You are on page 1of 12

Simplified Procedure for Estimating Earthquake-Induced

Deviatoric Slope Displacements


Jonathan D. Bray, F.ASCE1; and Thaleia Travasarou2

Abstract: A simplified semiempirical predictive relationship for estimating permanent displacements due to earthquake-induced devia-
toric deformations is presented. It utilizes a nonlinear fully coupled stick-slip sliding block model to capture the dynamic performance of
an earth dam, natural slope, compacted earth fill, or municipal solid-waste landfill. The primary source of uncertainty in assessing the
likely performance of an earth/waste system during an earthquake is the input ground motion. Hence, a comprehensive database contain-
ing 688 recorded ground motions is used to compute seismic displacements. A seismic displacement model is developed that captures the
primary influence of the system’s yield coefficient 共ky兲, its initial fundamental period 共Ts兲, and the ground motion’s spectral acceleration
at a degraded period equal to 1.5Ts. The model separates the probability of “zero” displacement 共i.e., 艋1 cm兲 occurring from the
distribution of “nonzero” displacement, so that very low values of calculated displacement do not bias the results. The use of the seismic
displacement model is validated through reexamination of 16 case histories of earth dam and solid-waste landfill performance. The
proposed model can be implemented rigorously within a fully probabilistic framework or used deterministically to evaluate seismic
displacement potential.
DOI: 10.1061/共ASCE兲1090-0241共2007兲133:4共381兲
CE Database subject headings: Dams; Earthquakes; Displacement; Structural reliability; Seismic effects; Slope stability.

Introduction a simple predictive relationship that includes quantification of the


uncertainty of key input variables. Pertinent previous studies are
The assessment of the likely performance of earth/waste struc- summarized, and while describing the model used to develop the
tures during earthquakes is important, because the failure of an “database” of simulated seismic displacements, the key improve-
earth dam, solid-waste landfill, natural slope, or constructed earth ments of the proposed approach are delineated. The proposed
embankment can result in significant human and financial losses seismic displacement model is presented, and following valida-
and severe environmental impact. Incorporating probabilistic tion by case histories of earth dam and solid-waste landfill
methodologies allow the engineer to quantify the uncertainty in seismic performance, recommendations are made on how this
the assessment of the likely performance of earth/waste slopes. procedure can be used in practice.
Simplified procedures for evaluating earth/waste slope stability
are required to bring probabilistic methods into daily practice.
Although some previous efforts have partially addressed this Previous Studies
need, the wealth of new earthquake recordings that have recently
become available with more realistic nonlinear, coupled stick-slip The seismic stability of earth and waste slopes is commonly
models and probabilistic procedures offers the opportunity to ad- evaluated with seismically induced permanent deformation proce-
vance the manner that engineers evaluate seismic stability of dures based on the landmark works of Newmark 共1965兲 and Mak-
earth/waste slopes in practice. disi and Seed 共1978兲. The calculated seismic displacement from
The goal of this paper is to describe a simplified semiempirical these procedures, whether the procedure is simplified or
procedure that can be used in the assessment of the likely perfor- advanced, is viewed appropriately as an index of seismic perfor-
mance of earth/waste structures that may slide during earth- mance. Seismic displacement estimates will always be approxi-
quakes. The procedure is developed to work within a fully mate in nature due to the complexities of the dynamic response of
probabilistic seismic hazard assessment, but it can be used also as the earth/waste materials involved and the variability of the earth-
quake ground motion. However, when viewed as an index of
1
Professor, Dept. of Civil and Environmental Engineering, Univ. of potential seismic performance, the calculated seismic displace-
California, Berkeley, CA 94720-1710. E-mail: bray@ce.berkeley.edu ment can and has been used effectively in practice to evaluate
2
Project Engineer, Fugro-West Inc., 1000 Broadway, Suite 200, earth/waste structure designs. Comprehensive discussions of these
Oakland, CA 94607-4099. E-mail: ttravasarou@fugro.com procedures and their application in engineering practice have
Note. Discussion open until September 1, 2007. Separate discussions been presented previously by investigators, such as Lin and Whit-
must be submitted for individual papers. To extend the closing date by
man 共1983兲, Ambraseys and Menu 共1988兲, Jibson 共1993兲, Bray et
one month, a written request must be filed with the ASCE Managing
Editor. The manuscript for this paper was submitted for review and pos-
al. 共1995兲, Kramer and Smith 共1997兲, Bray et al. 共1998兲, and
sible publication on June 1, 2005; approved on September 28, 2006. This Rathje and Bray 共2000兲. Most of these refinements of the New-
paper is part of the Journal of Geotechnical and Geoenvironmental mark procedures have been cast in a deterministic framework. A
Engineering, Vol. 133, No. 4, April 1, 2007. ©ASCE, ISSN 1090-0241/ smaller number of researchers have explored this problem from a
2007/4-381–392/$25.00. probabilistic perspective.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007 / 381

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
Lin and Whitman 共1986兲 studied the probability of failure of 共PEER兲 Center developed a probabilistic framework for the com-
sliding blocks using a rigid block assumption and modeling the putation of probabilities of exceedance of specified thresholds for
strong ground motion as a Gaussian stationary process. They variables that relate to the performance of engineered systems.
characterized ground motion pulses by their peak ground accel- The PEER framework, which is described in Deirlein et al.
eration 共PGA兲, root mean square acceleration, and central 共2003兲, is based on deconvolving overall system performance into
frequency, and characterized the slope strength by its yield coef- independent steps by introducing intermediate random variables.
ficient 共i.e., the seismic coefficient that brings the slope to a factor It is applied to seismic slope stability with the goal of computing
of safety of one in a pseudostatic analysis兲. In the end, seismic probabilities of exceeding specified displacement thresholds as
displacements are only conditioned on PGA and yield coefficient,
and the computation of the annual probability of exceeding speci-
fied displacement thresholds involves computing the seismic haz-
ard for one scalar parameter, i.e., PGA. Also using the Newmark
␭共D兲 = 冕
IM
G共D兩IM兲兩d␭共IM兲兩 共1兲

rigid sliding block method, Ambraseys and Menu 共1988兲 devel-


oped predictive equations for seismic displacement as functions where ␭⫽annual rate of exceedance; D⫽seismic displacement;
of the ratio of yield acceleration to PGA, the ground motion’s IM⫽intensity measure that characterizes one or more important
PGA and predominant period 共T p兲, and the event’s magnitude aspects of the ground motion, G⫽conditional probability of the
共M w兲 and source-to-site distance 共R兲. They used 50 earthquake seismic displacement exceeding D given IM, and
records, and restrict the use of their equations within the magni- 兩d␭共IM兲兩⫽absolute value of the derivative of the hazard curve for
tude range of 6.6–7.3. Ambraseys and Srbulov 共1994兲 developed the selected intensity measure. The goal of this study is to develop
an updated attenuation relationship for seismic displacement a simplified model for G, the conditional probability of seismic
based on additional earthquake records. More recently, Jibson et displacement exceeding a specific amount given an intensity mea-
al. 共1998兲 proposed a relationship for seismic displacement of a sure that characterizes the earthquake ground motion. This model
rigid block as a function of Arias intensity and yield coefficient may be incorporated in Eq. 共1兲 to estimate the seismic displace-
based on 280 independent earthquake records for use in develop- ment hazard. Alternatively, this model may be used directly to
ing probabilistic landslide hazard maps. estimate the expected range of seismic displacement for a speci-
Yegian et al. 共1991a,b兲 also adopted the Newmark rigid sliding fied earthquake scenario.
block assumption, and they normalized seismic displacements by
the ground motion’s PGA, T p, and number of equivalent cycles of
loading 共Neq兲. Based on simulated displacement data computed Basis for Proposed Procedure
from 86 earthquake records, they developed a relationship be-
tween the normalized seismic displacements and the ratio of yield
General
acceleration to PGA. As a result, seismic displacement is condi-
tioned on more than one ground motion parameter, which requires In this application, probabilistic methodologies usually involve
the computation of the joint hazard of these parameters. In a three steps: 共1兲 establishing a model for prediction of seismic
subsequent study, Ghahraman and Yegian 共1996兲 proposed a slope displacements, where seismic displacements are condi-
probabilistic procedure for calculating seismic displacement as a tioned on a number of variables characterizing the important
function of magnitude and distance, and yield coefficient. The ground motion characteristics and slope properties; 共2兲 computing
proposed relationship can be directly programmed in seismic haz- the joint hazard of the conditioning ground motion variables, and
ard analysis software to calculate annual probabilities of specified 共3兲 integrating the above-mentioned two steps to compute the
displacements being exceeded. The relationship was developed by seismic displacement hazard. Focusing on the first step, the cur-
making all primary variables 共e.g., PGA, T, and Neq兲 functions of rent study proposes a new relationship for seismic slope displace-
magnitude and distance. Consequently, the standard deviation of ment, which can be cast in a probabilistic framework similar to
the random error in this relationship is significant, i.e., on the that proposed by PEER. The use of the proposed relationship in
order of 5 in ln units, which results in the computation of a wide predicting seismic displacement hazard is discussed in Travasarou
range of seismic displacements. et al. 共2004兲.
The Bray et al. 共1998兲 seismic slope stability procedure pro- With regards to a predictive relationship for seismic slope dis-
vides median and standard deviation estimates of the seismic placements the differences between models proposed by research-
demand 共i.e., maximum horizontal equivalent acceleration兲 and ers relate to the following issues: 共1兲 type 共i.e., simulated or re-
normalized seismic displacement, but do so only in an approxi- corded兲 and number of earthquake records used; 共2兲 type of
mate manner to develop a sense of the variability of the estimated idealized model used for the calculation of seismic displacements;
displacement. However, Stewart et al. 共2003兲 were able to use this 共3兲 selection of the conditioning variables; and 共4兲 mathematical
procedure to develop a probabilistic screening analysis for decid- model used to develop the predictive equation for seismic slope
ing if detailed project-specific seismic slope stability investiga- displacement.
tions are required by the 1990 California Seismic Hazards
Mapping Act. Kim and Sitar 共2003兲 investigated key factors that
Earthquake Ground Motions
influenced seismic slope displacements using simulated rather
than recorded acceleration-time histories. Their study found that Currently available simplified slope displacement estimation pro-
the variability in calculated seismic displacement is primarily cedures were developed based on artificial simulations of earth-
controlled by the significant variability in the earthquake ground quake ground motion or on a modest number of actual earthquake
motion, and it is relatively less affected by the variability in the recordings. This study takes advantage of the recently augmented
earth slope properties. This finding is consistent with that of Ye- database of earthquake recordings, which provides the opportu-
gian et al. 共1991b兲. nity to characterize better the important influence of ground
Research at the Pacific Earthquake Engineering Research motions on the seismic performance of an earth/waste slope. As

382 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
viscoelastic modal analysis that uses strain-dependent material
properties to capture the nonlinear response of earth and waste
materials. It considers a single mode shape, but the effect of in-
cluding three modes was shown to be small. The results from this
model have been shown to compare favorably with those from a
fully nonlinear D-MOD-type stick-slip analysis 共Rathje and Bray
2000兲, but this model can be utilized in a more straightforward
and transparent manner. The model used herein is one dimen-
sional 共i.e., a relatively wide vertical column of deformable soil兲
to allow for the use of a large number ground motions with wide
range of properties of the potential sliding mass in this study.
One-dimensional 共1D兲 analysis has been found to provide a rea-
sonably conservative estimate of the dynamic stresses at the base
of two-dimensional 共2D兲 sliding systems 共e.g., Vrymoed and Cal-
Fig. 1. 共a兲 Generic seismic slope displacement problem of height H zascia 1978; Elton et al. 1991兲 and the calculated seismic dis-
and initial stiffness Vs and 共b兲 idealized nonlinear coupled stick-slip placements 共Rathje and Bray 2001兲. However, 1D analysis can
deformable sliding mass model with one-way sliding used in study underestimate the seismic demand for shallow sliding at the top
of 2D systems where topographic amplification is significant. For
this special case, the input PGA can be amplified as recommended
discussed previously, the uncertainty in the ground motion char- by Rathje and Bray 共2001兲 for moderately steep slopes 共i.e.,
acterization is by far the dominant source of uncertainty in calcu- ⬃1.25 PGA兲 and by Ashford and Sitar 共2002兲 for steep slopes
lating seismic displacements. 共i.e., ⬃1.5 PGA兲. In this study, the sliding mass was assigned a
The ground motion database used to generate the seismic dis- constant unit weight of 17.6 kN/ m3, the strain-dependent shear
placement data comprises available records from shallow crustal modulus reduction and material damping ratio curves for a soil
earthquakes that occurred in active plate margins 共PEER strong with a plasticity index of 30 共Vucetic and Dobry 1991兲. Sensitiv-
motion database 具http://peer.berkeley.edu/smcat/index.html典兲. ity analyses indicate that reasonable adjustments of these param-
These records conform to the following criteria: 共1兲 5.5艋 M w eters do not have a significant effect on the computed
艋 7.6, 共2兲 R 艋 100 km, 共3兲 Simplified Geotechnical Sites B, C, or displacements.
D 关i.e., rock, soft rock/shallow stiff soil, or deep stiff soil, respec- The Newmark sliding block mechanism used in this nonlinear
tively, Rodriguez-Marek et al. 共2001兲兴, and 共4兲 frequencies in the coupled stick-slip slope model captures that part of the seismi-
range of 0.25– 10 Hz have not been filtered out. Earthquake cally induced permanent displacement attributed to deviatoric
records totaling 688 from 41 earthquakes comprise the ground shear deformation 共i.e., either rigid body slippage along a distinct
motion database for this study 关see Travasarou 共2003兲 for a list of failure surface or distributed deviatoric shearing within the de-
records used兴. The two horizontal components of each record formable sliding mass兲. Ground movement due to volumetric
were used to calculate an average seismic displacement for each compression is not explicitly captured by Newmark-type models.
side of the records, and the maximum of these values was as- This is an important distinction of this slope displacement model.
signed to that record. The top of a slope can displace downward due to deviatoric de-
formation or volumetric compression of the slope-forming mate-
rials. However, top of slope movements resulting from distributed
Idealized Sliding Mass Model
deviatoric straining within the sliding mass or stick-slip sliding
Most of the available simplified probabilistic slope displacement along a failure surface are mechanistically different than top of
procedures employ the original Newmark rigid sliding block as- slope movements that result from seismically induced volumetric
sumption 共e.g., Lin and Whitman 1986; Ambrasseys and Menu compression of the materials forming the slope. The Newmark
1988; and Yegian et al. 1991b兲, which is known not to capture sliding block provides a simplified mechanistic analogy to capture
accurately the dynamic response of the deformable earth/waste deviatoric deformation. The finding that a Newmark procedure
potential sliding mass during earthquake shaking 共Seed and Mar- sometimes captures the overall top of slope displacement for
tin 1966; and Makdisi and Seed 1978兲. Some of the methods cases where seismic compression due to volumetric straining of
discussed previously recognize the deformability of the earth/ the earth or waste mass is the dominant mechanism is fortuitous,
waste system, but they do so in a decoupled fashion, which has because those seismic forces that tend to produce large volumetric
been shown to be overly conservative or slightly unconservative compression strains also often induce large calculated displace-
depending on the slope properties 共Kramer and Smith 1997; and ments in a Newmark method. There are cases where the
Rathje and Bray 2000兲. Compared to the rigid sliding block Newmark method does not capture the overall top of slope dis-
model, a nonlinear coupled stick-slip deformable sliding block placement, such as the seismic compression of large compacted
model offers a more realistic representation of the dynamic re- earth fills 共e.g., Stewart et al. 2001兲. Hence, it is preferred to
sponse of an earth/waste structure by accounting for the deform- separate these effects and use procedures based on the sliding
ability of the sliding mass and by considering the simultaneous block model to estimate deviatoric-induced displacements and use
occurrence of its nonlinear dynamic response and periodic sliding procedures based on the seismic compression of soils 共e.g., Toki-
episodes. In addition, its validation against shaking table experi- matsu and Seed 1987兲 to estimate volumetric-induced
ments provides confidence in its use 共Wartman et al. 2003兲. displacements.
The nonlinear coupled stick-slip deformable sliding model This nonlinear coupled stick-slip deformable sliding model
proposed by Rathje and Bray 共2000兲 for one-directional sliding, can be characterized by: 共1兲 its strength as represented by its yield
which is shown in Fig. 1, is used in this study. The seismic re- coefficient 共ky兲, and 共2兲 its dynamic stiffness as represented by its
sponse of the sliding mass is captured by an equivalent-linear initial fundamental period 共Ts兲. Seismic displacement values were

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007 / 383

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
Fig. 2. Efficiency of relating seismic displacement to three ground motion intensity measures: 共a兲 Arias intensity for a stiff, weak slope; 共b兲
spectral acceleration for a flexible, weak slope; and 共c兲 peak ground velocity for a flexible, weak slope

generated by computing the response of the idealized sliding mass quire that the ground motion hazard be available for the vector of
model with specified values of its yield coefficient 共i.e., ky = 0.02, IMs, which is significantly more complex. The goal of this study
0.05, 0.075, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, and 0.4兲 and its initial is to develop a simplified procedure that can be used in practice,
fundamental period 共i.e., Ts = 0, 0.2, 0.3, 0.5, 0.7, 1.0, 1.4, and so use of a single IM is desirable.
2.0 s兲 to the entire set of recorded earthquake motions described The efficiency and sufficiency of estimating seismic displace-
previously. For the baseline case, the overburden-stress corrected ment given an IM of an earthquake ground motion for 29 differ-
shear wave velocity 共Vs1兲 was set to 250 m / s, and the shear wave ent IMs were investigated. The optimal IMs included PGA,
velocity profile of the sliding block was developed using the re- spectral acceleration 共Sa兲, root mean square acceleration 共arms兲,
lationship that shear wave velocity 共Vs兲 is proportional to the peak ground velocity 共PGV兲, Arias intensity 共Ia兲, effective peak
fourth-root of the vertical effective stress. The sliding block velocity 共EPV兲, Housner’s response spectrum intensity 共SI兲, and
height 共H兲 was increased until the specified value of Ts was ob- Ang’s characteristic intensity 共Ic兲. Partial results from this study
tained. For common Ts values from 0.2 to 0.7 s, another reason- are shown in Fig. 3 关see Travasarou and Bray 共2003a兲 for full
able combination of H and average Vs were used to confirm that definitions and a more complete comparison兴. For period-
the results were not significantly sensitive to these parameters independent IMs 共i.e., no knowledge of the fundamental period of
individually. For nonzero Ts values, H varied between 12 and the potential sliding mass is required兲, Arias intensity was found
100 m, and the average Vs was between 200 and 425 m / s. All to be the most efficient IM for a stiff, weak slope, and response
sliding block systems would be classified as 2003 International spectrum intensity was found to be the most efficient for a flexible
Building Code Sites C or D. Hence, realistic values of the initial slope. No one period-independent IM was found to be adequately
fundamental period and yield coefficient for a wide range of earth efficient for slopes of all dynamic stiffnesses and strengths. How-
dams, earth fills, natural slopes, and solid-waste landfills were ever, the 5% damped elastic spectral acceleration at the degraded
used. fundamental period of the slope was found to be the optimal IM
across the wide range of slope periods and strengths considered in
this study.
Selection of Independent Variables
An estimate of the fundamental period of the slope is required
The amount of earthquake-induced seismic displacement of an with this IM, but an estimate of Ts is useful in characterizing the
embankment depends on the characteristics of the strong ground slope anyway, so it is accepted that spectral acceleration can be
motion and the slope properties. In the majority of the simplified
procedures for estimating seismic slope displacement, researchers
have used PGA as the primary ground motion intensity measure.
PGA has sometimes been supplemented by additional parameters
characterizing the frequency content and duration of the ground
motion. For example, Makdisi and Seed 共1978兲 used earthquake
magnitude as a proxy for duration in combination with the esti-
mated PGA at the crest of the embankment; Yegian et al. 共1991b兲
used predominant period and equivalent number of cycles of
loading in combination with PGA; and Bray et al. 共1998兲 used the
mean period and significant duration of the design rock motion in
combination with its PGA.
To compute the displacement hazard in a framework compat-
ible to that outlined in Eq. 共1兲, however, it is desirable to use a
single ground motion intensity measure 共IM兲, which satisfies the
requirements for efficiency and sufficiency 共i.e., it minimizes the
variability in the correlation with seismic displacement 共Fig. 2兲,
and it renders the relationship independent of other variables, re-
spectively, Cornell and Luco 2001兲. No one ground motion IM
could be expected to be fully sufficient in capturing all influenc-
ing aspects of the ground motion 共i.e., its intensity, frequency
content, and duration兲. Thus, the goal is to identify the optimal IM
that is as efficient and sufficient as possible. A vector containing Fig. 3. Relative efficiency of selected ground motion intensity
several IMs could be used, but then implementation would re- parameters 共data from Travasarou 2003兲

384 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
used to characterize the earthquake shaking. Due to material non-
linearity, it would be expected that the spectral acceleration at a
degraded period would be better than that at its initial period.
After evaluating 11 cases, it was found that the spectral accelera-
tion at a degraded period equal to 1.5 times the initial fundamen-
tal period of the slope, i.e., Sa共1.5Ts兲, was overall the most
efficient IM 共Travasarou and Bray 2003a兲. This degraded funda-
mental period captures the overall average stiffness reduction for
the earth/waste slopes considered in this study. Although
Sa共1.5Ts兲 does not satisfy sufficiency for all slopes, it was found
to be the most sufficient of the IMs examined. Spectral accelera-
tion does directly capture the important ground motion character-
istics of intensity and frequency content in relation to the
fundamental period of the potential sliding mass, and it indirectly
partially captures the influence of duration in that it tends to in-
crease as earthquake magnitude 共i.e., duration兲 increases. An ad-
ditional benefit of selecting spectral acceleration to represent the
ground motion is that the ground motion hazard for spectral ac-
celeration can be computed relatively easily due to the existence
of several robust attenuation relationships.
The slope’s yield coefficient 共ky兲 and initial fundamental pe-
Fig. 4. Distribution of simulated displacement data for D ⬎ 1 cm
riod 共Ts兲 were selected to represent the dynamic strength and with moment magnitude, rupture distance, yield coefficient, initial
stiffness, respectively, of the earth/waste slope in the seismic dis- fundamental period, and spectral acceleration at 1.5 times the initial
placement model. The yield coefficient represents the dynamic fundamental period
resistance of the slope, so it is second in importance to only the
key characteristics of the ground motion, which represents the
dynamic loading. The yield coefficient has always been used in slope yield coefficient, the slope’s initial fundamental period, and
simplified sliding block procedures due to its important effect on the ground motion’s spectral acceleration at 1.5 times the slope’s
seismic displacement. Simplified equations for calculating ky as a initial fundamental period. The scatter in these graphs is signifi-
function of slope geometry and strength are found in Bray et al. cant. However, examining the trends of the data, seismic displace-
共1998兲 among several other works. The primary issue in calculat- ment increases with increasing magnitude and decreases with
ing ky is estimating the dynamic strength of the critical strata increasing distance as expected. Spectral acceleration can capture
within the slope. Several publications include extensive discus- these effects as a ground motion parameter. As shown in the bot-
sions of the dynamic strength of soil 共e.g., Blake et al. 2002; tom graph, the amount of calculated seismic displacement is cor-
Duncan and Wright 2005兲, so the reader is referred to the litera- related to the spectral acceleration of the input earthquake ground
ture for a full discussion of this important topic. For this study, it motion at the degraded period of the slope, with displacement
is assumed that ky is constant, so consequently, the earth materials increasing significantly as Sa共1.5Ts兲 increases. Additionally, seis-
do not undergo severe strength loss as a result of earthquake mic displacement decreases with increasing yield coefficient as
shaking 共e.g., no liquefaction兲. expected. There is also a modest sensitivity of seismic displace-
Research by investigators 共e.g., Bray et al. 1998兲 has found ment to the slope’s fundamental period. The results shown in Fig.
that seismic displacement also depends on the dynamic response 4 and the calculated seismic displacement values that are less than
characteristics of the potential sliding mass or its stiffness. All 1 cm are the “simulated data” used to develop the regression
other factors held constant, seismic displacements increase when equations for estimating seismic displacements.
the sliding mass is near resonance compared to that calculated for
very stiff or very flexible slopes 共e.g., Kramer and Smith 1997;
Functional Forms of Predictive Equations
Rathje and Bray 2000; Wartman et al. 2003兲. This effect can be
taken into account by including the initial fundamental period in Currently available probabilistic procedures only address the case
the predictive equation. For these seismic stability calculations, of nonzero displacements occurring due to seismic loading. How-
the initial fundamental period of the sliding mass 共Ts兲 can be ever, situations commonly arise where a combination of earth-
estimated using the expression: Ts = 4H / Vs for the case of a rela- quake loading and slope properties will result in no significant
tively wide potential sliding mass that is either shaped like a deformation of an earth/waste system. Consequently, the finite
trapezoid or segment of a circle where its response is largely 1D probability of obtaining negligible 共“zero”兲 displacement should
共e.g., Rathje and Bray 2001兲, where H⫽the average height of the be modeled as a function of the independent random variables.
potential sliding mass, and Vs⫽average shear wave velocity of the Thus, during an earthquake, an earth slope may experience zero
sliding mass. For the case of a triangular-shaped sliding mass that or finite permanent displacements depending on the characteris-
largely has a 2D response, the expression: Ts = 2.6H / Vs should be tics of the strong ground motion and the slope’s dynamic proper-
used. ties and geometry. As discussed in Travasarou and Bray 共2003b兲,
The seismic displacements 共⬎1 cm兲 calculated for the set of seismically induced permanent displacements can be modeled as
idealized slopes defined previously, each with its specific yield a mixed random variable, which has a certain probability mass at
coefficient and initial fundamental period, undergoing the 688 zero displacement and a probability density for finite displace-
earthquake ground motions described previously are shown in ment values. Displacements smaller than 1 cm are not of
Fig. 4. The variation of the calculated seismic displacement is engineering significance and can for all practical purposes be con-
plotted against earthquake magnitude, source-to-site distance, sidered as negligible or zero. Additionally, the regression of dis-

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007 / 385

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
Fig. 5. 共a兲 Probability density function and 共b兲 probability of
exceedance for a mixed and a continuous random variable

placement as a function of a ground motion intensity measure


should not be dictated by data at negligible levels of seismic
displacement.
For the sake of clarity in the formulation of the relevant equa-
tions, the values of seismic displacement that are smaller than Fig. 7. Comparison of predicted probability of zero displacement
1 cm can be lumped to d0 = 1 cm. The probability density function 共i.e., D 艋 1 cm兲 versus the simulated displacement data for a slope
of seismic displacement is then described as with ky = 0.2
f D共d兲 = p̃␦共d − d0兲 + 共1 − p̃兲f̃ D共d兲 共2兲
where f D共d兲⫽displacement probability density function; ing the probability of occurrence of zero displacement was guided
by the trends shown in Fig. 6. A probit regression model was used
p̃⫽probability mass at D = d0; ␦共d − d0兲⫽Dirac delta function; and
for this analysis 共Green 2003兲.
f̃ D共d兲⫽displacement probability density function for D ⬎ d0. Con- In the case where a finite probability of “nonzero” displace-
trary to a continuous random variable, the mixed random variable ment is calculated, the amount of nonzero displacement needs to
can take on discrete outcomes with finite probabilities at certain be estimated. The distribution of seismic displacement is com-
points on the line as well as outcomes over one or more continu- puted, given that nonzero displacement has occurred. A truncated
ous intervals with specified probability densities. Fig. 5 illustrates regression model was used as described in Green 共2003兲. The
the case of a mixed variable with finite probability mass at D estimation of the values of the model coefficients was performed
= d0 and continuous probability density for D ⬎ d0. Using this using the principle of maximum likelihood.
formulation, the probability of exceedance at small displacements
can be smaller than 1, recognizing the possibility of a finite prob-
ability of nonfailure 关Fig. 5共b兲兴. Model for Estimating Seismic Deviatoric
Compatible with the concept of a mixed random variable, the Displacements
predictive model for seismic displacement consists of two dis-
crete steps. First, the probability of occurrence of zero displace- The predictive model for estimating seismic displacement con-
ment 共i.e., D 艋 1 cm兲 is computed as a function of the primary sists of two discrete computations of: 共1兲 The probability of neg-
independent variables ky, Ts, and Sa共1.5Ts兲. The dependence of ligible “zero” displacement, and 共2兲 the likely amount of nonzero
the probability of zero displacement on the three independent displacement. The model for computing the probability of zero
variables is illustrated in Fig. 6. The probability of zero displace- displacement is
ment increases significantly as the yield coefficient increases 关Fig.
6共a兲兴, and decreases significantly as the ground motion’s spectral P共D = 0兲 = 1 − ⌽共− 1.76 − 3.22 ln共ky兲 − 0.484Ts ln共ky兲
acceleration at the degraded period of the slope increases 关Fig. + 3.52 ln共Sa共1.5Ts兲兲兲 共3兲
6共c兲兴. The influence of the slope’s fundamental period is shown in
Fig. 6共b兲. The probability of zero displacement decreases initially where P共D = 0兲⫽probability 共as a decimal number兲 of occurrence
as the fundamental period increases from zero, because the slope of zero displacements; D-seismic displacement; ⌽-standard nor-
is being brought near to the mean period of most ground motions. mal cumulative distribution function 共i.e., NORMSDIST in
However, this probability increases sharply as the slope’s period Excel兲; ky⫽yield coefficient; Ts⫽initial fundamental period of the
continues to increase as it is now moving away from the reso- sliding mass in seconds, and Sa共1.5Ts兲 is the spectral acceleration
nance condition. The selection of the functional form for model- of the input ground motion at a period of 1.5Ts in the units of g.
An example of the model predictions versus the simulated data is
shown in Fig. 7 for four cases of slopes with yield coefficient
equal to 0.2 and four different values of the initial fundamental
period.
The amount of the nonzero displacement 共D兲 in centimeters
was first estimated as
ln共D兲 = − 1.10 − 2.83 ln共ky兲 − 0.333共ln共ky兲兲2
+ 0.566 ln共ky兲ln共Sa共1.5Ts兲兲 + 3.04 ln共Sa共1.5Ts兲兲
Fig. 6. Dependence of the probability of zero displacement on the 共a兲 − 0.244共ln共Sa共1.5Ts兲兲兲2 + 1.50Ts ± ␧ 共4兲
yield coefficient; 共b兲 initial fundamental period; and 共c兲 spectral where ky, Ts, and Sa共1.5Ts兲 are as defined previously for Eq. 共3兲,
acceleration at 1.5 times the initial fundamental period and ␧⫽normally distributed random variable with zero mean and

386 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
Fig. 8. Residuals 共lnDdata − lnDpredicted兲 of Eq. 共5兲 plotted versus
magnitude, rupture distance, the yield coefficient, and the initial
fundamental period

standard deviation ␴ = 0.67. When the residuals 共i.e.,


lnDdata − lnDpredicted兲 are plotted versus magnitude, there is a mag- Fig. 9. Model trends: 共1兲 for an M w = 7 strike-slip earthquake at a
nitude dependence, because Sa共1.5Ts兲 is not fully sufficient with soil site at a distance of 10 km. 共a兲 Probability of negligible
respect to magnitude 共or duration兲. This dependence was ad- displacements as a function of yield coefficient and initial
dressed by incorporating a magnitude term in the predictive equa- fundamental period using Eq. 共3兲; 共b兲 median displacement estimate
tion. Hence, the recommended relationship for estimating the using Eq. 共5兲; 共c兲 median seismic displacement versus yield
amount of nonzero seismic displacement 共D兲, which is not biased coefficient using Eq. 共7兲 and 共8兲. For an M w = 7.5 earthquake at
due to magnitude, is various ground motion intensity levels: 共d兲 rigid block; 共e兲 initial
fundamental period of 0.3 s with ±1 s . d. for PGA= 0.8 g; and 共f兲
ln共D兲 = − 1.10 − 2.83 ln共ky兲 − 0.333共ln共ky兲兲2 effect of magnitude for one spectral acceleration value and one initial
+ 0.566 ln共ky兲ln共Sa共1.5Ts兲兲 + 3.04 ln共Sa共1.5Ts兲兲 fundamental period.

− 0.244共ln共Sa共1.5Ts兲兲兲2 + 1.50Ts + 0.278共M − 7兲 ± ␧


共5兲
where ky, Ts, and Sa共1.5Ts兲 are as defined previously for Eq. 共3兲, for cases where Ts ranges from 0.05 to 2 s, and the first term of
and ␧ is a normally distributed random variable with zero mean these equations need only be replaced with −0.22 as was done in
and standard deviation ␴ = 0.66. All common coefficients of Eq. Eq. 共6兲 if evaluating nearly rigid potential sliding masses 共i.e.,
共5兲 are identical to those in Eq. 共4兲, except now the additional Ts ⬍ 0.05 s兲. The residuals shown in Fig. 8 are significant, but this
magnitude term has been added. As will be discussed later, to is due to the inherent variability of estimating seismic
eliminate the bias in the model when Ts ⬃ 0 s, the first term of displacement.
Eqs. 共4兲 and 共5兲 should be replaced with −0.22 when Ts ⬍ 0.05 s. Although the standard deviation is not significantly lower for
For the Newmark rigid sliding block case 共Ts = 0兲, Eq. 共5兲 Eq. 共5兲 when compared to that of Eq. 共4兲, Eq. 共5兲 is preferred
becomes because it eliminates the bias in the estimates of seismic displace-
ment with magnitude. Inclusion of the magnitude term, which
ln共D兲 = − 0.22 − 2.83 ln共ky兲 − 0.333共ln共ky兲兲2
captures in part the influence of duration, is intuitively appealing,
+ 0.566 ln共ky兲ln共PGA兲 + 3.04 ln共PGA兲 and it does not complicate the estimate of seismic displacement,
because in many cases the seismic hazard at a site is governed by
− 0.244共ln共PGA兲兲2 + 0.278共M − 7兲 ± ␧ 共6兲
earthquake events across a narrow range of magnitudes. Eq. 共4兲 is
where PGA⫽peak ground acceleration of the ground motion 关i.e., included, however, for cases where the inclusion of magnitude
Sa共Ts = 0兲兴. causes too much complication, although its use causes a system-
The residuals of Eq. 共5兲 are plotted in Fig. 8 versus some of atic bias in the results with seismic displacements being underes-
the key independent variables. The residuals of displacement ver- timated for large magnitude earthquakes 共M w ⬎ 7兲.
sus magnitude and yield coefficient show no bias. There remains It is often useful to establish a threshold displacement for ac-
only a slight dependence on distance, which is not significant, and ceptable seismic performance and then estimate the probability of
a moderate bias in the estimate at Ts = 0 and 2 s. The overestima- this threshold displacement being exceeded. Additionally, often a
tion at 2 s is not critical, because it is rare to have earth/waste range of expected seismic displacements is desired. The proposed
sliding masses with periods greater than 1.5 s, and Eq. 共5兲 is methodology can be used to calculate the probability of the seis-
conservative. However, the rigid body case 共i.e., Ts = 0兲 can be mic displacement exceeding a selected threshold of displacement
important for very shallow slides, and Eq. 共5兲 is unconservative 共d兲 for a specified earthquake scenario and slope properties. For
for this case. The rigid body case is a special case that could only example, consider a potential sliding mass with an initial funda-
be captured by incorporating a complicated high order function of mental period Ts, yield coefficient ky, and an earthquake scenario
Ts in Eqs. 共4兲 and 共5兲. Alternatively, the estimation at Ts = 0 s can that produces a spectral acceleration of Sa共1.5Ts兲. The probability
be corrected by replacing the first term 共i.e., −1.10兲 in Eqs. 共4兲 of the seismic displacement 共D兲 exceeding a specified displace-
and 共5兲 with −0.22. Hence, it is reasonable to use Eqs. 共4兲 and 共5兲 ment threshold 共d兲 is

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007 / 387

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
P共D ⬎ d兲 = 关1 − P共D = 0兲兴P共D ⬎ d兩D ⬎ 0兲 共7兲 seismic displacement increases rapidly. However, fundamental
period is also an important parameter to consider in addition to
The term P共D = 0兲 is computed using Eq. 共3兲. The term ground motion intensity and yield coefficient.
P共D ⬎ d 兩 D ⬎ 0兲 may be computed assuming that the estimated The proposed model has been developed for values of the
displacements are lognormally distributed as yield coefficient between 0.02 and 0.4, fundamental period be-

P共D ⬎ d兩D ⬎ 0兲 = 1 − P共D 艋 d兩D ⬎ 0兲 = 1 − ⌽ 冉 ln d − ln d̂



冊 tween 0 and 2.0 s, and Sa共1.5Ts兲 between 0.002 and 2.7 g, and
they should be used for cases within these ranges to provide rea-
sonable results. The yield coefficient is assumed to be constant, so
共8兲 caution is warranted when evaluating slopes comprised of mate-
rials susceptible to significant strength loss as a result of earth-
where ln共d̂兲 is computed using Eq. 共5兲; ␴⫽standard deviation of quake shaking 共i.e., liquefaction兲. The calculated seismic
the random error, which in this case is 0.66; and ⌽⫽standard displacement is moderately dependent on foundation conditions
normal cumulative distribution function. 共i.e., rock versus stiff soil兲, which was not modeled in the pro-
The trends in the estimates from the proposed model are posed equations to facilitate their use in seismic displacement
shown in Fig. 9. The upper graphs are results for a M w = 7 strike- hazard calculations. Due to the relative abundance of earthquake
slip earthquake at a soil site at a distance of 10 km. The probabil- motions recorded at soft rock/shallow stiff soil and deep stiff soil
ity of negligible seismic displacements depends significantly on sites, the proposed relationships can be used for these sites with-
yield coefficient and the initial fundamental period of the slope at out adjustment. However, reexamination of the seismic slope dis-
high yield coefficients. The influence of yield coefficient on seis- placements calculated with rock ground motions indicates that the
mic displacement is clearly apparent looking at Figs. 9共b and c兲. proposed relationships overestimate seismic displacements of
The lower graphs are for a M w = 7.5 earthquake at several dis- earth/waste structures situated on rock by about 15%.
tances from the fault so that the ground motion intensity param-
eters PGA or Sa共0.45 s兲 vary significantly for the case of a rigid
sliding block or deformable sliding block in Figs. 9共d and e兲, Model Validation
respectively. The partial effect of earthquake magnitude at a par-
ticular level of ground motion intensity 关i.e., Sa共0.45 s兲 = 0.8 g兴 is The ability of the proposed model to predict reliably observed
shown in Fig. 9共f兲. It is a partial effect, because the estimated performance was examined through comparison with observed
value of spectral acceleration typically increases with increasing displacements measured at 16 earth dams and solid-waste land-
earthquake magnitude, which increases seismic displacement. fills that underwent strong earthquake shaking. The observations
This effect is not shown in this figure, because spectral accelera- from these case histories were used solely to validate the model
tion was held constant. Fig. 9 illustrates that as yield coefficient and were not included in the data set for the development of the
decreases in relationship to the intensity of the ground motion the predictive equation. The case histories used in the model valida-

Table 1. Comparison of the Range of Computed Displacement Using Three Different Methods with the Maximum Observed Displacement
Proposed methodc
Makdisi and Bray
Obs. Seed 1978 et al. 1998
Dmax Site Sa共1.5Ts兲 Est. disp D D
System EQa 共cm兲 b ky Ts共s兲 class 共g兲 P共D = 0兲 共cm兲 P共D ⬎ Dmax兲 共cm兲 共cm兲
BuenaVista LF LP None 0.26 0.64 Alluvium 0.36 0.75 0–2 0.2 0 0
Guadalupe LF LP Minor 0.20 0.64 Rock 0.21 0.95 0 0.0 0 0–4
Pacheco Pass LF LP None 0.30 0.76 Rock 0.12 1.0 0 0.0 0 0
Marina LF LP None 0.26 0.59 Alluvium 0.30 0.9 0 0.05 0 0
Austrian Dam LP 50 0.14 0.33 Soft rock 0.94 0.0 20–70 0.3 1–30 20–100
Lexington Dam LP 15 0.11 0.31 Soft rock 0.78 0.0 15–65 0.9 0–10 30–110
Lopez Canyon C-A LF NR None 0.27 0.64 Soft rock 0.48 0.4 0–5 0.6 0 0
Lopez Canyon C-B LF NR None 0.35 0.45 Soft rock 0.43 0.85 0 0.1 0 0
Chiquita Canyon C LF NR 24 0.09 0.64 Soft rock 0.35 0.0 10–30 0.3 1–40 3–20
Chiquita Canyon D LF NR 30 0.10 0.64 Soft rock 0.35 0.0 5–20 0.1 0–10 2–15
Sunshine Canyon LF NR 30 0.31 0.77 Soft rock 1.40 0.0 20–70 0.6 0 0
Oll Section HH LF NR 15 0.08 0.00 Soft rock 0.24 0.1 4–15 0.1 3–30 2–25
Chabot Dam SF Minor 0.14 0.55 Rock 0.26 0.35 0–5 0.2 1–10 0–1
La Villita Dam S3 1 0.20 0.60 Alluvium 0.20 0.95 0 0.01 0 0
La Villita Dam S4 1.4 0.20 0.60 Alluvium 0.33 0.5 0–4 0.4 0 0
La Villita Dam S5 4 0.20 0.60 Alluvium 0.41 0.25 0–10 0.5 0–1 0

a
LP= 1989 Loma Prieta; NR= 1994 Northridge; SF= 1906 San Francisco; S3, S4, and S5 from Elgamal et al. 共1990兲.
b
Dmax⫽maximum observed displacement.
c
P共D = 0兲 from Eq. 共3兲. 0 displacement refers to negligible displacement 共i.e. D 艋 1 cm兲. Estimated displacement range is median ± ␴ from Eqs. 共3兲–共8兲.
P共D ⬎ Dmax兲 from Eqs. 共7兲 and 共8兲 using 1 or 5 cm for Dmax⫽None or Minor, respectively, and using observed Dmax value when available.

388 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
tion are described in Table 1. In all cases, the maximum observed There are four cases of observed moderate seismic displace-
displacement 共Dmax兲 is that portion of the permanent displacement ment for solid-waste landfills during the 1994 Northridge earth-
attributed to stick-slip type movement and distributed deviatoric quake 共i.e., Dmax = 15– 30 cm兲. In all of these cases, the proposed
shear within the deformable mass, and crest movement due to method estimates a very low chance of zero displacement occur-
volumetric compression was subtracted from the total observed ring. Moreover, the observed seismic displacement is within the
permanent displacement when appropriate to be consistent with range of the estimated seismic displacement for all cases, except
the mechanism implied by the Newmark method. The observed for one case. For this case, i.e., Chiquita Canyon D Landfill, the
seismic performance and best estimates of yield coefficient and observed value of 30 cm is only slightly above the upper esti-
initial fundamental period are based on the information provided mated value of 20 cm and within 1.5 s . d. of the median estimate
in Makdisi and Seed 共1978兲, Bray et al. 共1998兲, Harder et al. of seismic displacement, indicating that even for this case the
共1998兲, and Elgamal et al. 共1990兲. Complete details regarding model predictions are reasonable. It is also noteworthy that the
these parameters and pertinent seismological characteristics of the seismic displacement of Chiquita Canyon D Landfill is underes-
corresponding earthquakes can be found in Travasarou 共2003兲. timated by all simplified methods. For this case, the liner interface
The comparison of the proposed model estimates of seismic was subjected to significant strain prior to the earthquake, which
displacement 共Columns 8–10兲 with the maximum observed seis- may have caused high residual stresses, which induced more dis-
mic permanent displacement 共Column 3兲 is shown in Table 1. For placement during the earthquake episode 共Augello et al. 1995兲.
this comparison, only the best estimate of the slope’s yield coef- Lastly, there are two cases of moderate seismic displacement
ficient, its initial fundamental period, and the spectral acceleration of earth dams shaken by the 1989 Loma Prieta Earthquake. Lex-
at 1.5 times the initial fundamental period of the slope as repre- ington Dam and Austrian Dam underwent approximately 15 and
sented by their mean values for the first two slope parameters and 50 cm of seismic displacement, respectively, due to deviatoric
the median value for the last earthquake parameter are consid- straining from this event. For both cases the proposed method
ered. Hence, the computed displacement range is due to the vari- estimates essentially no chance of zero displacement occurring
ability in the seismic displacement given the value of the slope and ranges of expected seismic displacement that include the ob-
properties and the seismic load 关i.e., ␴ln D = 0.66 from Eq. 共5兲兴. served seismic displacement. In one case the observed seismic
The displacement range shown in Column 9 is the range for displacement is slightly above the median estimate from the pro-
which the probability of exceeding the given displacement is 84 posed model, and in the other case it is at the median minus 1s . d.
to 16%, which is computed using Eq. 共7兲 by solving for d for estimate level. Thus, the estimated seismic performance from the
P共D ⬎ d兲 = 0.84 and 0.16, respectively. Estimates from the pro- proposed model is consistent with the observed performance of
posed method are compared with those from two state-of-practice these earth dams, which provides confidence in its use to evaluate
methods for seismic displacement prediction, i.e., Makdisi and the seismic performance of these critical earth structures.
Seed 共1978兲 and Bray et al. 共1998兲 in Columns 11 and 12. For the The tenth column in Table 1 provides the probability of ex-
sake of consistency, the same yield coefficient computed using ceeding the maximum observed displacement computed using the
the best estimate of the soil strength is used in all cases. Addi- model proposed in this paper 关i.e., using Eqs. 共7兲 and 共8兲 as de-
tional details of the application of these two simplified procedures scribed previously兴. The proposed model offers satisfactory pre-
are provided in Travasarou 共2003兲. dictions for most of the case histories. The predicted probability
There are seven cases in which the observed seismic displace- of exceedance of the maximum observed displacements is low
ment was noted as being “None” or “Minor.” In all of these cases 共⬍20% 兲 for 8 cases where the observed seismic displacement
except for Lopez Canyon C-A Landfill, the proposed method es- was low or negligible, which is consistent with it being unlikely
timates relatively high probabilities of “zero” displacement 共i.e., to have significant seismic displacements for these cases, or on
D 艋 1 cm兲, low estimates of nonzero displacements, and low the order of 30–70% for 6 cases, suggesting that these case
probabilities of having seismic displacements exceed a low histories represent median model estimates. Exceptions to this
threshold value. For Lopez Canyon C-A Landfill, the estimates of generally reliable estimate of the probability of exceeding the
40% chance of “zero” displacement and a seismic displacement maximum observed displacement are Chiquita Canyon D Landfill
range of “0”–5 cm is still consistent with the observed good per- and Lexington Dam. As discussed previously, the proposed model
formance of this landfill. Hence, the proposed method’s estimates 共as do other simplified models兲 estimates a low probability of
of seismic performance are consistent with the good seismic per- exceeding the maximum observed displacement for Chiquita Can-
formance observed of these earth/waste structures. yon D Landfill 共i.e., 10%兲, which indicates it is slightly unconser-
Good seismic performance was also observed for the La Villita vative for this case. For Lexington Dam, the proposed model
Dam in Mexico 共Elgamal et al. 1990兲. Its performance to three estimates a high probability of exceeding the maximum observed
earthquake events is well-documented with increasing levels of displacement 共i.e., 90%兲, which indicates that it is slightly con-
observed seismic displacement from Events S3, S4, and S5. The servative for this case. Overall, however, the proposed model is
proposed model develops estimates of seismic displacement with judged to provide a reasonable estimate of the observed seismic
the same trends. There is a high probability of zero displacements performance for these 16 case histories.
共95%兲 for Event S3 where the maximum observed displacement In judging the proposed method, it is also important to com-
was 1 cm, which is considered a negligible level of displacement. pare its robustness relative to the other prevailing simplified pro-
There is a relatively low probability of zero displacement 共25%兲 cedures. For the cases in which None or Minor displacement was
with a range of estimated seismic displacement from 0–10 cm, observed, each method provides consistent estimates of negligible
which encompasses the observed value of 4 cm for Event S5. The to low seismic displacements. When small displacements
proposed model’s estimates are intermediate for Event S4, which 共艋5 cm兲 were observed, the proposed method captures the ob-
is consistent with the dam’s intermediate level of observed seis- served performance well, whereas the other methods’ estimates of
mic performance for this event. Thus, the proposed method is seismic displacement were low as they typically estimated no
judged to offer a reliable means for evaluating the observed seis- displacement. For cases with moderate observed displacements
mic performance of La Villita Dam for these earthquake events. 共5 cm艋 D 艋 50 cm兲, the proposed procedure generally predicted

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007 / 389

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
performance within the range observed, whereas the other pre-
vailing methods were more variable and often unconservative.
Overall, the proposed method is not significantly inconsistent
with existing methods, but it can capture the observed perfor-
mance better than existing procedures. Moreover, it is superior to
the prevalent simplified seismic displacement methods, because it
Fig. 10. Maximum cross section of dam with downstream slope
characterizes the uncertainty involved in the seismic displacement
potential sliding mass evaluated in illustrative example 共based on
estimate and can be used in a probabilistic seismic hazard
dam cross section in Harder et al. 1998兲
assessment.

rock, is shown in Fig. 10. The dam has been placed in San Lean-
Illustrative Seismic Evaluation Examples dro, Calif. at a rupture-distance of 1.1 km from the Hayward
fault. A deterministic analysis is performed first to evaluate a deep
To illustrate the use of the proposed relationships for estimating slide through the dam, which is followed by a pseudoprobabilistic
seismically induced permanent displacements, the Chabot earth analysis wherein the spectral acceleration is computed from a
dam is reanalyzed, and the likely seismic performance of a probabilistic seismic hazard assessment 共PSHA兲 at two levels of
representative earth dam situated in a highly seismic region is hazard.
evaluated. The compacted soil’s shear modulus was computed using the
pressure-dependent Seed and Idriss 共1970兲 relationship with a
mean value of k2,max ⬃ 100, which resulted in an average shear
Reanalysis of the Performance of the Chabot Dam
wave velocity of 450 m / s for the full height of the dam. For the
during the 1906 San Francisco Earthquake
case of base sliding at the maximum height of this triangular-
Chabot Dam is a 41 m high dam constructed of sandy clay which shaped potential sliding mass, the best estimate of its initial fun-
is located about 32 km away from the San Andreas fault. The dam damental period is
performance during the 1906 San Francisco earthquake is reana-
2.6H 2.6共57兲 m
lyzed with the proposed procedure. The initial fundamental period Ts = = ⬇ 0.33 s 共9兲
of the dam was estimated based on Makdisi and Seed 共1978兲 to be V̄s 450 m/s
0.55 s. Additionally, they used the average results from consoli-
dated static undrained tests for a deep failure surface to estimate The yield coefficient for a deep failure surface was estimated to
ky = 0.14. In the absence of a strong motion recording at the site, be 0.14 from pseudostatic slope stability analyses performed with
the median spectral acceleration at 1.5Ts was calculated using the the total stress strength properties of c = 14 kPa and ␾ = 21° based
relationship proposed by Abrahamson and Silva 共1997兲 for a on undrained triaxial compression tests.
M w = 7.9 event as Sa共T = 1.5共0.55 s兲兲 = 0.26 g. The deterministic design earthquake scenario is represented by
First, the probability of zero displacements is computed the mean maximum magnitude for the South Hayward fault,
using Eq. 共3兲 with the values of ky = 0.14, Ts = 0.55 s, and which is M w = 6.9, at the closest rupture-distance of R = 1.1 km.
Sa共1.5 s兲 = 0.26 g as: P共D = 0兲 = 1 − ⌽共0.35兲 ⬇ 0.36. Hence, there is The best estimate of the spectral acceleration at the degraded
about a 35% probability of negligible displacements 共D 艋 1 cm兲 period of the dam, i.e., 1.5Ts = 1.5共0.33 s兲 = 0.5 s, can be com-
occurring. This is consistent with observations that found no sig- puted as the average of the median predictions from multiple
nificant deformations. There is still a finite probability 共⬃65% 兲 attenuation relationships. Using Abrahamson and Silva 共1997兲
of nonzero displacements 共i.e., D ⬎ 1 cm兲. The 16 and 84% ex- and Sadigh et al. 共1997兲 for the “rock” site condition for a strike-
ceedance levels of expected seismic displacements are calculated slip fault with M w = 6.9 and R = 1.1 km, Sa共0.5 s兲 = 1.09 and
using Eq. 共7兲 by computing the value of the seismic displacement 1.04 g, respectively. Thus, the design value of Sa at the degraded
共d兲 for which P共D ⬎ d兲 = 0.16 and 0.84, respectively 共e.g., by per- period of sliding mass is 1.07 g, its initial period is 0.33 s, and ky
forming GOAL SEEK in Excel兲. The respective values are 0 and is 0.14.
5 cm. These values are slightly different than the minus 1 and The probability of zero displacement occurring is
plus 1 s.d. values estimated using only Eq. 共5兲, because there is a computed using Eq. 共3兲 as P共D = 0兲 = 1 − ⌽共−1.76− 3.22 ln共0.14兲
finite probability of negligible displacements. This range of esti- − 0.484共0.33兲 ln共0.14兲 + 3.52 ln共1.07兲兲 = 0, so there is zero prob-
mated displacements 共i.e., 0 – 5 cm兲 represents the variability of ability of negligible displacements occurring for this event, or a
the seismic displacement given the values of ky, Ts, and Sa共1.5Ts兲 100% probability for displacements greater than 1 cm occurring.
and does not include the variability of these predictive variables Because the probability of negligible displacements is zero, the
in the calculations. As discussed in Travasarou et al. 共2004兲, the median estimate, and 16 and 84% exceedance values may be
variability in these parameters can be included in the calculations computed directly from Eq. 共5兲 using a standard deviation for the
to compute seismic displacement hazard. natural logarithm of the displacement of 0.66. Seismic displace-
ment is estimated from Eq. 共5兲 using the same design parameters
with M w = 6.9 as
Evaluation of the Seismic Performance of a
Representative Earth Dam ln共D兲 = − 1.10 − 2.83 ln共0.14兲 − 0.333共ln共0.14兲兲2

The anticipated performance of a representative earth dam in + 0.566 ln共0.14兲ln共1.07兲 + 3.04 ln共1.07兲
terms of seismically induced permanent displacements is evalu- − 0.244共ln共1.07兲兲2 + 1.50共0.33兲 + 0.278共6.9 − 7兲 ± ␧
ated herein to provide an example of a forward analysis. The
共10兲
geometry and properties of the dam are similar to those for a dam
described in Harder et al. 共1998兲. The maximum cross section of Thus, the estimated displacements are, respectively,
the 57 m high rolled earth-fill dam, which is founded on fractured D = exp共ln共D兲兲 = exp共3.77兲 ⬇ 40 cm, D = exp共ln共D兲 − ␴兲 = exp共3.77

390 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
− 0.66兲 ⬇ 20 cm, and D = exp共ln共D兲 + ␴兲 = exp共4.43兲 ⬇ 80 cm. motion records that have recently become available. The spectral
Because the standard deviation for the natural logarithm of dis- acceleration at a degraded period of the potential sliding mass
placement is 0.66, the 16% exceedance displacement value is 共Sa共1.5Ts兲兲 is the most efficient and sufficient single ground mo-
approximately half of the median estimate and the 84% exceed- tion intensity measure. Hence, the seismic displacement model
ance displacement value is approximately twice the median esti- depends significantly on this intensity measure, which optimally
mate. Hence, one could simply use Eq. 共5兲 to estimate the median characterizes the seismic loading. The system’s seismic resistance
seismic displacement and use a range from half this value to twice is best captured by its yield coefficient 共ky兲, but the dynamic
this value as the seismic displacement estimate range from ap- response characteristics of the potential sliding mass are also an
proximately 16 to 84% exceedance. The seismic displacement important influence, which can be captured by its initial funda-
due to deviatoric deformation is estimated to be between 20 and mental period 共Ts兲. This model attempts to capture the mecha-
80 cm for the design earthquake scenario. The direction of this nisms that are consistent with the Newmark method, i.e.,
displacement should be oriented parallel to the direction of slope deviatoric-induced displacement due to sliding on a distinct plane
movement, which will be largely horizontal. For the total crest and distributed deviatoric shearing within the slide mass.
displacement of the dam, a procedure such as Tokimatsu and Seed The proposed mixed random variable model separates the
共1987兲 would be required to estimate the vertical settlement due probability of zero displacement 共i.e., 艋1 cm兲 occurring from the
to cyclic volumetric compression of the compacted earth fill. distribution of nonzero displacement, so that very low values of
The same predictive equations can be used in a pseudoproba- calculated displacement that are not of engineering interest do not
bilistic approach where the properties of the dam are treated as bias the results. In this procedure, the probability of negligible
deterministic constants with values equal to their best estimates displacement occurring is first calculated using Eq. 共3兲. Then, the
and the spectral acceleration at the degraded period of the dam is amount of nonzero displacement is estimated from Eq. 共5兲. The
computed from a PSHA for specified levels of hazard. A project- 16 and 84% exceedance seismic displacement values may be
specific PSHA or the U.S. Geological Survey 共USGS兲 general computed using Eqs. 共7兲 and 共8兲 by solving for the displacement
PSHA 共http://eqhazmaps.usgs.gov/兲 may be used. The estimated value d that gives P共D ⬎ d兲 = 0.16 and 0.84, respectively. The 16–
seismic displacements may be viewed as being consistent with the 84% exceedance seismic displacement range can be estimated
level of hazard at which the ground motions have been estimated. approximately as half to twice the median estimate from Eq. 共5兲.
Two levels of hazard are considered for this case, which are In cases where the controlling magnitude cannot be discerned, Eq.
defined by a ground motion return period of 475 years 共i.e., 10% 共4兲 can be used to estimate seismic displacement. The first term of
probability of exceedance in 50 years兲 and of 975 years 共i.e., 5% Eqs. 共4兲 and 共5兲 changes for the special case of a nearly rigid
probability of exceedance in 50 years兲. The spectral acceleration Newmark sliding block 共i.e., Ts ⬃ 0 s兲 to obtain Eq. 共6兲 for
at the degraded fundamental period of the sliding mass is given example.
by the USGS at the hypothetical dam location of N37.733- The proposed seismic displacement model provides estimates
W122.117 as Sa共0.5 s兲 = 1.2 g and 1.6 g for the 474 and 975 year of seismic displacements that are generally consistent with docu-
return periods, respectively. mented cases of earth dam and solid-waste landfill performance.
Using Eq. 共3兲 with the same slope parameters as in the deter- It also provides assessments that are not inconsistent with other
ministic case with Sa = 1.2 g, the probability of zero displacement simplified methods, but does so with an improved characteriza-
is computed to be 0. The estimated displacement can be computed tion of the uncertainty involved in the estimate of seismic dis-
from Eq. 共5兲 using the modal magnitude from the deaggregation placement. The proposed model can be implemented rigorously
of the hazard, which is dominated by one fault for this case, as within a fully probabilistic framework for the evaluation of the
M w = 6.85. The median seismic displacement for the 475 year re- seismic displacement hazard, or it may be used in a pseudodeter-
turn period ground motion level is estimated to be 50 cm and the ministic analysis. In all cases, however, the estimated range of
16–84% range of displacements is 25– 100 cm 共i.e., about half seismic displacement should be considered merely an index of the
and double, respectively, of the median estimate兲. Similarly, for expected seismic performance of the earth/waste structure.
the 975 year return period ground motion, Eq. 共3兲 estimates a zero
probability of zero displacements. The model magnitude is still
6.85 at this hazard level and period, but with the higher level of Acknowledgments
spectral acceleration of 1.6 g, large seismic displacements are es-
timated using Eq. 共5兲 with a median estimate of 85 cm and Support for this work was provided by the Earthquake Engineer-
45– 165 cm at the 16 to 84% exceedance levels. ing Research Centers Program of the National Science Founda-
tion under Award No. EEC-2162000 through the Pacific
Earthquake Engineering Research Center 共PEER兲 under Award
Conclusions Nos. NC5216 and NC7236. Additional support was provided by
the David and Lucile Packard Foundation. Discussions with Pro-
A simplified semiempirical predictive model for estimating seis- fessor Armen Der Kiureghian of the University of California at
mic deviatoric-induced displacements has been developed. The Berkeley, Professor Ross Boulanger of the University of Califor-
Newmark-type model for evaluating the seismic performance of nia at Davis, and Jacob Chacko of Fugro-West Inc. were
an earth dam, natural slope, compacted earth fill, or municipal invaluable.
solid-waste landfill is an advancement, because it is based on the
results of nonlinear fully coupled stick-slip sliding block analyses
using a comprehensive database of 688 pairs of horizontal com- References
ponents of recorded ground motions from 41 earthquakes. The
primary source of uncertainty in assessing the likely performance Abrahamson, N. A., and Silva, W. J. 共1997兲. “Empirical response spectral
of an earth/waste system during an earthquake is the input ground attenuation relations for shallow crustal earthquakes.” Seismol. Res.
motion, so this model takes advantage of the wealth of strong Lett., 68共1兲, 94–127.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007 / 391

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright
Ambraseys, N. N., and Menu, J. M. 共1988兲. “Earthquake-induced ground Lin, J.-S., and Whitman, R. V. 共1983兲. “Decoupling approximation to the
displacements.” J. Earthquake Eng., 16, 985–1006. evaluation of earthquake-induced plastic slip in earth dams.” Earth-
Ambraseys, N. N., and Srbulov, M. 共1994兲. “Attenuation of earthquake- quake Eng. Struct. Dyn., 11, 667–678.
induced displacements.” J. Earthquake Eng., 23, 467–487. Lin, J.-S., and Whitman, R. 共1986兲. “Earthquake induced displacements
Ashford, S. A., and Sitar, N. 共2002兲. “Simplified method for evaluating of sliding blocks.” J. Geotech. Engrg., 112共1兲, 44–59.
seismic stability of steep slopes.” J. Geotech. Geoenviron. Eng., Makdisi, F., and Seed, H. 共1978兲. “Simplified procedure for estimating
128共2兲, 119–128. dam and embankment earthquake-induced deformations.” J. Geotech.
Augello, A. J., Matasovic, N., Bray, J. D., Kavazanjian, E., Jr., and Seed, Engrg. Div., 104共7兲, 849–867.
R. B. 共1995兲. “Evaluation of solid waste landfill performance during Newmark, N. M. 共1965兲. “Effects of earthquakes on dams and embank-
the Northridge Earthquake.” Earthquake Design and Performance of ments.” Geotechnique, 15共2兲, 139–160.
Solid Waste Landfills—ASCE Geotechnical Special Publication No. Rathje, E. M., and Bray, J. D. 共2000兲. “Nonlinear coupled seismic sliding
54, M. K. Yegian and W. D. L. Finn, eds., New York, 17–50. analysis of earth structures.” J. Geotech. Geoenviron. Eng., 126共11兲,
Blake, T. F., Hollingsworth, R. A., and Stewart, J. P., eds. 共2002兲. Rec- 1002–1014.
ommended procedures for implementation of DMG Special Publica- Rathje, E. M., and Bray, J. D. 共2001兲. “One- and two-dimensional seismic
tion 117 guidelines for analyzing and mitigating landslide hazards in analysis of solid-waste landfills,” Can. Geotech. J., 38共4兲, 850–862.
California, Southern California Earthquake Center, Los Angeles. Rodriguez-Marek, A., Bray, J. D., and Abrahamson, N. 共2001兲. “An em-
Bray, J. D., Augello, A. J., Leonards, G. A., Repetto, P. C., and Byrne, R. pirical geotechnical seismic site response procedure.” Earthquake
J. 共1995兲. “Seismic stability procedures for solid waste landfills.” J. Spectra, 17共1兲, 65–87.
Geotech. Engrg., 121共2兲, 139–151. Sadigh, K., Chang, C.-Y., Egan, J. A., Makdisi, F., and Youngs, R. R.
Bray, J. D., Rathje, E. M., Augello, A. J., and Merry, S. M. 共1998兲. 共1997兲. “Attenuation relationships for shallow crustal earthquakes
“Simplified seismic design procedures for geosynthetic-lined, solid based on California strong motion data.” Seismol. Res. Lett., 68共1兲,
waste landfills.” Geosynthet. Int., 5共1–2兲, 203–235. 180–189.
Cornell, C., and Luco, N. 共2001兲. “Ground motion intensity measures for Seed, H. B., and Idriss, I. M. 共1970兲. “Soil moduli and damping factors
structural performance assessment at near-fault sites.” Proc., U.S.- for dynamic response analyses.” Rep. No. EERC 70-10, Univ. of Cali-
Japan Joint Workshop and Third Grantees Meeting, U.S.-Japan Co- fornia, Berkeley.
operative Research on Urban EQ. Disaster Mitigation, Seattle. Seed, H. B., and Martin, G. R. 共1966兲. “The seismic coefficient in earth
Deirlein, G., Krawinkler, H., and Cornell, C. 共2003兲. “A framework for dam design.” J. Soil Mech. and Found. Div., 92共SM3兲, 25–58.
performance-based earthquake engineering.” Proc., 2003 Pacific Con- Stewart, J. P., Blake, T. F., and Hollingsworth, R. A. 共2003兲. “A screen
ference on Earthquake Engineering, Christchurch, New Zealand, analysis procedure for seismic slope stability.” Earthquake Spectra,
Paper No. 140. 19共3兲, 697–712.
Duncan, J. M., and Wright, S. G. 共2005兲. Soil strength and slope stability, Stewart, J. P., Bray, J. D., McMahon, D. J., Smith, P. M., and Kropp, A.
Wiley, Hoboken, N.J. L. 共2001兲. “Seismic performance of hillside fills.” J. Geotech. Geoen-
Elgamal, A.-W., Scott, R., Succarieh, M., and Yan, L. 共1990兲. “La Villita viron. Eng., 127共11兲, 905–919.
dam response during five earthquakes including permanent deforma- Tokimatsu, K., and Seed, H. B. 共1987兲. “Evaluation of settlements in
tions.” J. Geotech. Engrg., 116共10兲, 1443–1462. sands due to earthquake shaking.” J. Geotech. Engrg., 113共8兲, 861–
Elton, D. J., Shie, C.-F., and Hadj-Hamou, T. 共1991兲. “One- and two- 878.
dimensional analysis of earth dams.” Proc., 2nd Int. Conf. on Recent Travasarou, T. 共2003兲. “Optimal ground motion intensity measures for
Advancements in Geotechnical Earthquake Engineering and Soil Dy- probabilistic assessment of seismic slope displacements.” Ph.D. dis-
namics, St. Louis, 1043–1049. sertation, Univ. of California at Berkeley, Berkeley, Calif.
Ghahraman, V., and Yegian, M. K. 共1996兲. “Risk analysis for earthquake- Travasarou, T., and Bray, J. D. 共2003a兲. “Optimal ground motion inten-
sity measures for assessment of seismic slope displacements,” 2003
induced permanent deformation of earth dams.” Proc., 11th World
Pacific Conf. on Earthquake Engineering, Christchurch, New
Conference on Earthquake Engineering, Acapulco, Mexico, Paper
Zealand.
No. 688, Pergamon, Oxford.
Travasarou, T., and Bray, J. D. 共2003b兲. “Probabilistically-based esti-
Green, W. 共2003兲. Econometric analysis, 5th Ed., Prentice-Hall, Engle-
wood Cliffs, N.J. mates of seismic slope displacements.” Proc., 9th Int. Conf. on Appli-
Harder, L. F., Bray, J. D., Volpe, R. L., and Rodda, K. V. 共1998兲. “Per- cations of Statistics and Probability in Civil Engineering, San Fran-
formance of earth dams during the Loma Prieta earthquake.” Profes- cisco, Calif. Paper No. 318, Millpress Science, Rotterdam, The
sional paper no. 1552-D, The Loma Prieta, California Earthquake of Netherlands.
October 17, 1989 earth structures and engineering characterization Travasarou, T., Bray, J. D., and Der Kiureghian, A. D. 共2004兲. “A proba-
of ground motion, U.S. Geological Survey, Reston, Va. bilistic methodology for assessing seismic slope displacements.” 13th
Jibson, R. W. 共1993兲. “Predicting earthquake induced landslide displace- World Conf. on Earthquake Engineering, Vancouver, B.C., Canada,
ments using Newmark’s sliding block analysis.” Transportation Re- Paper No. 2326.
search Record. 1411, Transportation Research Board, Washington, Vrymoed, J. L., and Calzascia, E. R. 共1978兲. “Simplified determination of
D.C. dynamic stresses in earth dams.” Proc., Earthquake Engineering and
Jibson, R. W., Harp, E. L., and Michael, J. A. 共1998兲. “A method for Soil Dynamics Conf., ASCE, New York, 991–1006.
producing digital probabilistic seismic landslide hazard maps: An ex- Vucetic, M., and Dobry, R. 共1991兲. “Effect of soil plasticity on cyclic
ample from the Los Angeles, California area.” USGS Open-File Rep. response.” J. Geotech. Engrg., 117共1兲, 89–107.
No. 98-113, U.S. Geological Survey, Washington, D.C. Wartman, J., Bray, J. D., and Seed, R. B. 共2003兲. “Inclined plane studies
Kim, J., and Sitar, N. 共2003兲. “Importance of spatial and temporal vari- of the Newmark sliding block procedure.” J. Geotech. Geoenviron.
ability in the analysis of seismically-induced slope deformation.” Eng., 129共8兲, 673–684.
Proc., 9th Int. Conf. on Applications of Statistics and Probability in Yegian, M., Marciano, E., and Ghahraman, V. 共1991a兲. “Earthquake-
Civil Engineering, Millpress Science, San Francisco. induced permanent deformations: Probabilistic approach.” J. Geotech.
Kramer, S. L., and Smith, M. W. 共1997兲. “Modified Newmark model for Engrg., 117共1兲, 35–30.
seismic displacements of compliant slopes.” J. Geotech. Geoenviron. Yegian, M., Marciano, E., and Ghahraman, V. 共1991b兲. “Seismic risk
Eng., 123共7兲, 635–644. analysis for earth dams.” J. Geotech. Engrg., 117共1兲, 18–34.

392 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / APRIL 2007

Downloaded 03 Apr 2009 to 128.193.15.52. Redistribution subject to ASCE license or copyright; see http://pubs.asce.org/copyright

You might also like