Georisk: Assessment and Management of Risk for
Engineered Systems and Geohazards
ISSN: 1749-9518 (Print) 1749-9526 (Online) Journal homepage: https://www.tandfonline.com/loi/ngrk20
The story of statistics in geotechnical engineering
Kok-Kwang Phoon
To cite this article: Kok-Kwang Phoon (2020) The story of statistics in geotechnical engineering,
Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 14:1,
3-25, DOI: 10.1080/17499518.2019.1700423
To link to this article: https://doi.org/10.1080/17499518.2019.1700423
Published online: 09 Dec 2019.
Submit your article to this journal
Article views: 3956
View related articles
View Crossmark data
Citing articles: 29 View citing articles
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=ngrk20
GEORISK
2020, VOL. 14, NO. 1, 3–25
https://doi.org/10.1080/17499518.2019.1700423
SPOTLIGHT ARTICLE
The story of statistics in geotechnical engineering
Kok-Kwang Phoon
Department of Civil and Environmental Engineering, National University of Singapore, Singapore, Singapore
ABSTRACT
ARTICLE HISTORY
The story of statistics in geotechnical engineering can be traced to Lumb’s classical Canadian
Geotechnical Journal paper on “The Variability of Natural Soils” published in 1966. In parallel,
the story of risk management in geotechnical engineering has progressed from design by
prescriptive measures that do not require site-specific data, to more refined estimation of
site-specific response using limited data from site investigation as inputs to physical models,
to quantitative risk assessment (QRA) requiring considerable data at regional/national scales.
In an era where data is recognised as the “new oil”, it makes sense for us to lean towards
decision making strategies that are more responsive to data, particularly if we have
zettabytes coming our way. In fact, we already have a lot of data, but the vast majority is
shelved after a project is completed (“dark data”). It does not make sense to reduce one
zettabyte to a few bytes describing a single cautious value. It does not make sense to
expect big data to be precise and to fit a particular favourite physical model as demanded
by the classical deterministic world view. This paper advocates the position that there is
value in data of any kind (good or not so good quality, or right or wrong fit to a physical
model) and the challenge is for the new generation of researchers to uncover this value by
hearing what data have to say for themselves, be it using probabilistic, machine learning, or
other data-driven methods including those informed by physics and human experience, and
to re-imagine the role of the geotechnical engineer in an immersive environment likely to be
imbued by machine intelligence.
Received 8 October 2019
Accepted 30 November 2019
1. Introduction
Errors using inadequate data are much less than those
using no data at all.
Charles Babbage
Every story has a beginning. The idea that statistics can be
used to quantify uncertainties in the properties of natural
soils (an intrinsic characteristic of site data) and this statistical approach can provide a rational basis for the selection of a suitably cautious design value may arguably be
traced to Lumb’s classical Canadian Geotechnical Journal
paper on “The Variability of Natural Soils” published in
1966. One may view this paper as being ahead of its
time. The First International Conference on Applications
of Statistics and Probability to Soil and Structural Engineering was organised in Hong Kong in 1971. It was not
surprising that Professor Peter Lumb played a key role
in launching this important conference series. The first
ICASP in Hong Kong was followed by Aachen (1975),
Sydney (1979), Florence (1983), Vancouver (1987),
Mexico City (1991), Paris (1995), Sydney (1999),
San Francisco (2003), Tokyo (2007), Zurich (2011), Vancouver (2015) and Seoul (2019).
KEYWORDS
Big Indirect Data (BID);
generic databases; MUSIC-X;
transformation models; site
challenge; similarity index
In response to a question on the “importance of statistics as a tool in engineering applications” posed during
an interview before his retirement (Lam and Li 1986),
Professor Lumb opined that
Traditionally engineering and civil engineering are very
deterministic in their teaching and in the attitude of
their practitioners. When something goes wrong, it
takes them by surprise. And yet all the things they are
handling, the raw materials, the input and output, are
random processes. If that can be taken seriously the
method of design can be improved considerably. Instead
of the old fashioned safety factor, the probability of failure type of approach is more satisfactory and practically
far more useful.
He further added that
once you think of all these things as being random processes, it does clear up the engineer’s mind as well as
improving his design. It makes him realize that he cannot predict what is going to happen precisely. This is
what most engineers try to do. That is what they taught
in schools and in universities in general: Engineering is
precise, it is a science. Yet, in reality, it is vague. It is not
a science. It is more an art.
CONTACT Kok-Kwang Phoon
kkphoon@nus.edu.sg
Department of Civil and Environmental Engineering, National University of Singapore, Block E1A,
#07-03, 1 Engineering Drive 2, Singapore 117576, Singapore
© 2019 Informa UK Limited, trading as Taylor & Francis Group
4
K.-K. PHOON
It was only in 1995 that a National Research Council
report “Probabilistic Methods in Geotechnical Engineering” recommended that
probabilistic methods, while not a substitute for traditional deterministic design methods, do offer a systematic and quantitative way of accounting for
uncertainties encountered by geotechnical engineers,
and they are most effective when used to organize and
quantify these uncertainties for engineering designs
and decisions.
A Recommended Practice DNV-RP-C207 (DNV 2012)
provides principles, guidance and recommendations for
use of statistical methods for analysis and representation
of soil data. The latest 4th edition of the international
standard “General Principles on Reliability for Structures” (ISO2394:2015) includes a new Annex D dedicated to the reliability of geotechnical structures.
Annex D recognises that geotechnical reliability-based
design should place site investigation and the interpretation of site conditions/profile/data as the cornerstone
of the methodology (Phoon et al. 2016). Despite these
notable advances, it is accurate to say that data plays a
supporting rather than a leading role in decision making
in practice. After all, data have not spoken for themselves. Phoon (2017) pointed out that data scarcity
(“curse of small sample size”) is more conspicuous in
geotechnical engineering than structural engineering.
Decision making strategies have evolved to be effective
in such a data poor environment, thus making it even
harder to monetise data because these predominant
strategies do not need much data – some such as design
by prescriptive measures requires almost none. One
could say there is selection pressure against data in evolutionary parlance.
This paper covers the estimation of useful statistics
from the original classical univariate setting to a more
realistic incomplete multivariate setting encountered in
a typical site investigation programme. It traces the evolution of uncertainties as an inconvenient feature out of
step with a deterministic world view that could be mitigated by adopting a judicious precedent-based cautious
stance, to something entirely undesirable that should
be minimised, to a fact of reality that should be coped
with explicitly using statistics, and to an asset that can
be exploited using Bayesian machine learning. The
focus is not on mathematical gymnastics, but to demonstrate that data can produce valuable insights to support
decision making in its own right, over and above its current value in physical modelling, ultimate/proof load
testing, and monitoring. The common lament is that
we do not have enough data to do this. This is not
true. We have a lot of data, but they are shelved in design
and regulatory offices after completion of a project. They
are stored primarily for compliance purposes, rather
than shared and further analysed to provide more
insights and to support future decision making. In
short, our data is mainly “dark”, a term defined in Gartner’s IT glossary (Gartner 2019). This is not true even for
a smaller set of data published in the literature that has
been progressively compiled into generic soil/rock databases in recent years. However, this is frequently true for
one specific site. In fact, site-specific data are more challenging to deal with than simply “not enough” and
“uncertain”. We now understand that there are at least
seven rather than two attributes that define our data.
Phoon, Ching, and Wang (2019) refer to these attributes
as “MUSIC-X”: Multivariate, Uncertain and Unique,
Sparse, Incomplete, and potentially Corrupted with “X”
denoting the spatial/temporal dimension. The “unique”
and “potentially corrupted” (in the sense of data containing outliers) attributes are very hard problems. It is
important to emphasise here that the term “uncertainty”
adopted in this paper refers to both imprecision in the
knowledge of a particular physical parameter, say
undrained shear strength, and the deeper imprecision in
modeling this imprecise knowledge, say the mean and
standard deviation of the random variable model or
the scale of fluctuation of the random field model. This
uncertainty of the uncertainty model, commonly
referred to as statistical uncertainty, is notoriously
difficult to address for sparse data. This paper shows
that reasonable solutions are available even for our
exceedingly modest site-specific data. There are also
non-probabilistic solutions (Beer et al. 2013). This is outside the scope of this paper.
All sensible engineers know that generic correlation
models must be used with caution. It is better to adopt
quasi-local correlation models supported by site-specific
data and data from “similar” sites possessing comparable
geology. At present, the engineer relies almost entirely on
his/her experience working on other sites to construct
such quasi-local models. One research challenge is how
to do this algorithmically. This is called the “site challenge” (Phoon 2018). This will potentially extend the
search for “similar” sites to anywhere anywhen, beyond
regional/municipal databases that the engineer is familiar with over the duration of his practice and beyond
the deep isolation that human experience encases each
and every one of us in because it cannot be shared in
full or with ease. Recent research shows that this site
challenge is tractable even under full MUSIC constraints.
The existence of a quasi-local correlation model that
maintains an optimal balance between a generic database
(that may not be directly and completely applicable
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
although extremely data rich) and a site database (that is
fully applicable but extremely data poor) (cf. “Goldilocks
dilemma” in Phoon 2020) remains an open research
question at this point.
2. Role of data in design
Simple calculations based on a range of variables are better than elaborate ones based on limited input. Professor
Ralph Peck’s Legacy Website (Geoengineer 2019)
All engineers make decisions in the face of uncertainty.
After all, why would we need a factor of safety if we
have omniscient access to perfect knowledge and information? Einstein and Baecher (1982) put this across
succinctly:
In thinking about sources of uncertainty in engineering
geology, one is left with the fact that uncertainty is inevitable. One attempts to reduce it as much as possible, but
it must ultimately be faced. It is a well-recognised part of
life for the engineer. The question is not whether to deal
with uncertainty but how?
Uncertainty, in its broadest sense, can range from known
unknowns where some knowledge and/or data exist for
characterisation to unknown unknowns where alternate
strategies such as robust or resilient design could be
more applicable. Phoon (2017) opined that in between
“white swans” (known unknowns) and “black swans”
(unknown unknowns), there will be “grey swans” covering events that are foreseeable even in the absence of data
or unforeseeable events that do not result in disproportionate consequences. In all likelihood, the factor of
safety is intended to cover only white and possibly
some grey swans.
Casagrande (1965)’s classic paper and Terzaghi Lecture on “calculated risk” remains relevant to our practice.
One should not confuse managing risk in the broad sense
articulated by Casagrande with quantitative risk
Table 1. Three-tier classification scheme of soil property
variability for reliability calibration (Source: Table 9.7, Phoon
and Kulhawy 2008).
Geotechnical parameter
Property variability
COV (%)
Undrained shear strength
Lowa
Mediumb
Highc
Lowa
Mediumb
Highc
Lowa
Mediumb
Highc
10–30
30–50
50–70
5–10
10–15
15–20
30–50
50–70
70–90
Effective stress friction angle
Horizontal stress coefficient
Typical of good quality direct lab or field measurements.
Typical of indirect correlations with good field data, except for the standard
penetration test (SPT).
c
Typical of indirect correlations with SPT field data and with strictly empirical
correlations.
a
b
5
assessment (QRA). One can manage risks cleverly with
experience alone. For example, Terzaghi and Peck
(1967) cited Kidder-Parker Architects’ and Builders’
Handbook (1931) in their Table 54.1 “Soil pressures
allowed by various building codes”. An engineer could
select an allowable bearing pressure based on the “character of foundation bed” and the name of a city. Table 1
of BS8004 (1986), “Presumed allowable bearing values
under static loading” provides a range of bearing values
for each type of rock and soil. Qualitative information
such as “strong limestones and strong sandstones” and
“schists and slates” for rocks and “firm clays” and “soft
clays and silts” for cohesive soils is sufficient. Section
2.5 of Eurocode 7 describes design by prescriptive
measures (EN 1997−1:2004). These presumed/prescribed values are conservative, but this is a sensible
approach to risk management in the absence of sitespecific data and calculation models.
In our current practice, building regulations typically
mandate minimum ground investigation at a specific site,
for example, the number of boreholes should be the greater
of (i) one borehole per 300 m2 or (ii) one borehole at every
interval between 10 and 30 m, but no less than 3 boreholes
in a project site. It is possible to “predict” a reasonable sitespecific response (say bearing pressure) based on such limited information through the mediation of a physical
model and a healthy dose of engineering judgment. We
need some site-specific data to use this approach. We
also need the “right” kind of data, which is chained to the
input side of the model. What is “right” for one model
may not be “right” for another model. Lambe (1973)’s Rankine Lecture explains the need to calibrate both data and
model together. In addition, a factor of safety and experience are still needed as noted by Burland (1987) in his
Nash Lecture. One may conclude that engineers are clever
in adopting risk management strategies that work with the
imperfect knowledge and the limited information they
have at hand. There is no doubt that we have been successful. Failures are rare.
So, what has changed? Some claimed that our internet
traffic has exceeded a zettabyte (1021 bytes or roughly the
number of sand grains on all the beaches on the planet)
throughput per year as of 2016. We may not have a zettabyte currently, but we have a lot of data. They are just
not directly useful such as not specific to the site of
interest or need to be transformed to fit the input
side of a physical model. Needless to say, they will
also be imperfect in the sense of being uncertain,
incomplete, and possibly even corrupted to some
extent. Phoon, Ching, and Wang (2019) coined the
term Big Indirect Data (BID) to refer to any data that
are potentially useful but not directly applicable to
the decision at hand. All engineering decisions are
6
K.-K. PHOON
ultimately black and white, be it choosing the dimensions of a structure, time interval between maintenance,
or issuance of an evacuation notice, notwithstanding
the imperfect nature of our data, methods, and understanding of reality. The adjective “useful” is used in the
context of supporting such real world decisions. The
generic soil/rock and load test databases presented in
the next section will be one type of BID. Monitoring
data is another BID.
It is timely to ask ourselves how existing strategies
that are tailored to work effectively in a data poor
environment can monetise BID. First, there is no mechanism to update presumptive bearing pressures or factors of safety using data. There is a formal mechanism
to do this for resistance factors that are calibrated from
statistics (e.g. Phoon, Kulhawy, and Grigoriu 2003;
Paikowsky et al. 2004; Fenton et al. 2016; Tang and
Phoon 2018). There is no space to engage in a full discussion on reliability-based design, simplified or otherwise,
but it is reasonable to lean towards mechanisms that are
responsive to data and self-improve with data, particularly if we have zettabytes coming our way. Second, a
physical model erects a significant computational barrier
between input and all other data. For example, monitoring data constitutes the basis for our observational
approach. However, it is assuredly “non-input” data. System identification techniques are needed to back-calculate the equivalent input data before updated
predictions are possible. For a large 3D finite element
model, forward calculations (inputs to outputs) can
take days using current computers. Backward calculations (outputs to inputs) will take longer. It is difficult
to leverage on our powerful physical models to glean
deeper insights from monitoring data, particularly for
risk management of big projects in real time.
System identification in its most general form can link
inputs and outputs without the mediation of a physical
model (black box approach as opposed to the physicsbased white box approach). The most common example
of system identification in geotechnical engineering is an
artificial neural network (Shahin, Jaksa, and Maier 2001;
Jaksa, Maier, and Shahin 2008). Interestingly, machinelearning methods are also data-, rather than physics-driven, in part because they accommodate all data, whether
they are good or not so good quality, or right or wrong fit
to a physical model. In fact, some machine-learning
methods have been successful even when the data used
have been judged “useless” by a human expert. Such a
physics-free and judgment-neutral approach can be
exceedingly powerful as demonstrated by Google’s
AlphaGo project. It is interesting that our practice is
almost entirely dominated by a white box approach,
although it is known in many fields that this approach
has limits in dealing with complex real world processes.
One should be mindful that real data emerge from such
complex processes, not from idealised physical models.
Another limitation is that a physical model cannot
“learn” on its own to be better as data and problem scenarios evolve in real time.
Actually, in the author’s opinion, our practice is more
accurately described as pure white box, because it does
not admit uncertainty explicitly. Although somewhat
exaggerated, one could argue that we are philosophically
aligned to Laplace’s Demon who famously said (Laplace,
not the hypothetical demon) in his “A Philosophical
Essay on Probabilities”:
We ought then to regard the present state of the universe
as the effect of its anterior state and as the cause of the
one which is to follow. Given for one instant an intelligence which could comprehend all the forces by which
nature is animated and the respective situation of the
beings who compose it – an intelligence sufficiently
vast to submit these data to analysis – it would embrace
in the same formula the movements of the greatest
bodies of the universe and those of the lightest atom;
for it, nothing would be uncertain and the future, as
the past, would be present to its eyes.
We do go to some lengths to collect good quality data
directly relevant to our physical models and if this is not
possible, we apply empirical rules to be conservative. Do
we collect only high quality data that are limited in quantity or only lower quality data in larger quantity? Do we
combine them? When shouldn’t they be combined? The
jury is still out on these important questions, but there is
no hope to make progress without an explicit uncertainty
model. Scott A. Barnhill left the following message on
Professor Ralph Peck’s legacy website: “Perhaps engineers trained in geology have an advantage. They are
more likely to accept mother nature as she exists, rather
than as created in the mind of the engineer” (Geoengineer 2019). This practical wisdom needs reinforcing if
we would like to engage emerging digital technologies
with greater haste. The Institution of Civil Engineers
(UK) State of the Nation Report (2017) observed that
“the infrastructure sector has been slow to engage with
the uptake of new digital technologies compared with
other industries”.
Let me call the future of geotechnical engineering
(unimaginatively) as Geo 4.0. If Geo 4.0 needs to operate
in an overwhelmingly data-rich cyber-physical environment, it is reasonable to question if a pure white box
approach will continue to be a winning strategy. A
black box or grey box (physics-informed data-driven)
approach may be more effective. The point is not to be
philosophical, but to be pragmatic. The statistician
George Box once said: “All models are wrong, but
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
some are useful”. This seems like a good adage to follow
when we explore how physics, data, and experience can
be combined in even more clever ways to support
decision making.
Peck (1980) adopted the intriguing question “Where
has all the judgment gone?” as the title for his fifth Laurits Bjerrum memorial lecture. The honest answer nowadays is no one knows. Philosophers are asking the same
question as the power of artificial intelligence expands
beyond what was previously thought to be reachable
only by the human mind. The Leverhulme Centre for
the Future of Intelligence (http://lcfi.ac.uk/) was established for this reason. One can safely say that machinehuman interactions will be transformed in unimaginable
ways and our human minds will be enhanced (to put it
mildly) as part of this transformation. At a more mundane level, it is already possible for Bayesian methods
to learn some limited aspects of expert judgment,
which will vastly expand the sharing of this “digitized
experience” beyond what we can do with conventional
“on the job” training at the individual level (Vick 2002).
The next two sections will briefly discuss: (1) generic
databases (BID) to provide an overview of the attributes
of geotechnical data and (2) preliminary research to
address the “site challenge” under the realistic data constraints of MUSIC as an example of what Bayesian
machine learning could do.
3. Generic databases
Numbers have an important story to tell. They rely on
you to give them a clear and convincing voice.
Stephen Few
Lumb (1966)’s classic paper on “The Variability of Natural Soils” showed that the variations in the properties of
four typical Hong Kong soils (a soft marine clay, an alluvial sandy clay, a residual silty sand, and a residual clayey
silt) about a mean trend can be characterised as random
variables following distributions such as normal, lognormal, and bi-normal distributions. The ensuing body of
work on soil properties was published in diverse venues
such as ICASP, ASCE symposiums (e.g. Characterisation
of soil properties: bridge between theory and practice,
Atlanta, Georgia, 1984; Uncertainty in the geologic
environment: from theory to practice, Madison, Wisconsin, 1996), and reports (e.g. Filippas, Kulhawy, and Grigoriu 1988; Orchant, Kulhawy, and Trautmann 1988;
Spry, Kulhawy, and Grigoriu 1988; Kulhawy, Birgisson,
and Grigoriu 1992; National Research Council 1995;
Vanmarcke and Fenton 2003), before culminating in
an extensive compilation of univariate statistics by
Phoon & Kulhawy (1999a, 1999b) that further generalised the characterisation of natural variations from a
7
random variable model (measurements at different
depths are independent) to a random field model
(measurements at different depths are correlated). The
theoretical application of a random field in geotechnical
engineering was popularised in an earlier paper by Vanmarcke (1977). Other seminal contributions were also
made by pioneers such as Wilson Tang (Tang 1984;
Lacasse, Liu, and Nadim 2017), Tien H Wu (Wu et al.
1989, 1996; Baecher and Christian 2019), Harr (1987),
Lacasse and Nadim (1996), Gregory Baecher and John
Christian (Baecher 1987; Baecher and Christian 2003),
Herbert Einstein (Einstein and Baecher 1983; Einstein
et al. 1996), and many others. The author is unable to
do even partial justice in this cursory overview of more
than five decades of work in trying to coax data that
are 100% accurate (within measurement limits) to say
something useful about the unknown state of the ground
between measured locations. It goes without saying that
these ground truths can only be approximate (no free
lunch!) and an explicit uncertainty model is again
necessary.
Over the years, the terms random field and spatial
variability have become synonymous, although the two
concepts are distinct. The former is a mathematical
model. The latter is a description of a geological reality.
There is no guarantee that a random field model, particularly the common second-order (stationary) version fully
described by an autocorrelation function, is an adequate
representation of this reality for all geologic settings.
The predominance of a second-order field model in the
literature arises in part from the difficulty of characterising
higher-order fields with limited data. Theoretical higherorder fields do exist (Shields and Kim 2017). Phoon,
Ching, and Wang (2019) pointed to several other limitations of the widely used second-order field model.
While observing that this model is a closer match to reality
compared to the independent and identically distributed
(i.i.d.) model, the authors observed that
it cannot be applied in its most general non-stationary
form because we do not have sufficient site investigation
data for statistical characterization. The current practice
is to assume a trend function can be removed from the
data and the residuals are second-order stationary
within a typical site. The reason for this assumption is
that pairs of measurements regardless of where they
are measured can be used to estimate the autocorrelation function. Needless to say, there is no trend, no
stationary residuals, and no autocorrelation function
in reality. These concepts exist purely within the stationary random field model.
The authors further highlighted that
trend removal can be difficult (Ching, Wu, and Phoon
2016; Ching et al. 2017; Ching and Phoon 2017).
8
K.-K. PHOON
Estimation of random field parameters is also computationally challenging (Tian et al. 2016; Wang H. et al.
2018; Xiao et al. 2018). Fine details of the autocorrelation function such as sample path “smoothness” are
important (Ching and Phoon 2019a). Characterization
of site stratigraphy is a major missing feature of past
random field studies until quite recently. (Wang,
Huang, and Cao 2013; Ching et al. 2015; Li et al. 2016;
Qi et al. 2016; Wang X. et al. 2016; Wang H. et al.
2017; Wang X. et al. 2018; Cao et al. 2019; Wang
H. et al. 2019; Wang X. et al. 2019; Wang, Hu, and
Zhao 2019)
The difficulties have nothing to do with theory. They
have everything to do with statistical characterisation
using actual data. Two statistics are needed to describe
a second-order (stationary) random field model, namely
the coefficient of variation (COV) and the scale of fluctuation (SOF). The COV is needed to describe the scatter
about the mean trend in the basic random variable
model, basically a characteristic amplitude of the fluctuations. The SOF is an additional statistics that roughly
describe the distance over which the measurements are
strongly correlated in a random field model (DeGroot
and Baecher 1993; Jaksa 1995; DeGroot 1996). It is a
characteristic wavelength of the fluctuations. If the SOF
is much larger than the mobilised volume of soil, the
basic random variable model can be adopted. It is accurate to say that measurements sampled at a depth interval larger than the SOF can be modelled as independent
random variables. Or to put this in another way, information on spatial variability cannot be captured by a
sparse sampling grid. From this practical perspective,
the random field model merely allows measurements
sampled at any depth interval, including near continuous
cone penetration test soundings, to be modelled with
greater realism.
The most complete compilation of COVs for both
soils and rocks to date is given by Phoon et al. (2016).
An updated compilation for the SOF is currently in progress (Cami, Javankhoshdel, and Phoon 2020). The practical value of characterising natural variations in the
form of a COV is that resistance factors can be calibrated
more realistically based on our knowledge of soil parameters, such as the three-tier classification scheme of
soil property variability shown in Table 1. A similar
scheme appears in the 2014 edition of the Canadian
Highway Bridge Design Code (CAN/CSAS614:2014)
that presents different resistance factors depending on
the “degree of understanding” (low, typical, high) (Fenton et al. 2016) and others (e.g. Paikowsky et al. 2004;
Bathurst, Javankhoshdel, and Allen 2017). For the first
time, we can establish a defensible link between site
investigation efforts and the economy of the design,
which gives us a fair chance of swaying more businessoriented clients to accept site investigation as an investment rather than a cost (Ching, Phoon, and Yu 2014).
The practical value of characterising spatial correlations
in the form of a SOF is that the COV of a spatial average
can be reduced, thus permitting higher resistance factors
to be used for problems governed by spatial averages.
Another practical value is that soil properties at
unsampled locations can be interpolated more accurately
using kriging or general regression (Yuen, Ortiz, and
Huang 2016; Yuen and Ortiz 2016, 2018) when spatial
correlations are available. This brief review is not
intended to be up-to-date or comprehensive on what
we know about spatial variability, its value, and its
impact on design. Research has advanced considerably
beyond Vanmarcke’s classic paper in 1977. The interested reader can refer to the Joint TC205/TC304 Working Group Report (2017) on “Discussion of statistical/
reliability methods for Eurocodes”, which has been
made available at the ISSMGE TC304 website: http://
140.112.12.21/issmge/tc304.htm.
The characterisation of geotechnical data has become
even more realistic with the compilation of multivariate
databases over the past decade. Momentum is gathering
worldwide to screen, organise, and share our valuable
data to hasten the pace of our digital transformation
such as Project 304 dB (TC304 2019). Ching, Li, and
Phoon (2016) provided a useful overview of generic
multivariate databases on soil/rock properties. Table 2
shows an updated summary of these databases, labelled
as (geo-material type)/(number of parameters of interest)/(number of data points). For example, the CLAY/
10/7490 database consists of 7490 records from 251
studies carried out in 30 countries. Each record contains
ten clay parameters measured at roughly the same depth,
although some may be missing. The CLAY/10/7490
database is global in coverage. In contrast, the SHCLAY/11/4051 municipal database covers 50 sites in
Shanghai (Zhang et al. 2019). Another source of information frequently collected comes from pile load tests.
The performance databases for other geotechnical structures (in addition to piles) are available, but less commonly reported in the literature. A comprehensive
survey of these databases was recently carried out by
Phoon and Tang (2019). Table 3 includes further
updates. The following geotechnical structures are covered: (1) shallow and deep foundations, (2) offshore
spudcans, (3) mechanically stabilised earth and soil
nail walls, (4) pipes and anchors (plate, helical, and shoring), (5) slopes and base heave, (6) cantilever walls, and
(7) braced excavations. Details are given elsewhere
(Phoon and Tang 2019). Case studies are even more
informative, but no systematic compilation has been
Table 2. Summary of some soil/rock property databases (updated from Phoon and Ching 2017).
Range of parameters
Database
Reference
Parameters of interest
# Data points
# Sites/studies
Ching and Phoon (2012)
Ching, Phoon, and Yu (2014)
CLAY/7/6310
su from 7 different test procedures
6310
164 studies
CLAY/10/7490
Ching and Phoon (2013,
2015)
Ching and Phoon (2014)
7490
251 studies
CLAY/9/249
D’Ignazio et al. (2019)
LL, PI, LI, s′ v /Pa , St, Bq, s′ p /Pa , su /s′ v , (qt − sv )/s′ v ,
(qt − u2 )/s′ v
s′ v /Pa , σv/Pa, s′ v /Pa , qt/Pa, u2/Pa, u0/Pa, PI, wn, St
FG/KSAT/4/1358
FI-CLAY/7/216a
Feng and Vardanega (2019)
D’Ignazio et al. (2016)
e, LL, wn/LL, −ln(ksat)
′
′
sFV
u , s v , s p , wn, LL, PL, St
1358
216
JS-Clay/5/124b
Liu et al. (2016)
Mr, qc, fs, wn, γd
124
16
RFG/TXCU-278
Beesley and Vardanega
(2019)
Hov et al. (2019)
su /s′ v , γ50 CIU, OCR, γ50 CKU
278
21 studies
499
Sweden
c
SE-CLAY/4/499
DSS
sFV
u , su ,
sv )/s′ v , (qt − u2 )/s′ v , (u2 − u0 )/s′ v , Bq
′
s p , LL
345
535
249
37 sites
40 sites
18 sites
33 studies
24 sites
SH-CLAY/11/
4051
Zhang et al. (2019)
LL, PI, LI, e, K0, s′ v /Pa Su /s′ v(UCST) , St(UCST), Su /s′ v(VST) , St(VST),
ps /s′ v
4051
50 sites
(Shanghai)
SAND/7/2794
ROCK/9/4069
Ching et al. (2017)
Ching et al. (2018)
D50, Cu, Dr, s′ v /Pa , fʹ, qt1, (N1)60
n, γ, RL, Sh, σbt, Is50, Vp, σc, E
2794
4069
176 studies
184 studies
a
1–4
1–6
PI
St
–
Low to very high plasticity
Sensitive to quick clays
Insensitive to quick
clays
1–10 Low to very high plasticity
Insensitive to quick
clays
1–10 Low to very high plasticity
Insensitive to quick
clays
1–10 Low to very high plasticity
Insensitive to quick
clays
–
Low to very high plasticity
–
1–
Low to very high plasticity
Insensitive to quick
7.5
clays
Soft to stiff clayey soils and silty clay soils with high variability of
the strength and stiffness characteristicsMr = 12.54–95.82 MPa,
qc = 0.22–3.93 MPa, fs = 0.03–0.14 MPa, wn (%) = 6.91–78.11,
γd = 10.47–19.92 kN/m3
1–32 Low to medium-high
plasticity
10–20
s′ p = 13–505 kPa; sFC
u = 5–101 kPa;
sDSS
u = 6–53 kPa; LL = 22–145%
Normal consolidated to slightly over-consolidated clay; Very soft
clay (LI = 0.49–2.19) with slight to medium plasticity (PI = 10.4–
26.5) and with medium to high sensitivity (St = 2.7–7.8)
1–15 D50 = 0.1–40 mm, Cu = 1–1000 + Dr = −0.1–117%
γ = 15–35 kN/m3, n = 0.01–55%σc = 0.7–380 MPa, E = 0.03–
120 GPa
F-CLAY renamed as FI-CLAY to follow internet domain for Finland (FI).
J-CLAY renamed as JS-CLAY to follow phonetics abbreviation of Jiangsu (JS).
c
SE-CLAY/4/499 based on S. Larsson (personal communications, 2019).
Notes: LL = liquid limit; PL = plastic limit; PI = plasticity index; LI = liquidity index; wn = natural water content; e = void ratio; ksat=saturated hydraulic conductivity; Mr = resilient modulus; qc = cone tip resistance; fs = sleeve
friction; γd = dry density; D50 = median grain size; Cu = coefficient of uniformity; Dr = relative density; σv = vertical total stress; s′ v = vertical effective stress; s′ p = preconsolidation stress; su = undrained shear strength; sFV
u =
′
′
undrained shear strength from field vane; sre
u = remoulded su; fʹ = effective friction angle; St = sensitivity; OCR = overconsolidation ratio, (qt − sv )/s v = normalised cone tip resistance; (qt − u2 )/s v = effective cone tip
resistance; u0 = hydrostatic pore pressure; (u2 − u0 )/s′ v = normalised excess pore pressure; Bq = pore pressure ratio = (u2-u0)/(qt-σv); Pa = atmospheric pressure = 101.3 kPa; qt1 = (qt/Pa) × CN (CN is the correction factor
for overburden stress); (N1)60 = N60×CN (N60 is the N value corrected for the energy ratio); n = porosity; γ = unit weight; R = Schmidt hammer hardness (RL = L-type Schmidt hammer hardness); Sh = Shore scleroscope hardness;
σbt = Brazilian tensile strength; Is = point load strength index (Is50 = Is for diameter 50 mm); Vp = P-wave velocity; σc = uniaxial compressive strength; E = Young’s modulus; γ50 CIU = shear strain to mobilise 0.5su under isotropically-consolidated undrained conditions; γ50 CKU = shear strain to mobilise 0.5(su – τ0); τ0 = initial shear stress; ps = cone tip resistance from CPT which is unique in China without the measurement of pore pressure; SDSS
u =
undrained shear strength from direct simple shear test.
b
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
CLAY/5/345
CLAY/6/535
′
′
LI, su, sre
u , s p, s v
su /s′ v , OCR, (qt −
OCR
9
10
K.-K. PHOON
Table 3. Summary of performance databases for some geotechnical structures (updated from Table 1; Phoon and Tang 2019).
Geotechnical structures
Shallow foundations
Offshore spudcans
Drilled shafts (vertical load)
Drilled shafts (lateral load)
Augered cast-in-place piles
Driven piles
Helical piles
Driven cast-in-situ piles
Pile foundations
Micropiles
Foundations
Mechanically stabilised earth walls
Soil nail walls
Multi-anchor walls
Slopes
Excavations (base heave)
Pipes
Plate anchors
Plate anchors
Helical anchors
Database/reference
Data source
Test type
Geomaterial
N
UML-GTR ShalFound07 (Paikowsky et al. 2010)
UML-GTR RockFound07 (Paikowsky et al. 2010)
Akbas (2007)
Samtani and Allen (2018)
SpreadFound/1026 (Tang et al. 2019)
Tang and Phoon (2019a)
Ng et al. (2001)
AbdelSalam, Baligh, and El-Naggar (2015)
Asem, Long, and Gardoni (2018)
DSHAFT (Garder et al. 2012)
Motamed, Elfass, and Stanton (2016)
Stark et al. (2017)
TxDOT (Moghaddam et al. 2018)
Tang, Phoon, and Chen (2019)
EPRI (Chen and Kulhawy 1994)
Chen and Lee (2010)
Chen, Lin, and Kulhawy (2011)
Marcos and Chen (2013)
Reddy and Stuedlein (2017)
McVay et al. (2016)
AAU-NGI (Augustesen 2006)
Zhang et al. (2006)
Long et al. (2009)
PILOT (Roling, Sritharan, and Suleiman 2011)
PSU (Smith et al. 2011)
Long and Anderson (2014)
ZJU-ICL (Yang et al. 2016)
Long (2016)
Lehane et al. (2017)
Adhikari et al. (2018)
TxDOT (Moghaddam et al. 2018)
Tang and Phoon (2018a, 2018b, 2019b)
Tang and Phoon (2018c, 2019c)
Long (2013)
Flynn (2014)
FHWA DFTLD (Abu-Hejleh et al. 2015)
Dithinde et al. (2011)
IFSTTAR (Burlon et al. 2014)
Niazi (2014)
Galbraith, Farrell, and Byrne (2014)
AUT-CPT (Moshfeghi and Eslami 2018)
WBPLT (Chen et al. 2014)
LADOTD (Rauser and Tsai 2016)
Nanazawa et al. (2019)
Almeida and Liu (2018)
EPRI (Kulhawy et al. 1983)
Huang and Bathurst (2009)
Miyata and Bathurst (2012a)
Miyata and Bathurst (2012b)
Miyata, Bathurst, and Allen (2014)
Miyata and Bathurst (2015)
Miyata and Bathurst (2019)
Allen and Bathurst (2018)
Miyata, Yu, and Bathurst (2018)
Wood et al. (2012a, 2012b)
Lazarte (2011)
Cheung and Shum (2012)
Lin, Bathurst, and Liu (2017)
Liu et al. (2018)
Yuan et al. (2019)
Miyata, Bathurst, and Konami (2011)
Travis, Schmeeckle, and Sebert (2011)
Bahsan et al. (2014)
Wu, Ou, and Ching (2014)
White, Cheuk, and Bolton (2008)
Stuyts, Cathie, and Powell (2016)
Ismail, Najjar, and Sadek (2018)
White, Cheuk, and Bolton (2008)
Stuyts, Cathie, and Powell (2016)
Tang and Phoon (2016)
Global
Global
Global
USA/Europe
Worldwide
–
Hong Kong
Egypt
Global
Iowa, USA
Las Vegas Valley
Illinois, USA
Texas
Global
Global
Global
Global
Global
USA
Florida, USA
Global
Hong Kong
Wisconsin, USA
Iowa, USA
Global
Illinois, USA
Global
Wisconsin, USA
Global
Wyoming, USA
Texas
Global
Canada/USA
Wisconsin, USA
United Kingdom
Mainly in USA
South Africa
France
Global
Ireland
Global
Global
Louisiana, USA
Japan
Canada
USA
–
Japan
Japan
Japan
Japan
Global
–
–
Texas, USA
–
Hong Kong
Global
–
China
Japan
Global
–
Global
–
–
–
–
–
–
Laboratory/field
Field
Field
Field
Prototype
Centrifuge
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field (static/dynamic)
Field (dynamic)
Field
Field
Field (dynamic)
Field
Field (static/dynamic)
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field
Field (static/dynamic)
Field
Field
Field
Laboratory
Laboratory/in situ
Laboratory
Laboratory
Field
In situ
Field
In situ/laboratory
Laboratory
Field
Field
In situ
In situ
In situ
In situ
Field
Field
In situ
Small/full-scale
Small/full-scale
Small scale/centrifuge
Small/full-scale
Small/full-scale
Laboratory
Field
Cohesionless
Rock
Cohesionless
Cohesionless
Various
Clay with sand
Rock/saprolite
Various
Soft rock
Various
Caliche
Weak rock
Various
Various
Clay/sand
Clay/sand
Clay/sand
Gravel
Cohesionless
Various
Various
Weathered granite
Various
Various
Various
Various
Sand
IGM
Various
Soft rock
Various
Various
Various
Various
Sand
Various
Various
Various
Various
Various
Various
Various
Various
Various
Ontario soils
Various
Cohesionless
Cohesionless
Various
N/A
Various
Cohesionless
Various
Various
Cohesionless
–
CDG/CDV
–
–
Various
Various
Various
Clay
Cohesive
Sand
Sand
Sand
Sand
Sand
Cohesive
549
122
400
80
1026
159
38
318
190
38
41
155
27
320
88
99
40
24
112
78
420
1514
316
275
322
111
117
215
120
25
33
783
1010
182
116
1567
174
174
330
175
466
613
1465
441
47
804
318
652
503
362
520
113
378
202
650
166
913
123
95
144
28
157
43
24
61
108
143
54
192
78
25
(Continued )
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
11
Table 3. Continued.
Geotechnical structures
Database/reference
Data source
Test type
Shoring anchors
Chahbaz, Sadek, and Najjar (2019)
Beirut
Field
Cantilever wall
Excavation (stability)
Phoon et al. (2009)
Marsland (1953)
–
–
Excavation (wall displacement)
Long (2001)
Moormann (2004)
Wang, Xu, and Wang (2010)
Wu, Ching, and Ou (2013)
Global
Global
Shanghai
Taipei
Centrifuge
Small-scale
Large-scale
Field
Field
Field
Field
Geomaterial
Clay/marl/
limestone
Sand
Loose/dense sand
Various
Soft soil
Soft soil
Soft clay
N
70
20
23
10
296
530
300
22
Notes: CDG = completely decomposed granite; CDV = completely decomposed volcanic; IGM = intermediate geomaterial; N = number of load tests; NUS =
National University of Singapore; UWA = University of Western Australia; ZJU = Zhejiang University; ICL = Imperial College London.
carried out perhaps with the exception of liquefaction
(Andrus, Stokoe, and Chung 1999; Cetin et al. 2004;
Moss et al. 2006; Idriss and Boulanger 2010; Ku et al.
2012; Juang, Ching, and Luo 2013; Kayen et al. 2013).
We have a lot of data, but our data is mainly “dark”
because it is typically not exploited to provide insights
or to support decision making once the project that produced the data is completed.
For soil/rock properties, the most basic design decision
in current geotechnical practice is to estimate their values
from other test results, typically field test results. Phoon
and Kulhawy (1999a) identified at least three sources of
uncertainties in a comprehensive statistical study of a
broad range of laboratory and field test data: (1) spatial
variability, (2) measurement errors (including statistical
uncertainty due to limited measurements), and (3) transformation uncertainty. The third source of uncertainty
arising from transforming a soil/rock parameter such as
the overconsolidation ratio to a design parameter such
as the normalised undrained shear strength can be significant as shown in the data scatter in Figure 1.
Although it is widely known that local transformation
models (dashed lines in Figure 1) are preferred to those
calibrated from a generic database such as CLAY/10/
7490, there are no methods to quantify this “site effect”
for data routinely collected in a typical project (in contrast to data specially collected for a research study).
Nonetheless, the variety of dashed lines, each referring
to a local transformation model, clearly shows that this
site effect is important. The need for building regulations
to mandate a site investigation in every project is a recognition that every site is unique to some degree. Clearly,
geotechnical data are “uncertain” and “unique” to some
extent. The former characteristic is more familiar and
better studied.
Building regulations do not permit site investigation
efforts to vary as a function of how much is known at
neighbouring/comparable sites, even if the sites were to
be adjacent to the site of interest, possibly because
there are no statistical methods that can manage uncertain, sparse, and somewhat unique data from different
sites in an acceptable way. The million dollar question
(literally, if one were to consider how many mandatory
site investigations are carried out worldwide in any
time period) is whether we can get more value from
site data beyond establishing generic transformation
models.
4. Value of data
Data is not information, information is not knowledge,
knowledge is not understanding, understanding is not
wisdom.
Clifford Stoll
Geotechnical data are often referred to as “uncertain” in
the qualitative sense of “I don’t know”, rather than with
a mathematical formalism in mind. In fact, it is accurate to say that the majority harbours the sentiment
that formalism such as statistics is not possible, because
of data scarcity. One may venture to guess that the slow
progress in geotechnical reliability-based design or
other formal risk-informed design methodology is partially impeded by the lack of an accurate understanding
of the attributes of geotechnical data beyond broad generalities such as “uncertain” and “scarce”. In fact, many
practitioners do not appreciate the power of statistics in
its unusual ability to quantify even the uncertainties in
the models it posits. In short, statistics can quantify the
precision limits of its own models consistently based on
the available data at hand. The National Research
Council (1995) made this point rather clearly: “The
lack of a large data set does not preclude the use of
probability theory. Probability theory can be used to
evaluate the uncertainties involved in working with
meager information”. Nonetheless, this deep insight
has been lost, as practitioners continue to express
some reservations under the misconception that there
is insufficient data to characterise a probability model,
such as its parameters (mean, COV) and its shape (normal, lognormal, beta, etc.). Many miss the point that
even ignorance can be approximately quantified. This
is exceedingly powerful, because it allows an engineer
12
K.-K. PHOON
Figure 1. Correlation between normalised undrained shear strength (su /s′ v ) and overconsolidation ratio (OCR) (Ching and Phoon
2019b).
to weigh the cost of making a decision against the cost
of collecting more information to reduce imprecision in
some aspects of the problem. This paper does not cover
sensitivity analysis, but its usefulness to decision making is clear. A naïve one-at-a-time deterministic sensitivity analysis can produce misleading results for a
number of reasons, but one simple reason would be
the inability to take care of dependencies in deterministic analysis. Dependency is a feature of all multivariate
real world data (Ching, Li, and Phoon 2016). It is most
commonly captured by a correlation coefficient in basic
statistics (Ching, Phoon, and Li 2016).
Phoon (2018) suggested that the attributes of geotechnical data can be succinctly described as MUSIC:
Multivariate, Uncertain and Unique, Sparse, and
InComplete. It is useful to clarify in passing that
“scarce” refers to a small number of measurements,
while “sparse” refers to a small number of measurements widely distributed in space. Given that all site
data are situated in space (and sometimes, in time),
the term “sparse” is more descriptive as it is unlikely
for measurements to be taken at one corner of a site.
Phoon, Ching, and Wang (2019) further suggested
that MUSIC can be re-interpreted to cover extremes:
Multivariate, Uncertain and Unique, Sparse, Incomplete, and potentially Corrupted. Ching and Phoon
(2019b) subsequently extended MUSIC to MUSIC-X,
where “X” denotes the spatial/temporal context of the
data. The screening for extremes or outliers and spatial
variability are clearly important, but these aspects are
not covered in this paper. Table 4 is a site-specific
example of an actual MUSIC database from Taipei.
With the exception of the mobilised undrained shear
strength [su(mob)], each column contains the results
from a different and independent test. Good practice
requires a suite of tests to be conducted for cross-validation, identification of layer boundaries, estimation of
design properties, and others. Hence, geotechnical data
are intrinsically “multivariate” in nature. There is an
obvious tradeoff between conducting different tests in
different locations and conducting different tests in
the same location. The former strategy collects more
information on the spatial variability of the site. The
latter strategy collects information on the cross-correlations among all tests. In practice, it is common to
adopt an intermediate strategy that involves conducting
different test combinations at different depths and
locations. The greyed out cells in Table 4 denote absent
measurements. Hence, geotechnical data are typically
“incomplete”. A data table without missing entries is
an exception rather than the norm in geotechnical
engineering. It is useful to observe that the greyed out
cells are not randomly distributed. They occur more
frequently in columns where measurements are more
costly and the percentage of absent measurements can
be very high in these columns.
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
13
Table 4. Site investigation data for a silty clay layer at a Taipei site (Ou and Liao 1987).
Test results
Depth (m)
2
su (kN/m )
2
su(mob) (kN/m )
LL
PI
LI
′
s v /Pa
s′ p /Pa
su (mob)/s′ v
qt1
12.8
UU
55.2
46.9
30.1
9.1
1.20
1.26
1.71
0.37
3.35
14.8
VST
50.7
52.9
32.8
12.8
1.43
1.43
0.36
3.34
16.1
UU
61.9
51.7
36.4
14.5
1.24
1.54
0.33
3.15
17.8
UU
54.2
42.8
41.9
18.9
0.90
1.68
1.79
0.25
2.74
18.3
VST
59.5
59.3
1.72
0.34
2.76
20.2
UU
73.1
60.5
38.1
17.3
0.70
1.88
0.32
2.73
22.7
VST
63.3
64.4
37.0
16.0
0.58
2.08
0.31
2.97
24.0
UU
82.2
67.5
38.0
16.2
0.75
2.19
2.19
0.30
2.80
26.6
UU
98.1
82.1
34.8
13.8
0.80
2.41
0.34
3.92
Note: LL = liquid limit; PI = plasticity index; LI = liquidity index; s′ v = vertical effective stress; s′ p = preconsolidation stress; Pa = atmospheric pressure = 101.3 kPa;
qt = (corrected) cone tip resistance; qt1 = (qt − sv )/s′ v ; su = undrained shear strength; su(mob) = mobilised su values (Mesri and Huvaj 2007).
Each row (record) in a MUSIC database such as Table
4 refers to data collected at the same depth from different
tests conducted in close proximity. There are only n = 9
rows in Table 4. Hence, geotechnical data are “sparse”.
However, this is not true for a generic database such as
CLAY/10/7490. The CLAY/10/7490 database consists
of n = 7490 records from 30 countries for ten clay parameters. In short, site-specific data can be sparse, but
generic data are not sparse, although they may not be
complete and directly applicable to a specific site. To
counter the prevalent sentiment that there is no big
data in geotechnical engineering, Phoon, Ching, and
Wang (2019) explicitly refer to any data that are potentially useful but not directly applicable to the decision at
hand as Big Indirect Data (BID). A generic database will
be one type of BID. A compilation of case studies can be
regarded as another type of BID.
Although site effects are well known, they are mainly
characterised in research studies through a testing
programme that is more detailed than what is routinely
carried out in practice and for rather distinctive geomaterials. Kulhawy and Mayne (1990) pointed out that
“comprehensive characterization of the soil at a particular
site would require an elaborate and costly testing
programme, well beyond the scope of most project budgets”. To the knowledge of the author, no one has quantified site effects numerically based on more routine data
such as those shown in Table 4 commonly collected at a
project level. In practice, site effects are broadly appreciated
based on geology, soil mechanics, and experiences at comparable sites, rather than characterised quantitatively
through a detailed multivariate analysis of the site data
that meets MUSIC constraints. The typical caveat included
in design guides would include a general statement such as
caution must always be exercised when using broad,
generalized correlations of index parameters or in-situ
test results with soil properties. The source, extent, limitations of each correlation should be examined carefully
before use to ensure that extrapolation is not being done
beyond the original boundary conditions. ‘Local’ calibrations, where available, are to be preferred over the
broad, generalized correlations. (Kulhawy and Mayne
1990)
Notwithstanding this sensible caveat, the engineer is typically left with no recourse but to use these generalised correlations in the absence of “local” versions. Hence, BID is
already routinely used in practice in the form of Figure 1.
One could surmise that it has some real value. The
research challenge is to distil more value out of BID.
4.1. Bayesian machine learning
Can we address some attributes of MUSIC-X based on
the meagre information we have at hand? This is arguably the central question that practitioners are most
interested to know. The short answer is yes. Recently,
Ching and Phoon (2019c) proposed a novel Bayesian
machine learning method to do this, namely to construct
a site-specific distribution function for a MUSIC database such as that shown in Table 4. Each database is
effectively a table with m columns representing soil parameters (Y1, Y2, … , Ym) and n rows representing
measurements at different depths. The observed data
are denoted by Y o and unobserved data denoted by
Y u. Because soil parameters can be highly non-normal,
Ching and Phoon (2015) adopted an analytical transformation based on the Johnson distribution to convert
(Y1, … , Ym) to approximately normal data. The approximately normal data are denoted by x = (X1, … , Xm)T,
where “T” refers to vector/matrix transpose. A key
assumption made in Ching and Phoon (2019c) is that
x at a certain depth follows the multivariate normal
probability density function (PDF):
1
m
−
f (x|ms , Cs ) = |Cs | 2 (2p) 2
1
T −1
× exp − (x − ms ) Cs (x − ms ) (1)
2
−
14
K.-K. PHOON
The multivariate normal PDF has mean vector = μs and
covariance matrix = Cs; the subscript “s” is to highlight
that μs and Cs are “site-specific”. Because site-specific
data are sparse (small n), it is technically challenging to
estimate μs and Cs using conventional methods such as
matching moments or maximising likelihood. It is also
very challenging to estimate the statistical uncertainties
associated with μs and Cs, which are significant for a
fairly typical record size of around 10. Ching and
Phoon (2014) pointed out that it is impossible to guarantee Cs to be positive definite when the multivariate data is
incomplete. To address these critical limitations, Ching
and Phoon (2019c) developed a novel Gibbs sampler
(GS) to overcome this long standing challenge. The key
idea is to treat μs, Cs, and x u (transformed from Y u) as
unknown random quantities and to sequentially sample
one random quantity at a time from distributions conditioned on the rest of the quantities and the observed
data x o (transformed from Y o) using GS. Simulation is
practical because these conditional probabilities are
available in closed-form for suitably chosen conjugate
priors. While Bayesian methods are known to be very
powerful, there is no acknowledgment in the literature
that these methods are very complex and computationally intensive. Excessive emphasis on what works in principle rather than what works in practice will not attract
more users. This GS has been generalised to MUSIC-X
recently (Ching and Phoon 2019b).
Consider properties at a new depth (xnew) that does
not appear in the training data previously used in the
GS. Based on the total probability theorem, the conditional multivariate PDF f(xnew|X o) is a mixture of
multivariate normal PDFs:
f (x
new |X
0
) = f (x
new |m s ,
1
≈
T − tb
T
t=tb +1
Cs )f (m s , Cs |X 0 )dm s dCs
N(x
new |m s.t ,
Cs,t )
(2)
where (μs,t, Cs,t) are the GS samples at time step = t; tb is
the end of the burning-period; and T is the total number
of GS time steps or samples. The simulation of a sitespecific probability distribution appears very complicated to the average engineer, but it can support a critical
design decision on how to choose soil/rock properties at
a particular site by “learning” from site-specific data
alone. An example based on actual data from a Taipei
site (Table 4) is shown in Figure 2. Although Table 4
contains 9 records, note that s′ p /Pa is only measured
at 3 depths. It is not surprising that the statistical uncertainties in Figure 2(b) is large. For Figure 2(c or d) where
9 site-specific data points are available, it can be seen that
the site-specific PDF (solid grey markers for MUSIC-X
and open black markers for MUSIC) is less scattered
and arguably more informative for this particular site
than the generic PDF (blue markers). The MUSIC and
MUSIC-X PDFs are similar, because the sampling intervals in Table 4 are large and spatial variability is thus not
well captured. Figure 2(d) further shows that Su /s′ v is
less correlated to qt1 at the Taipei site than the generic
version. The generic correlations in CLAY/10/7490 are
0.91, −0.57, −0.50, and 0.73 for the transformation
models shown in Figure 2(a–d), respectively, in standard
normal space (Ching and Phoon 2014). To the author’s
knowledge, this learning algorithm (MUSIC or
MUSIC-X) is the first of its kind. Even when used by
itself, the site-specific PDF can guide the engineers to
select conservative design values more appropriate for
a particular site by using the approximate lower bounds
of the solid grey markers (rather than generic blue markers) shown in Figure 2. If these values were ascertained
to be overly conservative, more measurements could be
taken and the favourite question “how many measurements are enough” can be addressed quantitatively by
the reduction in the scatter of the solid grey markers
that will improve the lower bounds. This MUSIC or
MUSIC-X PDF is basically a quantification of site
“uniqueness” from a data-informed perspective and
there is clear value to do this. Once site uniqueness can
be captured numerically, it opens all kinds of interesting
research avenues to combine site-specific data with generic data from other sites in the entire world, not merely
in the restricted region that an engineer practices in. The
similarity index approach outlined below is one such
example.
4.2. Similarity index approach
The next natural question is how to combine a sitespecific PDF with a generic PDF in a more discriminate
way that accounts for site “uniqueness”. Ching and
Phoon (2019d) developed a similarity index (S) based
on f(xnew|X o) to identify records from a generic database
that are “similar” to those from a specific site. A second
example of a MUSIC database from Onsøy, Norway is
given in Table 5. This set of site-specific data is shown
as red solid triangles in Figure 3 against a background
of generic data from CLAY/10/7490 shown as grey
solid circles. Figure 3 also presents data from another
site in Norway (Drammen) from Lacasse and Lunne
(1982). The Drammen and Onsøy sites are roughly
50 km apart with comparable geologic origins (Lacasse
et al. 1981; Lacasse and Lunne 1982). The data are identified as “similar” (S > 1) (black solid circles) or
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
15
Figure 2. Transformation models based on generic data (CLAY/10/7490) and site-specific data (MUSIC and MUSIC-X simulations) for a
Taipei site (Table 4) (J. Ching, personal communications, 2019).
Note: solid and dashed lines are the median and 95% confidence interval for the generic data.
“similarity” to the data at one site, and (3) perform a
weighted regression on a combined dataset containing
the site-specific data and the generic database. The
equivalent generic sample size (Neq) is the sum of the
weights produced by the generic records. Figure 4 presents the construction of quasi-site-specific transformation models using different number of records from
Table 5: (a) all 9 records; (b) 6 records (depths = 1.9,
3.5, 5.2, 9.5, 10.8, and 13.4 m); (c) 2 records (depths =
“dissimilar” (S < 1) (black open circles) using this similarity index. The practical benefit of doing this is that
we can replace a generic transformation model such as
Figure 1 by a quasi-site-specific version that is based
on the site-specific data and appropriately weighted generic data. Ching and Phoon (2019d) recommended the
following procedure to do this: (1) assume the weight
of a site-specific data point is 1, (2) assign weights to
records in a generic database as a function of their
Table 5. Site investigation data for a marine clay layer in Onsøy (Norway) (Lacasse and Lunne 1982).
Site-specific data Y
Index
1
2
3
4
5
6
7
8
9
Depth (m)
1
1.9
3.5
5.2
7.6
9.5
10.8
13.4
16.3
LL
56.2
50.2
59.9
56.8
66.3
65.1
74.4
71.4
72.7
PI
LI
s′ v /Pa
s′ p /Pa
su (mob)/s′ v
St
Bq
qt1
qtu
OCR
20
18.1
30.5
22.9
31.5
29.6
36.1
35.8
34.7
1.54
1.82
0.93
1.07
0.87
0.97
0.81
0.87
0.76
0.06
0.12
0.22
0.32
0.47
0.58
0.65
0.81
0.99
0.85
0.6
0.48
0.45
0.54
2.03
0.91
0.48
0.37
0.24
0.25
0.25
0.24
0.24
6
14
15
7
14
12
9
0.16
0.24
0.3
0.35
0.47
0.41
0.46
0.47
0.55
29.11
17.69
10.52
7.7
5.89
6.19
5.93
5.95
6.13
25.57
14.58
8.41
6.11
4.25
4.74
4.31
4.24
3.88
13.99
5.2
2.26
1.42
1.17
0.84
1.05
0.99
1.28
1.29
1
Notes: LL = liquid limit; PI = plasticity index; LI = liquidity index; s′ v = vertical effective stress; s′ p = preconsolidation stress; Pa = atmospheric pressure =
101.3 kPa; qt = (corrected) cone tip resistance; qt1 = (qt − sv )/s′ v ; su = undrained shear strength; su(mob) = mobilised su values (Mesri and Huvaj 2007); St
= sensitivity; qt = (corrected) cone tip resistance; u2 = pore pressure behind cone; Bq = pore pressure ratio = (u2-u0)/(qt-σv); u0 = hydrostatic pore pressure;
qt1 = (qt − sv )/s′ v ; qtu = (qt − u2 )/s′ v .
16
K.-K. PHOON
Figure 3. Automatic detection of records from a generic database CLAY/10/7490 that are “similar” to those from a specific site in Onsøy,
Norway (Ching and Phoon 2019d).
1.9 and 13.4 m); and (d) no site-specific data is available.
Note that s′ p /Pa is not measured at a depth = 9.5 m.
Hence, the number of site records for this specific transformation model is Ns = 8 and 5 for Figure 4(a and b),
respectively. It could be seen that the quasi-site-specific
transformation model is more customised to Onsøy
when there are 8 site records but revert back to the generic form when there are only 2 site records. Note that
the black solid circles are meant for reference only; the
full set of generic records (grey solid circles) with appropriate weights is used for regression. This similarity
index approach treats the records in a generic database
as independent, although a natural grouping based on
site exists. Data records measured within a site tend to
be more similar with each other than those measured
in other sites. Ching, Wu, and Phoon (2020) proposed
a hierarchical Bayesian model to capture this additional
site information commonly made available in generic
databases. No quantitative information related to the
site location such as GPS location, nearest city, region,
country and others is needed. Only qualitative knowledge that a group of data records are measured within
the same project site is needed. In the author’s opinion,
the above research is preliminary, because many structural elements of a generic database such as groups
have not been studied and the algorithms are not in a
full learning mode, including learning from expert judgment. However, they are founded on Bayesian theory
and they will set the scene for more powerful algorithms
to emerge in the near future. It is quite likely that some
aspects of our engineering experience would be “digitized” in the future in the sense of being captured by
algorithms that can learn from both data and their interactions with the engineers.
5. Let data speak for themselves
Hand (2014) said:
In general, when building statistical models, we must
not forget that the aim is to understand something
about the real world. Or predict, choose an action,
make a decision, summarize evidence, and so on, but
always about the real world, not an abstract mathematical world: our models are not the reality.
Let’s take a step back and ask ourselves why we need a
model in the first place. One answer is that we do not
have sufficient data to make a decision without mediation
by a model. The simplest statistical model is to assume
data are independent and identically distributed (i.i.d.).
Limited data are needed to characterise this model, but
it clearly deviates from a reality that exhibits spatial variability. The classical random field model tries to do better,
but it requires more data for statistical characterisation.
More recent multiple point methods in geostatistics can
consider more than two-point autocorrelation (or
second-order) information (Mariethoz and Caers
2015), but they require richer data in the form of
training images. Phoon, Ching, and Wang (2019) opined
that a “taxonomy of methods based on the type/amount
of data available could help guide future development in
data-driven algorithms and strengthen a virtuous cycle of
data collection hardware developing hand in hand with
algorithms”. Wang and Zhao (2016, 2017) explored a
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
17
Figure 4. Construction of quasi-site-specific transformation models by combining different amount of Onsøy data with appropriately
weighted records in CLAY/10/7490 (Ching and Phoon 2019b).
Note: median (solid line) and 95% confidence interval (dashed lines).
new sampling paradigm in digital signal processing
called compressive sampling (or sensing, CS) that can
reconstruct a near replica of the original signal from a
small number of measurements. Wang, Zhao, and
Phoon (2018) subsequently developed a Bayesian Compressive Sampling-Karhunen-Loève (BCS-KL) expansion version that can generate random field samples
(RFSs) directly from sparse measurements. This BCSKL generator is shown to be capable of dealing with
much more general non-Gaussian and non-stationary
RFSs, including RFSs with unknown non-stationary
auto-covariance structure without explicit estimation of
the autocorrelation function (Montoya-Noguera et al.
2019) and RFSs with unknown trend function without
de-trending (Wang Y et al. 2019). In addition, the BCSKL generator may be readily extended to simulate
cross-correlated bivariate RFSs (Zhao and Wang 2018).
An important open question is whether the basis functions in BCS (which are prescribed) can reproduce the
finer details of some sample paths such as those produced
by the Whittle-Matérn autocorrelation function correctly
when there are sufficient measurements (convergence). A
second related question is whether it can retain its key
practical advantage of representing such “rough” sample
paths using sparse measurements (rate of convergence).
It will be interesting to further explore the possibility of
decoupling BCS from the KL expansion, which contains
a fairly strong multivariate Gaussianity assumption.
18
K.-K. PHOON
Notwithstanding the above, it is clear that more recent
non-classical methods can handle much more realistic
data up to bivariate vector fields that are potentially
non-stationary and non-Gaussian, without making
strong demands on data such as sample sizes and completeness that cannot be fulfilled in practice. Classical
models have reigned supreme as we have always assumed
one can collect sufficient and appropriate data for characterisation. Many models were developed without a
characterisation method in mind and in fact, many
models exist without a satisfactory characterisation
method in place even years after they were introduced.
One cannot help but ask if this supposedly minor footnote pertaining to data collection and empirical characterisation thereof is really minor. It will be fruitful to
ask ourselves the inverse question: what data we have
and what models can we develop to make full use of
the data at hand, warts and all? In some sense, these
non-classical models that can function under very general conditions are allowing our data to speak for
themselves.
decisions to account for uncertainties explicitly. Geo
4.0 will do a lot more, but we (the entire geotechnical
engineering community) need to engage in re-imagining
the future of our profession with greater boldness. So,
where did engineering judgment go? To quote Professor
Ralph Peck, who is widely recognised for his practical
engineering wisdom: “Theory and calculation are not
substitute for judgement, but are the basis for sounder
judgments” (NGI 2019). Professor Peck may not have
imagined digitalisation, but he would agree that better
calculations grounded on actual data will make our
decisions even better. In the much longer term, no one
really knows the role of humans in an immersive
cyber-physical reality.
In fact, why don’t we initiate an “AlphaGeo”
project to see how far we can monetise our data and
to sharpen the role of engineering judgment? One suspects (with good reasons) that the story of statistics
will unfold in exciting and unexpected ways in the near
future. It is not far-fetched to wonder if data may be
the only reality.
6. Concluding thoughts
Acknowledgements
The moral of this story is that the value of geotechnical
data is significantly under-appreciated and not fully
exploited for decision making. Our data is “dark” in
the sense that it is stored primarily for compliance purposes, rather than shared and actively mined for insights
that can inform future decision making. The world is
being revolutionised by new and powerful ways of collecting, sharing, analysing, and monetising data. Clearly,
there is a pressing need for the geotechnical engineering
community to engage in this digital transformation.
There is no doubt that reasoned judgment is further
enhanced when it is guided by relevant data and analytical tools that make the most sensible use of data, be it
using physics, statistics, machine learning, or some combinations thereof. One Geo 4.0 approach could be to
develop a clever physics-informed and data-driven
“grey box” algorithm to shortlist “similar” sites from
BID for the engineer to further refine based on his/her
experience and for the algorithm to “learn” and thus
become even more discriminating in the future. In this
way, our collective experiences could be partially “digitized” as well. For the first time, we are starting to realise
we can combine experience and data in even more clever
ways. This paper describes some preliminary research on
Bayesian learning that grew out of a simple idea studied
by pioneers such as Professor Peter Lumb since the
sixties. The idea is that uncertainties and even limited
knowledge of these uncertainties can be quantified
numerically using statistics, thus allowing design
This paper is an update of the 10th Lumb Lecture, delivered at
the University of Hong Kong, 6 December 2018. The author
would like to thank Professor Limin Zhang, Editor in Chief
of Georisk, for his encouragement to prepare this paper. The
author is also grateful to the Department of Civil Engineering,
The University of Hong Kong and the Geotechnical Division,
The Hong Kong Institution of Engineers for their kind invitation to deliver this lecture. In particular, the generous hospitality extended by Professor Zhongqi Quentin Yue, Honorary
Professor Chack Fan Lee, and Dr Victor Li is deeply appreciated. The author also thanked Dr Victor Li for sharing the
article: “Excerpts from interview with Professor Peter
Lumb”, Hong Kong Statistical Society Newsletter, Vol. 9,
Issue 1, 1986. This paper was drafted during the author’s sabbatical at the Institute for Risk and Reliability, Leibniz University, which was funded by the Alexander von Humboldt
Foundation. Last but not least, the author is deeply indebted
to Prof Jianye Ching, National Taiwan University, for sharing
his many deep insights and research in Bayesian learning and
for preparing all the figures, to Dr Chong Tang for his extensive editorial assistance, and to the following colleagues for
their invaluable comments: Zijun Cao, Marco D’Ignazio (for
updating CLAY/9/249), Sina Javankhoshdel, C. Hsein Juang,
Leena Korkiala-Tanttu, Tim Länsivaara, Stefan Larsson (for
updating SE-CLAY/4/499), Andy Yat-fai Leung, Monica Löfman, Sukumar Pathmanandavel, Anders Prästings, Mengfen
Shen (for updating liquefaction databases), Yu Wang (for discussions on Bayesian compressive sampling), and Dongming
Zhang (for updating SH-CLAY/11/4051).
Disclosure statement
No potential conflict of interest was reported by the author.
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
ORCID
Kok-Kwang Phoon
http://orcid.org/0000-0003-2577-8639
References
AbdelSalam, S., F. Baligh, and H. M. El-Naggar. 2015. “A
Database to Ensure Reliability of Bored Pile Design in
Egypt.” Proceedings of the Institution of Civil Engineers –
Geotechnical Engineering 168 (2): 131–143.
Abu-Hejleh, N., M. Abu-Farsakh, M. Suleiman, and C. Tsai. 2015.
“Development and Use of High-Quality Databases of Deep
Foundation Load Tests.” Transportation Research Record:
Journal of the Transportation Research Board 2511: 27–36.
Adhikari, P., Y. Gebreslasie, K. Ng, T. Sullivan, and S. Wulff.
2018. “Static and Dynamic Analysis of Driven Piles in Soft
Rocks Considering LRFD Using a Recently Developed
Electronic Database.” In Installation, Testing, and Analysis
of Deep Foundations (GSP 294), 83–92. Reston, VA: ASCE.
Akbas, S. 2007. “Deterministic and Probabilistic Assessment of
Settlements of Shallow Foundations in Cohesionless Soils.”
Ph.D. thesis, Cornell University
Allen, T., and R. Bathurst. 2018. “Application of the Simplified
Stiffness Method to Design of Reinforced Soil Walls.”
Journal of Geotechnical and Geoenvironmental Engineering
144 (5): 04018024.
Almeida, A. P. R. P., and J. Liu. 2018. “Statistical Evaluation of
Design Methods for Micropiles in Ontario Soils.” DFI
Journal - The Journal of the Deep Foundation Institute 12
(3): 133–146.
Andrus, R. D., K. H. Stokoe, and R. M. Chung. 1999. Draft
Guidelines for Evaluating Liquefaction Resistance Using
Shear Wave Velocity Measurements and Simplified
Procedures. Gaithersburg, MD: US Department of
Commerce, Technology Administration, National Institute
of Standards and Technology.
Asem, P., J. Long, and P. Gardoni. 2018. “Probabilistic Model
and LRFD Resistance Factors for the Tip Resistance of
Drilled Shafts in Soft Sedimentary Rock Based on Axial
Load Tests.” In Innovations in Geotechnical Engineering:
Honoring Jean-Louis Briaud (GSP 299), 1–46. Reston, VA:
ASCE.
Augustesen, A. 2006. “The Effects of Time on Soil Behaviour
and Pile Capacity.” Ph.D. thesis, Aalborg University
Baecher, G. B. 1987. Statistical Analysis of Geotechnical Data.
Report No. GL-87-1, U.S. Vicksburg: Army Engineer
Waterways Experiment Station.
Baecher, G. B., and J. T. Christian. 2003. Reliability and
Statistics in Geotechnical Engineering. London and
New York: John Wiley and Sons.
Baecher, G. B., and J. T. Christian. 2019. “TH Wu and the
Origins of Geotechnical Reliability.” Georisk 13 (4): 242–246.
Bahsan, E., H. J. Liao, J. Y. Ching, and S. W. Lee. 2014.
“Statistics for the Calculated Safety Factors of Undrained
Failure Slopes.” Engineering Geology 172: 85–94.
Bathurst, R. J., S. Javankhoshdel, and T. M. Allen. 2017. “LRFD
Calibration of Simple Soil-structure Limit States Considering
Method Bias and Design Parameter Variability.” Journal of
Geotechnical and Geoenvironmental Engineering 143 (9):
04017053.
Beer, M., Y. Zhang, S. T. Quek, and K. K. Phoon. 2013.
“Reliability Analysis with Scarce Information: Comparing
19
Alternative Approaches in a Geotechnical Engineering
Context.” Structural Safety 41: 1–10.
Beesley, M. E., and P. J. Vardanega. 2019. “Parameter
Variability of Undrained Shear Strength and Strain Using
a Database of Reconstituted Soil Tests.” Canadian
Geotechnical Journal, in press.
BS8004. 1986. Code of Practice for Foundations. London:
British Standards Institution.
Burland, J. B. 1987. “The Teaching of Soil Mechanics – a
Personal View.” Proceedings, 9th European Conference on
Soil Mechanics and Foundation Engineering, Dublin, 3
vols., 1427–1447.
Burlon, S., R. Frank, F. Baguelin, J. Habert, and S. Legrand.
2014. “Model Factor for the Bearing Capacity of Piles
from Pressuremeter Test Results: Eurocode 7 Approach.”
Géotechnique 64 (7): 513–525.
Cami, B., S. Javankhoshdel, and K. K. Phoon. 2020. “A Review
of Scale of Fluctuation for Spatially Varying Soils:
Estimation Methods and Values.” ASCE-ASME Journal of
Risk and Uncertainty in Engineering Systems, Part A: Civil
Engineering, under review.
Cao, Z., S. Zheng, D. Q. Li, and K. K. Phoon. 2019. “Bayesian
Identification of Soil Stratigraphy Based on Soil Behavior
Type Index.” Canadian Geotechnical Journal 56 (4): 570–586.
Casagrande, A. 1965. “Role of the ‘Calculated Risk’ in
Earthwork and Foundation Engineering.” Journal of Soil
Mechanics and Foundations Division 91 (SM4): 1–40.
Cetin, K. O., R. B. Seed, A. Der Kiureghian, K. Tokimatsu, L. F.
Harder, R. E. Kayen, and R. E. S. Moss. 2004. “Standard
Penetration Test-based Probabilistic and Deterministic
Assessment of Seismic Soil Liquefaction Potential.”
Journal of Geotechnical and Geoenvironmental Engineering
130 (12): 1314–1340.
Chahbaz, R., S. Sadek, and S. Najjar. 2019. “Uncertainty
Quantification of the Bond Stress – Displacement
Relationship of Shoring Anchors in Different Geologic
Units.” Georisk: Assessment and Management of Risk for
Engineered Systems and Geohazards 13 (4): 276–283.
Chen, Y. J., and F. H. Kulhawy. 1994. Case History Evaluation
of Behavior of Drilled Shafts under Axial and Lateral
Loading. Report EPRI TR-104601. Palo Alto, CA: Electric
Power Research Institute (EPRI)
Chen, Y. J., and Y. H. Lee. 2010. “Evaluation of Lateral
Interpretation Criteria for Drilled Shaft Capacity.” Journal
of Geotechnical and Geoenvironmental Engineering 136
(8): 1124–1136.
Chen, Y. J., M. R. Liao, S. S. Lin, J. K. Huang, and M. C. M.
.Marcos. 2014. “Development of an Integrated Web-Based
System with a Pile Load Test Database and pre-Analyzed
Data.” Geomechanics and Engineering 7 (1): 37–53.
Chen, Y. J., S. W. Lin, and F. H. Kulhawy. 2011. “Evaluation of
Lateral Interpretation Criteria for Rigid Drilled Shafts.”
Canadian Geotechnical Journal 48 (4): 634–643.
Cheung, R. W. M., and K. W. Shum. 2012. Review of the
Approach for Estimation of Pullout Resistance of Soil Nails.
GEO Report No. 264. Geotechnical Engineering Office,
Civil Engineering and Development Department, Hong
Kong
Ching, J., D. Q. Li, and K. K. Phoon. 2016. “Statistical
Characterization of Multivariate Geotechnical Data.” In
Reliability of Geotechnical Structures in ISO2394, edited by
K. K. Phoon and J. V. Retief, 89–126. Balkema: CRC Press.
20
K.-K. PHOON
Ching, J., K. H. Li, K. K. Phoon, and M. C. Weng. 2018.
“Generic transformation models for some intact rock properties.” Canadian Geotechnical Journal 55 (12): 1702–1741.
Ching, J., and K. K. Phoon. 2012. “Modeling Parameters of
Structured Clays as a Multivariate Normal Distribution.”
Canadian Geotechnical Journal 49 (5): 522–545.
Ching, J., and K. K. Phoon. 2013. “Multivariate Distribution for
Undrained Shear Strengths Under Various Test Procedures.”
Canadian Geotechnical Journal 50 (9): 907–923.
Ching, J., and K. K. Phoon. 2014. “Correlations Among Some
Clay Parameters – the Multivariate Distribution.” Canadian
Geotechnical Journal 51 (6): 686–704.
Ching, J., and K. K. Phoon. 2015. “Constructing Multivariate
Distribution for Soil Parameters.” Chap. 1 in Risk and
Reliability in Geotechnical Engineering, 3–76. Balkema:
CRC Press.
Ching, J., and K. K. Phoon. 2017. “Characterizing Uncertain
Site-specific Trend Function by Sparse Bayesian Learning.”
Journal of Engineering Mechanics 143 (7): 04017028.
Ching, J., and K. K. Phoon. 2019a. “Impact of Auto-correlation
Function Model on the Probability of Failure.” Journal of
Engineering Mechanics 145 (1): 04018123.
Ching, J., and K. K. Phoon. 2019b. “Constructing a Sitespecific Multivariate Probability Distribution Using
Sparse, Incomplete, and Spatially Variable (MUSIC-X)
Data.” Journal of Engineering Mechanics, under review.
Ching, J., and K. K. Phoon. 2019c. “Constructing Site-specific
Multivariate Probability Distribution Model Using Bayesian
Machine Learning.” Journal of Engineering Mechanics 145
(1): 04018126.
Ching, J., and K. K. Phoon. 2019d. “Measuring Similarity
between Site-specific Data and Records from Other Sites.”
ASCE-ASME Journal of Risk and Uncertainty in
Engineering Systems, Part A: Civil Engineering, in press.
Ching, J., K. K. Phoon, J. L. Beck, and Y. Huang. 2017.
“Identifiability of Geotechnical Site-specific Trend
Functions.” ASCE-ASME Journal of Risk and Uncertainty
in Engineering Systems, Part A: Civil Engineering 3 (4):
04017021.
Ching, J., K. K. Phoon, and D. Q. Li. 2016. “Robust Estimation
of Correlation Coefficients Among Soil Parameters Under
the Multivariate Normal Framework.” Structural Safety
63: 21–32.
Ching, J., K. K. Phoon, and J. W. Yu. 2014. “Linking Site
Investigation Efforts to Final Design Savings with
Simplified Reliability-based Design Methods.” Journal of
Geotechnical and Geoenvironmental Engineering 140 (3):
04013032.
Ching, J., J. S. Wang, C. H. Juang, and C. S. Ku. 2015. “Cone
Penetration Test (CPT)-based Stratigraphic Profiling
Using the Wavelet Transform Modulus Maxima Method.”
Canadian Geotechnical Journal 52 (12): 1993–2007.
Ching, J., S. S. Wu, and K. K. Phoon. 2016. “Statistical
Characterization of Random Field Parameters Using
Frequentist and Bayesian Approaches.” Canadian
Geotechnical Journal 53 (2): 285–298.
Ching, J., S. Wu, and K. K. Phoon. 2020. “Constructing Quasisite-specific Multivariate Probability Distribution Using
Hierarchical Bayesian Model.” Journal of Engineering
Mechanics, ASCE, under review.
DeGroot, D. J. 1996. “Analyzing Spatial Variability of in Situ
Soil Properties.” In Uncertainty in the Geologic
Environment: From Theory to Practice (GSP 58), edited by
C. D. Shackelford, P. P. Nelson, and M. J. S. Roth, 210–
238. Reston, VA: ASCE.
DeGroot, D. J., and G. B. Baecher. 1993. “Estimating
Autocovariance of In-situ Soil Properties.” Journal of
Geotechnical Engineering 119 (1): 147–166.
D’Ignazio, M., T. Lunne, K. H. Andersen, S. L. Yang, B. Di Buò,
and T. T. Länsivaara. 2019. “Estimation of Preconsolidation
Stress of Clays from Piezocone by Means of High-quality
Calibration Data.” AIMS Geosciences 5 (2): 104–116.
D’Ignazio, M., K. K. Phoon, S. A. Tan, and T. T. Länsivaara.
2016. “Correlations for Undrained Shear Strength of
Finnish Soft Clays.” Canadian Geotechnical Journal 53
(10): 1628–1645.
Dithinde, M., K. K. Phoon, M. Wet, and J. Retief. 2011.
“Characterization of Model Uncertainty in the Static Pile
Design Formula.” Journal of Geotechnical and
Geoenvironmental Engineering 137 (1): 70–85.
DNV. 2012. “Statistical Representation of Soil Data.”
Recommended Practice DNVGL-RP-C207, Det Norske
Veritas (DNV), Oslo, Norway.
Einstein, H. H., and G. B. Baecher. 1982. “Probabilistic and
Statistical Methods in Engineering Geology I. Problem
Statement
and
Introduction
to
Solution.”
In
Ingenieurgeologie und Geomechanik als Grundlagen des
Felsbaues/Engineering Geology and Geomechanics as
Fundamentals of Rock Engineering, edited by L. Muller,
47–61. Vienna: Springer.
Einstein, H. H., and G. B. Baecher. 1983. “Probabilistic and
Statistical Methods in Engineering Geology. Specific
Methods and Examples Part I: Exploration.” Rock
Mechanics and Rock Engineering 16 (1): 39–72.
Einstein, H. H., V. B. Halabe, J.-P. Dudt, and F. Descoeudres.
1996. “Geologic Uncertainties in Tunnelling.” In
Uncertainty in the Geologic Environment: From Theory to
Practice (GSP 58), edited by C. D. Shackelford, P. P.
Nelson, and M. J. S. Roth, 239–253. Reston, VA: ASCE.
EN 1997−1:2004. Eurocode 7: Geotechnical Design — Part 1:
General Rules. Brussels, Belgium: European Committee for
Standardization (CEN).
Feng, S., and P. J. Vardanega. 2019. “Correlation of the
Hydraulic Conductivity of Fine-grained Soils with Water
Content Ratio Using a Database.” Environmental
Geotechnics 6 (5): 253–268.
Fenton, G. A., F. Naghibi, D. Dundas, R. J. Bathurst, and D. V.
Griffiths. 2016. “Reliability-based Geotechnical Design in
the 2014 Canadian Highway Bridge Design Code.”
Canadian Geotechnical Journal 53 (2): 236–251.
Filippas, O. B., F. H. Kulhawy, and M. D. Grigoriu. 1988.
Reliability-based Foundation Design for Transmission Line
Structures: Uncertainties in Soil Property Measurement.
Report EL-5507(3). Palo Alto: Electric Power Research
Institute.
Flynn, K. 2014. “Experimental Investigations of Driven Cast
in-Situ Piles.” Ph.D. thesis, National University of Ireland,
Galway
Galbraith, A., E. Farrell, and J. Byrne. 2014. “Uncertainty in
Pile Resistance from Static Load Tests Database.”
Proceedings of the Institution of Civil Engineers Geotechnical Engineering 167 (5): 431–446.
Garder, J., K. Ng, S. Sritharan, and M. Roling. 2012. An
Electronic Database for Drilled SHAft Foundation Testing
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
(DSHAFT). Report No. InTrans Project 10-366, Iowa
Department of Transportation
Gartner. 2019. “IT Glossary.” Gartner. Accessed October 3,
2019. https://www.gartner.com/it-glossary/dark-data.
Geoengineer. 2019. “Professor Ralph Peck’s Legacy Website.”
Geoengineer.org. Accessed October 3, 2019. https://peck.
geoengineer.org/resources/words-of-wisdom.
Hand, D. J. 2014. “Wonderful Examples, but Let’s not Close
Our Eyes.” Statistical Science 29: 98–100.
Harr, M. E. 1987. Reliability-based Design in Civil Engineering.
New York: McGraw-Hill.
Hov, S., A. Prästings, E. Persson, and S. Larsson. 2019. “On
Empirical Correlations for Normalised Shear Strengths
from Fall Cone and Direct Simple Shear Tests in Soft
Swedish Clays.” Geotechnical and Geological Engineering,
under review.
Huang, B. Q., and R. J. Bathurst. 2009. “Evaluation of SoilGeogrid Pullout Models Using a Statistical Approach.”
Geotechnical Testing Journal 32 (6): 489–504.
Idriss, I. M., and R. W. Boulanger. 2010. SPT-based
Liquefaction Triggering Procedures. Davis, CA: Center for
Geotechnical Modeling, Department of Civil and
Environmental Engineering, University of California at
Davis. Report No. UCD/CGM-10/02.
Ismail, S., S. S. Najjar, and S. Sadek. 2018. “Reliability Analysis
of Buried Offshore Pipelines in Sand Subjected to Upheaval
Buckling.” In Proceedings of the Offshore Technology
Conference (OTC). Houston, TX: American Petroleum
Institute. OTC-28882-MS.
ISO2394:1973/1986/1998/2015. General Principles on Reliability
for Structures. Geneva, Switzerland: International
Organization for Standardization.
Jaksa, M. B. 1995. “The Influence of Spatial Variability on the
Geotechnical Design Properties of a Stiff, Overconsolidated
Clay.” Ph.D. Dissertation, University of Adelaide.
Jaksa, M. B., H. R. Maier, and M. A. Shahin. 2008. “Future
Challenges for Artificial Neural Network Modelling in
Geotechnical Engineering.” Proceedings, 12th International
Conference of International Association for Computer
Methods and Advances in Geomechanics (IACMAG),
1710–1719, Goa, India, October 1–6.
Joint TC205/TC304 Working Group Report. 2017. “Discussion
of Statistical/Reliability Methods for Eurocodes. International
Society for Soil Mechanics and Geotechnical Engineering.”
http://140.112.12.21/issmge/tc304.htm.
Juang, C. H., J. Ching, and Z. Luo. 2013. “Assessing SPT-based
Probabilistic Models for Liquefaction Potential Evaluation:
A 10-Year Update.” Georisk: Assessment and Management
of Risk for Engineered Systems and Geohazards 7 (3): 137–
150.
Kayen, R., R. E. S. Moss, E. M. Thompson, R. B. Seed, K. O.
Cetin, A. D. Kiureghian, Y. Tanaka, and K. Tokimatsu.
2013. “Shear-wave Velocity–Based Probabilistic and
Deterministic Assessment of Seismic Soil Liquefaction
Potential.” Journal of Geotechnical and Geoenvironmental
Engineering 139 (3): 407–419.
Ku, C. S., C. H. Juang, C. W. Chang, and J. Ching. 2012.
“Probabilistic Version of the Robertson and Wride
Method for Liquefaction Evaluation: Development and
Application.” Canadian Geotechnical Journal 49 (1): 27–44.
Kulhawy, F. H., B. Birgisson, and M. D. Grigoriu. 1992.
Reliability-based Foundation Design for Transmission Line
21
Structures: Transformation Models for In-Situ Tests.
Report EL-5507(4). Palo Alto: Electric Power Research
Institute.
Kulhawy, F. H., and P. W. Mayne. 1990. Manual on Estimating
Soil Properties for Foundation Design. Report EL-6800. Palo
Alto, CA: Electric Power Research Institute.
Kulhawy, F. H., T. D. O’Rourke, J. P. Stewart, and J. F. Beech.
1983. Transmission Line Structure Foundations for UpliftCompression Loading: Load Test Summaries. Appendix to
EPRI Final Report EL-2870. Report No. EL-3160-LD. Palo
Alto, CA: Electric Power Research Institute (EPRI)
Lacasse, S., M. Jamiolkowski, R. Lancellotta, and T. Lunne.
1981. “In Situ Characteristics of Two Norwegian Clays.”
Proceedings, 10th International Conference on Soil
Mechanics and Foundation Engineering, Stockholm, 2
vols., 507–511.
Lacasse, S., Z. Liu, and F. Nadim. 2017. “Probabilistic
Characterization of Soil Properties—Recognition of
Wilson Tang’s Contribution to Geotechnical Practice.” In
Geotechnical Safety and Reliability: Honoring Wilson
H. Tang (GSP 286), edited by C. Hsein Juang, R. B.
Gilbert, L. Zhang, J. Zhang, and L. Zhang, 2–26. Reston,
VA: ASCE.
Lacasse, S., and T. Lunne. 1982. “Penetration Tests in Two
Norwegian Clays.” Proceedings, Second European
Symposium on Penetration Testing, Amsterdam, 661–670.
Lacasse, S., and F. Nadim. 1996. “Uncertainties in
Characterising Soil Properties.” In Uncertainty in the
Geologic Environment: From Theory to Practice (GSP 58),
edited by C. D. Shackelford, P. P. Nelson, and M. J. S.
Roth, 49–75. Reston, VA: ASCE.
Lam, K., and W. K. Li. 1986. “Excerpts from Interview with
Professor Peter Lumb.” Hong Kong Statistical Society
Newsletter 9 (1): 2–5.
Lambe, T. W. 1973. “Predictions in Soil Engineering.”
Géotechnique 23 (2): 151–202.
Lazarte, C. A. 2011. Proposed Specifications for LRFD SoilNailing Design and Construction. NCHRP Report 701.
Washington, DC: Transportation Research Board
Lehane, B. M., J. K. Kim, P. Carotenuto, F. Nadim, S. Lacasse,
R. J. Jardine, and B. F. J. Van Dijk. 2017. “Characteristics of
Unified Databases for Driven Piles.” In 8th International
Conference of Offshore Site Investigation and
Geomechanics. Vol. 1, 162–191. London: Society for
Underwater Technology.
Li, Z., X. Wang, H. Wang, and R. Y. Liang. 2016. “Quantifying
Stratigraphic Uncertainties by Stochastic Simulation
Techniques Based on Markov Random Field.” Engineering
Geology 201: 106–122.
Lin, P. Y., R. Bathurst, and J. Y. Liu. 2017. “Statistical
Evaluation of the FHWA Simplified Method and
Modifications for Predicting Soil Nail Loads.” Journal of
Geotechnical and Geoenvironmental Engineering 143 (3):
04016107.
Liu, H. F., L. S. Tang, P. Y. Lin, and G. X. Mei. 2018. “Accuracy
Assessment of Default and Modified Federal Highway
Administration (FHWA) Simplified Models for
Estimation of Facing Tensile Forces of Soil Nail Walls.”
Canadian Geotechnical Journal 55 (8): 1104–1115.
Liu, S., H. Zou, G. Cai, B. V. Bheemasetti, A. J. Puppala, and J.
Lin. 2016. “Multivariate Correlation among Resilient
Modulus and Cone Penetration Test Parameters of
22
K.-K. PHOON
Cohesive Subgrade Soils.” Engineering Geology 209: 128–
142.
Long, M. 2001. “Database for Retaining Wall and Ground
Movements due to Deep Excavations.” Journal of
Geotechnical and Geoenvironmental Engineering 127 (3):
203–224.
Long, J. 2013. Improving Agreement between Static Method
and Dynamic Formula for Driven Cast-in-Place Piles in
Wisconsin.
Report
No.
0092-10-09,
Wisconsin
Department of Transportation
Long, J. 2016. Static Pile Load Tests on Driven Piles into
Intermediate-geo Materials. Report No. WHRP 0092-1208, Wisconsin Department of Transportation
Long, J., and A. Anderson. 2014. Improved Design for Driven
Piles Based on a Pile Load Test Program in Illinois: phase
2. Report No. FHWA-ICT-14-019, Illinois Department of
Transportation
Long, J., J. Hendrix, and D. Jaromin. 2009. Comparison of Five
Different Methods for Determining Pile Bearing Capacities.
Report No. WisDOT 0092-07-04, Wisconsin Department
of Transportation.
Lumb, P. 1966. “Variability of Natural Soils.” Canadian
Geotechnical Journal 3 (2): 74–97.
Marcos, M. C. M., and Y. J. Chen. 2013. “Evaluation of Lateral
Interpretation Criteria for Drilled Shaft Capacity in
Gravels.” Geotechnical and Geological Engineering 31 (5):
1411–1420.
Mariethoz, G., and J. Caers. 2015. Multiple-point Geostatistics Stochastic Modeling with Training Images. West Sussex:
John Wiley & Sons.
Marsland, A. 1953. “Model Experiments to Study the Influence
of Seepage on the Stability of a Sheeted Excavation in Sand.”
Géotechnique 3 (6): 223–241.
McVay, M., S. Wasman, L. Huang, and S. Crawford. 2016.
Load and Resistance Factor Design (LRFD) Resistance
Factors for Auger Cast in Place Piles. Tallahassee, FL:
Florida Department of Transportation.
Mesri, G., and N. Huvaj. 2007. “Shear Strength Mobilized in
Undrained Failure of Soft Clay and Silt Deposits.” In
Advances in Measurement and Modeling of Soil Behavior
(GSP 173), 1–22. Reston, VA: ASCE.
Miyata, Y., and R. J. Bathurst. 2012a. “Measured and Predicted
Loads in Steel Strip Reinforced c-Φ Soil Walls in Japan.”
Soils and Foundations 52 (1): 1–17.
Miyata, Y., and R. J. Bathurst. 2012b. “Analysis and Calibration
of Default Steel Strip Pullout Models Used in Japan.” Soils
and Foundations 52 (3): 481–497.
Miyata, Y., and R. J. Bathurst. 2015. “Reliability Analysis of
Geogrid Installation Damage Test Data in Japan.” Soils
and Foundations 55 (2): 393–403.
Miyata, Y., and R. J. Bathurst. 2019. “Statistical Assessment of
Load Model Accuracy for Steel Grid-Reinforced Soil Walls.”
Acta Geotechnica 14 (1): 57–70.
Miyata, Y., R. J. Bathurst, and T. M. Allen. 2014. “Reliability
Analysis of Geogrid Creep Data in Japan.” Soils and
Foundations 54 (4): 608–620.
Miyata, Y., R. J. Bathurst, and T. Konami. 2011. “Evaluation of
Two Anchor Plate Capacity Models for MAW Systems.”
Soils and Foundations 51 (5): 885–895.
Miyata, Y., Y. Yu, and R. J. Bathurst. 2018. “Calibration of SoilSteel Grid Pullout Models Using a Statistical Approach.”
Journal of Geotechnical and Geoenvironmental Engineering
144 (2): 04017106.
Moghaddam, R. B., P. W. Jayawickrama, W. D. Lawson, J. G.
Surles, and H. Seo. 2018. “Texas Cone Penetrometer
Foundation Design Method: Qualitative and Quantitative
Assessment.” DFI Journal – The Journal of the Deep
Foundations Institute 12 (2): 69–80.
Montoya-Noguera, S., T. Zhao, Y. Hu, Y. Wang, and K. K.
Phoon. 2019. “Simulation of Non-stationary NonGaussian Random Fields from Sparse Measurements
Using Bayesian Compressive Sampling and KarhunenLoève Expansion.” Structural Safety 79: 66–79.
Moormann, C. 2004. “Analysis of Wall and Ground
Movements due to Deep Excavations in Soft Soil Based on
a new Worldwide Database.” Soils and Foundations 44
(1): 87–98.
Moshfeghi, S., and A. Eslami. 2018. “Study on Pile Ultimate
Capacity Criteria and CPT-Based Direct Methods.”
International Journal of Geotechnical Engineering 12 (1):
28–39.
Moss, R. E. S., R. B. Seed, R. E. Kayen, J. P. Stewart, A. Der
Kiureghian, and K. O. Cetin. 2006. “CPT-based
Probabilistic and Deterministic Assessment of in Situ
Seismic Soil Liquefaction Potential.” Journal of
Geotechnical and Geoenvironmental Engineering 132 (8):
1032–1051.
Motamed, R., S. Elfass, and K. Stanton. 2016. LRFD Resistance
Factor Calibration for Axially Loaded Drilled Shafts in the
Las Vegas Valley. Report No. 515-13-803, Nevada
Department of Transportation.
Nanazawa, T., T. Kouno, G. Sakashita, and K. Oshiro. 2019.
“Development of Partial Factor Design Method on
Bearing Capacity of Pile Foundations for Japanese
Specifications for Highway Bridges.” Georisk: Assessment
and Management of Risk for Engineered Systems and
Geohazards 13 (3): 166–175.
National Research Council. 1995. Probabilistic Methods in
Geotechnical Engineering. Washington, DC: National
Academies Press.
Ng, C. W. W., T. L. Y. Yau, J. H. M. Li, and Tang. 2001. “New
Failure Load Criterion for Large Diameter Bored Piles in
Weathered Geomaterials.” Journal of Geotechnical and
Geoenvironmental Engineering 127 (6): 488–498.
NGI. 2019. “Peck Library.” Norwegian Geotechnical Institute.
Accessed October 3, 2019. https://www.ngi.no/eng/
Publications-and-library/NGI-s-Historical-Libraries/PeckLibrary.
Niazi, F. S. 2014. “Static Axial Pile Foundation Response Using
Seismic Piezocone Data.” Ph.D. thesis, Georgia Institute
ofTechnology
Orchant, C. J., F. H. Kulhawy, and C. H. Trautmann. 1988.
Reliability-based Foundation Design for Transmission Line
Structures: Critical Evaluation of In-situ Test Methods.
Report EL-5507(2). Palo Alto: Electric Power Research
Institute.
Ou, C. Y., and J. T. Liao. 1987. Geotechnical Engineering
Research Report. GT96008. Taipei: National Taiwan
University of Science and Technology.
Paikowsky, S. G., B. Birgisson, M. McVay, T. Nguyen, C. Kuo,
G. B. Baecher, B. Ayyub, et al. 2004. Load and Resistance
Factors Design for Deep Foundations. NCHRP Report 507.
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
Washington, DC: Transportation Research Board of the
National Academies.
Paikowsky, S., M. Canniff, K. Lesny, A. Kisse, S. Amatya, and
R. Muganga. 2010. “LRFD Design and Construction of
Shallow Foundations for Highway Bridge Structures.”
NCHRP Report 651. Washington, DC: Transportation
Research Board.
Peck, R. B. 1980. “Where has All the Judgment Gone?”
Canadian Geotechnical Journal 17 (4): 584–590.
Phoon, K. K. 2017. “Role of Reliability Calculations in
Geotechnical Design.” Georisk 11 (1): 4–21.
Phoon, K. K. 2018. “Editorial for Special Collection on
Probabilistic Site Characterization.” ASCE-ASME Journal
of Risk and Uncertainty in Engineering Systems, Part A:
Civil Engineering 4 (4): 02018002.
Phoon, K. K. 2020. “The Goldilocks Dilemma - too Little or
too Much Data.” GeoStrata 24 (1), in press.
Phoon, K. K., and J. Ching. 2017. “Better Correlations for
Geotechnical Engineering.” A Decade of Geotechnical
Advances, Geotechnical Society of Singapore (GeoSS),
73–102.
Phoon, K. K., J. Ching, and Y. Wang. 2019. “Managing Risk in
Geotechnical Engineering – from Data to Digitalization.”
Proceedings,
7th
International
Symposium
on
Geotechnical Safety and Risk (ISGSR 2019), Taipei,
Taiwan, in press.
Phoon, K. K., and F. H. Kulhawy. 1999a. “Characterization of
Geotechnical Variability.” Canadian Geotechnical Journal
36 (4): 612–624.
Phoon, K. K., and F. H. Kulhawy. 1999b. “Evaluation of
Geotechnical Property Variability.” Canadian Geotechnical
Journal 36 (4): 625–639.
Phoon, K. K., and F. H. Kulhawy. 2008. “Serviceability Limit
State Reliability-based Design.” In Reliability-based Design
in Geotechnical Engineering: Computations and
Applications, 344–383. New York: Taylor & Francis.
Phoon, K. K., F. H. Kulhawy, and M. D. Grigoriu. 2003.
“Multiple Resistance Factor Design for Shallow
Transmission Line Structure Foundations.” Journal of
Geotechnical and Geoenvironmental Engineering 129 (9):
807–818.
Phoon, K. K., S. L. Liu, and Y. K. Chow. 2009.
“Characterization of Model Uncertainties for Cantilever
Walls in Sand.” Journal of GeoEngineering 4 (3): 75–85.
Phoon, K. K., W. A. Prakoso, Y. Wang, and J. Ching. 2016.
“Uncertainty Representation of Geotechnical Design
Parameters.” Chap. 3 in Reliability of Geotechnical
Structures in ISO2394, 49–87. Balkema: CRC Press.
Phoon, K. K., J. V. Retief, J. Ching, M. Dithinde, T.
Schweckendiek, Y. Wang, and L. M. Zhang. 2016. “Some
Observations on ISO2394:2015 Annex D (Reliability of
Geotechnical Structures).” Structural Safety 62: 24–33.
Phoon, K. K., and C. Tang. 2019. “Characterization
of Geotechnical Model Uncertainty.” Georisk 13 (2):
101–130.
Qi, X. H., D. Q. Li, K. K. Phoon, Z. Cao, and X. S. Tang. 2016.
“Simulation of Geologic Uncertainty Using Coupled
Markov Chain.” Engineering Geology 207: 129–140.
Rauser, J., and C. Tsai. 2016. “Beneficial use of the Louisiana
Foundation Load Test Database.” In Transportation
Research Board 95th Annual Meeting, 1–17. Washington,
DC: Transportation Research Board.
23
Reddy, S., and A. Stuedlein. 2017. “Ultimate Limit State
Reliability-Based Design of Augered Cast-in-Place Piles
Considering
Lower-Bound
Capacities.”
Canadian
Geotechnical Journal 54 (12): 1693–1703.
Roling, M., S. Sritharan, and M. Suleiman. 2011. Development
of LRFD Procedures for Bridge Pile Foundations in Iowa.
Vol. 1: An electronic Database for Pile Load Tests (PILOT).
Report No. IHRB Project TR-573, Iowa Department of
Transportation
Samtani, N. C., and T. C. Allen. 2018. “Implementation Report
– Expanded Database for Service Limit State Calibration of
Immediate Settlement of Bridge Foundation on Soil. Report
No. FHWA-HIF-18-008. Washington, DC: Federal
Highway Administration (FHWA).
Shahin, M. A., M. B. Jaksa, and H. R. Maier. 2001. “Artificial
Neural
Network
Applications
in
Geotechnical
Engineering.” Australian Geomechanics 36 (1): 49–62.
Shields, M. D., and H. Kim. 2017. “Simulation of Higherorder Stochastic Processes by Spectral Representation.”
Probabilistic Engineering Mechanics 47: 1–15.
Smith, T., A. Banas, M. Gummer, and J. Jin. 2011.
Recalibration of the GRLWEAP LRFD Resistance Factor
for Oregon DOT. Report No. FHWA-OR-RD-11-08,
Oregon Department of Transportation
Spry, M. J., F. H. Kulhawy, and M. D. Grigoriu. 1988.
Reliability-based Foundation Design for Transmission Line
Structures: Geotechnical Site Characterization Strategy.
Report EL-5507(1). Palo Alto: Electric Power Research
Institute.
Stark, T., J. Long, A. Baghdady, and A. Osouli. 2017. Modified
Standard Penetration Test-based Drilled Shaft Design
Method for Weak Rocks (phase 2 study). Report No.
FHWA-ICT-17-018, Illinois Department of Transportation
State of the Nation Report. 2017. “Digital Transformation.”
Institution of Civil Engineers. https://www.ice.org.uk/
news-and-insight/policy/state-of-the-nation-2017-digitaltransformation.
Stuyts, B., D. Cathie, and T. Powell. 2016. “Model Uncertainty
in Uplift Resistance Calculations for Sandy Backfills.”
Canadian Geotechnical Journal 53 (11): 1831–1840.
Tang, W. H. 1984. “Principles of Probabilistic Characterization
of Soil Properties.” In Probabilistic Characterization of Soil
Properties: Bridge Between Theory and Practice, edited by
D. S. Bowles and H.-Y. Ko, 74–89. Atlanta, GA: ASCE.
Tang, C., and K. K. Phoon. 2016. “Model Uncertainty of
Cylindrical Shear Method for Calculating the Uplift
Capacity of Helical Anchors in Clay.” Engineering Geology
207: 14–23.
Tang, C., and K. K. Phoon. 2018a. “Evaluation of Model
Uncertainties in Reliability-Based Design of Steel H-Piles
in Axial Compression.” Canadian Geotechnical Journal 55
(11): 1513–1532.
Tang, C., and K. K. Phoon. 2018b. “Statistics of Model Factors
in Reliability-Based Design of Axially Loaded Driven Piles
in Sand.” Canadian Geotechnical Journal 55 (11): 1592–
1610.
Tang, C., and K. K. Phoon. 2018c. “Statistics of Model Factors
and Consideration in Reliability-Based Design of Axially
Loaded Helical Piles.” Journal of Geotechnical and
Geoenvironmental Engineering 144 (8): 04018050.
Tang, C., and K. K. Phoon. 2018. “Statistics of Model Factors
and Consideration in Reliability-Based Design of Axially
24
K.-K. PHOON
Loaded Helical Piles.” Journal of Geotechnical and
Geoenvironmental Engineering 144 (8): 04018050.
Tang, C., and K. K. Phoon. 2019a. “Evaluation of Stress
Dependent Methods for the Punch-Through Capacity of
Foundations in Clay with Sand.” ASCE-ASME Journal of
Risk and Uncertainty in Engineering Systems, Part A: Civil
Engineering 5 (3): 04019008.
Tang, C., and K. K. Phoon. 2019b. “Characterization of Model
Uncertainty in Predicting Axial Resistance of Piles Driven
In to Clay.” Canadian Geotechnical Journal 56 (8): 1098–
1118.
Tang, C., and K. K. Phoon. 2019c. “Statistical Evaluation of
Model Factors in Reliability Calibration of High
Displacement Helical Piles Under Axial Loading.”
Canadian Geotechnical Journal. doi:10.1139/cgj-2018-0754.
Tang, C., K. K. Phoon, and Y. J. Chen. 2019. “Statistical
Analyses of Model Factors in Reliability-Based Limit State
Design of Drilled Shafts under Axial Loading.” Journal of
Geotechnical and Geoenvironmental Engineering 145 (9):
04019042.
Tang, C., K. K. Phoon, D. Q. Li, and S. O. Akbas. 2019.
“Expanded Database Assessment of Design Methods for
Spread Foundations Under Axial Uplift and Compression.”
under preparation.
TC304. 2019. “304dB – TC304 Databases.” TC304
(Engineering Practice of Risk Assessment and
Management), International Society for Soil Mechanics
and Geotechnical Engineering. Accessed October 3, 2019.
http://140.112.12.21/issmge/tc304.htm.
Terzaghi, K., and R. B. Peck. 1967. Soil Mechanics in
Engineering Practice. New York: John Wiley & Sons Inc.
Tian, M., D. Q. Li, Z. Cao, K. K. Phoon, and Y. Wang. 2016.
“Bayesian Identification of Random Field Model Using
Indirect Test Data.” Engineering Geology 210: 197–211.
Travis, Q., M. Schmeeckle, and D. Sebert. 2011. “Meta-analysis
of 301 Slope Failure Calculations. I: Database Description.”
Journal of Geotechnical and Geoenvironmental Engineering
137 (5): 453–470.
Vanmarcke, E. H. 1977. “Probabilistic Modeling of Soil
Profiles.” Journal of the Geotechnical Engineering Division
103 (11): 1227–1246.
Vanmarcke, E. H., and G. A. Fenton. 2003. Probabilistic Site
Characterization
at
the
National
Geotechnical
Experimentation Sites (GSP 121). Reston, VA: ASCE.
Vick, S. G. 2002. Degrees of Belief: Subjective Probability and
Engineering Judgment. Reston, VA: ASCE.
Wang, Y., Y. Hu, and T. Zhao. 2019. “CPT-based Subsurface
Soil Classification and Zonation in a 2D Vertical Cross-section Using Bayesian Compressive Sampling.” Canadian
Geotechnical Journal. doi:10.1139/cgj-2019-0131.
Wang, Y., K. Huang, and Z. Cao. 2013. “Probabilistic
Identification of Underground Soil Stratification Using
Cone Penetration Tests.” Canadian Geotechnical Journal
50 (7): 766–776.
Wang, X., Z. Li, H. Wang, Q. Rong, and R. Y. Liang. 2016.
“Probabilistic Analysis of Shield-driven Tunnel in
Multiple Strata Considering Stratigraphic Uncertainty.”
Structural Safety 62: 88–100.
Wang, X., H. Wang, R. Y. Liang, and Y. Liu. 2019. “A Semisupervised Clustering-based Approach for Stratification
Identification Using Borehole and Cone Penetration Test
Data.” Engineering Geology 248: 102–116.
Wang, X., H. Wang, R. Y. Liang, H. Zhu, and H. Di. 2018. “A
Hidden Markov Random Field Model Based Approach for
Probabilistic Site Characterization Using Multiple Cone
Penetration Test Data.” Structural Safety 70: 128–138.
Wang, H., X. Wang, J. F. Wellmann, and R. Y. Liang. 2018.
“Bayesian Stochastic Soil Modeling Framework Using
Gaussian Markov Random Fields.” ASCE-ASME Journal
of Risk and Uncertainty in Engineering Systems, Part A:
Civil Engineering 4 (2): 04018014.
Wang, H., X. Wang, J. F. Wellmann, and R. Y. Liang. 2019. “A
Bayesian Unsupervised Learning Approach for Identifying
Soil Stratification Using Cone Penetration Data.”
Canadian Geotechnical Journal 56 (8): 1184–1205.
Wang, H., J. F. Wellmann, Z. Li, X. Wang, and R. Y. Liang.
2017. “A Segmentation Approach for Stochastic
Geological Modeling Using Hidden Markov Random
Fields.” Mathematical Geosciences 49 (2): 145–177.
Wang, J., Z. Xu, and W. Wang. 2010. “Wall and Ground
Movements due to Deep Excavations in Shanghai Soft
Soils.” Journal of Geotechnical and Geoenvironmental
Engineering 136 (7): 985–994.
Wang, Y., and T. Zhao. 2016. “Interpretation of Soil Property
Profile from Limited Measurement Data: A Compressive
Sampling Perspective.” Canadian Geotechnical Journal 53
(9): 1547–1559.
Wang, Y., and T. Zhao. 2017. “Statistical Interpretation of Soil
Property Profiles from Sparse Data Using Bayesian
Compressive Sampling.” Géotechnique 67 (6): 523–536.
Wang, Y., T. Zhao, Y. Hu, and K. K. Phoon. 2019. “Simulation
of Random Fields with Trend from Sparse Measurements
without De-trending.” Journal of Engineering Mechanics
145 (2): 04018130.
Wang, Y., T. Zhao, and K. K. Phoon. 2018. “Direct Simulation
of Random Field Samples from Sparsely Measured
Geotechnical Data with Consideration of Uncertainty in
Interpretation.” Canadian Geotechnical Journal 55 (6):
862–880.
White, D., C. Cheuk, and M. Bolton. 2008. “The Uplift
Resistance of Pipes and Plate Anchors Buried in Sand.”
Géotechnique 58 (10): 771–779.
Wood, T., P. Jayawickrama, J. Surles, and W. Lawson. 2012a.
“Pullout Resistance of MSE Reinforcements in Backfills
Typically Used in Texas: Vol. 2, Test Reports for MSE
Reinforcements in Type B (sandy) Backfill.” Research
Report:
FHWA/TX-13/0-6493-3,
Vol.
2,
Texas
Department of Transportation
Wood, T., P. Jayawickrama, J. Surles, and W. Lawson. 2012b.
“Pullout Resistance of MSE Reinforcements in Backfills
Typically Used in Texas: Vol. 3, Test Reports for MSE
Reinforcements in Type A (gravelly) Backfill.” Research
Report:
FHWA/TX-13/0-6493-3,
Vol.
3,
Texas
Department of Transportation
Wu, T. H., M. A. Abdel-Latif, M. A. Nuhfer, and B. B.
Curry. 1996. Uncertainty in the Geologic Environment:
from Theory to Practice (GSP 58), 76–90. Reston, VA:
ASCE.
Wu, S. H., J. Y. Ching, and C. Y. Ou. 2013. “Predicting Wall
Displacements for Excavations with Cross Walls in Soft
Clay.” Journal of Geotechnical and Geoenvironmental
Engineering 139 (6): 914–924.
Wu, S. H., C. Y. Ou, and J. Y. Ching. 2014. “Calibration of
Model Uncertainties in Base Heave Stability for Wide
GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS
Excavations in Clay.” Soils and Foundations 54 (6): 1159–
1174.
Wu, T. H., W. H. Tang, D. A. Sangrey, and G. B. Baecher.
1989. “Reliability of Offshore Foundations-state of the Art.”
Journal of Geotechnical Engineering 115 (2): 157–178.
Xiao, T., D. Q. Li, Z. Cao, and L. M. Zhang. 2018. “CPTbased Probabilistic Characterization of Three-dimensional
Spatial Variability Using MLE.” Journal of Geotechnical and
Geoenvironmental Engineering 144 (5): 04018023.
Yang, Z., R. Jardine, W. Guo, and F. Chow. 2016. A
Comprehensive Database of Tests on Axially Loaded Piles
Driven in Sand. London: Academic Press.
Yuan, J., P. Y. Lin, R. Huang, and Y. Que. 2019. “Statistical
Evaluation and Calibration of Two Methods for
Predicting Nail Loads of Soil Nail Walls in China.”
Computers and Geotechnics 108: 269–279.
Yuen, K. V., and G. A. Ortiz. 2016. “Bayesian Nonparametric
General Regression.” International Journal for Uncertainty
Quantification 6 (3): 195–213.
25
Yuen, K. V., and G. A. Ortiz. 2018. “Multi-resolution Bayesian
Nonparametric General Regression for Structural Model
Updating.” Structural Control and Health Monitoring 25
(2): e2077.
Yuen, K. V., G. A. Ortiz, and K. Huang. 2016. “Novel
Nonparametric Modeling of Seismic Attenuation and
Directivity Relationship.” Computer Methods in Applied
Mechanics and Engineering 311: 537–555.
Zhang, L. M., L. M. P. Shek, H. W. Pang, and C. F. Pang. 2006.
“Knowledge-based Design and Construction of Driven
Piles.” Proceedings of the Institution of Civil Engineers Geotechnical Engineering 159 (3): 177–185.
Zhang, D. M., Y. Zhou, K. K. Phoon, and H. W. Huang. 2019.
“Multivariate Probability Distribution of Shanghai Clay
Properties.” Engineering Geology, under review.
Zhao, T., and Y. Wang. 2018. “Simulation of Cross-correlated
Random Field Samples from Sparse Measurements Using
Bayesian Compressive Sensing.” Mechanical Systems and
Signal Processing 112: 384–400.