Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: www.elsevier.com/locate/soildyn

T
Examination of differences between three SPT-based seismic soil
liquefaction triggering relationships
K. Onder Cetina, , Raymond B. Seedb, Robert E. Kayenb, Robb E.S. Mossc, H. Tolga Bilged,
⁎

Makbule Ilgaca, Khaled Chowdhuryb,e

a
Dept. of Civil Engineering, Middle East Technical University, Ankara, Turkey
b
Dept. of Civil and Environmental Engineering, University of California, Berkeley, CA, USA
c
California Polytechnic State University, San Luis Obispo, CA, USA
d
GeoDestek Ltd., Ankara, Turkey
e
US Army Corps of Engineers, South Pacific Division Dam Safety Production Center, Sacramento, CA, USA

A R T I C LE I N FO A B S T R A C T

Keywords: The preceding companion paper presented the updating of the seismic soil liquefaction triggering relationship of
Soil liquefaction Cetin et al. [1], and compared the resulting updated relationship with the earlier version. In this second paper, a
Liquefaction triggering relationship detailed cross-comparison is made between three triggering relationships: (1) Seed et al. [2], as slightly updated
Earthquake by the NCEER Working Group (Youd et al. [3]), (2) Boulanger and Idriss [4], and (3) Cetin et al. [5]. Differences
Seismic hazard
between these three triggering relationships, and the apparent causes of them are examined. Also studied are the
Cyclic loading
Standard penetration test
impacts of these differences on levels of conservatism with regard to evaluation of liquefaction triggering hazard,
In-situ test and the resulting risks for engineering projects.
Probability

1. Introduction Fig. 1 shows the established soil liquefaction triggering “boundary

curves” associated with each of these relationships. All three relationships
The preceding companion paper of Cetin et al. [5] presented the have been re-plotted at the same scales to make visual cross-comparisons
updating of the seismic soil liquefaction triggering relationship of Cetin easier and more direct. The liquefaction triggering field case history data
et al. [1], and compared the updated relationship with its earlier version. points plotted in each figure are those of the original authors, and all data
With the aim of developing a fair comparison framework, when com- points (as well as the boundary curves) are normalized to a fines-corrected
piling Cetin et al. [6] database, field case histories from relatively more “clean sand” reference condition of N1,60,CS rather than N1,60.
recent events of 1999 Chi-Chi, 2008 Achaia-Ilia, Greece, 2010 Haiti, Plotting all three relationships on the same scale is helpful with regard
2010 Chile-Maule, 2011 Tohoku, 2010–2011 New Zealand-Canterbury, to making cross-comparisons, but it can be difficult to see in detail some of
2012 Emilia-Romanga (Northern Italy), etc., earthquakes were excluded the differences between the boundary curves of these three relationships.
since they were also not included in Idriss and Boulanger [7] database. Accordingly, Fig. 2(a) shows all three studies, with the BI2012 and
However, the presentation of a further expanded database with these CEA2018 relationships represented by contours of PL = 50%, and Fig. 2(b)
additional new case histories will be the scope of another manuscript. In repeats Fig. 2(a) but with these two probabilistic relationships represented
this second paper, a detailed cross-comparison is made between three by contours of PL = 20%. The SEA1985 relationship had no probabilistic
triggering relationships: (1) Seed et al. [2] as slightly updated by the basis, so the clean sand boundary curve for that relationship remains in the
NCEER Working Group (Youd et al. [3]), (2) Boulanger and Idriss [4], same position in both figures, and serves as a useful visual point of re-
and (3) Cetin et al. [5]. These three triggering relationships will be re- ference. All of these curves shown in Fig. 2 are presented on a “clean sand”
ferred to hereafter as SEA1985, BI2012 and CEA2018, respectively. basis (fines content ≤ 5%). As shown in Fig. 2, there are significant dif-
Differences between these three triggering relationships, and the ap- ferences between the triggering boundary curves at these two important
parent causes of these differences are examined. Also examined are the levels of hazard or probability of liquefaction.
impacts of these differences on levels of conservatism with regard to It must also be noted that examination of the boundary curves alone
evaluation of likelihood of triggering of liquefaction. does not fully characterize overall levels of hazard or conservatism. Each of

⁎
Corresponding author.
E-mail address: [email protected] (K.O. Cetin).

https://doi.org/10.1016/j.soildyn.2018.03.013

Available online 05 June 2018

Received 11 July 2017; Accepted 12 March 2018

0267-7261/ © 2018 Elsevier Ltd. All rights reserved.

K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

Notation list PL Probability of liquefaction

R Distance to source (km) [31]
amax Peak horizontal acceleration rd Stress reduction coefficient
CN Overburden correction S Site class. S = 0 (for rock), S = 1 (for soil site)
CR Correction factor for the rod length [31]
CPT Cone penetration test SPT Standard penetration test
CSR Cyclic stress ratio Vs Shear wave velocity
CSR σv′, Mw, α Cyclic stress ratio at a depth where vertical Vs,12 m Shear wave velocity for the upper 12 m
effective stress and shear stress ratio are σ′v γmax Maximum shear strain
and α due to a Mw magnitude earthquake γbelow-GWT Unit weight below ground water table
CSR σv′= 1atm, Mw = 7.5, α = 0 CSR normalized to σ'v = 1 atm, Mw = 7.5 and γabove_GWT Unit weight above ground water table
α=0 α Initial static driving shear stress ratio;
CRR Cyclic resistance ratio α = τhv,static / σ'v
dcr. d = Critical depth for liquefaction σN1,60 Standard deviation of the N1,60
DR Relative density σln(CSRσv′ , α, Mw ) Standard deviation of ln(CSR σv′, α, Mw )
FC Fines content σln(Mw ) Standard deviation of ln(Mw )
g Acceleration of gravity σFC Standard deviation of ln(FC )
Ko Coefficient of earth pressure at rest σln(σv′ ) Standard deviation of ln(σv′)
Kσ Correction for overburden stress σε Standard deviation of the model uncertainty
KMw Magnitude (duration) scaling factors σ'v Vertical effective stress
Kα Correction for sloping sites σv Vertical total stress
N1,60 Standard penetration test blow count corrected θ Limit state model parameters
for overburden, energy and procedural differ- τav Average shear stress
ences. τhv,cyclic,peak Peak cyclic horizontal shear stress
N1,60,CS Fines -corrected N1,60 value ΔN1,60 SPT penetration resistance correction for fines
M Mw = Moment magnitude content
Pa Atmospheric pressure (1 atm)

Fig. 1. Liquefaction triggering relationships as proposed by (a) SEA1985 as modified slightly by Youd et al. [3], (b) CEA2018 and (c) BI2012 (CSR values are plotted
after correcting for typographical errors described in Boulanger and Idriss [8]).

the three sets of boundary curves are developed to act in conjunction with a relationships, and (3) the recommended associated engineering protocols
number of prescribed or recommended engineering protocols in terms of for forward analyses of projects, in evaluating differences between the three
parameter assessment (e.g. evaluation of earthquake-induced cyclic stress triggering relationships.
ratio (CSR), N1,60 etc.), and with a number of additional (“secondary”) re- Figs. 1 and 2 also show that differences between the three triggering
lationships that result in further adjustments for effective overburden stress relationships are less pronounced at the “upper” portions of the boundary
(σ'v), causative earthquake magnitude (M or Mw), and fines adjustments curves (N1,60,CS ≥ 20 blows/ft). It is important to note, however that (1)
(∆N1,60 as a function of fines content). These “secondary” relationships can the ratios of the differences here (in terms of CSR) are lesser in magni-
also have potentially significant impacts on forward assessments of lique- tude than at the lower portions of the curves applicable to lower pene-
faction hazard for engineering projects. They can either compound or par- tration resistances, and (2) differences at higher N1,60,CS values generally
tially offset levels of conservatism or unconservatism in the baseline have lesser engineering implications because soils with higher corrected
boundary curves shown in Figs. 1 and 2, and their impacts differ over SPT N1,60,CS values have higher post-liquefaction strengths and tend to
varying ranges of parameters. Accordingly, it is necessary to jointly examine have more limited cyclic deformation potential as well.
both (1) the proposed sets of boundary curves, as well as (2) the secondary For most engineered projects, it is soils with lower corrected SPT

76
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

Fig. 2. Clean sand boundary curves (N1,60,CS) for all three relationships co-plotted on the same two figures; (a) the deterministic curve of SEA1985 (as modified
slightly by Youd et al. [3]) and the PL = 50% contours of BI2012 and CEA2018, and (b) the deterministic curve of SEA1985 and the PL = 20% contours of BI2012 and
CEA2018.

penetration resistances (N1,60,CS ≤ 20blows/ft) that are of principal these discussions, any effect which i) increases the CSR or ii) decreases
concern. Unfortunately, it is in this lower range of N1,60,CS values, the N1,60,CS median values of the case histories, or iii) both, will be
where the potential consequences can often be high, that the greatest referred to hereafter as “unconservative” since these effects will trans-
differences between the three triggering relationships occur. late case history data points up and/or left, and consequently cyclic
resistance ratio will be overestimated.

2. Examination of the six main differences between the three

triggering relationships 2.1. Differences in rd relationships

An abbreviated summary of the six main differences between the Differences in the stress reduction (shear mass participation) factor,
three triggering relationships is presented in Supplementary material or rd relationships, is the first of the issues listed in Table S1. The earlier
Table S1, which is a useful summary guide to the discussion that fol- work of Seed et al. [2,9] employed the rd relationship developed by
lows. Similarly, Supplementary material Fig. S1 provides a visual cross- Seed and Idriss [10] in the “simplified” framework, in which CSR was
comparison of input parameters of case histories present commonly in evaluated by using Eq. (3) from the accompanying paper of Cetin et al.
both the Idriss and Boulanger [7] and Cetin et al. [6] databases. Table 6 [5] in back-analyses of case histories. To implement this simplified
in the companion paper of Cetin et al. [5] presents a summary overview approach, first it was necessary to develop the “rd curves”. In early
of average (mean) values of key parameters and indices, for each of the 1970's, the rd curves were developed by analyzing a limited number of
three databases of SEA1985, Idriss and Boulanger [7] and Cetin et al. site profiles. Those profiles consisted of 100 feet (33 m.) of sand, and
[6], and this will also be useful in the discussions that follow. As part of were not representative of the broad ranges of natural site conditions,

Fig. 3. Plots of rd values calculated based on seismic site response analyses by (a) Golesorkhi [11], (b) Imai et al. [12], (c) Cetin and Seed [13], and (d) Seed and Idriss
[10] and Idriss [14]. The red line and red arrows show the rd values estimated at the approximate median critical depth of the databases of Idriss and Boulanger [7]
and Cetin et al. [6].

77
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

which involves variable and often layered stratigraphy, present at many provided by Idriss and Boulanger [7], rd curves of Idriss [14] were
of the liquefaction field performance case history sites employed in the developed by performing one-dimensional equivalent linear seismic site
development the triggering relationship of SEA1985. response analyses. In these analyses, suites of input motions of three
The resulting non-representativeness of the limited number of sand- different magnitudes (Mw = 5.5, 6.5 and 7.5) with varying intensities
only sites was then further compounded because at that early juncture, were applied as input motions to six soil profiles, to conduct a total of
“input” ground motions were typically having lower intensities com- 512 seismic site response analyses. Five of these six sites were again
pared to the ones commonly used today. The rd curves of Seed and monolithically comprised of 100 feet (33 m.) of sand, underlain by
Idriss [10] are considered to be non-representative of the field case “rock”. Thus, it is considered they are non-representative of the lique-
history sites and the shaking levels of many of the case histories in the faction case history database sites, and they again produce overly “stiff”
SPT-based soil liquefaction triggering catalog. These rd values tend to rd behaviors. The sixth site was a modification of the USGS strong
systematically overestimate CSR. That, in turn, causes the field case motion recording site at La Cienegain Los Angeles, California. This is a
history data points to be plotted “too high” (vertically) on plots like site with actual stratigraphy, but the Vs profile at this site was "mod-
Fig. 1(a), which then unconservatively biases (i.e.: shifts liquefaction ified", and due to this modification it produces even a “stiffer” rd be-
triggering boundary curves up) the resulting triggering relationship. havior (rd values further to the right). Additionally, the “representative”
Prof. H. Bolton Seed's last Ph.D. student, Dr. Ramin Golesorkhi, rd curves developed based on the results of these analyses were not
performed one-dimensional site response analyses using both the selected at the mean or median values, but rather at the 65th-percentile
equivalent linear approach and also fully nonlinear models, to evaluate values. This selection further shifted the resulting rd curves further to
these rd effects and also to assess the ranges over which the equivalent the right and it is not complying with the use of mean input parameters
linear analyses would be an adequately reliable proxy for nonlinear within the maximum likelihood framework employed to develop a li-
behaviors in a variety of other applications as well. Dr. Golesorkhi quefaction triggering relationship.
applied his rd-related response analyses mainly to monolithic sites again As shown in Fig. 3, rd curves of Idriss [14] are not in good overall
comprised entirely of sand strata. Fig. 3(a) shows the rd curves devel- agreement with the curves developed by the other three research
oped by Golesorkhi [11] based on his analyses at sand-only sites. This groups, and they produce values that are biased to the high side at all
figure is re-scaled to the same vertical and horizontal axes as the other depths.
site response analysis results shown in Fig. 3(b), (c) and (d) for direct An important pair of postulates affecting this and other elements of
comparisons. back-analyses of liquefaction performance field case histories for pur-
Imai et al. [12] had also advanced the assessment of rd, and per- poses of developing liquefaction triggering relationships are listed as
formed a total of 143 one-dimensional site response analyses based on follows;
multiple reflection theory by using 5 different input motions (with
Postulate 1. It is often true that decisions and approximations that
maximum horizontal accelerations varying in the range of
would be “conservative” in forward engineering analyses of actual
0.052–0.233 g), for a suite of layered soil sites (with actual strati-
projects (e.g. over-estimation of rd, and of resulting CSR values), are
graphy) to investigate this issue. Their results are also presented in
instead un-conservative when applied to the back-analyses of field
Fig. 3(b) and again scaled to matching axes.
performance case histories for purposes of developing triggering
Cetin and Seed [13] performed a significantly larger number of site
relationships.
response analyses to real sites with real stratigraphy as described in the
companion paper to approximate nonlinear response effects. Results of As an example here, the over-estimation of rd (and of CSR) that
these 2153 site response analyses are presented in Fig. 3(c). These re- resulted from using curves developed for the non-representative and
sults differed from the other three sets of rd curves presented in Fig. 3 as overly “stiff” site conditions would have been conservative in a forward
(1) all 50 of the analyzed sites were actual ones, they were from the analysis for an actual engineering project. But in back-analyses of li-
larger liquefaction field performance case history database, variable quefaction performance field case histories, the overestimation of rd
conditions and stratigraphy were covered so they were considered to be (and of resulting CSR's) serves instead to displace the plotted case
“representative”, and (2) a suite of 42 carefully selected input motions histories vertically upwards on the eventual "boundary curve" plots (e.g.
was applied to all 50 of these sites (the input motions cover the range Fig. 1(a), (b) and (c)), producing an unconservative bias in the trig-
from low to high intensities of shaking from low to high magnitude gering relationships developed based on these data. This same principal
causative events, and they include actual strong motions that are re- applies to multiple other parameters and relationships, not just rd and
presentative of near-, mid- and far-field events in each magnitude CSR, and so it will be discussed further throughout this manuscript.
range. In addition, for 53 of the liquefaction field case history sites
Postulate 2. Engineers are often taught that it is important to "repeat
where a nearby strong motion recording was available to be scaled for
the same mistakes going forward (executing engineering analyses of
use as an “input” motion, actual site- and earthquake event-specific site
real projects) as were made going backwards (e.g.: back-analyzing field
response analyses were performed. Regressions were then conducted to
case histories for development of triggering relationships)". But it does
develop rd relationships as a function of: (1) site stiffness and strati-
not guarantee that errors or bias in the back-analyses will be suitably
graphy, (2) levels and intensity of shaking, (3) causative earthquake
mitigated in forward engineering analyses for a specific site or
magnitude (as a proxy for duration effects), and (4) depth.
engineering project.
The dark dots in Fig. 3(d) show the middle values of the early re-
commended rd curve of Seed and Idriss [10], which was not magnitude Unconservative bias in back-analyses is not necessarily mitigated
dependent. This early curve was developed from stiff monolithic (sand) (or counter-balanced) by performing forward analyses using identical
sites lacking of layering or stratigraphy, and lower levels of shaking. protocols. The overall bias from inappropriate back-analyses is largely
Moreover, the associated range is narrow and does not span as broadly “averaged” into the relationships developed, but forward analyses will
as the other three suites of analysis results. The differences are sig- be applied to a specific project site and conditions rather than to an
nificant at the relatively shallow depths of principal interest for back- “average” site with average conditions. It is true that some counter-
analyses of the liquefaction field performance case histories. As shown balancing occurs if the “simplified” (e.g. the rd-based) approach is used
in Table 6 of the companion paper of Cetin et al. [5], the mean depth of to evaluate CSR's in situ for forward analyses of an actual project, and if
critical strata tends to be in the order of approximately 5 m for the the same rd curves are employed as were used for the back-analyses
liquefaction field performance case history database. upon which the triggering curves are based. But the degree of counter-
The suite of four solid lines in Fig. 3(d) present the rd curves de- balancing is variable, and dependent upon the juxtaposition of actual
veloped and recommended by Idriss [14]. According to the information site conditions vs. the “average” conditions in the liquefaction field

78
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

as an important issue in the mid-1970's and the early 1980's because

triggering relationships were increasingly being applied to analyses of
major earth dams (with potentially liquefiable strata under high ef-
fective overburden stresses). Laboratory research and principles of
critical state soil mechanics show that soils of a given density (or nor-
malized SPT penetration resistance) would be more likely to liquefy at
the same CSR if they were under higher initial effective overburden
stress, because the higher effective stress would suppress cyclic dilation
and increase cyclic compression on shear stress reversal during cyclic
loading.
Accordingly, research efforts were undertaken to develop Kσ re-
lationships based on laboratory undrained cyclic testing. By the second
half of 1980's, (1) a body of laboratory test data were available, and Kσ
relationships were being proposed by different sets of experts, and (2) it
was becoming increasingly common practice to assume that the
“shallow” liquefaction triggering relationships developed based on field
performance case history data were suitably representative for condi-
tions conforming to σ'v ≤ 1 atm, and then Kσ relationships would be
applied to these triggering “curves” to extrapolate in order to evaluate
liquefaction resistance (e.g. CRR) at higher initial effective stresses
(greater depths) where σ'v > 1 atm.
Fig. 4. The recommended Kσ relationships of (1) Youd et al. [3] as appended to The NCEER Working Group (Youd et al. [3]) proposed a Kσ re-
the triggering relationship of SEA1985, (2) BI2012, and (3) CEA2018. lationship of this type for application to the triggering relationship of
SEA1985. This Kσ relationship was based on cyclic undrained labora-
performance database. tory test data, and Kσ was a function of both σ'v and also relative den-
More importantly, for sites and projects of sufficient importance, or sity. That was a suitable approach based on the state of knowledge at
sufficient challenge and complexity, that site-specific seismic site re- that time. This Kσ relationship is shown in Fig. 4.
sponse analyses (or even site-specific site and soil-structure interaction Boulanger [16], Boulanger and Idriss [17] and Idriss and Boulanger
seismic response analyses) are warranted, in-situ CSR's are then directly [7,18,19] produced less conservative Kσ relationships for extrapolating
(and correctly) calculated in these more detailed site- and project-spe- triggering curves that had been developed based on shallow data to
cific forward analyses (employing higher-order seismic response ana- higher overburden stresses. These Kσ relationships were again based on
lyses to directly calculate CSR at each point of interest, rather than interpretations of available laboratory cyclic test data and are a func-
employing the “simplified” rd curves approach), and thus no counter- tion of both σ'v and also N1,60,CS. These relationships provide for a lesser
balancing (or compensating) error occurs in the forward analyses. This, rate of decrease in liquefaction triggering resistance (CRR) with in-
in turn, means that the unconservative bias associated with the use of creasing σ'v. This is also a valid approach.
non-representative rd curves in back-analyses and formulation of the Cetin et al. [1,5] recognized the difficulties and uncertainties in-
triggering relationship is not totally compensated for in large, chal- volved in extrapolating the results of (1) undrained cyclic triaxial tests,
lenging, or complicated projects that can often be among those of and (2) of undrained cyclic simple shear tests with uni-directional
greatest importance (e.g. large dams, major buildings, significant in- horizontal loading, both with uniform cycles, for evaluation of the fully-
frastructure, challenging site conditions, etc.). This issue was also re- multi-directional (in a horizontal plane) and non-uniform cyclic shear
cognized by Boulanger and Idriss [19] and their liquefaction triggering loadings of principal interest for prediction of field behavior in real
correlations were recommended to be used only with the same re- earthquakes. Accordingly, they preferred developing a Kσ relationship
lationships that were used in the development of their correlations (e.g.: as part of the overall probabilistic regressions of the full liquefaction
only with their rd relationship; but not with a site-response-estimated rd triggering field performance case history database, which intrinsically
or CSR). embodies these three-dimensional and irregular “real” world seismic
cyclic loadings.
The resulting Kσ relationship of Cetin et al. [5] is a function of σ'v
2.2. Differences in Kσ relationships
only, and it is shown by the solid red line in Fig. 4. An additional
variable was regressed, as part of the overall regression of the field case
The second issue addressed in Table S1 is the development (or se-
history data, to investigate whether an effect of N1,60,CS could be dis-
lection) of Kσ relationships for normalization of CSR's at arbitrary ef-
cerned from the large field case history database. However, only a very
fective overburden stresses to “equivalent” CSR that would just trigger
slight influence of N1,60,CS on Kσ could be distinguished by this effort.
liquefaction at an effective vertical stress of σ'v = 1 atm by using Eq.
Cetin and Bilge [20] performed a large number of undrained cyclic
(1). Kσ is directly a function of vertical effective stress yet some studies
simple shear tests to investigate Kσ, and discovered that this scaling
also use relative density (or N1,60,CS) while defining this stress scaling
factor is strongly a function of the cyclic shear strain level at which
factor.
“triggering” of liquefaction or cyclic mobility is deemed to have oc-
CSR σv′, α, Mw τhv,cyclic,average 1 a σ 1 curred. If the conventional levels of laboratory cyclic shear strain are
CSR σv′= 1atm ., α, Mw = = = 0.65⋅ max ⋅ v ⋅rd⋅ employed as “triggering strain” criteria for all tests (regardless of
Kσ σv′ Kσ g σv′ Kσ
density or N1,60,CS), then the Kσ curves of Youd et al. [3], Idriss and
(1)
Boulanger [7] and Cetin et al. [5], appear generally reasonable in form.
The early triggering relationships (e.g. Seed [15] etc.) were not However, the field case history data likely represent differing levels of
normalized for vertical effective stress, because it was tacitly re- cyclic strains. It should also be noted that denser strata (with higher
cognized that they were based on liquefaction field performance case N1,60,CS values) typically requires more limiting cyclic shear strain for
histories for “shallow” soil strata; as discussed previously. The re- triggering of liquefaction as discussed by Cetin and Bilge [21]. It could
lationship of SEA1985 was the most recent of these early relationships, then be inferred, based on the laboratory data set of Cetin and Bilge
and it was also not normalized for Kσ effects. However, it had emerged [20] that a slightly “inverted” relationship between Kσ and N1,60,CS

79
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

might be expected; with the effect that increasing N1,60,CS values affects σ'v≈ 1 atm, but the field case history database of SEA1985 was actually
Kσ in a manner slightly inverse to the trends posited by Youd et al. [3] more closely correlated with an overall average field case history stress
and others, as was actually observed in the regression of the liquefac- level of σ'v = 0.67 atm. Assuming that it was 1 atm, and then appending
tion triggering field case history database. However, there still exist a Kσ relationship from 1 atm to progressively higher values of σ'v had
significant uncertainty associated with this issue. Thus, pending addi- approximately the equivalent effect of “truncating” Kσ to Kσ ≤ 1.0, as
tional investigation by other researchers, it was decided to take a illustrated in Fig. 4 for the relationship recommended by Youd et al.
middle position in developing the triggering relationship of CEA2018 [3]. Based on Postulates 1 and 2, the truncation of Kσ would be con-
and employ a Kσ relationship (1) that is a function of σ'v only, (2) that is servative for forward analyses of actual projects with low effective
regressed from the overall field performance case history database, and stresses. Yet, for back-analyses of liquefaction field case histories and
(3) that is not necessarily recommended to be used outside the vertical for subsequent development of liquefaction triggering relationships,
effective stress range of the field performance case history database. As this type of truncation creates a significant unconservative bias in the
discussed in the companion paper, the Kσ relationship of Cetin et al. [5] resulting triggering relationships. Truncating at Kσ ≤ 1.0 has the effect
is employed internally within the development of the triggering re- of increasing the overall average “normalized” CSR values which were
lationship, and it is used mainly to correctly “center” (or normalize) the back-calculated from the case histories, because (1) a majority of those
relationship to a reference effective overburden stress of σ'v = 1 atm. cases had σ'v < 1 atm, and thus Kσ > 1.0, and (2) in back-analyses the
Extrapolation of the resulting normalized triggering relationship can back-calculated CSR is multiplied by 1/Kσ before plotting the results in
then be accomplished by means of any of a number of other proposed the triggering boundary curve plots of Fig. 1. That, in turn, means that
Kσ relationships. This will be discussed further. truncation of Kσ biases the triggering boundary curves unconservatively
The histogram given in the lower part of Fig. 4 shows the dis- by pushing CSR values vertically upwards in plots like those of Fig. 1(a)
tribution of vertical effective stresses in the field case history database through (c).
of Cetin et al. [6], which is for these purposes very similar to that of Cetin et al. [1] addressed this issue and did not truncate Kσ values
Idriss and Boulanger [7]. The case history regression-based Kσ re- for back analyses of case histories and subsequent development of their
lationship of Cetin et al. [5] agrees reasonably well with the relation- triggering relationship to mitigate truncation error. However, a limit of
ship proposed by Youd et al. [3] and available cyclic laboratory test Kσ ≤ 1.5 was then recommended at very shallow depths for forward
results as presented in Fig. 4. On the other hand, The Kσ relationship engineering assessments. Similarly, Cetin et al. [5] specifically did not
proposed by Idriss and Boulanger [7] shown in Fig. 4 does not employ apply a truncation of Kσ in processing the case history back analyses.
field case history data. Instead, it is based on the assessment of un- However, for forward (design) assessments, it is recommended that Kσ
drained cyclic laboratory test data and engineering judgment. For the to be limited (truncated) to Kσ ≤ 1.6. This affects only soils at very
purpose of enabling a direct visual comparison, available cyclic la- shallow depths, and very low effective overburden stresses
boratory test results as compiled and interpreted by Montgomery et al. (σ'v < 0.25 atm). Only 5 of the 210 field performance case histories
[22], which would be later referred by Boulanger and Idriss [8] to be in back-analyzed by Cetin et al. [5] would have been affected by this Kσ
reasonable agreement with their Kσ relationship, are also shown in ≤ 1.6 truncation if it had been employed in the development of the
Fig. 4, along with additional data developed by Cetin and Bilge [20] triggering relationship, but it was not.
based on undrained cyclic simple shear testing. The data set gathered Idriss and Boulanger [23] elected to truncate their Kσ relationship at
and compiled by Montgomery et al. [22] was based on both cyclic Kσ ≤ 1.0, and later Idriss and Boulanger [19] truncated at Kσ ≤ 1.1.
triaxial and cyclic simple shear testing, with a majority of the data Out of 230 cases 47 are affected (and unconservatively biased) due to
developed by cyclic triaxial testing. this truncation. Without truncation these 47 case history points would
There are significant uncertainties with regard to interpretation of have produced normalized CSR values that would have plotted lower
cyclic triaxial test data for purposes of development of liquefaction on the CSR vs. N1,60,CS triggering curve plots. Examining their overall
triggering Kσ relationships for forward application to field conditions field case history database, it is observed that Kσ truncation effects
that will involve both non-uniform and three-dimensionally, randomly appear to be more pronounced for N1,60,CS > 20 blows/ft, and the re-
directionally varying in plan view (3-D) cyclic simple shear loadings. sulting triggering curves would thus likely be somewhat more affected
There are lesser levels of uncertainty in employing the more limited (unconservatively biased) at this larger range of N1,60,CS values due to
one-directional uniform cyclic simple shear test data that make up part this truncation.
of the data set compiled by Montgomery et al. [22] and all of the data There is no physical reason or basis for truncation of Kσ in back-
developed by Cetin and Bilge [20], but these uncertainties are not analyses and development of triggering relationships. More sig-
negligible. nificantly, it is simply a straightforward imposed constraint (or bias)
The Kσ curves of Youd et al. [3] were developed by a large group of that produces unconservative bias in the resulting triggering relation-
researchers involving many of the world's top liquefaction experts. It is ships. Both Postulates 1 and 2 apply here, and either (1) not correcting
suggested that engineers might consider this overall situation (and for Kσ effects (failing to “center” or normalize the triggering relation-
Fig. 4) and then judge that the Youd et al. [3] curves still appear to be a ship correctly at σ'v = 1 atm) as is commonly done when extrapolating
reasonable and defensible basis for forward analyses for very high σ'v. It the “shallow” relationship of SEA1985 to higher levels of effective
is the view of the authors of this paper that more research is needed overburden stress, or (2) truncating at either Kσ ≤ 1.0 or Kσ ≤ 1.1, as
here, and that in the face of current uncertainty it would be good to was done in the development of the triggering relationships of Idriss
avoid potentially serious unconservatism; especially at very high ef- and Boulanger [23] and of BI2012, respectively, are considered as
fective stresses, as those high stress levels often occur in conjunction sources of unconservative bias in the liquefaction triggering relation-
with major dams or other critical and/or potentially high risk struc- ships.
tures.
2.4. Differences in probabilistic treatment used in the development of the
2.3. Differences in truncations of Kσ relationships triggering relationships

Truncation of Kσ is the third issue addressed in Table S1. As dis- Probabilistic treatment in the development of the triggering re-
cussed earlier, it had often been assumed that the earlier “shallow” li- lationships is the fourth main difference between the three relationships
quefaction triggering relationships (e.g. SEA1985) were appropriate to addressed in Table S1. SEA1985 did not perform formal probabilistic
σ'v ≤ 1 atm, and as a result it had become somewhat standard practice regressions. Instead they plotted the data in several “bins”, separating
to assume that those triggering curves were representative for them based on fines contents, and then hand-drew three sets of

80
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

boundary curves for data with (a) FC ≤ 5%, (b) 5% < FC < 35%, and model error. This was a challenging analytical effort. In the end, a total
(c) FC ≥ 35% based on engineering judgment. Prof. H. Bolton Seed had of seven model fitting coefficients (θ1 to θ6 and σє) were regressed in a
targeted these three boundary curves at approximately a 10–20% manner similar to that employed by Cetin et al. [1], and an eighth
probability of liquefaction (Seed RB (1988) personnel communication), model coefficient (θ7 ) addressing the relative distributions of input
but he recognized that (1) the sparseness of the available field perfor- parameter uncertainty and model error was also (simultaneously) re-
mance case history data, and (2) the lack of a formal probabilistic re- gressed in the same overall maximum likelihood regression as discussed
gression, raised questions as to the degree to which this target was in the companion manuscript of Cetin et al. [5].
actually met. As shown in Fig. 1(b) and (c) the resulting shapes of the median (PL
Boulanger and Idriss [4] state that their liquefaction triggering = 50%) boundary curves for the two regressed relationships differ
boundary curves are based on regressions performed by the maximum significantly, and so do the associated contours of PL = 5%, 20%, 50%,
likelihood method. The formal application of the maximum likelihood 80% and 95% for these two relationships. The “spread” of these prob-
method would require at least the following two attributes to be in- abilistic contours is a representation of the “uncertainty” (or model
corporated in the required assessments and analyses: (1) the regressions errors) of these two triggering relationships.
would have to be performed using functional shapes (or equational Fig. 5 is an enlarged view of Fig. 1(c), and it is annotated with a
forms) providing both (a) suitable general characteristics and also (b) vertical dashed line at N1,60,CS = 20 blows/ft, so that the impacts of the
sufficient (regressable) degrees of freedom so that the overall re- three principal differences in the regression performed by BI2012 can
lationship could suitably adapt its shape and position to conform to the be more clearly examined.
dictates of the data, and (2) the regressions would have to employ The first of these is the failure of the mean (PL = 50%) triggering
correct and appropriate treatment of both (a) model error or un- boundary curve to suitably “fit” the field case history data at low values
certainty, and (b) input parameter uncertainty. of N1,60,CS ≤ 20 due primarily to the overly “rigid” equational shape
As shown in Fig. 1(c), the functional shape employed for the de- selected and the lack of sufficient degrees of regressible model fitting
velopment of BI2012 “boundary curves” has an equation of a “bent” parameters. Upon close inspection, even when weighting factors of 0.8
shape much like a boomerang, with somewhat more curvature near the and 1.2 are applied to the "Liquefied" and "Non-Liquefied" cases, re-
middle and less at the two ends. This selected functional shape (equa- spectively, the uneven (unconservatively biased) positioning of PL
tional form) has only one single regressable degree of freedom (model = 50% boundary curve in the range of N1,60,CS ≤ 20, relative to the
coefficient) which named as C0 by Boulanger and Idriss [4]. As a result, data as developed and plotted by BI2012, can be visually observed.
based on regression, this shape can only translate in one single direction Re-examining Fig. 1(b) and (c), it can be noted that both prob-
(vertically), but it cannot independently translate in two orthogonal abilistic relationships have PL-based boundary curves that are relatively
directions (e.g. laterally and vertically), and rotate or adjust its bend or closely spaced (indicating relatively low levels of uncertainty) in their
curvature. This lack of sufficient degrees of freedom represents a “stiff” bottom left corners. As the curves begin to rise, the uncertainty (and
regression, in which the resulting regressed relationship cannot well thus also the “spread” between the PL-based curves) begins to increase.
adapt itself to conform to the dictates of the database. This violates the In the Cetin et al. [5] relationship, this spread continues to further in-
first of the two requirements of a maximum likelihood regression as set crease as the curves rise to the top right extent of the figure. But in the
forth above. BI2012 relationship, the spread between the PL-based curves initially
A second issue is that the input parameter uncertainty of each case increases, and then begins to decrease again and the curves draw closer
history needs to be consistently incorporated, and then both input together in the top right corner.
parameter uncertainties and overall model uncertainty need to be sui-
tably handled. The performance of a maximum likelihood regression is
governed by the need to correctly model both (a) model error, ε , and (b)
input parameter variance or uncertainty of individual case histories
(i.e.: uncertainties in N1,60, i , CSRσ ′v , α, Mw, i , Mw, i , FC i,σ ′v, i of the ithcase
history, represented by standard deviations designated as σN1,60, i ,
σln(CSRσ ′v , α, Mw ), i , σln(Mw ), i , σFC , i and σln(σ ′v ), i etc.). Neither the model error
nor the variance or uncertainty of individual case history input para-
meters were fully and systematically modeled within the maximum
likelihood regressions performed by BI2012. Instead the standard de-
viation of the model error was assumed to be 0.13. As presented and
discussed in Cetin [24], the model coefficients of liquefaction triggering
relationships are also inter-correlated; hence assuming a value for one
model coefficient (e.g.: assuming standard deviation of model error, σє
as 0.13) inevitably constrains the other model coefficients (e.g.: C0 in
BI2012), even if an attempt is made to independently (but actually
conditionally) regress them due to intercorrelation of these coefficients.
Cetin et al. [5], in contrast, repeated the same level of effort that
had previously been devoted to the estimation of individual input
parameter error (or uncertainties) in processing and back-analyses of
the field performance case histories by Cetin et al. [1]. The results
differed slightly from the 2004 estimates due primarily to (1) differ-
ences in the case histories included in the database, and (2) differences
in some of the details involved the back-analyses of the case histories.
As discussed in the companion paper of Cetin et al. [5], it was then
necessary to appropriately distribute the overall uncertainty between
input parameter uncertainty and model error. An element of the solu-
tion here was to extend the regression (still by the maximum likelihood
method) to include determination of the most appropriate (maximum Fig. 5. The probabilistic triggering relationship boundary curves of BI2012, and
likely best fit) distribution between input parameter uncertainty and their field case history data points.

81
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

Additional investigators have performed similar probabilistic re- 2.6. Differences in transparency of case history processing and
gressions for liquefaction triggering relationships, employing a variety documentation
of different field liquefaction performance case history data sets.
Significant examples include: Liao et al. [25], Youd and Noble [26], The sixth issue addressed in Table S1 is “Transparency”, and with it
Toprak et al. [27] and Juang et al. [28]. the corollary issue of appropriate technical review. Transparency refers
Figs. S2(a) and S2(b) show two of these four additional relation- to the adequacy and transparency of documentation of (1) the back-
ships. Fig. S2(a) shows the relationship developed by Liao et al. [25], ground source data, (2) the selection, processing and analyses of those
which was a noteworthy early effort of this type, by early experts in the data, and (3) the many details and judgments made at each stage along
field of geotechnical probability and reliability. Fig. S2(b) shows the the way in developing these types of complicated triggering relation-
more recent relationship developed by Juang et al. [28] by using the ships. Better, or worse, background documentation and transparency
Bayesian mapping approach. All four of these previous relationships does not directly affect the likely bias of a given triggering relationship.
developed suites of probabilistic boundary curves with (1) significantly But it makes it easier for other engineers and researchers to understand,
higher overall uncertainty, and (2) all four of them produced boundary and to back-check and review, the development of these types of im-
curves with the lowest uncertainty in the lower left corner (at low pe- portant relationships.
netration resistances) and with then progressively increasing model Cetin et al. [1] tried to be comprehensively transparent in the pre-
uncertainty towards the upper right corner, where the highest un- sentation of the background development of their triggering relation-
certainty occurs at the highest penetration resistances. ship. As a result, it was studied by other engineers and investigators,
The authors of this current paper have, as an ensemble, been per- and issues or errors were identified and challenged. As a consequence,
sonally involved in developing a number of the liquefaction triggering Cetin et al. [5] were able to revise their derivations, employing a field
field case histories in the databases discussed. An important lesson from case history database that was well-checked and vetted by other ex-
those field investigations is the relatively high level of difficulty often perts. That is an ideal outcome; and the authors of this paper would
involved in determining whether or not a site “triggered” with regard to suggest that full and transparent documentation should be a funda-
liquefaction when penetration resistances are high (N1,60,CS ≥ 20–25 or mental requirement for all similar efforts to develop engineering ana-
so), because (1) these higher blowcount soils have limited cyclic strain lysis tools for important problems with broad ramifications for public
potential, and (2) they undergo lesser levels of post-earthquake volu- safety.
metric reconsolidation and so exude lesser levels of boil fluids and as- SEA1985 had also provided what was, at that time, an open and
sociated soil ejecta. As a result, uncertainty is intrinsically higher in this transparent documentation of their data and analyses. Their case his-
upper blowcount range as a result of uncertainty in characterizing ob- tory database was significantly smaller than the more recent efforts,
served field performance with regard to triggering (or non-triggering). and the relationship developed was “simpler” and lacked a formal
probabilistic basis. The level of documentation transparency was not as
extensive as that of Cetin et al. [1], and hence full and complete in-
2.5. Differences in probabilistic treatment in the development of Kσ, KMw, dependent examination of all details was not possible for a number of
and fines adjustments (∆N1,60) relationships their case histories, but most cases could be suitably checked and ex-
amined. Moreover, the following steps taken in development of the
Probabilistic treatment in the development of “secondary” re- triggering relationship were clearly explained.
lationships dealing with (1) effective overburden effects (Kσ), (2) cau- The work of Idriss and Boulanger [23] could not be properly
sative magnitude (duration) scaling effects (KMw), and (3) fines ad- checked or technically reviewed due to lack of documentation. The
justments (∆N1,60), is the fifth of the major issues addressed in Table S1. missing background documentation was later presented as a U.C. Davis
SEA1985 employed no formal probabilistic approaches in the devel- research report by Idriss and Boulanger [7]. With the release of this
opment of their main triggering curves, and they also used no formal 2010 document, it was possible to know which case histories were in-
probabilistic approaches in the development of their relationships for cluded in the development of their liquefaction triggering relationships.
these three additional issues. Relatively complete background details were presented, and that could
BI2012 preferred not to employ probabilistic regressions to develop, then be traced and checked for 101 of the 230 case histories used in
or assist in the development of, “secondary” relationships dealing with their relationships. However, the remaining 129 case histories are dif-
(1) effective overburden effects (Kσ), (2) causative magnitude scaling ficult to be fully evaluated and back-analyzed.
effects (KMw), and (3) fines adjustments (∆N1,60). A combination of non-
probabilistic regressions and engineering judgments were instead used 2.7. Additional differences between the three triggering relationships
to develop these three secondary relationships, and the resulting re-
lationships are discussed in Table S1. Table S2 lists 11 additional issues and factors that result in differ-
Cetin et al. [5] followed a significantly different set of approaches ences between the three triggering relationships. They are not all of the
here. All three relationships (Kσ, KMw, and ∆N1,60) were developed as remaining factors. Instead they are selected either because they can
part of a combined overall regression of the full field case history da- make a potentially non-negligible difference in certain ranges of ap-
tabase along with the development of other triggering relationship plications, or because engineers tend to ask about them and their ef-
elements (e.g. the probabilistic triggering boundary curves, etc.). As a fects. The issues discussed in Table S2 are generally less significant than
result, (1) the large field performance case history database was a those of Table S1. As this manuscript is over length, an indepth dis-
dominant contributor to the defining of these “secondary” relation- cussion of these additional differences are presented in Cetin et al. [6]
ships, and (2) the resulting “secondary” relationships were intrinsically which can be accessed online at http://users.metu.edu.tr/ocetin/
compatible with the overall probabilistic regressions performed, and Database_Report_2016.pdf, and they will not be repeated in full
the resulting probabilistically-based triggering relationships. Two of the herein. Instead, only two of the additional eleven issues will be briefly
resulting case history-based relationships (Kσ, and KMw) were then discussed here. The additional factors discussed in Table S2 are num-
checked and were confirmed to agree suitably well with fundamentally bered 7 through 17. Of these, Factor No's. 13 and 16 will be discussed
different and independently developed laboratory testing-based re- below.
lationships and data, as discussed previously.
2.8. Fines adjustment, (∆N1,60,CS)

Each of the three sets of triggering relationships employs a

82
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

correction (or adjustment) for effects of fines content. The fines cor-
rections developed and adopted by SEA1985, BI2012, Cetin et al. [1]
and CEA2018 are presented and compared in Fig. 6.
There were significant differences in the procedures employed to
develop these fines corrections, and in the case history data sets upon
which they were based. Seed et al. [2] had sparse field case history
data, and they manually plotted the data (in CSR vs. N1,60 space) with
the data “binned” into three sets of cases with (1) FC < 5%, (2)
5% ≤ FC ≤ 35% (assumed to represent approximately FC = 15%), and
(3) FC > 35%. They then drew lines, by hand, based on engineering
judgement, to develop the three triggering boundary curves shown in
Fig. 1(a). No formal regressions were performed. Prof. H. Bolton Seed
later became concerned that subsequent accumulating data suggested
that the resulting (inferred) fines adjustments were a bit too large, and
he employed a slightly lower set of fines adjustments (∆N1,60) in his
final paper (Seed [29]).
Fig. 7. Magnitude (duration) scaling factors.
The basis of Boulanger and Idriss fines correction scheme cannot be
fully traced. The presentation in Idriss and Boulanger [19] indicates
that they initially repeated this same “binning” process, employing the developed. It is important to employ fines adjustments compatible with
plotting of three sets of binned data and then the hand-drawing of the triggering relationship selected.
several sets of boundary curves based on engineering judgment, as a
basis for inferring their new fines corrections.
2.9. Magnitude-correlated duration (KMw) correction
Cetin et al. [1,5] developed fines corrections based on the overall
(formal) probabilistic regressions of the large field performance case
Scaling of the triggering relationships for numbers of equivalent
history database, so that these fines corrections are the only set of
uniform cycles of seismic loading (or duration of shaking) is another
formally regressed fines corrections among the three sets of triggering
correction that affects liquefaction triggering assessments, particularly
relationships examined in this paper. The purpose of performing formal
for small magnitude cases. Duration, or numbers of cycles, are corre-
regressions is to develop keener and more reliable insights than what
lated with causative earthquake magnitude, so the correction factor
can be obtained based on visual judgments.
employed here is referred as KMw. As shown Fig. 7, there exist sig-
The fines adjustments (∆N1,60) of SEA1985, shown in Fig. 6, are a
nificant differences among KMw recommendations proposed by various
function of N1,60 (or CRR) and they are the largest of the four fines
research teams for very low magnitudes (e.g. Mw = 5.5), but differ-
adjustment relationships shown. The fines adjustment of Idriss and
ences can also be significant at very high magnitudes (e.g. Mw = 8.0
Boulanger [18,19] does not vary as a function of N1,60, and it is of
and greater).
intermediate size (generally lower than that of SEA1985 and generally a
The KMw relationships of (1) Idriss [30] which defines the lower
bit higher than that of CEA2018, except at very high values of N1,60).
bound of the range recommended by the NCEER Working Group (Youd
On the other hand, the fines adjustments of Cetin et al. were regressed
et al. [3]) for application to the triggering relationship of SEA1985 and
based on the field case history database. Fig. 6 reveals that the fines
(2) BI2012 were both developed based on (i) laboratory undrained
adjustments of SEA1985 were indeed a bit on the large side, and that
cyclic test data and (ii) processing of large numbers of strong ground
there is somewhat better (but imperfect) agreement between the fines
motion recordings, and no use was made of the liquefaction triggering
adjustments of Idriss and Boulanger [18,19] and those of CEA2018.
field case histories. The KMw relationships of Cetin et al. [1,5] are based
As shown in Table 6 of the companion paper Cetin et al. [5] and Fig.
on regressions of the large liquefaction triggering field case history
S1, the corresponding difference is only approximately 2.4% in the
databases, and make no use of laboratory data. As a result, these two
resulting overall "median" fines corrected N1,60,CS values for the data
sets of relationships are based on fully independent sets and types of
sets of Idriss and Boulanger [7] and Cetin et al. [6]. These are relatively
data.
modest, but non-zero, differences and they would be expected to have
The relationship of CEA2018 is located between the relationships of
relatively modest effects on the overall triggering relationships
Idriss [30] and Boulanger and Idriss [8], as shown in Fig. 7. There is

Fig. 6. Comparative illustration of fines correction.

83
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

relatively good agreement among these despite the very different ap- 4. The early fines adjustments of this relationship were not developed
proaches, and fully independent data sets, upon which they are based. based on regression, but instead they were estimated based on
In addition, all of the large field case history data sets discussed here manual plotting of the field data and engineering judgment. The
have median values of magnitude approximately equal to 7.1. As a sparse available field data led to an initial level of fines adjustments
result, the impacts of KMw on differences between the triggering re- that the late Prof. Seed later came to view as somewhat un-
lationships is relatively small. conservative (too large) as additional liquefaction triggering field
data continued to become available. That was later confirmed by the
3. Conclusion subsequent relationships developed by both Boulanger and Idriss [4]
and by Cetin et al. [1,5] which developed smaller fines corrections
For a given N1,60,CS value, the corresponding cyclic resistance values based on the significantly larger field case history databases avail-
(i.e.: CRR) associated with any target level of likelihood of liquefaction able to them.
triggering based on the relationships of (1) SEA1985, (2) BI2012 and 5. There was no formal probabilistic basis for this triggering relation-
(3) Cetin et al. [5] are observed to be significantly different. These ship. Thus, it is not known with any accuracy what level of prob-
differences occur at essentially all locations on the triggering curves, ability of liquefaction triggering is represented by the “determi-
but they are most pronounced at low N1,60,CS values. For conditions nistic” triggering curve developed. Accordingly, this relationship
corresponding to σ'v = 1 atm, the differences between the estimated cannot be used directly in performing probabilistic assessments of
CRR values reach as high as 50–80% in the critical region of liquefaction triggering hazard, nor in associated/resulting risk eva-
N1,60,cs < 20 blows/ft within which the consequences of triggering of luations.
liquefaction can be especially significant due to low post-liquefaction 6. This early triggering relationship was presented and documented in
strengths and high cyclic shear strain potential. a manner that was open and transparent for its time, but the level of
This paper has examined the principal sources and causes of dif- transparency was incomplete and so it was difficult to fully back-
ferences between the three triggering relationships, addressing each check some elements of the work.
source of differences in turn. It is now useful to summarize by addres-
sing each of the three overall triggering relationships in turn. 3.2. BI2012

3.1. SEA1985 This relationship began as the Idriss and Boulanger [18,22] trig-
gering relationship, and it has subsequently been modified to produce
SEA1985 was the last of the “first generation” of empirical trig- the current relationship of BI2012. Changes have been incremental, and
gering relationships based on field case histories, and it was the first the character of the overall relationship remains largely intact. Key
essentially “complete” relationship of this type as it was the first com- attributes and issues associated with this relationship include the fol-
prehensive effort to address both (1) fines corrections, and (2) adjust- lowing:
ments of measured SPT penetration resistances to account for variations
in SPT equipment and procedures. Key attributes and issues associated 1. Boulanger and Idriss largely accepted and adopted the significantly
with this early relationship include the following: enlarged case history database of Cetin et al. [1], which involves
197 field performance cases, and then added 33 additional cases to
1. The field case history data set was sparse. Moreover, Seed et al. [2] develop a database of 230 cases. The result was a significantly larger
had to accept and use some cases of lower quality, reliability or level database of generally higher overall quality compared to the one
of documentation (compared to the databases used in development employed by SEA1985. On the other hand, some of the 33 cases
of other two triggering relationships) due to the limited number of added by Boulanger and Idriss (2010) subsequently failed to meet
available cases. At the end, a total of 125 liquefaction field case the screening criteria employed by Cetin et al. [6] and were not used
histories were utilized in developing this triggering relationship. in that study. However, overall the databases of Idriss and Boulanger
2. The early “rd” curves of Seed and Idriss [10] were used for back- [7] and Cetin et al. [6] were largely similar, and quality was gen-
analyses of the field case histories. These rd curves had been de- erally good in both.
veloped based on one-dimensional site response analyses of sim- 2. Similar to SEA1985, BI2012 also employed rd curves that were de-
plified, monolithic, sand-only site conditions that were not re- veloped based on analyses of overly “stiff” site conditions. These
presentative of the more layered and complex site conditions were newly developed rd curves, but they had strikingly similar
present at many of the field case history sites. The acceleration le- attributes to those previously employed by SEA1985, and the use of
vels applied to these sites were not sufficiently high as to encompass these non-representative rd curves again produced estimates of
the levels of shaking of some of the case histories. Use of these non- back-calculated CSR that were systematically biased to the high
representative rd curves produced estimates of back-calculated CSR side, tending to push the resulting CSR values vertically upwards on
that were systematically biased to the high side, tending to push the the triggering plots. This, again, introduced a source of un-
resulting CSR values vertically upwards on the triggering plots, and conservative bias in the resulting triggering relationship developed.
producing a source of unconservative bias in the resulting triggering This issue was also recognized by Boulanger and Idriss [19] and
relationship. their liquefaction triggering correlations were recommended to be
3. No normalization for effective overburden stress was performed (no used only with the same relationships that were used in the devel-
Kσ corrections were applied) because this triggering relationship opment of their correlations (e.g.: only with their rd relationship; but
was recognized to be generally applicable to “shallow” site condi- not with a site-response-estimated rd or CSR).
tions corresponding to most of the liquefaction triggering field case 3. Boulanger and Idriss were aware of the need for normalization for
histories. Because the most “representative” vertical overburden effective overburden stress effects, and so Kσ correction relation-
stress of the overall case history database, and thus the triggering ships were developed and applied. It was decided to truncate Kσ to a
relationship as well, was more nearly σ'v = 0.67 atm than σ'v value of Kσ ≤ 1.1 in processing their case history data. This trun-
= 1 atm, not correcting for Kσ effects was largely equivalent to cation affected 47 of the 230 case histories, and in a similar manner
“truncation” of Kσ to Kσ ≤ 1.0. Due to this “truncation”, CSR's of as the inadvertent truncation of Kσ implicit in the use of the trig-
cases with σ'v < 1 atm (a majority of the cases) move vertically gering relationship of SEA1985. As a result, this truncation of Kσ
upwards on the triggering plots and it introduced a significant again emerges a source of unconservative overall bias in this trig-
source of unconservative bias developed relationship. gering relationship. Boulanger and Idriss developed their own Kσ

84
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

relationship, and this relationship was also less conservative than all case histories were therefore correctly normalized for effective
previous relationships when applied to back-analyses of field case overburden stress effects, and so were the overall database and the
histories. resulting triggering curves developed.
4. The initial fines corrections of Idriss and Boulanger [18] were again 4. Fines corrections were developed based on formal probabilistic re-
based on plotting binned sub-sets of the field case history data and gressions of the large liquefaction field performance case history
engineering judgment, as SEA1985 had done. The resulting fines database.
adjustments are significantly smaller (more conservative) than those 5. Formal probabilistic regressions of the field case history database
of SEA1985, and they are on average somewhat larger than those of were performed to develop the resulting probabilistic liquefaction
Cetin et al. [5]. These fines adjustments are not a function of N1,60, triggering relationship. Suitable degrees of freedom were available
as is the case for the fines adjustments of both SEA1985 and Cetin in the regression so that the triggering curves could conform
et al. [5], and this appears to result in locally unconservative (over- themselves to the dictates of the large field database, and the very
sized) fines adjustments at very low N1,60 values. Yet, this localized difficult and time-consuming task of evaluating and treating both
issue does not appear to significantly affect the overall triggering individual parameter uncertainties as well as overall model un-
relationship at higher N1,60 values. certainty were suitably performed. As a result, this triggering re-
5. Formal probabilistic regressions were performed to develop this lationship provides an unbiased framework for application to
overall triggering relationship, but an unconservative “fit” to the probabilistic liquefaction triggering and overall risk evaluations.
field data was observed due to “stiffly” regressed model and model 6. Because the initial work of Cetin et al. [1] was transparently well-
uncertainty was underestimated at all locations. Both of these issues documented, the back-analyses, assumptions, etc. involved in those
were particularly pronounced at higher values of N1,60,CS ≥ 20. studies were well examined by other researchers. Issues, questions,
6. The Kσ relationship proposed by BI2012 was based on laboratory and challenges resulting from the examinations and reviews were
undrained cyclic test data, rather than regressions of the field case thus able to be implemented to develop (1) a resulting database that
history database, and it differs from the Kσ relationship developed is more closely reviewed, and (2) regressed updated triggering re-
by Cetin et al. [5]. Compared to the other Kσ relationships, it is less lationships that benefitted from both challenges and discussions of
conservative for (i) use in back-analyses of field case histories and details of the previous work. A similar level of transparency and
development of triggering relationships, and (ii) forward en- documentation is aimed for the updated work of CEA2018.
gineering analyses for cases with very high effective overburden 7. The Kσ relationship employed in development of the overall trig-
stresses. gering curves was based on regression of the large field case history
7. Documentation and transparency was lacking when the initial re- database, and so was specifically appropriate over the range of
lationship of Idriss and Boulanger [18,19,22] was first published. As vertical effective stresses well represented in this database
a result, their work could not be properly evaluated and fully (0.25 atm ≤ σ'v ≤ 1.8 atm). This Kσ relationship was used to
checked. Improved documentation was eventually provided by “center” or normalize the overall triggering curves to a condition
Idriss and Boulanger [7], but independent checking of the proces- representative of σ'v = 1 atm. Having accomplished that in an un-
sing and back-analyses of the remaining 129 field case histories biased manner, the resulting triggering curves can then be extra-
continues to be difficult to impossible. polated to much higher effective vertical stresses employing suitable
Kσ relationships of the engineer's choice. A discussion of the merits
3.3. CEA2018 of various Kσ relationships for extrapolation to higher effective
vertical stresses is presented in this paper, in conjunction with
1. Cetin et al. [1] developed systematic screening criteria to evaluate Fig. 4.
the suitability and reliability of potential candidate case histories. 8. Documentation of the data and analyses involved in the develop-
They applied those criteria to the case history database of SEA1985, ment of the new triggering relationship of Cetin et al. [5] is again
and consequently eliminated 35 of the 125 cases. They next ex- presented in a complete and transparent manner, so that other en-
amined more than 200 potential new candidate field case histories, gineers and researchers can thoroughly examine and check all de-
and based on the same screening criteria they adopted only 110 of tails.
those. Cetin et al. [6] deleted 3 cases, and then screened 33 addi-
tional cases added by BI2012 and found 13 of them to meet the Re-examining the trigging relationship plots of Fig. 1(a), (b) and (c)
screening criteria. These cases were added and producing a final more closely, with the benefit of the discussions above, it can now be
database with 210 field performance case histories, all of them clearly seen that the case history data tend to plot higher on the plots in
systematically screened for quality and reliability. Fig. 1(a) and (b), than in Fig. 1(c). It can also be noticed that the
2. CEA2018 employed rd curves of Cetin and Seed [13] which had been triggering curves plot higher (especially for N1,60,CS < 20). The main
developed based on 2153 site response analyses of 50 actual sites reasons for this are now hopefully well understood.
from the case history database. These probabilistically-based re- The triggering curves of both SEA1985 and BI2012 can be demon-
lationships were defined as a function of site conditions (layering strated to produce higher cyclic resistance ratio values (i.e.: un-
and stiffness) as well as intensity and duration of shaking. They were conservatively biased), especially at N1,60,CS < 20. Despite a number of
used in the assessment of CSR in 162 of the field case histories. The other relatively more minor issues, the governing factors leading to this
remaining 48 case histories were back-analyzed by means of site- unconservatism are the use of unrepresentatively i) higher rd and ii)
and event-specific one – dimensional site response analyses using lower Kσ values in the processing of case history data.
(1) available nearby ground motion records from the actual earth- This unconservatism can be expected to be most significant for en-
quakes (scaled to transpose them to the local site), and (2) actual gineering projects where, (1) critical strata have representative values
site stratigraphy and soil properties. The resulting back-calculated of N1,60,CS < 20, (2) site-specific seismic site response (or site response
values of CSR for all 210 case histories were thus specifically un- and soil-structure interaction) analyses are performed to directly cal-
biased best-estimates with case-specific uncertainties also evaluated. culate CSR values, rather than using the “simplified” (rd-based) ap-
Moreover, they are compatible for use in forward engineering ana- proach, and/or (3) liquefaction of soils with σ'v significantly greater
lyses employing either (1) “simplified” (rd-based) evaluations of than 1 atmosphere is of potential concern.
CSR's, or (2) direct determination of CSR's by means of event-spe- As a concluding remark, one of the most significant underlying
cific seismic response analyses. messages of this manuscript is likely the importance of fully transparent
3. No truncation of Kσ was employed. Back-calculated CSR values for documentation of the background details and data involved in the types

85
K.O. Cetin et al. Soil Dynamics and Earthquake Engineering 113 (2018) 75–86

of back-analyses and regressions, etc. employed in the development of soil liquefaction resistance evaluations, (Earthquake Engineering Research Center
these complex types of engineering analysis tools. This enables the Report No. UCB/EERC-84/15). Berkeley, CA: Dept. of Civil and Environmental
Engineering, University of California at Berkeley; 1984.
other engineers and researchers (and oversight agencies) to fully review [10] Seed HB, Idriss IM. Simplified procedure for evaluating soil liquefaction potential. J
and understand the work, which is important for engineering analysis Soil Mech Found Div, ASCE 1971;97:1249–73. [SM9, Proc. Paper 8371].
tools with significant ramifications with regard to public safety. That [11] Golesorkhi R. Factors influencing the computational determination of earthquake-
induced shear stresses in sandy soils [Dissertation submitted in partial fulfillment of
lesson is already being implemented. A multi-year Next Generation the requirement for the degree of doctor of philosophy]. University of California at
Liquefaction (NGL) program coordinated through the Pacific Berkeley; 1989.
Earthquake Engineering Research Center (PEER) is now underway and [12] Imai T, Tonouchi K, Kanemori T. The simple evaluation method of shear stress
generated by earthquakes in soil ground, report no 3. Bur Pract Geol Investig
involving an unusually large number of researchers in an effort de- 1981:39–58.
signed to develop improved liquefaction triggering relationships. The [13] Cetin KO, Seed RB. Nonlinear shear mass participation factor (rd) for cyclic shear
NGL effort is currently targeted at producing new triggering relation- stress ratio evaluation. Soil Dyn Earthq Eng J 2004;24(2):103–13.
[14] Idriss IM. An update to the Seed-Idriss simplified procedure for evaluating lique-
ships over the next five years, or so, and experience from the NGA
faction potential. In: Proceedings of TRB workshop on new approaches to lique-
program suggests that (1) there may be some delays, and (2) the faction, Publication No. FHWA-RD-99-165, Federal Highway Administration; 1999.
eventual results are likely to be well-reviewed and well-checked new [15] Seed B. Evaluation of soil liquefaction effects on level ground during earthquakes,
engineering analysis tools, with good communal support within both Liquefaction Problems in Geotechnical Engineers, Preprint 2725, presented at the
ASCE National Convention, Philadelphia; 1976. p. 1–104.
the research and practice communities. [16] Boulanger RW. High overburden stress effects in liquefaction analyses. J Geotech
In the meantime, engineers will continue to have to sort through the Geoenviron Eng ASCE 2003;129(12):1071–82.
thicket of confusion surrounding the existing liquefaction triggering [17] Boulanger RW, Idriss IM. State normalization of penetration resistance and the
effect of overburden stress on liquefaction resistance. In: Proceddings of 11th in-
relationships. It is the hope of the authors that the materials presented ternational conference on soil dynamics and earthquake engineering and 3rd in-
in this paper, and in the companion paper by Cetin et al. [5], will be ternational conference on earthquake geotechnical engineering, Univ of California,
helpful in that regard. Berkeley, CA; 2004.
[18] Idriss IM, Boulanger RW. Semi-empirical procedures for evaluating liquefaction
potential during earthquakes. In: Doolin D, editor. Proceedings of 11th interna-
Acknowledgements tional conference on soil dynamics and earthquake engineering and 3rd
International conference on earthquake geotechnical engineering, 1. Stallion Press;
2004. p. 32–56.
The authors are deeply grateful to the many engineers and re- [19] Idriss IM, Boulanger RW. Soil liquefaction during earthquakes. Monograph MNO-
searchers who developed the invaluable liquefaction triggering field 12. Oakland, CA: Earthquake Engineering Research Institute; 2008. [261 pp].
performance case history data upon which these types of correlations [20] Cetin KO, Bilge HT. Stress scaling factors for seismic soil liquefaction engineering
problems: a performance-based approach. In: Proccedings of international con-
are based. We are also grateful to the many engineers and colleagues
ference on earthquake geotechnical engineering from case history to practice in
who encouraged this current work, and whose discussions and com- honor of Prof. Kenji Ishihara, Istanbul, Turkey; 2013.
ments were of great value. [21] Cetin KO, Bilge HT. Performance-based assessment of magnitude (duration) scaling
factors. J Geotec Geoenviron Eng ASCE 2012;138(3):324–34.
[22] Montgomery J, Boulanger RW, Harder Jr. LF. Examination of the Kσ overburden
Appendix A. Supplementary material correction factor on liquefaction resistance. J Geotech Geoenviron Eng 2014.
http://dx.doi.org/10.1061/(ASCE)GT.
Supplementary data associated with this article can be found in the [23] Idriss IM, Boulanger RW. Semi-empirical procedures for evaluating liquefaction
potential during earthquakes. J Soil Dyn Earthq Eng 2006;26:115–30.
online version at http://dx.doi.org/10.1016/j.soildyn.2018.03.013. [24] Cetin KO. Reliability-based assessment of seismic soil liquefaction initiation hazard
[Dissertation submitted in partial fulfillment of the requirement for the degree of
References doctor of philosophy]. University of California at Berkeley; 2000.
[25] Liao SSC, Veneziano D, Whitman RV. Regression models for evaluating liquefaction
probability. J Geotech Eng ASCE 1988;114(4):389–409.
[1] Cetin KO, Seed RB, Der Kiureghian A, Tokimatsu K, Harder Jr LF, Kayen RE, et al. [26] Youd TL, Noble SK. Magnitude scaling factors. In: Proceedings of NCEER workshop
SPT-Based probabilistic and deterministic assessment of seismic soil liquefaction on evaluation of liquefaction resistance of soils, Nat Ctr For Earthquake Engrg Res,
potential. ASCE J Geotech Geoenviron Eng 2004;130(12):1314–40. State Univ of New York at Buffalo; 1997. p. 149–65.
[2] Seed HB, Tokimatsu K, Harder LF, Chung RM. The influence of SPT procedures in [27] Toprak S, Holzer TL, Bennett MJ, Tinsley JC. CPT- and SPT-based probabilistic
soil liquefaction resistance evaluations. J Geotech Eng ASCE assessment of liquefaction potential. In: Proceedings of seventh U.S.-Japan work-
1985;111(12):1425–45. shop on earthquake resistant design of lifeline facilities and countermeasures
[3] Youd TL, Idriss IM, Andrus RD, Arango I, Castro G, Christian JT, et al. Liquefaction against liquefaction; 1999.
resistance of soils. Summary report from the 1996 NCEER and 1998 NCEER/NSF [28] Juang CH, Jiang T, Andrus RD. Assessing probability-based methods for liquefac-
workshops on evaluation of liquefaction resistance of soils. J Geotech Geoenviron tion potential evaluation. J Geotech Geoenviron Eng ASCE 2002;128(7):580–9.
Eng 2001;127(10):817–33. [29] Seed HB. Design problems in soil liquefaction. J Geotech Eng ASCE
[4] Boulanger RW, Idriss IM. Probabilistic standard penetration test–based lique- 1987;113(8):827–45.
faction–triggering procedure. J Geotech Geoenviron ASCE 2012;138(10):1185–95. [30] Idriss IM. Seed memorial lecture. University of California at Berkeley; 1995.
[5] Cetin KO, Seed RB, Kayen RE, Moss RES, Bilge HT, Ilgac M, et al. SPT-based [31] Liu AH, Stewart JP, Abrahamson NA, Moriwaki Y. Equivalent number of uniform
probabilistic and deterministic assessment of seismic soil liquefaction triggering stress cycles for soil liquefaction analysis. J Geotech Geoenviron Eng
hazard. Soil Dyn Earthq Eng 2018. 2001;127(12):1017–26.
[6] Cetin KO, Seed RB, Kayen RE, Moss RES, Bilge HT, Ilgac M. et al. Summary of SPT-
Based field case history data of the updated 2016 database. METU Soil Mechanics
and Found Eng Research Center. Report no: METU/GTENG 08/16-01; 2016. Further reading
[7] Idriss IM, Boulanger RW. SPT-based liquefaction triggering procedures. (Report
UCD/CGM-10/02). Davis, CA: Center for Geotechnical Modeling, Department of [1] Seed HB, Idriss IM. Ground motion and soil liquefaction during earthquakes.
Civil and Environmental Engineering, University of California; 2010. p. 136. Oakland, CA: Earthquake Engineering Res. Insitute Monograph; 1982.
[8] Boulanger RW, Idriss IM. CPT and SPT based liquefaction triggering procedures. [2] Kayen R, Moss RES, Thompson E, Seed RB, Cetin KO, Kiureghian A, et al. Shear-wave
(Report No. UCD/CGM-14/01). Davis, CA: Center for Geotechnical Modeling, velocity-based probabilistic and deterministic assessment of seismic soil liquefaction
Department of Civil and Environmental Engineering, University of California; 2014. potential. J Geotech Geoenviron Eng 2013;139(3):407–13.
p. 134.
[9] Seed HB, Tokimatsu K, Harder LF, Chung RM. The influence of SPT procedures in

86