been made (e.g. Saade
were developed using different sliding assumptions (e.g., Meehan and et al., 2016; Tsai et al., 2019; Peng et al., 2009; Shinoda and Miyata,

* Corresponding author.
E-mail addresses: [email protected] (C. Li), [email protected] (G. Wang), [email protected] (J. He), [email protected] (Y. Wang).

https://doi.org/10.1016/j.enggeo.2022.106616
Received 27 September 2021; Received in revised form 17 February 2022; Accepted 8 March 2022
Available online 15 March 2022
0013-7952/© 2022 Elsevier B.V. All rights reserved.
C. Li et al. Engineering Geology 301 (2022) 106616

2017). However, these mapping methods of seismic landslide hazards analytical approach should be sufficient to consider other causative
are still considered deterministic approaches because a single empirical factors.
displacement model and a set of specific input parameters associated This study proposes an integral probabilistic method to promote the
with a slope are adopted in the calculation. That is, the epistemic un­ application of seismic landslide hazard maps that considers the
certainties of the slope properties and the aleatory variability of seismic epistemic uncertainties associated with slope properties using Monte
motions were not considered sufficiently. The influence of different Carlo simulation. The definition of the critical slope angle is described
displacement models and input parameters on the comparison between for a convenient application of region mapping incorporating the un­
observed and predicted landslides is significant (Dreyfus et al., 2013). A certainties of different empirical displacement models and the aleatory
robust method incorporating the uncertainty in strength parameters variability of seismic motions. Instead of using sliding displacement
should be developed. threshold, a seismic landslide hazard category defined by critical slope
Capolongo et al. (2002) and Refice and Capolongo (2002) applied angle was proposed in this study. Subsequently, such categories were
physically based models to probabilistic seismic landslide hazard applied to the regional mapping process of Anchorage, Alaska, for
assessment, considering the uncertainties of input parameters. Wang validation and comparison with previous works. The effects of proba­
et al. (2008) introduced a method to estimate the probabilities of bility distribution and digital elevation model (DEM) resolution are
displacement exceeding 10 cm, which were shown as a map of seismic discussed. Finally, a parametric analysis was conducted for the coeffi­
landslide hazards using Monte Carlo simulation. Saygili and Rathje cient of variation (COV) and unit weight of the soil.
(2009) proposed a probabilistic approach using the concept of
displacement hazard curves to consider both the variability of ground 2. Probability-based seismic landslide hazard mapping
motion and the uncertainty in the prediction of sliding displacement. procedures
Wang and Rathje (2015) extended the works of Saygili and Rathje
(2009) to consider the uncertainties within slope properties and In this section, the construction of a seismic landslide hazard map­
empirical displacement models using logic-tree analysis. The application ping procedure is elaborated by comprehensively expressing the un­
of probability-based seismic landslide hazard mapping procedures has certainties of relevant factors. In regional analysis, the rigid sliding
received increasing attention (Liu et al., 2016; Martino et al., 2018; model based on the assumption of an infinite slope model has been
Wang et al., 2020). As indicated by Wasowski et al. (2011), the widely used in earthquake-triggered landslide hazard assessment and
improvement of the regional-scale assessment of seismic landslide sus­ seismic landslide hazard maps. This method was implemented in the
ceptibility and hazard represents another top research hotspot owing to present analysis. Following the general mapping logic of existing liter­
their potential in guiding land use planning and the development of ature, this study includes three main steps for a region. A seismic
major lifelines. Hence, a thorough probability-based method is neces­ landslide hazard category was established after storing the DEM map
sary for seismic landslide hazard mapping. Meanwhile, the computation along with the topographic map in the ArcGIS platform, and then the
cost is best to be limited to a suitable range, and the extensibility of the mapping procedure was performed by combining the slope map and the

Fig. 1. Flowchart for seismic landslide hazard mapping procedures in the present study.

2
C. Li et al. Engineering Geology 301 (2022) 106616

hazard category. As shown in Fig. 1, the flowchart of the application of hypercube sampling (LHS) technique produced a realistic distribution
the proposed probability-based procedures for seismic landslide hazard from a relatively limited number of samples and allowed less loss of
zonation illustrates the main routine of the three steps. Among them, the accuracy, which led to efficient computation in the Monte Carlo simu­
seismic landslide hazard category based on the critical slope angle is the lation process (Refice and Capolongo, 2002). In this study, the slope
most important work in this study. It consists of three sections including angle α is taken as a controlling variable, ranging from 1◦ to 90◦ with 1◦
the yield acceleration statistical samples, weighted mean values of the increments, and then each input parameter with assumed probability
annual rate of exceedance, and critical slope angle distributions. Each distribution is sampled using the Monte Carlo simulation together with
part of this section is detailed as follows. the LHS technique in the same geologic unit. These statistical samples,
along with a controlled slope angle, are substituted into eqs. (1) and (2),
2.1. Monte Carlo simulation of yield acceleration and the majority of the yield acceleration values that form the yield
acceleration samples can be calculated simultaneously. Many groups of
This is the first step in the seismic landslide hazard category. The ky samples exist, each of which is related to a specific geologic unit and a
epistemic uncertainty of input parameters (i.e., geotechnical properties selected slope angle. Thus, the repeat procedure leads to many ky values
of soil) is associated with the process of yield acceleration calculation. for all slope angle ranges and for the entire region composed of some
Based on the concept of permanent displacement, which embodies the geologic units.
characterizations of earth slopes subjected to seismic environments; the
susceptibility of slopes under earthquake loading can be characterized 2.2. Weighted mean annual rate of exceedance
by an index of yield acceleration, as stated by Newmark (Newmark,
1965). The yield acceleration (ky) can be explicitly described as The displacement hazard curve has been demonstrated as an effec­
tive method to combine the displacement threshold of the described
ky = (FSstatic − 1)⋅sinα⋅g (1)
seismic slope performance with different seismic hazard levels, based on
the yield acceleration. It describes the seismic hazard level (mean annual
where g is the acceleration owing to gravity, α is the slope angle, and
rate of exceedance λD) for different displacement values x and was
FSstatic represents the static factor of safety of the slope deduced from an
considered as a fully probabilistic approach in permanent sliding
infinite slope model. The static factor of safety can be expressed as
displacement quantification with a specific hazard level. Wang and
′ ( )
Rathje (2015) detailed a complete description of the displacement
′
c tanφ γ
FSstatic = + 1− m w (2)
γtsinα tanα γ hazard curve (Section 2.4 associated with Eqs. 3–4 of Wang and Rathje
(2015)), and the relevant concept was applied in this study. As shown in
where c’ is the effective cohesion (kPa), φ’ is the effective friction angle, Fig. 2, one ky value corresponds to one displacement hazard curve,
γ is the unit weight of soil (kN/m3), γw is the unit weight of water (kN/ where λD(x) represents the mean annual rate of exceedance for a
m3), t is the sliding block thickness normal to the failure surface or displacement level of x cm. λD(x) increases with an increase in the yield
sliding surface in meters, and m is the proportion of the block thickness acceleration value given a selected displacement value (e.g., 5 cm in
saturated (saturation ratio). Eq. (2) implies that three elements Fig. 2 and λD(5) increases as ky increased from 0.1 g to 0.2 g). Un­
contributed to the overall stability analysis, which contains the effects of doubtedly, numerous yield acceleration samples from the Monte Carlo
cohesive strength (first item of the right of the equation), friction angle simulation above deduce lots of λD(x) values with a selected displace­
(second item of the right of the equation), and pore water pressure (the ment x under consideration within the same empirical displacement
last item of the right of the equation). model (e.g., RS09 model in Fig. 2).
A specific value of the yield acceleration of a slope can be calculated Owing to the importance of the displacement threshold (e.g., 1 cm, 5
from Eqs. 1–2 under a deterministic group of relevant parameters. cm, and 15 cm) for seismic landslide hazard zonation, a convenient
Therefore, the index of yield acceleration indicates the intrinsic char­ method that compromises the accuracy of dealing with numerous yield
acteristics of the slope that resist the effect of an earthquake. However, acceleration values and the annual rate of exceedance accompanied is to
uncertainties in slope properties play an important role in the quanti­ directly establish a regression relationship between ky and λD(x) for a
fication of seismic landslide hazard assessment. Ignoring it and selecting given threshold x. The regression relationship was detailed by Wang
a group of specific parameters would lead to a conservative or non- (2014) and Wang and Rathje (2015). In general, as shown in Fig. 2, after
conservative result for yield acceleration and landslide hazard assess­
ment. To consider the epistemic uncertainties associated with these
parameters (c’, φ’, γ, t, m), Wang and Rathje (2015) applied a logic-tree
analysis method in which the branches represent possible selections of
input parameters, and each node is a specific parameter with an assigned
weight value. Unlike logic-tree treatment, which selects discrete limit
values, Monte Carlo simulation is an efficient approach to address un­
certain distributed problems (Wang et al., 2008).
The proposed method adopts the Monte Carlo simulation method,
considering that each particular parameter is assumed to obey a
reasonable probability distribution. For example, Fenton and Griffiths
(2003 and 2004) promote a lognormal distribution suitable for geologic
parameters and cohesion while assuming an independent normal dis­
tribution for the friction angle with a truncated range. Cherubini (2000)
recommended a lognormal distribution suitable for simulation and
advise that paid more attention to the connection between c’ and φ’. The
unit weight of soil (γ) can be considered to follow a normal distribution.
A similar probability distribution is suggested by assigning the geo­
metric parameters of the slope (t and m), which adopt a uniform dis­
tribution. Any suitable distribution with relevant statistical parameters
can be used with no restriction imposed on the selection of the assumed Fig. 2. Displacement hazard curves for ky values between 0.1 and 0.2 g
probability distribution for the input parameters. In addition, the Latin (adapted from Wang, 2014).

3
C. Li et al. Engineering Geology 301 (2022) 106616

selecting a displacement threshold (e.g., x = 5 cm) in the abscissa of the

plot of x and λD(x), a set of λD(x) values are obtained from the
displacement hazard curves associated with a selected suite of ky., which
is assumed to increase with a small increment (e.g., 0.01 g) from a small
ky value. Therefore, dozens of pairs of data (λD(x), ky) were regressed to
deliver an empirical formulation for the target which directly connects
the yield acceleration with the annual rate of exceedance. The rela­
tionship between ky and λD(x) can be obtained for each empirical
displacement model and each displacement threshold in the study re­
gion. A general form of regression formulation with a 4th order poly­
nomial that properly fits with a suite of data (λD(x), ky) is expressed as
follows (Wang, 2014)
( ( ) )4 ( ( ) )3 ( ( ) )2 ( )
ln(λD (x) ) = a1 ln ky + a2 ln ky + a3 ln ky + a4 ln ky + a5 (3)

where a1 to a5 are coefficients of the regression model changed over

empirical displacement models and displacement thresholds (different
empirical displacement models or displacement thresholds correspond
to a different group of coefficients, as shown in Table 4). Eq. 3 leads to
the convenient acquisition of a series of λD(x) samples by substituting
into the ky samples computed using the Monte Carlo method. Thus, the
mean value for these annual rates of exceedance samples is obtained
from the arithmetic average sampling values of λD(x) above. Every mean
value of the λD(x) samples, which is related to the empirical displace­
ment model, are assigned a specific weight value that considers the
uncertainties of different empirical displacement models. A weighted
average of these arithmetic mean λD(x) values for all possible empirical
displacement models was calculated under a given displacement
threshold. Thus, the weighted mean annual rate of exceedance can be
computed by controlling parametric variables slope angle (α) and
geologic unit (k) as follows (Wang and Rathje, 2015)
∑
n
λD (x, α, k) = ωi ⋅λD (x, α, k)i (4)
i=0

where λD (x; α, k) represents the weighted mean annual rate of exceed­

ance for displacement threshold (x) and shows different results for
various pairs of slope angle (α) and geologic unit (k), λD(x; α, k)i repre­
sents the arithmetic mean value of the annual rate of exceedance sam­
ples for the ith empirical displacement prediction model, and ωi is the
corresponding weight of this empirical model. n is the total number of
empirical displacement models applied in this study. The weighted
mean value considers the difference between different displacement
prediction models.
Fig. 3. Weighted mean annular rate of exceedance curves for different geologic
units (G-1, G-2, and G-3 represent three illustrative geologic units).
2.3. Critical slope angle

As expressed in Eq. 4, within the same geologic unit (k) along with a λD (x) equaled the target hazard level λ*. Alternatively, the critical slope
specific displacement threshold (e.g., x = 5 cm), the value λD (x; α, k) angle is taken as the closest abscissa if only the target hazard level de­
represents a type of correspondence between slope angle(α) ranging viates from the discrete value of λD (x). Taking the geologic unit G-1 in
from 1◦ to 90◦ and the weighted mean annual rate of exceedance. That Fig. 3 as an example, α*5 cm equals 9◦ and 21◦ according to two types of
is, one slope angle (α) corresponded to one λD (x) provided the selected hazard levels in the threshold x = 5 cm, whereas the critical
slope angle α*15 cm approaches 22◦ and 30◦ in threshold x = 15 cm,
displacement threshold x is determined. Once the value λD (x; α, k)
respectively. Therefore, the critical slope angle (α*x), corresponding to
equals a target hazard level λ*(e.g., λ* = 0.000404 1/year or λ* =
different displacement thresholds (x), for all geological units of a region
0.0021 1/year represent 2% probability and 10% probability of ex­
can be obtained under the selected hazard level λ* with the following
ceedance in 50 years, respectively), the corresponding slope angle is
analysis procedures. The magnitude of the critical slope angle represents
defined as the critical slope angle (α*). Figs. 3 (a) and (b) show that
the levels of the seismic landslide hazard and works the same function
λD (x; α, k) varied with an increase in slope angle (α) for three illustrative
equivalent to the displacement threshold under a conditional hazard
geologic units (G-1, G-2, and G-3) under two different displacement
level. This suggests that the aforementioned critical slope angle can be
thresholds (i.e., x = 5 cm and x = 15 cm). The critical slope angle is
used for seismic landslide hazard mapping applications, similar to the
regarded as the abscissa value, which corresponds to the intersection
performance of the displacement threshold.
point of the horizontal target hazard level line and geologic unit λD
curve, considering the situation in which the slope angle varies with
discrete increments (increment set as 1◦ ). There are two situations 2.4. Seismic landslide hazard category
regarding the determination of the critical slope angle from the inter­
section point. First, the critical slope angle is the abscissa value when In regional analysis or seismic landslide hazard mapping, instead of

4
C. Li et al. Engineering Geology 301 (2022) 106616

exact predictions of sliding displacement, landslide hazard categories proposed procedures is the derivation of equivalent seismic landslide
are more applicable. When introducing critical slope angles corre­ hazard categories and how the hazard category works in this region.
sponding to different displacement thresholds and to a selected geologic
unit, a modified assessment criterion for seismic landslide hazards can 3.2. Yield acceleration simulation
be obtained. The proposed seismic landslide hazard categories can be
applied to create seismic landslide hazard maps in the process of land As stated in Section 2, the first preparation is the samples of yield
planning and preliminary evaluation of potential seismic slope failures. acceleration that require a series of input parameters. Seventeen
As shown in Table 1, the middle column represented a common seismic geologic units of this study area along with the assigned shear strength
landslide hazard category used by USGS which is illustrated by parameters are listed in Table 2, whose values are considered as the
displacement thresholds (x = 1 cm, 5 cm, and 15 cm) and the right expected values. The four geologic units (bc, c-bl, l, and s) only provide
column is the equivalent seismic landslide hazard evaluation criterion the value of cohesion; therefore, these parameters are regarded as un­
which is established from the concept of critical slope angle (α*1 cm, α*5 drained shear strength, whereas other groups of values with cohesion
cm, and α*15 cm) of the present study in the light of previous sections. and friction angles are considered as effective strength parameters. The
Each location can be assigned to a seismic landslide hazard zone by selection of the COV is consistent with the previous work of Wang and
combining the critical slope angle with the discrete values of the slope Rathje (2015). For example, 30% is selected for undrained shear
map developed from a DEM. For instance, a realistic slope angle of a strength (COV-undrained is 0.3 for bc, c-bl, l, and s), 10% was chosen as the
specified site located between α*1 cm to α*5 cm is defined as a moderate effective friction angle (COV-friction was 0.1), and 20% was assigned as
hazard, meaning the co-seismic sliding displacement of this grid cell is effective cohesion (COV-cohesion was 0.2). The mean value of the strength
located between 1 and 5 cm. The entire area landslide hazards under a parameters along with the selected COV above comprises a basic group
specific earthquake hazard level can be determined by comparing the of statistical parameters, which is known in advance in the present
real slope angle of each grid cell of geologic units with the corresponding study. Subsequently, universal probability distribution forms are
critical slope angle. assumed for strength parameters based on a consistent group of ex­
pected values and corresponding COVs. In this study, strength parame­
3. Application of the proposed approach in a study region of ters (c’, φ’) are considered to obey lognormal distributions (Fenton and
Anchorage, Alaska Griffiths, 2003 and 2004). Meanwhile, an independent normal distri­
bution along with a joint normal distribution with an assumed correla­
3.1. Relevant works tion coefficient (ρ = − 0.5) (Alamanis and Dakoulas, 2019) are assigned
to the strength parameters for a better sensitivity comparison. A more
Anchorage, Alaska, was selected as the case study for applicability rigorous probability distribution should be used in the simulation of
verification based on the availability of the required data and the geologic parameters, which is worthy of further study and is beyond the
accessibility of databases of soil properties in such areas. As a repre­ scope of this study. Based on the features of geologic units consisting of
sentative application, Jibson and Michael (2009) produced seismic different types of soils and on the relevant statistical results of shallow
landslide hazard maps for this region based on a deterministic approach soils of the region by Wang (2014), the sliding block thickness (t) is
that used a series of typical parameter values including shear strength,
sliding block thickness, and saturation ratio to compute yield accelera­
Table 2
tion and introduced a displacement prediction model for the calculation
Geologic units and strength parameters (adapted from Wang and Rathje, 2015).
of Newmark displacement. Thus, the corresponding seismic landslide
Unit Cohesion Friction Unit description
hazard of a site was determined by comparing the calculated displace­
angle
ment with the displacement thresholds. Wang and Rathje (2015) pro­
posed a probabilistic procedure for the mapping of co-seismic landslide af 24 36 Deposits in alluvial fans, alluvial cones, and
emerged deltas
hazards in this region, improving the limitations of the deterministic
al 19 36 Alluvium in abandoned stream channels and in
method above. The aleatory variability in the sliding displacement terraces along modern streams
prediction and the epistemic uncertainty of input parameters and an 21 36 Coarse-grained surficial deposits
different displacement empirical models were considered in their map­ b 192 40 Bedrock
bc 120 0 Bootlegger Cove clay
ping process with the help of logic tree analysis, and a mean λD threshold
c-br 38 38 Colluvium derived from bedrock on slopes of the
approach was applied for seismic landslide hazard zonation in the study Chugach Mountains
area. c-bl 38 0 Colluvium derived from glacial materials along
Considered as an improvement to previous studies, this study is coastal bluffs
applied to the same study area for a better comparison. The basic f 48 34 Human made fills
ga 38 32 Glacial outwash in irregularly shaped hills
datasets that contributed to seismic landslide hazard mapping were
(including kames, eskers, and kame terraces)
mainly topographic maps, DEM of the study area, slope map derived gm 48 38 Glacial and (or) marine deposits, typically in
from DEM, and geologic unit map connected with input parameters to elongated hills
calculate yield acceleration. A relatively detailed description of such l 144 0 Lake and pond deposits
ls 24 30 Landslide deposits, similar to a unit
basic datasets was accessed from previous studies. The central work
m 43 38 Morainal deposits, generally in long ridges
contributing to the seismic landslide hazard mapping application of the marking the margins of former glaciers
mg 38 37 Marine, glacial, and (or) lacustrine deposits
s 72 0 Silt
Table 1 sh 24 34 Sand deposits in broad, low hills, and windblown
sand deposits in cliff-head dunes near Point
Seismic landslide hazard categories.
Campbell
Hazard category Sliding displacement (D) Critical slope angle (α*) sl 19 34 Sand deposits in a wide low-lying belt around
(used by USGS) (present approach) Connors Lake
Low 0–1 cm 0-α*1cm (a) The first column represents the abbreviation of the geologic unit description.
Moderate 1–5 cm α*1cm –α*5cm (b) The specific value is assumed to be the expectation in probability distribu­
High 5–15 cm α*5cm –α*15cm
tions.
Very high >15 cm >α15cm
(c) Cohesion in the unit of Kilopascal (kPa).
Critical slope angle in the unit of degree. (d) Friction angle in the unit of degree.

5
C. Li et al. Engineering Geology 301 (2022) 106616

assumed to follow a uniform distribution with different representative Table 4

values for different geologic units (thinner thickness (2 to 4 m) for Statistical parameters of ky samples for all geologic units with two probability
geologic units bc, c-bl, l, and s with zero friction angle, whereas the other distribution.
geologic units have thicker thicknesses (3 to 9 m)). The saturation ratio Items Lognormal Normal
(m) satisfies the uniform distribution sequence in (0,1). Mean value Standard deviation Mean value Standard deviation
The unit weights of soil and water are constant and are assigned
af 0.205 0.140 0.206 0.140
values of 18.8 kN/m3 and 9.8 kN/m3, respectively. The effect of the
al 0.161 0.125 0.161 0.124
variability of the unit weight of soil is discussed below, and the results an 0.205 0.140 0.206 0.140
show that it has almost no influence on the mapping outcomes. The b 1.911 0.731 1.912 0.735
selected distribution patterns for the input parameters are listed in bc 0.683 0.528 0.676 0.518
Table 3. By performing a Monte Carlo simulation and the LHS technique c-br 0.374 0.190 0.374 0.188
c-bl 0.230 0.236 0.228 0.222
for the input parameters (c’, φ’, t, m) of every geologic unit based on the f 0.402 0.209 0.402 0.209
probabilistic assumptions of Table 3, a series of ky samples can be ga 0.273 0.171 0.273 0.170
calculated according to eqs. (1) and (2). These random samples of the gm 0.472 0.219 0.471 0.219
input parameters consist of 100,000 numbers in each simulation, which l 0.914 0.632 0.907 0.627
ls 0.119 0.109 0.119 0.108
leads to a realistic probability distribution (e.g. Refice and Capolongo,
m 0.423 0.205 0.423 0.204
2002). Only a few minutes are required in the MATLAB platform for all mg 0.356 0.186 0.356 0.186
of these input random samples, which combine to directly calculate s 0.242 0.287 0.239 0.278
sufficient ky samples without external uncertainty, except for the input sh 0.174 0.131 0.174 0.129
parameters. sl 0.131 0.112 0.131 0.111

In Table 4, the resulting statistical parameters of the yield accelera­

tion sample associated with the lognormal and independent normal
distributions are illustrated for 17 geologic units when the controlled
slope angle is 30◦ . As a result of the probabilistic calculation by Monte
Carlo simulation in this study, yield accelerations are distributed over a
large range for each geologic unit at every single slope angle, which is
superior to the deterministic result with a single value or to the logic
result with some discrete values. As illustrated, the values of both mean
and standard deviation in the lognormal are larger than those in the
normal distribution for geologic units bc, c-bl, l, and s. Only one value of
mean and standard deviation deviated by approximately 0.001 for other
geologic units. However, the overall deviations in Table 4 are small for
the two assumed probability distributions with respect to the mean
value and corresponding standard deviation. Owing to the truncation
operation of the ky computation which requires ky to be equal to 0.01 g
as soon as the resulting yield acceleration is less than 0.01 g or negative
(e.g., Wang, 2014), the final distribution of ky did not exhibit a signifi­
cant difference as shown in Table 4. The truncation operation of the
yield acceleration distributions of the independent normal distribution
with a slope angle of 30◦ for geologic unit bc is shown in Fig. 4.

3.3. Mapping process

Fig. 4. Yield acceleration distributions before truncation.
Following the yield acceleration samples above, the critical slope
distribution and corresponding seismic landslide hazard category are each empirical model were developed by Wang (2014), and are listed in
deduced in this section for final regional hazard maps by further Table 5. Thus, the mean value of the annual rate of exceedance λD(x; α,
incorporating the epistemic uncertainty. The regression formulation k)i associated with the empirical displacement model and the subse­
between ky and λD(x) is favorable when facing many yield acceleration quent weighted mean annual rate of exceedance can be obtained using
samples. Four types of empirical displacement models are applied in this eqs. (3) and (4). The critical slope angles, including three selected
study considering the epistemic uncertainty associated with the empir­ displacement thresholds (x = 1 cm, 5 cm, and 15 cm), for all geologic
ical displacement model, which applies for the rigid block sliding model units over the region were obtained by comparing λD (x; α, k) curves with
and were also used by Wang and Rathje (2015) (RS09 (Rathje and the target hazard level (λ* = 0.000404 1/year and 0.0021 1/year in this
Saygili, 2009), J07 (Jibson, 2007), BT07 (Bray and Travasarou, 2007) study). As shown in Table 6, the critical slope angle distribution along
and SR08 (Saygili and Rathje, 2008)). The corresponding regression with two selected hazard levels, which include three assumed distribu­
coefficients a1 to a5 of eq. (3) for this region and assigned weights (wi) of tion patterns, are listed for 17 geologic units corresponding to the
displacement threshold x = 5 cm (similar tables of critical slope angle
Table 3 distributions are obtained for the other two displacement thresholds and
Distribution patterns of input parameters in the present study. are not shown here). The critical slope angle in the lognormal distri­
Parameter Distribution type Remarks
bution for some geologic units (bc, c-bl, l, and s) is larger than that in the
other two probabilistic distributions, and it shows no more than
Cohesion c ′
Normal/Lognormal
σ = μ ∙ COV 1◦ discrepancy for other geologic units. The difference phenomenon of
Friction angle φ′ Normal/Lognormal
c′ and φ′ Joint Normal ρ = − 0.5 the critical slope angle distribution over the geologic unit is listed
t~U(2, 4) for (bc、c-bl、l、s) consistently with the ky distribution in Table 4.
Block thickness t Uniform
t~U(3, 9) for other With the help of the critical slope angle distribution, a critical
Saturation ratio m Uniform U(0, 1)
component of the proposed hazard mapping procedures-seismic land­
μ and σ represent expectation and standard deviation, respectively. slide hazard category is generated. As listed in Table 7, the integrated

6
C. Li et al. Engineering Geology 301 (2022) 106616

Table 5
Regression coefficient and corresponding weights for different models (adapted from Wang, 2014).
Displacement threshold Empirical model a1 a2 a3 a4 a5 wi

1 cm RS09 0.0067 0.0673 − 0.2122 − 3.5961 − 10.7348 0.22

J07 0.0089 0.1051 − 0.0672 − 3.7699 − 11.8035 0.22
BT07 0.0116 0.1377 0.0984 − 3.2727 − 11.0939 0.22
SR08 − 0.0024 − 0.0088 − 0.3859 − 3.7557 − 11.4605 0.34
5 cm RS09 0.0044 0.0529 − 0.1717 − 3.4082 − 11.3582 0.22
J07 − 0.0013 0.0061 − 0.3379 − 4.1157 − 13.0937 0.22
BT07 − 0.0036 − 0.0161 − 0.3892 − 3.9692 − 12.2431 0.22
SR08 − 0.0139 − 0.1396 − 0.8389 − 4.3842 − 12.7071 0.34
15 cm RS09 − 0.0029 − 0.0212 − 0.3917 − 3.6514 − 12.1558 0.22
J07 − 0.0224 − 0.2211 − 1.1387 − 5.3158 − 14.6797 0.22
BT07 − 0.0512 − 0.5404 − 2.35 − 7.037 − 15.0673 0.22
SR08 − 0.0397 − 0.4348 − 2.0095 − 6.3164 − 14.6476 0.34

Table 6
Critical slopes for all geologic units with selected hazard level (x = 5 cm).
Distribution pattern Hazard level af al an b bc c-br c-bl f ga gm l ls m mg s sh sl

Normal 2% 19 18 19 63 9 25 4 25 21 28 13 16 27 25 3 18 16
10% 24 22 24 76 20 30 12 30 26 33 26 20 31 30 11 23 21
Lognormal 2% 19 18 20 69 20 25 12 26 22 29 25 16 27 25 10 18 16
10% 24 22 24 80 27 30 17 31 27 33 33 20 32 29 14 23 21
Joint Normal 2% 20 19 20 66 9 27 4 27 23 30 13 16 28 26 3 19 17
10% 25 23 25 78 21 31 12 32 27 35 26 21 33 30 11 24 22

The symbol 2% represents a 2% probability of exceedance in 50 years and 10% represents a 10% probability of exceedance in 50 years.

percentage of each hazard level can be summarized by accounting for

Table 7
the corresponding raster cells. The entire process mentioned above can
Seismic landslide hazard category classified by critical slope angles (2% prob­
be implemented readily using the computation platform ArcGIS.
ability of exceedance in 50 years).
Category Low Moderate High Very high
3.4. Results of seismic landslide hazard mapping
af 0–16 16–19 19–22 >22
al 0–14 14–18 18–20 >20
an 0–16 16–20 20–22 >22 Fig. 5 shows the corresponding maps of landslide hazard with a DEM
b 0–66 66–69 69–72 >72 resolution of 6 m, which is consistent with previous studies. The
bc 0–17 17–20 20–22 >22 explanation for hazard category (i.e., moderate, high and very high) in
c-br 0–22 22–25 25–28 >28
the following figures and tables can be found in Table 1. Table 8 shows
c-bl 0–8 8–12 12–14 >14
f 0–23 23–26 26–28 >28 the resulting distributions of geologic units whose grid cells are
ga 0–18 18–22 22–24 >24 considered as high and very high hazards (permanent sliding displace­
gm 0–26 26–29 29–31 >31 ment D > 5 cm or realistic slope angle α > α*5 cm). It also provides
l 0–23 23–25 25–28 >28 analytical results that are similar to those of previous studies. The area
ls 0–12 12–16 16–18
proportion represents the percentage of the region covered by each
>18
m 0–24 24–27 27–29 >29
mg 0–22 22–25 25–27 >27 geologic unit. Unit proportion is the total percentage of grid cells
s 0–6 6–10 10–12 >12 considered as high and very high seismic landslide hazard within each
sh 0–15 15–18 18–21 >21 geologic unit, and contribution proportion reflects the landslide
sl 0–13 13–16 16–19 >19
contribution of each geologic unit to the overall study region, which
indicates the quantities of grid cells with a realistic slope angle over the
seismic landslide hazard category of all geologic units for this region, critical slope angle associated with the displacement threshold. In this
which specifies the target hazard level λ* = 0.000404 1/year and study, three geologic units c-br, c-bl, and ls are estimated to be most
lognormal distribution for c’ and φ’, are classified into four hazard levels susceptible to landslides with the highest percentage of unit proportion,
(low, moderate, high, and very high). Seismic landslide hazard cate­ and the total contribution proportion of these three units is close to
gories of similar forms for different hazard levels and probability dis­ approximately 58% while occupying only approximately 8% of the
tributions are anticipated. entire study area. In addition, because of the highest area proportion,
The prepared seismic landslide hazard categories and basic datasets the geologic unit (al) contributes 10.2% to the overall quantity of
above were used to generate probabilistic seismic landslide hazard maps landslides, although only 1.2% of this geologic unit is regarded as high
for the study area of Anchorage. Hence, the region is divided into mil­ and very high hazard. Wang and Rathje (2015) showed a lower value in
lions of grid cells whose cell sizes are on the scale of meters and are the unit proportion along with a higher contribution level; however, an
stored as raster datasets, and each grid cell is assigned a realistic slope opposing situation appeared in the results of Jibson and Michael (2009).
angle from the slope map. According to the proposed mapping proced­ The abnormal lowest value for the geologic unit c-br was less than 0.1%
ures, the landslide hazard grade of each location can be determined by in the logic-tree analysis in contrast to a higher level in this study and
directly comparing the specific slope angle with the critical slope angle deterministic method. Meanwhile, the geologic unit c-bl was predicted
of seismic landslide hazard categories. An overall seismic landslide as the most landslide-susceptible unit in the deterministic analysis, and
hazard zone can be described by summarizing the results grid by grid. the corresponding unit proportion reached 93%, which is much higher
Different seismic landslide hazard maps are categorized by different than other results. The significant differences among the three analytical
hazard levels (low, moderate, high, and very high hazards). The methods for the geologic units c-br and c-bl might result from the se­
lection of the sliding block thickness (t) and saturated ratio (m) and the

7
C. Li et al. Engineering Geology 301 (2022) 106616

comparison, the differences between the deterministic result and this

study are influenced by the variability of parameters, epistemic uncer­
tainty associated with empirical models, and the displacement hazard
curve that is used to consider the uncertainty of seismic motions. The
ranges of the current results for the unit proportion and contribution
proportion are mostly centered on the works of Wang and Rathje (2015)
and Jibson and Michael (2009), according to comparisons within the
same geologic unit.
The strength parameters (c’ and φ’) play an important role in the
yield acceleration calculation described using eq. (2). Consequently,
different probability distribution assumptions may result in different
results for seismic landslide hazard maps in terms of the statistical
consequences of the total landslide hazard zone. Table 9 shows the three
probabilistic distribution results of the percentage of the study area
regarded as low, moderate, high, and very high hazards, respectively,
which considered two typical hazard levels in this study. The results of
the lognormal distribution are mainly discussed in this study, and in­
dependent and joint normal distribution are supplemented as referenced
results. The percentage of landslide zones with D > 5 cm under high
seismic hazard level (2% probability of exceedance in 50 years) is almost
twice as high as the low hazard level (10% probability of exceedance in
50 years), within the same distribution model. For example, the per­
centage of lognormal distribution increased from 1.29% to 2.59%, and
that of joint normal distribution varied from 1.30% to 2.73%. The lowest
results are from the lognormal distribution, meaning a relatively safer
environment; however, the independent normal distribution shows
contrasting situations in which more percentage grid cells of the study
area are predicted to be landslide-susceptible. The middle one is the
joint normal distribution results that indicate a closer relationship with
the lognormal distribution when selecting a low hazard level. The dif­
ference between normal and joint normal distribution reflects the effect
of the correlation between cohesion and friction angle on the assessment
of seismic landslide hazards, and it is aggravated as the hazard level
increases, such as 0.17%, which increases to 0.46% corresponding to the
transformation of λ* = 0.00211 1/year descending to λ* = 0.000404 1/
year. In addition, combined with the results of Table 6, the reason why
the percentage of the lognormal is less than the other two probability
distributions result from such geologic units (bc, c-bl, l, and s), which
contain a large critical slope angle and a greater contribution propor­
tion. Only a few deviations for ky in Table 4 lead to more differences in
the critical slope angle in Table 6, and it further led to a large discrep­
ancy in the percentage of the area estimated as a high hazard level in
Table 9. This confirms that different probability distributions partially
influence the final accuracy of seismic landslide hazard mapping.
The use of a higher DEM resolution to create topographic maps re­
sults in steeper slopes, and the effect of resolution on slopes increases
significantly for slope angles above 40◦ (Wang, 2014). Wang et al.
(2017) discussed the influence of DEM resolution on the assessment of
seismic landslide occurrence and suggested that higher resolution pro­
vides higher accuracy at local convex topographies such as scarps and
ridge crests. However, a higher DEM resolution will detect minor fea­
tures that do not necessarily represent a landslide hazard for the infinite
Fig. 5. seismic landslide hazard maps showing three hazard categories slope model. This means the mapping results of seismic landslide haz­
(referred to Table 1) with 6 m DEM resolutions. ards not only depend on the assigned shear strengths discussed above
but are also controlled predominantly by the distribution of realistic
neglect of variability of these parameters. Wang and Rathje (2015) chose slope angles. As shown in Table 10, the value of the seismic landslide
a specific value for t = 3 m and m = 0.0 and attained a quite low per­ hazard level above the moderate hazard associated with higher a reso­
centage for unit c-br, whereas Jibson and Michael (2009) selected a lution (1.5 m) is almost twice the value of the rough resolution (6 m). For
block thickness of 15 m and a saturation ratio of 0.8 for these units, instance, 4.79% decreases to 2.59% for landslide hazard D > 5 cm,
which showed the highest value for unit c-bl. A lower ky resulting from a which corresponds to a change in resolution from 1.5 to 6 m. The in­
higher t and a higher m lead to an overestimated displacement, and a crease in hazard zone percentage of D > 5 cm (varied from 1.76% to
lower m or lower t contributes to the contrasting result. The results of 2.38%); owing to the effect of the resolution, it shows a growing ten­
this study and those of the logic tree analysis differ because of the dency with increasing hazard level. Figs. 5 and 6 show the predicted
simulation method employed for the consideration of epistemic uncer­ seismic landslide hazard maps of the two DEM resolutions in the ArcGIS
tainty of input parameters in the calculation process of ky samples. As a platform, respectively. A significant increase in the number of landslide
areas is explicitly shown by comparing the detailed views. This indicates

8
C. Li et al. Engineering Geology 301 (2022) 106616

Table 8
Distribution and comparison of high and very high seismic landslide hazard zone over geologic units of Anchorage, Alaska under 2% probability of exceedance in 50
years.
Geo units Area This study (Wang and Rathje, 2015) (Jibson and Michael, 2009)
proportion
Unit Contribution Unit Contribution Unit Contribution
proportion proportion proportion proportion proportion proportion

af 11.4% 2.1% 9.0% 1.1% 9.7% 2.2% 6.1%

al 22.0% 1.2% 10.2% 0.6% 10.5% 1.3% 7.0%
an 9.8% 0.5% 1.8% 0.3% 2.3% 0.6% 1.4%
b 2.2% 0.0% 0.0% 0.0% 0.0% 0.4% 0.2%
bc 2.8% 1.2% 1.2% 1.3% 2.8% 1.9% 1.2%
c-br 6.1% 13.2% 30.8% <0.1% 0.1% 18.9% 27.4%
c-bl 1.3% 38.2% 18.7% 34.7% 34.0% 93.0% 28.2%
f 3.0% 0.2% 0.3% 0.1% 0.3% 0.7% 0.5%
ga 5.3% 3.7% 7.5% 2.6% 10.7% 7.7% 9.6%
gm 3.9% 0.3% 0.4% 0.2% 0.5% 0.9% 0.8%
l 1.8% 0.2% 0.2% 0.3% 0.4% 0.7% 0.3%
ls 1.0% 19.4% 7.8% 15.0% 12.1% 22.4% 5.6%
m 6.0% 1.3% 3.0% 0.7% 3.0% 3.0% 4.2%
mg 9.5% 0.4% 1.3% 0.2% 1.5% 0.8% 1.8%
s 2.0% 1.4% 1.1% 1.8% 2.7% 2.8% 1.4%
sh 2.1% 5.6% 4.6% 4.2% 6.8% 5.8% 3.0%
sl 10.0% 0.6% 2.3% 0.3% 2.4% 0.5% 1.3%

(a) Area proportion represents the percentage of the region covered by each geologic unit.
(b) Unit proportion represents the percentage of grid cells considered as high and very high hazards within each geologic unit.
(c) Contribution proportion represents the contribution of each geologic unit to all the number of D > 5 cm grid cells in this region.

should be incorporated as well. Phoon and Kulhawy (1999a) noted that

Table 9
the unit weight of soil is approximately 13–20 kN/m3 and the COV is less
Distribution of four seismic landslide hazard levels of the region considering the
than 10% (9% is advised), which is the similar range of COV adopted by
effect of probability distribution.
Cherubini (2000), which is 3%–10% for unit weight. Hence, an expected
Hazard Distribution Low Moderate High Very D>5
value of 18 kN/m3 is assigned to the soil, and the corresponding COV
level high cm
value is taken as 9% to explore the influence of the unit weight of soil on
Lognormal 95.73% 1.67% 0.79% 1.81% 2.59% the critical slope angle distributions over the overall geologic units and
Normal 94.98% 1.84% 0.90% 2.28% 3.19%
2%
Joint
the corresponding seismic landslide hazard maps. Based on assump­
95.54% 1.73% 0.80% 1.94% 2.73% tions, the unit weight of the soil is a random variable that follows the
Normal
Lognormal 98.27% 0.44% 0.32% 0.97% 1.29% common independent normal distribution. A similar process was per­
10%
Normal 98.05% 0.48% 0.39% 1.08% 1.47% formed for the calculation of critical slope angles over all geologic units
Joint
98.27% 0.43% 0.33% 0.97% 1.30% in the region.
Normal
Fig. 7 shows the resulting critical slope angle distributions for all
geologic units with a 5 cm displacement threshold and two seismic
hazard levels. As shown in the figure legend, the constant value repre­
Table 10 sents a fixed value of 18.8 kN/m3, and the variational value incorporates
Distribution of four seismic landslide hazard levels of the region considering the the uncertainty of the unit weight of soil. There are few effects of the
effect of DEM resolution. variation in the unit weight of soil on the critical slope angle. A large
Hazard Resolution Low Moderate High Very D>5 proportion of consistent results are indicated before and after the
level high cm incorporation of the COV of the unit weight of soil over 17 geologic
6m 95.73% 1.67% 0.79% 1.81% 2.59% units. Therefore, a single representative value, within a common range
2%
1.5 m 92.94% 2.09% 1.08% 3.89% 4.97% from the viewpoint of engineering, for the unit weight of soil is valid to
10%
6m 98.27% 0.44% 0.32% 0.97% 1.29% calculate the critical slope angle and create seismic landslide hazard
1.5 m 96.28% 0.67% 0.54% 2.51% 3.05%
maps. The influence on the assessment of landslide occurrence is
negligible compared with other parameters such as c’, φ’, and t.
that a higher DEM resolution might lead to a higher percentage of this
study area estimated as susceptible to landslides. However, this does not 4.2. Effect of COV Combination
mean that a higher DEM resolution yields better predicted results
because of the local convex terrain, which may lead to misjudgment. The results derived from the previous mapping procedure only pro­
Therefore, a suitable resolution of the DEM is important when requiring vide a deterministic combination of COVs (COV-undrained is 0.3, COV-
a high-precision seismic landslide hazard map. friction is 0.1, and COV-cohesion is 0.2). Therefore, the influence of different
COV combinations on the seismic landslide hazard mapping results is
4. Parameter analysis discussed here. Phoon and Kulhawy (1999a, 1999b) provided an
approximate guideline to characterize inherent soil variability, in which
4.1. Incorporating the uncertainty of the unit weight of soil the range of COV was approximately 10%–50% for undrained shear
strength (COV-undrained) and 5% to 15% for effective friction angle (COV-
The application of the proposed approach to a region in Anchorage is friction). Based on the research of Cherubini (2000), COV was approxi­
assigned a uniform unit weight of soil of 18 kN/m3 over the entire study mately 10%–70% for effective cohesion c’, and the range for effective
area, to better compare this study and previous studies. However, the friction angle φ’ varied for different soil types, where the fluctuation of
unit weight of soil is considered as an input parameter, and its variability the lower value connects to coarse soil, whereas a higher value is
assigned to fine soil, such as 10%–50% for clay, 5%–25% for silt, and

9
C. Li et al. Engineering Geology 301 (2022) 106616

Fig. 7. Effect of the variability of the unit weight of soil on critical slope angle.

φ’. Other input parameters are the same as those mentioned above. The
reference value for the displacement threshold x is 5 cm and that for the
hazard level is 2% probability of exceedance in 50 years, and the
lognormal distribution is default satisfied for c’ and φ’.
Fig. 8 shows the critical slope angle comparisons of different COV
combinations and geologic units denoted by the continued number. The
variability of the COV combination has a significant influence on the
critical slope angle, especially for geologic units with friction of zero,
which are denoted as 5, 7, 11, and 15. For example, the result of the
geologic unit bc varies from 20◦ to 32◦ as the COV combination de­
creases by 8◦ when incorporating a higher COV combination. However,
COV combinations have less effect on other geologic units, and the
critical slope angle decreases slightly with an increase in COV. For
example, geologic Unit 4 (geologic unit b) shows the best capacity to
resist instability during shaking and expresses only a one-degree in­
crease between the lowest and highest combinations. This phenomenon
is attributed to the different selection of COV combinations in this sec­
tion. The large step of COV-undrained from 0.1 to 0.5 led to a considerable
decrease in the results for these geologic units containing only

Fig. 6. seismic landslide hazard maps showing three hazard categories

(referred to Table 1) with 1.5 m DEM resolutions.

5%–15% for sand. To analyze the influence of different COV combina­

tions of soil strength parameters on the final critical slope, critical an­
gles, and seismic landslide hazard mapping results, two types of COV
combinations are presented in this section. Group 1 represents the
lowest variation, where COV-undrained is 0.1, COV- friction is 0.05, and
COV- cohesion is 0.2, which is unchanged. Group 2 represents the highest
variation, where COV-undrained is 0.5, COV- friction is 0.15, and COV-
cohesion is 0.2, which is unchanged. The lowest variation means less
discrete or less variability of strength parameters in contrast to the
highest variations that represent more variability associated with c’ and Fig. 8. Effect of different COV combination of c’ and φ’ on the critical
slope angle.

10
C. Li et al. Engineering Geology 301 (2022) 106616

undrained parameters (denoted as 5, 7, 11, and 15 in Fig. 8). The cor­ 6. Conclusions
responding outcomes of other geologic units, which have both strength
parameters and show minor variations, are caused by the slight decrease This paper introduces a comprehensive probabilistic method for
in COV- friction, from 0.15 to 0.05, and an unchanged value of 0.2, which seismic landslide hazard mapping procedures that include the prepara­
is assigned to COV- cohesion. For example, if the friction angle is set to 30◦ , tion of a slope map from DEM and equivalent landslide hazard cate­
then +σ leads to 34.5◦ , 33, ◦ and 31.5, ◦ for three COVs (corresponding gories. Instead of using displacement threshold, the critical slope angle,
to 0.15, 0.1, and 0.05, respectively). The difference between them is determined by selecting the seismic hazard level, leads to a modified
almost close to 1◦ , which coincides with the disparities of the critical seismic landslide hazard category, which is more convenient for seismic
slope angle in Fig. 8. Generally, the lower the COV value, the higher the landslide hazard mapping of a region. The proposed hazard category
slope angle, which indicates a low risk of co-seismic landslides, and less was applied to the region of Anchorage for the seismic landslide hazard
area will be regarded as a high landslide hazard. This suggests that the mapping process. The results in hazard maps show that the area of
accurate determination of soil strength parameters is of great signifi­ hazardous zone of most geologic units in present study is smaller than
cance for seismic landslide hazard mapping. The comparison results that in the deterministic analysis, except for the geologic units that were
further indicate that ignoring the uncertainty of slope properties leads to assigned a set of fixed values and ignored the variability of relevant
a considerable difference; therefore, a deterministic group value of input parameters. As far as these geologic units are concerned, logic-tree
parameters might result in an overestimation of resistance to landslides analysis shows an underestimated result, whereas the deterministic
or an underestimated mapping result. method overestimates the hazard levels of these units. Hence, an
appropriate probability distribution, such as a lognormal distribution, is
5. Discussion sufficient for the assessment of the occurrence of a seismic landslide.
However, a suitable DEM resolution needs to be considered in the
The proposed method provides a new type of probabilistic seismic mapping process for a considerable difference with a factor of 2 between
landslide hazard mapping procedure that incorporates epistemic un­ the comparisons of higher and lower resolutions. The seismic landslide
certainty associated with the variability of input parameters (i.e., c’, φ’, hazard mapping results or critical slope angles are more sensitive to
γ, t, and m) and includes aleatory variability of seismic motions and variations in the COV combination than the unit weight of soil. A single
different empirical displacement models. The influence of the input representative value of the unit weight of soil was sufficient for the
parameters and probability distributions on the final mapping results are seismic landslide hazard mapping process of a region.
contained in the middle process of the yield acceleration calculation. In This study describes a feasible method to explore the effects of DEM
addition, the effects of seismic motions and different empirical resolution. However, the proposed method is applicable to shallow
displacement models on the mapping procedure are expressed by landslides because of the hypothesis of the infinite slope model and rigid
displacement hazard curves along with regression empirical formula­ sliding block modeling. Hence, caution should be exercised when the
tions. An abnormal situation existed in the simulation of the strength proposed method is used to analyze deep landslides or soft stratum
parameters by the assumed normal distributions. A negative c’ and φ’ slopes. The development of more advanced technology together with the
might be obtained randomly, which leads to negative yield acceleration proposed method, which considers complex factors such as spatial cor­
based on eqs. (1) and (2). However, this situation has been completely relations and topographic amplification, would considerably promote
avoided by the truncation operation for the ky distribution, which is the development of seismic landslide hazard maps on a regional scale.
discussed in the section above. This abnormal situation for negative c’
and φ’ in the Monte Carlo simulation could barely exist because of a CRediT authorship contribution statement
small COV and a large expectation value in this study, even for high COV
combinations. Chao Li: Conceptualization, Methodology, Software, Investigation,
As shown in Fig. 1, the entire process for seismic landslide hazard Formal analysis, Writing – original draft. Gongmao Wang: Visualiza­
mapping is divided into two separate parts: the preparation of the slope tion, Investigation. Jianjian He: Resources. Yubing Wang: Conceptu­
map and the critical slope angle distributions over all geologic units. alization, Funding acquisition, Resources, Supervision, Writing – review
Because the calculation of the critical slope angle and the accompanying & editing.
seismic landslide hazard category is independent of the mapping process
using a computation platform such as ArcGIS, it is convenient to adopt
the Monte Carlo method and to consider the effect of DEM resolutions Declaration of Competing Interest
without being limited to the computation cost. This differs significantly
from previous studies that were thought to be grid cell-dependent The authors declare that they have no known competing financial
analytical methods and were limited to the number of grid cells in a interests or personal relationships that could have appeared to influence
region. For instance, single and low resolution (such as 6 m, 10 m or 30 the work reported in this paper.
m) were usually adopted by previous studies (Jibson and Michael, 2009;
Wang et al., 2017; Wang and Rathje, 2015) to accelerate the computa­ Acknowledgments
tion speed and for the limited data storage capacity, considering that the
number of grid cells increases significantly as the resolution changes. The authors acknowledge the National Natural Science Foundation
However, the entire process with a resolution of 1.5 m, compared to a of China (No. 51808490) and the Chinese Program of Introducing Tal­
resolution of 6 m adds few computational cost in this study. Therefore, it ents of Discipline to University (the 111 Project, B18047) for their
shows the superiority of the proposed approach in regional mapping, financial support. Technical discussions with Dr. Ellen M. Rathje of the
compared with common methods of grid cell dependency. The proposed University of Texas at Austin were very beneficial and gratefully
mapping procedure in this study employs Newmark’s method and an acknowledged.
infinite slope model. Hence, caution should be exercised when applied
to the analysis of deep landslides or soft stratum slopes. The variability References
of the input parameters was considered to be consistent within the same
geologic units in this study. Therefore, a better improvement in the Alamanis, N., Dakoulas, P., 2019. Effect of spatial variability of soil properties on the
simulation of regional application necessitates studies on the spatial stability and permanent seismic displacements of highway slopes. In: Proceedings of
the XVII ECSMGE-2019.
correlation between grid cells and a rigorous probability distribution. Bojadjieva, J., Sheshov, V., Bonnard, C., 2017. Hazard and risk assessment of
earthquake-induced landslides—case study. Landslides 15, 161–171.

11
C. Li et al. Engineering Geology 301 (2022) 106616

Bray, J.D., Travasarou, T., 2007. Simplified procedure for estimating earthquake-induced Peng, W.F., Wang, C.L., Chen, S.T., Lee, S.T., 2009. Incorporating the effects of
deviatoric slope displacements. J. Geotech. Geoenviron. 133, 381–392. topographic amplification and sliding areas in the modeling of earthquake-induced
Capolongo, D., Refice, A., Mankelow, J., 2002. Evaluating earthquake-triggered landslide landslide hazards, using the cumulative displacement method. Comput. Geosci. 35,
hazard at the basin scale through GIS in the Upper Sele river Valley. Surv. Geophys. 946–966.
23, 595–625. Phoon, K.K., Kulhawy, F.H., 1999a. Characterization of geotechnical variability. Can.
Cherubini, C., 2000. Reliability evaluation of shallow foundation bearing capacity on c’, Geotech. J. 36, 612–624.
φ’ soils. Can. Geotech. J. 37, 264–269. Phoon, K.K., Kulhawy, F.H., 1999b. Evaluation of geotechnical property variability. Can.
Dreyfus, D., Rathje, E.M., Jibson, R.W., 2013. The influence of different simplified Geotech. J. 36, 625–639.
sliding-block models and input parameters on regional predictions of seismic Rathje, E.M., Saygili, G., 2009. Probabilistic assessment of earthquake-induced sliding
landslides triggered by the Northridge earthquake. Eng. Geol. 163, 41–54. displacements of natural slopes. Bull. N. Z. Soc. Earthq. Eng. 42, 18–27.
Fenton, G.A., Griffiths, D.V., 2003. Bearing-capacity prediction of spatially random c - ϕ Refice, A., Capolongo, D., 2002. Probabilistic modeling of uncertainties in earthquake-
soils. Can. Geotech. J. 40, 54–65. induced landslide hazard assessment. Comput. Geosci. 28, 735–749.
Fenton, G.A., Griffiths, D.V., 2004. Reply to the discussion by R. Popescu on “Bearing Saade, A., Abou-Jaoude, G., Wartman, J., 2016. Regional-scale co-seismic landslide
capacity prediction of spatially random c - ϕ soils”. Can. Geotech. J. 41, 368–369. assessment using limit equilibrium analysis. Eng. Geol. 204, 53–64.
Gallen, S.F., Clark, M.K., Godt, J.W., Roback, K., Niemi, N.A., 2017. Application and Saygili, G., Rathje, E.M., 2008. Empirical predictive models for earthquake-induced
evaluation of a rapid response earthquake-triggered landslide model to the 25 April sliding displacements of slopes. J. Geotech. Geoenviron. 134, 790–803.
2015 Mw 7.8 Gorkha earthquake, Nepal. Tectonophysics 714-715, 173–187. Saygili, G., Rathje, E.M., 2009. Probabilistically based seismic landslide hazard maps: an
Jibson, R.W., 2007. Regression models for estimating coseismic landslide displacement. application in Southern California. Eng. Geol. 109, 183–194.
Eng. Geol. 91, 209–218. Shinoda, M., Miyata, Y., 2017. Regional landslide susceptibility following the Mid
Jibson, R.W., 2011. Methods for assessing the stability of slopes during earthquakes—a NIIGATA prefecture earthquake in 2004 with NEWMARK’S sliding block analysis.
retrospective. Eng. Geol. 122, 43–50. Landslides 14, 1887–1899.
Jibson, R.W., Michael, J.A., 2009. Maps showing seismic landslide hazards in Anchorage, Tao, W., Shuren, W., Jusong, S., Peng, X., 2013. Application and validation of seismic
Alaska. In: US Geological Survey Scientific Investigations Map 3077, scale 1:25,000, landslide displacement analysis based on Newmark model: a case study in Wenchuan
11-p. pamphlet. Available at URL. https://pubs.usgs.gov/sim/3077. earthquake. Acta Geologica Sinica (English Edition) 87, 393–397.
Jibson, R.W., Harp, E.L., Michael, J.A., 1998. A Method for Producing Digital Tsai, H.Y., Tsai, C.C., Chang, W.C., 2019. Slope unit-based approach for assessing
Probabilistic Seismic Landslide Hazard Maps: An Example from the Los Angeles, regional seismic landslide displacement for deep and shallow failure. Eng. Geol. 248,
California, Area. US Department of the Interior, US Geological Survey. 124–139.
Jibson, R.W., Harp, E.L., Michael, J.A., 2000. A method for producing digital Wang, Y., 2014. Probabilistic Assessments of the Seismic Stability of Slopes:
probabilistic seismic landslide hazard maps. Eng. Geol. 58, 271–289. Improvements to Site-Specific and Regional Analyses(Dotoral Dissertation).
Lee, S.T., Yu, T.T., Peng, W.F., Wang, C.L., 2010. Incorporating the effects of topographic University of Texas at Austin.
amplification in the analysis of earthquake-induced landslide hazards using logistic Wang, K.L., Lin, M.L., 2009. Development of shallow seismic landslide potential map
regression. Natural Hazards and Earth System Science 10, 2475–2488. based on Newmark’s displacement: the case study of Chi-Chi earthquake, Taiwan.
Liu, J.M., Gao, M.T., Wu, S.R., Wang, T., Wu, J., 2016. A Hazard Assessment Method for Environmental Earth Sciences 60, 775–785.
potential Earthquake-Induced Landslides - a Case Study in Huaxian County, Shaanxi Wang, Y., Rathje, E.M., 2015. Probabilistic seismic landslide hazard maps including
Province. Acta Geologica Sinica-English Edition 90, 590–603. epistemic uncertainty. Eng. Geol. 196, 313–324.
Liu, J., Shi, J., Wang, T., Wu, S., 2017. Seismic landslide hazard assessment in the Wang, H., Wang, G., Wang, F., Sassa, K., Chen, Y., 2008. Probabilistic modeling of
Tianshui area, China, based on scenario earthquakes. Bull. Eng. Geol. Environ. 77, seismically triggered landslides using Monte Carlo simulations. Landslides 5,
1263–1272. 387–395.
Martino, S., Battaglia, S., Delgado, J., Esposito, C., Martini, G., Missori, C., 2018. Wang, Y., Song, C., Lin, Q., Li, J., 2016. Occurrence probability assessment of
Probabilistic Approach to provide scenarios of Earthquake-Induced Slope failures earthquake-triggered landslides with Newmark displacement values and logistic
(PARSIFAL) Applied to the Alcoy Basin (South Spain). Geosciences 8. regression: the Wenchuan earthquake, China. Geomorphology 258, 108–119.
McCrink, T.P., 2001. Regional earthquake-induced landslide mapping using Newmark Wang, T., Liu, J., Shi, J., Wu, S., 2017. The influence of DEM resolution on seismic
displacement criteria, Santa Cruz County, California. In: Ferriz, H., Anderson, R. landslide hazard assessment based upon the Newmark displacement method: a case
(Eds.), Engineering Geology Practice in Northern California, California Division of study in the loess area of Tianshui, China. Environmental Earth Sciences 76.
Mines and Geology Bulletin 210/Association of Engineering Geologists Special Wang, T., Liu, J.M., Shi, J.S., Gao, M.T., Wu, S.R., 2020. Probabilistic seismic landslide
Publication, 12, pp. 77–92. hazard assessment: a case study in Tianshui, Northwest China. J. Mt. Sci. 17,
Meehan, C.L., Vahedifard, F., 2013. Evaluation of simplified methods for predicting 173–190.
earthquake-induced slope displacements in earth dams and embankments. Eng. Wasowski, J., Keefer, D.K., Lee, C.T., 2011. Toward the next generation of research on
Geol. 152, 180–193. earthquake-induced landslides: current issues and future challenges. Eng. Geol. 122,
Miles, S.B., Keefer, D.K., 2009. Evaluation of CAMEL — comprehensive areal model of 1–8.
earthquake-induced landslides. Eng. Geol. 104, 1–15. Zhang, Y.B., Xiang, C.L., Chen, Y.L., Cheng, Q.G., Xiao, L., Yu, P.C., Chang, Z.W., 2019.
Newmark, N.M., 1965. Effects of earthquakes on dams and embankments. Géotechnique Permanent displacement models of earthquake-induced landslides considering near-
15, 139–160. fault pulse-like ground motions. J. Mt. Sci. 16, 1244–1257.
Niño, M., Jaimes, M.A., Reinoso, E., 2014. Seismic-event-based methodology to obtain Zhao, H., Song, E.X., 2012. A method for predicting co-seismic displacements of slopes
earthquake-induced translational landslide regional hazard maps. Nat. Hazards 73, for landslide hazard zonation. Soil Dyn. Earthq. Eng. 40, 62–77.
1697–1713.

12