Performance-Based Probabilistic Seismic

Slope Displacement Procedure

Jorge Macedo,a) Jonathan Bray,a) M.EERI, Norman Abrahamson,a) M.EERI,
and Thaleia Travasaroub)

Engineers often use simplified seismic slope displacement procedures to

evaluate the seismic performance of earth structures and natural slopes. Current
state of practice procedures typically separate the estimation of the ground
motion intensity measure (IM) from the estimate of seismic displacement
(D), given the selected IM hazard level. Thus D is estimated based on a single
IM value. A straightforward performance-based seismic slope assessment
procedure is proposed, which considers the full range of potential IM values
to estimate seismic slope displacements directly related to a hazard level.
Seismic performance is assessed through either a Newmark-type seismic dis-
placement estimate or a calibrated seismic coefficient that can be used in pseu-
dostatic slope stability analyses. The procedures were developed for a wide
range of earth systems for shallow crustal earthquakes and subduction zone
earthquakes. Currently employed simplified slope displacement procedures
do not provide consistent assessments of the actual seismic slope displacement
hazard. The proposed procedures can be readily used in practice to perform
rigorous performance-based seismic slope displacement hazard assessments.
[DOI: 10.1193/122516EQS251M]

INTRODUCTION
The seismic performance of an earth/waste system or natural slope is typically assessed
through Newmark-type sliding block analyses that provide an estimate of the expected seis-
mic displacement (e.g., Bray and Travasarou 2007, Makdisi and Seed 1978, Saygili and
Rathje 2008). Dynamic nonlinear effective stress analyses using finite element or finite dif-
ference methods with robust soil constitutive models may be employed for critical earth sys-
tems or when strength loss mechanisms, such as from soil liquefaction, are involved. In many
other cases, if dynamic analyses are performed, they are used to calculate the seismic
response of the earth system, which is then used to calculate seismic displacements in an
approach referred to as the decoupled approximation (i.e., the calculation of seismic displa-
cement is decoupled from the calculation of the seismic response; Lin and Whitman 1983).
Even in cases when dynamic analyses are performed, Newmark-type sliding block analyses
form the basis for a preliminary assessment of the expected seismic performance of the earth
system. Pseudostatic slope stability procedures, wherein a horizontal seismic coefficient is

a)
Department of Civil and Environmental Engineering, Univ. of California, Berkeley, CA 94720-1710
b)
Fugro Consultants Inc., Oakland, CA, USA

673
Earthquake Spectra, Volume 34, No. 2, pages 673–695, May 2018; © 2018, Earthquake Engineering Research Institute
674 MACEDO ET AL.

applied to a potential sliding mass in a conventional slope limit equilibrium analysis, are also
commonly employed in practice.
Current state of practice simplified seismic slope displacement procedures separate the
estimation of the ground motion intensity measure (IM) from the calculated seismic displa-
cement (D) given the selected IM hazard level. This approach is referred to in this study as
pseudo-probabilistic because the IM hazard curve is estimated probabilistically, but D is
estimated based on only a single IM value. In this approach, the ground motion IM is esti-
mated for a given design hazard level (or return period) though a probabilistic seismic hazard
analysis (PSHA), and then D is estimated for that IM value. The underlying assumption is the
hazard level for D is consistent with that of the IM. For example, Bray and Travasarou (2007)
performed regression analyses of the results of over 55,000 fully coupled deformable sliding
block analyses using nearly 700 ground motion records to develop an estimate of the seismic
displacement as a function of the yield coefficient of the earth system (ky , which is the seismic
coefficient that results in a pseudostatic factor of safety [FS] of one); the 5% damped spectral
acceleration of the ground motion at the base of the potential sliding, assuming there is no
material above it at its degraded fundamental period, which will be denoted as Sa (the
degraded period is taken as 1.5 times the system’s initial fundamental period); the initial
fundamental period of the sliding mass (T s ); and earthquake moment magnitude (M).
Each of these parameters are uncertain, with the highest level of uncertainty being commonly
associated with the input ground motion. Thus the mean ground motion IM, which in this
procedure is Sa , is often estimated considering a design seismic hazard level (e.g., at the 10%
probability of exceedance in 50 years level), while the other parameters are assigned best-
estimate, median values to develop a 16% to 84% exceedance estimate range of the seismic
displacement given the specified value of Sa .
An alternative is to use a fully probabilistic approach. The proposed performance-based
assessment of seismic slope displacement uses a single IM hazard curve, which is usually the
mean hazard curve, whereas a fully probabilistic approach would include all fractals of the
IM hazard curve. The uncertainties associated with the ground motion IM parameter are not
usually included in the logic tree for the estimation of seismic displacements (D) because they
have already been taken into account in the ground motion hazard analysis for the IM, as also
pointed out by Wang and Rathje (2015). Only a few of the current available procedures are
suitable to be implemented in a performance-based probabilistic seismic slope assessment for
slope properties expressed as ky and T s (e.g., Bray and Travasarou 2007, Rathje et al. 2014,
Travasarou et al. 2004). A performance-based seismic slope assessment is more robust than
commonly used procedures and can provide additional insights. However, performance-
based seismic slope displacement procedures are rarely used in engineering practice because
they are not easy to use, and straightforward implemented procedures are not available. In
this paper, a straightforward performance-based procedure for the estimation of seismic slope
displacement is presented. Additionally, a compatible pseudostatic slope stability procedure
is proposed. The proposed procedures are applicable for a wide range of earth systems shaken
by shallow crustal earthquakes along active margins (e.g., California), as well as for sub-
duction zone earthquakes (e.g., South America). They are easily implemented in practice
using information that an engineer commonly has access to when evaluating the seismic
performance of an earth/waste system or natural slope.
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 675

PREVIOUS STUDIES
There are several seismic sliding block displacement procedures available in the litera-
ture. Several of these well-known procedures are formulated only for the case of rigid sliding
blocks wherein the dynamic response of the potential sliding mass is neglected (e.g., Jibson
2007, Lin and Whitman 1986, Newmark 1965, Richards and Elms 1979, Saygili and Rathje
2008, Watson-Lamprey and Abrahamson 2006). The rigid sliding block analyses should not
be used except in those few cases wherein the potential sliding mass is shallow and stiff, so
that its fundamental period is nearly zero, or as a screening tool in regional assessments when
there is insufficient information to characterize the actual fundamental period of the potential
sliding mass. In all other cases, the dynamic response of what is commonly a “non-rigid”
sliding mass should be considered, because the dynamic response of the sliding mass has
been shown to be a key factor (Rathje and Bray 1999, Rathje and Bray 2000). As mentioned
previously, many of the seismic slope displacement methods that capture the flexibility of the
sliding mass do so by employing the decoupled approximation that assumes no relative dis-
placements during the seismic response analysis, whose results are then used to calculate the
seismically induced permanent displacement (e.g., Bray and Rathje 1998, Makdisi and Seed
1978). Due to the inherent limitations of the decoupled approximation, fully coupled stick-
slip sliding block analyses are preferred (Rathje and Bray 1999, Rathje and Bray 2000).
There are simplified seismic slope displacement procedures that capture the simultaneous
occurrence of the nonlinear dynamic response of the potential sliding mass and the effects
of periodic sliding episodes (e.g., Bray and Travasarou 2007, Bray et al. 2017). Few sim-
plified seismic slope displacement procedures are based on the results of two-dimensional
(2-D) or three-dimensional (3-D) dynamic analyses. As mentioned previously, the largest
source of uncertainty is typically due to the input earthquake ground motion. Hence a simpler
one-dimensional (1-D) model is used with many earthquake ground motions to develop the
data to support the development of a simplified seismic slope stability procedure. This
assumption is reasonably conservative for most cases (Rathje and Bray 2001); the exception
is for shallow sliding adversely affected by topographic amplification. Lastly, most seismic
slope displacement procedures have been formulated only for shallow crustal earthquakes
along active plate margins. However, the Bray et al. (2017) procedure has been formulated
for subduction earthquake zones.
Figure 1 shows the different ways sliding block seismic slope displacement procedures
are implemented. As described in Rathje and Saygili (2008), a deterministic approach first
requires the selection of a design earthquake scenario, defined by M and distance (R) and the
selection of the number of standard deviations above the median (εGM ). Several earthquake
scenarios should be considered. The IM is estimated based on ground motion prediction
equations (GMPE) and the selected εGM (often arbitrarily set to 0 or 1). Once the IM is
defined, D is estimated using a seismic slope displacement model, which is usually formu-
lated in terms of properties of the earth/waste system (e.g., ky , T s ) and the IM. The number of
standard deviations below and above the median displacement estimate (εd ) is used to
account for the variability introduced in the calculation of seismic displacement, given
the specified IM and earth system parameters.
In the pseudo-probabilistic approach (Figure 1), an IM hazard curve that defines the mean
annual rate of exceedance (λGM ) for different levels of the IM is developed through a PSHA
676 MACEDO ET AL.

Figure 1. Different approaches for the estimation of Newmark-based seismic slope

displacements.

that considers all potential earthquake/ground motion scenarios (M, R, and εGM ). The design
IM value is selected from the hazard curve for an acceptable λGM (or return period). This
particular IM value is then used to estimate the seismic slope displacement with the physical
properties of the earth/waste system (e.g., k y , T s ). The number of standard deviations (εd ) in
the estimate of D given these inputs is used to account for the displacement variability.
In a performance-based approach (Figure 1), the variability in the ground motion IM and
the seismic displacement are considered explicitly through a probabilistic analysis that con-
volves the IM hazard curve with a seismic displacement model and the variability of the
system properties used in the model. Several fractals of the IM hazard curve may be
used, but only the mean IM hazard curve is typically used, and different seismic displacement
models may also be considered to include this source of epistemic uncertainty, but only one
model is typically used. The result is a seismic displacement hazard curve that provides the
mean annual rate of exceedance for different levels of seismic displacement (λd ). The dis-
placement hazard curve can be used to estimate the expected seismic displacements directly
related with the design hazard level. It better captures the uncertainty in the estimate of seis-
mic slope displacement. Thus it is well suited for making engineering decisions.
Pseudostatic slope stability procedures are often used in engineering practice in prelimin-
ary assessments because they are straightforward. However, they are limited unless the para-
meters utilized in the analysis accurately reflect the potential seismic demand and its impact
on the seismic performance of the system. The seismic coefficient that is employed in a
pseudostatic slope stability analysis should be selected in a rational manner if this procedure
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 677

is to form a sound basis for a seismic slope stability assessment (e.g., Bray and Travasarou
2009). The selection of the seismic coefficient employed in the analysis is often based on
precedence, regulatory design guidance, and engineering judgment, without due considera-
tion of the seismic displacement that constitutes satisfactory performance for each particular
project, and without incorporating the vastly different seismic exposure for sites around the
world. Commonly used pseudostatic methods are often formulated under the assumption of a
fixed maximum level of seismic displacements that can be considered acceptable (e.g., 1 m
for earth dams in the Seed 1979 method). These methods usually consider the seismic coef-
ficient as a fixed number based on precedence (e.g., 0.15) or as a fixed fraction of the peak
ground acceleration (PGA). Several recent efforts relate directly the selection of the seismic
coefficient to be used in pseudostatic slope stability analysis to the allowable level of seismic
displacement that the earth/waste system can sustain (e.g., Bray and Travasarou 2009,
Papadimitriou et al. 2014). Although these efforts provide a rational basis for selecting
the seismic coefficient based on the allowable level of displacement, they are not formulated
in a rigorous probabilistic framework (e.g., they do not consider the entire seismic hazard
curve in the selection of the seismic coefficient). A pseudostatic slope stability method based
on the performance-based seismic slope displacement procedure developed in this study is
also presented in this paper.

CRITICAL DESIGN ISSUE

Before discussing the performance-based seismic slope displacement procedure further,
it is imperative that a key design issue be addressed. In the evaluation of the seismic
performance of an earth/waste system or natural slope, the engineer should first assess if
there are materials in the earth structure or its foundation that can lose significant strength
as a result of cyclic loading (e.g., soil liquefaction). If so, this should be the primary focus of
the evaluation because large displacement flow slides are possible. There is abundant
research on this topic, and it is not the focus of this study. This study addresses the seismic
performance of earth/waste systems or natural slopes wherein flow slides are not likely
(i.e., the post-liquefaction slope stability factor of safety is greater than one).

PERFORMANCE-BASED ASSESSMENT OF SEISMIC DISPLACEMENT

The framework of Bray and Travasarou (2007) is used in this study because it provides an
explicit definition of the probability of negligible displacement and the median seismic dis-
placement; it has been formulated based on the fully coupled stick-slip deformable sliding
block analysis; its framework has been used to develop procedures for earthquakes in shallow
crustal settings and subduction interface settings; and it is used widely in practice. It models
seismically induced permanent displacements as a mixed random variable that has a prob-
ability mass at “zero” displacement (d 0 ) and a probability density for finite displacement
values greater than d 0 (Figure 2). Displacements smaller than d 0 are typically not of
engineering significance and can, for all practical purposes, be considered to be negligible
(i.e., zero). The values of seismic displacements that are smaller than d 0 are lumped together
to d 0 . The probability of “zero” displacements (d ≤ d 0 ) and the median “non-zero” displace-
ments are estimated based on ky , T s , Sa , and M using ground motion recordings from shallow
crustal earthquakes. Using this framework, the seismic displacement model for shallow crus-
tal earthquake from Bray and Travasarou (2007), referred to as BT07, has been formulated
678 MACEDO ET AL.

Figure 2. (a) Probability density function for a mixed random variable and (b) probability of
exceedance for a mixed (solid line) and a continuous (dotted line) random variable (after
Bray and Travasarou 2007).

with and without magnitude dependence, considering d 0 ¼ 1 cm. Bray et al. (2017), referred
to as BMT17, used the same framework to develop predictive equations for seismic slope
displacement for subduction zone earthquakes, but considered d 0 ¼ 0.5 cm.
The goal of a performance-based seismic slope displacement assessment is to develop the
seismic displacement hazard curve, which defines the annual probability of the seismic dis-
placement exceeding a specified seismic displacement threshold. This probability is given by
Equation 1 for the case in which the seismic displacement model is not magnitude dependent,
and by Equation 2 for the case in which the model is magnitude dependent.
nts ð ∞
nky X
X
λd ¼
EQ-TARGET;temp:intralink-;e1;41;351 PðD > d j Sa ,kyi ,T sj ÞΔλðSa ÞdðSa Þwij (1)
i¼1 j¼1 0

nts ð ð ∞
nky X
X
λd ¼
EQ-TARGET;temp:intralink-;e2;41;306 PðD > d j Sa ,M,kyi ,T sj ÞPðM j Sa ÞΔλðSa ÞdðSa Þwij (2)
i¼1 j¼1 M 0

Consistent with the Pacific Earthquake Engineering Research Center (PEER) performance-
based engineering framework (Deierlein et al. 2003), P is used for probability and λ is used
for annual rate of exceedance. Δλ represents the rate of occurrence. PðD > d j Sa ,kyi ,T sj Þ in
Equation 1 is the conditional probability the displacement level d is exceeded given Sa , k yi
and T sj , and ΔλðSa Þ is the annual probability of occurrence of Sa . In Equation 2,
PðD > d j Sa ,M,kyi ,T sj Þ is the conditional probability the displacement level d is exceeded
given Sa , M, kyi and T sj . PðM j Sa Þ is the probability of occurrence of M given Sa ; it represents
the contribution to the total hazard Sa from M. The term PðM j Sa Þ can be estimated from the
hazard deaggregation for Sa . The uncertainties of ky and T s are treated as epistemic, so a logic
tree is employed, in which nky values for ky are defined with weighting factors
wkyi ði ¼ 1∶ nkyÞ and nts values for T s are defined with weighting factors
wtsj ð j ¼ 1∶ ntsÞ. The weighting factor for each combination of ky and T s is defined as wij .
ky and T s may be considered to be represented by a lognormal distribution with its mean
and covariance (cov) to assign their alternative values and weighting factors (i.e., consider
an equal partition in the log space to evaluate the weighting factors). In the case in which
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 679

the uncertainty of ky and T s is not considered, the calculations are performed using the mean
values of k y and T s with weighting factors equal to 1.
The annual probability of occurrence  motion level Sai , ΔλðSa Þ, is equal to the
 of a ground
a ÞÞ
derivative of the hazard curve for Sa i:e:; dðHaðS
dSa evaluated at Sai . Travasarou et al. (2004)
used a function that fit approximately the hazard curve at the degraded period of the system to
calculate its derivative and then evaluate the required integrals. However, this procedure
could lead to inconsistencies especially for low or large values of spectral acceleration.
In this study, the integrals are estimated based on the “integration by parts” and “rate of
occurrence” methods. The integration by parts method does not require approximations
for the seismic hazard curve derivative and is used by the U.S. Geological Survey
(USGS) for risk analyses. The rate of occurrence method is an approximation, but believed
to be sufficiently accurate.
The shallow crustal earthquake seismic slope displacement model without magnitude
dependence of Bray and Travasarou (2007) estimates the probability of “zero” displacement
and the median value of the “non-zero” seismic displacement with Equations 3 and 4,
respectively.

PðD ¼ “0”Þ ¼ 1  Φð1.76  3.22Lnðky Þ  0.484T s þ 3.52LnðSa ÞÞ

EQ-TARGET;temp:intralink-;e3;62;423 (3)

^ ¼ 1.10  2.83Lnðk y Þ  0.333ðLnðky ÞÞ2 þ 0.566Lnðk y ÞLnðSa Þ þ 3.04LnðSa Þ

LnðdÞ
EQ-TARGET;temp:intralink-;e4;62;392

 0.244ðLnðSa ÞÞ2 þ 1.5T s (4)

where Φ is the standard normal cumulative distribution function, and d^ is the median seismic
displacement. The first term in Equation 4 should be replaced by −0.22 when T s < 0.05s. The
first term inside the integral in Equation 1 can be evaluated as:

PðD > d j Sa ,ky ,T s Þ ¼ ½1  PðD ¼ “0” j Sa ,ky ,T s Þ½PðD > d j Sa ,k y ,T s ,D > “0”Þ
EQ-TARGET;temp:intralink-;e5;62;307 (5)

Considering a lognormal distribution for seismic displacements (D), the second term is esti-
mated as:
 ^ 
LnðdÞ  LnðdÞ
PðD > d j ky ,T s ,Sa ,D > “0”Þ ¼ 1  Φ (6)
σ LnD
EQ-TARGET;temp:intralink-;e6;62;245

where σ LnD ¼ 0.67, which corresponds to the standard deviation of the error in Equation 4.
In the “rate of occurrence” approach, the ΔλðSai Þ term in Equation 1 (for a Sai value
representative of a spectral acceleration bin from Sa1 to Sa2 ) can be estimated from the hazard
curve in terms of the mean annual rate of exceeding the ground motion levels in the bin, λGM1
and λGM2 , in terms of the rate of occurrence, ROi ¼ λGM1  λGM2 , as:
 
 dHa

ΔλðSai Þ ¼   ¼ λGM1  λGM2 ¼ ROi (7)
dSa Sai
EQ-TARGET;temp:intralink-;e7;62;133

d Sai d Sai
680 MACEDO ET AL.

Thus Equation 1 becomes:

nky X
X nts X
λd ¼
EQ-TARGET;temp:intralink-;e8;41;627 PðD > d j Sak ,kyi ,T sj ÞROk wij (8)
i¼1 j¼1 k

Equation 8 can be directly evaluated. In the integration by parts approach, using the inte-
gration by part theorem for the integral in Equation 1, the following result is obtained (the
detailed derivation is included on the electronic Appendix A)
ð∞    
LnðdÞ  LnðdÞ
I Sa ¼ 3.52PDF1ðHaÞ 1  Φ þ ðHaÞðΦð1.76  3.22Lnðky Þ
σ LnD
EQ-TARGET;temp:intralink-;e9;41;537

∞
 
PDF2 0.566Lnðk y Þ
 0.484T s Lnðk y Þ þ 3.52LnðSa ÞÞÞ dLnðSa Þ
σ LnD 0.488LnðSa Þ þ 3.04
(9)

Equation 9 can be used directly in Equation 1 to estimate the displacement hazard curve if the
best estimated values (e.g., mean) for k y and T s are considered (i.e., their weighting factors are
set to 1 in Equation 1). To account for material inhomogeneity and variability in the strength of
the soil, the epistemic variability for the yield coefficient (k y ) and the system’s initial funda-
mental period (T s ) can be considered through a logic tree scheme, as previously discussed.
Bray and Travasarou (2007) prefer the use of their equation to estimate the median
“non-zero” seismic displacements for shallow crustal earthquakes, which includes a magni-
tude term:
^ ¼ 1.10  2.83Lnðky Þ  0.333ðLnðk y ÞÞ2 þ 0.566Lnðk y ÞLnðSa Þ þ 3.04LnðSa Þ
LnðdÞ
EQ-TARGET;temp:intralink-;e10;41;342

 0.244LnðSa Þ2 þ 1.5T s þ 0.278ðM  7Þ (10)

The probability of “zero” displacement relationship remains as defined previously as

Equation 3. The probability of a seismic displacement D greater than a given value d
and for given values of ky , T s , Sa , and M, is estimated from Equation 5 with σ LnD equal
to 0.66 (which corresponds to the standard deviation of the error term in Equation 10).
In the “rate of occurrence” approach, the ΔλðSai Þ term in Equation 2 (for a Sa1 value
representative of a spectral acceleration bin from Sa1 to Sa2 ), is estimated using Equation 7
as in the previous case and PðMjSa Þ can be estimated from the hazard deaggregation for the
magnitude term. The seismic slope displacement hazard curve is evaluated by approximating
the integrals numerically for weighted values of k y and T s , as discussed previously.
Alternative equations for the “integration by parts” approach are provided in the electronic
Appendix A.
Bray et al. (2017) provide a predictive equation to estimate the probability of “zero”
displacements and the median “non-zero” displacement for subduction interface earthquakes.
These equations include a dependence on the magnitude term for the median “non-zero”
seismic displacement and can be used in the same way as described before for the predictive
equations for the shallow crustal settings with magnitude dependence. The necessary
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 681

equations to use the Bray et al. (2017) procedure in a performance-based framework are
provided in the electronic Appendix A.

UNCERTAINTY TREATMENT OF SYSTEM PROPERTIES

The uncertainty for the system properties (i.e., ky and T s ) can be treated as epistemic
(e.g., Rathje and Saygili 2011, Wang and Rathje 2015). However, their uncertainty has
also been considered through probability density functions (PDF; e.g., Yegian et al.
1991, Travasarou et al. 2004). In the latter case, which is also consistent with the PEER
performance-based design framework, Equations 1 and 2 become:
ð ð ð∞
λd ¼ PðD > d j Sa ,k y ,T s ÞPðSa Þf ðk y Þf ðT s ÞdSa dky dT s
EQ-TARGET;temp:intralink-;e11;62;527 (11)
Ts ky 0

ð ð ð ð∞
λd ¼
EQ-TARGET;temp:intralink-;e12;62;480 PðD > d j Sa ,M,k y ,T s ÞPðM j Sa ÞðΔλðSa ÞÞf ðT s ÞdSa dky dT s (12)
Ts ky M 0

The functions f ðk y Þ and f ðT s Þ are the probability density functions for the yield coefficient
and the fundamental period, respectively; the other terms were defined previously. Equa-
tions 11 and 12 can be evaluated with the procedures already presented approximating
the integrals over ky and T s as summations.
The two approaches for categorizing the uncertainty in the system properties are compared
considering a system with mean value of k y ¼ 0.15 with cov = 0.2 and mean value of
T s ¼ 0.67 s with cov = 0.1, which is located in Yerba Buena (YB), California. The hazard
curve for the spectral acceleration at the degraded period of the system and the deaggregation
information were obtained from the USGS website (https://earthquake.usgs.gov/hazards/
interactive/, last accessed 4 September 2017). A lognormal distribution was considered for
k y and T s in the PDF approach. For the case that considers an epistemic treatment for k y
and T s , three alternative ky and three alternative T s , values are defined through a logic
tree with weighting factors consistent with a lognormal distribution. The alternative values
of ky are 0.10, 0.15, and 0.23 with weighting factors of 0.14, 0.72 and 0.14, and the alternative
values of T s are 0.54, 0.67, and 0.83 with weighting factors 0.14, 0.72, and 0.14.
Figure 3 shows the comparisons between the two approaches for the system’s uncertainty
treatment, the different curves from each branch of the logic tree are also included (i.e., nine
curves given the three alternative values each for k y and T s ). The mean slope displacement
hazard curve calculated considering the epistemic treatment for k y and T s matches the curve
calculated using PDF functions for ky and T s . Thus if one only wanted the mean slope dis-
placement hazard curve, either approach could be used. However, the epistemic approach has
the advantage of allowing an assessment in the uncertainty of the displacement hazard curve,
and it is recommended.

EPISTEMIC UNCERTAINTY OF IM AND D MODELS

Usually the epistemic uncertainties associated with the ground motion are not included in
the logic tree for the estimation of seismic displacements because they have already been taken
into account in the ground motion hazard analysis for the IM (e.g., Wang and Rathje 2015).
682 MACEDO ET AL.

Figure 3. Comparison between the mean hazard displacement curve considering an epistemic
uncertainty for ky and T s and the hazard displacement curve considering PDFs for ky and T s .

Figure 4. Illustrative logic tree for PSHA analysis.

However, the epistemic uncertainty in the IM may also be integrated into the logic tree for the
estimation of D. To show the consideration of epistemic uncertainty in the IM and D predictive
models, a simplified PSHA analysis is conducted for two seismic sources. The logic tree con-
sidered for the PSHA is shown in Figure 4. There are three possible magnitudes and two pos-
sible values for the activity rate for each seismic source, and two GMPEs are used to calculate
PGA for the rigid sliding block case (i.e., T s ¼ 0 s). Thus a total of 72 IM hazard curves, each
one with its respective weighting factor, are calculated.
The epistemic uncertainty in the parameter ky is also considered using values of 0.1, 0.15,
and 0.23 with weighting factors of 0.14, 0.72, and 0.14, respectively. The epistemic uncer-
tainty in the seismic displacement model is captured using the BT07 model with a weighting
factor of 0.7 and the Saygili and Rathje (2008) scalar model with a weighting factor of
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 683

Figure 5. Mean, median, and 10%–90% percentiles seismic displacement hazard curves.

0.3. Each IM hazard curve is convoluted with each D model for all the realizations of k y ,
which produces 432 seismic displacement hazard curves. They are shown in Figure 5
with the mean hazard curve and the 10%, 50%, and 90% fractal hazard curves. Figure 5
also includes the 10% and 90% fractal hazard curves when only the mean IM hazard
curve is considered. Including several sources of uncertainty in this simple example shows
the range of results possible. Notice also that the range of results is broader if all IM hazard
curves are considered. It is standard practice in performance-based design to use the mean
hazard curve to estimate seismic displacement at a specified hazard level.

DIFFERENT TECTONIC SETTINGS

Sometimes there are significant contributions from seismic sources in more than one
tectonic setting (e.g., shallow crustal and subduction zones). Bray et al. (2017) found
there were not large differences in the estimates resulting from a shallow crustal earthquake
seismic displacement model (BT07) and subduction earthquake zone seismic displacement
model (BMT17) if they were both conditioned on the same seismic demand, such as Sa . It is
more important to use GMPEs with the appropriate characterization for each tectonic setting
to estimate Sa than to use different seismic displacement models. However, as there are some
cases in which the secondary effect of using the seismic displacement model appropriate for
each tectonic setting is important, recommendations are presented.
In those cases, the IM hazard curve from each tectonic setting should be convoluted with
their respective representative displacement model, and their annual rates of exceedance
should be summed up to produce the total seismic displacement hazard curve. Figure 6a
shows the spectral acceleration hazard curves from shallow crustal and subduction earth-
quake zones for a system with T s ¼ 0.67 s and k y ¼ 0.1 located in Washington, which
has contributions from two seismic settings. A PSHA provides the hazard curves for Sa
and the deaggregation information, which, for this example, is obtained from the USGS
684 MACEDO ET AL.

Figure 6. (a) Spectral acceleration hazard curves by tectonic settings, 1.5T s ¼ 1.00 s and (b) total
displacement hazard curve and components by tectonic setting.

website (https://earthquake.usgs.gov/hazards/interactive/). The website provides the deaggre-

gation by GMPE, which can be used to estimate the deaggregation for the seismic sources in
each tectonic setting. The shallow crustal Sa hazard curve was convoluted with the BT07
model, and the subduction Sa hazard curves (i.e., slab and interface) were convoluted
with the BMT07 model. Figure 6b shows the total seismic displacement hazard curve
and the contributions from each tectonic setting.

METHODOLOGY IMPLEMENTATION
The proposed procedures for developing the seismic slope displacement hazard curve in
shallow crustal earthquake and subduction interface tectonic settings are implemented in the
electronic Appendix B, which contains Excel spreadsheets and executables files to perform
the necessary calculations. They can be readily used in engineering practice after the engineer
defines the required inputs, which are:
(1) Estimate the mean value of the yield coefficient (k y ) and fundamental period (T s ) of
the potential sliding mass, as described by Bray (2007). The uncertainty of these
variables may be included by estimating their coefficient of variation and defining
alternative values with corresponding weights in a logic tree scheme. The engineer
can evaluate the mean and cov values for ky and T s using available information
(e.g., Travasarou et al. 2004), and use these parameters to define the alternative values
and weighting factors that are consistent with a lognormal distribution. In cases where
uncertainty in these variables is not judged to be significant, the calculations are com-
puted at the mean values of ky and T s with weighting factors equal to 1.
(2) Compute the seismic hazard curve for Sa (the spectral acceleration at 1.5 times the
fundamental period of the system) and the deaggregation information for magnitude
bins at different values of Sa in the hazard curve. In general, this can be done through
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 685

a site-specific probabilistic seismic hazard analysis. For the United States, the seis-
mic hazard curve and deaggregation may be obtained from https://earthquake.usgs.
gov/hazards/interactive/, or using the nshmp-haz software from: https://github.com/
usgs/nshmp-haz/ (last accessed 4 September 2017) Interpolated values may be used
to define the Sa seismic hazard curve at periods not available at these websites.
(3) Specify a list of seismic displacement values used for the estimation of the seismic
displacement hazard curve. The provided tools will estimate the annual rate of
exceedance (i.e., λd ) for each value in the list.
(4) Define a target hazard level (or return period, e.g., 475 years) for the estimation of
the seismic displacement.
Figure 7 shows the use of the spreadsheets and executable files in the electronic appendix
(the diagram is for the shallow crustal case, but it is similar for other cases). The spreadsheets
in the appendix also contain explanatory notes for their use. In this straightforward applica-
tion of performance-based seismic slope displacement hazard calculations, the engineer can
focus on the estimation of the spectral acceleration hazard curve with its deaggregation infor-
mation and the system properties defined by ky and T s and their variability.

ILLUSTRATIVE EXAMPLE
The anticipated performance of a representative earth dam is evaluated in terms of
seismically-induced permanent slope displacement. Return periods of 475 and 2,475 years
are considered. The calculations are performed with the files provided in electronic
Appendix B. The geometry and properties of the dam are identical to the example used
by Bray and Travasarou (2007). The maximum cross section of the 57 m-high rolled
earth-fill dam, which is founded on rock, is shown in Figure 8.

Figure 7. (a) Flow diagram for the performance-based estimation of seismic slope displacements.
686 MACEDO ET AL.

Figure 8. Cross section of dam used in illustrative example (from Bray and Travasarou 2007).

Step 1: The analysis is performed with best-estimate (mean) values of k y and T s to eval-
uate a deep slide through the dam. For the case of base sliding at the maximum height of this
triangular-shaped potential sliding mass, the best estimate of its initial fundamental period is:

2.6H 2.6ð57mÞ
Ts ¼
EQ-TARGET;temp:intralink-;e13;41;444 ¼ ≈ 0.33s, therefore, 1.5T s ≈ 0.5s (13)
Vs 450m∕s

where V s is the mean value of the shear wave velocity of the sliding mass. The yield
coefficient for a deep failure surface was estimated to be 0.14 from a pseudostatic slope
stability analyses. The stability analysis was performed with the total stress strength proper-
ties of c ¼ 14 kPa and ϕ ¼ 21° based on undrained triaxial compression tests. The mean
values of ky and T s are entered in the Input sheet of the Input_Crust.xls spreadsheet with
cov ¼ 0 for this example.
Step 2: The dam is assumed to be at GPS coordinates N37.81 W122.36 in Northern
California, where there is negligible contribution to the seismic hazard from subduction
earthquakes. The hazard curve for the spectral acceleration at the degraded period of the
dam (0.5 s) and the deaggregation information are obtained using the nshmp-haz software.
Different combinations of return period were entered to develop the seismic hazard curve
shown in Figure 9a. The corresponding deaggregation information is pasted on the Input
sheet of the Input_Crust.xls spreadsheet (i.e., the gray cells).
Step 3: A list of seismic displacements for their annual rate of exceedance is needed.
Twenty-three values between 1 and 100 cm are considered and entered in the Input
sheet of the Input_Crust.xls.
Step 4: The design hazard levels are defined in this example for return periods of 475 and
2,475 years.
With this information, the files in the electronic Appendix B are used to develop the
seismic slope displacement hazard curve shown in Figure 9b. The estimated seismic
slope displacement for the 475- and 2,475-year return periods are approximately 20 cm
and 63 cm, respectively.
In a pseudo-probabilistic analysis using the BT07 procedure, first, the spectral accelera-
tions for the return periods of interest are estimated, and these values are used in the
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 687

Figure 9. Illustrative example: (a) spectral acceleration hazard curve, and (b) seismic slope dis-
placement hazard curve.

procedure with the properties of the sliding mas (T s and ky ) and a magnitude value, which
can be obtained from the deaggregation at each return period. From Figure 9a, the spectral
accelerations for 475- and 2,475-year return periods are 0.63 g and 1.0 g, respectively, and the
modal magnitude is 8.0 in both cases.
With these input values, the median seismic slope displacements from the BT07 model
are estimated to be 20 cm and 52 cm, respectively. For this case, the use of a spectral accel-
eration that corresponds to 2,475-year return period in the pseudo-probabilistic approach
yields a seismic slope displacement value (52 cm) with approximately a return period of
1,750 years (Figure 9b). Conversely, the performance-based approach enables the seismic
slope displacement at each return period to be estimated directly. As shown with this exam-
ple, the return period of the seismic slope displacement does not necessarily correspond with
the return period of the ground motion intensity measure. The seismic slope displacement
hazard should be used to evaluate the seismic performance of an earth/waste system or nat-
ural slope.

COMPARISON OF PERFORMANCE-BASED ANALYSES

Performance-based seismic slope displacement hazard assessments are compared for
earth systems located at three different sites in the United States. The sites are selected
to represent areas with different seismic activity and tectonic settings. Figure 10 shows a
comparison of seismic slope displacement hazard curves, considering the “integration by
parts” and the “rate of occurrence” methods, as well as with magnitude and no magnitude
dependence. The curves are shown for a system with a fundamental period T s ¼ 0.67 s
located in Yerba Buena Island (YB), California (N37.81 W122.36); a system with a funda-
mental period T s ¼ 0.33 s located in Salt Lake City (SL), Utah (N40.75 W111.88); and two
systems with fundamental periods equal to T s ¼ 0.33 s and T s ¼ 0.67 s located west of
Seattle (S1 and S2, respectively), Washington (N47.42 W123.56). The sites in YB and
688 MACEDO ET AL.

Figure 10. (a) Spectral acceleration hazard curves: S1, S2 = Seattle; SL = Salt Lake; YB =
Yerba Buena; (b) comparison of “integration by parts,” “rate of occurrence” methods, and exam-
ination of magnitude dependence for YB site; (c) sensitivity analyses in terms of cov values of ky
and T s , considering magnitude dependence for YB site. Results are for the mean weighted hazard
curve; and (d–f) comparison of “integration by parts” and “rate of occurrence” methods for SL,
S1, and S2 cases. Horizontal dashed lines represent hazard levels for return periods of 475 and
2,475 years.
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 689

SL have contributions primarily from crustal seismic sources; whereas, the site west of
Seattle has contribution primarily from the subduction seismic sources.
The USGS website (https://earthquake.usgs.gov/hazards/interactive/) and nshmp-haz
software provide the hazard curves for the spectral acceleration at the system’s degraded
period and magnitude deaggregation information. Mean values of k y (i.e., 0.04, 0.1, 0.2,
and 0.3) and T s were used in the analyses. However, sensitivity analyses were performed
using cov values of 0.2 and 0.5 for k y , and cov values of 0.1 and 0.25 for T s . As described
previously, weighting factors were applied to three alternative values of each parameter to
capture a lognormal distribution. The selected values of ky cover the range typically found in
practice. The selected cov values represent typical lower and upper ranges (e.g., Travasarou
et al., 2004).
The results for the cases shown in Figure 10 indicate the “integration by parts” and “rate
of occurrence” methods provide consistent results for the estimation of the seismic slope
displacement hazard curve. Also, the inclusion of the magnitude dependence term in the
predictive equations for median seismic displacements for the cases analyzed in the shallow
crustal earthquake setting does not affect greatly the seismic slope displacement hazard
curve. The different slopes of the seismic slope displacement hazard curves for these
cases (Figures 10b, 10d, 10e, and 10f ) generally follow the trends of the slopes of the spectral
acceleration hazard curves (Figure 10a), with the YB site displaying the steepest slope and
the SL site displaying the flattest slope over the return periods examined in this study.
The inclusion of the uncertainty of ky and T s did not significantly affect the estimated seismic
displacements for the cases with the lower cov values of 0.2 and 0.1 for ky and T s , respec-
tively (Figure 10c). However, use of the higher cov values of 0.5 and 0.25 for ky and T s ,
respectively, influenced considerably the estimated seismic slope displacements.
Table 1 compares the results obtained using the conventional pseudo-probabilistic
approach and the proposed performance-based approach to estimate seismic slope displace-
ments for the cases discussed previously. The pseudo-probabilistic approach underestimates
the seismic slope displacement value estimated with the performance-based approach at both
return periods for the YB case.

Table 1. Comparison pseudo-probabilistic and performance-based approaches to estimate

seismic slope displacements for selected cases at 475 and 2,475-year return periods (ky ¼ 0.1)

Spectral accel. (g) Pseudo-prob. D (cm) Perf-based D (cm) Δd **

475 2,475 475 2,475 475 2,475 475 2,475

Site 1.5 T S * (s) years years years years years years years years
YB 1.0 0.33 0.58 15 50 16 57 0.07 0.13
SL 0.5 0.30 1.10 6 76 5 76 −0.18 0.00
S1 0.5 0.47 1.03 25 100 16 106 −0.44 0.06
S2 1.0 0.25 0.60 11 64 7 70 −0.45 0.09
Note: S1, S2 = Seattle; SL = Salt Lake; YB = Yerba Buena.
* T s is the initial fundamental period of the system.
** Δd is defined as Ln(Perf.-Based)-Ln(Pseudo-Prob.).
690 MACEDO ET AL.

The pseudo-probabilistic approach overestimates the seismic slope displacement value

estimated with the performance-based approach at the 475-year return period for the SL case.
Results are consistent at the 2,475-year return period. These results occur because of the
higher recurrence of earthquakes in California relative to the recurrence of earthquakes
in Utah, which implies that the slope of the seismic hazard curve in YB is steeper than
the slope of the hazard curve in SL. Thus the rate of occurrence of spectral acceleration values
is higher in YB than in SL, causing more contribution to the annual rate of exceedance for a
given seismic displacement. For the Seattle cases, the pseudo-probabilistic approach overes-
timates the seismic slope displacement value estimated with the performance-based approach
at the 475-year return period and underestimates the seismic slope displacement values at the
2,475-year return period. This result can be explained by considering the activity rate of earth-
quakes in the Cascadia zone (∼1/600 per year). For a median ground motion
(i.e., its probability of exceedance is 0.5) and a simple seismic source, the annual rate of excee-
dance for the median ground motion will be about 1/1,200 per year (i.e., 0.5  1/600). Thus a
negative value for delta (delta represents the difference between the performance-based and
the pseudo-probabilistic estimations) would be expected for the first return period (associated
with a rate of 1/475 per year), and a positive value for delta would be expected for the second
return period (associated with a rate of 1/2,475 per year). A performance-based seismic slope
displacement hazard analysis is not in general more or less conservative than a pseudo-
probabilistic analysis. There may be systematic differences in the same region and the
same return period, but, ultimately, the comparison depends on the seismotectonic setting
for the site and the return period.
Engineers often use the ground motion associated with a specified hazard level to cal-
culate seismic slope displacements (i.e., pseudo probabilistic approach) with the implicit
assumption that the hazard level associated with the seismic slope displacement is the
same as the hazard level associated with the ground motion parameter being used. However,
this is not generally valid (as shown in Table 1). Similar inconsistencies with the pseudo-
probabilistic approach were found by Franke et al. (2016), Kramer and Mayfield (2007), and
Rathje and Saygili (2011). The engineer should target directly the hazard levels associated
with the calculated seismic slope displacements, because, ultimately, the potential seismic
performance of an earth/waste system or natural slope depends on seismic slope displacement
and not on a particular ground motion intensity measure.

PROBABILISTIC EVALUATION OF SEISMIC COEFFICENT

The proposed performance-based seismic slope displacement hazard procedure can be
reformulated to estimate a hazard-compatible seismic coefficient (k) to be used in a pseudo-
static slope stability analysis using these steps:
1. In consultation with the client and regulators, define the allowable level of seismic
slope displacement of the earth/waste system to achieve the desired perfor-
mance level.
2. Select the appropriate return period for the allowable seismic slope displacement.
Several values may be selected, each with their respective return period.
3. Develop a best-estimate mean value of the fundamental period of the potential slid-
ing mass (T s ). Assign an appropriate covariance value of its uncertainty if epistemic
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 691

uncertainty is to be modeled. Typically, this is not required because a majority of the

uncertainty is in the seismic demand parameter.
4. Assume an initial mean value of the system’s yield coefficient (ky ). The uncertainty
of ky could also be modeled using an appropriate covariance value of this variable.
Calculate an initial displacement hazard curve following the procedures described
previously.
5. Iterate using different values of ky to recalculate the seismic slope displacement
hazard until the seismic slope displacement hazard curve passes through the inter-
section of the horizontal line that goes through the annual frequency of exceedance
defined in Step 2 and the vertical line that goes through the tolerable level of seismic
displacement defined in Step 1. The corresponding ky value is the seismic coefficient
(k) compatible with the tolerable level of seismic slope displacement and the
specified return period. This procedure is implemented in the files included in
the electronic Appendix B to automate the iterative procedure.

Figure 11. Probabilistic estimation of seismic coefficient. Horizontal lines represent return per-
iods of 475 and 2,475 years. Allowable displacement (vertical line) of 15 cm: (a) probabilistic
estimation of seismic coefficient YB site, (b) probabilistic estimation of seismic coefficient SL
site, (c) probabilistic estimation of seismic coefficient S1 site, and (d) probabilistic estimation of
seismic coefficient S2 site.
692 MACEDO ET AL.

Table 2. Comparison pseudo-probabilistic and performance-based approaches to estimate a

pseudostatic coefficient for selected cases at 475 and 2,475-year return periods. (Da ¼ 15cm)

Spectral accel. (g) Pseudo-prob. k*** Perf-based k

1.5 T S * Da ** 475 2,475 475 2,475 475 2,475

Site. (s) (cm) years years years years years years
YB 1.0 15 0.33 0.58 0.103 0.184 0.100 0.200
YB 1.0 30 0.33 0.58 0.070 0.127 0.080 0.140
SL 0.5 15 0.30 1.10 0.057 0.280 0.060 0.270
S1 0.5 15 0.47 1.03 0.133 0.280 0.100 0.270
S2 1.0 15 0.25 0.60 0.090 0.200 0.070 0.210
Note: S1, S2 = Seattle; SL = Salt Lake; YB = Yerba Buena.
* T s is the fundamental period of the system.
** Da is the tolerable level of displacements. For YB, Da ¼ 15 and 30 cm for comparison purposes.
*** k is the seismic coefficient.

6. The calculated k value is applied to the potential sliding mass in a pseudostatic slope
stability analysis, and if FS ≥ 1.0, the seismic slope displacement is at or below the
tolerable level of seismic slope displacement for the specified return period.
Figure 11 shows the results of the probabilistic estimation of the pseudostatic seismic
coefficient for the sites described previously for a tolerable seismic slope displacement
of 15 cm. Table 2 provides a comparison (also including the case where tolerable seismic
slope displacement is set at 30 cm) of the estimated seismic coefficients using the proposed
performance-based method and the median values estimated based on a pseudo-probabilistic
method. The estimation based on a pseudo-probabilistic method is performed by inverting the
Equations 10 and A.2 (Appendix A) and solving for ky . The inversion of the Equation 9 is
performed for shallow crustal settings (as recommended by Bray and Travasarou 2009) and
the inversion of Equation A.2 for subduction interface earthquakes (as recommended by Bray
et al. 2017). In general, the estimation of the seismic coefficient using the performance-based
and pseudo-probabilistic methods are consistent (the highest difference is in the order of
30%), but the performance-based method is preferred because it more rationally incorporates
the uncertainties of the variables and the influence of the selected hazard level.

CONCLUSIONS
Performance-based procedures are presented for the straightforward assessment of the seis-
mic performance of earth/waste systems or natural slopes in shallow crustal and subduction
earthquake settings. Deterministic or pseudo-probabilistic simplified seismic slope displace-
ment procedures are typically used in engineering practice. However, performance-based pro-
cedures are preferred because they more completely consider the actual uncertainty in the
ground motion intensity measure and earth system properties. Moreover, a performance-
based procedure provides seismic slope displacement estimates that are consistent with design
hazard levels. Performance-based procedures are not commonly used because they are thought
to be too complicated to be used for noncritical projects. The proposed procedures enable the
straightforward development of the seismic slope displacement hazard or hazard-compatible
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 693

seismic coefficient through the files provided in the electronic appendix. Thus they may be
readily used in engineering practice.
The seismic slope displacement models of Bray and Travasarou (2007) for shallow
crustal earthquakes and Bray et al. (2017) for subduction interface earthquakes are incor-
porated in the proposed probabilistic procedure. Consistent results are obtained with these
models whether the seismic slope displacement hazard curve is estimated using “integra-
tion by parts” or “rate of occurrence” approaches. The integration by parts approach is more
fundamental if the analytical derivatives of the displacement models can be calculated (as it
is the case for BT07 or BMT17). It has been used to confirm the accuracy of the local
approximations in the “rate of occurrence” approach. For the shallow crustal earthquake
cases studied, the seismic slope displacement hazard curve was not highly sensitive to the
use of predictive equations with or without the earthquake magnitude term. It is sufficient
to use the modal magnitude from the seismic hazard deaggregation in the magnitude-
dependent model.
The performance-based seismic slope displacement hazard procedure does not neces-
sarily provide higher or lower values than the commonly used pseudo-probabilistic pro-
cedure. Their relative values depend on the particular seismo-tectonic setting and the
design hazard level. The performance-based method provides seismic slope displacement
estimates that are directly compatible with the target design levels. Hence it should be used,
because the hazard is developed based on a relevant index of seismic performance, which is
the calculated seismic slope displacement, and not on an indirect index of performance such
as a ground motion intensity measure. Other alternatives to be used as an index of seismic
performance, such as the measure of damage (DM in the PEER approach) or decision
variables in loss analyses (DV in the PEER approach) may also be considered in the
performance-based framework.
The proposed performance-based seismic slope displacement hazard procedure was
recast to estimate a hazard-compatible seismic coefficient to be used in a pseudostatic
slope stability analysis. The value of the seismic coefficient is a function of the selected
threshold of allowable seismic slope displacement, among other factors. Thus the design
seismic coefficient value is estimated in a rational manner. The procedure relies on the
estimation of the displacement hazard curve and uses standard information that is currently
used in practice so it can be readily implemented. A comparison with conventionally
employed pseudo-probabilistic procedures for selecting the seismic coefficient (i.e., Bray
and Travasarou 2009) indicates that the results of performance-based procedure can differ
by up 30%. However, the trends in the results from both procedures are consistent. The
performance-based procedure is recommended because it directly addresses the selected
hazard level and the associated uncertainties.

ACKNOWLEDGMENTS
The first author acknowledges INNÓVATE-PERÚ under grant 092-14 for providing
partial financial support to perform this research. This work was also supported by funding
from the Faculty Chair in Earthquake Engineering Excellence, University of California,
Berkeley.
694 MACEDO ET AL.

APPENDIX
Please refer to the online version of this paper to access the supplementary material pro-
vided in the Appendix.

REFERENCES
Bray, J. D., 2007. Chapter 14: Simplified Seismic Slope Displacement Procedures. Earthquake
Geotechnical Engineering, in Proceedings of the 5th International Conference on Earthquake
Geotechnical Engineering Invited Lectures, in Geotechnical, Geological, and Earthquake
Engineering Series, Pitilakis, D. Kyriazis, Ed., Springer, Vol. 6, 327–353.
Bray, J. D., and Rathje, E. R., 1998. Earthquake-induced displacements of solid-waste landfills,
Journal of Geotechnical and Geoenvironmental Engineering, ASCE 124(3), 242–253.
Bray, J. D., and Travasarou, T., 2007. Simplified procedure for estimating earthquake-induced
deviatoric slope displacements, Journal of Geotechnical and Geoenvironmental Engineering
133(4), 381–392.
Bray, J. D., and Travasarou, T., 2009. Pseudostatic coefficient for use in simplified seismic slope
stability evaluation, Journal of Geotechnical and Geoenvironmental Engineering 135(9),
1336–1340.
Bray, J. D., Macedo, J. L., and Travasarou, T., 2017. Simplified procedure for estimating seismic
slope deviatoric displacements in subduction zones, Journal of Geotechnical and Geoenvir-
onmental Engineering, accepted for publication.
Deierlein, G., Krawinkler, H., and Cornell, C., 2003. A framework for performance-based earth-
quake engineering, Paper No. 140, in Proceedings of the Pacific Conference on Earthquake
Engineering, 13–15 February 2003, Christchurch, New Zealand.
Franke, K. W., Ulmer, K. J., Ekstrom, L. T., and Meneses, J. F., 2016. Clarifying the difference
between traditional liquefaction hazard maps and probabilistic liquefaction reference para-
meter maps, Soil Dynamics Earthquake Engineering 90(2016), 240–249.
Jibson, R. W., 2007. Regression models for estimating coseismic landslide displacement,
Engineering Geology 91(2–4), 209–218.
Kramer, S. L., and Mayfield, R. T., 2007. Return period of soil liquefaction, Journal of
Geotechnical and Geoenvironmental Engineering, ASCE 133(7), 802–813.
Lin, J. S., and Whitman, R. V., 1983. Decoupling approximation to the evaluation of earthquake-
induced plastic slip in earth dams, Earthquake Engineering and Structural Dynamics 11(5),
667–678.
Lin, J.-S., and Whitman, R. V., 1986. Earthquake induced displacements of sliding blocks,
Journal of Geotechnical Engineering 112(1), 44–59.
Makdisi, F., and Seed, H., 1978. Simplified procedure for estimating dam and embankment earth-
quake induced deformations, Journal of Geotechnical Engineering 104(7), 849–867.
Newmark, N., 1965. Effects of earthquakes on dams and embankments, Geotechnique 15(2),
139–160.
Papadimitriou, A. G., Bouckovalas, G. D., and Andrianopoulos, K. I., 2014. Methodology for
estimating seismic coefficients for performance-based design of earth dams and tall embank-
ments, Soil Dynamics Earthquake Engineering 56, 57–73.
Rathje, E. M., and Bray, J. D., 1999. An examination of simplified earthquake induced displace-
ment procedures for earth structures, Canadian Geotechnical Journal 36(1), 72–87.
doi:10.1139/t98-076.
PERFORMANCE-BASED PROBABILISTIC SEISMIC SLOPE DISPLACEMENT PROCEDURE 695

Rathje, E. M., and Bray, J. D., 2000. Nonlinear coupled seismic sliding analysis of earth struc-
tures, Journal of Geotechnical and Geoenvironmental Engineering 126(11), 1002–1014.
Rathje, E. M., and Bray, J. D., 2001. One- and two-dimensional seismic analysis of solid-waste
landfills, Canadian Geotechnical Journal 38(4), 850–862.
Rathje, E. M., and Saygili, G., 2008. Probabilistic seismic hazard analysis for the sliding dis-
placement of slopes: scalar and vector approaches, Journal of Geotechnical and Geoenviron-
mental Engineering, ASCE 134(6), 804–814.
Rathje, E. M., and Saygili, G., 2011. Pseudo-probabilistic versus fully probabilistic estimates of
sliding displacements of slopes, Journal of Geotechnical and Geoenvironmental Engineering,
ASCE 137(3), 208–217.
Rathje, E. M., Wang, Y., Stafford, P., Antonakos, G., and Saygili, G., 2014. Probabilistic assessment
of the seismic performance of slopes, Bulletin of Earthquake Engineering 12(3), 1071–1090.
Richards, R., and Elms, D. G., 1979. Seismic behaviour of gravity retaining walls, Journal of
Geotechnical Engineering Division, ASCE 105(4), 449–464.
Saygili, G., and Rathje, E. M., 2008. Empirical predictive models for earthquake-induced sliding
displacements of slopes, Journal of Geotechnical and Geoenvironmental Engineering, ASCE
134(6), 790–803.
Seed, H. B., 1979. Considerations in the earthquake-resistant design of earth and rockfill dams,
Geotechnique 29(3), 215–263.
Travasarou, T., Bray, J. D., and Der Kiureguian, A., 2004. A probabilistic methodology for
assessing seismic slope displacements, Paper No. 2326, in Proceedings of the 13th World
Conference on Earthquake Engineering, 1–6 August 2004, Vancouver, B.C., Canada.
Wang, Y., and Rathje, E., 2015. Probabilistic seismic landslide hazard maps including epistemic
uncertainty, Engineering Geology 196, 313–324.
Watson-Lamprey, J., and Abrahamson, N., 2006. Selection of ground motion time series and
limits on scaling, Soil Dynamics and Earthquake Engineering 26(5), 477–482.
Yegian, M. K., Marciano, E. A., and Ghahraman, V. G., 1991. Earthquake-induced permanent
deformations: probabilistic approach, Journal of Geotechnical Engineering 117, 35–50.
(Received 25 December 2016; accepted 23 December 2017)