Professional Documents
Culture Documents
Kramer Wang
Kramer Wang
Kramer Wang
Abstract: An empirical model for the estimation of residual strength is proposed. The proposed model predicts residual strength as a nonlinear
function of both penetration resistance and initial effective stress. The model is consistent with steady-state concepts and behavior observed in
laboratory tests, but was calibrated on the basis of residual strengths back-calculated from flow-side case histories. The back-calculated
strengths considered inertial effects, hydroplaning, uncertainties, the conditions under which flow sides are known not to have occurred
in past earthquakes, and the relative quality of each case history. The proposed model predicts residual strengths that are generally between
those predicted by direct approaches and those predicted by normalized strength approaches. The proposed model also allows estimation of the
probability distribution of residual strength. DOI: 10.1061/(ASCE)GT.1943-5606.0001317. © 2015 American Society of Civil Engineers.
Fig. 1. Stress-strain and stress path behavior for elements of soil consolidated to three different initial states (adapted from Yoshimine and Ishihara
1998, soils and foundation, The Japanese Geotechnical Society, with permission)
(a) (b)
Fig. 2. Conceptual illustration of ultimate steady-state strength: (a) relationship between one-dimensional normal consolidation line, ultimate steady-
state line, and state parameter; (b) resulting trend of ultimate steady-state strength with vertical effective stress
Laboratory Testing-Based Approach show shearing resistances continuing to change at shear strains of
30% or more. The rapid softening that occurs upon triggering of
Numerous laboratory investigations of the USS behavior of sands
liquefaction generally prevents testing equipment from maintaining
have been undertaken (Olson and Stark 2003). Early experimental
stress-controlled loading through the mobilization of the USS.
investigations formed the basis for current understanding of the
Although the steady-state concept postulates that USS strength
behavior of liquefiable soils, and led to the development of a labo-
depends only on soil density (e.g., Poulos 1981; Been et al. 1991;
ratory testing-based procedure for evaluation of USS strength.
Ishihara 1993), a number of experimental investigations (e.g., Miura
Procedures and Toki 1982; Hanzawa 1980; Kuerbis et al. 1988; Vaid et al.
Poulos et al. (1985) proposed a method for evaluation of USS 1990; Riemer and Seed 1997; Yoshimine and Ishihara 1998) have
strength based on laboratory testing of carefully retrieved undis- concluded that USS strength is influenced by stress path. Given that
turbed specimens and testing of reconstituted specimens of the flow failures often involve multiple stress paths, estimation of the in
same soil. The procedure involves determining the in situ void ratio situ USS strength from tests performed using one particular stress
from one or more undisturbed specimens, measuring the USS path may be inaccurate.
strength of undisturbed specimens consolidated to confining pres- Discussion
sures sufficiently high to ensure contractive behavior, determining Although conceptually useful, experimentally based procedures for
the slope of the USS line by testing specimens reconstituted at dif- measurement of the residual strength mobilized in the field suffer
ferent void ratios, and correcting the measured USS strengths of the from a number of practical limitations that have tempered their use
undisturbed specimens by assuming the USS line of the undis- in geotechnical engineering practice. While some of these limita-
turbed specimens is parallel to that of the reconstituted specimens. tions could potentially be reduced by using, for example, frozen
Sadrekarimi (2013) used a framework similar to that described samples and multiple stress path testing, a number of them still re-
in Eqs. (1)–(4) to express the normalized USS strength ratio as a main. Even more significantly, however, it must be recognized that
function of the state parameter, soil compressibility, and steady- the USS conditions that can exist under idealized conditions in
state friction angle. The relationship showed good agreement with laboratory tests usually do not exist in actual flow sides. As a flow
laboratory data for a number of sands with different relative com- failure develops in the field, drainage can occur, leading to changes
pressibilities, i.e., ratios of ψ to λss . The use of this procedure for in effective stresses, volume, and density; strain rates are variable as
determination of USS strength, therefore, requires knowledge or the failing soil accelerates and then comes to rest; stresses and
estimation of the state parameter and of the in situ compressibility stress paths are different; and different soils can be mixed together
of the soil. during flow. All of these factors tend to limit the direct applicability
of laboratory-measured shear strengths to actual flow sides.
Experimental Issues
Experimental investigations have involved different soils prepared
in different manners and tested at different rates using different Case History-Based Approach
types of equipment. In a number of instances, however, they have Even before all of the issues associated with the laboratory
produced conflicting and/or inconsistent results, which have iden- testing-based approach were identified and explored, the notion
tified important issues in the problem of estimating the strength of of using back-calculated strengths from flow failure case histories
liquefied soil. was proposed (Seed 1987). In the case history–based approach,
The effects of sampling disturbance can be pronounced back-calculated strengths are correlated to the penetration resis-
for loose, saturated sands. Even with disturbance corrections tance of the liquefied soil. Since actual flow failures frequently
(e.g., Poulos et al. 1985), the flat slopes of typical USS lines involve drainage, pore pressure (or void) redistribution, mixing,
can lead to significant uncertainty in estimated USS strength variable strain rates, and other factors that violate the assumptions
(Kramer 1989). Laboratory strength tests also may not be able to of the USS of deformation, back-calculated strengths are frequently
reach the very large strains associated with the USS while main- referred to as residual strengths. Two basic approaches, termed here
taining reasonable uniformity of stresses and strains; many labora- the direct approach and the normalized strength approach, have
tory tests (e.g., Verdugo and Ishihara 1996; Yamamuro and Lade been proposed. The primary versions of both approaches are de-
1998; Yoshimine and Ishihara 1998; Yamamuro and Covert 2001) scribed in the following sections.
tions to standard penetration test (SPT) resistance. Only three of the (in atm) computed as
case histories (Calaveras Dam, Fort Peck Dam, and Lower San
Fernando Dam) were from case histories with clean sand SPT re- Sr ≈ 0.0236 exp½ðN 1 Þ60 ð6Þ
sistances greater than 10. The residual strength was back-calculated
by varying the residual strength in a zone considered to have lique- Wride et al. (1999) reviewed the case histories used by Seed
fied until limit equilibrium procedures produced a factor of safety (1987), Seed and Harder (1990), and Stark and Mesri (1992), and
of unity. The postfailure slope geometry was analyzed for most of related their interpretations of the most reliable undrained strength
the case histories, but the prefailure geometry was analyzed for to the minimum SPT resistance, which was described as “a reason-
some. The back-calculated residual strength was plotted against a able lower bound ðN 1 Þ60in situ .” Wride et al. (1999) noted the diffi-
representative clean sand SPT resistance, ðN 1 Þ60−cs , which was culty of expressing the conditions in a particular flow-side case
defined using history by a single undrained strength and a single SPT resistance.
Gutierrez and Eddy (2011) back-calculated residual strengths
ðN 1 Þ60−cs ¼ ðN 1 Þ60 þ ΔðN 1 Þ60 ð5Þ from postfailure geometries of 38 flow-side case histories, a num-
ber of which have been identified by others as lateral spreads. The
back-calculation did not consider inertial effects, and residual
where ΔðN 1 Þ60 is a fines correction (Fig. 3). The procedures by strengths were correlated to minimum SPT resistances with no fines
which the representative SPT resistance or the fines correction were correction. These analyses established the relationship
obtained were not explicitly described, but the fines correction was
the same as that recommended for evaluation of liquefaction poten- Sr ¼ 0.87ðN 1 Þ60 þ 0.1ðN 1 Þ260 ð7Þ
tial. The results of the back-calculation analyses fell within a band
that showed a trend of increasing residual strength with increasing Gutierrez and Eddy (2011) also characterized uncertainties in
SPT resistance, as shown in Fig. 3. residual strengths using first-order reliability methods. Uncertainty
Seed and Harder (1990) expanded (to 17 case histories) and re- in the residual strength was characterized by a beta distribution and
interpreted the case history database and developed an updated re- only shown graphically; the 16th percentile curve is about 60% of
lationship for residual strength estimation. Back-calculated residual the 50th percentile curve, which would correspond to a σln Sr value
strengths were correlated to a representative clean sand SPT resis- of 0.51 if Sr was a lognormally distributed quantity.
tance, with the fines correction identical to that of Seed (1987). The
variation of residual strength with clean sand SPT resistance is Normalized Strength Procedures
shown graphically in Fig. 3. Seed and Harder (1990) recommended Castro (1987) observed that the density of a given soil increased
use of “the lower-bound, or near lower-bound relationship” be- with increasing initial effective stress and that residual strength
tween residual strength and SPT resistance “owing to scatter and increased with increasing density, and concluded that the USS
strength should be related to initial effective stress. As indicated
previously, the ratio of USS strength to effective stress should
be constant for a given soil if the USS line and consolidation curve
are parallel (λss ¼ λc ).
Castro (1987), Castro and Troncoso (1989), and Castro (1991)
investigated several tailings dams in South America and reported
ratios of USS strength to initial major principal effective stress,
0 , ranging from 0.12 to 0.19.
Sus =σ1c
Stark and Mesri (1992) developed a database of 20 case histor-
ies for which the ratios of residual strength to initial vertical effec-
tive stress, hereafter referred to as the normalized residual strength
0 , were computed. These stress ratios were found to
ratio, Sr =σvo
correlate better to a representative clean sand SPT resistance when
a different fines correction (Table 1) than that of Seed (1987) was
used. Noting that the strength of a liquefied soil at a given vertical
effective stress should increase with factors (e.g., gradation, particle
angularity, particle roughness) that also increase SPT resistance,
Fig. 3. Variation of residual strength with equivalent clean sand SPT 0 to representative clean sand
Stark and Mesri (1992) related Sr =σvo
resistance using direct approach residual strength models; upper and
SPT resistance. Supplementing the case history database with in-
lower bound curves are shown for Seed (1987) and Seed and Harder
terpreted laboratory test results, Stark and Mesri (1992) proposed
(1990) models
that the normalized residual strength ratio could be estimated as
(2005) showed that the large-strain strength of a mixture of finer and structural engineering, is a function of geometry and material
and coarser soils can have a lower residual strength than that of properties. Back-analysis of shear strength from a slope failure case
either constituent material at the same void ratio. To the extent that history, therefore, requires characterization of the geometry and
layered soils exist in many of the case histories, this effect will be material properties of the slope. Geometric variables include those
reflected in back-calculated residual strengths. that describe the external geometry of the slope, the geometry of
A number of the case histories involved flow sides into bodies of individual layers of soil/rock within the slope, the geometry of any
water. When such slides occur at high velocities, water in front of phreatic surface within the slope, and the position of the failure
the slide may become trapped under the leading edge of the flow, surface used in the back-analysis. Material property variables in-
resulting in hydroplaning (Horne and Joyner 1965; Mohrig et al. clude measures of density and available shearing resistance.
1998). When hydroplaning occurs, the length of the failure surface
is effectively shortened, which can lead to underestimation of back- Prefailure and Postfailure Geometry
calculated residual strength when not accounted for. Although hy- Uncertainties in slope coordinates were related to the manner in
droplaning may occur at displacement levels greater than those at which the prefailure and postfailure geometries were measured
which residual strength is mobilized, it can affect final geometries and documented. For some case histories, the geometries were sur-
that are used in common inertial correction procedures. veyed with accurate instruments and reported in clear drawings; in
Models using the normalized strength approach predict other cases they were measured more crudely and documented
extremely low residual strengths at low initial vertical effective in less detail. Standard deviations of external geometry points typ-
stresses (i.e., at shallow depths), regardless of SPT resistance. This ically varied from 8 to 15 cm in the horizontal direction and from 30
aspect of the model suggests that flow sliding should be expected to 60 cm in the vertical direction. For internal boundaries, standard
in moderately dense to dense soils subjected to very strong ground deviations ranged from 30 cm to 1.2 m in the horizontal and vertical
shaking; while such soils may develop displacements due to lateral directions, respectively. Water table elevations were assumed to
spreading, the absence of flow slides in such events calls this aspect vary with standard deviations ranging from 10 to 30 cm at the slope
of the normalized model into question. surface and from 30 cm to 1.0 m within the slope.
the ZIF, which was estimated from the sliding block model of of the height of the slope). The nature of the failure mechanism also
Olson (2001). Uncertainty in the zero-inertia geometry was taken affected the accuracy with which the position of the top of the fail-
as a function of the level of documentation of the individual flow ure surface could be determined; the horizontal position of this
slide and of the degree to which the failure mechanism was con- point was taken to vary over a range related to the length of the
sistent with the simplified sliding block model used in estimation of failure surface and the accuracy with which the available informa-
the ZIF. A coefficient of variation ranging from 0.1 to 0.3 was as- tion constrained that location; the range varied from 0 to 6% of the
signed to the ZIF; lower values were used for block-type failures on length of the failure surface. Case histories with broadly distributed
well-defined failure surfaces and higher values for slumping fail- failure zones, slumping mechanisms, and/or retrogressive deforma-
ures or failures with highly distributed zones of shear strain pro- tion mechanisms were assigned higher levels of uncertainty in this
ducing poorly defined failure surfaces. geometric parameter.
Fig. 6. Estimation of zero-inertia geometry for Wachusett Dam: (a) identification of reference points on prefailure and postfailure geometries;
(b) estimated geometry under zero-inertia conditions
16 0.736 0.585 0.528 0.491 0.463 0.449 0.455 ibration process also considered conditions under which flow slides
have been observed not to occur. The details of the model develop-
ment process are described in the following sections.
Back-Analysis Results
Form of Predictive Equation
Each primary case history back-analysis produced a distribution of
50,000 average ðN 1 Þ60 and Sr values, each representing an individ- Multiple functional forms of equations relating residual strength
ual Monte Carlo realization of the slope failure. The mean values of (direct approach) and residual strength ratio (normalized strength
the SPT resistances and residual strengths, along with mean fines approach) to SPT resistance were investigated. The former used
contents and initial vertical effective stresses, are given for each a relationship of the direct form
case history in Table 2. Sr ¼ a1 exp½a2 ðN 1 Þa603 ð12Þ
Direct comparison of the back-calculated residual strengths
for the primary case histories with those obtained by previous in- and the latter a normalized relationship
vestigators is complicated by fundamental differences in the mod-
eling techniques. The consideration of spatial variability in the Sr b3
0 ¼ b1 þ b2 ðN 1 Þ60 ð13Þ
back-analyses performed in this investigation results in average σvo
residual strength values that are greater than the value that would
both of which are capable of producing models very similar to
be obtained using the average SPT resistance with no spatial vari-
those shown in Figs. 3 and 4. Calibration of both models using
ability; this effect increases with increasing residual strength and
nonlinear least-squares regression produced model coefficients that
slope height. Fig. 7 shows the residual strengths from the primary
fit the data well but had residuals that were heteroscedastic and/or
case histories that have been evaluated by other investigators. The
exhibited trends with respect to the predictor variables. The accu-
agreement is generally good with the average residual strengths
racy of the normalized strength model was somewhat better than
from the current back-analyses falling near the high ends of the
that of the direct model, and was found not to be improved by
ranges obtained by others. This behavior results from consideration
the addition of fines content terms.
of spatial variability and hydroplaning, and from differences in the
These preliminary results motivated the development of a model
way inertial effects were calculated.
of different form that has elements of both the direct and normal-
ized strength modeling approaches. The search for an appropriate
model form was guided by the following basic criteria:
Development of an Empirical Model 1. Residual strength should increase with increasing SPT
resistance.
The development of a model for evaluation of residual strength
2. Residual strength should increase with increasing initial ver-
must consider the many complex issues discussed earlier in this
tical effective stress.
3. The sensitivity of residual strength to SPT resistance should
increase with increasing SPT resistance.
4. The residual strength need not vary in direct proportion to in-
itial vertical effective stress.
5. Residual strengths at low initial vertical effective stresses
should be consistent with lateral spreading activity observed
in actual earthquakes.
6. The model should allow probabilistic characterization of resi-
dual strength.
A number of potential predictive models satisfying as many of
these criteria as possible were investigated using nonlinear least-
squares regression. The results of these analyses showed that a
model of the basic form
0 Þ θ4 g
Sr ¼ θ1 expfθ2 ½ðN 1 Þ60 þ θ3 ðσvo ð14Þ
could potentially satisfy all of the criteria and produce a better fit to
Fig. 7. Comparison of back-calculated primary case history residual
the case history data than either the direct or normalized residual
strengths with residual strengths computed by others (LSFD = Lower
strength models. This basic equation was modified to include a
San Fernando Dam; FPD = Fort Peck Dam)
fines content term and an indicator variable for void redistribution
1.2
N = 10 NN==10
10
S = 1 atm SS==11atm
atm
1.0 COVS = 0 COVNS==00
COV
COVN = 0 COVS = 0
0.8 COVN = 0.1 COVS = 0.01
COVN = 0.2 COVS = 0.05
fSr(Sr)
0.4
0.2
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.0 0.1 0.2 0.3 0.4 0.5 0.6
(a) Residual Strength, Sr (atm) (b) Residual Strength, Sr (atm)
Fig. 9. Probability density functions for residual strength: (a) different uncertainties in SPT resistance; (b) different uncertainties in initial vertical
effective stress
Fig. 10. Predicted variation of residual strength with initial vertical effective stress for (a) ðN 1 Þ60 ¼ 5; (b) ðN 1 Þ60 ¼ 10; (c) ðN 1 Þ60 ¼ 15 bpf
(Idriss–Boulanger curves shown with and without void redistribution; (d) for ðN 1 Þ60 ¼ 5 and 10 with minimum strengths implied by lateral spreading
case histories and back-calculated strengths from flow-side case histories at low initial vertical effective stress levels; Seed–Harder values taken at
lower-third point; shallow soil friction angle of 25° used for drained strength constraint
These case histories have ðN 1 Þ60 values ranging from 2.9 to 10.6 and ultimate steady-state line slope. While the theoretical frame-
with an average of 5.2 and standard deviation of 2.2. It should work cannot be expected to accurately predict strengths that are
therefore be expected that the predicted strengths for ðN 1 Þ60 ¼ 5 mobilized under field conditions that violate its basic assumptions,
should be consistent with the strengths from those case histories. the fact that the proposed model shows trends that are consistent
Fig. 10(d) confirms that the proposed model predictions exceed the with those implied by the theoretical framework help support its
lateral spreading data (gray circles) for ðN 1 Þ60 ¼ 10 and are con- applicability. The direct and normalized strength approaches are
sistent with the flow-side data (black circles) for ðN 1 Þ60 ¼ 5. not consistent with the theoretical framework given available ex-
perimental data for typical liquefiable soils.
Several aspects of the proposed model are particularly note-
Discussion worthy. First, the model allows probabilistic characterization of
The model described in this paper presents an alternative to residual residual strength. As such, it does not attempt to be conservative
strength models currently used in engineering practice. It shares as most deterministic models do. Second, the model does not make
some characteristics with the direct residual strength models and use of a fines correction; the addition of fines content-related terms
some with normalized strength models; it was formulated in a to the predictive model did not result in improved residual strength
way that could have produced results consistent with either if the predictions. This characteristic was also observed by Olson and
data had indicated that either was most appropriate. The residual Stark (2002). Third, the use of a void redistribution potential indi-
strengths back-calculated from the flow-side case history database, cator term that would distinguish between cases where void redis-
supplemented by relevant data from a lateral spreading case history tribution would and would not be expected also did not result in
database, however, provided better support for the proposed model. improved predictions. This observation could be taken to indicate,
The manner in which the residual strength predicted by the as some have suggested, that void redistribution effects occur to
proposed model varies with SPT resistance and effective stress is some extent in virtually all flow-side case histories; the accuracy
consistent with that predicted by a theoretical steady-state frame- of that interpretation, however, cannot be proven with available
work that considers the relationship between soil compressibility data. Finally, the proposed model predicts residual strengths at low
Ross Boulanger, Liam Finn, and Tom Shantz for their constructive
comments on a draft of this paper.
Conclusions
Liquefaction-induced flow sides are complex events with many fac- References
tors affecting the mobilized shearing resistance and deformation on
liquefied soils. The wide and generally unknown variability of these Been, K., Jefferies, M. G., and Hachey, J. E. (1991). “The critical state of
sands.” Geotechnique, 41(3), 365–381.
factors from one case to another prevents back-calculated residual
Boulanger, R. W., and Idriss, I. M. (2011). “Challenges in estimating the in-
strengths from being treated as properties of the soil alone but
situ strength of liquefied soil.” Proc., 8th Int. Conf. on Urban Earth-
rather as parameters that reflect the soil properties and character- quake Engineering, Tokyo Institute of Technology, Tokyo, 9–14.
istics of the individual slopes. Further complicating matters, only a Boulanger, R. W., and Truman, S. P. (1996). “Void redistribution in sand
small number of flow-side case histories have been identified and under post-earthquake loading.” Can. Geotech. J., 33(5), 829–834.
investigated. The case histories encompass a range of soil types, Burlington, R. S., and May, D. C., Jr., (1970). Handbook of probability and
slope geometries, and failure mechanisms, and few have been thor- statistics, McGraw-Hill, New York, 463.
oughly investigated and carefully documented. It is not surprising, Casagrande, A. (1976). “Liquefaction and cyclic deformation of sands: A
therefore, that different approaches have been taken to the difficult critical review.” Harvard Soil Mechanics Series No. 88, Harvard Univ.,
problem of predicting the residual strength of liquefied soil. Cambridge, MA, 27.
Previous residual strength models have fallen into two main Castro, G. (1969). “Liquefaction of sands.” Harvard soil mechanics series
categories: one that assumes that residual strength is a function 87, Harvard Univ., Cambridge, MA.
of penetration resistance alone, i.e., that it is independent of initial Castro, G. (1987). “On the behavior of soils during earthquakes—
Liquefaction.” Soil dynamics and liquefaction, A. S. Cakmak, ed.,
effective stress, and another that assumes residual strength is pro-
Elsevier Science, Amsterdam, Netherlands, 169–204.
portional to initial effective stress. The former models generally
Castro, G. (1991). “On the behavior of soils during earthquakes—
predict higher residual strength at low effective stress levels (shal- Liquefaction.” Proc., NSF/EPRI Workshop on Dynamic Soil Properties
low depths) and the latter predict higher strengths at high effective and Site Characterization, EPRI NP-7337, Vol. 2, Electric Power
stress levels (greater depths). Research Institute, Palo Alto, CA, 1–36.
A third type of residual strength estimation model has been pro- Castro, G., and Poulos, S. J. (1977). “Factors affecting liquefaction and
posed. The proposed model predicts residual strength as a nonlinear cyclic mobility.” J. Geotech. Eng. Div., 106(GT6), 501–506.
function of both penetration resistance and initial effective stress. Castro, G., and Troncoso, J. (1989). “Effects of 1989 Chilean earthquake
The proposed model is consistent with behavior observed in labo- on three tailings dams.” Proc., 5th Chilean Conf. on Seismology and
ratory tests, but was calibrated on the basis of residual strengths Earthquake Engineering, Chilean Association of Seismology and
back-calculated from flow-side case histories. The back-calculated Earthquake Engineering, Santiago, Chile.
strengths considered inertial effects, hydroplaning, and uncertain- Davis, A. P., Jr., Poulos, S. J., and Castro, G. (1988). “Strengths backfig-
ured from liquefaction case histories.” Proc., 2nd Int. Conf. on Case
ties in the individual case histories. The calibration process con-
Histories in Geotechnical Engineering, Univ. of Missouri and Rolla,
sidered the potential effects of fines and of void redistribution, the
MO, 1693–1701.
conditions under which flow sides are known not to have occurred Gutierrez, M., and Eddy, M. (2011). “Probabilistic analysis of liquefied
in past earthquakes, and the relative quality of each case history. shear strength from case histories of slope and embankment failures.”
The proposed model allows estimation of the probability distri- Proc., 5th Int. Conf. on Earthquake Geotechnical Engineering, Chilean
bution of residual strength. The median residual strengths predicted Geotechnical Society, Santiago, Chile.
by the proposed model are consistent with those of direct models Hanzawa, H. (1980). “Undrained strength and stability of quick sand.”
such as Seed and Harder (1990) at low initial effective stress levels, Soils Found., 20(2), 17–29.
and with normalized strength models such as Olson and Stark Horne, W. B., and Joyner, U. T. (1965). “Pneumatic tire hydroplaning and
(2002) and Idriss and Boulanger (2007) at higher initial effective some effects of vehicle performance.” Society of Automotive Engineers
stress levels. At very high initial effective stress levels (>2–3 atm), Technical Paper 970 C, Society of Automotive Engineers, Detroit.
however, the proposed model predicts lower residual strengths than Idriss, I. M. (1998). “Evaluation of liquefaction potential, consequences
the normalized strength models. and mitigation—An update.” Presentation Notes, Vancouver Geotech-
nical Society Meeting, Vancouver, Canada.
Given the lack of case history data, and the significant scatter in
Idriss, I. M., and Boulanger, R. W. (2007). “SPT- and CPT-based relation-
the data that are available, the existence of different models for pre- ships for the residual shear strength of liquefied soils.” Earthquake
diction of residual strength is beneficial in allowing consideration Geotechnical Engineering, 4th Int. Conf. on Earthquake Geotech-
of epistemic uncertainty. When multiple plausible predictive mod- nical Engineering—Invited Lectures, K. D. Pitilakis, ed., Springer,
els are available, expected values of the predicted quantities can be Netherlands, 1–22.
computed as weighted averages of the values given by the different Ishihara, K. (1993). “Liquefaction and flow failure during earthquakes:
models. The 33rd Rankine lecture.” Géotechnique, 43(3), 351–414.