Engineering Geology 122 (2011) 5160

Contents lists available at ScienceDirect

Engineering Geology
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e n g g e o

A unied model for predicting earthquake-induced sliding displacements of rigid and

exible slopes
Ellen M. Rathje , George Antonakos
University of Texas at Austin, 1 University Station C1792, Austin, TX 78712 USA

a r t i c l e

i n f o

Article history:
Received 26 February 2010
Received in revised form 12 October 2010
Accepted 17 December 2010
Available online 24 December 2010
Keywords:
Earthquake
Landslide
Seismic hazards
Seismic slope stability

a b s t r a c t
Permanent sliding displacement represents a common damage parameter for evaluating the seismic stability
of slopes. Recently developed empirical models for the sliding displacement of shallow (rigid) sliding masses
have demonstrated that including multiple ground motion parameters in the predictive model (e.g., peak
ground acceleration and peak ground velocity) improves the displacement prediction and reduces it
uncertainty. A unied framework is developed that extends these empirical displacement models for
application to exible sliding masses, where the dynamic response of the sliding mass is important. This
framework includes predicting the seismic loading for the sliding mass in terms of the maximum seismic
coefcient (kmax) and the maximum velocity of the seismic coefcient-time history (kvelmax). The predictive
models are a function of the peak ground acceleration (PGA), peak ground velocity (PGV), the natural period
of the sliding mass (Ts), and the mean period of the earthquake motion (Tm). The empirical predictive models
for sliding displacement utilize kmax and kvelmax in lieu PGA and PGV, and include a term related to the
natural period of the sliding mass. This unied framework provides a consistent approach for predicting the
sliding displacement of rigid (Ts = 0) and exible (Ts N 0) slopes.
2011 Elsevier B.V. All rights reserved.

1. Introduction
Permanent sliding displacement represents a common damage
parameter for evaluating the seismic stability of slopes. This
displacement represents the cumulative, downslope movement of a
sliding mass due to earthquake shaking. The magnitude of sliding
displacement relates well with observations of seismic performance
of slopes (e.g., Jibson et al., 2000), and thus has been a useful
parameter in seismic design and hazard assessment.
Fig. 1 outlines the process commonly used to compute the
earthquake-induced sliding displacement (D) of a slope with yield
acceleration, ky (ky = seismic coefcient that when multiplied by the
weight of the sliding mass and applied to the slope yields a factor of
safety of 1.0). If the sliding mass is relatively shallow and stiff, a rigid
sliding block analysis is appropriate. In this case, the natural period of
the sliding mass (Ts) is essentially zero and the dynamic response of
the sliding mass can be ignored. The seismic loading is simply the
acceleration-time (at) history at the base of the sliding mass, with
the destabilizing force-time history (F(t))on the slope equal to the at
history (in units of gravity, g) times the weight of the sliding mass.
Seismic loading parameters for the slope can be derived that

Corresponding author. Tel.: + 1 5122323683; fax: + 1 5124716548.

E-mail addresses: e.rathje@mail.utexas.edu (E.M. Rathje), antonakosg@gmail.com
(G. Antonakos).
0013-7952/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.enggeo.2010.12.004

represent various ground motion characteristics (GM) of an acceleration-time history, such as the peak ground acceleration (PGA), peak
ground velocity (PGV), Arias Intensity (Ia), etc. Generally, these
ground motion parameters are specied based on ground motion
prediction equations (e.g., Next Generation Attenuation (NGA)
models) and probabilistic seismic hazard analysis. The seismic loading
parameters can be used, along with the ky of the slope, to predict D
from empirical models (e.g., Jibson, 2007; Saygili and Rathje, 2008).
Alternatively, a suite of acceleration-time histories can be selected
that represents the expected ground shaking at the site and each time
history can be used to directly compute D for the given ky. The Java
program by Jibson and Jibson (2003) makes performing these
calculations quite easy, although the acceleration-time histories
must be selected appropriately.
Deeper and/or softer sliding masses are exible and have natural
periods greater than zero, such that the rigid sliding block model is not
appropriate. In these cases, the dynamic response of the exible
sliding mass must be taken into account (Fig. 1). Two-dimensional
nite element analysis can be used to model this dynamic response, or
alternatively the sliding mass at its maximum thickness can be
modeled as a one-dimensional soil column. Previous research (e.g.,
Rathje and Bray, 2001; Vrymoed and Calzascia, 1978) has shown that
the one-dimensional simplication provides an adequate estimate of
the seismic loading for deeper sliding masses. A decoupled sliding
block analysis (e.g., Makdisi and Seed, 1978; Bray and Rathje, 1998)
computes the dynamic response of the sliding mass without any
consideration of the sliding displacement, and then uses the results of

52

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

Fig. 1. Approaches for computing earthquake-induced sliding displacements for rigid and exible sliding masses.

the dynamic response analysis to compute the sliding displacement. A

coupled analysis (e.g., Rathje and Bray, 1999, 2000) simultaneously
computes the dynamic and sliding responses. Within either approach,
the seismic loading time history for the sliding mass is related to the
seismic coefcient (k)-time history, in which k represents the average
acceleration within the sliding mass as well as the shear force at the
base of the sliding mass. The destabilizing force-time history (F(t)) is
then simply equal to the k-time history times the weight of the sliding
mass. For a coupled analysis, k cannot exceed ky, and the dynamic
equations of equilibrium change during sliding to enforce this
equilibrium condition. For a decoupled analysis, the k-time history
may exceed ky, and the k-time history is used in a rigid sliding block
analysis in lieu of the acceleration-time to compute displacements.
It has become common practice to use rigid sliding block analyses
for sliding masses with Ts below some threshold, while using a exible
sliding block analysis for Ts greater than some threshold (Jibson,
2011-this issue). In actuality, the response of slopes does not abruptly
change from rigid to exible at some value of Ts, but rather there is a
transition from rigid to exible behavior as Ts increases. This paper
develops a unied framework that models the full range of dynamic
response conditions from rigid through exible sliding mass behavior.
This unied approach predicts the seismic loading parameters and
permanent displacements of sliding masses as a function of Ts (as
dened by the one-dimensional period computed for the maximum
thickness of the sliding mass), such that the approach tracks the
response from rigid conditions (Ts = 0) to very exible conditions
(Ts 0). The unied model is also built upon recently developed
empirical models for rigid sliding displacement (i.e., Saygili and
Rathje, 2008; Rathje and Saygili, 2009), which use PGA and PGV to
predict sliding displacement. The unied approach denes the seismic
loading parameters in terms of these same ground motion parameters, except that these parameters are computed from a k-time
history rather than an acceleration-time history. Predictive models for
these seismic loading parameters are provided, and empirical models
that predict D for rigid and exible conditions are developed.

2. Rigid sliding block displacements

A large number of empirical models are available that predict the
sliding displacement of rigid sliding masses. Newmark (1965) rst
proposed using rigid sliding block displacement to assess the seismic
performance of slopes, and later various researchers (e.g., Sarma,
1975; Franklin and Chang, 1977; Sarma, 1980; Ambraseys and Menu,
1988; Yegian et al., 1991, and Ambraseys and Srbulov, 1994; Sarma
and Kourkoulis, 2004) developed charts and/or predictive equations
for D. Recent research (e.g., Watson-Lamprey and Abrahamson, 2006;
Jibson, 2007; Bray and Travasarou, 2007) includes more robust
empirical models developed from larger ground motion datasets. The
unied model presented here is based on the recent empirical
displacement models of Saygili and Rathje (2008) and Rathje and
Saygili (2009), and thus these models are discussed in detail below.
Saygili and Rathje (2008) presented a suite of empirical predictive
models for the sliding displacement of rigid slopes, and these models
considered various ground motion parameters, such as PGA, PGV, Ia,
and mean period (Tm, Rathje et al., 2004), as well as combinations of
these ground motion parameters. Rathje and Saygili (2009) slightly
modied the PGA model from Saygili and Rathje (2008) by adding a
term related to earthquake magnitude (M). The Rathje and Saygili
(2009) modication is repeated in Rathje and Saygili (2011). The
recommended single (scalar) ground motion parameter model is the
(PGA, M) model from Rathje and Saygili (2009), and the recommended two (vector) ground motion parameter model is the (PGA,
PGV) model from Saygili and Rathje (2008). For simplicity, these
models will be called the SR08/RS09 models.
Fig. 2 plots predicted values of D from the SR08/RS09 models as a
function of ky for different earthquake scenarios of M = 6, 7, and 8,
each with a site-to-source distance (R) equal to 2 km. Also considered
are rock (Vs30 = 760 m/s) and soil (Vs30 = 400 m/s) site conditions.
The Boore and Atkinson (2008) ground motion prediction equation
was used to predict the median values of PGA and PGV for each
scenario, and these values are listed in Table 1. Note that the PGA

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

53

different as earthquake magnitude, and the associated PGV, increases.

The (PGA, PGV) model generally predicts smaller displacements for
these scenarios, on the order of 30 to 40% smaller. These differences
are caused by the fact that the empirical models were developed using
rock and soil motions from the Next Generation Attenuation (NGA)
ground motion dataset. Because soil motions tend to have larger PGV
values than rock motions and the (PGA, M) model does not include
the effects of PGV, the (PGA, M) model predicts larger displacements
than the (PGA, PGV) model for rock sites. This effect is demonstrated
in Fig. 2(b), which uses soil ground motion parameters (Vs30 = 400 m/s)
to predict D. When utilizing ground motion parameters for soil sites, the
differences between the (PGA, M) and (PGA, PGV) models are much
smaller.
In addition to the differences in median displacements from the
(PGA, M) and (PGA, PGV) models, there are signicant differences in
the standard deviations of the predictions. The standard deviation
(lnD) for each model increases with increasing ky/PGA, with values
ranging between 0.75 and 1.0 (in natural log units) for the (PGA, M)
model (Rathje and Saygili, 2009), and values ranging between 0.4 and
0.9 for the (PGA, PGV) model (Saygili and Rathje, 2008). To illustrate
these differences, the median and 1lnD displacements for the (PGA,
M) and (PGA, PGV) models are shown in Fig. 3 for the M = 7, R = 2 km
scenario event with Vs30 = 760 m/s. At larger ky, the 1lnD range in
displacement is close to a factor of 10 for the (PGA, M) model, while at
smaller ky the 1lnD range represents a factor of about 5. For the
(PGA, PGV) model, the 1lnD displacement range is much smaller by
comparison, with the range representing a factor of 4.0 at larger ky
and a factor of 2.5 at smaller ky. Thus, there is signicantly less
uncertainty in the displacement prediction when PGV is used in the
displacement calculation.
One limitation of the SR08/RS09 empirical models is that they only
represent rigid sliding block conditions, yet exible sliding block
conditions are very common. It is proposed to use a framework similar
to SR08/RS09 for exible sliding conditions, but application of this
framework to exible sliding conditions rst requires appropriate
quantication of the seismic loading.
3. Seismic loading parameters for rigid and exible sliding masses

Fig. 2. Rigid sliding block displacements calculated from the SR08/RS09 (PGA, M) and
(PGA, PGV) models for different earthquake scenarios and site conditions.

values become similar (i.e., saturate) at larger magnitudes, while the

PGV values continue to increase (Table 1). Additionally, the soil
conditions amplify PGV more than they amplify PGA. Predicted values
of D are shown in Fig. 2(a) for both the (PGA, M) and (PGA, PGV)
models and Vs30 = 760 m/s. For M = 6, the (PGA, M) and (PGA, PGV)
models predict similar displacements, but the displacements become

Table 1
Ground motion parameters for each earthquake scenario and site conditions.
Vs30 = 760 m/s

Vs30 = 400 m/s

R (km)

PGA (g)

PGV (cm/s)

Tm (s)

PGA (g)

PGV (cm/s)

Tm (s)

6.0
7.0
8.0

2
2
2

0.30
0.43
0.48

19
42
74

0.37
0.44
0.46

0.34
0.47
0.52

27
59
102

N/A
N/A
N/A

First consider the seismic loading parameters for rigid sliding blocks.
The SR08/RS09 models use (PGA, M) and (PGA, PGV) to characterize the
seismic loading for these systems. The recorded acceleration-time
history from the GIL067 station during the 1989 Loma Prieta (M= 6.9)
earthquake is shown in Fig. 4(a) (PGA = 0.36 g), and the velocity-time
history derived from numerical integration of the acceleration-time
history is shown in Fig. 4(b) (PGV = 29 cm/s). For a rigid sliding mass
subjected to the GIL067 motion, the acceleration-time history represents the seismic loading and the characteristics of the both the
acceleration- and velocity-time histories will inuence the level of
induced displacement.
The seismic loading for a exible sliding mass subjected to the
GIL067 motion is not the acceleration-time history due to the dynamic
response of the sliding mass. Rather, the seismic loading is the k-time
history (e.g., Seed and Martin, 1966; Bray and Rathje, 1998), which
represents the average acceleration within the sliding mass as well as
the shear force at the base of the sliding mass. Consider the dynamic
response of a 30-m thick sliding mass (H = 30 m) with a shear wave
velocity of 250 m/s (Vs = 250 m/s) and associated site period of 0.5 s
(Ts = 4 H/Vs = 0.5 s). The k-time history for this site, computed using
one-dimensional, equivalent-linear site response analysis, is shown in
Fig. 4(c). The k-time history displays much less high frequency motion
than the acceleration-time history due to the averaging of accelerations within the sliding mass. Additionally, its peak value (kmax) is
smaller than the input PGA (kmax = 0.12 g versus PGA = 0.36 g). The
k-time history and its associated kmax represent the appropriate
seismic loading for this exible sliding mass.

54

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

To use the SR08/RS09 predictive models for exible sliding masses,

the appropriate seismic loading parameters must be specied. Based
on the above descriptions, kmax should be used to replace PGA in the
SR08/RS09 models and kvelmax should be used to replace PGV.
Earthquake magnitude does not need to be modied. Predictive
models for kmax and kvelmax are required such that engineers do not
need to perform dynamic response analysis to estimate these seismic
loading parameters. These predictive models are along the same lines
as the design charts for kmax developed by Bray and Rathje (1998) and
Bray et al. (1998).
Predictive models for kmax and kvelmax are developed based on
one-dimensional site response calculations of ve sites subjected to
80 input motions using the equivalent-linear site response code Strata
(Kottke and Rathje, 2008). The sites consist of one 15-m prole
(Vs = 400 m/s), two 30-m proles (Vs = 400 m/s and 250 m/s) and
two 100-m proles (Vs = 400 m/s and 265 m/s). The resulting values
of site period (Ts) are 0.15 s, 0.30 s, 0.48 s, 1.0 s, and 1.5 s. The
nonlinear soil properties are modeled with the curves of Darendeli
and Stokoe (2001) using PI = 0 and appropriate values of conning
pressure based on the thicknesses of the proles. The 80 input
motions represent motions from M = 6 to 7.9 earthquakes recorded a
distances between 0.1 and 60 km with Vs30 = 200 to 1000 m/s.
However, most of the Vs30 values are between 400 and 800 m/s. The
input PGA values range from 0.02 to 1.0 g, and the input PGV values
range from 1.2 cm/s to 70 cm/s. k-time histories were computed at
the base of each one-dimensional site prole, from which kmax and
kvelmax values were derived. Further details about the analyses
performed can be found in Antonakos (2009).
The computed kmax values are plotted versus input PGA in Fig. 5(a)
for the 400 analyses performed. There is trend of increasing kmax with
increasing PGA, although at a decreasing rate and with more scatter at
larger values of PGA. Bray and Rathje (1998) investigated the ratio of
kmax to PGA and showed that the period ratio (Ts/Tm) has a strong
inuence on this value. kmax/PGA is plotted versus Ts/Tm in Fig. 5(b),
and several important observations can be made. First, kmax/PGA
approaches 1.0 as Ts/Tm approaches 0.1. This trend is consistent with
kmax = PGA for rigid sliding masses, and indicates that Ts/Tm = 0.1
essentially represents rigid sliding conditions. Next, kmax is greater
than PGA at moderate period ratios (Ts/Tm = 0.1 to 0.7), while kmax is
less than PGA at larger period ratios. The data in Fig. 5(b) are shown
for different ranges of input PGA. These data indicate that the ratio of
kmax/PGA decreases with increasing input PGA.
A predictive equation for kmax/PGA is developed to model these
trends. This model assumes a log-normal distribution for kmax/PGA,
and predicts ln(kmax/PGA) as a function of ln[Ts/Tm] and PGA.
Fig. 3. Median and 1lnD rigid sliding block displacements predicted by the SR08/
RS09 (PGA, M) and (PGA, PGV) models for M = 7, and R = 2 km.

ln kmax = PGA = 0:4590:702PGA

for Ts = Tm 0:1

f lnTs = Tm  = 0:1g
+ 0:228 + 0:076PGA
2

f lnTs =Tm =0:1g

In the same way that an acceleration-time history can be
numerically integrated to generate a velocity-time history, the ktime history can be numerically integrated to generate a velocity-time
history of the k-time history. This velocity is called kvel, and while it
does not represent the average velocity of motion within the sliding
mass, it does provide information regarding the frequency content of
the k-time history. The maximum value of the kvel-time history is
called kvelmax. As expected, the kvel-time history contains less high
frequency motion than the velocity-time history. Surprisingly,
however, the value of kvelmax (31 cm/s) is similar to the value of
PGV (29 cm/s). Because the integrated kvel-time history is inuenced by both the amplitude and frequency content of the k-time
history, the increase in long period motion in the k-time history is
balanced by the reduction in its peak such that kvelmax is similar in
amplitude to PGV.

lnkmax = PGA = 0

for Ts = Tm b 0:1

The standard deviation for this model in natural log units is 0.25.
Given the predicted value of kmax/PGA and the input motion PGA, kmax
can be estimated.
Fig. 6 presents the model predictions of kmax/PGA as a function of
input PGA and Ts/Tm. Generally, kmax/PGA is greater than 1.0 at
smaller values of Ts/Tm, and then falls below 1.0 at larger period ratios.
The range of Ts/Tm values that predict kmax greater than PGA (i.e.,
kmax/PGA N 1.0) decreases with increasing PGA, and at large input
intensities kmax is less than PGA at all period ratios. All curves predict
kmax/PGA = 1.0 for Ts/Tm 0.1, i.e., rigid sliding conditions.
Bray and Rathje (1998) developed a predictive model for kmax that
uses a power law relationship to predict a normalized kmax (kmax/

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

55

Fig. 4. (a) Acceleration and (b) velocity-time histories for a rigid sliding block. (c) k-time history and (d) kvel-time history for a exible sliding mass with Ts = 0.5 s.

[NRFPGA]) as a function of Ts/Tm. The power law relationship results

in a log-linear relationship between kmax and Ts/Tm for a constant
value of PGA. The PGA normalization effectively scales kmax linearly
with PGA, although the nonlinear response factor (NRF) recommended by Bray and Rathje (1998) takes into account some nonlinear
scaling. The NRF is a parameter that decreases with increasing input

PGA, such that for a given Ts/Tm kmax/PGA decreases with increasing
input PGA. Bray and Rathje (1998) state that their model is
appropriate for Ts/Tm greater than 0.5. The predictive model from
Eq. (1) is compared to the predictions from Bray and Rathje(1998) in
Fig. 7 for input PGA values of 0.2 and 0.8 g. For PGA = 0.2 g, the Bray
and Rathje (1998) predictions agree favorably with Eq. (1) in the
period range of 0.5 to 2.0 where the log-linear shape is most valid. At
larger period ratios the Bray and Rathje (1998) model predicts larger
values of kmax because the log-linear shape cannot represent the
nonlinear relationship. The second-order polynomial used in Eq. (1)
more accurately models the variation of kmax/PGA over a wide range of
period ratios. For PGA=0.8 g, the Bray and Rathje (1998) model is
consistently larger than Eq. (1), although the difference is most
pronounced at large period ratios. This difference indicates that the NRF
factor incorporated in Bray and Rathje (1998) does not model as much soil
nonlinearity as the model developed in this study.
The additional information required to use the SR08/RS09
predictive models is k-velmax. Fig. 8(a) shows the computed values
of kvelmax versus PGV. Based on the example shown in Fig. 4, we
should not expect signicant differences in kvelmax and PGV. A
signicant amount of the data in Fig. 8(a) centers about a 1:1 line, but
there are some considerably smaller values. To further explore this
variability, the ratio of kvelmax to PGV was computed for each data
point and plotted versus Ts/Tm (Fig. 8(b)) for different ranges on input
PGA. The kvelmax/PGV data display similar trends to the kmax/PGA

Fig. 5. (a) Variation of kmax with PGA, and (b) kmax/PGA verus Ts/Tm.

Fig. 6. kmax/PGA model predictions from Eq. (1).

56

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

PGV is not signicant at small period ratios, only the coefcient for the
second-order term is a function of PGA. The predictive model for k
velmax/PGV is given by:
ln kvelmax = PGV= 0:240f ln Ts = Tm  = 0:2g

for Ts=Tm 0:2

+ 0:0910:171PGA
2

flnTs =Tm =0:2g

lnkvelmax = PGV= 0

Fig. 7. Comparisons of kmax predictions from Eq. (1) and from Bray and Rathje (1998).

data. The data indicate kvelmax equal to PGV at very small period
ratios (Ts/Tm 0.2 in this case), kvelmax greater than PGV at smaller
period ratios (Ts/Tm = 0.2 to 1.5) and kvelmax less than PGV at larger
period ratios. The range of period ratios where kvelmax/PGV N 1.0 is
larger than the range of period ratios where kmax/PGA N 1.0. Again,
there is an input amplitude effect, with smaller values of kvelmax/
PGV observed at larger values of input PGA. However, this input
amplitude effect is not pronounced at smaller period ratios.
A predictive model for kvelmax/PGV was developed with a similar
functional form to Eq. (1). Because the input PGA effect for kvelmax/

Fig. 8. (a) Variation of kvelmax with PGV, and (b) kvelmax/PGV versus Ts/Tm.

for Ts = Tm b0:2

The standard deviation for this model in natural log units is 0.25.
Fig. 9 presents the model predictions of kvelmax/PGV as a function
of input PGA and Ts/Tm. At period ratios less than 0.3 the predicted
values of kvelmax/PGV are similar for all input intensities. At larger
period ratios, kvelmax/PGV is smaller for larger input intensities. The
model predicts kvelmax/PGV = 1.0 (i.e., rigid sliding conditions) for
Ts/Tm 0.2.
4. Displacement predictions for rigid and exible sliding masses
The objective of this study is to modify the SR08/RS09 rigid block
empirical models such that they can be used to predict the decoupled
displacements of rigid and exible sliding systems. The initial
hypothesis is that the original SR08/RS09 empirical models can be
used, but with PGA replaced by kmax and PGV replaced by kvelmax. To
test this hypothesis, decoupled sliding displacements were calculated
using the computed k-time histories for the ve sites and 80 input
motions (400 time histories). Displacements were calculated for
ky = 0.04, 0.08, 0.12, and 0.16. The resulting dataset included 569 nonzero values of displacement (i.e., instances where ky b kmax). These
values of displacement were compared with the median values
predicted by the SR08/RS09 empirical models given the computed
values of kmax and kvelmax for each calculated k-time history.
Additionally, rigid sliding block displacements were computed for the
80 input time histories and the four values of ky for comparison with
the median values predicted by the SR08/RS09 empirical models.
The residuals (i.e., ln(data)ln(predicted)) of the computed values
of D (i.e., data) with respect to the empirically predicted values of D
were calculated for both the (PGA, M) model and the (PGA, PGV)
model. For both models, the average residuals over the complete
dataset are greater than 0.0, with an average of 0.24 for the (PGA, M)
model and an average of 0.42 for the (PGA, PGV) model. These positive
values indicate that the computed values of D for these exible sliding
masses are larger, on average, than the values predicted by the SR08/
RS09 empirical models. The difference is caused by the fact that the
frequency content of a k-time history is signicantly different than for

Fig. 9. kvelmax/PGV model predictions from Eq. (2).

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

an acceleration-time history (Fig. 4), which results in larger displacements. While kvelmax attempts to take into account this difference in
frequency content, the time histories in Fig. 4 demonstrate that PGV
and kvelmax do not vary signicantly from one another although the
k-time histories display signicantly different frequency contents.
Thus, the original SR08/RS09 empirical models require an additional
modication to capture this effect.
The residuals for the rigid sliding block displacement were
investigated to evaluate how the selected ground motion dataset may
be inuencing the results. The average residuals for rigid sliding block
conditions should be equal to 0.0, because the SR08/RS09 model
represents rigid sliding conditions. For the (PGA, M) model the average
residual for rigid sliding (Ts = 0.0 s) was 0.8, which signies that, on
average, the computed values of D from the 80 motion dataset are
smaller than those predicted by SR08/RS09. The computed values of D
are smaller than predicted by SR08/RS09 because the average Vs30 for
the motions used in this study (Vs30 ~ 550 m/s) is larger than the average
for those used in the SR08/RS09 studies (Vs30 ~ 400 m/s). Motions from
sites with larger Vs30 display less long period energy, which results in
smaller displacements. For the (PGA, PGV) model, the average residual
for rigid sliding conditions was essentially zero. The Vs30 effect is not
apparent for this model because the inclusion of PGV takes into account
the different frequency contents for rock and soil motions.
The residuals for the exible sliding masses were investigated
together with the residuals for rigid sliding masses to identify the site/
ground motion parameters that inuence the difference between the
computed and predicted displacements. Fig. 10 plots the residuals
versus site period for the two displacement models. These data

57

indicate that the residuals increase with increasing site period (Ts),
but at a decreasing rate. The residuals increase with Ts because sliding
masses with larger values of Ts generate k-time histories with more
long period energy that lead to larger displacements. The scatter at
any one period in Fig. 10 is larger for the (PGA, M) model than the
(PGA, PGV) model, and this observation is consistent with the relative
values of lnD reported for the two models.
For the (PGA, M) model (Fig. 10(a)), the average residual is equal to
0.8at Ts =0.0 s (rigid conditions) and increases to 1.95 at Ts =1.5 s. A
positive residual of 1.95 corresponds to computed displacements that are
7 times larger than those predicted by the empirical model. A second
order polynomial was t to the average residuals, and this expression can
be used to modify the SR08/RS09 (PGA, M) rigid sliding block model for
the effects of decoupled, exible sliding. However, the residuals in Fig. 10
(a) are inuenced by the fact that the ground motion dataset is not fully
consistent with the dataset used in the SR08/RS09 studies (i.e., different
Vs30) which causes the average residual to be non-zero at Ts =0.0 s.
Therefore, the recommended modication involves translating the curve
shown in Fig. 10(a) such that the average residual is equal to zero at
Ts =0.0 s. The resulting modication to the SR08/RS09 (PGA, M) model to
account for exible sliding is:


2
lnDflexible = ln DPGA;M + 3:69Ts 1:22Ts


lnDflexible = ln DPGA;M + 2:78

for Ts 1:5

for Ts N 1:5

where DPGA,M represents the median displacement predicted by the

(PGA, M) SR08/RS09 rigid sliding block model and Ts is the natural
period of the sliding mass. For the calculation of DPGA,M, kmax is used in
lieu of PGA.
For the (PGA, PGV) model (Fig. 10(b)), the average residual is zero
at Ts = 0.0 s and increases to a value as large as 0.70. However, the
average residuals become relatively constant at periods greater than
0.5 s. A linear relationship was t through the average residuals at
Ts 0.5 s, with no further increase modeled at larger periods. The
resulting modication to the SR08/RS09 (PGA, PGV) model to account
for exible sliding is:


lnDflexible = ln DPGA;PGV + 1:42Ts


lnDflexible = ln DPGA;PGV + 0:71

for Ts 0:5
for Ts N 0:5

where DPGA,PGV represents the median displacement predicted by the

(PGA, PGV) SR08/RS09 rigid sliding block model and Ts is the natural
period of the sliding mass. For the calculation of DPGA,PGV, kmax is used
in lieu of PGA and kvelmax is used in lieu of PGV.
After correcting the biases observed in the residuals shown in
Fig. 10, the standard deviation of lnD (lnD) was computed from the
corrected residuals. Considering that the SR08/RS09 models display a
variation of lnD with ky/PGA, the models from this study should
display a variation of lnD with ky/kmax. The computed values of lnD
are plotted versus ky/kmax in Fig. 11 for the (PGA, M) and (PGA, PGV)
models. The lnD values for the (PGA, M) model follow a linear trend
(Fig. 11(a)), and are about 10% smaller than the lnD values from the
SR08/RS09 (PGA, M) model. The reduction in standard deviation for
exible sliding masses is expected because the dynamic response
calculation lters out any high frequency peaks that contribute to the
variability in rigid block displacements. The recommended lnD
relationship for the (PGA, M) model for exible sliding masses is
given by:

Fig. 10. (a) Displacement residuals versus Ts for the SR08/RS09 (PGA, M) model and
(b) displacement residuals versus Ts for the SR08/RS09 (PGA, PGV) model.

lnD = 0:694 + 0:322ky = kmax

for PGA; M model

58

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

Fig. 11. (a) Standard deviation of lnD for exible sliding masses using revised (PGA, M)
model, (b) standard deviation of lnD for exible sliding masses using revised (PGA,
PGV) model.

The lnD values for the (PGA, PGV) model (Fig. 11(b)) are also
smaller than those from SR08/RS09 (0 to 25% smaller), particularly at
large values of ky/kmax. A revised linear relationship is used to predict
lnD for exible sliding masses for the (PGA, PGV) model:
lnD = 0:40 + 0:284ky = kmax

for PGA; PGV model

5. Example applications
To illustrate the application of the unied model for predicting the
dynamic response and sliding displacement of slopes, consider the
following example. The critical sliding mass for a slope is 20-m thick
with an average Vs = 400 m/s and resulting Ts = 0.2 s. The ky is equal
to 0.1. The design event is M = 8 and R = 2 km, with the input rock
motions described by Table 1 (PGA = 0.48 g, PGV = 74 cm/s) and with
Tm = 0.46 s (Rathje et al., 2004). Based on the site and ground motion
characterizations, Ts/Tm = 0.43.
Eqs. (1) and (2) are used to predict kmax and kvelmax based on
the PGA (0.48 g), PGV (74 cm/s), and Ts/Tm (0.43). Using these values,
Eq. (1) predicts kmax/PGA = 0.79, while Eq. (2) predicts kvelmax/
PGV = 1.08. Thus, the seismic loading for this sliding mass is predicted
as: k m a x = 0.38 g (= 0.790.48 g) and kvel m a x = 80 cm/s
(=1.0874 cm/s).
Using the seismic loading parameters of kmax = 0.38 g and M = 8
along with ky = 0.1, the SR08/RS09 (PGA, M) model predicts 63.1 cm
when kmax is used in place of PGA. This value must be adjusted using the
modication for exible sliding given in Eq. (3). For Ts = 0.2 s, this

expression predicts a median displacement value of 126 cm for exible

sliding conditions. For the SR08/RS09 (PGA, PGV) model, the appropriate
seismic loading parameters are kmax = 0.38 g and kvelmax = 80 cm/s.
Using kmax in lieu of PGA and kvelmax in lieu of PGV in the SR08/RS09
(PGA, PGV) model generates a displacement of 36.9 cm. Adjusting this
value for exible sliding and Ts = 0.2 s (Eq. (4)), the median predicted
value of displacement is 49 cm for exible sliding conditions. Note that
this value is less than half the value predicted by the SR08/RS09 (PGA, M)
model.
It is interesting to note the signicant difference between the
displacements predicted by the (PGA, M) and (PGA, PGV) models.
Depending on the seismic conditions and slope parameters, the (PGA,
M) model may predict displacements as much as 2.5 times larger than
the (PGA, PGV) model. The larger displacements for the (PGA, M)
model are a result of two issues: the neglected Vs30 effect (Fig. 2) and
the proposed modication in Eq. (3). As noted previously, Vs30
inuences the frequency content of shaking and leads to larger
displacements. Because the dataset used to develop the (PGA, M)
model included both rock and soil motions and because the (PGA, M)
model does not take into account Vs30, this model tends to predict
larger displacements. The modication in Eq. (3) was generated by
translating up the residuals in Fig. 10(a) so that the mean residual
would be zero for rigid sliding conditions. This signicant translation
leads to even larger displacements. There is uncertainty in the
decision to translate the entire curve in Fig. 10(a) based on the
residuals at Ts = 0.0 s, and thus there is less condence in the (PGA,
M) model for exible sliding. Therefore, the (PGA, PGV) model is
recommended over the (PGA, M) model for use in practice.
To illustrate the dynamic and displacement responses of slopes
under rigid through exible conditions using the developed framework, consider sliding masses with site periods (Ts) ranging from 0.0
to 1.0 s. Fig. 12 shows both the dynamic responses (i.e., kmax and k
velmax) and sliding displacements for these sliding masses subjected
to motions from M = 6, 7, and 8 earthquakes at a distance of 2 km. For
these earthquake scenarios, the median values of PGA, PGV, and Tm for
Vs30 = 760 m/s are given in Table 1. Fig. 12(a) demonstrates that kmax
for a exible system is generally smaller than kmax for a rigid system
(Ts = 0.0), except for sliding masses with very small site periods (Ts
less than about 0.1 s). The reduction in kmax with increasing Ts is
signicant, with kmax decreasing by 80% at large periods and large
input intensities. Alternatively, the reduction in kvelmax with site
period is not as dramatic. Flexible systems with periods of up to 0.4 s
display larger values of kvelmax than rigid systems (Fig. 12(b)), and
at larger periods kvelmax is never more than 35% smaller than the
rigid value. At larger periods kvelmax gets smaller, but the rate of
reduction is much smaller than for kmax.
Fig. 12(c) and (d) shows the variation of displacement as a
function of Ts for ky = 0.05 and 0.1 for the revised (PGA, PGV) model.
At shorter periods (Ts b 0.3 s for ky = 0.05, Ts b 0.15 s for ky = 0.1), the
exible systems displace more than the rigid systems due to the
enhanced amplitudes and increase in long period content of the
seismic loading induced by the dynamic response. At longer periods,
the displacements for exible systems are smaller than for rigid
systems due to the nonlinear response of the soil and the reductions in
the amplitude of the seismic loading (i.e., kmax). The period range over
which exible systems displace more than rigid systems depends on
ky, and generally this range is larger for smaller values of ky.
6. Conclusions
Evaluating the seismic performance of slopes involves predicting
sliding block displacements for critical sliding masses. Current
practice typically uses a rigid sliding block approach for shallow
sliding masses and a decoupled, exible sliding block approach for
deeper/softer sliding masses. Empirical predictive models are available to predict the sliding displacements of rigid sliding masses and

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

59

Fig. 12. (a) Predicted values of kmax as a function of Ts, (b) predicted values of kvelmax as a function of Ts, (c) predicted values of sliding displacement as a function of Ts with
ky = 0.05 for the revised (PGA, PGV) model developed in this study, and (d) predicted values of sliding displacement as a function of Ts with ky = 0.1 for the revised (PGA, PGV) model
developed in this study.

exible sliding masses, but these models do not adequately model the
transition from rigid to exible behavior.
This paper presents a unied empirical model to predict the sliding
displacements of rigid and exible sliding masses. The unied model
is an extension of the empirical models for rigid sliding masses
developed by Saygili and Rathje (2008) and Rathje and Saygili (2009).
The main advancements contributed by the SR08/RS09 models
include: (1) the use of a large ground motion dataset, (2) the addition
of a frequency content parameter (PGV) to better predict displacements, and (3) a better description of the standard deviation
associated with each model.
The unied approach involves rst predicting the seismic loading
parameters for a potential sliding mass, and then using these seismic

loading parameters to predict sliding displacement. The seismic

loading parameters are given by kmax and kvelmax, dened as the
maximum value of the k-time history and the maximum velocity of
the k-time history, respectively. A predictive model for kmax was
developed as a function of PGA and Ts/Tm, and a predictive model for
kvelmax was developed as a function of PGV, PGA, and Ts/Tm. To
predict sliding displacement, kmax is used in lieu of PGA and kvelmax
is used in lieu of PGV in the SR08/RS09 models.
In addition to the change in seismic loading parameters, the SR08/
RS09 models must be further modied to account for the differences
in frequency characteristics between acceleration-time histories and
k-time histories. This modication is a function of Ts and increases the
predicted displacement. Modication for both the (PGA, M) and (PGA,

60

E.M. Rathje, G. Antonakos / Engineering Geology 122 (2011) 5160

PGV) models are developed, but the (PGA, PGV) model is recommended because of the signicant frequency content information
provided by PGV (for rigid sliding) and by k-velmax (for exible
sliding).
References
Ambraseys, N.N., Menu, J.M., 1988. Earthquake-induced ground displacements.
Earthquake Engineering and Structural Dynamics 16, 9851006.
Ambraseys, N.N., Srbulov, M., 1994. Attenuation of earthquake-induced displacements.
J. Earthquake Engineering and Structural Dynamics 23, 467487.
Antonakos, G. 2009. "Models of Dynamic Response and Decoupled Displacements of
Earth Slopes during Earthquakes", M.S. Thesis, University of Texas at Austin, Austin,
TX.
Boore, D.M., Atkinson, G.M., 2008. Ground-motion prediction equations for the average
horizontal component of PGA, PGV and 5%-damped PSA at spectral periods
between 0.01 s and 10.0 s. Earthquake Spectra, EERI 24 (1), 99138.
Bray, J.D., Rathje, E.M., 1998. Earthquake-induced displacements of solid-waste landlls.
Journal of Geotechnical and Geoenvironmental Engineering, ASCE 124 (3),
242253.
Bray, J.D., Travasarou, T., 2007. Simplied procedure for estimating earthquake-induced
deviatoric slope displacements. J. Geotech. and Geoenvir. Engrg. Volume 133 (Issue 4),
381392.
Bray, J.D., Rathje, E.M., Augello, A.J., Merry, S.M., 1998. Simplied seismic design
procedure for lined solid-waste landlls. Geosynthetics International 5 (12),
203235.
Darendeli, M.B., Stokoe II, K.H., 2001. Development of a new family of normalized
modulus reduction and material damping curves. Geotech. Engrg. Rpt. GD01-1.
University of Texas, Austin, Texas.
Franklin, A.G., Chang, F.K., 1977. Earthquake resistance of earth and rock-ll dams: U.S.
Army Corps of Engineers Waterways Experiment Station. Miscellaneous Paper S71-17 59 pp.
Jibson, R.W., 2007. Regression models for estimating coseismic landslide displacement.
Engineering Geology 91, 209218.
Jibson, R.W., 2011. "Methods for assessing the stability of slopes during earthquakesa
retrospective". Engineering Geology 122, 4350 (this issue).
Jibson, R.W., Jibson, M.W., 2003. Java programs for using Newmark's method and
simplied decoupled analysis to model slope performance during earthquakes. U.S.
Geological Survey Open-le Report 03-005, version 1.1.
Jibson, R.W., Harp, E.L., Michael, J.A., 2000. A method for producing digital probabilistic
seismic landslide hazard maps. Engineering Geology Vol. 58, 271289.
Kottke, E.M., Rathje, E.M., 2008. Technical Manual for Strata, PEER Report 2008/10.
Pacic Earthquake Engineering Research Center, University of California at
Berkeley. 84 pp.

Makdisi, F.I., Seed, H.B., 1978. Simplied procedure for estimating dam and
embankment earthquake induced deformations. Journal of the Geotechnical
Engineering Division, ASCE 104 (GT7), 849867.
Newmark, N.M., 1965. Effects of earthquakes on dams and embankments. Geotechnique 15, 139159.
Rathje, E.M., Bray, J.D., 1999. An examination of simplied earthquake-induced
displacement procedures for earth structures. Canadian Geotechnical J. 36 (1),
7287.
Rathje, E.M., Bray, J.D., 2000. Nonlinear coupled seismic sliding analysis of earth structures.
Journal of Geotechnical and Geoenvironmental Engineering, ASCE 126 (11),
10021014.
Rathje, E.M., Bray, J.D., 2001. One and two dimensional seismic analysis of solid-waste
landlls. Canadian Geotechnical Journal 38, 850862.
Rathje, E.M., Saygili, G., 2009. Probabilistic assessment of earthquake-induced sliding
displacements of natural slopes. Bull. of the New Zealand Society for Earthquake
Engineering 42 (1), 1827.
Rathje, E.M., and G. Saygili 2011. "Pseudo-Probabilistic versus Fully Probabilistic
Estimates of Sliding Displacements of Slopes," Journal of Geotechnical and
Geoenvironmental Engineering, ASCE 137 (3).
Rathje, E.M., Faraj, F., Russell, S., Bray, J.D., 2004. Empirical relationships for frequency
content parameters of earthquake ground motions. Earthquake Spectra, EERI 20 (1),
119144.
Sarma, S.K., 1975. Seismic stability of earth dams and embankments. Geotechnique 25 (4),
743761.
Sarma, S.K., 1980. A simplied method for the earthquake resistant design of earth
dams. Dams and Earthquakes. Proc. ICE Conference, London, pp. 155160.
Sarma, S., Kourkoulis, R., 2004. Investigation into the prediction of sliding block
displacements in seismic analysis of earth dams. Proc. 13 World Conference on
Earthquake Engineering, Paper no. 1957, Vancouver, Canada.
Saygili, G., Rathje, E.M., 2008. Empirical predictive models for earthquake-induced
sliding displacements of slopes. Journal of Geotechnical and Geoenvironmental
Engineering, ASCE 134 (6), 790803.
Seed, H.B., Martin, G.R., 1966. The seismic coefcient in earth dam design. Journal of Soil
Mech. and Found. Div. 92 (SM3), 2558.
Vrymoed, J.L., Calzascia, E.R., 1978. Simplied determination of dynamic stresses in
earth dams. Proceedings, Earthquake Engineering and Soil Dynamics Conference,
ASCE, NY, pp. 9911006.
Watson-Lamprey, J., Abrahamson, N., 2006. Selection of ground motion time series and
limits on scaling. Soil Dynamics and Earthquake Engineering Vol. 26 (no. 5),
477482.
Yegian, M.K., Marciano, E.A., Ghahraman, V.G., 1991. Earthquake-induced permanent
deformations: probabilistic approach. Journal of Geotechnical Engineering 117,
3550.