Professional Documents
Culture Documents
Abstract: The paper describes a performance-based approach to the evaluation of liquefaction potential, and shows how it can be used
to account for the entire range of potential ground shaking. The result is a direct estimate of the return period of liquefaction, rather than
a factor of safety or probability of liquefaction conditional upon ground shaking with some specified return period. As such, the
performance-based approach can be considered to produce a more complete and consistent indication of the actual likelihood of lique-
faction at a given location than conventional procedures. In this paper, the performance-based procedure is introduced and used to
compare likelihoods of the initiation of liquefaction at identical sites located in areas of different seismicity. The results indicate that the
likelihood of liquefaction depends on the position and slope of the peak acceleration hazard curve, and on the distribution of earthquake
magnitudes contributing to the ground motion hazard. The results also show that the consistent use of conventional procedures for the
evaluation of liquefaction potential produces inconsistent actual likelihoods of liquefaction.
DOI: 10.1061/共ASCE兲1090-0241共2007兲133:7共802兲
CE Database subject headings: Earthquakes; Liquefaction; Sand; Penetration tests; Hazards.
motion duration. It should be noted that two pieces of loading Research 1997兲. The liquefaction evaluation procedure described
information—amax and earthquake magnitude—are required for by Youd et al. 共2001兲 will be referred to hereafter as the NCEER
estimation of the CSR. procedure. The NCEER procedure has been shown to produce
reasonable predictions of liquefaction potential 共i.e., few cases of
Characterization of Liquefaction Resistance nonprediction for sites at which liquefaction was observed兲 in
past earthquakes, and is widely used in contemporary geotechni-
The CRR is generally obtained by correlation to in situ test re- cal engineering practice. For the purposes of this paper, a conven-
sults, usually standard penetration 共SPT兲, cone penetration 共CPT兲, tionally liquefaction-resistant site will be considered to be one for
or shear wave velocity 共Vs兲 tests. Of these, the SPT has been most which FSL 艌 1.2 for a 475-year ground motion using the NCEER
commonly used and will be used in the remainder of this paper. A
procedure. This standard is consistent with that recommended by
number of SPT-based procedures for deterministic 共Seed and
Martin and Lew 共1999兲, for example, and is considered represen-
Idriss 1971; Seed et al. 1985; Youd et al. 2001; Idriss and Bou-
tative of those commonly used in current practice.
langer 2004兲 and probabilistic 共Liao et al. 1988; Toprak et al.
1999; Youd and Noble 1997; Juang and Jiang 2000; Cetin et al.
Probabilistic Approach
2004兲 estimation of liquefaction resistance have been proposed.
Recently, a detailed review and careful reinterpretation of lique-
Deterministic Approach faction case histories 共Cetin 2000; Cetin et al. 2004兲 was used to
Fig. 1共a兲 illustrates the widely used liquefaction resistance curves develop new probabilistic procedures for the evaluation of lique-
recommended by Youd et al. 共2001兲, which are based on discus- faction potential. The probabilistic implementation of the Cetin et
sions at a National Center for Earthquake Engineering Research al. 共2004兲 procedure produces a probability of liquefaction, PL,
共NCEER兲 workshop 共National Center for Earthquake Engineering which can be expressed as
冋
PL = ⌽ −
共N1兲60共1 + 1FC兲 − 2 ln CSReq − 3 ln M w − 4 ln共⬘vo/pa兲 + 5FC + 6
册 共2兲
where ⌽ = standard normal cumulative distribution function; faction includes both loading terms 共again, peak acceleration, as
共N1兲60 = corrected SPT resistance; FC= fines content 共in percent兲; reflected in the CSR, and magnitude兲 and resistance terms 共SPT
CSReq = cyclic stress ratio 关Eq. 共1兲 without MSF兴; M w = moment resistance, FC, and vertical effective stress兲. Mean values of the
magnitude; ⬘vo = initial vertical effective stress, pa is atmospheric model coefficients are presented for two conditions in Table 1—a
pressure in same units as ⬘vo; = measure of the estimated model case in which the uncertainty includes parameter measurement/
and parameter uncertainty; and 1–6 are model coefficients ob- estimation errors and a case in which the effects of measurement/
tained by regression. As Eq. 共2兲 shows, the probability of lique- estimation errors have been removed. The former would corre-
spond to uncertainties that exist for a site investigated with a found that the average effective stress for their critical layers were
normal level of detail and the latter to a “perfect” investigation at lower effective stresses 共⬃0.65 atm兲 instead of the standard 1
关i.e., no uncertainty in any of the variables on the right-side of atm, and made allowances for those differences. Also, the basic
Eq. 共2兲兴. Fig. 1共b兲 shows contours of equal PL for conditions in shapes of the cyclic resistance curves are different—the Cetin
which measurement/estimation errors are included; the et al. 共2004兲 curves 共Fig. 3兲 have a smoothly changing curvature
measurement/estimation errors have only a slight influence on the
while the Youd et al. 共2001兲 curve 共Fig. 1兲 is nearly linear at
model coefficients but a significant effect on the uncertainty term,
intermediate SPT resistances 关共N1兲60 ⬇ 10– 22兴 with higher curva-
.
Direct comparison of the procedures described by Youd et al. tures at lower and higher SPT resistances. An approximate com-
共2001兲 and Cetin et al. 共2004兲 is difficult because various aspects parison of the two methods can be made by substituting CRR for
of the procedures are different. For example, Cetin et al. 共2004兲 CSReq in Eq. 共2兲 and then rearranging the equation in the form
where ⌽−1 = inverse standard normal cumulative distribution Table 1兲 shows equivalence of FSL when a value of PL ⬇ 0.6 is
function. The resulting value of CRR can then be used in the used. Cetin et al. 共2004兲 suggest the use of a deterministic curve
common expression for FSL. Arango et al. 共2004兲 used this for- equivalent to that given by Eq. 共3兲 with PL = 0.15, which would
mulation without measurement/estimation errors 共Case II in Table produce a more conservative result than the NCEER procedure.
1 and found that the Cetin et al. 共2004兲 and NCEER procedures The differences between the two procedures are most pronounced
yielded similar values of FSL for a site in San Francisco when a at high CRR values; the NCEER procedure contains an implicit
value of PL ⬇ 0.65 was used in Eq. 共3兲. A similar exercise for a assumption of 共N1兲60 = 30 as an upper bound to liquefaction sus-
site in Seattle with measurement/estimation errors 共Case I in ceptibility while Cetin et al. 共2004兲, whose database contained
Fig. 2. Magnitude and distance deaggregation of 475-year peak acceleration hazard for site in Seattle
considerably more cases at high CSR levels, indicate that lique- analyzed by the U.S. Geological Survey 共USGS兲 共http://
faction is possible 共albeit with limited potential effects兲 at 共N1兲60 eqhazmaps.usgs.gov兲 are shown in Fig. 3. The decreasing signifi-
values above 30. cance of lower magnitude earthquakes for longer return periods,
evident in Fig. 3, is a characteristic shared by many other
locations.
Seismic Hazard Analysis
Ground shaking levels used in seismic design and hazard evalu- Performance-Based Liquefaction Potential
ations are generally determined by means of seismic hazard Evaluation
analyses. Deterministic seismic hazard analyses are used most
often for special structures or for estimation of upper bound In practice, liquefaction potential is usually evaluated using de-
ground shaking levels. In the majority of cases, however, ground terministic CRR curves, a single ground motion hazard level, for
shaking levels are determined by probabilistic seismic hazard example, for ground motions with a 475-year return period, and a
analyses. single earthquake magnitude, usually the mean or mode. In con-
Probabilistic seismic hazard analyses consider the potential trast, the performance-based approach incorporates probabilistic
levels of ground shaking from all combinations of magnitude and CRR curves and contributions from all hazard levels and all
distance for all known sources capable of producing significant earthquake magnitudes.
shaking at a site of interest. The distributions of magnitude and The roots of performance-based liquefaction assessment are in
distance, and of ground shaking level conditional upon magnitude the method of seismic risk analysis introduced by Cornell 共1968兲.
and distance, are combined in a way that allows estimation of the The first known application of this approach to liquefaction as-
mean annual rate at which a particular level of ground shaking sessment was presented by Yegian and Whitman 共1978兲, although
will be exceeded. The mean annual rate of exceeding a ground earthquake loading was described as a combination of earthquake
motion parameter value, y *, is usually expressed as y*; the recip- magnitude and source-to-site distance rather than peak accelera-
rocal of the mean annual rate of exceedance is commonly referred tion and magnitude. Atkinson et al. 共1984兲 developed a procedure
to as the return period. The results of a PSHA are typically pre- for the estimation of the annual probability of liquefaction using
sented in the form of a seismic hazard curve, which graphically linearized approximations of the CRR curves of Seed and Idriss
illustrates the relationship between y* and y *. 共1983兲 in a deterministic manner. Marrone et al. 共2003兲 described
The ground motion level associated with a particular return liquefaction assessment methods that incorporate probabilistic
period is therefore influenced by contributions from a number of CRR curves and the full range of magnitudes and peak accelera-
different magnitudes, distances, and conditional exceedance prob- tions in a manner similar to the PBEE framework described
ability levels 共usually expressed in terms of a parameter, , de- herein. Hwang et al. 共2005兲 described a Monte Carlo simulation-
fined as the number of standard deviations by which ln y * exceeds based approach that produces similar results.
the natural logarithm of the median value of y for a given M and PBEE is generally formulated in a probabilistic framework to
R兲. The relative contributions of each M-R pair to y* can be evaluate the risk associated with earthquake shaking at a particu-
quantified by means of a deaggregation analysis 共McGuire, 1995兲; lar site. The risk can be expressed in terms of economic loss,
the deaggregated contributions of magnitude and distance are fre- fatalities, or other measures. The Pacific Earthquake Engineering
quently illustrated in diagrams such as that shown in Fig. 2. Be- Research 共PEER兲 has developed a probabilistic framework for
cause both peak acceleration and magnitude are required for PBEE 共Cornell and Krawinkler 2000; Krawinkler 2002; Deierlein
cyclic stress-based evaluations of liquefaction potential, the mar- et al. 2003兲 that computes risk as a function of ground shaking
ginal distribution of magnitude can be obtained by summing the through the use of several intermediate variables. The ground
contributions of each distance and value for each magnitude; motion is characterized by an intensity measure, IM, which could
magnitude distributions for six return periods at a site in Seattle be any one of a number of ground motion parameters 共e.g., amax,
冋
P关FSL ⬍ FSL* 兩amaxi,m j兴 = ⌽ −
共N1兲60共1 + 1FC兲 − 2 ln共CSReq,i · FSL* 兲 − 3 ln m j − 4 ln共⬘vo/pa兲 + 5FC + 6
册 共7兲
P关Nreq ⬎ Nreq
*
兩amaxi,m j兴 = ⌽ −冋 *
Nreq共1 + 1FC兲 − 2 ln CSReq,i − 3 ln m j − 4 ln共⬘vo/pa兲 + 5FC + 6
册 共9兲
*
The value of Nreq can be interpreted as the SPT resistance re- ing. The wide range of smoothly increasing SPT resistance, while
quired to produce the desired performance level for shaking with perhaps unlikely to be realized in a natural depositional environ-
a return period of 1 / N* . ment, is useful for illustrating the main points of this paper.
req
Locations
Comparison of Conventional and
Performance-Based Approaches In order to illustrate the effects of different seismic environments
on liquefaction potential, the hypothetical site was assumed to be
Conventional procedures provide a means for evaluating the liq- located in each of the ten United States cities listed in Table 2. For
uefaction potential of a soil deposit for a given level of loading. each location, the local seismicity was characterized by the proba-
When applied consistently to different sites in the same seismic bilistic seismic hazard analyses available from the USGS 共using
environment, they provide a consistent indication of the likeli- the 2002 interactive deaggregation link with listed latitudes and
hood of liquefaction 共expressed in terms of FSL or PL兲 at those longitudes兲. In addition to being spread across the United States,
sites. The degree to which they provide a consistent indication of these locations represent a wide range of seismic environments;
liquefaction likelihood when applied to sites in different seismic the total seismic hazard curves for each of the locations are shown
environments, however, has not been established. That issue is in Fig. 6. The seismicity levels vary widely—475-year peak ac-
addressed in the remainder of this paper. celeration values range from 0.12g 共Butte兲 to 0.66g 共Eureka兲.
Two of the locations 共Charleston and Memphis兲 are in areas of
low recent seismicity with very large historical earthquakes, three
Idealized Site 共Seattle, Portland, and Eureka兲 are in areas subject to large-
Potentially liquefiable sites around the world have different like- magnitude subduction earthquakes, and two 共San Francisco and
lihoods of liquefaction due to differences in site conditions San Jose兲 are in close proximity 共⬃60 km兲 in a very active envi-
共which strongly affect liquefaction resistance兲 and local seismic ronment.
environments 共which strongly affect loading兲. The effects of seis-
mic environment can be isolated by considering the liquefaction Conventional Liquefaction Potential Analyses
potential of a single soil profile placed at different locations.
Fig. 5 shows the subsurface conditions for an idealized, hypo- Two sets of conventional deterministic analyses were performed
thetical site with corrected SPT resistances that range from rela- to illustrate the different degrees of liquefaction potential of the
tively low 关共N1兲60 = 10兴 to moderately high 关共N1兲60 = 30兴. Using the
cyclic stress-based approach, the upper portion of the saturated Table 2. Peak Ground Surface 共Quaternary Alluvium兲 Acceleration
sand would be expected to liquefy under moderately strong shak- Hazard Information for Ten U.S. Cities
Latitude Longitude 475-year 2,475-year
Location 共N兲 共W兲 amax amax
Butte, MT 46.003 112.533 0.120 0.225
Charleston, SC 32.776 79.931 0.189 0.734
Eureka, CA 40.802 124.162 0.658 1.023
Memphis, TN 35.149 90.048 0.214 0.655
Portland, OR 45.523 122.675 0.204 0.398
Salt Lake City, UT 40.755 111.898 0.298 0.679
San Francisco, CA 37.775 122.418 0.468 0.675
San Jose, CA 37.339 121.893 0.449 0.618
Santa Monica, CA 34.015 118.492 0.432 0.710
Fig. 5. Subsurface profile for idealized site Seattle, WA 47.530 122.300 0.332 0.620
Fig. 7. Profiles of: 共a兲 factor of safety against liquefaction; 共b兲 required SPT resistance obtained using NCEER deterministic procedure with
475-year ground motions
ses for an element of soil near the center of the saturated zone 共at application of conventional procedures for evaluation of liquefac-
a depth of 6 m, at which 共N1兲60 = 18 for the hypothetical soil pro- tion potential. For each site location, the process is as follows:
file兲. Fig. 9共a兲 shows factor of safety hazard curves, and Fig. 9共b兲 1. At the depth of interest, determine the SPT resistance re-
PB
shows hazard curves for Nreq , the SPT resistance required to resist quired to produce a factor of safety of 1.2 using the conven-
liquefaction. Note that the SPT resistances shown in Fig. 9共b兲 are tional approach 关from either Figs. 7共b兲 or 8共b兲兴. At that SPT
those at which liquefaction would actually be expected to occur, resistance, the soils at that depth would have an equal lique-
rather than the values at which FSL would be as low as 1.2 共cor- faction potential 共i.e., FSL = 1.2 with a 475-year ground mo-
responding to a conventionally liquefaction resistant soil as de- tion兲 at all site locations as evaluated using the conventional
fined previously兲, which were plotted in Figs. 7 and 8. Therefore, approach.
the mean annual rates of exceedance in Fig. 9 are equal at each 2. Determine the mean annual rate of exceedance for the SPT
PB
site location for FSL = 1.0 and Nreq = 18. resistance from Step 1 using results of the type shown in Fig.
9共b兲 for each depth of interest. Since Fig. 9共b兲 shows the SPT
Equivalent Return Periods resistance for FSL = 1.0, this is the mean annual rate of lique-
The results of the conventional deterministic analyses shown in faction for soils with this SPT resistance at the depth of
Figs. 7 and 8 can be combined with the results of the interest.
performance-based analyses shown in Fig. 9 to evaluate the return 3. Compute the return period as the reciprocal of the mean an-
periods of liquefaction produced in different areas by consistent nual rate of exceedance.
Fig. 9. Seismic hazard curves for 6-m depth: 共a兲 factor of safety against liquefaction, FSL for 共N1兲60 = 18; 共b兲 required SPT resistance, Nreq
PB
, for
FSL = 1.0
4. Repeat Steps 1–3 for each depth of interest. Table 3 shows the return periods of liquefaction at a depth of
This process was applied to all site locations in Table 2 to evalu- 6 m 共the values are approximately equal to the averages over the
ate the return period for liquefaction as a function of depth for depth of the saturated zone for each site location兲 for conditions
each location; the calculations were performed using 475-year that would be judged as having equal liquefaction potential using
ground motions and again using 2,475-year ground motions. conventional procedures. Using both the NCEER and NCEER-C
Fig. 10 shows the results of this process for both sets of con- procedures, the actual return periods of liquefaction can be seen
ventional analyses. It is obvious from Fig. 10 that consistent ap- to vary significantly from one location to another, particularly for
plication of the conventional procedure produces inconsistent the case in which the conventional procedure was used with
return periods, and therefore different actual likelihoods of lique- 2,475-year motions.
faction, at the different site locations. Examination of the return The actual return periods depend on the seismic hazard curves
period curves shows that they are nearly vertical at depths greater and deaggregated magnitude distributions, and are different for
than about 4 m, indicating that the deterministic procedures are the NCEER and NCEER-C procedures. Using the NCEER pro-
relatively unbiased with respect to SPT resistance. The greater cedure with 475-year motions, the actual return periods of lique-
verticality of the curves based on the NCEER-C analyses results faction range from as short as 236 years 关for Eureka, which is
from the consistency of the shapes of those curves and the con- affected by the 共N1兲60 = 30 limit implied by the NCEER proce-
stant PL curves given by Eq. 共2兲, which were used in the dure兴 to 565 years 共Memphis兲; the corresponding 50-year prob-
performance-based analyses. Differences between the shapes of ability of liquefaction 共under the Poisson assumption兲 in Eureka
the NCEER curve 关Fig. 1共a兲兴 and the curves 关Fig. 1共b兲兴, particu- would be more than double that in Memphis. The return periods
larly for sites subjected to very strong shaking 共hence, very high computed using the NCEER-C procedure with 475-year motions
CSRs兲 such as San Francisco and Eureka, contribute to depth- are more consistent, but still range from 341 years 共San Jose兲 to
dependent return periods for the NCEER results. about 570 years 共Charleston and Memphis兲.
If deterministic liquefaction potential evaluations are based on
Table 3. Liquefaction Return Periods for 6 m Depth in Idealized Site at 2,475-year ground motions using the NCEER procedure, the im-
Different Site Locations Based on Conventional Liquefaction Potential plied limit of 共N1兲60 = 30 produces highly inconsistent actual re-
Evaluation Using 475-year and 2,475-year Motions turn periods—the 50-year probability of liquefaction in Eureka
would be more than six times that in Portland. All but two of the
475-year motions 2,475-year motions
ten locations would require 共N1兲30 = 30 according to that proce-
Location NCEER NCEER-C NCEER NCEER-C dure and, as illustrated in Fig. 9共b兲, the return periods for
PB
Nreq = 30 vary widely in the different seismic environments. The
Butte, MT 348 418 1,592 2,304
variations are smaller but still quite significant using the
Charleston, SC 532 571 1,433a 2,725
NCEER-C procedure.
Eureka, CA 236a 483 236a 1,590
Differences in regional seismicity can produce significant dif-
Memphis, TN 565 575 1,277a 2,532
ferences in ground motions at different return periods. Leyen-
Portland, OR 376 422 1,675 1,508 decker et al. 共2000兲 showed that short-period 共0.2 s兲 spectral
Salt Lake City, UT 552 543 1,316a 2,674 acceleration, for example, increased by about 50% when going
San Francisco, CA 355a 503 355a 1,736 from return periods of 475 years to 2,475 years in Los Angeles
San Jose, CA 360 341 532a 1,021 and San Francisco but by 200–500% or more in other areas of the
Santa Monica, CA 483 457 794a 1,901 country. The position and slope of the peak acceleration hazard
Seattle, WA 448 427 1,280a 2,155 curve clearly affect the return period of liquefaction. However,
a
Upper limit of 共N1兲60 = 30 implied by NCEER procedure reached. the regional differences in magnitude distribution 共i.e., the rela-
tive contribution to peak acceleration hazard from each magni- 14.3 for a 475-year return period of liquefaction 共performance-
tude兲 also contribute to differences in return period; San based procedure兲. For Portland, the relative unconservatism in the
Francisco and San Jose have substantially different return periods conventional approach means that the required SPT resistance
for liquefaction despite the similarity of their hazard curves be- increases slightly from 17.0 共deterministic procedure兲 to 17.5
cause the relative contributions of large magnitude earthquakes on 共performance-based procedure兲.
the San Andreas fault are higher for San Francisco than for San Because cyclic resistance ratio varies nonlinearly with SPT
Jose. resistance, it is also useful to consider the difference between the
deterministic and performance-based approaches from a factor of
safety standpoint. Since FSL is proportional to CRR, the ratio of
Conditions for Consistent Liquefaction Potential the CRR values corresponding to the SPT resistances in Table 4
can be thought of as factor of safety ratios that describe the
The differences in actual liquefaction return periods produced by “extra” liquefaction resistance required by the deterministic crite-
conventional liquefaction potential evaluations make it difficult to rion 共FSL = 1.2 for 475-year ground motion兲 relative to that re-
establish uniform and consistent procedures for conventional quired by the performance-based criterion 共475-year return period
evaluation of seismic hazards such as liquefaction potential. The for liquefaction兲. These values, shown in Table 4, range from 0.88
performance-based approach, however, provides a framework in to 1.28 for the NCEER procedure and from 0.91 to 1.32 for the
which design and evaluation can be based on a specified return NCEER-C procedure; higher values are associated with locations
period for liquefaction rather than on the basis of a factor of that are “penalized” by conventional deterministic procedures.
safety or probability of liquefaction for a ground motion with a The zero values shown for Eureka and San Francisco result from
specified return period. It is useful to compare the differences in the fact that NreqPB
⬎ 30 for those highly active areas; the corre-
possible requirements for acceptable liquefaction resistance pro- sponding CRR from the NCEER procedure is infinite. In Mem-
duced by the conventional and performance-based approaches. phis and Charleston, for example, the deterministic criterion
For the purposes of this paper, two alternative criteria will be would result in a factor of safety some 30% higher than that
considered: associated with an actual return period of 475 years. At several
1. The previously described conventional criterion of a mini- other locations, the same deterministic criterion would result in
mum FSL = 1.2 with 475-year ground motions; and factors of safety lower than those associated with the same actual
2. A performance-based criterion of a 475-year return period liquefaction hazard. In effect, the conventional procedure results
for liquefaction, i.e., TR共FSL = 1.0兲 = 475 years. in an owner in Memphis designing for an equivalent factor of
The 475-year return period used in the second criterion is in- safety that is about 45% higher than an owner in Portland.
tended as an example; a suitable specific return period for an
actual performance-based liquefaction criterion would need to be
identified and endorsed by a group of experienced professionals. Summary and Conclusions
The SPT resistances required to satisfy the first criterion at a
depth of 6 m, computed using the NCEER and NCEER-C proce- The evaluation of liquefaction potential involves comparison of
dures, are listed in Table 4, as well as the SPT resistances required consistent measures of loading and resistance. In contemporary
to satisfy the second criterion at the same depth. For those loca- geotechnical engineering practice, such comparisons are com-
tions at which the return periods for liquefaction in Table 3 were monly made using cyclic shear stresses expressed in terms of
less than 475 years, the SPT resistances required for liquefaction normalized cyclic stress and cyclic resistance ratios. The CSR is
with an actual return period of 475 years are increased, and vice usually estimated using a simplified procedure in which the level
versa for locations at which the Table 2 return periods were of ground shaking is related to peak ground surface acceleration
greater than 475 years. For a location like Memphis, the relative and earthquake magnitude. Criteria by which liquefaction resis-
conservatism in the conventional approach means that the re- tance is judged to be adequate are usually expressed in terms of a
quired SPT resistance of 19.1 for FSL = 1.2 with the 475-year single level of ground shaking.
motion 共NCEER procedure兲 is reduced to an SPT resistance of For a given soil profile at a given location, liquefaction can be