Professional Documents
Culture Documents
Liquefaction–Triggering Procedure
Ross W. Boulanger, M.ASCE1; and I. M. Idriss, Dist.M.ASCE2
Abstract: A probabilistic version of the Idriss and Boulanger standard penetration test (SPT)–based liquefaction triggering correlation is
Downloaded from ascelibrary.org by University of Nebraska-Lincoln on 04/07/13. Copyright ASCE. For personal use only; all rights reserved.
derived using a maximum likelihood approach and an updated case history database. Measurement and estimation uncertainties in the cyclic
stress ratio (CSR) and SPT (N1)60cs values and the effects of the choice-based sampling bias in the case history database are taken into account.
The results of sensitivity analyses show that the position of the most likely triggering curve is well constrained by the case history data and that
the magnitude of the total error term is also reasonably constrained. The most likely value for the SD of the error term in the triggering correlation
is, however, found to be dependent on the uncertainties assigned to the CSR and (N1)60cs. The results of the sensitivity study appear to provide
reasonable bounds on the effects of different interpretations on the positions of the triggering curves for various probabilities of liquefaction.
Methods for including model and parameter uncertainties in probabilistic liquefaction analyses are briefly discussed. The derived triggering
correlation is compared with relationships developed from cyclic laboratory test results for specimens obtained using frozen sampling
techniques. DOI: 10.1061/(ASCE)GT.1943-5606.0000700. © 2012 American Society of Civil Engineers.
CE Database subject headings: Soil liquefaction; Earthquakes; Cyclic loads; Penetration tests; Probability.
Author keywords: Liquefaction; Earthquakes; Cyclic loads; Standard penetration test; Probability.
Fig. 1. SPT-based probabilistic correlations for the CRR of sands for M 5 7.5 and s9v5 1 atm: (a) Toprak et al. (1999); (b) Cetin et al. (2004)
means that a number of generalized assumptions and approx- it is computed for a specific earthquake magnitude (moment mag-
imations regarding dominant sources of uncertainty and their nitude, M) and s9v at depth z. The value of t max can be estimated from
distributions across the case history database are necessary as part dynamic response analyses, but such analyses must include a suf-
of the probabilistic analysis. The remainder of this section describes ficient number of input acceleration time series and adequate site
the liquefaction and probabilistic analysis frameworks, including the characterization details to be reasonably robust. Alternatively, the
various assumptions regarding the primary sources of uncertainty maximum shear stress can be estimated using the Seed and Idriss
and their distributions. Details of the underlying case history data- (1971) simplified procedure
base are provided separately in Idriss and Boulanger (2010) and
sv amax
Boulanger et al. (2012). Sensitivity analyses are used in the fol- CSRM;s9v ¼ 0:65 rd ð2Þ
lowing section to evaluate the effects that these various assumptions s9v g
and approximations may have had on the resulting probabilistic
liquefaction triggering correlations. where sv 5 vertical total stress at depth z; amax/g 5 maximum
horizontal acceleration (as a fraction of gravity) at the ground
surface; and rd 5 shear stress reduction factor that accounts for the
Liquefaction Analysis Framework
dynamic response of the soil profile.
The stress-based approach for evaluating the potential for liquefaction The soil’s CRR is affected by the duration of shaking [which is
triggering, developed by Seed and Idriss (1967), has been used widely expressed through an earthquake magnitude scaling factor (MSF)]
for the last 45 years (e.g., Seed and Idriss 1971; Shibata 1981; and effective overburden stress (which is expressed through a Ks
Tokimatsu and Yoshimi 1983; NRC 1985; Seed et al. 1985; Youd factor). These effects are accounted for in the processing of case
et al. 2001; Cetin et al. 2004; Idriss and Boulanger 2004). The basic histories by adjusting the earthquake-induced CSR to a reference
framework of this approach compares the earthquake-induced CSR M 5 7.5 and s9v 5 1 atm, using the following expression:
with the cyclic resistance ratios (CRRs) of the soil. The components of
sv amax
this framework, as briefly summarized in the following, were con- CSRM57:5; sv951 atm ¼ 0:65 rd 1 1 ð3Þ
structed to provide a rational treatment of the various factors that affect s9v g MSF Ks
penetration resistance and cyclic resistance.
The earthquake-induced CSR, at a given depth within the soil- The soil’s CRR is also affected by the presence of sustained static
profile, is usually expressed as a representative value (or equivalent shear stresses, such as may exist beneath foundations or within
uniform value) equal to 65% of the maximum cyclic shear stress slopes. This effect, which is expressed through a Ka factor, is
ratio generally small for nearly level ground conditions. It was not in-
cluded in the updated case history database by Idriss and Boulanger
t max (2010) because the case histories are dominated by level or nearly
CSRM;s9v ¼ 0:65 ð1Þ
s9v level ground conditions.
The soil’s CRR is usually correlated to an in situ parameter, such
where t max 5 maximum earthquake-induced shear stress; s9v 5 as SPT blow count (number of blows per foot), CPT penetration
vertical effective stress; and the subscripts on the CSR indicate that resistance, or shear wave velocity, Vs. SPT blow counts are affected
correction factor for using split spoons with room for liners but
with the liners absent, and Nm 5 measured SPT blow count. The The uncertainties in the limit state function are represented by
factors CB and CS are set equal to unity if standard procedures are three contributors in the present analysis. Measurement or estima-
followed. tion uncertainties in the case history data points are assumed to be
The correlation of CRR to (N1)60 is affected by the soil’s fines adequately represented by including uncertainties in the (N1)60cs and
content (FC). For mathematical convenience, this effect can be CSRM57.5, s951 atm values. The uncertainty in (N1)60cs is represented
expressed in terms of an equivalent clean-sand (N1)60cs, which is by inclusion of a random model error term that is assumed to be
obtained using the following expression: normally distributed with a constant coefficient of variation (COVN)
(e.g., Orchant et al. 1988; Baecher and Christian 2003) and a mean of
ðN1 Þ60cs ¼ ðN1 Þ60 þ DðN1 Þ60 ð5Þ zero, which is consistent with the listing of mean (N1)60cs values in
the database by Idriss and Boulanger (2010). The uncertainty in
CRR can then be expressed in terms of (N1)60cs CSRM57.5, s951 atm is assumed to be log-normally distributed, which
is consistent with log-normal distributions for the uncertainty in
CRRM57:5; s9v 51 atm ¼ f ðN1 Þ60cs ð6Þ predictions of peak ground accelerations (e.g., Abrahamson et al.
2008). Uncertainty in the CRRM57.5, s951 atm expression is repre-
The SPT-based liquefaction triggering procedure by Idriss and sented by inclusion of a random model error term, which is also
Boulanger (2008) used the previous stress-based framework, as have assumed to be log-normally distributed with mean of zero.
many other similar procedures. The development of the various The uncertainty in the representative (N1)60cs value assigned to any
functions [i.e., rd, CN, CR, Ks, MSF, and D(N1)60] and the details of case history includes contributions from two major sources. One source of
the case history database used herein are described in Idriss and uncertainty is variability in the SPT equipment and procedures used at
Boulanger (2008) and Idriss and Boulanger (2010). The logic for different case history sites. The second major source of uncertainty is the
selecting representative (N1)60cs values for individual case histories degree to which the available SPT data are truly representative of the
is illustrated with examples in Idriss and Boulanger (2010); the critical strata, which depends on the degree to which the geologic con-
reported (N1)60cs values are mean values for the critical stratum, ditions are understood, the heterogeneity of the deposits, the number of
which are consistent with their treatment in the analyses described in borings, and the placement of the borings relative to the strata of concern.
this paper. These materials are essential components of the prob- Either of these sources of uncertainty can dominate the total uncertainty in
abilistic liquefaction triggering relationship presented in this paper the selection of a representative (N1)60cs value for a given site. The
but are not repeated here for brevity. majority of the liquefaction case histories lack sufficient information to
justify attempting to develop site-specific estimates of the magnitudes and
distributions of these uncertainties for each case history. In addition, the
Limit State Function and Assumed Distributions literature does not provide a clear basis for refined estimates of uncer-
The model for the limit state function (g) was taken as the dif- tainties across the range of testing methods and geologic environments
ference between the natural logs of the CRRM57.5, s951 atm and represented in the case history database without site specific data (e.g.,
CSRM57.5, s951 atm values, such that liquefaction is assumed to Baecher and Christian 2003; Phoon 2008). For these reasons, it was
have occurred if g # 0 and to have not occurred if g . 0. The assumed that the uncertainty in the (N1)60cs values in the liquefaction
CRRM57.5, s951 atm value was estimated using the following form database could be approximated, at least to a first order, using a normal
of the Idriss and Boulanger [2008; Eq. (70)] relationship distribution with the same value of COVN for all case histories.
The uncertainty in the CSRM57.5, s951 atm values estimated for
( any case history similarly depends on numerous factors, including
ðN1 Þ60cs ðN1 Þ60cs 2
CRRM57:5; s9v 51 atm ¼ exp þ the proximity of strong ground motion recordings, spatial variability
14:1 126 in ground motions, and availability and quality of indirect measures
) of shaking levels (e.g., damage to structures, disruption of non-
ðN1 Þ60cs 3 ðN1 Þ60cs 4
2 þ 2 Co ð7Þ structural contents). The estimates of amax at liquefaction and
23:6 25:4 no-liquefaction sites by various researchers are often based on a
combination of these types of information, and can be considered to
where Co 5 unknown fitting parameter that serves to scale the Idriss- have smaller variances than estimates obtained from ground motion
Boulanger relationship while maintaining its shape; note that the prediction equations alone. The uncertainty in estimates of amax for
Idriss-Boulanger deterministic relationship corresponds to Co 5 2.8. each case history depends on the quality of information available,
The use of a single fitting parameter provides the means for ex- but it was found that quantifying these uncertainties on a case-history
amining the uncertainty in the previously noted relationship; the specific basis was generally not justified, except for those few cases
effect of including additional fitting parameters that modify the that had a strong ground motion recording directly at the site. For this
shape of the relationship is evaluated later. reason, the SD in log(CSRM57.5, s951 atm) was set to one of two
^ þ «N
N ¼ N ð13Þ where F 5 standard normal cumulative probability function. For
example, the probability of having observed liquefaction becomes
^ .0.84 if the case history data point plots more than one sT above the
sN ¼ COVN × N ð14Þ
triggering curve. In this regard, it is important to recognize that case
history data points are plotted at the CSRM57.5, s951 atm value ex-
lnðSÞ ¼ ln ^
S þ «lnðSÞ ð15Þ pected in the absence of liquefaction and that this CSRM57.5, s951 atm
value may be significantly greater than the value attained if liq-
uefaction was triggered early in strong shaking. For this reason, the
^ þ «lnðRÞ
lnðRÞ ¼ ln R ð16Þ
case history data points that fall well above the triggering curve
have probabilities close to unity, and thus, they have very little
The corrected limit state function with inclusion of the uncer-
influence on the overall likelihood function. The reverse is true for
tainties can then be written as
the no-liquefaction cases. The likelihood function can now be
gðN; S; Co Þ ¼ lnðRÞ 2 lnðSÞ ð17Þ written as
" L Co ; «lnðRÞ
^ þ «N
N ^ þ «N 2
N
gðN; S; Co Þ ¼ þ ¼ ∏ P½gðN; S; Co Þ # 0 ∏ P½gðN; S; Co Þ . 0
14:1 126
Liquefied sites Nonliquefied sites
#
N^ þ «N 3 ^ þ «N 4
N ð23Þ
2 þ 2 Co
23:6 25:4
L Co ; «lnðRÞ
þ «lnðRÞ 2 ln ^
S 2 «lnðSÞ ð18Þ
" # " #
g ^ ^
^ N; S; Co ^ ^ ^
g N; S; Co
¼ ∏ F 2 ∏ F
Liquefied sites sT Nonliquefied sites sT
This expression can be simplified by multiplying out the poly-
nomial terms and then neglecting the higher-order terms with «N ð24Þ
squared, cubed, or raised to the fourth power. The resulting ex-
pression is where ∏ 5 product of the sequence of terms.
" # The case history database is, however, believed to contain an
2
N^ N^ N^ 3 ^ 4
N uneven sampling of liquefaction and no-liquefaction case histories
gðN; S; Co Þ ¼ þ 2 þ 2 Co (relative to their true populations in the field) because researchers
14:1 126 23:6 25:4
more often choose to investigate liquefaction sites. Manski and
1 ^
2N 3N^2 ^3
4N Lerman (1977) suggested that the bias from an uneven choice-based
þ þ 2 þ «N sampling process could be corrected by weighting the observations
14:1 1262 23:63 25:44
to better represent the actual population. Cetin et al. (2002) noted that
þ «lnðRÞ 2 ln ^S 2 «lnðSÞ ð19Þ this amounted to rewriting the likelihood function as
solution for these first four cases is plotted versus N in Fig. 2; the
1 2 Qliq;true results show that the rate at which sT increases with N is dependent
wnonliquefied ¼ ð27Þ
1 2 Qliq;sample on the assumed value of COVN, but that the overall values for sT are
relatively similar for all four cases. Because the assumed values for
where Qliq,true 5 true proportion of the occurrences of liquefaction in the uncertainties in S and N are increased, the most likely value for
the population; and Qliq,sample 5 proportion of occurrences of liq- the uncertainty in R is decreased. For the fourth case, the values of
uefaction in the sample set. Cetin et al. (2002) reported that a panel of sln(S) 5 0.25 and COVN 5 0.20 are sufficiently large that the most
eight experts agreed that the ratio wnonliquefied/wliquefied should be likely solution is a unique relationship for R (i.e., sln(R) 5 0). In
.1.0 and ,3.0, with the most common estimate being between 1.5 addition, the differences in the values of the likelihood function for
and 2.0. They further allowed this ratio to be a parameter in the these first three cases were not significant, such that the maximum
Bayesian updating analyses, and found that a ratio of 1.5 minimized likelihood solution did not provide strong support for one case over
their overall model variance. Accordingly, they adopted weighting another. These results illustrate how the maximum likelihood
values of wnonliquefied 5 1.2 and wliquefied 5 0.8, producing the ratio analysis of the case history data provides insight on the total un-
wnonliquefied/wliquefied 5 1.5. It is not clear how these weighting values certainty, but does not provide clear guidance on the appropriate
can currently be estimated for analysis of the liquefaction case partitioning of that uncertainty into the components of N, S, and R.
history database except by judgment, and in this regard, the values The four cases, however, give practically the same values for the
adopted by Cetin et al. (2002) appear reasonable. fitting parameter Co (i.e., 2.66–2.67), such that the most likely
position of the triggering curve is well constrained.
Results of Parameter Estimation The last two cases listed in Table 1 vary the ratio wnonliquefied/
wliquefied from 1.2 to 1.8 (bracketing the value of 1.5 adopted by Cetin
et al. 2002), while keeping sln(S) and COVN constant. The smaller
Results and Sensitivity to Assumed Parameter weighting ratio causes the solution for the most likely triggering
Estimation Errors curve to shift slightly downward (i.e., Co increases from 2.66 to
There are five parameters that can be either estimated or left as fitting 2.70), whereas the greater weighting ratio causes the most likely
parameters in determining the maximum likelihood solution; these triggering curve to shift slightly upward (i.e., Co decreases to 2.64).
parameters are Co, sln(R), sln(S), COVN, and wnonliquefied/wliquefied. The solution for sln(R) is relatively unaffected by the choice of
Cetin et al. (2002) estimated uncertainties in S and N for the in- weighting ratio between 1.2 and 1.8.
dividual case histories that were used in their Bayesian analyses and Curves for probabilities of liquefaction (PL) equal to 15% and
then suggested that sln(S) would be about 0.2 and COVN would be 50%, with inclusion of the estimation errors in CSRM57.5, s951 atm
about 0.15 in applications with good practices. Others have sug- and (N1)60cs, for the six cases listed in Table 1 are plotted together
gested that COVN can range from 0.15 to 0.50, depending on the with the case history data in Fig. 3. Fig. 3 illustrates how the median
quality of the data (e.g., Orchant et al. 1988; Phoon and Kulhawy curves (PL 5 50%) from the six solution cases listed in Table 1 are
1999). Ground motion prediction equations have SDs of about almost identical. In addition, the PL 5 15% curves are also very
0.45–0.55 in the natural log of the peak ground acceleration similar because all six solution cases have similar total error terms,
(Abrahamson et al. 2008), which suggest that sln(S) could be around as illustrated in Fig. 2.
0.45–0.55, if it was estimated solely on the basis of a ground motion The triggering curves for a probability of liquefaction (PL) equal
prediction equation; smaller values of sln(S) would be expected for to 15% based on model uncertainty alone (i.e., including only the
most case histories given the additional information provided by uncertainty in CRRM57.5, s951 atm) for the six solution cases listed
strong ground motion recordings and site-specific observations (e.g., in Table 1 are plotted together with the case history data in Fig. 4.
damage to structures, disruption of nonstructural contents). As These curves do not include the estimation errors in S or N and, thus,
previously discussed, the data available for most sites in the liq- are located higher than the PL 5 15% curves in Fig. 3. The highest
uefaction database are inadequate to quantify the site-specific un- PL 5 15% curve in Fig. 4 is from Case 4, which is identical to the
certainty in N or S, and thus, the approach adopted in this study was PL 5 50% curve for this case because sln(R) 5 0. The other PL 5 15%
to solve for Co and sln(R) using a range of estimated values for sln(S), curves in Fig. 4 are closely spaced. The deterministic triggering
COVN, and wnonliquefied/wliquefied. correlation recommended by Idriss and Boulanger (2004, 2008)
Maximum likelihood solutions for Co and sln(R) are listed in essentially overlies the PL 5 15% curves for these five cases.
Table 1 for six cases with different assumptions regarding the values
Additional Sensitivity Studies
for sln(S), COVN, and wnonliquefied/wliquefied. The first four cases in-
volve varying the values for sln(S) from 0.15 to 0.25 and COVN from Sensitivity of the results to the form of the liquefaction triggering
0.15 to 0.20 while keeping wnonliquefied/wliquefied 5 1.5. For the seven equation was evaluated by introducing additional flexibility into the
sites that had strong ground motion recordings at the site, a reduced expression for CRRM57.5, s951 atm as
Fig. 2. SD in the total error term (sT) and CRR relationship (sln(R)) from the maximum likelihood solution for different estimates of sln(S) and COVN
Fig. 3. Curves of CRRM 5 7.5, s9v 5 1 atm versus (N1)60cs for probabilities of liquefaction of 15 and 50% with inclusion of estimation errors in CSR
and (N1)60cs; solutions for the six cases listed in Table 1
(
ðN1 Þ60cs ðN1 Þ60cs 2 in Fig. 5 for the case where sln(S) 5 0.2, COVN 5 0.15, and
CRRM57:5; s9v 51 atm ¼ exp þ wnonliquefied/wliquefied 5 1.5 (i.e., the same assumptions as for Case 2
14:1 126
in Table 1). The median triggering curve is almost equal to the
)
ðN1 Þ60cs 3 ðN1 Þ60cs 4 deterministic curve at (N1)60cs values ,10, and becomes increas-
2 þ 2 Co ingly higher than the deterministic curve with increasing values
23:6 C4
of (N1)60cs. This result indicates that the currently available case
ð28Þ history data would, on their own, support a triggering curve that
turns sharply upward at values of (N1)60cs smaller than does the
where the additional fitting parameter C4 controls the sharpness with deterministic curve. This issue is discussed in the following section,
which the liquefaction triggering relationship curves upward at along with a comparison of these relationships to those constructed
higher (N1)60cs values. The median triggering curve was found to be based on the results of cyclic laboratory tests on specimens obtained
relatively insensitive to the assumed values for the uncertainties in using frozen sampling techniques.
S and N or to the weighting factors, as was demonstrated for the The results obtained with the original form of the liquefaction trig-
original form of the liquefaction triggering equation in the previous gering equation were also checked for their sensitivity to the use of
section. Triggering curves for PL 5 15%, 50%, and 85% are shown the 50th percentile rather than 67th percentile rd values (Idriss and
Fig. 4. Curves of CRRM 5 7.5, s9v 5 1 atm versus (N1)60cs for probabilities of liquefaction of 15% excluding uncertainties in CSR and (N1)60cs; solutions
for the six cases listed in Table 1
Fig. 5. Curves of CRRM57.5, s9v51atm versus (N1)60cs for probabilities of liquefaction of 15, 50, and 85% using the expression for
CRRM57.5, s9v51atm with two fitting parameters
Boulanger 2010). The relationships originally proposed by Idriss (1999) 1.5 (i.e., the same assumptions as for Case 2 in Table 1), the derived
corresponded to about the 67th percentile rd values from those analyses, fitting parameter Co was increased from 2.67 to 2.69, and sln(R) was
with this percentile chosen so that the curve for M 5 7.5 was consistent unchanged at 0.15. This change in Co lowered the triggering curves by
with the earlier average rd curve recommended by Seed and Idriss 2% [i.e., exp(2.67–2.69) 5 0.98].
(1971). The median rd values were slightly lower, with the difference
progressively increasing from 0% at the ground surface to about 5%
Recommended Relationships
at a depth of 10 m for all M. The case histories were reprocessed with
the median rd values, and the probabilistic triggering relationships Selecting the most appropriate values for Co and sln(R) from the
recomputed. The median triggering curve was again relatively results of these maximum likelihood solutions involves subjective
insensitive to the assumed values for the uncertainties in S and N or to evaluation of the most appropriate partitioning of the total un-
the weighting factors, as was demonstrated in the previous section. certainty in the liquefaction case history database. This evaluation
Furthermore, the median triggering curve was found to be only must also consider the limitations of the statistical models and case
slightly lowered by the use of median rd values, as expected. For ex- history database, including uncertainties that are not explicitly
ample, using sln(S) 5 0.2, COVN 5 0.15 and wnonliquefied=wliquefied 5 accounted for, and the other available information regarding cyclic
8 9
>
> ðN1 Þ60cs ðN1 Þ60cs 2 ðN1 Þ60cs 3 ðN1 Þ60cs 4 >
>
>
< þ 2 þ 2 2:67 2 ln CSR M57:5; s9v 51 atm >
=
14:1 126 23:6 25:4
PL ðN1 Þ60cs ; CSRM57:5; s9v 51 atm ¼ F 2
>
> slnðRÞ >
>
>
: >
;
ð31Þ
The recommended triggering curves for probabilities of lique- derived herein (Fig. 6). The reasons for these differences are ex-
faction (PL) equal to 15%, 50%, and 85% with model uncertainty amined in detail in Idriss and Boulanger (2010, 2012), where the
alone [i.e., conditional on known values of CSRM57.5, s951 atm and authors proposed that the primary causes of the differences in the
(N1)60cs] are plotted together with the case history data in Fig. 6. The liquefaction triggering correlations were the interpretations and
previously noted probabilistic triggering relationship [Eq. (30)] is treatment of eight key case histories with vertical effective stresses
equal to the deterministic triggering correlation recommended by between 0.65 and 1.5 atm in the Cetin et al. (2000, 2004) database.
Idriss and Boulanger (2004, 2008) for «ln(R) 5 20.13. The de- These included having classified four key case histories as lique-
terministic relationship is therefore 1 SD below the expected trig- faction cases in conflict with the original sources’ classification as
gering curve and, accordingly, corresponds to a probability of being no-liquefaction cases, and having significant numerical errors
liquefaction of about 16%, as illustrated in Figs. 4 and 6. between the rd values used to develop their correlation and the rd
The probabilistic triggering curves by Cetin et al. (2002, 2004), values computed using their applicable equation. The consequence
as shown in Fig. 1(b), are located significantly lower than the curves of these misclassifications and errors was the regression of an
Fig. 6. Curves of CRRM57.5, s9v51 atm versus (N1)60cs for probabilities of liquefaction of 15, 50, and 85%
uations, the uncertainties in estimating the latter parameters are much Nishio 1988; Yoshimi et al. 1989, 1994).
greater than the uncertainty in the liquefaction triggering model. For The liquefaction triggering relationships by Tokimatsu and
this reason, the formal treatment of uncertainties in the seismic hazard Yoshimi (1983), which are almost identical to those of Yoshimi et al.
analysis and a detailed site characterization effort are generally more (1994), except at very low (N1)60cs values, are compared with the case
important to a liquefaction evaluation analysis than the uncertainty in history data in Fig. 7. Fig. 7 also shows the PL 5 15% curves derived
the liquefaction triggering model. using the recommended probabilistic relationship [Eq. (30)] and the
Similarly, probabilistic relationships similar to that of Toprak relationship derived using Eq. (28) with two fitting parameters. All
et al. (1999) shown in Fig. 1(a) must be recognized as already in- three correlations are in generally good agreement for (N1)60cs values
cluding some uncertainty in the input parameters (i.e., the mea- less than about 15. For (N1)60cs values greater than about 15, the
surement uncertainties in the case history database). For example, Tokimatsu and Yoshimi (1983) curves follow the results of the frozen
a probabilistic liquefaction hazard analysis can be structured so sand test data in curving sharply upward near (N1)60cs of about 40. The
that it sequentially branches through a range of seismic hazards PL 5 15% curve from Eq. (28) with two fitting parameters follows
the case history data and curves sharply upward near (N1)60cs of about
(which would account for the majority of the uncertainty in the
26, whereas the recommended PL 515% relationship curves sharply
CSRM57.5, s951 atm values) and a range of site characterizations
upward near (N1)60cs of about 30. The case history database contains
[which would account for the majority of the uncertainty in the no cases of liquefaction for (N1)60cs values .26 despite the very strong
(N1)60cs values] before it gets to the liquefaction triggering shaking levels and numerous sites with representative (N1)60cs in the
analysis. In that approach, it may be reasonable to only include range of 25 to 40. There were also 10 no-liquefaction cases with
model uncertainty in the liquefaction triggering analysis because values of (N1)60cs between 21 and 28 that plotted above the rec-
the parameter uncertainties had already been accounted for in the ommended PL 5 15% triggering curve.
previous branches of the analysis. If the seismic hazard and site The position of liquefaction triggering curves at the upper range
characterization branches are instead followed by a probabilistic of (N1)60cs values should consider the information provided by both
liquefaction triggering relationship such as that by Toprak et al. the case history and frozen sand sample test databases. The cyclic
(1999), then the inclusion of the model and measurement tests on frozen sand samples show that it may be possible to generate
uncertainties in the liquefaction triggering analysis, in addition to peak excess pore pressure ratios of 100% in very dense sands with
Fig. 7. Comparison of the Tokimatsu and Yoshimi (1983) frozen sand sample test-based liquefaction triggering correlation with the updated case
history database by Idriss and Boulanger (2010) and the derived probabilistic liquefaction triggering curves
uation of in-situ test methods.” Rep. EL-5507, Vol. 2, Electric Power Yoshimi, Y., Tokimatsu, K., and Hosaka, Y. (1989). “Evaluation of liq-
Research Institute, Palo Alto, CA. uefaction resistance of clean sands based on high-quality undisturbed
Phoon, K.-K. (2008). Reliability-based design in geotechnical engineering: samples.” Soil Found., 29(1), 93–104.
Computations and applications, Taylor and Francis, New York. Yoshimi, Y., Tokimatsu, K., and Ohara, J. (1994). “In situ liquefaction
Phoon, K.-K., and Kulhawy, F. H. (1999). “Characterization of geotechnical resistance of clean sands over a wide density range.” Geotechnique, 44(3),
variability.” Can. Geotech. J., 36(4), 612–624. 479–494.
Seed, H. B., and Idriss, I. M. (1967). “Analysis of liquefaction: Niigata Youd, T. L., et al. (2001). “Liquefaction resistance of soils: Summary report
earthquake.” J. Soil Mech. Found. Div., 93(SM3), 83–108. from the 1996 NCEER and 1998 NCEER/NSF workshops on evaluation
Seed, H. B., and Idriss, I. M. (1971). “Simplified procedure for evaluating soil of liquefaction resistance of soils.” J. Geotech. Geoenviron. Eng., 127(10),
liquefaction potential.” J. Soil Mech. Found. Div., 97(SM9), 1249–1273. 817–833.
Seed, H. B., Tokimatsu, K., Harder, L. F. Jr., and Chung, R. (1985). “In- Youd, T. L., and Noble, S. K. (1997). “Liquefaction criteria based on
fluence of SPT procedures in soil liquefaction resistance evaluations.” statistical and probabilistic analyses.” Proc., NCEER Workshop on
J. Geotech. Eng., 111(12), 1425–1445. Evaluation of Liquefaction Resistance of Soils, Technical Rep. NCEER-
Shibata, T. (1981). “Relations between N-value and liquefaction potential 97-022, T. L. Youd and I. M. Idriss, eds., State Univ. of New York,
of sand deposits.” Proc., 16th Annual Convention of Japanese Society Buffalo, NY, 201–205.