Professional Documents
Culture Documents
Abstract: Liquefaction assessment has primarily been performed using in situ penetration testing, but this practice has become problematic
for gravelly soils. For example, standard penetration test (SPT)- or cone penetration test (CPT)-based correlations can become unreliable
owing to interference with large gravel particles, while the Becker Penetration Test, commonly used for gravelly soil, can be relatively
expensive and requires conversion to an equivalent sand blow count. As an alternative, probabilistic liquefaction-triggering curves have
been developed based on shear-wave velocity (V s ) using gravel sites in the M w 7.9 Wenchuan earthquake. These curves have significant
uncertainty because of the small data set. In this study, new probabilistic triggering curves for gravel liquefaction have been developed based
on a V s data set. The data set consists of 174 data points (96 liquefaction and 78 no liquefaction) obtained from 17 earthquakes in seven
countries within different geological environments. The larger data set better constrained the curves and reduced the range between the 15%
and 85% probability of liquefaction curves, indicating less uncertainty. These triggering curves for gravel are shifted to the right relative
to comparable curves for sand, indicating that higher V s values are necessary to preclude liquefaction. To account for the influence of
the different earthquake magnitudes on liquefaction, a magnitude scaling factor (MSF) was developed specifically for gravel. This curve
falls within the range of other MSF curves for sands based on V s . DOI: 10.1061/(ASCE)GT.1943-5606.0002784. © 2022 American Society
of Civil Engineers.
Author keywords: Gravel liquefaction; Probabilistic liquefaction-triggering curve; Shear-wave velocity; Magnitude scaling factor
(MSF).
Introduction
1
Professor, Dept. of Civil and Environmental Engineering, Brigham
Young Univ., 430 Engineering Bldg., Provo, UT 84602 (corresponding Liquefaction of loose saturated granular soils produces significant
author). ORCID: https://orcid.org/0000-0002-8977-6619. Email: rollinsk@ damage to civil infrastructure such as bridges, roadways, and ports
byu.edu in most major earthquakes. Liquefaction and the resulting loss of
2
Ph.D. Candidate, Dept. of Civil and Environmental Engineering, shear strength causes landslides, lateral spreading, and bearing
Brigham Young Univ., 430 Engineering Bldg., Provo, UT 84602. Email: capacity failure for foundations, along with excessive settlement
jashod.roy@gmail.com and rotation of foundations. Direct and indirect economic losses
3
Associate Professor, Dept. of Civil Engineering, Univ. of California, resulting from liquefaction are substantial costs to society. Evi-
Berkeley, CA 94720. Email: adda.zekkos@berkeley.edu
4 dence of gravel liquefaction has been reported in multiple case
Associate Professor, Dept. of Civil Engineering, Univ. of California,
Berkeley, CA 94720. Email: zekkos@berkeley.edu histories during 26 earthquake events over the past 130 years, as
5
Associate Professor, Dept. of Engineering and Geology, Univ. of summarized in Table 1. Assessing the potential for liquefaction
Chieti-Pescara, Viale Pindaro, 42, Pescara 65129, Italy; Researcher, Roma of gravelly soils in a cost-effective and reliable way has often posed
1 Section, Istituto Nazionale di Geofisica e Vulcanologia, Viale Crispi, 42, a significant challenge in geotechnical engineering.
L’Aquila 67100, Italy. ORCID: https://orcid.org/0000-0001-5835 Typical laboratory investigation techniques have proven ineffec-
-079X. Email: sara.amoroso@unich.it tive for characterizing gravelly soil due to the cost and difficulty of
6
Professor, Guangxi Key Laboratory of Geomechanics and Geotechni- extracting undisturbed samples from gravelly deposits (Cao et al.
cal Engineering, Guilin Univ. of Technology, Guilin 541004, China. Email:
2013). In addition, the large particle size of gravels can lead to ar-
iemczz@163.com
7
Technologist, Roma 1 Section, Istituto Nazionale di Geofisica e
tificially high penetration resistance values from traditional in situ
Vulcanologia, Via di Vigna Murata, 605, Rome 00143, Italy. Email: tests such as the cone penetration test (CPT) and standard penetra-
giuliano.milana@ingv.it tion test (SPT) (DeJong et al. 2017). The 168-mm-diameter Becker
8 penetration test (BPT) (Harder and Seed 1986; Harder 1997) re-
Researcher, Roma 1 Section, Istituto Nazionale di Geofisica e
Vulcanologia, Viale Crispi, 43, L’Aquila 67100, Italy. Email: maurizio duces the potential for artificially high penetration values; however,
.vassallo@ingv.it this method is relatively expensive and is not available outside
9
Researcher, Roma 1 Section, Istituto Nazionale di Geofisica e of North America. In addition, the method requires a correlation
Vulcanologia, Viale Crispi, 43, L’Aquila 67100, Italy. Email: giuseppe between the BPT blow count and the SPT blow count, which leads
.digiulio@ingv.it
Note. This manuscript was submitted on July 14, 2021; approved on
to greater uncertainty relative to methods that are directly correlated
January 6, 2022; published online on April 5, 2022. Discussion period open with field liquefaction resistance.
until September 5, 2022; separate discussions must be submitted for indi- As an alternative, in situ measurement of shear-wave velocity
vidual papers. This paper is part of the Journal of Geotechnical and (V s ) is a popular way of characterizing the liquefaction resistance
Geoenvironmental Engineering, © ASCE, ISSN 1090-0241. of soil deposits. V s is a basic mechanical property of soil materials,
Borah Peak, Idaho, USA 1985 6.9 Youd et al. (1985), Andrus (1994), and Harder and Seed (1986)
Armenia 1988 6.8 Yegian et al. (1994)
Limon, Costa Rica 1991 7.7 Franke and Rollins (2017)
Roermond, Netherlands 1992 5.8 Maurenbrecher et al. (1995)
Hokkaido, Japan 1993 7.8 Kokusho et al. (1995)
Kobe, Japan 1995 7.2 Kokusho and Yoshida (1997)
Chi-Chi, Taiwan 1999 7.8 Chu et al. (2000), and Lin et al. (2004)
Kocaeli, Turley 1999 7.6 Bardet et al. (2000)
Wenchuan, China 2008 7.9 Cao et al. (2011, 2013)
Tohoku, Japan 2010 9.0 Tatsuoka et al. (2017)
Cephalonia Island, Greece 2012 6.1 Nikolaou et al. (2014), and Athanasopoulos-Zekkos et al. (2019, 2021)
Iquique, Chile 2014 8.2 Rollins et al. (2014), and Morales et al. (2020)
Muisne, Ecuador 2016 7.8 Lopez et al. (2018)
Kaikoura, New Zealand 2016 7.8 Cubrinovsky et al. (2017)
Durres, Albania 2019 6.4 Pavlides et al. (2020)
Petrinja, Croatia 2020 6.4 Amoroso et al. (2021)
directly related to the small strain shear modulus (G0 ), that is an dispersion characteristics of the ground (Stokoe et al. 1994; Andrus
essential parameter for performing soil–structure interaction analy- 1994; Kayen et al. 2002). These noninvasive test methods can sig-
sis and liquefaction evaluation under earthquake loading. The use nificantly reduce the cost of in situ V s estimation, although they are
of V s as an index of liquefaction resistance is based on the fact that affected by uncertainties related to poor ability in resolving very
V s and liquefaction resistance are similarly influenced by void ra- thin layers and very short wavelengths. They also suffer from un-
tio, effective vertical stresses, stress history, and geologic age (Youd certainty in inverting the dispersion curve and become less reliable
et al. 2001). In addition, V s is considerably less sensitive to the with depth below the ground surface (Vantassel and Cox 2021).
problems of soil compression and reduced penetration resistance Despite these limitations, which also affect techniques based on
when fines are present, compared with SPT and CPT methods. body waves, the noninvasive techniques are a good compromise
Moreover, V s only requires relatively minor corrections for fines for deriving an average V s profile at a rather low operational cost,
content (FC), at least for sands (Kayen et al. 2013), unlike the SPT especially at stiffer sites where penetration techniques are not pos-
or CPT, which requires significant correction for FC for liquefac- sible. In fact, testing at three field test sites indicated that variability
tion evaluation. The primary advantage of the in situ V s approach is of V s profiles from noninvasive methods were generally compa-
that testing can be performed at sites where boring is not possible, rable to those from invasive methods when performed by experts
or the penetration test results may be unreliable. Hence, V s meas- (Garofalo et al. 2016b).
urement can be considered a reliable and economical alternative to Dobry et al. (1982) proposed a strain-based approach for assess-
overcoming the difficulties of penetration testing through gravelly ing liquefaction potential and developed the concept of a cyclic
strata. Nevertheless, some concern exists about the applicability of threshold strain for which pore pressures are generated. To develop
V s measurement for liquefaction assessment because V s is essen- a more simplified method of liquefaction-triggering analysis com-
tially a small-strain parameter, whereas liquefaction is a medium- to patible with in situ tests like the SPT or CPT, Seed et al. (1983)
large-strain phenomenon (Jamiolkowski and Lo Presti 1990; Kramer developed a cyclic stress ratio ðCSRÞ − V s triggering model using
1996). However, the ultimate strength, at large strain, typically SPT versus V s correlations. Later, in the early 1990s, several cor-
increases as the initial shear stiffness increases, so V s may still be relations were developed between the effective stress normalized
correlated with liquefaction resistance. shear-wave velocity (V s1 ) and the cyclic resistance ratio (CRR)
The traditional methods of measuring V s require a penetrometer causing liquefaction based on an increasing number of direct field
or instrumented boreholes to measure the travel time of shear waves measurements of V s at liquefaction sites (Robertson et al. 1992;
at various depths. A downhole test requires one borehole to mea- Kayen et al. 1992; Lodge 1994). These correlations were based
sure the vertically propagating wave, while a cross-hole test re- on the liquefaction of sandy soils.
quires at least two boreholes to directly measure the horizontally Andrus and Stokoe (2000) increased the population of V s data
propagating wave (Stokoe et al. 1994). These invasive test methods by collecting a large database from locations around the world
are usually quite expensive due to the cost of drilling, casing, and where both sandy and gravelly soils had liquefied in various seis-
grouting boreholes. In the last two decades, some advanced non- mic events. Based on this data set, improved triggering curves were
invasive test methods [spectral analysis of surface waves (SASW) developed for sands and gravels for different FC percentages. The
and multichannel analysis of surface waves (MASW)] have been database of Andrus and Stokoe (2000) only contained a limited
developed that indirectly estimate the V s profile using surface-wave number of data points, where the V s1 was higher than 200 m=s
Based on the Andrus and Stokoe (2000) data set, Juang et al. or did not take place in 17 major earthquake events and adding
(2001, 2002) developed probabilistic V s1 -based liquefaction- them to the existing 47 data points from China reported by Cao
triggering curves using reliability-based concepts. Kayen et al. et al. (2011). Based on this new data set, a new family of probabi-
(2013) subsequently developed V s -based probabilistic liquefaction- listic liquefaction-triggering curves was developed using logistic
triggering curves for sandy soils by compiling a large database of regression techniques. The triggering equations developed in this
422 case histories of sand liquefaction. However, all these probabi- study include the moment magnitude for each case history as an
listic correlations are based on the liquefaction of sands rather than independent variable. This formulation made it possible to develop
gravels. a new MSF curve expressly for gravel liquefaction based on V s .
Using logistic regression techniques, Cao et al. (2011) produced This paper provides details regarding the collection and processing
liquefaction probability curves based on V s data collected from of the expanded data set along with the development of the new
the Chengdu Plain in China, where gravel liquefaction took place liquefaction-triggering procedure. The proposed V s -based trigger-
during the 2008 M w 7.9 Wenchuan earthquake [Fig. 1(a)]. These ing curves are compared with existing V s -based probabilistic trig-
probability curves were based on 47 data points (19 liquefaction gering curves developed for gravel by Cao et al. (2011), as well
and 28 no liquefaction) for a single earthquake and in a similar those developed for sand by Kayen et al. (2013).
geological environment that consisted of loose alluvial fan gravel
deposits that are typically overlain by a clay surface layer that is Collection of Additional V s -Based Case
2–4 m thick (Cao et al. 2011). Owing to the limited number of data History Data
points and the possibility of false negatives (e.g., sites where lique-
faction may have occurred but did not produce surface manifesta- As a part of this study, additional gravel liquefaction case history
tion), the 15% and 85% probability of liquefaction-triggering data were obtained by performing or collecting in situ V s test
curves are relatively far apart. In comparison, V s -based probabilis- data from sites around the world where gravelly soil liquefaction
tic liquefaction-triggering curves for sands (Kayen et al. 2013) has been identified. The sites where gravel liquefaction features
shown in Fig. 1(b) have more closely grouped 15% and 85% (e.g., surface ejecta of gravelly soils, lateral spreading, settlement
probability curves because of the much larger size of the data set. manifestation) were found were treated as liquefaction points.
Moreover, the Cao et al. (2011) probability curves were developed In addition, V s data were also collected from sites where no
for a single M w 7.9 earthquake, which made it impossible to liquefaction was evident despite having gravel layers below the
develop appropriate correction factors for different earthquake water table. These no-liquefaction points may provide an important
Fig. 1. Comparison of (a) V s -based probabilistic liquefaction-triggering curves for gravel (Cao et al. 2011); and (b) V s -based probabilistic
liquefaction-triggering curves for sand (Kayen et al. 2013). Liquefaction points are denoted by solid symbols while nonliquefied points are denoted
by open symbols.
Fig. 2. Range of grain size distribution curves for liquefiable gravelly soil based on collected case history database (according to ASTM criteria).
(Arai and Tokimatsu 2005; Picozzi et al. 2005) for some of the
Performance of In Situ Tests to Collect Additional investigated sites. The HVSR curve using ambient noise (Nogoshi
V s Data and Igarashi 1971; Nakamura 1989) was computed using ambient
At many sites V s testing was performed using the MASW method. noise recorded for a few hours by collocated seismic stations using
The MASW method considers the dispersion of Rayleigh waves to a three-component velocimeter with an eigenfrequency of 5 s. The
generate an apparent phase-velocity dispersion relationship that is HVSR curve was considered to be connected to the ellipticity of
then used in an inversion analysis to derive a V s profile. MASW Rayleigh-wave during the inversion step, with the HVSR peak pro-
surveys were performed using a linear array at each site composed viding information on the fundamental resonance frequency of the
of vertical geophones (4.5 Hz) spaced at 1-m intervals to increase site (Fäh et al. 2003).
resolution near the surface and 3-m intervals for greater depths. A This procedure, based on a joint inversion, helped extend the
sledgehammer (usually 5.5 kg) striking on a plastic plate was used investigated frequency range toward lower frequency values and
as the seismic source. The source was aligned to the geophones and to reduce the ambiguities typically encountered in velocity profile
located at several offsets (three to five) for each linear array. For inversions (Tokimatsu 1997; Wathelet et al. 2004). A joint inver-
each offset, a stack of three to five measurements was considered sion, combining active and passive data analysis as well as any a
adequate to increase the signal-to-noise ratio. The recorded phase priori information, allows one, in principle, to increase the reliabil-
offset of different frequency waves (f-k analysis) is used to develop ity of the estimates of the average shear-wave velocity profile and to
a relationship between phase velocity and frequency (or wave- achieve a greater investigation depth compared to that obtained by
length) called a dispersion curve. Based on the dispersion curves, inverting only the dispersion curve from active data (Bard et al.
inversion analyses were conducted using the Park et al. (1999) 2010). Details of these inversion analyses for every site to obtain
methodology and typically without a priori subsurface information the best estimate of V s can be found in the corresponding reference
to derive a shear-wave velocity model. It is well known that major given in the Table S1.
differences in interpretation arise not from the dispersion analyses Besides MASW, other tests, e.g., SASW, cross-hole testing, and
in particular, but from the inversion algorithm used to estimate the downhole testing, were also performed at several sites, as men-
V s profile from the dispersion curve. Therefore, the inversion pro- tioned in Table S1, to obtain the V s profile. The distribution of
cess has a much stronger effect on the final V s model compared to different test methods used to investigate the gravel case history
the dispersion curve generation method (Garofalo et al. 2016a). data in the present study is plotted in Fig. 5. The plot shows that
At some Italian sites (Avasinis and Bordano), two-dimensional the highest percentage of the V s data was obtained by performing
arrays of 24 vertical geophones with a diameter of approximately MASW testing at sites in Alaska, China, Italy, Ecuador, Greece,
50 m were installed in addition to the MASW acquisition. For and New Zealand. A major portion of the Chinese data (Zhou et al.
the array’s configuration, 23 geophones were installed around 2 2020) was also obtained by performing SASW and downhole test-
concentric circles (with radii of 12 and 24 m, respectively) with ing. Moreover, SASW testing was also performed at sites in Idaho
a spacing between adjacent geophones of approximately 10 m. by Andrus (1994). At a few sites in Alaska, Friuli, and Idaho, cross-
The last geophone was installed in the centers of circles. These hole testing was also performed. The seismic dilatometer was also
arrays acquired ambient seismic noise for at least 1.5 h, and the used to measure V s at one site in L’Aquila.
data were analyzed through the frequency-wavenumber (f-k) algo-
rithm to define a more refined dispersion curve generally charac- Data Interpretation and Selection of Critical Layers
terized by a higher resolution toward low frequencies (<10 Hz),
The V s values obtained by various in situ methods were corrected
compared to that inferred from the linear MASW analysis.
for overburden pressure to obtain V s1 using the equation
The surface-wave inversion is affected by the nonuniqueness of
the solution (Foti et al. 2009), and to better constrain the inversion 0 0.25
V s1 ¼ V s ðPa =σvo Þ ð1Þ
procedure, a joint inversion of surface Rayleigh-wave dispersion
0
curves and ellipticity curves derived from ambient vibration analy- where σvo = initial vertical effective stress; and Pa = atmospheric
ses [horizontal-to-vertical spectral ratio (HVSR)] was performed pressure approximated by a value of 100 kPa, as suggested by
quired to explore the correction factors specifically for gravelly than approximately 300 m=s. This range of critical layer depth is
soils, which is beyond the scope of this study.
In addition, the CSR for each site and earthquake event was
obtained using the simplified equation
0
CSR ¼ 0.65ðamax =gÞðσvo =σvo Þrd ð2Þ
Seed and Idriss (1971) developed Eq. (2), where σvo is the initial
vertical total stress; amax is the peak ground acceleration (PGA);
and rd is a depth reduction factor as reported in Youd et al. (2001).
The PGAs (amax ) for each site were obtained either from nearby
strong ground motion stations (SGMSs) or from USGS ShakeMaps
(Worden et al. 2010) depending on the available data. This is a sim-
ilar approach to that adopted by Idriss and Boulanger (2008) while
developing CPT-based liquefaction-triggering procedure for sandy
soil. According to the references given in the Table S1, the PGAs at
the sites of Manta, Hokkaido, Borah Peak, Wellington, Lixouri, and
Argostoli were estimated from the nearby SGMS records. The
PGAs at the Chengdu Plain in China were obtained from the con-
tour maps developed from the nearby SGMS records. At the Italian
sites, the SGMSs were located within the area of Avasinis but not
close to the liquefied areas. Hence, the PGAs estimated from the
nearby SGMSs were verified with the results of a ground motion Fig. 6. Distribution of earthquake magnitude (Mw ) and corresponding
prediction equation (GMPE) (Bindi et al. 2011) and USGS Shake- PGA for improved data set.
Map estimations, as reported in Rollins et al. (2020). At the sites in
Alaska, PGAs were obtained from a USGS ShakeMap. At the sites
in L’Aquila, PGAs were determined using the SGMSs for sites
(Aquila3 and Aquila4) close to a seismic station in operation
during the earthquake and using the information derived from the
ShakeMap (Faenza and Michelini 2011) for the remaining sites.
Overall, 88% of PGA data are directly based on SGMS records
and the remaining 12% are from USGS ShakeMaps. It should
be noted that estimation of PGAs from different kinds of sources
may add considerable uncertainty to the liquefaction potential
evaluation. However, the fact that a large percentage of the PGAs
in this study are based on direct measurement in the field can ac-
tually reduce the uncertainty to a moderate level.
Besides CSR, the moment magnitude (M w ) was treated as an-
other independent variable while developing the liquefaction-
triggering procedure. The values of M w for all seismic events are
also provided in the Table S1. The distribution of M w versus PGA
for all the data points is plotted in Fig. 6. This shows that the data
set contains a broad range of M w from 5.3 to 9.2 along with the
PGAs varying from 0.116g to 1.457g. In Fig. 6, several points over-
lap each other that actually represent multiple sites for similar earth-
quake magnitudes but different PGAs.
Based on the V s1 and CSR plots with depth, the critical layer
for each location was identified as the layer with the highest po-
tential to trigger and manifest liquefaction at the ground surface
(Cubrinovski et al. 2018). Typically, this was the gravelly layer
Fig. 7. Distribution of critical layer depth and corresponding V s1 for
corresponding to the lowest range of V s1 value below the water
expanded gravel liquefaction case history data set.
table (Kayen et al. 2013). There can be multiple critical layers along
data point, the converted CSR was plotted with the corresponding In this study, a logistic regression analysis (Liao et al. 1988; Youd
V s1 along with the modified triggering curves, as shown in Fig. 8, and Noble 1997; Cao et al. 2011, 2013; Rollins et al. 2021) has
to provide a magnitude standard. Liquefaction data points are de- been performed to obtain a new set of probabilistic liquefaction-
noted by solid symbols, while no-liquefaction data points are triggering curves based on the whole database given in the Table S1.
denoted by open symbols. Unlike the correlation of Cao et al. (2011) which involved only one
Fig. 8 shows that most of the newly added liquefaction points seismic event, the moment magnitude (M w ) has been considered to
with probabilities of liquefaction greater than 50% are satisfactorily be an independent variable along with the other explanatory vari-
evaluated by the Cao et al. (2011) triggering curves. In contrast, the ables i.e., normalized shear-wave velocity (V s1 ) and the natural log
no-liquefaction points from the Port of Wellington (Cook Strait and of the cyclic stress ratio [lnðCSRÞ] while performing the logistic
Lake Grassmere earthquakes) and Argostoli (1983 earthquake) fall regression analysis. To compute the CSR, the formulation of rd
above the 30% triggering line. In other words, these sites are evalu- in Eq. (2) has been updated to Eq. (7) to include the effect of mag-
ated as potentially liquefiable, but this is inconsistent because these nitude on the variation of rd with depth
sites did not liquefy during the actual seismic event. Furthermore,
the no-liquefaction points from L’Aquila in Italy at the middle rd ¼ e½αðzÞþβðzÞMw ð7Þ
portion of the plot fall between the 30% and 50% triggering curve,
indicating a relatively high chance of liquefaction, whereas the where
gravelly deposits at these sites did not liquefy during the actual
earthquake events. z
αðzÞ ¼ −1.012–1.126 sin þ 5.133 ð8Þ
In addition, a large number of no-liquefaction data points from 11.73
the Chengdu Plain in China newly added from Zhou et al. (2020)
during the Wenchuan earthquake fall considerably above the 30%
z
triggering line, suggesting that those points could be liquefiable. βðzÞ ¼ 0.106 þ 0.118 sin þ 5.142 ð9Þ
Because the existing Cao et al. (2011) triggering curve incorrectly 11.28
Fig. 8. Comparison of CSR-V s1 points for all collected data with probabilistic liquefaction-triggering curves developed by Cao et al. (2011). Solid
markers denote liquefaction sites, while open markers denote no-liquefaction sites.
where the series of X values represents the soil and the earthquake are statistically admissible, there are slight variations in their per-
parameters. In this study, these variables are V s1 , M w , and ln(CSR) formance in evaluating the liquefaction and no-liquefaction points
along with various combinations of these primary variables. The of various ranges. To illustrate this fact, 30% triggering curves for
vectors θ are the model coefficients, which are unknowns that need each of the different solutions were plotted together with all the
to be estimated to define the function of the probability of lique- points in Fig. 9.
faction (PL ). These parameters were obtained using the maximum Fig. 9 shows that the triggering curve based on the solution
likelihood estimation (MLE) technique. The likelihood function of Trial 1, which is a function of only a first-order polynomial
can be expressed as in V s1 , starts with a much lower range of CRR that does not prop-
Y erly evaluate several no-liquefaction points in the low range of V s1 .
Lðθ; XÞ ¼ ½PL ðθ; XÞn ½1 − PL ðθ; XÞm−n ð11Þ
The triggering curve obtained from the solution of Trial 2, which
contains a second-order polynomial in V s1 , is much improved com-
where m = total number of data in sample; and n = total number
pared to the first-order solution at the lower range, but it becomes
of liquefied sites, i.e., m − n = total number of nonliquefied sites.
slightly slanted toward the right at the upper end of the curve, which
To maximize this likelihood function, the following condition
may present some limitations in evaluating the no-liquefaction
should be satisfied:
points at the high range of V s1 . On the other hand, the Trial 4
∂Lðθ; XÞ solution having V s1 polynomial up to the fourth order makes the
¼ 0; i ¼ 1; 2; : : : ; n ð12Þ triggering curve too steep and vertical from the middle to high
∂θi
range of CRR, which may produce limitations in evaluating some
By satisfying Eq. (12), n number of equations can be obtained, liquefaction points in the middle or higher range of CSR. Hence, in
which need to be solved to estimate the model coefficients θ. judging the performance of all these curves for the given data set,
However, solving this set of equations required extensive numerical the Trial 3 solution based on a third-order polynomial of V s1 was
programming that was accomplished by simply using the variable determined to be the most suitable way of determining the lique-
selection procedure (Youd and Noble 1997) with the JMP Pro 13 faction potential of the existing points in a reliable manner. There-
software package in this study. A similar approach was taken by fore, the recommended equation for obtaining the probability of
Rollins et al. (2021) to develop the liquefaction-triggering pro- liquefaction from this updated regression analysis is as follows:
cedure based on the dynamic penetration test (DPT) penetration
resistance. However, a brief illustration of the logistic regression 1
PL ¼ ð13Þ
analysis associated with the stepwise variable selection procedure 1 þ expf1.6M w þ 4.95 lnðCSRÞ − 3.88 × 10−7 V 3s1 g
following the approach of Rollins et al. (2021) is given as follows:
Step 1: First a set of explanatory variables is defined by Rearranging Eq. (13), the revised CRR can be written in the
combining normalized shear-wave velocity (V s1 ) and earthquake following form:
parameters [ln(CSR) and M w ] in different forms. The explanatory 8 9
variables in the first set are [V s1 , V 2s1 , V 3s1 , V 4s1 , M w , ln(CSR)] and >
<3.88 × 10−7 V 3s1 − 1.6M w − ln 1−P L >
=
different combinations of these variables, for example, M w lnðCSRÞ, PL
CRR ¼ exp ð14Þ
M w V s1 , and V s1 lnðCSRÞ. Hence, the likelihood function is formu- >
: 4.95 >
;
lated as a function of this set of variables as given by Eq. (11), and
the first cycle of regression is performed to estimate the first set of
model parameters based on Eq. (12). The relationship between V s1 and CRR can be obtained by
Step 2: Following completion of the first cycle, chi-square test substituting Mw ¼ 7.5 and various probabilities of liquefaction.
results and p-values (the probability that the level of prediction of
the model would decrease if the explanatory variable provided zero
contribution to the model) are checked for the estimated model Table 2. Results of model parameter estimates from logistic regression
parameters for all the variables. The variable having the maximum analysis
p-value and minimum chi-square value is eliminated from the set of Parameter estimations (p < 0.05)
explanatory variables because a high p-value essentially indicates Candidate
that the associated variable makes the least contribution to the variable Trial 1 Trial 2 Trial 3 Trial 4
predictive capacity of the model. Hence, following elimination of V s1 −0.048 — — —
the variable with the highest p-value, a new regression analysis is V 2s1 — −1.48 × 10−4 — —
performed based on the new set of variables following Step 1. V 3s1 — — −3.88 × 10−7 —
Step 3: Steps 1 and 2 are repeated iteratively and the insignifi- V 4s1 — — — −1.3 × 10−9
cant variables are excluded from the prediction model after each Mw 2.02 2.01 1.6 1.52
lnðCSRÞ 3.046 5.11 4.95 5.152
cycle of regression based on the chi-square test and probability
Fig. 9. Comparison among different logistic regression solutions for 30% probability of liquefaction relative to liquefaction and no-liquefaction
points. Solid markers denote liquefaction sites, while open markers denote no-liquefaction sites.
Development of Magnitude Scaling Factor that the MSF model developed for gravelly soil falls about midway
between the Andrus and Stokoe (2000) curve at the high end and
While comparing the data points with the Cao et al. (2011) trigger- the Kayen et al. (2013) curve at the low end, both of which were
ing curves, the original CSRs for all the points with various M w developed for sandy soil based on V s1 . In addition, the MSF curve
values were converted to CSRM¼7.5 using Eq. (5), the lower-bound from this study falls just below the lower-bound range for MSF
MSF curve (Idriss 1999) recommended by the National Center for recommended by the NCEER/NSF liquefaction workshop (Youd
Earthquake Engineering Research (NCEER)/National Science et al. 2001) but slightly higher than the MSF curve proposed by
Foundation (NSF) workshop (Youd et al. 2001). However, the Idriss and Boulanger (2008). Hence, the MSF model developed
Idriss (1999) MSF equation was primarily developed for the in this study for gravelly soil appears to be reasonably consistent
liquefaction of sand, which could be a reason for the improper with existing MSF curves for sands.
liquefaction assessment of several gravel liquefaction case histories
using the Cao et al. (2011) triggering curves. In the new
liquefaction-triggering procedure described in the previous section,
the moment magnitude, M w , was treated as a separate variable as
shown in Eqs. (13) and (14). Therefore, as a part of this regression
analysis, a new MSF model was developed specifically for gravelly
soils that may help improve liquefaction assessment at some grav-
elly sites considered in this study. However, more data from other
earthquakes would still be desirable to refine the regression model
and further constrain the set of triggering curves.
To develop the MSF curve, CSR values were first obtained from
Eq. (14) for M w 5.5 through 9 with an increment of 0.5, keeping V s1
and PL constant. Then the CSRs for different magnitudes were
divided by CSRM¼7.5 to obtain the magnitude scaling factor. The
same process was then repeated for different values of V s1 and PL
in Eq. (14) to obtain the variation of MSF with these variables.
But, notably, the MSF pattern did not vary with the normalized
shear-wave velocity (V s1 ) and the probability of liquefaction (PL ).
Hence, the MSF was found to be formulated as a function of
magnitude with the best-fit exponential equation given by
MSF ¼ 10.667 expð−0.316M w Þ ð15Þ Fig. 10. Comparison of MSF curve from logistical regression analysis
of gravel liquefaction case histories based on V s with MSF curves
The MSF curve obtained with Eq. (15) is plotted and compared
proposed previously for sand.
with several other MSF curves in Fig. 10. The comparison shows
Fig. 11. Proposed liquefaction-triggering curves along with collected data points. Solid markers denote liquefaction sites, while open markers denote
no-liquefaction sites.
Based on the new MSF model [Eq. (15)], the CSRs for all the the no-liquefaction points from Argostoli, KNK, and Coyote Creek
case history data points were converted to CSRM¼7.5 and plotted played a significant role in constraining the lower branch of the
with the newly developed triggering curves obtained by Eq. (14), triggering curves to move upwards. Likewise, the triggering curves
as shown in Fig. 11. Fig. 11 shows that all the data points were at the higher range of V s1 have been tightened and slightly steep-
slightly relocated and the liquefaction potential of the marginal ened compared to the Cao et al. (2011) triggering curves by the
points evaluated as being relatively better compared to Fig. 8. This additional no-liquefaction data points from Chengdu, L’Aquila
fact is particularly applicable to the case histories at the Port and Valdez. However, more data points would certainly be desir-
of Wellington in New Zealand and at the port of Argostoli in able to better constrain the shape of the triggering curves in the high
Cephalonia, Greece, where liquefaction and no-liquefaction data range of V s1 .
points are now better distributed with respect to the 50% probability In the middle range of the triggering curves, there are a few no-
curve. Moreover, the larger database including the additional lique- liquefaction points from the Chengdu Plain that fall above the 50%
faction and no-liquefaction data points better constrained the trig- triggering curve and a few liquefaction points from the same region
gering curves by reducing the uncertainty at both the higher and that fall below the 30% triggering curve. Because of these points,
lower ranges of CSR. Therefore, the newly developed V s -triggering the set of triggering curves remains slightly sloped above the V s1
procedure in association with the new MSF model provides a much value of 200 m=s, causing several no-liquefaction points to fall
improved alternative for evaluating the liquefaction potential of marginally on the 30% triggering curve instead of falling distinctly
gravelly deposits. below the line. These points may belong to the false negative or
false positive categories, as explained previously, leading to incon-
sistent evaluation of the actual incident. These points might also
Comparison with Cao et al. (2011) Triggering Curves
be governed by some other in situ parameters, for example, the
The newly developed probabilistic triggering curves with liquefac- permeability of the soil strata, level of FC, and the presence of
tion probabilities of 15% to 85% are plotted in Fig. 12(a) with solid an impervious cap layer, that were not included in the regression
lines along with the similar curves developed by Cao et al. (2011) analysis due to a lack of information available for each site. These
marked as dashed lines to allow for a distinct comparison between points remain a source of uncertainty in the overall liquefaction-
the two triggering procedures. For lower values of V s1 (around triggering procedure.
150 m=s), the CRR for the new 50% probability of liquefaction As shown in Fig. 12(a), for V s1 values above 200 m=s, the PL ¼
curve is about 0.10, while it is only 0.04 for the Cao et al. (2011) 50% curve for the new regression is very similar to that for the Cao
curves. This adjustment produces much better agreement with et al. (2011) regression. However, the addition of new liquefaction
observed field performance. These higher CRR values at lower points has pulled the new PL ¼ 85% curve to the right, while the
shear-wave velocities are also more typical of those predicted by addition of no-liquefaction data points has pulled the new PL ¼
the V s -based triggering curves developed by Kayen et al. (2013) 15% curve to the left, relative to the Cao et al. (2011) curves. Mov-
for sands. In fact, it can be seen from Fig. 11 that the marginal ing the new PL ¼ 15% curve to the left is particularly significant
liquefaction and no-liquefaction points from the Port of Wellington because this curve is often recommended for deterministic evalu-
for the 2013 Cook Strait and Lake Grassmere events of M w ¼ 6.6, ations (Kayen et al. 2013). However, the slope of the new set of
Boulanger, R. W., and I. M. Idriss. 2014. CPT and SPT based liquefaction
triggering procedures. Rep. No. UCD/CGM-14/01. Davis, CA: Center
Data Availability Statement of Geotechnical Modeling, Univ. of California-Davis.
Cao, Z., T. L. Youd, and X. Yuan. 2011. “Gravelly soils that liquefied
Some or all data, models, or code that support the findings of this during 2008 Wenchuan, China earthquake, Ms = 8.0.” Soil Dyn. Earth-
study are available from the corresponding author upon reasonable quake Eng. 31 (8): 1132–1143. https://doi.org/10.1016/j.soildyn.2011
.04.001.
request.
Cao, Z., T. L. Youd, and X. Yuan. 2013. “Chinese dynamic penetration test
for liquefaction evaluation in gravelly soils.” J. Geotech. Geoenviron.
Eng. 139 (8): 1320–1333. https://doi.org/10.1061/(ASCE)GT.1943
Acknowledgments -5606.0000857.
Cao, Z., X. Yuan, T. L. Youd, and K. M. Rollins. 2012. “Chinese dynamic
Funding for this study was provided by Grant G16AP00108 from
penetration tests (DPT) at liquefaction sites following 2008 Wenchuan
the USGS Earthquake Hazard Reduction Program and Grants earthquake.” In Proc., 4th Int. Conf. on Geotechnical and Geophysical
CMMI-1663546 and CMMI- 1663288 from the National Science Site Characterization, 1499–1504. London: Taylor & Francis Group.
Foundation. This funding is gratefully acknowledged. This work is Chang, W. J. 2016. “Evaluation of liquefaction resistance for gravelly sands
also part of a research project funded by ReLUIS (University using gravel content–corrected shear-wave velocity.” J. Geotech. Geo-
Network of Seismic Engineering Laboratories) Consortium 2019– environ. Eng. 142 (5): 04016002. https://doi.org/10.1061/(ASCE)GT
2021, WP 16 Geotechnical Engineering, Task 16.1: Site response .1943-5606.0001427.
and liquefaction. However, the opinions, conclusions, and recom- Chen, L., X. Yuan, Z. Cao, R. Sun, W. Wang, and H. Liu. 2018. “Character-
mendations in this paper do not necessarily represent those of the istics and triggering conditions for naturally deposited gravelly soils
sponsors. We also express sincere appreciation to Luca Minarelli that liquefied following the 2008 Wenchuan M w 7.9 earthquake, China.”
for help in arranging access to geophysical surveys at sites in Italy Earthquake Spectra 34 (3): 1091–1111. https://doi.org/10.1193
and to G. Athanasopoulos for arranging access to these tests at sites /032017EQS050M.
in Cephalonia, Greece. Chu, B. L., S. C. Hsu, S. E. Lai, and M. J. Chang. 2000. Soil liquefaction
potential assessment of the Wu Feng area after the 921 Chi-Chi earth-
quake. Report of National Science Council. [In Chinese.] Taipei,
Taiwan: National Research Council of Taiwan.
Supplemental Materials
Coulter, H. W., and R. R. Migliaccio. 1966. Effect of the earthquake of
March 22, 1964 at Valdez, Alaska. US Geological Survey Professional
Table S1 is available online in the ASCE Library (www.ascelibrary
Paper 542-C. Washington, DC: USGS.
.org).
Cubrinovski, M., J. Bray, C. de la Torre, M. Olsen, B. Bradley, G. Chiaro,
E. Stocks, and L. Wotherspoon. 2017. “Liquefaction effects and asso-
ciated damages observed at the Wellington CentrePort from the 2016
References Kaikōura earthquake.” Bull. N.Z. Soc. Earthquake Eng. 50 (2): 152–173.
https://doi.org/10.5459/bnzsee.50.2.152-173.
Amoroso, S., et al. 2021. “Liquefaction field reconnaissance following the
Cubrinovski, M., J. D. Bray, C. de la Torre, M. Olsen, B. Bradley,
29th December 2020 M w 6.4 Petrinja earthquake (Croatia).” In Proc.,
G. Chiaro, E. Stocks, L. Wotherspoon, and T. Krall. 2018. “Liquefac-
EGU General Assembly Conf. Abstracts 2021.
tion-induced damage and CPT characterization of the reclamations at
Andrus, R. D. 1994. “In situ characterization of gravelly soils that liquefied
in the 1983 Borah Peak earthquake.” Ph.D. dissertation, Dept. of Civil CentrePort, Wellington.” Bull. Seismol. Soc. Am. 108 (3B): 1695–1708.
Engineering, Univ. of Texas at Austin. https://doi.org/10.1785/0120170246.
Andrus, R. D., and K. H. Stokoe II. 2000. “Liquefaction resistance of DeJong, J. T., M. Ghafghazi, A. P. Sturm, D. W. Wilson, J. den Dulk, R. J.
soils from shear-wave velocity.” J. Geotech. Geoenviron. Eng. 126 (11): Armstrong, A. Perez, and C. A. Davis. 2017. “Instrumented Becker
1015–1025. https://doi.org/10.1061/(ASCE)1090-0241(2000)126:11 penetration test. I: Equipment, operation, and performance.” J. Geotech.
(1015). Geoenviron. Eng. 143 (9): 04017062. https://doi.org/10.1061/(ASCE)
Arai, H., and K. Tokimatsu. 2005. “S-wave velocity profiling by joint GT.1943-5606.0001717.
inversion of microtremor dispersion curve and horizontal-to-vertical Dobry, R., R. Ladd, F. Yokel, R. Chung, and D. Powell. 1982. Prediction of
(H/V) spectrum.” Bull. Seismol. Soc. Am. 95 (5): 1766–1778. https://doi pore water pressure buildup and liquefaction of sands during earth-
.org/10.1785/0120040243. quakes by the cyclic strain method. NBS Building Science Series 138.
Athanasopoulos-Zekkos, A., D. Zekkos, K. M. Rollins, J. Hubler, Washington, DC: National Bureau of Standards.
J. Higbee, and A. Platis. 2019. “Earthquake performance and charac- Faenza, L., and A. Michelini. 2011. “Regression analysis of MCS intensity
terization of gravel-size earth fills in the ports of Cephalonia, Greece, and ground motion spectral accelerations (SAs) in Italy.” Geophys. J.
following the 2014 earthquakes.” In Proc., Earthquake Geotechnical Int. 186 (3): 1415–1430. https://doi.org/10.1111/j.1365-246X.2011
Engineering for Protection and Development of Environment and Con- .05125.x.
structions, Proc. 7th Intl. Conf. on Earthquake Geotechnical Engineer- Fäh, D., F. Kind, and D. Giardini. 2003. “Inversion of local S-wave velocity
ing, 1212–1219. London: Taylor & Francis Group. structures from average H/V ratios, and their use for the estimation of
comparison of surface wave methods.” Soil Dyn. Earthquake Eng. quake.” In Proc., 3rd Int. Conf. on Recent Advances in Geotechnical
82 (3): 222–240. https://doi.org/10.1016/j.soildyn.2015.12.010. Earthquake Engineering and Soil Dynamics, 235–242. St. Louis:
Garofalo, F., et al. 2016b. “InterPACIFIC project: Comparison of invasive Missouri Univ. of Science and Technology.
and non-invasive methods for seismic site characterization. Part II: Kokusho, T., and Y. Yoshida. 1997. “SPT N-value and S-wave velocity for
Inter-comparison between surface-wave and borehole methods.” Soil gravelly soils with different grain size distribution.” Soils Found. 37 (4):
Dyn. Earthquake Eng. 82 (3): 241–254. https://doi.org/10.1016/j 105–113. https://doi.org/10.3208/sandf.37.4_105.
.soildyn.2015.12.009. Kozeny, J. 1927. “Ueber kapillare Leitung des Wassers im Boden.”
Golesorkhi, R. 1989. “Factors influencing the computational determination Sitzungsber. Akad. Wiss. Wien 136 (2): 271–306.
of earthquake-induced shear stresses in sandy soils.” Ph.D. dissertation, Kramer, S. L. 1996. Geotechnical earthquake engineering, 653. Hoboken,
Dept. of Civil and Environmental Engineering, Univ. of California. NJ: Prentice-Hall.
Harder, L. F. 1997. “Application of the Becker penetration test for Liao, S. S. C., D. Veneziano, and R. V. Whitman. 1988. “Regression
evaluating the liquefaction potential of gravelly soils.” In Proc., models for evaluating liquefaction probability.” J. Geotech. Geoen-
NCEER Workshop on Evaluation of Liquefaction Resistance. Buffalo, viron. Eng. 114 (4): 389–411. https://doi.org/10.1061/(ASCE)0733
NY: National Center for Earthquake Engineering Research. -9410(1988)114:4(389).
Harder, L. F., Jr., and H. B. Seed. 1986. Determination of penetration Liao, S. S. C., and R. V. Whitman. 1986. “Overburden correction factors
resistance for coarse-grained soils using the Becker hammer drill. for SPT in sand.” J. Geotech. Eng. 112 (3): 373–377. https://doi.org/10
Rep. No. UCB/EERC-86/06. Berkeley, CA: College of Engineering, .1061/(ASCE)0733-9410(1986)112:3(373).
Univ. of California.
Lin, P., C. Chang, and W. Chang. 2004. “Characterization of liquefaction
Hubler, J., A. Athanasopoulos-Zekkos, and D. Zekkos. 2017. “Monotonic,
resistance in gravelly soil: Large hammer penetration test and shear
cyclic, and postcyclic simple shear response of three uniform gravels in
wave velocity approach.” Soil Dyn. Earthquake Eng. 24 (9–10):
constant volume conditions.” J. Geotech. Geoenviron. Eng. 143 (9):
675–687. https://doi.org/10.1016/j.soildyn.2004.06.010.
04017043. https://doi.org/10.1061/(ASCE)GT.1943-5606.0001723.
Lodge, A. L. 1994. “Shear wave velocity measurements for subsurface
Hubler, J., A. Athanasopoulos-Zekkos, and D. Zekkos. 2018. “Monotonic
characterization.” Ph.D. dissertation, Dept. of Civil and Environmental
and cyclic simple shear response of gravel-sand mixtures.” Soil Dyn.
Engineering, Univ. of California.
Earthquake Eng. 115 (Dec): 291–304. https://doi.org/10.1016/j.soildyn
Lopez, J. S., X. Vera-Grunauer, K. Rollins, and G. Salvatierra. 2018.
.2018.07.016.
“Gravelly soil liquefaction after the 2016 Ecuador earthquake.” In
Idriss, I., and R. W. Boulanger. 2008. Soil liquefaction during earthquakes.
Proc., Conf. on Geotechnical Earthquake Engineering and Soil
Oakland, CA: Earthquake Engineering Research Institute.
Dynamics V, 273–285. Reston, VA: ASCE.
Idriss, I. M. 1999. “An update to the Seed-Idriss simplified procedure
Maurenbrecher, P. M., A. Den Outer, and H. J. Luger. 1995. “Review of
for evaluating liquefaction potential.” In Proc., TRB Workshop on
New Approaches to Liquefaction. Washington, DC: Federal Highway geotechnical investigations resulting from the Roermond April 13, 1992
Administration. earthquake.” In Proc., 3rd Int. Conf. on Recent Advances in Geotech-
nical Earthquake Engineering and Soil Dynamics, 645–652. Reston,
Ishihara, K. 1985. “Stability of natural deposits during earthquakes.” In
Vol. 1 of Proc., 11th Int. Conf. on Soil Mechanics and Foundation VA: ASCE.
Engineering, 321–376. Accord, MA: A.A. Balkema. McCulloch, D. S., and M. G. Bonilla. 1970. Effects of the earthquake of
Jamiolkowski, M., and D. C. F. Lo Presti. 1990. “Correlation between March 27, 1964, on the Alaska Railroad. Washington, DC: US
liquefaction resistance and shear wave velocity.” Soils Found. 32 (2): Government Printing Office.
145–148. https://doi.org/10.3208/sandf1972.30.2_33. Morales, C., C. Ledezma, E. Saez, S. Boldrini, and K. M. Rollins. 2020.
Juang, C. H., C. J. Chen, and T. Jiang. 2001. “Probabilistic framework “Seismic failure of an old pier during the 2014 M w 8.2, Pisagua, Chile
for liquefaction potential by shear wave velocity.” J. Geotech. Geoen- earthquake.” Earthquake Spectra 36 (2): 880–903. https://doi.org/10
viron. Eng. 127 (8): 670–678. https://doi.org/10.1061/(ASCE)1090 .1177/8755293019891726.
-0241(2001)127:8(670). Nakamura, Y. 1989. “A method for dynamic characteristics estimation of
Juang, C. H., T. Jiang, and R. D. Andrus. 2002. “Assessing probability subsurface using microtremor on the ground surface.” Quarterly Rep.
based methods for liquefaction evaluation.” J. Geotech. Geoenviron. Railway Techn. Res. 30 (1): 25–33.
Eng. 128 (7): 580–589. https://doi.org/10.1061/(ASCE)1090-0241 Nikolaou, S., D. Zekkos, D. Assimaki, and R. Gilsanz. 2014. “GEER/
(2002)128:7(580). EERI/ATC earthquake reconnaissance January 26th/February 2nd 2014
Kayen, R., R. E. S. Moss, E. M. Thompson, R. B. Seed, K. O. Cetin, A. Der Cephalonia, Greece Events, version 1.” Accessed November 18, 2021.
Kiureghian, Y. Tanaka, and K. Tokimatsu. 2013. “Shear-wave velocity– http://www.geerassociation.org/index.php/component/geer_reports/?view=
based probabilistic and deterministic assessment of seismic soil lique- geerreports&id=32.
faction potential.” J. Geotech. Geoenviron. Eng. 139 (3): 407–419. Nogoshi, M., and T. Igarashi. 1971. “On the amplitude characteristics of
https://doi.org/10.1061/(ASCE)GT.1943-5606.0000743. microtremor (part 2).” J. Seismol. Sco. Japan 24 (1): 26–40.
Kayen, R., E. Y. Tanaka, T. Kishida, and S. Sugimoto. 2002. “Liquefaction Ohta, Y., and N. Goto. 1978. “Physical background of the statistically ob-
potential of native ground in West Kobe, Japan by the spectral analysis tained S-wave velocity equation in terms of soil indexes.” [In Japanese.]
of surface waves (SASW) method.” In Proc., 8th US-Japan Workshop ButsuriTanko(Geophys. Explor.) 31 (1): 8–17.
on Earthquake Resistant Design of Lifeline Facilities and Counter- Park, C. B., R. D. Miller, and J. Xia. 1999. “Multichannel analysis of sur-
measures against Liquefaction. Buffalo, NY: Multidisciplinary Center face waves.” Geophysics 64 (3): 800–808. https://doi.org/10.1190/1
for Earthquake Engineering Research. .1444590.