Regional Modeling of Liquefaction-

Induced Ground Deformation

Jean-Pierre Bardet,a) M.EERI, Tetsuo Tobita,a) Nicholas Mace,a)

and Jianping Hua)

Liquefaction-induced ground deformations are permanent ground dis-

placements resulting from earthquakes, which can extend over areas as large
as a few square kilometers and have amplitudes ranging from a few centime-
ters to few tens of meters. This type of ground deformation caused substantial
damage to lifelines and pile-foundations of buildings and bridge piers along
the Kobe shoreline during the 1995 Hyogoken-Nanbu, Japan, earthquake.
This paper presents a four-parameter multiple-linear-regression model for es-
timating the amplitude of liquefaction-induced ground displacement for both
ground-slope and free-face conditions at a regional scale. The applicability of
the model for mapping the amplitude of liquefaction-induced ground defor-
mation is investigated over selected regions. The paper also presents a re-
gional model for estimating the probability for the displacements to exceed
some threshold amplitude, and to fall within confidence intervals. Both mod-
els are useful for risk assessment to spatially distributed lifeline networks re-
sulting from future earthquakes. [DOI: 10.1193/1.1463409]

INTRODUCTION
During past earthquakes, areas of ground surface as large as a few square kilometers
were observed to shift laterally due to soil liquefaction. These liquefaction-induced lat-
eral ground deformations had amplitudes ranging from a few centimeters to a few tens
of meters. They took place in gently sloping ground conditions (0.1% to 6% slope) and
along river dikes, quaywalls and embankments. During the 1995 Hyogoken-Nanbu
(Kobe), Japan, earthquake, liquefaction induced lateral ground deformations were nu-
merous (Hamada et al. 1996a, b) and caused substantial damage to lifelines and pile-
foundations of buildings and bridge piers along the Kobe shoreline (Hamada et al.
1996a, Karube and Kimura 1996, Matsui and Oda 1996, Tokimatsu et al. 1996). One of
main damaging characteristics of liquefaction-induced ground deformations is that they
are not confined to a few isolated locations but extend over areas as large as a few
square kilometers, which may cause widespread damage to spatially distributed lifeline
networks. There have been very few attempts to map the regional extent of potential haz-
ards related to liquefaction-induced ground deformation (e.g., Martin and Andrews
1995). These preliminary studies revealed mostly difficulties and uncertainties in evalu-
ating over large areas the geotechnical parameters of existing ground deformation mod-
els (e.g., Bartlett and Youd 1992). The potential risks to lifeline networks has prompted
some utility companies to seek new probabilistic methods for assessing the extent of

a)
Civil Engineering Department, University of Southern California, Los Angeles, CA 90080-2531

19
Earthquake Spectra, Volume 18, No. 1, pages 19–46, February 2002; © 2002, Earthquake Engineering Research Institute
20 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

damage and repair to lifelines after future earthquakes and for implementing retrofit
policies for existing lifeline networks. To our knowledge however, besides the model in-
troduced by Bartlett and Youd (1992), no probabilistic approach has been proposed and
tested for evaluating the probability of liquefaction-induced ground displacement.
This paper presents a four-parameter MLR (multiple-linear regression) and probabi-
listic model for assessing the amplitude and probability of liquefaction-induced ground
deformation over large areas. One of the possible model applications is risk assessment
of spatially extended lifeline networks during future earthquakes. Following the intro-
duction, the second and third sections of the paper review the existing databases and
models of liquefaction-induced deformation. The fourth section presents the new MLR
model for predicting the amplitude of liquefaction-induced ground deformation. The
fifth section investigates the capability of the model to predict the spatial distribution of
liquefaction-induced displacements in particular case histories. The final section pre-
sents the probabilistic model for assessing confidence intervals and probability for
liquefaction-induced deformation.

DATABASES OF LIQUEFACTION CASE HISTORIES

The case histories of liquefaction-induced ground deformation during past earth-
quakes are essential for understanding and characterizing the effects of liquefaction, and
for developing physical and empirical models to predict liquefaction damage. A large
number of case histories of liquefaction-induced lateral spreads have been compiled in
the two volumes edited by Hamada and O’Rourke (1992). Some examples of case his-
tories include those from the following earthquakes: 1964 Niigata, Japan; 1971 San
Fernando, California; 1983 Ninhonkai-Chubu, Japan; 1983 Borah Peak, Idaho (Youd
et al. 1985); 1987 Superstition Hills, California (Youd and Holtzer 1994); 1989 Loma
Prieta, California (Holtzer et al. 1994); 1993 Hokkaido-Nansei-Oki, Japan (Isoyama
1994); 1994 Northridge, California (Holzer et al. 1996; Bardet and Davis 1996a, b;
Davis and Bardet 1996); and 1995 Hyogoken-Nanbu, Japan (Hamada et al. 1996b, Ishi-
hara et al. 1996, 1997).
Bartlett and Youd (1992) assembled one the most comprehensive databases on
liquefaction-induced ground deformation, which was recently revised by Bartlett (1998).
The main features of the database, which are used in the present modeling, are summa-
rized. The database has 467 entries from seven earthquakes listed in Table 1. The data
fall into four main categories:
• Ground displacement amplitudes
• Ground-slope and free-face topographies
• Seismic parameters
• Boreholes

GROUND DISPLACEMENT AMPLITUDE DATA

Table 1 summarizes the number of ground displacement data per earthquake and
site. A site is an area delineating a consistent slide in which the measured displacement
vectors can be regrouped. The vectors inside the slide area are included in the database,
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 21

Table 1. Number of sites and displacement vectors in Bartlett and Youd (1992) database

Accuracy
Earthquake Name Number of sites Number of vectors Horizontal (m) Vertical (m)
1906 San Francisco 4 4 ⫾0.1 to ⫾0.5
1964 Alaska 5 7 ⫾0.1 to ⫾0.5
1964 Niigata 14 299 ⫾0.72 ⫾0.66
1971 San Fernando 2 28 ⫾0.47 ⫾0.42
1979 Imperial Valley 2 32 ⫾0.02
1983 Nihonkai-Chubu 1 72 ⫾0.17 ⫾0.28
1987 Superstition Hills 1 6 ⫾0.1 to ⫾0.5

while the vectors that are isolated or too distant from available geotechnical data are ex-
cluded. The delineation of liquefaction-induced slides from aerial maps is not straight-
forward in all cases, and requires some engineering judgment. This task becomes diffi-
cult when the displacement vectors are scarce and small. Table 1 also lists the
measurement accuracy (Bartlett and Youd 1992). For most of the earthquakes, the mea-
surement error ranges from 10 to 50 cm, at the exception of the 1964 Niigata earthquake
(72 cm) and 1979 Imperial Valley earthquake (2 cm).
In the present analysis, the data are divided in two data sets: (A) complete data for all
ranges of displacement amplitude, and (B) data limited to displacement amplitudes
smaller than 2 meters. The data set B does not consider large displacement amplitudes
against which it may be difficult to conceive engineering countermeasures. The data
were further subdivided into ground-slope and free-face cases. As illustrated in Figure 1,
free faces (FF) are abrupt variations of ground surface elevation, such as quaywalls and
embankments. In some case histories, the displacements of free faces seem to have in-
duced the deformation of nearby liquefied ground (Hamada and Wakamatsu 1998).
Ground-slope (GS) cases correspond to free-field conditions far away from free faces.
Table 2 gives the number of entries in data sets A and B, together with their respective
partitions into free-face and ground-slope cases. Figure 2 shows the distribution of dis-
placement amplitudes. Most of the data fall in the 4-m range. As shown in Figure 3, the
difference between the databases of Barlett and Youd (1992) and Bartlett (1998) is sig-
nificant enough to be detectable from the histograms of displacement amplitude. As
shown in Bardet et al. (1999), this difference leads to MLR models slightly different
from that of Bartlett and Youd (1992).

Figure 1. Definition of free-face ratio H/L and ground slope S (Bartlett and Youd 1992).
22 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Table 2. Data sets used in MLR models

Notation Definition No. of data points

A All data (FFGS) 467
All free-face (FF) data 213
All ground-slope (GS) data 254
B All data (FFGS) with displacement smaller than 2 m 283
Free-face (FF) data with displacement smaller than 2 m 118
Ground-slope (GS) data with displacement smaller than 2 m 165

GROUND-SLOPE AND FREE-FACE TOPOGRAPHICAL DATA

In the present analysis, as in that of Bartlett and Youd (1992), the ground-slope and
free-face topographical data are used to characterize the effects of static shear loads that
drive liquefaction-induced displacements. As shown in Figure 1, the ground slope S is
the inclination of the ground surface at the location of the displacement vector. The free
face is characterized by the ratio W⫽H/L, where H is the height of the free face (i.e.,
difference between crest elevation and toe elevation) and L the horizontal distance from
the toe of the free face to the displacement vector. Figure 4 shows the distributions of
free-face ratio W and ground slope S for data sets A and B. The values of W in the da-
tabase are smaller than 55% and those of S smaller than 6%.

SEISMIC DATA
Table 3 summarizes the seismic parameters of the seven earthquakes in the database
and the corresponding number of observations for ground-slope and free-face cases.
Site-specific parameters, i.e., epicentral distance (hereafter defined as the nearest hori-
zontal distance to seismic energy source or fault rupture), and peak ground acceleration
(PGA) have values that vary within relatively small ranges. During the earthquakes prior
to 1979, the transient ground motion was poorly recorded, mainly because there were
very few strong motion instruments deployed before 1979. Most liquefaction-induced
ground deformations were measured for earthquakes having a magnitude between 7 and
7.8 and for epicentral distances smaller than 30 km.

Figure 2. Histograms of displacement-amplitudes of free-face (FF) and ground-slope (GS)

cases for (a) data set A and (b) data set B.
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 23

Figure 3. Comparison of histograms of displacement amplitude for database of Bartlett and

Youd (1992) and Bartlett (1998) on liquefaction-induced lateral displacement.

As shown in Figure 5, the data in the database of Bartlett (1998) fall within two
boundaries in the R-M plane. Above the upper bound, there is no data available in the
database, and any prediction in this region is uncertain. Below the lower bound, which is
formed by Ambraseys (1988) data, there has been no observed liquefaction-induced
ground deformation.

BOREHOLE DATA
The Bartlett (1998) database contains a total of 239 borehole tests consisting of 224
Standard Penetration Test (SPT) and 15 Cone Penetration Test (CPT) soundings. These
data are intended to characterize the geometry of soil deposits and soil properties in the
vicinity of displacement vectors. In the present analysis, the parameters for characteriz-
ing borehole data are selected to be the normalized SPT blow count N160 (Seed et al.
1985), and the borehole visual soil classification. The other parameters used by Bartlett
and Youd (1992), i.e., the mean grain diameter D50 and the fines contents F (percent by
weight smaller than 75 µm) are not used in the present analysis. Based on our experi-
ence, the data on F and D50 from boreholes are usually too sparse to obtain a meaningful
average over the depth in question, and over large areas. For SPT boreholes, the vertical
resolution of N160 data is at best 50 cm. Bartlett and Youd (1992) refined the SPT reso-
lution and accounted for the change in soil layering as noted in the borehole visual clas-

Figure 4. Histograms of (a) free-face ratio W and (b) ground slope S in data sets A and B.
24 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Table 3. Seismic parameters for earthquakes and number of observations for ground-slope and
free-face cases in the Bartlett and Youd (1992) database

Epicentral
distance
Earthquake Name Magnitude PGA (g) (km) GS-A FF-A GS-B FF-B
1906 San Francisco 7.9 0.26–0.28 13–27 2 2 2 2
1964 Alaska 9.2 0.21–0.39 35–100 3 4 2 4
1964 Niigata 7.5 0.19 21 160 139 91 65
1971 San Fernando 6.4 0.55 1 5 23 5 12
1979 Imperial Valley 6.5–6.6 0.21–0.51 2–6 2 30 2 20
1983 Nihonkai-Chubu 7.7 0.25 27 72 0 53 0
1987 Superstition Hills 6.6 0.21 23 0 6 0 6

sifications. This refinement is not considered in the present analysis as it becomes also
uncertain over large areas. In the case of the 15 CPT boreholes, N160 values were cor-
related from CPT tip resistances using the correlation of Seed and DeAlba (1986).

SOIL PROPERTIES AT VECTOR LOCATION

The present analysis uses only one soil property T15 , the cumulated thickness (m) of
saturated cohesionless soils having a normalized blow count N160 smaller than or equal
to 15, as defined in Bartlett and Youd (1992). Boreholes and displacement vectors usu-
ally have different locations, although boreholes are selected for their proximity to the
displacement vectors. The average value of T15 is first estimated at the borehole location
then interpolated to the location of the displacement vectors as suggested by Bartlett and
Youd (1992). As shown in Figure 6, the average value of T15 (i.e., X̄) located at the dis-
placement vector is estimated from the value of T15 (i.e., Xi) at each borehole surround-
ing the displacement vector through the following weighted average:

Figure 5. Ranges of available data and no-observed liquefaction-induced ground deformation

in the database of Bartlett (1998).
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 25

Figure 6. Relation of displacement vectors and boreholes (after Bartlett and Youd, 1992).

n
Xi
X̄⫽ 兺
冉冊
n (1)
i⫽1 1
di 兺
j⫽1 dj
where n is the number of boreholes used for averaging (the maximum value of n is 4),
and di is the distance between the ith borehole and the displacement vector. Figure 7
shows the distribution of the minimum distance between displacement vectors and bore-
hole data. In some cases, the minimum distance exceeds 400 m, which raises legitimate
questions on the correlation between geotechnical and displacement data. Figure 8
shows the histogram of T15 for data sets A and B. In both data sets, the distributions of
T15 for GS and FF cases are bimodal with two peaks at 3 m and 9 m, and have a maxi-
mum range of 15 m.

DATABASES OF GROUND DEFORMATION AND LIQUEFACTION OCCURRENCE

Figure 9 compares the entries in the ground deformation databases of Bartlett and
Youd (1992) and those in the database of liquefaction occurrence of Harder (1991). The
database of Harder (1991) includes case histories for which there was either evidence of
liquefaction or no evidence for liquefaction. Such a database is the basis of most lique-
faction analysis procedures, as described by Youd and Idriss (1997).
Other databases of liquefaction occurrence have been developed based on shear
wave velocity (Stokoe et al. 1988) and CPT data (Robertson and Campanella 1985).

Figure 7. Minimum distance between boreholes and displacement vectors (Bartlett and Youd
1992).
26 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Figure 8. Histograms of thickness T15 for free-face (FF) and ground-slope (GS) cases in data
sets A and B.

However, these databases are not considered in the present study. As shown in Figure 9,
almost all cases of lateral spreading fall within the boundaries of liquefaction occur-
rence. The rare exceptions deserve to be studied in detail. In default of detailed infor-
mation on these particular cases, it is speculated that these lateral deformations were
generated by the deformation of nearby liquefied soils.

REVIEW OF EMPIRICAL MODELS

Several analytical approaches have been proposed to model liquefaction-induced
ground deformation (e.g., Baziar et al. 1992; Byrne 1991; Jibson 1994; Towhata et al.
1996, 1997; Yegian et al. 1991). These analytical models are capable of explaining a few,
but not all aspects of liquefaction-induced deformations. Most analytical models require
numerous parameters for predicting liquefaction-induced deformation, and are therefore
impractical to apply over the large areas covered by lifeline networks. The empirical
methods based on case histories of liquefaction-induced deformation are alternate ap-

Figure 9. Comparison of databases of Harder (1991) and Bartlett and Youd (1992).
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 27

Figure 10. LSI model (Youd and Perkins 1987): measured versus predicted liquefaction-
induced lateral displacement (data points from Bartlett and Youd [1992] database).

proaches readily applicable for assessing damage to lifeline networks after earthquakes.
To our knowledge, there are five models for assessing liquefaction-induced lateral dis-
placements: Youd and Perkins (1987), Hamada (1986), Rauch (1997), Bartlett and Youd
(1992, 1995), and Youd et al. (1999).
The Liquefaction Severity Index (LSI) model (Youd and Perkins 1987) has similari-
ties to attenuation curves for peak ground acceleration. It relates the amplitude of
ground deformations, distance, and earthquake magnitude as follows:

log LSI⫽⫺3.49⫺1.86 log R⫹0.98 M (2)

where LSI is the general maximum amplitude of ground failure displacement (inch), R is
the epicentral distance (km), and M is the earthquake moment magnitude. LSI cannot
exceed 100 (Youd and Perkins 1987). Figure 10 compares the measured displacements
and those calculated using Equation 2. The points should fall on the line with a 1:1 slope
for a perfect prediction, and on the lines with 2:1 and 1:2 slope when the prediction is
half or twice the measured value, respectively. As shown in Figure 10, there is a poor
agreement between measured and calculated displacements, which implies that distance
R and magnitude M are not sufficient for predicting liquefaction-induced displacement.
Hamada et al. (1986) predict the amplitude of horizontal ground deformation only in
terms of slope and thickness of liquefied layer:

D⫽0.75H 0.5␪ 0.33 (3)

where D is the horizontal displacement (m), ␪ is the slope (%) of ground surface or base
of liquefied soil, and H is the thickness (m) of liquefied soil. The Hamada model is only
based on topographic and geotechnical parameters (i.e., ␪ and H), and no seismic pa-
rameters (e.g., R and M ).
Rauch (1997) regrouped liquefaction-induced ground deformations within slide ar-
eas, instead of considering them as individual displacement vectors, and applied
28 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Table 4. Values of MLR coefficients and adjusted R2 for the Bartlett and Youd (1992) and
FFGS4 models

Bartlett and Youd (1992) FFGS4

Original F15⫽13% and Data set A Data set B
Coefficients D5015⫽0.292 mm
b0 ⫺15.787 ⫺7.274 ⫺6.815 ⫺6.747
boff ⫺0.579 ⫺0.579 ⫺0.465 ⫺0.162
b1 1.178 1.178 1.017 1.001
b2 ⫺0.927 ⫺0.927 ⫺0.278 ⫺0.289
b3 ⫺0.013 ⫺0.013 ⫺0.026 ⫺0.021
b4 0.657 0.657 0.497 0.090
b5 0.429 0.429 0.454 0.203
b6 0.348 0.348 0.558 0.289
b7 4.527 - - -
b8 ⫺0.922 - - -
R2 adjusted 82.60% 61.00% 64.25% 64.27%
Number of data 467 467 467 213

multiple-linear-regression methods to these liquefaction-induced slides. Rauch proposed

three different models for the average lateral ground displacement, which are referred to
as regional, site and geotechnical. However, Rauch provides no guidance on how to as-
sess the spatial extent of liquefaction-induced slides.
Based on the database described earlier, Bartlett and Youd (1992) proposed
the following relation for predicting the amplitude of liquefaction-induced ground
deformation:

log共D⫹0.01兲⫽b0⫹bof f ⫹b1M⫹b2 log共R兲⫹b3R⫹b4 log共W兲

⫹b5 log共S兲⫹b6 log共T15兲⫹b7 log共100⫺F15兲⫹b8D5015 (4)
where D is the horizontal displacement (m); M the moment magnitude; R the epicentral
distance (km); S the slope (%) of ground surface; W the free-face ratio (%); T15 the
thickness (m) of saturated cohesionless soils (excluding depth ⬎20 m and ⬎15% clay
content) with N160⬍15; F15 the average fine content (% finer than 75 µm); and D5015
the average D50 grain size (mm) in T15 .
The values of the ten constant coefficients—b0 , bof f , and b1 to b8—are given in
Table 4. Equation 4 applies to both ground-slope and free-face cases. For free-face
cases, b5 is equal to zero and the value of bof f is specified in Table 4. For ground-slope
cases, b4 and bof f are both set equal to zero. As shown in Figure 11, most of the model
predictions are scattered within the lines with 2:1 and 1:2 slope, while they should fall
close to the line with a 1:1 slope for a perfect prediction.
Youd et al. (1999) slightly revised the Bartlett and Youd (1992) model due to the cor-
rections stemming from the 1983 Nihonkai-Chubu data set and to account for the
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 29

Figure 11. Measured liquefaction-induced ground displacements from the database of Bartlett
(1998) versus predicted displacement values by the Bartlett and Youd (1992) model.

liquefaction-induced ground deformation in coarser soils in Kobe, Japan, during the

1995 Hyogoken-Nanbu earthquake. This recent model is not investigated hereafter.

FOUR-PARAMETER MLR MODEL

CONTROLLING VARIABLES
The identification of controlling variables is a critical step in the MLR analysis of
liquefaction-induced ground deformation. The controlling parameters in our MLR
analysis were identified as a subset of those of Bartlett and Youd (1992) after a careful
consideration of other combinations of variables. One of the main criteria for selecting
variables was their direct relation to measured data, with as little influence as possible
from analysis assumptions. For instance, the thickness T15 can be directly obtained from
SPT profiles. In contrast, the average depth, which is related to the minimum factor of
safety (Rauch 1997), depends on the selected type of liquefaction analysis. Given the
limitations of the seismic and geotechnical data in the existing database, combinations
of variables other than those used by Bartlett and Youd (1992) did not seem promising.
However, future studies should consider other variables (e.g., CPT data) when new data
become available.
The parameters F15 and D5015 of Bartlett and Youd (1992) are not used in the present
analysis, because they are rather difficult to obtain from borehole data and to determine
over large areas. They require taking soil samples from boreholes, performing grain size
analysis in the laboratory, and computing averages within the layers with a SPT blow
count smaller than 15. It is obvious that this formidable task is impractical when the
areas under investigation become large. The uncertainties of the variables F15 and D5015
are certainly the largest of any MLR variables in Bartlett and Youd (1992).
For this reason, a four-parameter MLR model with variables—M, R, W, S, and
T15—is proposed and calibrated using the data sets A and B. The free-face ratio W and
ground slope S are independent variables and are not used simultaneously. The MLR
30 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Table 5. Range of values for MLR variables in data sets A and B, and their free-face and
ground-slope subsets

Data Set A Data Set B

Complete Free-Field Ground-Slope Complete Free-Field Ground-Slope
Variables FFGS-A FF-A GS-A FFGS-B FF-B GS-B
D (m) 0–10.15 1–10.15 0–5.35 0–1.99 0–1.98 0–1.99
M 6.4–9.2 6.4–9.2 6.4–9.2 6.4–9.2 6.4–9.2 6.4–9.2
R (km) 0.2–100 0.5–100 0.2–100 0.2–100 0.5–100 0.2–100
W 1.64–55.68 1.64–55.68 - 1.64–48.98 1.64–48.98 -
S (%) 0.05–5.90 - 0.05–5.90 0.05–2.5 - 0.05–2.5
T15 (m) 0.2–19.7 0.2–16.7 0.7–19.7 0.2–19.7 0.2–13.6 0.7–19.7

model was developed to provide a first-order approximation of liquefaction-induced dis-

placement over large areas, and to avoid the uncertainties arising from the determination
of F15 and D5015 . The MLR model distinguishes free-face and ground-slope cases, al-
though this distinction may not always be obvious in all circumstances.
Table 5 shows the range of values for the MLR variables in the data sets A and B,
and their corresponding free-face and ground-slope data subsets. The data sets are re-
ferred to as FF, GS and FFGS, which stand for Free Face, Ground Slope, and both Free
Face and Ground Slope, respectively. In data set A, the maximum ground-slope displace-
ment (i.e., 5.35 m) is half the maximum free-face displacement. In data set B, all dis-
placement amplitudes are smaller or equal to 2 m. The range of variables M, R, W, S,
and T15 are almost identical for data sets A and B, and their free-face and ground-slope
subsets. Table 5 is useful to define the domain of applicability of the MLR models. It is
not recommended to use MLR models for variable values that fall outside the ranges of
Table 5.
The model obtained in the present analysis is referred to as FFGS4, which stands for
free-face and ground-slope cases and four parameters. It has the following equation:

log共D⫹0.01兲⫽b0⫹bof f ⫹b1M⫹b2 log共R兲⫹b3R⫹b4 log共W兲⫹b5 log共S兲⫹b6 log共T15兲

(5)
where the variables M, R, W, S, and T15 are defined below Equation 4, and the values of
the seven coefficients—b0 , bof f , and b1 to b6—are given in Table 4. Like Equation 4,
Equation 5 applies to both free-face (b5⫽0) and ground-slope (b4⫽0) cases. The values
of variables M, R, W, S, and T15 should be confined to the observation ranges listed in
Table 5. The displacement is considered to be negligible when the values of distance R
and earthquake magnitude M fall below the lower bound of Figure 5.
The values of the b-coefficients were obtained by performing a regression analysis
with the program Minitab (1989). As shown in Table 4, the coefficients b1 , b2 , and b3
control the magnitude-dependent attenuation of liquefaction-induced displacements with
distance. For large epicentral distance, the attenuation relation is largely controlled by
the data points from Ambraseys (1988). In contrast to Bartlett and Youd (1992), FFGS4
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 31

Figure 12. Measured versus predicted displacements for four-parameter model FFGS4 cali-
brated from (a) data set A and (b) data set B.

has different attenuation coefficients for free-face and ground-slope cases. FFGS4 was
calibrated for data sets A and B, therefore producing models FFGS4A and FFGS4B. The
accuracy of multiple-linear regressions is evaluated using the adjusted R2 coefficient,
which accounts for the difference in the number of data points used in regression. As
shown in Table 4, the coefficient values of these two models are similar, and their ad-
justed R2 values almost coincide. Based on the results of Table 4, it is concluded that

Figure 13. Comparison of amplitudes of liquefaction-induced lateral displacements measured

and predicted for data set A by model FFGS4 calibrated from (a) data set A and (b) data set B.
32 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Figure 14. Comparison of relative errors between measured displacement and displacement
predicted by the Bartlett and Youd (1992) model and FFGS4 model for (a) data set A and (b)
data set B.

models FFGS4A and FFGS4B fit the data sets A and B with similar accuracy, and that
they can be used indifferently. Due to the decrease in the number of MLR variables,
FFGS4 fits measured ground displacement with less accuracy than the six-parameter
model of Bartlett and Youd (1992). However, as shown in Table 4, FFGS4 is more ac-
curate than a four-parameter version of Bartlett and Youd (1992) in which the values of
F15 and D5015 were fixed to average values (i.e., F15⫽13% and D5015⫽0.292 mm).
Figure 12 shows the measured displacements plotted against those predicted by
model FFGS4 for data sets A and B. The points are more scattered about the 1:1 line
than in Figure 12, which corresponds to lower values of adjusted R2 coefficients.
Figure 13 compares the measured displacements in data sets A and B to those pre-
dicted by model FFGS4 in a way different from that in Figures 11 and 12. The observed
and predicted values of displacement are plotted as a function of the entry number in the
data sets. Overall, this alternate representation indicates that model FFGS4 is capable of
modeling the ground displacement over a wide range of displacement amplitude. Figure
14 compares the predictions of FFGS4 and Bartlett and Youd (1992) by plotting the rela-
tive error between the measured and predicted displacements. The relative error is de-
fined as follows:

␧⫽100兩Dm⫺D兩/Dm共%兲 (6)
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 33

Figure 15. Location of free-face (G10-FF⬘) and ground-slope (H10-MM⬘) cases in

Niigata, Japan.

where Dm is the measured displacement amplitude, and D is the predicted displacement

amplitude. As shown in Figure 14, the six-parameter model of Bartlett and Youd (1992)
is more accurate than the four-parameter model FFGS4. Based on the present analysis,
some preliminary recommendations can be made regarding the selection of MLR mod-
els for predicting liquefaction-induced ground displacement. The choice of a particular
model depends on the availability of geotechnical data. When there is information avail-
able on the grain size distribution of soils, the Bartlett and Youd (1992) model is rec-
ommended. When there is little information on the soil grain size distribution, model
FFGS4 is recommended.

APPLICATION TO MAPPING OF LIQUEFACTION-INDUCED

GROUND DEFORMATION
In the previous section, the predictions of the MLR model were compared to obser-
vations as individual data points, without examining their spatial distributions over the
earthquake-impacted regions. However, during past earthquakes, liquefaction-induced
ground deformations were not confined at a few isolated locations but extended over
wide areas. This section investigates the model predictive capabilities over two particular
areas in Niigata, Japan. As shown in Figure 15, these areas are labeled G10-FF⬘ and
H10-MM⬘, and correspond to a free-face case and a ground-slope case, respectively.
Figure 16 shows the surface elevation and the contours of measured displacement
amplitudes for the area referred to as G10-FF⬘ (Kawagishi-cho) at Niigata, Japan, after
the 1964 Niigata earthquake. The surface elevation is obtained from pre-earthquake el-
evation data (Hamada 1999). In this area, there was a liquefaction-induced slide approxi-
mately 400 m by 400 m, which was adjacent to the Echigo Railway Bridge and directly
beneath the Hakusan Substation. Cables attached to a tower at the substation dipped into
the river due to movement towards the river (Hamada and O’Rourke 1992). The mea-
sured displacement amplitudes are the largest along the bank of Shinano River (i.e., free
face), and decrease with the distance from the free face.
As shown in Figure 17, the displacements that are measured on the ground surface
and at feet of bridge footings are represented as solid gray vectors, and those measured
34 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Figure 16. Ground surface elevation (m) and contours of measured amplitude of lateral ground
displacement (cm) for area G10-FF⬘ during 1964 Niigata, Japan, earthquake (data after Ha-
mada 1999; coordinates are in meters).

on buildings are shown as dashed vectors. Not all measured displacements in Figure 17
were used in the multilinear regression analysis; only those that originate from a circular
symbol were used for MLR model calibration. The displacement vectors were predicted
using model FFGS4, which was originally developed to predict only the amplitude of
liquefaction-induced ground displacement. Additional assumptions were required for
predicting the direction of liquefaction-induced lateral displacement. For free-face cases,
the horizontal displacements are assumed to be perpendicular to the free face. This as-
sumption is justified as the measured vectors are almost perpendicular to the free face to
the south of the Echigo Railway Line. The displacements measured on top of roofs were
not predicted as they do not represent accurately the liquefaction-induced ground defor-
mation. Overall FFGS4 predicts well the amplitude of ground deformation, even in areas
from which the model was not calibrated. To the north of the Echigo Railway Line and
400 m to the north of the free face along the Shinano River, there are some discrepancies
between predicted and measured displacement directions, which result from the local de-
formation of the railway embankment.
Figure 18 shows the observed and predicted displacements, surface map, borehole
location, and spatial distribution of the MLR parameters, thickness T15 and free-face ra-
tio for the area G10-FF⬘ at Niigata, Japan, during the 1964 Niigata earthquake. The val-
ues of T15 and free-face ratio W, which are required by model FFGS4, were calculated at
50 by 34 grid points evenly spaced at a 10.5-m grid interval. The values of T15 at the grid
points were generated from the values of T15 at the boreholes represented as pins in Fig-
ure 18d using the Kriging method (e.g., Golden Software 1998). The values of W at the
grid points were calculated by determining the minimum distance L between the grid
points and the free face along the riverside, which was approximated by two straight
segments. The height H of the free face was selected equal to 5.2 m. The value of W,
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 35

Figure 17. Measured (Hamada 1999) and predicted liquefaction-induced displacements for
area G10-FF⬘ at Niigata, Japan, during the 1964 Niigata earthquake.

which theoretically becomes infinite along the free face, was set equal to the maximum
value of 56%, which is the maximum observed value in the database of Bartlett (1998).
As shown in Figure 18b, the contours indicate that the amplitude of predicted dis-
placements decrease uniformly with the distance from the free face, which translates the
significant effects of free-face ratio and the relative small effects of T15 spatial variation
in model FFGS4. The model predicts maximum ground deformation in excess of 5 m
along the free face. However, the contours in Figure 18a display a different and more
concentrated spatial distribution of observed ground deformation, which may have been
caused by the local failure of the embankment along the Shinano River. One may also
speculate that displacements would have been more accurately predicted if there were
more available data on boreholes and height H of free face along the Shinano River.
There are, however, definite uncertainties in generating spatially the averages from bore-
hole data, which leads generally to overestimating the extent of liquefaction-induced
slides. There is a need to collect data outside slide areas for constraining more definitely
the spatial extent of liquefaction-induced lateral spreads.
Figure 19 shows the ground surface elevation and the contours of measured displace-
ment amplitudes for the area H10-MM⬘ (Bandai 5 cho-me) at Niigata, Japan, during the
1964 Niigata earthquake. As shown in Figure 15, a slide occurred in the vicinity of the
Hotel Niigata and covered an area of approximately 400 by 200 m. Catastrophic failure
of the hotel piles was documented in Hamada and O’Rourke (1992). The magnitudes
and directions of the liquefaction-induced displacement vectors were related to the
36 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Figure 18. Representation of (a) observed and (b) predicted amplitude of liquefaction-induced
lateral ground deformation, (c) ground surface, (d) borehole location, spatial distribution of (e)
thickness T15 , and (f) free-face ratio for area G10-FF⬘ in Niigata during the 1964 Niigata, Ja-
pan, earthquake (data after Hamada 1999; coordinates are in meters).

ground slope. This slide is considered to be a representative example of ground-slope

cases. However, the western part of the H10-MM⬘ area is closer to the Shinano River,
and ground deformations in this western part may belong to free face cases. In Figure
20, as for Figure 17, the displacements measured on the ground surface are represented
using solid lines, and those measured on buildings are drawn using dashed lines.
As shown in Figure 20, not all measured displacements, but only those with a cir-
cular symbol were used in the MLR calibration of model FFGS4. The measured dis-
placements are small compared to those of the free-face case of Figure 17. For this par-
ticular ground-slope case, the predicted displacements are assumed to be collinear to the
slope gradient direction. Overall, the FFGS4 model is capable of predicting the ampli-
tudes of ground displacement even in areas where it was not calibrated. The discrepancy
between predicted and measured directions of displacement, which are observed to the
west of area H10-MM⬘ originate from free-face effects close to the Shinano River
(Figure 15).
As shown in Figure 21, the liquefaction-induced displacement vectors were pre-
dicted over a 900 by 660-m area using model FFGS4. As shown in Figure 21c, the area
was covered with a 50 by 30 grid, the points of which were evenly spaced at 18 m and
22 m intervals in the north-south and east-west directions, respectively. The ground-
slope data (i.e., S) was smoothened using a 16-point moving average to filter out high-
frequency noises resulting from small erratic changes in closely spaced elevation data.
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 37

Figure 19. Ground surface elevation (m) and contour of measured amplitude of lateral ground
displacement (cm) for area H10-MM⬘ during 1964 Niigata, Japan, earthquake (Hamada 1999;
coordinates are in meters).

The values of T15 at the grid points were calculated from those at the discrete boreholes
represented as pins in Figure 21d using the same Kriging techniques as in Figure 18.
As shown in Figures 20 and 21, model FFGS4 predicts reasonably well the ampli-
tude and spatial distribution of liquefied displacement within the slide area. As in

Figure 20. Measured (Hamada 1999) and predicted liquefaction-induced displacements for
area H10-MM⬘ at Niigata, Japan, during 1964 Niigata earthquake.
38 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Figure 21. Representation of (a) observed and (b) predicted amplitude of liquefaction-induced
lateral ground deformation, (c) ground surface, (d) borehole location, spatial distribution of (e)
thickness T15 , and (f) average ground slope for area G10-FF⬘ in Niigata during the 1964 Nii-
gata, Japan, earthquake (data from Hamada 1999; coordinates are in meters).

the free-face case of Figures 16 through 18, there are still definite uncertainties in de-
fining the extent of the liquefaction-induced slides, which future studies need to
resolve.

PROBABILISTIC MODEL OF LIQUEFACTION-INDUCED

GROUND DEFORMATION
Based on the previously developed MLR model, this section determines (1) the con-
fidence limits for liquefaction-induced ground deformation and (2) the probability of
exceeding some level of ground deformation. The reader is referred to Draper and Smith
(1981) for details on probability analysis.

MODEL FOR CONFIDENCE LIMITS AND PROBABILITY

After introducing the variable D⬘⫽log(D⫹0.01), the MLR model FFGS4 (i.e.,
Equation 5) predicts the value of D̂⬘ in the following generic form:

D̂⬘⫽b0⫹b1X1⫹¯⫹bpXp (7)
where b0 ,..., bp are constant coefficients, X1 ,..., Xp are the model variables and p the total
number of model variables. The variance of D̂ can be expressed as follows:
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 39

Table 6. Components of vector X0

Component Variable Definition

X1 1 for free face and 0 for ground slope
X2 M Earthquake moment magnitude
X3 R Epicentral distance (km)
X4 log R
X5 log W Free-face ratio
X6 log S Ground slope (%)
X7 log T15 Thickness (m) of saturated cohesionless soils with N160⬍15

V共D̂⬘兲⫽V共b0兲⫹X 21V共b1兲⫹¯⫹X 2pV共bp兲⫹2X1covar共b0 ,b1兲⫹¯⫹2Xpcovar共b0 ,bp兲

⫹2X1X2covar共b1 ,b2兲⫹¯⫹2Xp⫺1Xpcovar共bp⫺1 ,bp兲 (8)
where V(bi) is the variance of coefficient bi , and covar(bi , b j) is the covariance of co-
efficient bi and b j . Equation 8 can be written in a matrix form as follows:

V共D̂⬘兲⫽s2XT0 CX0 (9)

s2 is the residual mean square:
n
1
s⫽
2
兺 共D̂i⬘⫺Di⬘兲2
n⫺p⫺1 i⫽1
(10)

where D⬘i is the ith observed value of D⬘, and D̂⬘i is ith predicted value corresponding to
D⬘i (i⫽1 to n). The matrix C with p⫹1 columns and p⫹1 rows is defined from the ma-

冉 冊
trix X with p⫹1 columns and n rows as follows:

1 X1,1 ¯ X1,p
1 X2,1 ¯ X2,p
C⫽共XTX兲⫺1 and X⫽ (11)
⯗ ⯗ ⯗ ⯗
1 Xn,1 ¯ Xn,p
th
where Xi,j is the value of X j for the i observation. In Equation 9, the vector X0 repre-
sents the values of the model variables for an individual observation. X0 has the same
type of components as a row of X. XT0 is the transpose of X0 :

XT0 ⫽共1 X1 X2 ¯ Xp 兲 (12)

In the case of model FFGS4, the components of X0 are defined in Table 6. The com-
ponents of matrix C and residual mean square s2 for the MLR model were calculated by
using the computer program Minitab (1989). The calculated results are as follows:
40 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

冢 冣
0.46 0.14 0.00 0.08 ⫺0.03 ⫺0.40 0.02 0.01
46.52 ⫺0.77 ⫺0.25 0.13 ⫺39.92 ⫺11.30 ⫺5.13
0.60 ⫺2.25 0.00 ⫺0.39 0.63 ⫺1.19
24.74 ⫺0.53 6.07 ⫺2.71 ⫺5.63
C⫽10⫺3⫻
0.03 ⫺0.18 0.06 0.11
Symmetric 46.84 0.04 ⫺0.16
44.29 10.62
22.33
and s2⫽0.084 (13)
Based on the values of s and matrix C in Equation 13, the confidence limits and
probability for future ground deformations can be predicted. As described in Draper and
Smith (1981), a future observation of D⬘ at X0 is predicted to fall within the bounds D⬘⫺
⬘ with a confidence value equal to 1⫺␣:
and D⫹

⬘ ⫽D̂⬘⫾z共D̂⬘⫹z⌽⫺1关1⫺␣/2兴兲,
D⫾ (14)
where z⫽s冑XT0 CX0⫹1; ⌽⫺1 is the inverse of the standard normal distribution ⌽; and
D̂⬘ is calculated using Equation 5. The corresponding bounds for displacement D are
D⫾⫽10D⫾⬘ ⫺0.01. The probability for a future ground deformation D at X0 to exceed
some value D is predicted as follows:
*

P共D⬎D 兲⫽P共D⬘⬎D⬘ 兲⫽1⫺⌽

* *
冋 D⬘ ⫺D̂⬘
*
z
册 (15)

where D⬘ ⫽log(D ⫹0.01).

* *
CONFIDENCE INTERVALS FOR LIQUEFACTION-INDUCED GROUND
DEFORMATION
Figure 22 shows the confidence limits of liquefaction-induced deformation predicted
by Equation 14 for all the individual observations in the Bartlett (1998) database. The
confidence limits corresponding to 90% confidence (i.e., ␣⫽10%) are represented by
error bars. The numbers along the horizontal axis refer to the numbering system in data
set A. Figure 22 also shows the measured displacement values and the values of mean
displacement predicted by model FFGS4 as a line centered at the error bars. The confi-
dence intervals enclose most of the measured displacement values.

PROBABILITY CALCULATION OVER LARGE AREAS

Equations 15 defines the probabilistic model for assessing the probability of the
liquefaction-induced ground deformation D to exceed some threshold value D*, given
some local conditions characterized by the parameter values X0 . The probability of fu-
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 41

Figure 22. Model FFGS4 for liquefaction-induced lateral displacement: measured displace-
ment, predicted mean displacement and confidence interval (90% normal-distribution).

ture liquefaction-induced ground deformations to exceed some threshold can be plotted

over large areas in the same way as the ground displacement amplitude in Figures 18b
and 21b.
Figure 23 shows an example of the probability map generated using model FFGS4
over the area G10-FF in Niigata, Japan, which would take place for an earthquake hav-
ing the same magnitude and distance as the 1964 Niigata earthquake. For comparison,
Figures 17 and 18a show the measured displacements in the same area. The probabilities
for displacement amplitude to exceed 2 m were evaluated using Equation 15 at the grid
points of Figure 18c, from the values of T15 and free-face ratio W previously calculated

Figure 23. Predicted probability for liquefaction-induced lateral spread to exceed 2 m in the
area G10-FF⬘ at Niigata, Japan, during the 1964 Niigata earthquake.
42 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Figure 24. Predicted probability for liquefaction-induced lateral spread to exceed 2 m in the
area H10-MM⬘ at Niigata, Japan, during the 1964 Niigata earthquake.

at these grid points. As shown in Figure 23, the probability of displacement exceeding 2
m near the free face is 100%. To the west, the probability is slightly higher due to the
increase in thickness T15 of liquefied layer. The probability contours of Figure 23 are
similar to those of predicted displacement in Figure 18b. As expected from probability
analysis, the contour line in Figure 23 along which displacements have a 50% probabil-
ity to exceed 2 m coincides with the contour line in Figure 18b along which predicted
displacement amplitudes are equal to 2 m.
Figure 24 shows an example of the probability map calculated for ground-slope con-
ditions over the area H10-MM⬘ in Niigata, Japan, which would occur for an earthquake
having parameters identical to those of the 1964 Niigata earthquake. The probabilities
for the ground deformation to exceed 2 m were evaluated at the grid points of Figure 21.
As expected, the 50% probability contours of Figure 24 correspond to the 2 m contours
of Figure 21b.

DISCUSSION
Mapping the probability of liquefaction-induced displacement amplitudes requires
the use of spatial interpolation techniques similar to those used for mapping the pre-
dicted mean values of displacement. As previously mentioned, there are definite uncer-
tainties in extending the probability maps outside the areas of measured ground defor-
mation, because there is usually little geotechnical information available for the areas
that did not undergo ground deformation.
The probabilistic model in this study is based on the database of Bartlett (1998),
which regroups data collected prior to 1994. There is a need to improve and extend the
present database on liquefaction-induced lateral spreads by including the data sets from
the 1994 Northridge earthquake and the 1995 Hyogoken-Nanbu earthquake. The data
should be collected inside and outside the areas where liquefaction-induced displace-
ments were observed to take place. This effort is presently ongoing utilizing GIS (Geo-
graphic Information Systems) techniques similar to those used here for representing the
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 43

spatial extent of liquefaction-induced deformation hazards. Following the completion of

the new database on liquefaction-induced ground deformation, new generations of
probabilistic models will be proposed.

CONCLUSION
A four-parameter multiple-linear regression (MLR) model and a companion proba-
bilistic model have been presented for estimating the amplitude and probability of
liquefaction-induced ground deformation at a regional scale. Both models apply to
ground-slope and free-face conditions, and are calibrated from measured displacements,
topographical data, borehole information, and earthquake data prior to 1994. The MLR
calibrations yielded practically identical results when performed from a data set encom-
passing all deformation amplitudes and a smaller data set limited to deformation ampli-
tudes smaller than 2 meters. The resulting MLR model (i.e., FFGS4) is recommended
for assessing liquefaction-induced ground deformation over large areas when there are
limited geotechnical data. The capabilities of FFGS4 for assessing liquefaction-induced
ground deformation over large areas have been investigated. FFGS4 was shown to gen-
erate a comprehensive representation of the free-face and ground-slope deformations
case histories during the 1964 Niigata earthquake.
The probabilistic model accompanying FFGS4 is useful for assessing the confidence
interval for predicting ground deformation and the probability of exceeding certain
ground deformation levels. The model is applicable for assessing the extent of potential
damage to spatially distributed lifeline networks resulting from future earthquakes. Both
MLR and probabilistic models are preliminary because they are only based on data col-
lected from earthquakes prior to 1994. Ongoing research effort is now focusing on data
collection of case histories of liquefaction-induced ground deformation in the 1994
Northridge earthquake, 1995 Hyogoken-Nanbu earthquake, and other recent earth-
quakes. Following the completion of new databases, new generations of models will be
proposed.

ACKNOWLEDGMENTS
The financial support of the Pacific Earthquake Engineering Research center
(PEER) and the Pacific Gas and Electric Company (PG&E) is acknowledged. The fi-
nancial support of the National Science Foundation (grant CMS9812521) is also ac-
knowledged. The authors thank Norm Abrahamson and William Savage of PG&E for
their help and advice during the course of this research project. The authors thank Prof.
L. Youd of Brigham Young University, and Dr. S. Bartlett of the Utah Department of
Transportation, for sharing their databases. The authors also thank Prof. M. Hamada of
Waseda University, Japan, and his colleagues for providing us with reports and digital
data.

REFERENCES
Ambraseys, N. N., 1988. Engineering Seismology, Earthquake Eng. Struct. Dyn. 17, 1–105.
Bardet, J. P., and Davis, C. A., 1996a. Performance of San Fernando dams during the 1994
Northridge Earthquake, J. Geotech. Eng. 122 (7), 554–564.
44 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Bardet, J. P., and Davis, C. A., 1996b. Engineering observations on ground motion at the Van
Norman Complex after the 1994 Northridge earthquake, Bull. Seismol. Soc. Am. 86 (1B),
S333–S349.
Bardet, J. P., Mace, N., and Tobita, T., 1999. Liquefaction-induced ground deformation and fail-
ure, Report to PEER/PG&E, University of Southern California, Civil Engineering Depart-
ment, May.
Bartlett, S. F., 1998. Revised liquefaction-induced database (private communication).
Bartlett, S. F., and Youd, T. L., 1992. Empirical analysis of horizontal ground displacement gen-
erated by liquefaction-induced lateral spreads, Technical Report NCEER-92-0021, National
Center for Earthquake Engineering Research, State University of New York, Buffalo.
Bartlett, S. F., and Youd, T. L., 1995. Empirical prediction of lateral spread displacement, J.
Geotech. Eng. 121 (4), 316–329.
Baziar, M. H., Dobry, R., and Elgamal, A. W. M., 1992. Engineering evaluation of permanent
ground deformation due to seismically-induced liquefaction, Technical Report NCEER-92-
0007, National Center for Earthquake Engineering Research, State University of New York,
Buffalo, 2 vols.
Byrne, P. M., 1991. A model for predicting liquefaction induced displacement, Proceedings of
the Second International Conference on Recent Advances in Geotechnical Earthquake En-
gineering and Soil Dynamics, St. Louis, Missouri, Vol. 2, 1027–1035.
Davis, C., and Bardet, J. P., 1996. Performance of two reservoirs during the 1994 Northridge
earthquake, J. Geotech. Eng. 122 (8), 613–622.
Draper, N. R., and Smith, H. 1981. Applied Regression Analysis, 2nd ed. (Wiley, New York),
709.
Golden Software, 1998. Surfer for Windows, Version 6, Golden Software, Inc., Golden, CO.
Hamada, M., 1999. Digital data of liquefaction-induced lateral spread for the 1964 Niigata
earthquake (private communication).
Hamada, M., Isoyama, R., and Wakamatsu, K., 1996a. Liquefaction-induced ground displace-
ment and its related damage to lifeline facilities, Soils and Foundations, Special Issue 1,
Japanese Geotechnical Society, 81–97.
Hamada, M., Isoyama, R., and Wakamatsu, K., 1996b. The 1995 Hyogoken-Nanbu (Kobe)
earthquake, liquefaction, ground displacement and soil condition in Hanshin area, Associa-
tion for Development of Earthquake Prediction, The School of Science and Engineering,
Waseda University, Japan Engineering Consultants.
Hamada, M., and O’Rourke, T. D., 1992. Case histories of liquefaction and lifeline performance
during past earthquakes, Technical Report NCEER-92-0001, National Center for Earthquake
Engineering Research, State University of New York, Buffalo, 2 vols.
Hamada, M., and Wakamatsu, K., 1998. Liquefaction-induced ground displacement triggered
by quaywall movement, Soils and Foundations, Special Issue 2, Japanese Geotechnical So-
ciety, 85–96.
Hamada, M., Yasuda, S., Isoyama, R., and Emoto, K., 1986. Study on liquefaction induced per-
manent ground displacement, Report for the Association for the Development of Earthquake
Prediction.
Harder, L., 1991. Evaluation of liquefaction potential using the SPT, Seismic Short Course,
Evaluation and Mitigation of Earthquake Induced Liquefaction Hazards, University of
Southern California, Los Angeles, February.
Holzer, T. L., Bennett, M. J., Tinsley, J. C. III, Ponti, D. J., and Sharp, R. V., 1996. Causes of
REGIONAL MODELING OF LIQUEFACTION-INDUCED GROUND DEFORMATION 45

ground failure in alluvium during the Northridge, California, earthquake of January 17,
1994, Proceedings of the 6th U.S.-Japan Workshop on Earthquake Resistant Design of Life-
line Facilities and Countermeasures for Soil Liquefaction, NCEER, edited by M. Hamada
and T. D. O’Rourke, pp. 345–360.
Holzer, T. L., Tinsley, J. C. III, Bennett, M. J., and Mueller, C. S., 1994. Observed and predicted
ground deformation-Miller Farm lateral spread, Watsonville, California, Proceedings of the
6th U.S.-Japan Workshop on Earthquake Resistant Design of Lifeline Facilities and Coun-
termeasures for Soil Liquefaction, NCEER, edited by M. Hamada and T. D. O’Rourke, pp.
79–98.
Ishihara, K., Yasuda, S., and Nagase, H., 1996. Soil characteristics and ground damage, Soils
and Foundations, Special Issue 1, Japanese Geotechnical Society, 109–118.
Ishihara, K., Yoshida, K., and Kato, M., 1997. Characteristics of lateral spreading in liquefied
deposits during the 1995 Hanshin-Awaji earthquake, J. Earthquake Eng. 1 (1), 23–55.
Isoyama, R., 1994. Liquefaction-induced ground failures and displacements along the
Shiribeshi-Toshibetsu River caused by the 1993 Hokkaido-Nansei-Oki earthquake, Proceed-
ings of the 5th U.S.-Japan Workshop on Earthquake Resistant Design of Lifeline Facilities
and Countermeasures for Soil Liquefaction, NCEER, edited by T. D. O’Rourke and M. Ha-
mada, pp. 1–26.
Jibson, R., 1994. Predicting earthquake-induced landslide displacement using Newmark’s slid-
ing block analysis, Transportation Research Record 1411, Transportation Research Board,
Washington, DC, 9–17.
Karube, D., and Kimura, M., 1996. Damage to foundation of railways structures, Soils and
Foundations, Special Issue 1, Japanese Geotechnical Society, 201–210.
Martin, G. R., and Andrews, D. C., 1995. Map-based characterization of liquefaction potential
for Southern California, Task H-8 Report, Southern California Earthquake Center, Univer-
sity of Southern California, December.
Matsui, T., and Oda, K., 1996. Foundation damage of structures, Soils and Foundations, Special
Issue 1, Japanese Geotechnical Society, 189–200.
Minitab, 1989. Minitab Statistical Software, Release 7, Minitab Inc., 3081 Enterprise Drive,
State College, PA 16801.
Rauch, A. F., 1997. EPOLLS: An empirical method for predicting surface displacement due to
liquefaction-induced lateral spreading in earthquakes, Ph.D. thesis, Virginia Polytechnic In-
stitute, Blacksburg, VA.
Robertson, P. K., and Campanella, R. G., 1985. Liquefaction potential of sands using the cone
penetration test, J. Geotech. Eng. 111 (3), 298–307.
Seed, H. B., and De Alba, P., 1986. Use of SPT and CPT tests for evaluating the liquefaction
resistance of sands, Use of In-Situ Tests in Geotechnical Engineering, ASCE, Geotechnical
Special Publication 6, 281–302.
Seed, H. B., Tokimatsu, K., Harder, L. F., and Chung, R. M., 1985. Influence of SPT procedures
in soil liquefaction resistance evaluation, J. Geotech. Eng. Div., ASCE, 111 (12), 1425–1445.
Stokoe, K. H. II, Roesset, J. M., Biecherwale, J. G., and Aouad, M., 1988. Liquefaction poten-
tial of sands from shear wave velocity, Proceedings of the 9th World Conference on Earth-
quake Engineering, III, Tokyo, Japan, 213–218.
Tokimatsu, K., Mizumo, H., and Kakurai, M., 1996. Building damage associated with geotech-
nical problems, Soils and Foundations, Special Issue 1, Japanese Geotechnical Society, 219–
234.
46 J. P. BARDET, T. TOBITA, N. MACE, AND J. HU

Towhata, I., Kogai, Y., Amimoto, K., and Putra, H. G., 1996. Theory and model tests on miti-
gation measures against lateral flow of liquefied ground, Proceedings of the 6th U.S.-Japan
Workshop on Earthquake Resistant Design of Lifeline Facilities and Countermeasures for
Soil Liquefaction, NCEER, edited by M. Hamada and T. D. O’Rourke, pp. 403–419.
Towhata, I., Orense, R. P., and Toyota, H., 1997. Mathematical principles in prediction of lateral
ground displacement induced by seismic liquefaction, Soils and Foundations, Japanese Geo-
technical Society.
Yegian, M. K., Marciano, E. A., and Gharaman, V. G., 1991. Earthquake-induced permanent
deformation: probabilistic approach, J. Geotech. Eng. 117 (1), 35–50.
Youd, T. L., Hanson, C. M., and Bartlett, S. F., 1999. Revised MLR equations for predicting
lateral spread displacements, Proceedings of the 7th U.S.-Japan Workshop on Earthquake
Resistant Design of Lifeline Facilities and Countermeasures for Soil Liquefaction, MCEER,
edited by T. D. O’Rourke, J. P. Bardet, and M. Hamada, pp. 99–114.
Youd, T. L., Harp, E. L., Keefer, D. K., and Wilson, R. C., 1985. The Borah Peak, Idaho, earth-
quake of October 28, 1983—Liquefaction, Earthquake Spectra 2 (1), 71–89.
Youd, T. L., and Holzer, T. L., 1994. Piezometer Performance at the Wildlife Liquefaction Site,
J. Geotech. Eng. 120 (6), 975–995.
Youd, T. L., and Idriss, I. M., 1997. Proceedings of NCEER workshop on evaluation of lique-
faction resistance of soils, Technical Report NCEER-97-0022, National Center for Earth-
quake Engineering Research, State University of New York, Buffalo.
Youd, T. L., and Perkins, D. M., 1987. Mapping of Liquefaction Severity Index, J. Geotech.
Eng. 11, 1374–1392.
(Received 15 August 1999; accepted 22 October 2001)