You are on page 1of 16

Computers and Geotechnics 55 (2014) 195–210

Contents lists available at ScienceDirect

Computers and Geotechnics


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

Insight into the seismic response of earth dams with an emphasis


on seismic coefficient estimation
Konstantinos I. Andrianopoulos a,⇑, Achilleas G. Papadimitriou b, George D. Bouckovalas a,
Dimitrios K. Karamitros a
a
Geotechnical Department, School of Civil Engineering, National Technical University of Athens, Greece
b
Department of Civil Engineering, University of Thessaly, Greece

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

Article history: This paper presents a numerical investigation of the seismic response of earth dams by employing results
Received 17 December 2012 from 110 nonlinear two-dimensional (2D) dynamic analyses of four different cross-sections with heights
Received in revised form 6 September 2013 ranging from 20 to 120 m. The analyses were of a parametric nature, considering the effects of seismic
Accepted 9 September 2013
excitation characteristics (intensity and frequency content), foundation soil stiffness, and the existence
Available online 5 October 2013
of typical stabilising berms and/or an impounded reservoir. The results of these studies indicated that
the predominant period of a dam’s vibration was strongly affected by its height and the input motion
Keywords:
characteristics. The results also indicated that the peak acceleration at the dam’s crest was governed
Earth dams
Earthquake
by its height, the input motion characteristics, and the stiffness of the foundation soil, but not by the
Embankments other parameters. These same analyses yielded results on pseudo-static seismic coefficients for a total
Non-linear soil response of 1084 potential sliding masses within the analysed cross-sections, demonstrating that the seismic coef-
Numerical analysis ficients decreased as the sliding mass became deeper and bulkier, increased if the mass was located
Pseudo-static analysis upstream rather than downstream, and were strongly affected by the seismic excitation characteristics
Seismic coefficient and stiffness of the foundation soil. Moreover, these results allowed for a thorough evaluation of existing
methodologies for seismic coefficient estimation, quantifying their accuracy and depicting their limita-
tions. This evaluation process also illustrated the fact that there is currently no methodology accounting
for all significant problem parameters.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction dam during the 2011 Tohoku earthquake. Thus, questions have
been raised regarding the seismic safety of existing dams that have
Reports of severe seismic earth dam failures have been limited not been designed to withstand earthquakes (small and/or old
worldwide and related mainly to liquefaction (e.g., the Lower San dams) or have been designed using methods of analysis that are
Fernando Dam during the 1971 San Fernando earthquake, numer- now considered outdated [8]. New, taller dams have also been con-
ous tailings dams such as the Mochikoshi dams during the 1978 structed in unfavourable foundation conditions, creating an in-
Izu–Ohshim–Kinkai earthquake [1], and a number of Indian dams creased need for more accurate design methods.
such as the Chang and Shivlakha dams during the 2001 Bhuj earth- In general, the assessment of the seismic stability of new or
quake [2]). In contrast, as reported by numerous reconnaissance existing dams can be performed via (a) pseudo-static analyses
reports and studies (e.g. [3–7]), there have been many more cases [9], (b) displacement-based (Newmark or sliding block) methods
in which some sort of cracking or slope sliding has occurred during [10–12], and (c) dynamic stress-deformation numerical analysis
strong shaking, which is not necessarily related to liquefaction [13]. Although such robust numerical analyses as method (c) are
(e.g., the Austrian dam during the 1989 Loma Prieta earthquake). now quite common, methods (a) and (b) still form the basis of
Moreover, earthquake-related damages to earth dams continue to engineering practice in the seismic design of earth dams around
occur worldwide and unfortunately still include total collapses, the world, at least in the design preliminary stages of new dams
such as the complete failure of the 18.5 m Fujinuma embankment or the safety assessment of existing dams. Compared to (b), pseu-
do-static analyses have the benefit of accumulated experience and
user friendliness because they were first employed in the 1950s
⇑ Corresponding author. Address: Mourgkanas 6, Maroussi, 15126, Athens, and require merely the estimation of a pseudo-static factor of
Greece. Tel.: +30 2108069604. safety, FSd, against seismic ‘‘failure’’ of the geostructure’s slopes.
E-mail addresses: kandrian@tee.gr (K.I. Andrianopoulos), apapad@civ.uth.gr
An example of such a pseudo-static problem is illustrated in
(A.G. Papadimitriou), gbouck@central.ntua.gr (G.D. Bouckovalas), dimkaram@
gmail.com (D.K. Karamitros). Fig. 1, which also depicts significant problem parameters, such as

0266-352X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compgeo.2013.09.005
196 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

sliding mass’s permanent downslope displacement [14,16]. Such


analyses are always demanding in terms of software, expertise,
and cost and thus defeat the purpose of selecting method (a) over
(c). To avoid these downsides, researchers and practitioners around
the world have devised various simplified methods for estimating
appropriate seismic coefficient values for use in pseudo-static
analyses. Still, it is unclear how well such simplified methods agree
with measured responses and the results of robust dynamic anal-
yses of earth dams’ seismic responses.
Thus, this study attempted to clarify the compatibility between
existing simplified methods for estimating seismic coefficients
Fig. 1. Definition of the critical geometric and geotechnical parameters for the (outlined in Section 2) and robust dynamic numerical analyses
seismic slope stability of earth dams and tall embankments. (performed using the methodology described in Section 3). These
analyses provided insight into the seismic response of zoned earth
dams and homogeneous tall embankments and ascertained the rel-
the peak values of the seismic acceleration at the crest, PGAcrest, at ative significance of various problem parameters. As fully detailed
the outcropping (bed)rock, PGArock, and in the ‘‘free field’’ of the in Section 4, such parameters included excitation characteristics
foundation soil, PGA. The most critical point of the approach is (intensity and frequency content), dam geometry (height and exis-
the value of the horizontal inertial force, Fh, applied at the sliding tence of stabilising berms), foundation soil conditions, and the
mass’s centre of gravity. This force Fh is equal to the weight of dam’s operation phase (‘‘end of construction’’ and ‘‘steady-state
the sliding mass, W, multiplied by a dimensionless seismic coeffi- seepage’’ conditions). Note that in addition to the excitation charac-
cient, kh. Therefore, the selection of an appropriate value for kh is teristics, the other parameters under study are crucial to the static
crucial for the estimation of FSd and of paramount importance to stability of earth dams, but their significance in terms of seismic
a safe, rational design or safety assessment. loading is unknown. Whereas Section 4 considers the seismic re-
The value of kh should reflect the vibration of the sliding mass sponse of the whole dam, Section 5 focuses on seismic coefficients
during the design earthquake. Given that the sliding mass is gener- and pinpoints which of the foregoing parameters are important to
ally not rigid, different locations within this mass do not vibrate in designating their values. Finally, Section 6 presents a critical eval-
phase (especially in tall dams where the predominant wave length, uation of the methodologies outlined in Section 2, providing a
kd, of the seismic shear waves within the dam body is comparable quantification of their accuracy and limitations based on the
to their height, H [14]), and their vibrations are not of equal inten- numerical results.
sity (e.g., vibrations are less intense deep within the dam body than As clarified above, the paper deals with the seismic response of
on its surface). Therefore, the value of kh should be related to the ‘‘zoned earth dams and homogeneous tall embankments’’. Yet, for
resultant acceleration time history of the sliding mass, which is re- reasons of brevity, in the remainder of the paper the studied geo-
lated to the resultant (horizontal) force time history along the structures will be referenced as ‘‘dams and embankments’’.
shear band delineating the sliding mass within the dam body. This
resultant acceleration time history is generally expected to be a
2. Methodologies for seismic coefficient estimation
function of the characteristics of the dam and excitation and the
geometry of the sliding mass [15]. Yet, it can also be affected if slip-
This section begins with an overview of proposals from the
page has been initiated along the shear band delineating the slid-
literature for ‘‘effective’’ seismic coefficients, khE, for use in pseu-
ing mass within the dam body [16]. In any case, the peak value
do-static slope stability analyses of dams and embankments (Sec-
of the resultant acceleration time history is observed only momen-
tion 2.1). It then provides estimators from the literature of peak
tarily, and therefore, the design of earth dams using the respective
values of the seismic coefficient, khmax, for use in displacement-
peak value of the seismic coefficient khmax, along with a require-
based (Newmark-type) methods (Section 2.2).
ment for a pseudo-static factor of safety against seismic ‘‘failure’’
FSd P 1.0, leads to an overly conservative approach. Hence, com-
mon practice dictates the use of an ‘‘effective’’ value of the seismic 2.1. Estimating ‘‘effective’’ seismic coefficients
coefficient, khE (a percentile of khmax), combined with the existing
requirement of FSd P 1.0, at the expense of generally ‘‘small’’ (but According to Kramer [14], it was Terzaghi [9] who first depicted
still unknown) permanent downslope displacements. Based on values of khE = 0.1, 0.2, and 0.5 for ‘‘severe’’, ‘‘violent, destructive’’,
Papadimitriou et al. [17], literature values for the khE/khmax ratio and ‘‘catastrophic’’ earthquakes, respectively. Until the mid-1970s,
have ranged from 0.5 to 0.8, with the most commonly used value the depiction of khE had been based on local experience and typi-
equal to 0.67 (e.g., in the British Standards for pseudo-static anal- cally led to values between 0.10 and 0.15, with the assumed value
yses of the slopes of dams or embankments [18]). Alternatively, increasing as a function of the design earthquake magnitude, M, or
there have been recent efforts to relate the appropriate selection the significance of the civil engineering work, without exceeding a
of an ‘‘effective’’ seismic coefficient, khE (for use in pseudo-static value of 0.2 (the old design guidelines of the US Army Corps of
analyses), directly to the allowable downslope displacement [18– Engineers). However, Kramer [14] has increased the range of khE
22], thus introducing performance-based design principles in the up to 0.25 in both the USA and Japan for more recent practice.
well-established pseudo-static analysis method. The older values (up to 0.15) have been corroborated by Seed
Given the complexity of the problem, an accurate estimation of [24], who has proposed values of 0.1 for M = 6.5 and 0.15 for
the resultant acceleration time history of a (generally) flexible slid- M = 8.5, together with a requirement of FSd = 1.15 to ensure that
ing mass has required robust dynamic numerical analyses. Two displacements will not exceed 1 m (a value considered allowable
types of numerical procedures have been used for this purpose: for tall dams).
‘‘decoupled’’ analyses, in which the dynamic response of the exam- Seed’s proposals have more recently been validated as being
ined dam (and sliding mass) is calculated separately from possible quite reasonable, even for high PGA values [22]. However, the
slippage [15,23], and ‘‘coupled’’ analyses, in which the dynamic re- influence of the PGA level on the value of khE, which has been intu-
sponse is considered simultaneously with the accumulation of the itively acknowledged as early as Terzaghi [9], was incorporated
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 197

explicitly in literature proposals in the mid-1980s. In particular, The use of Makdisi and Seed’s correlation [15] and that of Mar-
based on a technical report from USCOLD [25], the typical practice cusson [32] (in Section 2.1) has created a practical need for esti-
of the time in the USA was to use khE values ranging from mating PGAcrest, a value that is generally not equal to the PGA or
0.25(PGA/g) to 1 PGA/g, with the largest values accounting the PGArock values that may be known from a seismic hazard study
elastic amplification of motion within the body of the dam. Accord- of the site. In general, the accurate estimation of PGAcrest requires
ing to Pyke [26], khE has ranged from 0.2(PGA/g) to 0.6(PGA/g), but the execution of non-linear numerical analyses, such as those per-
based on earthquake magnitude values M ranging from 6.5 to 8.5, formed for this study (Section 3). Alternatively, an estimate may be
respectively. Note that although there are different measures of obtained by using inelastic response spectra for the free-field sur-
earthquake magnitude, the values of M mentioned above must face of a dam’s foundation soil, which should be available from the
be considered equal to the most ‘‘accurate’’ moment magnitude aforementioned seismic hazard study. In doing so, the value of
Mw. Nevertheless, after Idriss [27], one may consider the local mag- PGAcrest can be estimated by accounting for the dam’s first two to
nitude ML as accurate (ML ffi Mw) for ML 6 6.25 and the surface three modes of vibration [15].
wave magnitude MS ffi Mw as accurate for 6.25 < MS 6 7.5, but not Although less sound from a theoretical perspective, if khmax is
for higher-magnitude events. related to the PGA itself, then the foregoing practical problem no
Similarly, the respective British Standards [18] have proposed longer exists. This scenario has been studied by Zania et al. [21],
the use of khE = 0.67(PGA/g). This proposal essentially implies that who performed a ‘‘decoupled’’ procedure for estimating downslope
any sliding mass within the dam is rigid and therefore the peak va- displacements of embankments. In particular, they correlated the
lue of its resultant acceleration time history, denoted by khmaxg, is ratio of khmax/(PGA/g) with the tuning period ratio of the embank-
equal to the PGA, or equivalently that khmax = PGA/g. A similar cor- ment’s eigenperiod over the predominant period of excitation. This
relation between the ‘‘effective’’ seismic coefficient and the PGA has correlation has enabled the proposal of a ‘‘seismic coefficient spec-
also been proposed by seismic codes (e.g., EC8 and the Greek code trum’’ for zero-slope displacements (the term ‘‘spectrum’’ implying
EAK) for the pseudo-static analysis of slopes, but not dams or a maximum plateau value for khmax/(PGA/g) at (tuning) period ra-
embankments. For example, EC8 proposes the use of khE = 0.5(- tios of approximately one due to resonance).
PGA/g), which is estimated using intensity maps for rock outcrop- Much more recently, Rathje and Antonakos [33] performed
ping locations in National Annexes and the soil factor according to parametric ‘‘coupled’’ analyses for flexible sliding masses and
ground category, thus introducing one-dimensional (1D) site established a predictive equation for the khmax/(PGA/g) ratio. This
amplification effects for each value. In the Greek code, the intensity equation indicates that the ratio of khmax/(PGA/g) decreases with
maps provide values for the ‘‘effective’’ ground acceleration (=0.8 an increasing PGA input, and its value is governed by the (Tmass/
PGA), which essentially translates to khE = 0.4(PGA/g). In any case, Tm) period ratio. In particular, the period Tmass is the fundamental
Salgado [28] has criticised the use of a unique percentile value period of the sliding mass, which is assumed to vibrate approxi-
(0.5 above) in estimating khE in EC8 because this value should be mately as a 1D horizontal soil layer over bedrock with a thickness
a function of earthquake magnitude, M, and PGA level, which vary equal to the maximum thickness of the sliding mass. Concurrently,
according to country and region, among other code-related issues. Tm is the mean period of the seismic excitation, which is similar but
Moreover, EC8 accounts for the expected amplification of seismic not equal to the predominant period, Te (see [16] for its definition).
motion near topographic irregularities (e.g., near slopes) due to In particular, this predictive equation indicates that as the sliding
the scattering and diffraction of incident seismic waves, a phenom- mass becomes more flexible (i.e., it becomes thicker), the value
enon commonly referred to as ‘‘topography effects’’ [30]. In particu- of khmax/(PGA/g) decreases significantly, particularly for Tmass/
lar, it has been proposed that an estimate of khE increases linearly Tm > 0.20.3.
from its minimum foregoing value when z = H to its maximum va-
lue, which is 20% higher for very shallow failure surfaces (z ? 0)
due to topography effects that are applied if the slope has a height 3. Description of numerical analyses
H P 30 m and its inclination is between 15° and 30° (higher angles
appear in natural slopes but rarely in earth dams). Thus, EC8 is con- 3.1. Overview
sistent with explicit recommendations for increasing the PGA by
25% for moderately steep slopes [29]. This increase could be larger To shed light on the seismic response of geostructures, such as
for steeper slopes (e.g., 40% for slope angles >30°, according to EC8) dams and embankments, 110 non-linear 2D dynamic analyses
or even moderately sloped dams and embankments if their crests were performed using the methodology described herein. Four dif-
are relatively narrow [30,31]. ferent cross-sections were considered to investigate the effect of
According to Marcusson [32], the slope stability of dams should the geostructure’s height on its dynamic response. In particular,
be performed using khE values related to PGAcrest (Fig. 1), a peak va- one case describing a tall uniform embankment of maximum
lue of acceleration that accounts for dam vibration, which PGA height H = 20 m was considered along with three cases of zoned
does not. earth dams:

2.2. Estimating peak seismic coefficients  a small dam of maximum height H = 40 m,


 a medium dam of maximum height H = 80 m, and
In their seminal work, Makdisi and Seed [15] were the first to  a tall dam of maximum height H = 120 m.
correlate the value of the peak seismic coefficient khmax with both
the value of (PGAcrest/g) and a decreasing function of the ratio (z/H), The dams’ slope inclinations were chosen to vary between 1:2
representing the maximum depth z of the failure surface (mea- and 1:2.5 (vertical:horizontal) to achieve an adequate factor of
sured from the crest) over the dam’s height H (Fig. 1). In particular, slope safety under static conditions. Excluding the case of the tall
by employing ‘‘decoupled’’ analyses, they have reported values of embankment (H = 20 m), that was considered uniform, the
khmax 6 PGAcrest/g, with the equal sign holding for z/H = 0. The ben- remaining cross-sections included a clay core with slope inclina-
efit arising from their proposal is that khmax accounts not only for tions of the core–shell interfaces ranging between 4:1 and 5:1
dam vibration through PGAcrest but also for the sliding mass’s geo- (vertical:horizontal).
metric characteristics through z/H, both of which are not depicted The effect of typical (upstream and downstream) stability
with the PGA. berms on the overall response was also investigated. The heights
198 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

Table 1 the following generalised Ramberg and Osgood [42] relation type
Type and plasticity index PI(%) for each geomaterial comprising the various for monotonic and cyclic loading paths:
earthdams, and related model constants for the employed non-linear hysteretic
model (po is the mean effective pressure at equilibrium (in kPa), used to vary the Gmax
(small strain) shear wave velocity V (in m/s) with depth). Gt ¼ ð1Þ
A
Geomaterial PI (%) V (m/s) a1 c1 (%) m 8   9
Uniform rockfill (H = 20 m) 0–7.5 115 (po)0.27 0.64 0.016 0.33 >
< 1 þ 2 a1  1 2Xg ; for cyclic loading >
=
1 1
Non-cohesive shells (H = 40, 0–7.5 86 (po)0.31 0.64 0.016 0.33 A¼   P1
80, 120 m) >
: 1 þ 2 a1  1 gX ; for monotonic loading >
;
1 1
Clay core (H = 40, 80, 120 m) 7.5–15 60 (po)0.29 0.64 0.027 0.33

where Gmax is the maximum (small-strain) shear modulus and A is a


dimensionless scalar that decreases during loading that introduces
and widths of typical stability berms were set approximately equal
Gt degradation and thus hysteretic damping, denoted as n. In partic-
to H/3 and 2H/3, respectively (where H is the ever-current dam
ular, the Gmax of Eq. (1) is either quantified as a function of the
height). Apart from the tall dam (H = 120 m), the remaining
(small strain) shear wave velocity V of the soil (and its mass density
cross-sections were analysed both with and without typical
q), using Gmax = qV2, or estimated as a function of the mean effec-
stability berms to depict the berms’ relative effect on dynamic
tive pressure p and void ratio e of the soil:
responses.
 rffiffiffiffiffi
A multi-stage construction of each cross-section was initially Bpa p
Gmax ¼ ð3Þ
simulated to establish an admissible stress field under static condi- 0:3 þ 0:7e2 pa
tions at the end of construction. Based on the computed mean
effective stresses, the initial shear-wave velocity profile of each where B is a model constant and pa is the atmospheric pressure. In
cross-section was estimated according to Table 1. For H = 20 m, turn, scalar A of Eq. (2) is a function of the ‘‘distance’’ X in the gen-
the empirical correlation of Kokusho and Eshashi [34] for dense eralised stress-ratio space of the ever-current deviatoric stress ratio
(void ratio e = 0.4) gravels was used. For the other cross-sections, tensor r (=s/p, where s is the deviatoric stress tensor and p is the
an initial dynamic shear modulus, Gmax, was computed according mean effective stress) from its value rref at the last reference state.
to Hardin and Drnevich [35,36] for both clay core and core–shells. The latter is estimated by
Thus, the H = 20 m section could be considered a tall, dense rockfill qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
embankment, whereas the remaining cross-sections simulated the X¼ 1=2ðr  rref Þ : ðr  rref Þ ð4Þ
dynamic response of zoned earth dams of various heights.
where ‘:’ denotes the double inner product of the two tensors. The
The effect of foundation conditions was examined by assuming
reference state for monotonic loading is the equilibrium state,
various shear wave velocities for the foundation layer, ranging
whereas the reference state is updated at each shear reversal state
from Vb = 250 m/s to 1500 m/s (the minimum value was used only
for cyclic loading. In addition, scalar g1, which divides ‘‘distance’’ X
for H = 20 m). Finally, to investigate the effect of reservoir
in Eq. (2), provides a correlation to the (updating) reference state:
impoundment on seismic response, the same dam cross-sections !
were analysed for both the ‘‘end of construction’’ and ‘‘steady-state Gref
max
seepage’’ conditions. In the former case, the water table was as- g1 ¼ a1 c1 ð5Þ
pref
sumed to lie at the dam base, whereas in the latter case, the water
table followed the surface of the impounded reservoir until the up- where pref and Gref max correspond to the values of p and Gmax (using
stream crest and then decreased linearly within the clay core, Eq. (1) for A = 1) at the last reference state (where rref is also up-
reaching the dam base at the core–downstream shell interface. dated), respectively, and c1 and a1 are model constants. For simplic-
Furthermore, in the analyses conducted for ‘‘steady-state seepage’’ ity, shear reversal is assumed to be triggered when the X value
conditions, the reservoir was modelled as an incompressible mate- reduces from its previous step. Then, pref, Gref
max , and r
ref
are updated,
rial with a density of q = 1 Mg/m3. The elastic stress–strain relation rendering ‘‘distance’’ X to be zero, as it was initially at the equilib-
was used for its modelling, with Poisson’s ratio of approximately rium state, and A = 1 for this time step. Given this updating upon
m = 0.5, a bulk modulus of 2 GPa, and a shear modulus of approxi- shear reversal and the form of Eqs. (1)-(5), the predicted soil re-
mately zero. This calibration corresponded to a compressional sponse is non-linear hysteretic. The non-linear model has four
wave velocity equal to 1500 m/s and a shear wave velocity of zero dimensionless constants: the elastic Poisson’s ratio m (with 0.33
within the reservoir. being a commonly assumed value), constant B (or velocity V) that
The parametric analyses also studied the effects of seismic mo- scales the Gmax value, positive scalar a1 (61) that introduces non-
tion characteristics, as described in Section 3.4. linearity (a1 = 1 for A = 1 and a linear response), and c1, a reference
cyclic shear strain level. Fig. 2 provides sample simulation runs for
3.2. Constitutive model for geomaterials calibrating constants a1 and c1 to best fit the experimental curves of
the normalised secant shear modulus G/Gmax degradation and hys-
The mechanical response under dynamic loading of the various teretic damping n increase curves with a cyclic strain amplitude c,
geomaterials comprising the dams was simulated using the non- from Vucetic and Dobry [43].
linear hysteretic constitutive model described below, which has For the purposes of this paper, the (small strain) shear wave
been implemented as a user-defined model routine in FLAC [37] velocities, V (related to the Gmax and model constant B values),
by the authors. Given the time-domain analysis performed, the for the dam shells and clay core were considered functions of the
equations were written in an incremental format, and the soil re- initial mean effective stress po, as this stress was estimated using
sponse was simulated by employing a step-by-step integration a staged construction analysis of each dam under study (Table 1).
based on the ever-current tangential values of the material moduli. The constants a1 and c1 were chosen based on the calibration of
The soil was modelled as a non-linear hysteretic material using Fig. 2, depending on the plasticity index PI(%): 0–7.5% for the shells,
constantly updated values of the tangential bulk Kt and shear Gt modelled as a PI = 0% material, and 7.5–15% for the clay core, mod-
moduli. Following isotropic elasticity, Kt and Gt are interrelated elled as a PI = 15% material (Table 1).
via a constant elastic Poisson’s ratio m (a model constant). Then, Fig. 2 illustrates that the foregoing non-linear hysteretic model
based on [38–41], the non-linear hysteretic form of Gt is given by used in this study yields almost no damping at very small cyclic
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 199

quency numerical perturbations, stiffness Rayleigh damping was


also used in the numerical analyses, calibrated to 1% damping for
the predominant frequency of the excitation.

3.3. Mesh discretisation and boundary conditions

When employing FLAC, the continuum was divided into a finite-


difference mesh composed of quadrilateral elements (‘‘zones’’ in
FLAC terminology). Proper care was taken in creating a dense
mesh, with a maximum zone dimension equal to 1/10 of the pre-
dominant shear wavelength in each part of the medium, to avoid
the numerical distortion of the propagating waves’ frequency con-
tent. The mesh extended laterally and to the bottom sufficiently far
from the dam body (at least two times the dam height H) to min-
imise the effects of artificial reflections. Transmitting boundaries
[44] were applied at the base of the mesh for the same reason,
whereas boundaries simulating the free-field response [45] were
applied at its lateral sides. The seismic excitation, represented by
vertically propagating in-plane SV waves (i.e. shear waves with
polarisation in the vertical plane), was applied as a time history
of shear stress along the mesh’s horizontal base to avoid artificial
wave reflections at the bottom boundary that are unavoidable
when time histories of displacement, velocity, or acceleration are
applied.

3.4. Seismic excitations


Fig. 2. Calibration of model constants a1 and c1 in the employed model to fit the
experimental curves of Vucetic and Dobry [43].
Recorded ground motions from strong earthquakes in Greece
were used as seismic excitations in all 110 analyses:

shear strains, which is not in accordance with experimental data. (a) Splb time history recorded during the 1999 Athens earth-
Thus, to incorporate non-zero damping in the discretised soil med- quake (M = 5.9, longitudinal component, 12 km epicentral
ium irrespective of cyclic shear strain level and avoid high-fre- distance).

Fig. 3. Acceleration time histories a(t) normalised by their peak acceleration value, amax, that were used as input excitations.
200 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

the investigated parameters (input motion characteristics, founda-


tion soil stiffness, existence of berms, and/or impounded reser-
voirs, dam height) on fundamental indices of the dam’s response,
such as the non-linear fundamental dam vibration period, To (Sec-
tion 4.2), and peak acceleration at the crest, PGAcrest (Section 4.3).
Section 4.1 presents results pertaining to the spatial variation of
the peak horizontal acceleration along the dam’s surface and with-
in its body, providing a clearer picture of how seismic waves atten-
uate in a dam’s vicinity.
Fig. 5a presents typical results corresponding to the response of
the short dam (H = 40 m) founded on firm ground (Vb = 1000 m/s)
at the ‘‘end of construction’’ when subjected to the Splb time history
scaled at PGA = 0.05 g. In particular, Fig. 5a provides information
on how the peak horizontal acceleration varies spatially with the
horizontal distance at the ground surface along the dam (surface
Fig. 4. Elastic response spectra (5% critical damping) normalised by their peak
level) but also at its base (foundation level). As shown, the peak
ground acceleration value, amax, compared to the elastic design spectra for ground
categories A and B of EC8.
horizontal acceleration exhibited intense spatial variability at the
surface level along the dam’s inclined shells, as opposed to the
more uniform response along the horizontal ground away from
(b) Aigio time history recorded during the 1995 Aigio earth- the dam. By directly comparing the results for the surface and
quake (M = 6.2, east–west component, 18 km epicentral foundation levels, it became obvious that significant amplification
distance). of seismic motion existed within the dam body (a factor of almost
(c) Argostoli time history recorded during the 1983 Cephallonia three, on average, between the two levels), which was a clear man-
earthquake (M = 7.0, east–west component, 34 km epicen- ifestation of topography effects [31]. Fig. 5a also illustrates that due
tral distance). to the presence of a heavy geostructure, the peak horizontal accel-
(d) Kozani time history recorded during the 1995 Kozani–Greve- eration was clearly de-amplified at the foundation level by a factor
na earthquake (M = 6.6, east–west component, 16 km epi- of approximately 1.5 compared to the peak horizontal acceleration
central distance). at the free-field (away from the dam) of PGA = 0.05 g. Hence, the
average amplification of peak horizontal acceleration at the dam’s
Fig. 3 presents the acceleration time histories employed, nor- surface level was equal to approximately two times that of the
malised with respect to their peak acceleration values for compar- PGA. However, the amplification at the crest was significantly
ison purposes. These recordings are characterised by durations of higher (approximately 2.6 times that of the PGA), with seismic mo-
strong motion that vary from 1 to 2 s (Aigio) up to approximately tion that was far from uniform despite its narrow width. This result
10 s (Argostoli). The Aigio time history includes a directivity pulse was similar to what analyses of 2D uniform hills have shown [31].
at approximately 3 s, whereas the other recordings may be consid- Fig. 5b compares these results in terms of the peak horizontal
ered relatively far field motions. In addition, Fig. 4 presents (and acceleration’s spatial variability at the surface and foundation lev-
compares) their elastic response spectra (for 5% critical damping). els of a dam, yielding similar results for conditions identical to
The selected group of seismic recordings covers a wide range of those described before but with typical stability berms at the toes
predominant periods, Te = 0.14–0.5 s, which is the expected range of both the dam’s sides. Fig. 5b illustrates that the existence of
of predominant periods of bedrock motion in EC8 (at least for berms did not essentially affect the peak horizontal accelerations
Southern Europe, where M > 5.5). Fig. 4 also illustrates that the se- at the crest, foundation level, or ground surface away from the
lected recordings include relatively monochromatic excitations dam. In contrast, the existence of berms intensified the spatial var-
(Aigio, Splb), as well as an excitation with a wide range of signifi- iation in the peak horizontal acceleration along the dam surface.
cant periods (Argostoli). Furthermore, the selected group of excita- This intensification could again be attributed to the topographic
tions covers almost the entire range of significant periods for the amplification of the seismic ground motion observed in the vicin-
elastic design spectra of ground categories A and B of EC8, which ities of the two berms. Fig. 5b also illustrates that these local
correspond to relatively stiff ground conditions that are generally amplification phenomena could be quite significant but did not af-
used as foundation layers for dams and embankments. fect the seismic responses at the centre of the dam or its crest.
The seismic excitations applied at the base of the mesh were Fig. 5c compares the results of the combination of dam-founda-
appropriately scaled in each analysis to attain the desired peak tion-excitation from Fig. 5a (at the ‘‘end of construction’’ condition)
ground acceleration (PGA) at the foundation layer’s free-field sur- with those for the same dam after impoundment (at the ‘‘steady-
face. The PGA values in the performed analyses ranged from state seepage’’ condition). The peak accelerations are amplified
0.05 g to 0.5 g, thus enabling the study of the entire spectrum of along the upstream shell and remain unaltered or slightly reduced
seismic dam responses, from the linear range for low-intensity mo- along the crest and downstream shell in the latter condition. These
tions up to the highly non-linear range for high-intensity motions. differentiations were observed at surficial parts of the dam body
because the peak accelerations were not affected by the existence
of the reservoir at the dam’s base. Overall, the impoundment of the
4. Insights into the seismic responses of earth dams reservoir made the spatial variation in the peak accelerations at
surface level more non-uniform than in previously tested cases
4.1. Spatial variation of the peak ground accelerations with or without berms.
This amplification of peak accelerations in the upstream shell,
Section 4 presents typical results from the numerical analyses compared to those in the downstream shell under ‘‘steady-state
with the goal of providing insights into the seismic responses of seepage’’ conditions, were attributed to the contrast between shear
dams as geostructures, with no reference being made to seismic wave velocities, V (proportional to the effective stresses in Table 1),
coefficients of any potential sliding mass within the dam’s body. that existed between the non-saturated downstream shell and
In particular, this section presents results related to the effects of submerged dam body (the upstream shell and part of the core).
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 201

Fig. 5. Spatial variation in the maximum horizontal acceleration along the dam’s surface and base (H = 40 m, Splb time history, PGA = 0.05 g, Vb = 1000 m/s): (a) reference case
(without berms and at the end of construction), (b) effect of stabilising berms at the end of construction, and (c) effect of reservoir impoundment for the dam without berms.

An example of this contrast in shear wave velocity values is pre- a first approximation. Thus, comparing this period with the range
sented in Fig. 6, which depicts the shear wave velocity contours of predominant periods, Te, expected at the dam’s location roughly
in the portion of the mesh close to the dam with H = 40 m. Observe determined whether resonance would occur and thus whether sig-
that the V values in the downstream shells exceed 500 m/s at large nificant amplification of accelerations from the base to the crest
depths, while at symmetric locations in the upstream shells the V would be observed. The geomaterials perform elastically for low-
values remain much lower. Thus, the stiffer downstream shell ap- intensity motions, so the dam’s fundamental vibration period is
peared to partially reflect the vertically impinging SV waves to- considered essentially equal to its elastic eigenperiod, Toe. How-
ward the upstream shell, producing seismic amplification effects ever, most geomaterials are prone to stiffness degradation and an
in the upstream region without affecting the peak acceleration at increase in hysteresis, particularly under strong shaking. Thus, dur-
the crest. There was no such stiffness contrast within the dam’s ing an maximum credible earthquake (MCE) design scenario (in
body in the ‘‘end of construction’’ conditions, and therefore, the spa- which the dam must survive without the loss of impounding
tial variation in the peak accelerations was uniform, and there was capacity), or even an operating basis earthquake (OBE) scenario
no difference between upstream and downstream accelerations for (in which the dam must survive without considerable damage) in
vertically impinging SV waves (see Table 2). high-seismicity areas, the geomaterials behaved in a non-linear
manner that may be reflected on the dam’s fundamental vibration
4.2. Factors affecting the non-linear fundamental dam vibration period, To.
period, To To validate this scenario, Fig. 7 presents results from four of the
performed analyses, all pertaining to the trapezoidal cross-section,
As with any structure, knowledge of a dam’s fundamental vibra- without berms, of a medium height dam (H = 80 m) founded on
tion period provides a good index of its seismic response, at least as firm ground (Vb = 1000 m/s) in ‘‘end of construction’’ conditions. In

Fig. 6. Shear wave velocity contours under steady-state seepage conditions in the case of the short dam (H = 40 m).
202 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

Table 2 particular, Fig. 7a presents the results for very-low-intensity


List of symbols. (PGA = 0.05 g) and very-high-intensity (PGA = 0.5 g) shaking due
A Dimensionless scalar of employed model to the low-frequency Aigio time history, whereas Fig. 7b does the
B Constant of the employed model same for the relatively high-frequency Splb time history. The re-
e Void ratio sults are presented in terms of the dam’s base-to-crest spectral
Fh Horizontal inertial force of the sliding mass
FSd Factor of safety against seismic ‘‘failure’’
amplification ratios Sa,crest/Sa,base, normalised by their values for
Gmax Maximum (dynamic) shear modulus T = 0 (normalised transfer functions). This normalisation empha-
Gref
max
Maximum (dynamic) shear modulus at reference state in employed sises only the depicted value of the dam’s fundamental vibration
model period, To, and not the amount of seismic amplification, which var-
Gt Tangential shear modulus
ies widely given the very different PGA levels compared. Based on
H Dam height
kh Seismic coefficient of the sliding mass Fig. 7a for the Aigio time history, the dam’s fundamental vibration
khE ‘‘Effective’’ value of the seismic coefficient period almost doubles, from its value To = 0.8 s to a value of To = 1.4
khmax Peak value of the seismic coefficient based on the resultant s, as a result of a 10-fold increase in the PGA from 0.05 g to 0.5 g.
acceleration time history This 75% increase in To was attributed to severe stiffness degrada-
kh(t) Resultant (horizontal) acceleration time history of the sliding mass
Kt Tangential bulk modulus
tion and an increase in the hysteretic damping of the dam’s geoma-
M Earthquake magnitude terials during the strong seismic event compared to that of the
MCE Maximum credible earthquake very-low-intensity motion. However, when the same dam was
OBE Operating basis earthquake subjected to the Splb excitation, the initial value of the dam’s fun-
PGAcrest Peak value of the seismic acceleration at the crest
damental vibration period To = 0.73 s (for PGA = 0.05 g) was af-
PGArock Peak value of the seismic acceleration at the outcropping bedrock
PGA Peak value of the seismic acceleration at the ‘‘free-field’’ of the fected very little by the 10-fold increase in the PGA: a 16%
foundation soil increase to a value of To = 0.85 s.
PI Plasticity index This quantitative difference in the amount To increases between
p Mean effective stress Fig. 7a and b was attributed to the difference in the predominant
pref Mean effective stress at reference state in employed model
pa Atmospheric pressure
period, Te, of the applied excitation each time. In particular, the
r Deviatoric stress-ratio tensor low-frequency Aigio time history had a predominant period
rref Reference state in employed model Te = 0.4 s that was relatively similar to the elastic eigenperiod of
Sa,crest Spectral acceleration at crest dam vibration Toe = 0.73 s, making near-resonance phenomena ex-
Sa,base Spectral acceleration at dam’s base
pected. In contrast, the relatively high-frequency Splb time history
s Deviatoric stress tensor
Tmass Fundamental period of the sliding mass had a value of Te = 0.14 s that was very different from the elastic
Tm Mean period of the seismic excitation eigenperiod of dam vibration Toe = 0.73 s, and thus, the dam was
Te Predominant period of the seismic excitation expected to vibrate out-of-phase from its excitation. As an effect,
Toe Elastic fundamental dam vibration period the expected geomaterial non-linearity was minimal. Overall, it
To Non-linear fundamental dam vibration period
t Width of the sliding mass in the horizontal direction
was concluded that the dam’s fundamental vibration period To al-
V Shear wave velocity of the soil ways increased due to shaking intensity (with the PGA being a
Vb Shear wave velocity of the foundation layer good index for the shaking intensity), but this increase was signif-
Vsd Average (non-linear) shear wave velocity within the dam body icant only for cases near resonance; that is, for values of Toe/Te
W Weight of the sliding mass
slightly larger than one. Values of Toe/Te 6 1 are hardly expected
w Maximum distance between two lines that are parallel to the
points of entry and exit of the failure surface and adjoin the sliding in practice because they may appear for only regular (and not tall)
mass embankments under very-high-frequency motions.
z Maximum depth of the failure surface However, it was still uncertain whether other parameters also
a1 Model constant affected the increase in the dam’s fundamental vibration period
c1 Model constant
from its elastic eigenperiod value of Toe to its non-linear counterpart
c Cyclic strain amplitude
kd average shear wavelength within dam body To. To shed light on this issue, Fig. 8 compares the results in terms of
m Poisson’s ratio the dam’s base-to-crest spectral amplification ratios Sa,crest/Sa,base
n Hysteretic damping from three analyses pertaining to a dam with H = 40 m and founded
q Material density
on Vb = 1000 m/s, under the Splb seismic excitation for PGA = 0.05 g.
X Distance in generalized stress-ratio space used in employed model
The difference between these three analyses was that the dam had a
trapezoidal cross-section without stabilising berms and fell within

Fig. 7. Typical comparison of dam base-to-crest spectral amplification ratios normalised by their values for T = 0 (normalised transfer functions) and PGA = 0.05 g and 0.5 g
for two cases of seismic excitation (results for a trapezoidal dam section with H = 80 m, founded on Vb = 1000 m/s, ‘‘end of construction’’ condition): (a) Aigio time history
when Toe/Te = 1.85 (near resonance) and (b) Splb time history when Toe/Te = 5.2 (out-of-phase vibration).
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 203

To disregard local amplifications and spatial variations at the


dam’s crest that are of little practical importance (see Fig. 7 for
an example), the value of PGAcrest in this paper was estimated as
the maximum value of the resultant acceleration time history in
the area of the upper 10% of the dam’s height, i.e. the area of the
dam in the close vicinity of the crest. Papadimitriou et al. [17],
by compiling pertinent data from 2D non-linear analyses, have
demonstrated that this amplification ratio is affected by the dam’s
fundamental vibration period, To, and the predominant period of
excitation, Te. They have also demonstrated that scatter is reduced
if the PGAcrest/PGA ratio is correlated to the dam-to-excitation per-
iod ratio, To/Te, because this latter ratio acts as a tuning period
ratio.
Thus, Fig. 9 adopts this type of correlation for the analyses per-
formed in this study and presents only the pertinent data that cor-
Fig. 8. Typical comparison of dam base-to-crest spectral amplification ratios
(transfer functions) for a dam without stabilising berms or an impounded reservoir, respond to two distinct cases: tall embankments founded on a very
the same dam with stabilising berms but no reservoir and the same dam without compliant foundation soil (Vb = 250 m/s, represented by hollow
berms but with an impounded reservoir (H = 40 m, founded on Vb = 1000 m/s, under symbols) and dams founded on a solid rock foundation
the Splb seismic excitation for PGA = 0.05 g). (Vb = 1500 m/s, represented by solid symbols). Moreover, the
shapes of the symbol differentiate the data related to low-intensity
(PGA < 0.15 g) excitations from those related to high-intensity
the ‘‘end of construction’’ (without an impounded reservoir) condi-
(PGA > 0.15 g) excitations. The employed data verify the adopted
tions in one case, whereas in the remaining two analyses, the same
correlation to the tuning period ratio given the small relative scat-
dam appeared once with stabilising berms (again without an im-
ter, and the peak dam amplifications are observed for tuning ratio
pounded reservoir) and once with both a trapezoidal cross-section
values close to unity. Furthermore, the peak amplification ratios
and an impounded reservoir (‘‘steady-state seepage’’ conditions).
range from 1.1 to 3.6, comparable to but somewhat lower than
Observe, first, that the existence of typical stabilising berms did
the values reported by Hwang et al. [49] and Cascone and Rampello
not appear to affect the dam’s fundamental vibration period, To, be-
[48]. These relatively smaller values in the results presented here
cause the berms had to be much bulkier than H/3 in height and 2H/3
could be because the literature reports normalise the peak acceler-
in width (the dimensions used in this paper) to provide the ex-
ation at the crest to the peak acceleration at the dam base, not to
pected stiffening effect on the dam’s fundamental vibration period
the PGA. Thus, the literature-reported amplification ratios are sys-
To. Similarly, the impounded reservoir did not appear to affect the To
tematically higher because de-amplification of the motion (with
value, which could be attributed to the reduced amplitude of hydro-
respect to the PGA) is consistently observed at the dam base (see
dynamic pressures at the typically mildly inclined slopes of dams
examples in Fig. 5). Our numerical estimates depicted PGAcrest as
and embankments (assumed equal to 1:2–1:2.5 (vertical:horizon-
the peak value of the resultant acceleration in the close vicinity
tal) in this paper), compared to the full (Westergaard) amplitude
of the crest (i.e., the upper 10% of the dam’s height) and not the va-
expected in the almost vertical slopes of other types of concrete
lue at the very top of the dam, which has generally resulted in
dams or retaining walls [47].
slightly smaller yet more consistent values.
Focusing on the effect of foundation conditions, Fig. 9 indicates
4.3. Factors affecting the peak ground acceleration at the crest, PGAcrest that the dam amplification ratio was significantly affected by the
stiffness of the foundation layer. In particular, Fig. 9 clearly demon-
The significance of the peak acceleration at the crest, PGAcrest, strates that founding a dam on hard rock led to reduced radiation
was already highlighted in Section 2. Of special interest is the com- damping and thus much higher seismic amplification within the
parison of its value to the intensity of the applied seismic motion dam’s body. The opposite trend occurred for relatively soft soil
(PGA) because the PGAcrest/PGA ratio essentially depicts the seis-
mic amplification ratio within the dam body macroscopically. Re-
ported values of this ratio from case studies have not been found
in the literature. However, there have been reports of ground mo-
tion amplification values from the bases to the crests of various
dams. Cascone and Rampello [48] have presented real cases in
which a dam’s base-to-crest amplification ratio ranges from 1.5
to 10, with the higher values referring to stiff dams subjected to
weak motion and the lower values to more flexible dams under
strong motion. Similarly, Hwang et al. [49] have presented an anal-
ysis of seismic records for the Liyutan dam, reporting amplification
values of approximately two to six, with an average amplification
ratio of four under very weak excitations. However, these amplifi-
cation values decrease by 1.4–1.7 times under strong seismic shak-
ing, such as during the Chi–Chi earthquake. Such reports have
depicted significant amplifications within a dam’s body, and only
parametric numerical studies can systematically shed light on
the factors affecting these amplification phenomena. Thus, in addi-
tion to the previous section’s study of how various factors affect Fig. 9. Correlation of the dam amplification ratio, PGAcrest/PGA, to the tuning dam
period ratio, To/Te, for two extreme cases of foundation layer stiffness (Vb = 1500 m/s
the dam’s fundamental vibration period, To, the numerical investi- and 250 m/s) and two PGA levels (lower and higher than 0.15 g). The results
gation was extended to how the same factors affected the value of correspond to trapezoidal cross-sections with H = 20 and 120 m under the Splb and
PGAcrest, or, more accurately, the PGAcrest/PGA ratio. Aigio excitations and ‘‘end of construction’’ conditions.
204 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

conditions, which offered significant radiation damping to the trend line of PGAcrest/PGA versus To/Te for Vb = 1500 m/s that was
vibrating geostructure, thus reducing the seismic amplification established in Fig. 9 was still valid for cases of dams that were
within the dam. Furthermore, Fig. 9 fully describes the tuning role founded on hard rock but also included typical stabilising berms
of the period ratio To/Te because the seismic amplification was re- and/or an impounded reservoir. These cases are denoted by differ-
duced due to the dam’s increasingly out-of-phase vibration when ent symbols and comprise a relatively narrow band of data around
this ratio became greater than 1.0. De-amplification effects were the dashed trend line in the figure. Thus, it was concluded that the
even observed for unusually large values of To/Te (such as in very presence (or absence) of typical stabilising berms and the
tall dams under strong, high-frequency shaking) due to the dam’s impoundment of the reservoir did not affect the correlation be-
complete out-of-phase vibration. However, this de-amplification tween PGAcrest/PGA and To/Te. In addition, given that these param-
effect was much more probable when the foundation soil stiffness eters were also found insignificant to the value of the dam’s
was relatively low due to the contributing effect of increased radi- fundamental vibration period To (Section 4.2), it was concluded
ation damping. In such cases, de-amplification could be observed that they did not affect the value of PGAcrest. The reasons why these
even for relatively low, more probable values of To/Te. For example, parameters did not affect PGAcrest were identical to those given in
de-amplification effects within the dam body could be observed for Section 4.2, explaining why the parameters also did not affect the
values of To/Te as low as two in the extreme case of Vb = 250 m/s, dam’s fundamental vibration period To, and need not be repeated
whereas such effects could be observed only for To/Te > 6 when here.
Vb = 1500 m/s. Finally, Fig. 9 also investigates whether the PGAcrest/
PGA normalisation alleviated the effect of the PGA level on the em-
ployed correlation. It was observed that for both relatively low 5. Factors affecting the peak seismic coefficient khmax
(<0.15 g) and relatively high (>0.15 g) PGA values, the data fol-
lowed the same trend lines (roughly delineated by dashed lines) This section investigates the factors affecting the peak seismic
governed by the foundation layer stiffness (Vb value), but the data coefficient, khmax, of potential sliding masses within dams. To this
for the low PGA values corresponded to lower To/Te values than end, each of the 110 performed nonlinear ground response analy-
their counterparts for high PGA values. Thus, the effect of the ses was processed to provide the resultant horizontal acceleration
PGA level was satisfactorily introduced through the PGAcrest/PGA time history, kh(t), for a number of potential sliding masses within
normalisation but also correlated with the tuning period ratio To/ its body (approximately 10 sliding masses per analysed case). By
Te indirectly because an increase in PGA led to more intense geo- estimating the peak acceleration value of each of these time histo-
material nonlinearity and thus increased the value of To. ries, a database of 1084 khmax values was created for sliding masses
According to these observations, high amplification ratios were of different geometries. Given the parametric nature of the per-
expected for short dams characterised by low fundamental dam formed analyses, the results also enabled the study of the relative
periods, To, that were comparable to the usual predominant excita- importance of the investigated parameters (input motion charac-
tion periods, Te, especially for low PGA values and stiff foundation teristics, stiffness of foundation soil, existence of berms and/or im-
layers. For higher PGA values, when non-linearity increased the To pounded reservoir, dam height) on the values of khmax.
values of such short dams, the seismic amplification within the First, the effect of sliding mass geometry on the values of khmax
dams was smaller and de-amplification may even have been ob- was investigated, namely, the fundamental effect of the sliding
served, especially if these dams were founded on soft layers. For mass’s maximum depth (Section 5.1) and of its exact shape (Sec-
tall dams, low amplification ratios were generally expected, irre- tion 5.2), to evaluate pertinent literature proposals. Next, the ef-
spective of the usually expected seismic excitations. Nevertheless, fects of reservoir impoundment (Section 5.3) and the existence of
de-amplification effects were not generally expected for tall dams berms (Section 5.4) on the values of khmax, which have attracted
because such dams could not be readily founded on soft layers. less attention in the literature, were examined.
The foregoing conclusions with respect to PGAcrest were based
on observations made in Fig. 9, where the data correspond to trap-
ezoidal cross-sections and conditions at the ‘‘end of construction’’. 5.1. Effect of the maximum depth z of a sliding mass on khmax
Thus, it was worth investigating whether the presence of typical
stabilising berms and/or an impounded reservoir affected these Based on Makdisi and Seed [15], the values of khmax were suc-
conclusions. To this end, Fig. 10 investigates whether the dashed cessfully normalised by the peak acceleration at the crest, PGAcrest,
for generalisation purposes. These authors also demonstrated that
the dimensionless ratio khmax/(PGAcrest/g) decreases with an in-
crease in the normalised ratio z/H, i.e. the maximum depth z of
the sliding mass within a dam body divided by the dam’s height
H (see Fig. 1 for definitions). The analyses performed in this paper
enabled the testing of this geometric normalisation of the maxi-
mum depth z. Fig. 11a and b present a typical comparison of the
normalised horizontal acceleration time histories kh/(PGAcrest/g)
for the same sliding mass (with z/H = 0.5) of an dam with
H = 40 m (‘‘end of construction’’ conditions, Vb = 1000 m/s without
berms) under two widely different excitations, Splb and Aigio, with
the same low-intensity PGA = 0.05 g. In addition to the differences
in the duration of the sliding mass vibration, the results exhibited
different values of khmax/(PGAcrest/g) with 0.60 for the high-fre-
quency Splb and 0.66 for the low-frequency Aigio. This type of dif-
ference in the khmax/(PGAcrest/g) values for sliding masses with
identical z/H ratios under different excitations is systematic and
Fig. 10. Effect of typical stabilising berms and/or an impounded reservoir on the
correlation of the dam amplification ratio, PGAcrest/PGA, to the tuning dam period
characterises the majority of the numerical results. Thus, the
ratio, To/Te. The results correspond to earth dams with H = 40 and 80 m, founded on proposal made by Makdisi and Seed [15], which employs a purely
hard rock (Vb = 1500 m/s), under the Splb and Aigio excitations with PGA = 0.25 g. geometrical normalisation of maximum depth z (for generalisation
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 205

Fig. 11. Typical effect of the maximum depth z of a sliding mass on the value of khmax/(PGAcrest/g) for sliding masses with different z/H values and under different excitations,
namely: (a) with z/H = 0.5, z/kd = 0.29 under the Splb excitation, (b) with z/H = 0.5, z/kd = 0.21 under the Aigio excitation, and (c) with z/H = 0.67, z/kd = 0.28 under the Aigio
excitation (H = 40 m, PGA = 0.05 g, Vb = 1000 m/s, ‘‘end of construction’’ conditions, without berms).

purposes) and neglects input motion characteristics, needs The validity of simplifying Eq. (7) was verified in Papadimitriou
refinement. et al. [50]. Of interest here was the determination of whether the
Thus, a new normalisation scheme was needed that would z/kd normalisation correlated better to khmax/(PGAcrest/g) than the
introduce the input motion characteristics and combine them with z/H normalisation. Fig. 11c presents the normalised horizontal
the geometry of the sliding mass. The maximum depth of the slid- acceleration time history kh/(PGAcrest/g) for another sliding mass
ing mass z can be normalised over kd, i.e. the average shear wave- with z/H = 0.67 in the same dam as that depicted in Fig. 11a and
length within the dam’s body. By definition, this wavelength b, under the Aigio excitation with PGA = 0.05 g from Fig. 11b. Ob-
depends on the average (non-linear) shear wave velocity within serve that the khmax/(PGAcrest/g) values in Fig. 11a and c almost coin-
the dam’s body, Vsd, and the predominant vibration period of the cide despite the different z/H values, which could be attributed to
dam, Td. The former can be estimated based on the dam’s non-lin- the fact that the two different sliding masses corresponded to the
ear fundamental vibration period, To, and the dam height, H, using same z/kd value given the different predominant excitation periods
the following equation: of the Splb (Fig. 11a) and Aigio (Fig. 11c) excitations. Interestingly,
this type of coincidence in the khmax/(PGAcrest/g) values for sliding
V sd ¼ ð2:6HÞ=T o ð6Þ
masses with the same z/kd under different excitations was system-
which is a simplification of the equation proposed by Dakoulas and atic and could provide the basis for correlating khmax/(PGAcrest/g) to
Gazetas [46] for dams. The predominant vibration period of the the dimensionless thickness z/kd for design purposes. This alterna-
dam, Td, is equal to neither the predominant excitation period, Te, tive is investigated in Fig. 12, which includes peak seismic coeffi-
nor the dam’s non-linear fundamental vibration period, To, but cient data for dams in ‘‘end of construction’’ conditions and with
generally takes a value in between the two. Thus, for simplicity, H = 40, 80, and 120 m, founded on stiff soil or rock (Vb P 1000 m/
its value can be estimated as the mean value of the two period s) under all four excitation waveforms, and with PGA values ranging
values, and hence the average shear wavelength, kd, within the from 0.05 to 0.50 g. Fig. 12 illustrates that the well-known decreas-
dam’s body can then be approximately estimated by: ing effect of the maximum depth z on the value of khmax/(PGAcrest/g)
  was corroborated by our analyses. More importantly, this decreas-
2:6H T o þ T e
kd ¼ V sd T d ¼ ð7Þ ing effect could be successfully generalised if the maximum thick-
To 2
ness z was normalised over kd.

5.2. Effect of the exact shape of a sliding mass on khmax

Given the relatively small scatter of the data in Fig. 12, it was
understood that the correlation of khmax/(PGAcrest/g) to the nor-
malised maximum depth z/kd provided a good overall estimate
of the acceleration regime in any sliding mass. However, from
a geometrical perspective, the maximum depth z of a sliding
mass cannot fully describe its geometry because many differ-
ently shaped sliding masses with the same z value could be
defined within a single dam. Two additional geometrical param-
eters, w and t, were introduced to better describe the shape of
any sliding mass. As illustrated in Fig. 1, parameter ‘‘t’’ corre-
sponded to the width of the sliding mass in the horizontal direc-
tion, whereas parameter ‘‘w’’ corresponded to the maximum
distance between two lines that were parallel to the failure sur-
face’s points of entry and exit and adjoined the sliding mass.
Small (t/w) ratios corresponded to relatively elongated ‘‘thin’’
sliding masses, whereas large (t/w) ratios corresponded to rela-
tively ‘‘bulky’’ sliding masses. To investigate the relative impor-
tance of the normalised thickness (t/w) on the resultant
acceleration time history and its related peak seismic coefficient,
khmax, Fig. 13 provides a typical comparison of the normalised
Fig. 12. Correlation of khmax/(PGAcrest/g) ratio to z/kd ratio of a sliding mass resultant acceleration time histories for two sliding masses with
(founded on Vb P 1000 m/s, all seismic excitations and PGA = 0.05–0.5 g). the same dam-foundation-excitation combination and the same
206 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

5.3. Effect of reservoir impoundment on khmax

In Section 4, it was demonstrated that reservoir impoundment


does not affect the PGAcrest/PGA dam amplification ratio (Fig. 10),
nor does it affect the value of the dam’s non-linear fundamental
vibration period, To (Fig. 8). However, it was also shown that a res-
ervoir does affect the spatial variation of the peak horizontal accel-
erations at the dam’s surface, mainly by amplifying the motion in
the upstream shell for ‘‘steady-state seepage’’ conditions (Fig. 5).
Thus, it was still uncertain whether reservoir impoundment would
have a different effect on the vibration of upstream or downstream
sliding masses within the same dam. Therefore, Fig. 14a investi-
Fig. 13. Typical effect of the normalised thickness of a sliding mass, t/w, on the gates this issue by presenting a typical comparison of resultant
value of khmax/(PGAcrest/g) for ‘‘thin’’ and ‘‘bulky’’ sliding masses with the same z/kd acceleration time histories for upstream and downstream sliding
ratio. H = 80 m, Vb = 1000 m/s, under the Aigio excitation with PGA = 0.25 g and masses with the same geometry that are of intermediate depth
‘‘end of construction’’ conditions. (z/kd = 0.37) but apply to the same dam-foundation-excitation
combination. The time histories were relatively similar, but there
depth z and width w but different thicknesses t. In particular, was a clear amplification depicted in the upstream sliding mass,
the sliding mass with (t/w) = 0.12 could be considered ‘‘thin’’, reflected by a higher value of khmax/(PGAcrest/g) = 0.62 compared
whereas its counterpart with (t/w) = 0.24 was ‘‘bulky’’. The to 0.53 for the downstream sliding mass with identical geometry.
comparison indicated that the time histories were qualitatively This type of difference was found systematically in all relatively
similar, yet the ‘‘thin’’ sliding mass was characterised by higher shallow to medium-depth sliding masses but not in relatively dee-
peak seismic coefficient values because khmax/(PGAcrest/g) = 0.62, per sliding masses. To illustrate this phenomenon, Fig. 14b pre-
compared to 0.55 for its ‘‘bulky’’ counterpart that had the same sents a similar comparison of upstream and downstream sliding
z/kd value. Such differences were systematic and attributed to masses with z/kd = 0.55, in which the differences are diminished.
the fact that the former included mostly surficial locations on These effects could be attributed to amplification phenomena
the dam body, where higher accelerations were expected, com- due to the stiffness contrast between the saturated (and thus soft-
pared to the latter sliding masses, which were heavier and er) upstream shell compared to the non-saturated (and thus stiffer)
reached deeper locations within the dam body that vibrated less downstream shell of a zoned earth dam at ‘‘steady-state seepage’’
intensely (Fig. 5). conditions. Note the differences in the shear wave velocity values

Fig. 14. Typical effect of reservoir impoundment on the value of khmax/(PGAcrest/g) for upstream and downstream sliding masses with identical geometries that are: (a)
shallow (z/kd = 0.37) and (b) deep (z/kd = 0.55). H = 80 m, Vb = 1000 m/s, under the Aigio excitation with PGA = 0.25 g and ‘‘steady-state seepage’’ conditions.

Fig. 15. Typical effect of stabilising berms on the value of khmax/(PGAcrest/g) for sliding masses with identical geometrical properties z, w, and t but within earthen dams
without or with berms. The results for sliding masses that are: (a) shallow (z/kd = 0.37) and (b) deep (z/kd = 0.55). H = 80 m, Vb = 1000 m/s, under the Aigio excitation with
PGA = 0.25 g and ‘‘end of construction’’ conditions.
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 207

of the two shells in Fig. 6. However, the effect of reservoir


impoundment on resultant accelerations diminished for deeper
sliding masses because the peak accelerations at the dam base
were unaffected by the existence of an impounded reservoir (see
typical results in Fig. 5).

5.4. Effect of typical stabilising berms on khmax

Section 4 also demonstrated that the existence of typical berms


did not affect the PGAcrest/PGA dam amplification ratio (Fig. 10) or
the value of the dam’s non-linear fundamental vibration period, To
(Fig. 8). In contrast, such berms did affect the spatial variation of
the peak horizontal accelerations at the dam’s surface, mainly by
amplifying the motion in their vicinity (see typical results in
Fig. 5). Thus, it was worth investigating whether these local ampli-
fication effects also applied to the vibration of sliding masses that
contained stabilising berms. Fig. 15a presents a typical comparison
of resultant acceleration time histories for a sliding mass of inter-
mediate maximum depth (z/kd = 0.37) based on two distinct anal-
yses: a) a case in which the dam exists without berms
(symmetric trapezoidal cross-section) and b) a case in which the
same dam also has typical stabilising berms, as described in Sec-
Fig. 16. Evaluation of empirical estimates for the ‘‘effective’’ seismic coefficient, khE,
tion 3. The foundation and excitation conditions were identical in based on numerical data.
both analyses. The results indicated that the time histories were
largely identical, proving that the existence of a stabilising berm
played no role when the endpoint of the sliding mass was above with the proposal given by the British Standards [18]. However, the
the berm. This is not the case in Fig. 15b, which presents a similar large scatter (standard deviation of ±0.33) related to this crude
comparison of time histories for a deep sliding mass (z/kd = 0.55 average value demonstrated that other problem parameters ex-
and an endpoint within the berm). The presence of the berm pro- isted apart from the PGA that should be considered (e.g., character-
duced a higher peak seismic coefficient, khmax/(PGAcrest/g) = 0.39 istics of the dam, frequency content of the excitation, and
compared to 0.31 in the dam without berms. Again, these effects geometrical characteristics of the sliding mass; see Sections 4
were systematic and demonstrated that the seismic amplification and 5 for their significance).
phenomena in the vicinity of berms affected only sliding masses Similarly, Fig. 17 evaluates various estimates for the ‘‘effective’’
that contained these berms. seismic coefficient, khE, that are based on correlations to the PGA.
Additionally, the existence of an additional effect of the normalised
maximum sliding mass depth z/H was assessed. An increase in the
6. Evaluation of existing methodologies for seismic coefficient
normalised maximum depth z/H consistently reduced the value of
estimation
the seismic coefficient for the same PGA level, an issue that has
been qualitatively addressed only by the EC8. From a quantitative
6.1. Overview
perspective, the overall quantitative consistency of the proposal
made by the British Standards [18], implied by Fig. 17, masks the
The insights made into the seismic response of dams (Section 4)
and the knowledge obtained on the factors affecting the peak seis-
mic coefficient, khmax (Section 5), could prove invaluable when
evaluating existing methodologies for estimating seismic coeffi-
cients (outlined in Section 2). Such an evaluation is performed in
this section based on the pertinent numerical data from the 110
2D analyses described in Section 3.
The aforementioned methodologies found in the literature were
applied as proposed and their predictions were compared to the
respective numerical data. To evaluate the khE methodologies, the
numerical estimates of the ‘‘effective’’ seismic coefficient value,
khE, were estimated by assuming khE/khmax = 0.67, i.e. the typically
ad hoc selected value for this ratio based on the literature.

6.2. Evaluation of methodologies for ‘‘effective’’ seismic coefficients

Fig. 16 evaluates various empirical estimates for the ‘‘effective’’


seismic coefficient, khE, and illustrates that the intuitive proposals
of Terzaghi [9] are rational. Furthermore, the typical empirical val-
ues of khE (=0.1–0.25 in [14]) were considered safe options for val-
ues of PGA 6 0.25 g but could prove very non-conservative for
dams or embankments that were designed to withstand earth-
quakes of larger intensities. In addition, Fig. 16 corroborates the
intuitively rational and increasing effect of PGA on the value of Fig. 17. Evaluation of estimates for the ‘‘effective’’ seismic coefficient, khE, related to
khE, leading to an average khE/(PGA/g) = 0.66 that largely coincides the PGA and based on numerical data.
208 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

fact that it leads to an overestimation of khE values for deep sliding


masses (z/H P 0.8) and an underestimation for their shallow coun-
terparts (z/H 6 0.4). In addition, the range of empirical estimates
compiled by USCOLD [25] includes the majority of the numerical
results for sliding masses of intermediate and large depths (z/
H P 0.5) but is nonconservative for shallower sliding masses. Fi-
nally, the proposal made by Pyke [26] appears appropriate only
for deep sliding masses (z/H P 0.8). Fig. 17 illustrates that the
EC8 code provisions, which qualitatively address the issue of z/H,
have generally underestimated the numerical results for deep slid-
ing masses (z/H P 0.8). Nevertheless, these code provisions must
be evaluated with caution because they are intended to be used
for natural slopes only.

6.3. Evaluation of methodologies for maximum seismic coefficients

Fig. 18 evaluates estimates of the maximum seismic coefficient


khmax based on the values of the maximum acceleration at the
crest, PGAcrest, and the normalised maximum depth, z/H. Estimat-
ing the maximum seismic coefficient, khmax, based on PGAcrest re-
duced the scatter of the numerical results compared to similar
estimates based on PGA (Fig. 17). This result could be attributed
to the fact that PGAcrest reflected the dam’s vibration characteris- Fig. 19. Evaluation of estimates for the peak seismic coefficient, khmax, related to the
PGA and based on numerical data.
tics, whereas the PGA did not. In addition, Fig. 18 illustrates that
the reducing effect of the normalised maximum depth, z/H, on
the value of khmax, also depicted in Fig. 17, remained consistent.
With respect to the pertinent proposals from the literature, it non-linear fundamental vibration period, To, to the predominant
was observed that the guidelines of Marcusson [32] are oversim- period, Te, of the excitation, and the different symbols in the figure
plistic and lead to conservative estimates of khmax for sliding denote different normalised maximum depths, z/H, of the sliding
masses of intermediate to large depths (with z/H > 0.5). In contrast, masses. These numerical estimates corroborated the resonance ef-
the proposal made by Makdisi and Seed [15] is qualitatively accu- fects in the khmax values, and the out-of-phase dam vibration, cor-
rate and leads to accurate estimates of khmax, largely for medium responding to large values of To/Te, led to reduced values of khmax,
size dams (H = 40 and 80 m) comparable to the height of the dams as depicted by the ‘‘seismic coefficient spectrum’’ given by Zania
considered in their publication. Nevertheless, Fig. 18 also illus- et al. [21]. However, in quantitative terms, this spectrum clearly
trates that their design charts are conservative for tall dams underestimated the values of khmax for To/Te P 1.5. Overall,
(H = 120 m) and nonconservative for short dams or tall embank- Fig. 19 underlines the significance of the (tuning) period ratio To/
ments (H = 20 m). Te on the response and demonstrates that estimates of khmax made
Fig. 19 attempts an evaluation of khmax estimates based on the based on the PGA led to larger scatter than that shown in Fig. 18
PGA that do not require an a priori estimation of the PGAcrest (un- because seismic coefficients appeared to be better correlated to
like those evaluated in Fig. 18). The numerical estimates of khmax/ PGAcrest than to PGA.
(PGA/g) were correlated to the (tuning) period ratio of the dam’s Furthermore, as shown in Fig. 19, the ratio of khmax/(PGA/g)
clearly decreased with an increase in z/H for all (tuning) period ra-
tios, To/Te (see denoted trend lines). This result was in qualitative
agreement with what Rathje and Antonakos [33] have demon-
strated by ‘‘coupled’’ analyses, in which their predictive equation
for this ratio generally reduces with Tmass/Tm, where Tmass is a linear
function of the maximum depth z.
Finally, no methodology was found in the literature to account
for such issues as reservoir impoundment and the existence of typ-
ical stabilising berms. This fact alone contributed to the relatively
large scatter of the evaluated methodologies when compared to
the numerical data obtained from robust analyses.

7. Conclusion

This paper provided insights into the seismic response of earth


dams and tall embankments, with an emphasis on the factors
affecting the peak seismic coefficients of potentially sliding
masses. This insight was achieved by performing and processing
results obtained from 110 non-linear 2D dynamic ground response
analyses of a parametric nature, which yielded results for 1084 po-
tential sliding masses. These results were also used to evaluate
methodologies in the existing literature for the estimation of seis-
Fig. 18. Evaluation of estimates for the peak seismic coefficient, khmax, related to the mic coefficients for pseudo-static slope stability analyses. The fol-
PGAcrest and based on numerical data. lowing conclusions were drawn based on this study:
K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210 209

 In addition to the area of the crest, local seismic amplification and the anonymous reviewers for their constructive comments
phenomena were expected at the upstream shell of an that helped in enhancing the quality of the manuscript.
impounded reservoir and in the vicinity of typical stabilising
berms. Due to the local nature of the foregoing phenomena, References
the impoundment of the reservoir and the existence of typical
stabilising berms did not affect the peak acceleration at the [1] Seid-Karbasi M, Byrne PM. Embankment dams and earthquakes. Int J
Hydropower Dams 2004;11(2):96–102.
crest, PGAcrest, or the dam’s non-linear fundamental period, To. [2] Singh R, Roy D, Jain SK. Investigation of liquefaction failure in earthen dams
 PGAcrest was efficiently used as a normalising parameter for the during the Bhuj earthquake. Kharagpur, India: Indian Institute of Technology;
estimation of the peak seismic coefficient, khmax, of potential 2003.
[3] Elgamal AM, Scott RF, Succarieh MF, Yan L. La Villita Dam response during
sliding masses because it accounted for the dam’s vibration
earthquakes including permanent deformation. J Geotech Eng ASCE
characteristics, as opposed to the most frequently used peak 1990;116(10):1443–62.
ground acceleration at the free field, PGA, which did not. [4] Ozkan MY, Erdik M, Tuncer MA, Yilmaz C. An evaluation of Surgu dam response
during 5 May 1986 earthquake. Soil Dynam Earthq Eng 2006;15(1):1–10.
 The correlation of khmax/(PGAcrest/g) to the dimensionless maxi-
[5] Ozkan MY, Ozyazicioglou M, Aksar UD. An evaluation of Guldurcek dam
mum depth, z/H, of the sliding mass proposed by Makdisi and response during 6 June 2000 Orta earthquake. Soil Dynam Earthq Eng
Seed [15], although accurate in concept, has been characterised 2006;26:405–19.
by large scatter. More accurate estimates of khmax/(PGAcrest/g) [6] Singh R, Roy D, Das D. A correlation for permanent earthquake-induced
deformation of earth embankments. Eng Geol 2007;90:174–85.
could be achieved if this ratio was correlated to the dimension- [7] Liping J, Haian L, Yongqiang L, Chunhui L. Characteristics and factors that
less maximum depth, z/kd, of the sliding mass, where kd is the influenced damage to dams in the Ms 8.0 Wenchuan earthquake. Earthq Eng &
predominant shear wavelength within the dam’s body. The Eng Vib 2011;10:349–58.
[8] ICOLD. Dam safety and Earthquakes. Position paper of International
increased accuracy was because kd accounted for the frequency Commission on Large Dams, chaired by Dr. Martin Wieland; 2012.
content of the excitation, as opposed to the height, H, which did [9] Terzaghi K. Mechanisms of landslides. Engineering geology (Berkey)
not. volume. Geological Society of America; 1950.
[10] Newmark N. Effects of earthquakes on dams and embankments. Geotechnique
 ‘‘Thin’’ sliding masses had higher khmax/(PGAcrest/g) values than 1965;15(2):139–60.
‘‘bulky’’ sliding masses with the same z/kd ratio and dam-foun- [11] Franklin AG, Chang FK. Permanent displacements of earth embankments by
dation-excitation combination. Similarly, relatively increased Newmark sliding block analysis. Report 5, Miscellaneous paper S-71-17, US
Army Corps of Engineers, Waterways Experiment Station, Vicksburg,
values of khmax/(PGAcrest/g) were observed for upstream sliding Mississippi; 1977.
masses in ‘‘steady-state seepage’’ conditions and/or deep sliding [12] Yegian MK, Marciano EA, Gharaman VG. Earthquake-induced permanent
masses that included typical stabilising berms. deformations: probabilistic approach. ASCE, J Geotech Eng
1991;117(1):35–50.
 Existing methodologies for estimating seismic coefficients for
[13] Seed HB, Lee KL, Idriss IM, Makdisi R. Analysis of the slides in the San Fernando
pseudo-static analyses were not proved stand-alone, suffi- dams during the earthquake of Feb. 9 1971. Report No. EERC 73–2, Earthquake
ciently accurate, widely applicable, or any combination thereof. Engineering Research Center, University of California, Berkeley; 1973. p. 150.
Furthermore, there was no methodology that could account for [14] Kramer S. Geotechnical earthquake engineering. Prentice Hall; 1996.
[15] Makdisi FH, Seed HB. Simplified procedure for estimating dam and
all of the significant problem parameters depicted above (e.g., embankment earthquake-induced deformations. J Geotech Eng Division,
frequency content of excitation, exact geometry and upstream ASCE 1978;104(7):849–67.
or downstream location of sliding mass, existence of berms). [16] Rathje EM, Bray JD. ‘‘Nonlinear coupled seismic sliding analysis of earth
structures’’. J Geotech Geoenviron Eng ASCE 2000;126(11):1002–14.
Nevertheless, the database of peak seismic coefficient, khmax, [17] Papadimitriou AG, Andrianopoulos KI, Bouckovalas GD, Anastasopoulos K.
values created for this study could be used to establish a Improved methodology for the estimation of seismic coefficients for the
rational, stand-alone, and easy-to-use methodology, and an pseudo-static stability analysis of earthdams. In: Proceedings, 5th
international conference on recent advances in geotechnical earthquake
effort along these lines has been presented by Papadimitriou engineering and soil, dynamics; 2010. p. 24–9.
et al. [50]. [18] Charles JA, Abiss CP, Gosschalk EM, Hinks JL. An engineering guide to seismic
risk to dams in the United Kingdom. Build Res Establish Rep 1991.
[19] Biondi G, Cascone E, Rampello S. Performance-based pseudo-static analysis of
These conclusions’ range of applicability is identical to the var- slopes. In: Proceedings of 4th international conference on earthquake
iation of the problem parameters in the analyses on which they geotechnical engineering, Greece; 2007.
were based. Thus, they are valid for the following scenarios: [20] Bray JD, Travasarou T. Pseudostatic coefficient for use in simplified slope
stability evaluation. J Geotech Geoenviron Eng, ASCE 2007;135(9):1336–40.
[21] Zania V, Tsompanakis Y, Psarropoulos PN. Seismic slope stability of
(a) Earth dams or tall embankments, with heights, H, ranging embankments: a comparative study on EC8 provisions. In: Maugeri M,
from 20 to 120 m, that are sufficiently long to allow for an editor. Proceedings, ERTC-12 workshop on evaluation of geotechnical
accurate 2D approximation. These geostructures may or aspects of EC8, Athens, Sept. 11. Patron Editore – Quarto Inferione –
Bologna; 2011.
may not have typical stabilising berms (height and width [22] Bozbey I, Gundogdu O. A methodology to select seismic coefficients based on
up to H/3 and 2H/3, respectively), and their reservoir may upper bound ‘‘Newmark’’ displacements using earthquake records from
or may not be impounded. In all cases, they should be Turkey. Soil Dynam Earthq Eng 2011;31:440–51.
[23] Lin J-S, Whitman RV. Decoupling approximation to the evaluation of
founded on firm ground with shear wave velocities, Vb, earthquake induced plastic slip in earth dams. Earthquake Eng Struct Dynam
greater than 250 m/s. 1983;11:667–78.
(b) Seismic excitations with predominant periods, Te, between [24] Seed HB. Considerations in the earthquake-resistant design of earth and
rockfill dams. Geotechnique 1979;29(3):215–63.
0.14 and 0.50 s and peak accelerations, PGA, at the free field [25] USCOLD. Guidelines for selecting seismic parameters for dam projects. Report
of the foundation soil reaching up to 0.50 g. of Committee on Earthquakes. U.S. Committee on Large Dams; 1985.
[26] Pyke R. Selection of seismic coefficients for use in pseudo-static slopestability
analyses; 1991. <http://www.tagasoft.com>.
[27] Idriss IM. Evaluating seismic risk in engineering practice. In: Proceedings, 11th
international conference on soil mechanics and foundation engineering, San
Acknowledgements Francisco, vol. 1; 1985. p. 255–320.
[28] Salgado F. Re-evaluation of the slope stability pseudo-static method of
analysis. In: Proceedings, ERTC-12 workshop on evaluation of EC8, Athens;
The authors would like to thank the Public Power Corporation September 11, 2011.
of Greece S.A. for funding this research, as well as Civil Engineers [29] Rathje EM, Bray JD. One- and two-dimensional seismic analysis of solid-waste
Angelos Zographos, Sofia Tsakali and Stavroula Stavrou for execut- landfills. Can Geotech J 2001;38(4):850–62.
[30] Bouckovalas GD, Papadimitriou AG. Numerical evaluation of slope topography
ing the numerical analyses, whose results are partly presented effects on seismic ground motion. Soil Dynam Earthq Eng 2005;25(7–
herein. Furthermore, the authors would like to thank the editor 10):547–55.
210 K.I. Andrianopoulos et al. / Computers and Geotechnics 55 (2014) 195–210

[31] Papadimitriou AG. Topographic aggravation of the peak seismic acceleration School of Civil Engineering, National Technical University of Athens; 2006 [in
near two dimensional hills and slopes. In: Proceedings, 5th international Greek].
conference on earthquake geotechnical engineering, Santiago, Chile; 2011, [41] Papadimitriou AG. Elastoplastic modelling of monotonic and dynamic
January 10–13. behavior of soils. Doctorate thesis. Department of Geotechnical Engineering,
[32] Marcusson WF III. Moderator’s report for session on ‘Earth dams and stability Faculty of Civil Engineering, National Technical University of Athens; June,
of slopes under dynamic loads’. Proceedings, international conference on 1999 [in Greek].
recent advances in geotechnical earthquake engineering and soil dynamics, [42] Ramberg W, Osgood WR. Description of stress-strain curve by three
vol. 3; 1981. p. 1175. parameters. Technical note 902, National Advisory Committee for
[33] Rathje EM, Antonakos G. A unified model for predicting earthquake-induced Aeronautics; 1943.
sliding displacements of rigid and flexible slopes. Eng Geol 2011;122:51–60. [43] Vucetic M, Dobry R. Effect of soil plasticity on cyclic response. J Geotech Eng,
[34] Kokusho T, Esashi Y. Cyclic triaxial test on sands and coarse materials. In: ASCE 1991;117(1):89–107.
Proceedings of 10th international conference on soil mechanics and [44] Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media. J Eng Mech
foundation engineering, Stockholm; 1981. 1969;95(EM4):859–77.
[35] Hardin BO, Drnevich VP. Shear modulus and damping in soils: measurement [45] Cundall PA, Hansteen H, Lacasse S, Selnes PB. NESSI – soil structure interaction
and parameter effects. J Soil Mech Found Division, ASCE program for dynamic and static problems. N.G.I., Report 51508-9; 1980.
1972;98(SM6):603–24. [46] Dakoulas P, Gazetas G. A class of inhomogeneous shear models for seismic
[36] Hardin BO, Drnevich VP. ‘‘Shear modulus and damping in soils: design response of dams and embankments. Soil Dynam Earthq Eng
equations and curves’’. J Soil Mech Found Division, ASCE 1985;4(4):166–82.
1972;98(SM7):667–92. [47] Zangar C., Hydrodynamic pressures on dams due to horizontal earthquake
[37] Itasca Consulting Group Inc. FLAC – Fast Lagrangian Analysis of Continua. effects. Monograph, http://www.usbr.gov/pmts/hydraulics_lab/pubs/EM/
Version 5.0, User’s Manual; 2005. EM11.pdf
[38] Andrianopoulos KI, Papadimitriou AG, Bouckovalas GD. Bounding surface [48] Cascone E, Rampello S. Decoupled seismic analysis of an earth dam. Soil
plasticity model for the seismic liquefaction analysis of geostructures. Soil Dynam Earthq Eng 2003;23:349–65.
Dynam Earthq Eng 2010;30(10):895–911. [49] Hwang J-H, Wu C-P, Wang S-C. Seismic record analysis of the Liyutan earth
[39] Andrianopoulos KI, Papadimitriou A, Bouckovalas G. Explicit integration of dam. Can Geotech J 2007;44:1351–77.
bounding surface model for the analysis of earthquake soil liquefaction. Int J [50] Papadimitriou AG, Bouckovalas GD, Andrianopoulos KI. Methodology for
Numer Anal Meth Geomech 2010;34(15):1586–614. estimating seismic coefficients for performance-based design of earthdams
[40] Andrianopoulos KI. Numerical modeling of static and dynamic behavior of and tall embankments. Soil Dynam Earthq Eng; 2014, submitted for
elastoplastic soils. Doctorate thesis. Department of Geotechnical Engineering, publication.

You might also like