Engineering Geology 222 (2017) 124–139

Contents lists available at ScienceDirect

Engineering Geology

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

Empirical predictive relationships for rigid sliding displacement based on

directionally-dependent ground motion parameters
Jian Song a, Yufeng Gao a, Adrian Rodriguez-Marek b, Tugen Feng a,⁎
a
Key Laboratory of Ministry of Education for Geomechanics and Embankment Engineering, Hohai University, Nanjing 210098, China
b
Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg, VA 24060, USA

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

Article history: The seismic performance of earth slopes is typically quantified by the predicted rigid-sliding-block displacement
Received 9 December 2016 of a simplified sliding mass. Current empirical predictive relationships for earthquake-induced sliding displace-
Received in revised form 2 March 2017 ments of slopes are generally developed based on the computed displacement data from a suite of earthquake
Accepted 29 March 2017
ground motion time histories. The displacement predicted from these relationships is for the ground motion in-
Available online 5 April 2017
tensity measures associated with a specific ground motion time history. These intensity measures are different
Keywords:
from those for a single definition of bidirectional ground motion that are used in ground motion prediction equa-
Earthquake-induced landslides tions and the distribution of ground shaking following an earthquake (e.g., ShakeMap), which take into consid-
Newmark displacement eration ground motion directionality. Therefore, the use of ground motion intensity measures is not consistent
Predictive model throughout the assessment process of seismic sliding displacement of slopes. This paper presents rigid sliding
Ground motion directionality displacements calculated for a set of ground motion records by rotating the horizontal components through all
Orientation-independent estimation angles. The degree of the azimuthal variation of sliding displacement of slopes with different yield accelerations
is examined by analyzing the distribution of sliding displacements in all orientations. Empirical predictive rela-
tionships for the orientation-independent earthquake-induced sliding displacement of slopes are developed as
a function of directionally-dependent definitions of ground motion parameters. The proposed relationships en-
sure consistency between the derivation of the ground motion intensity measures and its application in the pre-
diction of sliding displacement of slopes, and consider the potential effects of ground motion directionality on
displacement predictions.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction
Earthquake-induced landslides have proved to be some of the most
damaging seismic hazards in numerous earthquakes. It is hence impor-
tant to estimate the probability of occurrence of these landslides both in
site-specific and regional-scale assessments. Sliding displacement due
to earthquake shaking is commonly used to assess the seismic perfor-
mance of slopes. Newmark (1965) proposed a rigid sliding block
model, which assumes that downslope sliding is initialized when the
shaking acceleration exceeds the yield acceleration (ky) of the block,
and the block continues to move along a shear surface until the veloci-
ties of the block and ground coincide. The sliding displacement (D) is
defined as the cumulative relative displacement at the end of ground
shaking. Although the rigid sliding block model is a simplified represen-
tation of the field conditions, sliding displacements computed in this
fashion have been demonstrated to correlate strongly with the occur-
rence of landslides in previous well documented earthquakes
(e.g., Wilson and Keefer, 1983; Jibson et al., 2000).
⁎ Corresponding author.
E-mail address: tgfeng75@163.com (T. Feng).
The yield acceleration of the simplified sliding block and a site-
specific ground motion time history are required to compute sliding
block displacements. However, it is generally complicated and time con-
suming to generate an appropriate ground motion time history for cal-
culation of sliding displacements, particularly for regional-scale
assessments. Alternatively, empirical predictive relationships for sliding
displacement are commonly used. These relationships have been devel-
oped as a function of the slope parameter (ky) and one or more intensity
measures (IMs) of earthquake shaking (e.g., Makdisi and Seed, 1978;
Ambraseys and Menu, 1988; Jibson, 2007; Bray and Travasarou, 2007;
Saygili and Rathje, 2008; Rathje and Saygili, 2009; Rathje and
Antonakos, 2011; Hsieh and Lee, 2011; Lee and Green, 2015; Song and
Rodriguez-Marek, 2015; Song et al., 2016). While these empirical dis-
placement models cannot replace site-specific seismic response analy-
ses of slopes, they are valuable for the assessment of seismic risk of
slopes both at the local and regional levels.
Various deterministic and probabilistic methodologies based on em-
pirical predictive relationships are used to estimate the seismic dis-
placement hazard of slopes (e.g., Rathje and Saygili, 2008, 2009, 2011;
Rathje et al., 2014; Du and Wang, 2014; Rodriguez-Marek and Song,
2016). These methodologies involve the prediction of ground motion

http://dx.doi.org/10.1016/j.enggeo.2017.03.025
0013-7952/© 2017 Elsevier B.V. All rights reserved.
 J. Song et al. / Engineering Geology 222 (2017) 124–139
Fig. 1. Flow chart for current methods used for the estimation of seismic slope
displacement.
Fig. 2. Distribution of earthquake ground motion records used in this study in terms of
earthquake magnitude (Mw) and closest distance to the rupture fault (Rrup).
Fig. 3. Time histories of acceleration, velocity and sliding displacement of a rigid sliding block (ky = 0.05 g) for the positive and negative polarities of the El Centro Array #4 ground motion
from the 1979 Mw6.5 Imperial Valley-06 earthquake.
125
IMs, which is generally accomplished through the use of ground motion
prediction equations (GMPEs) or, for the analyses of past earthquakes,
using other estimates of ground shaking following an earthquake
(e.g., ShakeMap developed by U.S. Geological Survey, USGS, Worden
and Wald, 2016). Earthquake ground motions produce translational ac-
celerations in two horizontal components and one vertical component;
however, most existing GMPEs, as well as ShakeMap, predict ground
motion intensity for a single definition of bidirectional ground motions.
For example, the NGA-West2 research program has produced models
for predicting the median IMs of a ground motion when rotated over
all horizontal orientations (this is referred as the IMsRotD50, Boore,
2010) and the motions shown in maps in ShakeMap are the larger
ground motion values, i.e., the larger value observed on the two hori-
zontal components (IMsLarger).
Current empirical predictive relationships for earthquake-induced
sliding displacements of slopes were developed by using the calculated
displacement data from a suite of ground motion time histories. Hence,
the displacements predicted from these relationships are for ground
motion IMs associated with a specific ground motion time history.
These IMs are different from the predicted ground shaking for future
earthquakes from GMPEs, or the best estimates of ground shaking fol-
lowing an earthquake from ShakeMap, and do not take into consider-
ation ground motion directionality. Therefore, the use of ground
motion IMs is not consistent throughout the assessment process of seis-
mic performance of earth slopes (Fig. 1). In addition, sliding displace-
ments may be different for ground shaking in different orientations
for given predicted or estimated directionally-dependent ground mo-
tion parameters (IMsRotD50 or IMsLarger). Traditional predictive relation-
ships do not consider the potential effects of ground motion
directionality on sliding displacements.
The relationship between ground motion directionality and the dy-
namic response of potentially unstable slopes has been observed in
past earthquakes. For example, Del Gaudio and Wasowski (2011) pro-
vide evidence that seismic ground motion on slopes covered by thick
colluvia or on deep-seated landslides can have a pronounced directional
character, with maxima oriented along the maximum slope direction.
Del Gaudio and Wasowski (2011) based these observations on instru-
mental recordings at Caramanico Terme (Italy), and also point to similar
observations in the literature. These observations point to the need to
126 J. Song et al. / Engineering Geology 222 (2017) 124–139

Fig. 4. Computation of the rigid block sliding displacement by rotating the horizontal ground motion (GM) components through all non-redundant angles.

Fig. 5. Orientation variation of computed sliding displacements, PGA and PGV for the El Centro Array #4 ground motion from the 1979 Mw6.5 Imperial Valley-06 earthquake. (a) Sliding
displacement for ky = 0.05 g, (b) sliding displacement for ky = 0.2 g, (c) PGA, (d) PGV.

improve the understanding of the degree of variability of sliding dis-
placements resulting from the orientation of ground motions. More-
over, given the widespread usage of the rigid sliding displacement in
earthquake-induced landslide hazard assessments, it is important to en-
sure that there is consistency in the orientation definition of inputs
(i.e., IMs) and the computed displacements.
This paper examines the influence of ground motion directionality
on the development of an empirical predictive relationship for
earthquake-induced sliding displacements of slopes. A subset of ground
motion records from the NGA-West2 database is selected and rigid slid-
ing displacements are computed for each ground motion record by ro-
tating the horizontal components through all angles. The distribution
of sliding displacement for all orientations is examined. Empirical pre-
dictive relationships, based on directionally-dependent ground motion
IMs, are developed for the maximum and the median sliding displace-
ments through all orientation. The proposed relationship ensures the
consistency between the derivation of the ground motion IMs and its
application in the prediction of sliding displacement of slopes. Although
the directionality of the motion relative to the orientation of a slope gen-
erally cannot be predicted a priori, the work in this study provides engi-
neers with a better understanding of the sensitivity of slope
displacements to azimuthal variations and the proposed relationships,
which consider ground motion directionality, can be used to help iden-
tify the range of sliding displacements that may occur.
2. Ground motion data
Motion-to-motion variability for a given IM is an important source of
variability when estimating sliding displacements of slopes. A large
number of earthquake recordings are desirable to characterize better
the influence of ground motions on the displacement prediction. The
latest NGA-West2 ground motion database (Ancheta et al., 2012) from

Fig. 6. Distributions of various ratios of the sliding displacement for different ky values through all orientations: Histograms of (a) the ratio between sliding displacement in all orientations
(Dall orientations) and the maximum sliding displacement (Dmax); (b) the ratio between sliding displacement in two recorded orientations (Drecorded orientations) and the maximum sliding
displacement (Dmax); (c) the ratio between the minimum sliding displacement (Dmin) and the maximum sliding displacement (Dmax); (d) the ratio between the median sliding
displacement (Dmedian) and the maximum sliding displacement (Dmax).
J. Song et al. / Engineering Geology 222 (2017) 124–139 127
128 J. Song et al. / Engineering Geology 222 (2017) 124–139
the Pacific Earthquake Engineering Research Center (PEER) includes
one of the most comprehensive sets of ground motions recorded in
worldwide shallow crustal earthquakes in active tectonic regimes
(http://ngawest2.berkeley.edu/).
The ground motion records used in this study were selected from the
NGA-West2 database. First, ground motions with moment magnitude
(Mw) between 5.5 and 7.9 and with closest distance to the rupture
fault (Rrup) within 100 km were selected to focus on the range of mag-
nitudes and distances that will generally control landslide hazard in
high seismicity regions. Then, ground motions recorded at certain
sites (i.e., soft soil sites, on the crest or abutments of dams, underground,
not at the ground floor of a building, or in buildings larger than four
stories), and ground motions with high-pass filter corner frequencies
larger than 0.25 Hz or low-pass filter corner frequencies b10 Hz were
removed from the database (Saygili and Rathje, 2008). Finally, records
with single horizontal component or with unknown component orien-
tations were also excluded. Note that when downloading the ground
motion time histories from the database, some recordings compiled in
the NGA-West2 flatfile are unavailable in the database. The resulting
dataset included 2496 motions. The distribution of records in the final
ground motion dataset in terms of Mw and Rrup is given in Fig. 2.
3. Rigid block sliding displacements
The sliding displacement of rigid blocks with various ky values (0.02,
0.05, 0.1, 0.15, 0.2, 0.3 g) subjected to the selected ground motions was
calculated. In the commonly used Newmark-type analysis, the upslope
resistance to sliding is assumed to be infinitely large such that only
downslope displacement is allowed. The polarity of shaking is hence
of great importance in the calculation of sliding displacement, as is illus-
trated in Fig. 3. In the subsequent analyses, displacements were com-
puted for positive and negative polarities for each motion, with the
largest displacement used for analyses. The use of the largest value
avoids the possible underestimation of the sliding displacement of
slopes. Note also that the sliding displacements in existing predictive re-
lationships are generally correlated to the peak ground acceleration
(PGA) and peak ground velocity (PGV), which are also the maximum
absolute acceleration and velocity in the two polarities of ground
motion.
In traditional studies on the development of predictive relationships
of sliding displacements (e.g., Saygili and Rathje, 2008; Hsieh and Lee,
2011; Lee and Green, 2015), the displacements were computed for or-
thogonal components recorded at the same station during the same
earthquake and were treated as separate data points in the analysis. In
this study, the ground motions are rotated through 0–180° (the remain-
ing orientations are redundant because these orientations have been
considered in terms of the polarity of shaking described previously)
with respect to the fault strike with an increment of 2°. The rigid sliding
displacements were then computed for all the horizontal components
with various rotation angles for each ground motion record. The proce-
dure for the derivation of displacement data is illustrated in Fig. 4.
4. Distribution of sliding displacement over all orientations
The ground motion intensity can be varied in different orientations
when the ground motion is polarized. Therefore, the sliding displace-
ment of a rigid block may also differ under ground shaking in different
orientations. For example, Fig. 5 illustrates the computed sliding dis-
placements for two values of ky, along with the PGA and PGV values
for all horizontal ground motion components of the El Centro Array
#4 ground motion from the 1979 Mw6.5 Imperial Valley-06 earthquake.
Fig. 7. Histogram of the orientation of the maximum sliding displacement direction relative to the fault strike at different ky values for ground motions with different magnitude and
distance. (a) Ground motions with Mw N 6.5, (b) ground motions with Mw ≤ 6.5, (c) ground motions with Rrup N 5 km, (d) ground motions with Rrup ≤ 5 km.
The sliding displacement exhibits pronounced azimuthal variability for
both small and large ky values. The rigid sliding block model describes
a plastic response of slopes under the ground shaking, and the direc-
tionality of this plastic sliding-displacement may be different from the
ground shaking, as can be observed in Fig. 5. The directional variation
is apparently different for sliding displacement, PGA and PGV. The ori-
entation of the maximum sliding displacement of slopes with ky =
0.05 g is 100° from the fault-parallel (FP) direction, while the maximum
PGA and PGV orientation is 156° and 62° from the FP direction, respec-
tively. The sliding displacement in the maximum PGA and PGV orienta-
tion is 66.41 cm and 87.38 cm, respectively, while the maximum sliding
displacement in all orientations is 142.51 cm for ky = 0.05 g. In addition,
the directional variation of sliding displacement is different for slopes
with different strength (ky values). The maximum displacement orien-
tation for ky = 0.05 g is 100° from the FP direction, while the maximum
displacement orientation for ky = 0.2 g is about 130° from the FP direc-
tion (the maximum displacement is 11.37 cm). The displacement in the
maximum PGA orientation (66.41 cm) is smaller than that in the max-
imum PGV orientation (87.38 cm) for ky = 0.05 g, while the displace-
ment in the maximum PGA orientation (8.86 cm) is significantly
larger than that in the maximum PGV orientation (0.73 cm) for ky =
0.2 g.
Various ratios of the sliding displacement through all orientations
for each ground motion record are compared to examine the degree of
the azimuthal variation of sliding displacement. Fig. 6 shows the ratio
of the sliding displacement in all orientations, in two recorded orienta-
tions, the minimum sliding displacement (Dmin) and the median sliding
displacement (Dmedian) to the maximum sliding displacement (Dmax)
for the rigid block with different ky values, respectively (referred to as
Dall orientations/Dmax, Drecorded orientations/Dmax, Dmin/Dmax and Dmedian/
Dmax). Note that only cases with a non-zero Dmax are considered. It
can be seen that most of the Dall orientations/Dmax are larger than 0.5
with about 30% ratios close to one for ky = 0.02 g, and the number of
the largest Dall orientations/Dmax values (close to one) decreases while
the number of the smallest Dall orientations/Dmax (close to zero) values in-
creases as increasing the strength of slope (ky). This is the result of the
increasing number of cases where the displacement is zero at some ori-
entations as the strength of slope is enhanced. Similar results are ob-
served for the distribution of Drecorded orientations/Dmax. Most of Dmin/
Dmax are b0.5 for all ky values, indicating that the maximum displace-
ment is N 100% larger than the minimum displacement. This is much
more prominent as the yield acceleration of the block ky increases. The
number of Dmin/Dmax that equals to zero becomes dominant for larger
ky values. Most of the Dmedian/Dmax values are larger than 0.5 for
ky ≤ 0.05 g, while they are b 0.5 for ky N 0.05 g. The orientation variations
of sliding displacement shown in Fig. 6 are also significantly larger than
those of ground shaking (Shahi and Baker, 2014), particularly for slopes
with large ky values. These observations imply that the sliding displace-
ment is significantly different for ground motion in different orienta-
tions. The use of computed sliding displacement based on individual
ground motion component may underestimate the seismic hazard of
slopes.
The orientation of the maximum sliding displacement direction is
also important, and is related to the predominant sliding direction of
earthquake-induced landslides. The distribution of the orientation of
the maximum sliding displacement direction relative to the fault strike
is examined for different ky values and for ground motions with differ-
ent magnitude and closest distance (Fig. 7). The maximum direction ori-
entation is defined as the minimum angle between the strike of the fault
and the orientation of maximum sliding displacement, and thus ranges
from 0 to 90°. It can be seen that the maximum displacement could
J. Song et al. / Engineering Geology 222 (2017) 124–139 129
130 J. Song et al. / Engineering Geology 222 (2017) 124–139

Table 1
Model parameters for the scalar (PGA, Mw) form using different displacement data.

Sliding displacement used in the regression a1 a2 a3 a4 a5 a6 a7 τ σ σtotal b1 b2 b3 b4

0° from the strike 5.054 −14.345 20.966 −25.355 9.390 0.681 0.913 0.51 0.83 0.97 – – – –
20° from the strike 5.052 −13.730 18.106 −20.560 6.905 0.693 0.878 0.52 0.84 0.99 – – – –
40° from the strike 5.100 −12.936 14.622 −14.572 3.438 0.734 0.896 0.52 0.84 0.99 – – – –
60° from the strike 5.236 −13.465 17.147 −19.044 5.943 0.763 0.821 0.53 0.84 0.99 – – – –
80° from the strike 5.258 −13.918 18.237 −19.479 5.453 0.749 0.792 0.55 0.84 1.00 – – – –
100° from the strike 5.270 −14.392 20.619 −23.795 7.938 0.740 0.793 0.56 0.85 1.02 – – – –
120° from the strike 5.203 −13.603 17.014 −17.727 4.591 0.732 0.846 0.49 0.84 0.97 – – – –
140° from the strike 5.164 −13.466 16.480 −17.294 4.760 0.706 0.816 0.46 0.84 0.96 – – – –
160° from the strike 5.117 −14.144 19.801 −23.086 7.985 0.692 0.862 0.48 0.83 0.96 – – – –
Two recorded components 5.187 −14.341 20.773 −24.604 8.774 0.727 0.828 0.56 0.84 1.01 – – – –
Median D and median IMs 5.214 −13.408 16.880 −18.458 5.342 0.748 0.850 0.52 0.77 0.93 26.044 −32.801 1.412 0.888
Max D and median IMs 5.596 −12.890 19.059 −22.997 9.447 0.821 0.754 0.51 0.86 1.00 18.656 −16.193 0.938 0.091
Median D and larger IMs 5.022 −16.313 28.270 −42.139 21.328 0.670 0.903 0.53 0.86 1.01 13.057 −21.263 0.615 0.771
Max D and larger IMs 5.393 −15.003 25.865 −34.085 15.160 0.770 0.816 0.51 0.88 1.02 17.239 −18.383 0.887 0.335

occur at all orientations from the fault strike for different ground mo-
tions. The magnitude does not seem to have any significant influence
on the distribution of orientation of the maximum sliding displacement.
However, different tendencies can be observed for different closest dis-
tances, i.e., the maximum direction orientation of sliding displacement
for sites with distances N 5 km is almost uniformly distributed, while a
pronounced polarity of the sliding displacements (directed towards
the fault-normal direction) is observed when the site is located within
5 km of the fault. This is similar to the observations of ground motion di-
rectionality for spectral acceleration at periods greater than about 1.0 s
(Shahi and Baker, 2014).
5. Orientation-independent estimation of sliding displacement
The observations presented in the previous sections indicate that the
sliding displacement of slopes can be significantly different for ground
shaking in different orientations. The reminder of this paper presents
the development of various empirical relationships for sliding displace-
ments based on directionally-dependent ground motion IMs. The coef-
ficients of all the relationships were determined using a mixed effects
regression approach that accounts for the inter-event (between earth-
quakes) and intra-event variability (within earthquakes). The mixed ef-
fects modelling assumes that displacement values follow a logarithmic
normal distribution. The statistical program R 3.2.3 (2015) was used
to perform the regression analyses.
5.1. Predictive relationships for sliding displacement at various orientations
Current predictive relationships for sliding displacements were
commonly developed using the displacements computed for two orien-
tations in which the ground motion was recorded. In order to check
whether there is a distinction among the predictive relationships devel-
oped using the sliding displacements of slopes at various orientations,
various relationships are developed by using computed displacements
from ground motion components with various rotation angles from
the fault strike. This is given as:

 h   i
ln Dij Rot ðkÞ
¼ f ky ; IMsij RotðkÞ þ ηi þ εij
where [lnDij]Rot(k) represents the computed logarithmic sliding dis-
placement for the k orientation from the fault strike of the j-th recording
and the i-th event; f[.] represents the functional form predicting the log-
arithmic sliding displacement; (IMsij)Rot(k) is the value of the ground
motion parameters used in the function of the k orientation of the j-th
recording and the i-th event; ηi refers to inter-event residual and has a
mean of zero and a variance of τ2; and εij denotes intra-event residual
and has a mean of zero and a variance of σ2, respectively. The total resid-
ual for the j-th record of the i-th event is defined as the sum of the
corresponding inter- and intra-event residuals (i.e. ηi + εij). The stan-
dard deviation of the total residual is given by:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
σ total ¼ σ 2 þ τ2 ð2Þ
The functional forms of the predictive relationships developed by
Saygili and Rathje (2008) and Rathje and Saygili (2009) are selected
to perform regressions, including a scalar (PGA, Mw) form (Eq. (3a))
and a vector (PGA, PGV) form (Eq. (3b)), referred to as RS09(PGA,
Mw) and SR08(PGA, PGV) forms, respectively:

    2  3
f ky ; PGA; M w ¼ a1 þ a2 ky =PGA þ a3 ky =PGA þ a4 ky =PGA
 4 ð3aÞ
þ a5 ky =PGA þ a6 ln ðPGAÞ þ a7 ðMw −6Þ

    2  3
f ky ; PGA; PGV ¼ a1 þ a2 ky =PGA þ a3 ky =PGA þ a4 ky =PGA
 4
þ a5 ky =PGA þ a6 ln ðPGAÞ þ a7 ln ðPGVÞ
ð3bÞ
The coefficients and variabilities for the two predictive relationships
using different displacement data are given in Tables 1 and 2.
Fig. 8 provides a comparison between the median displacements
predicted by the relationships developed from various orientations
using RS09(PGA, Mw) and SR08(PGA, PGV) forms. The sliding displace-
ment is for a specific scenario: Mw = 7, Rjb = 6 km, rock site conditions
(Vs30 = 760 m/s), and strike-slip faulting. The values of PGA and PGV
used to predict the displacement are 0.33 g and 30 cm/s, which are
the median values predicted from the Boore et al. (2014) ground motion
prediction relationship. In view of the prevalent use of the displacement
for two recorded ground motion components when developing the em-
pirical predictive relationship of sliding displacement, the relationship
to make comparisons is also developed by using computed displace-
ments from the two recorded components. The existing RS09(PGA,
ð1Þ
Mw) and SR08(PGA, PGV) models for the sliding displacement are not
used to avoid issues related to differences in predicted displacement
due to disparities in ground motion database, analysis techniques and
so on. The results in Fig. 8 imply that the median sliding displacement
is almost identical from different predictive relationships. The variation
of standard deviation with ky/PGA for different predictive relation-
ships is shown in Fig. 9, and the distribution of standard deviation
is generally similar for different relationships. The observations in
Figs. 8 and 9 indicate that the predictive relationships are indepen-
dent of the orientations of the displacement data used in their
derivation.
 J. Song et al. / Engineering Geology 222 (2017) 124–139 131

Table 2
Model parameters for the vector (PGA, PGV) form using different displacement data.

Sliding displacement used in the regression a1 a2 a3 a4 a5 a6 a7 τ σ σtotal b1 b2 b3 b4

0° from the strike −0.496 −13.463 16.490 −16.880 3.990 −0.529 1.400 0.25 0.57 0.62 – – – –
20° from the strike −0.544 −13.171 15.542 −15.624 3.498 −0.531 1.409 0.23 0.58 0.62 – – – –
40° from the strike −0.517 −12.362 12.357 −10.602 0.801 −0.504 1.396 0.24 0.58 0.63 – – – –
60° from the strike −0.401 −12.654 13.696 −12.959 2.104 −0.480 1.378 0.24 0.58 0.63 – – – –
80° from the strike −0.478 −12.957 13.802 −11.585 0.635 −0.518 1.400 0.24 0.58 0.63 – – – –
100° from the strike −0.445 −13.279 15.372 −14.404 2.209 −0.520 1.396 0.25 0.59 0.64 – – – –
120° from the strike −0.424 −12.868 13.843 −12.338 1.316 −0.512 1.387 0.23 0.59 0.63 – – – –
140° from the strike −0.539 −12.659 12.852 −10.770 0.603 −0.536 1.411 0.20 0.59 0.62 – – – –
160° from the strike −0.507 −13.358 16.194 −16.841 4.251 −0.538 1.407 0.21 0.58 0.62 – – – –
Two recorded components −0.403 −13.489 16.845 −17.509 4.329 −0.498 1.382 0.26 0.58 0.64 – – – –
Median D and median IMs −0.488 −13.042 15.426 −15.782 3.405 −0.508 1.403 0.24 0.51 0.56 29.904 −42.468 −0.403 2.735
Max D and median IMs −0.332 −12.249 16.860 −20.178 8.171 −0.472 1.441 0.21 0.62 0.65 15.150 −17.481 −0.144 1.297
Median D and larger IMs −0.690 −16.055 27.374 −40.406 19.633 −0.584 1.405 0.27 0.61 0.67 12.912 −25.073 −0.942 1.992
Max D and larger IMs −0.503 −14.762 25.155 −33.149 14.534 −0.518 1.441 0.21 0.63 0.66 14.825 −21.000 −0.462 1.716

5.2. Predictive relationships for the maximum and median sliding displace- is given as:
ment over all orientations

  
ln Dij RotD100
¼ f ky ; IMsij RotD50 þ ηi þ εij ð4Þ
5.2.1. Displacement predicted by the median IMs over all orientations


  
In this section, predictive relationships are developed for the maxi- ln Dij RotD50
¼ f ky ; IMsij RotD50 þ ηi þ εij ð5Þ
mum and median sliding displacement over all orientations as a func-
tion of the median ground motion IMs (i.e., the directionally- where (lnDij)RotD100 is the 100th percentile of the logarithmic sliding
dependent IMs used in the latest NGA-West2 GMPEs, IMsRotD50). This displacement over all orientations (the maximum sliding displacement)

Fig. 8. Comparison of the predicted median sliding displacement of various models developed based on the computed displacement data of ground motions in various orientations (Mw =
7, Rjb = 6 km, PGA = 0.33 g, PGV = 30 cm/s): (a) The RS09(PGA, Mw) form; (b) The SR08(PGA, PGV) form.

Fig. 9. Comparison of the standard deviation of various models developed based on the computed displacement data of ground motions in various orientations: (a) The RS09(PGA, Mw)
form; (b) The SR08(PGA, PGV) form.
132 J. Song et al. / Engineering Geology 222 (2017) 124–139

Fig. 10. Distributions of inter-event and intra-event residuals with respect to Mw, ky/PGA, ln(PGA), ln(PGV) and ln(Ia) (Ia is the Arias intensity) for models developed using the median
displacements in all orientations as a function of the median IMs in all orientations. The dotted lines and the gray zones represent the results of local regression and 95% confidence
intervals, respectively. (a) Median D and median IMs, and RS09(PGA, Mw) form, (b) Median D and median IMs, and SR08(PGA, PGV) form.

of the j-th recording and the i-th event; (lnDij)RotD50 is the 50th percen-
tile of the logarithmic sliding displacement over all orientations (the
median sliding displacement) of the j-th recording and the i-th event;
and (IMsij)RotD50 is the 50th percentile of the ground motion parameters
over all orientations (the median IMs) of the j-th recording and the i-th
event, which is the definition of IMs in the latest NGA-West2 GMPEs.
The regression analyses are performed by using Eqs. (3a) and (3b).
The parameters for the two predictive relationships are given in
Tables 1 and 2.
The distributions of inter- and intra-event residuals from
Eqs. (4) and (5) plotted versus various predictors are shown in
Figs. 10 and 11. Also shown in the figures are the average trends in
 J. Song et al. / Engineering Geology 222 (2017) 124–139 133

Fig. 11. Distributions of inter-event and intra-event residuals with respect to Mw, ky/PGA, ln(PGA), ln(PGV) and ln(Ia) for models developed using the maximum displacements in all
orientations as a function of the median IMs in all orientations. (a) Max D and median IMs, and RS09(PGA, Mw) form, (b) Max D and median IMs, and SR08(PGA, PGV) form.

residuals computed using local regression (loess) using the software R
3.2.3. No obvious biases between the residuals and the predictors can
be observed, with an exception of the intra-event residuals versus
ln(PGV) for the RS09(PGA, Mw) form, which displays an increasing
trend with increasing PGV values.
A comparison between the predicted median sliding displacement
of these models is shown in Fig. 12, along with the results from the
model developed using displacements of the recorded orthogonal
ground motion components (predictions are shown for a scenario
given by Mw = 7, Rjb = 6 km, PGA = 0.33 g, PGV = 30 cm/s). It can
134 J. Song et al. / Engineering Geology 222 (2017) 124–139

Fig. 12. Comparison of the predicted median sliding displacement of models developed using the computed displacements of the recorded orthogonal ground motion components as a
function of IMs in the time histories, using the median and maximum displacements in all orientations as a function of the median IMs in all orientations, respectively (Mw = 7, Rjb =
6 km, PGA = 0.33 g, PGV = 30 cm/s): (a) The RS09(PGA, Mw) form; (b) the SR08(PGA, PGV) form.

be seen that the predicted displacement is consistent between the rela-  

tionships developed for the median displacement and for the displace-
ment for the two recorded components. However, the predicted
displacement from the relationship developed for the maximum dis-
placement in all orientations is significantly larger than that from the
former two relationships. The predicted maximum displacement is
about 20%–70% larger than the predicted median displacement when
ky b 0.1 g, and is N 100% larger when ky N 0.1 g for both functional
forms. We note that the use of the IMsRotD50 as the predictors ensures
consistency between the derivation of the ground motion IMs and its
application in the prediction of sliding displacement of slopes. Fig. 13
shows the variation of standard deviation with ky/PGA for different pre-
dictive relationships. Observe that the standard deviation decreases
when using the DRotD50 and IMsRotD50 in the predictive relationship com-
pared to the traditional predictive relationship developed using dis-
placements from recorded ground motion time histories. The standard
deviation for the model developed by DRotD100 and IMsRotD50 is also
lower than that of traditional predictive relationship at small ky/PGA
values, while it is higher for larger ky/PGA values. The standard devia-
tion for the scalar (PGA, Mw) model and vector (PGA, PGV) model of
sliding displacement are represented as a second and first order polyno-
mial, respectively (Rathje and Saygili, 2009):
   2
σ ln D ¼ c1 þ c2 ky =PGA þ c3 ky =PGA for ðPGA; Mw Þ form
σ ln D ¼ c1 þ c2 ky =PGA for ðPGA; PGVÞ form ð6bÞ
The coefficients in Eqs. (6a) and (6b) are given in Table 3.
When considering the displacement of a rigid sliding block subject
to a given ground motion time history, displacements occur only if the
PGA exceeds the value of the yield acceleration (e.g., ky/PGA b 1). This
is also the case when the intended use of a model is to predict displace-
ments for a given component of a ground motion. Eqs. (4) and (5), how-
ever, use the median ground motion as the input. This raises the
possibility that non-zero displacements occur for ground motions
where the median PGA is lower than the yield acceleration. Thus, the
proposed relationships should be considered as conditioned on non-
zero displacements and a model for the probability of occurrence of
zero displacement is needed. This can be expressed as:
E½D ¼ E½DjD N0  ½1−P ðD ¼ 0Þ ð7Þ
where E[.] and P(.) represent expected value and probability, respectively.
A logistic regression model by program R 3.2.3 was used for this analysis,
and the probability of zero displacement can be expressed as:
1
P ðD ¼ 0Þ ¼
   for ðPGA; Mw Þ form
1 þ exp b1 þ b2 ky =PGA þ b3 ln ðPGAÞ þ b4 Mw
ð6aÞ ð8aÞ

Fig. 13. Comparison of the standard deviation of models developed using the computed displacements of the recorded orthogonal ground motion components as a function of IMs in the
time histories, using the median and maximum displacements in all orientations as a function of the median IMs in all orientations, respectively: (a) The RS09(PGA, Mw) form; (b) the
SR08(PGA, PGV) form.
 J. Song et al. / Engineering Geology 222 (2017) 124–139 135

Table 3 following an earthquake. For example, the online mapping product

Coefficients in the relationships of standard deviation for the scalar (PGA, Mw) and vector ShakeMap, developed by the USGS, displays the distribution of ground
(PGA, PGV) forms using different displacement data.
shaking following an earthquake. The overall mapping philosophy is
Coefficients Scalar (PGA, Mw) form Vector (PGA, to combine information from individual stations, site amplification
PGV) form characteristics, and ground motion prediction equations to create the
c1 c2 c3 c1 c2 best composite map. Note that the motions depicted in the map are
Median D and median IMs 0.753 0.809 −0.772 0.429 0.280 the larger value from the two recorded horizontal components of mo-
Max D and median IMs 0.799 0.651 −0.379 0.363 0.547 tion (Worden and Wald, 2016), which is also a directionally-
Median D and larger IMs 0.736 1.261 −1.157 0.473 0.454 dependent ground motion IMs, and deviates from the IMs in a particular
Max D and larger IMs 0.800 0.903 −0.684 0.434 0.480 ground motion time history. In view of this, predictive relationships of
sliding displacement of slopes based on the IMsLarger are developed.
This is given as:

 h   i
1
P ðD ¼ 0Þ ¼
   for ðPGA; PGVÞ form ln Dij ¼ f ky ; IMsij L arger þ ηi þ εij ð9Þ
1 þ exp b1 þ b2 ky =PGA þ b3 ln ðPGAÞ þ b4 ln ðPGVÞ RotD100

ð8bÞ
The regression coefficients determined from logistic regression anal-
yses are listed in Tables 1 and 2.
A comparison between the predicted probabilities of zero displace-
ment for different relationships is shown in Fig. 14. The probability is
for the specific scenario of Mw = 6.5, Rjb = 15 km, rock site conditions
(Vs30 = 760 m/s), and strike-slip faulting. The median values of PGA
and PGV are 0.16 g and 11.9 cm/s predicted from the Boore et al.
(2014) ground motion prediction relationship. As expected, the proba-
bility of zero displacement for the model developed by the maximum
displacement in all orientations decreases compared to the model de-
veloped by the median displacement.
It is interesting to note that the sliding displacement in the maxi-
mum IMs direction cannot represent the maximum sliding displace-
ment over all orientations (Fig. 5). Moreover, the displacement
predicted from the traditional vector model using the maximum PGA
and PGV values is for an unrealistic single orientation ground motion,
because the maximum PGA and maximum PGV generally occur in dif-
ferent orientations (Shahi and Baker, 2014). In addition, if someone
uses the maximum ground motion IMs in the traditional models to pre-
dict sliding displacements, relationships would be needed to predict the
maximum IMs as a function of median IMs. This would introduce an ad-
ditional conversion with the associated uncertainties for the ratio of
maximum to median IMs (Shahi and Baker, 2014).
5.2.2. Displacement predicted by the larger value of IMs in the two recorded
components
The assessment of sliding displacement after an earthquake can pro-
vide a rapid estimate of the potential for earthquake-induced slope fail-
ures, which helps guiding the emergency response. This is generally
performed on the basis of the observed distribution of ground shaking

 h   i
ln Dij RotD50
¼ f ky ; IMsij L arger þ ηi þ εij ð10Þ
where IMsLarger is the larger value of IMs in the two recorded compo-
nents, which is that used in ShakeMap. The regression analyses are
made by using Eqs. (3a) and (3b). The coefficients for the two predictive
relationships are given in Tables 1 and 2.
The distributions of inter- and intra-event residuals from
Eqs. (9) and (10) plotted versus various predictors are shown in
Figs. 15 and 16. Similar to the observations when using the median
IMs as predictors, no obvious biases can be observed, except for the
intra-event residuals versus ln(PGV) for the RS09(PGA, Mw) form.
Therefore, the predictive relationships developed using the vector
(PGA, PGV) functional form is recommended, in spite of the fact that
the correlation coefficient between the PGA and PGV would be need
in the probabilistic assessment of the sliding displacement of slopes.
A comparison between the predicted median sliding displacement
of these models is shown in Fig. 17, along with the results from the
model developed by displacements of the recorded orthogonal ground
motion components (Mw = 7, Rjb = 6 km, PGA = 0.33 g, PGV =
30 cm/s). Observe that the predicted displacements from the relation-
ship developed using the median displacement in all orientations as a
function of IMsLarger is smaller than that for the two recorded compo-
nents. The predicted displacement from relationship developed using
the maximum displacement in all orientations as a function of IMsLarger
is still larger than that for the two recorded components, but not as sig-
nificant as that for the relationship as a function of IMsRotD50. This is due
to that the IMsLarger is generally larger than the IMsRotD50 for a ground
motion [e.g., Bradley and Baker, 2014]. Fig. 18 shows the variation of
standard deviation with ky/PGA for different predictive relationships.
The standard deviation for the model using the orientation-
independent displacements and IMsLarger is larger than that of model

Fig. 14. Comparison of the predicted probability of zero displacement of models developed using the median and maximum displacements in all orientations as a function of the median
IMs in all orientations, respectively (Mw = 6.5, Rjb = 15 km, PGA = 0.16 g, PGV = 11.9 cm/s): (a) The RS09(PGA, Mw) form; (b) the SR08(PGA, PGV) form.
136 J. Song et al. / Engineering Geology 222 (2017) 124–139

Fig. 15. Distributions of inter-event and intra-event residuals with respect to Mw, ky/PGA, ln(PGA), ln(PGV) and ln(Ia) for models developed using the median displacements in all
orientations as a function of the larger IMs of the two recorded components. (a) Median D and larger IMs, and RS09(PGA, Mw) form, (b) Median D and larger IMs, and SR08(PGA, PGV) form.

developed using the computed displacements of the recorded orthogo-
nal ground motion components as a function of IMs in the time histo-
ries. This is different from the observations for the relationship as a
function of IMsRotD50. The standard deviations are represented using
Eqs. (6a) and (6b), and the coefficients are given in Table 3. The proba-
bility of zero displacement is represented using Eqs. (8a) and (8b) based
on the logistic regression analyses, and the coefficients are listed in
Tables 1 and 2. A comparison between the predicted probabilities of
zero displacement for different relationships is shown in Fig. 19 for a se-
lected scenario (Mw = 6.5, Rjb = 15 km, PGA = 0.16 g, PGV =
11.9 cm/s). The probability of zero displacement from the model devel-
oped by the median displacement is larger than that from the model de-
veloped by the maximum displacement. This observation is consistent
with the predicted sliding displacement.
 J. Song et al. / Engineering Geology 222 (2017) 124–139 137

Fig. 16. Distributions of inter-event and intra-event residuals with respect to Mw, ky/PGA, ln(PGA), ln(PGV) and ln(Ia) for models developed using the maximum displacements in all
orientations as a function of the larger IMs of the two recorded components. (a) Max D and larger IMs, and RS09(PGA, Mw) form, (b) Max D and larger IMs, and SR08(PGA, PGV) form.

6. Summary and conclusions
To ensure consistency between the derivation of ground motion IMs
and their application to the assessment of the seismic performance of
slopes, this paper provides an orientation-independent estimation of
earthquake-induced sliding displacements of slopes.
Rigid-sliding-block displacements are computed for all the horizon-
tal components with various rotation angles for the selected ground
motion records. The distribution of sliding displacements in all orienta-
tions is analyzed. In most cases, the maximum displacement can be
100% larger than the minimum displacement through all orientations,
which indicates that sliding displacements can be strongly dependent
138 J. Song et al. / Engineering Geology 222 (2017) 124–139

Fig. 17. Comparison of the predicted median sliding displacement of models developed using the computed displacements of the recorded orthogonal ground motion components as a
function of IMs in the time histories, using the median and maximum displacements in all orientations as a function of the larger IMs of the two recorded components, respectively
(Mw = 7, Rjb = 6 km, PGA = 0.33 g, PGV = 30 cm/s): (a) The RS09(PGA, Mw) form; (b) The SR08(PGA, PGV) form.

on orientation for a given a set of directionally-dependent ground mo-
tion IMs.
Empirical predictive relationships for sliding displacements are de-
veloped based on various definitions of both sliding displacements
and ground motion IMs. We found that the predictive relationships
are independent of the orientations of the displacement data used in
their development. The predictive relationships for the orientation-
independent earthquake-induced sliding displacement of slopes are de-
veloped based on directionally-dependent IMs which can be derived ei-
ther from ground motion prediction equations (GMPEs) or from
ShakeMap. In general, we show that traditional models (i.e., those de-
veloped using the two as-recorded ground motions) can be used to pre-
dict median slope displacements. However, the predicted standard
deviations are different. The predictive relationship for the maximum
displacement in all orientations is provided as an alternative. The pre-
dictive relationships using the larger value of IMs in the two recorded
components are also provided to estimate landslides triggered by a
past earthquake based on the shaking estimated from ShakeMap.
This study aims at providing a simplified method to include the ef-
fects of ground motion directionality on the assessment of
earthquake-induced landslide hazard. The proposed relationships are
more appropriate for regional-level studies for landslide hazard map-
ping. The empirical relationships should not replace careful three-
dimensional site-specific studies that rigorously include consideration
of ground motion directionality. The directionality associated with
slopes (3-D geometry) is not considered in the study. This is important
and should be considered, especially in site-specific analyses. Although
the directionality of the motion relative to the orientation of a slope gen-
erally cannot be predicted a priori, the work in this study could provide
engineers with a better understanding of the sensitivity of the displace-
ment to azimuthal variations in sliding displacements and the proposed
relationships, which consider ground motion directionality, can be used
to help identify the range of sliding displacements that may occur. The
maximum displacement model can provide an estimation of the
upper bound of the sliding displacements. The orientation-
independent relationships provided in this paper are not intended to re-
place current models, but are a complement to these models. Different
users may be interested in different definitions of the displacement,
and the examination on the directionality of sliding displacements in-
creases the awareness of possible biases related to ground motion
directionality.
Acknowledgements
This research has been supported by the National Natural Science
Foundation of China (grant no. 41602280 and 41630638), National
Key Basic Research Program of China (grant no. 2015CB057901), Na-
tional Key Research and Development Program of China (grant no.
2016YFC0800205), China Postdoctoral Science Foundation (grant no.
2016M601708), the Fundamental Research Funds for the Central Uni-
versities in China (grant no. 2016B01114). These supports are gratefully
acknowledged. The authors would also like to acknowledge anonymous
reviewers for their constructive comments. Their comments improved
the quality of this manuscript.

Fig. 18. Comparison of the standard deviation of models developed using the computed displacements of the recorded orthogonal ground motion components as a function of IMs in the
time histories, using the median and maximum displacements in all orientations as a function of the larger IMs of the two recorded components, respectively: (a) The RS09(PGA, Mw) form;
(b) the SR08(PGA, PGV) form.
 J. Song et al. / Engineering Geology 222 (2017) 124–139 139

Fig. 19. Comparison of predicted probability of zero displacement of models developed using the median and maximum displacements in all orientations as a function of the larger IMs of
the two recorded components, respectively (Mw = 6.5, Rjb = 15 km, PGA = 0.16 g, PGV = 11.9 cm/s): (a) The RS09(PGA, Mw) form; (b) the SR08(PGA, PGV) form.

References Newmark, N.M., 1965. Effects of earthquakes on dams and embankments. Géotechnique
15 (2), 139–160.
R, 2015. A Programming Environment for Data Analysis and Graphics, Version 3.2.3.
Ambraseys, N.N., Menu, J.M., 1988. Earthquake-induced ground displacements. Earthq.
Rathje, E.M., Antonakos, G., 2011. A unified model for predicting earthquake-induced slid-
Eng. Struct. Dyn. 16 (7), 985–1006.
ing displacements of rigid and flexible slopes. Eng. Geol. 122 (1–2), 51–60.
Ancheta, T.D., Bozorgnia, Y., Darragh, R., Silva, W.J., Chiou, B., Stewart, J.P., Boore, D.M.,
Rathje, E.M., Saygili, G., 2008. Probabilistic seismic hazard analysis for the sliding displace-
Graves, R., Abrahamson, N.A., Campbell, K.W., Idriss, I.M., Youngs, R.R., Atkinson,
ment of slopes: scalar and vector approaches. J. Geotech. Geoenviron. Eng. ASCE 134
G.M., 2012. PEER NGA-West2 database: A database of ground motions recorded in
(6), 804–814.
shallow crustal earthquakes in active tectonic regions. 15th Word Conference on
Rathje, E.M., Saygili, G., 2009. Probabilistic assessment of earthquake-induced sliding dis-
Earthquake Engineering, Lisbon, Portugal.
placements of natural slopes. Bull. N. Z. Soc. Earthq. Eng. 42 (1), 18–27.
Boore, D.M., 2010. Orientation-independent, nongeometric-mean measures of seismic in-
Rathje, E.M., Saygili, G., 2011. Estimating fully probabilistic seismic sliding displacements
tensity from two horizontal components of motion. America]–>Bull. Seismol. Soc.
of slopes from a pseudoprobabilistic approach. J. Geotech. Geoenviron. Eng. ASCE 137
Am. 100 (4), 1830–1835.
(3), 208–217.
Boore, D.M., Stewart, J.P., Seyhan, E., Atkinson, G.M., 2014. NGA-West2 equations for
Rathje, E.M., Wang, Y., Stafford, P.J., Antonakos, G., Saygili, G., 2014. Probabilistic assess-
predicting PGA, PGV, and 5% damped PSA for shallow crustal earthquake. Earthquake
ment of the seismic performance of earth slopes. Bull. Earthq. Eng. 12 (3),
Spectra 30 (3), 1057–1085.
1071–1090.
Bradley, B.A., Baker, J.W., 2014. Ground motion directionality in the 2010–2011 Canter-
Rodriguez-Marek, A., Song, J., 2016. Displacement-based probabilistic seismic demand
bury earthquakes. Earthq. Eng. Struct. Dyn. 44 (3), 371–384.
analyses of earth slopes in the near-fault region. Earthquake Spectra 32 (2),
Bray, J.D., Travasarou, T., 2007. Simplified procedure for estimating earthquake-induced
1141–1163.
deviatoric slope displacements. J. Geotech. Geoenviron. Eng. ASCE 133 (4), 381–392.
Saygili, G., Rathje, E.M., 2008. Empirical predictive models for earthquake-induced sliding
Del Gaudio, V., Wasowski, J., 2011. Advances and problems in understanding the seismic
displacements of slopes. J. Geotech. Geoenviron. Eng. ASCE 134 (6), 790–803.
response of potentially unstable slopes. Eng. Geol. 122 (122), 73–83.
Shahi, S.K., Baker, J.W., 2014. NGA-West2 models for ground-motion directionality. Earth-
Du, W., Wang, G., 2014. Fully probabilistic seismic displacement analysis of spatially dis-
quake Spectra 30 (3), 1285–1300.
tributed slopes using spatially correlated vector intensity measures. Earthq. Eng.
Song, J., Rodriguez-Marek, A., 2015. Sliding displacement of flexible earth slopes subject
Struct. Dyn. 43 (5), 661–679.
to near-fault ground motions. J. Geotech. Geoenviron. Eng. ASCE 141 (3), 04014110.
Hsieh, S.Y., Lee, C.T., 2011. Empirical estimation of the Newmark displacement from the
Song, J., Gao, G.Y., Rodriguez-Marek, A., Rathje, E.M., 2016. Seismic assessment of the rigid
Arias intensity and critical acceleration. Eng. Geol. 122 (1–2), 34–42.
sliding displacements caused by pulse motions. Soil Dyn. Earthq. Eng. 82, 1–10.
Jibson, R.W., 2007. Regression models for estimating coseismic landslide displacement.
Wilson, R.C., Keefer, D.K., 1983. Dynamic analysis of a slope failure from the 6 August 1979
Eng. Geol. 91 (2–4), 209–218.
Coyote Lake, California earthquake. Bull. Seismol. Soc. Am. 73, 863–877.
Jibson, R.W., Harp, E.L., Michael, J.A., 2000. A method for producing digital probabilistic
Worden, C.B., Wald, D., 2016. ShakeMap Manual Online: Technical Manual, User's Guide,
seismic landslide hazard maps. Eng. Geol. 58, 271–289.
and Software Guide. U. S. Geological Survey. http://usgs.github.io/shakemap/ 10.
Lee, J., Green, R.A., 2015. Empirical predictive relationship for seismic lateral displacement
1234/012345678.
of slopes. Géotechnique 65 (5), 374–390.
Makdisi, F.I., Seed, H.B., 1978. Simplified procedure for estimating dam and embankment
earthquake induced deformations. J. Geotech. Eng. Div. ASCE 104 (GT7), 849–868.