Selection of Characteristic Values for

Rock and Soil Properties using

Bayesian Statistics and Prior
Knowledge: Theory, Software and
Application Examples

Yu Wang
City University of Hong Kong

Innotech Forum on Geotechnology, Hong Kong, 17 April 2018

 Acknowledgement
• Joint TC205/TC304 Working Group on
“Discussion of statistical/reliability methods for
Eurocodes”
• Discussion group on “Selection of characteristic
values for rock and soil properties using Bayesian
statistics and prior knowledge” led by Yu Wang
• Discussers (alphabetical order):
– Marcos Arroyo (Spain), Zijun Cao (China), Jianye Ching
(Taiwan), Tim Länsivaara (Finland), Trevor Orr (Ireland), Kok-
Kwang Phoon (Singapore), Hansruedi Schneider (Switzerland),
Brian Simpson (UK)
 Outline
• Background and Motivation
• Bayesian Methods
• Software and Android App
• Illustrative Examples
• Concluding Remarks
 Background
• Soil or rock properties are necessary input
to geotechnical designs and analyses
• During site investigation, soil or rock
properties are often measured sparsely due
to time, resource, or technical constraints.
• How to select characteristic values for soil
or rock properties from limited data?
Example: what Cu value to use?
Example: what SPT profile to use?

• Bond and Harris

(2008) asked about
one hundred
engineers to select
the SPT profile
• The maximum
profile was about 3
to 5 times greater
than the minimum
one
 Motivation
• To select characteristic values for soil or rock
properties in a rational and consistent manner
using Bayesian statistics and prior knowledge
• Eurocode 7, Clause 2.4.5.2(10):
“If statistical methods are employed in the
selection of characteristic values for ground
properties, such methods should differentiate
between local and regional sampling and should
allow the use of a priori knowledge of comparable
ground properties.”
• Bayesian method combines information from
different sources in a rational manner
 Geotechnical Site Investigation
PROCEDURE INFORMATION UNCERTAINTY
I: Desk-study
Information updating process

Prior knowledge Uncertainties in

II: Site reconnaissance (e.g., geological maps, existing information
geotechnical reports, engineering
experience and judgment, etc.)
III: In-situ investigation
Site observation data Inherent variability
Measurement error
IV: Laboratory testing (e.g., data from test boring, in-situ
Statistical uncertainty
testing and/or laboratory testing)
V: Interpretation of site
observation data Transformation model Transformation
(e.g., empirical regression) uncertainty

VI: Inferring geotechnical

properties and How to
Updated knowledge combine them
underground strata Systematically?

Multi-sources Various Limited site-

CHALLENGES information uncertainties specific data
 Bayesian Framework for
Geotechnical Site Investigation
• Bayesian approach combines systematically information from
different sources for uncertainty quantification
Prior Prior Likelihood Posterior Posterior
knowledge distribution function distribution knowledge

Likelihood model Site observation data

ML XM

Probabilistic
Transformation
model MP of
model MT:
the design
property X D X D = fT ( X M ;ε T )

Inherent Transformation Statistical uncertainty and

variability uncertainty measurement errors
Bayesian Equivalent Sample Method
• Likelihood Function
Probability model MP of XD with model parameters Θp
(e.g., random variable, random field)
Transformation model, XD = fT(XM; εT)
• Prior Distribution
Non-informative – e.g., joint uniform distribution
Informative – subjective probability assessment framework (Cao et al., 2016)
• Posterior Distribution
P (Θ P Data , Prior ) = K × P (Data Θ P , Prior ) × P (Θ P | Prior )
P (X D Data , Prior ) = ∫ P( X D ΘP ) P(ΘP Data , Prior )dΘP
Markov Chain Monte Carlo simulation
Equivalent samples of XD
Example: Estimating Eu of clay
from SPT-N
• Limited site-specific SPT-N data

• Formulation of likelihood function

• Formulation of prior distribution
• Posterior distribution of Eu
• Bayesian equivalent samples of Eu from
Markov Chain Monte Carlo simulation
 Likelihood Function
A statistically homogenous clay layer
Undrained Young’s modulus
Eu~ LogNormal (µ, σ)
µ = Mean value of Eu
σ = Standard deviation of Eu
lnEu~ Normal (µN, σN)
µN = Mean value of lnEu
σN = Standard deviation of lnEu
lnEu = fN (µN, σN) = f(µ, σ)
Depth (D)
 Likelihood Function
In a log-log scale
ln(Eu/pa)=0.63lnNSPT+2.96+εT
Eu Eu values are obtained
pa from pressuremeter tests
pa=0.1MPa
pa: atmospheric pressure
εT: transformation uncertainty
Normal random variable
SPT N Value (NSPT) (Phoon and Kulhawy 1990)

(Ohya et al. 1982, Kulhawy and Mayne 1990)

 Likelihood Function
Transformation
lnNSPT = fT(lnEu) + ε uncertainty
lnNSPT ∼ Normal (µSPT,σSPT)
lnEu = f(µ, σ) µSPT ~ fm(µ, σ); σSPT ~fs(µ,σ)
Normal variable

P(lnNSPT|µ,σ) =
Normal Probability Density Function (PDF)
DATA={lnNSPT,i, i=1,…, ns}
ns

Likelihood function P(DATA|µ,σ) = P(lnNSPT,i|µ, σ)

i=1
Measurement errors Inherent
Statistical uncertainty variability
 Prior Distribution
• When only respective typical ranges [µmin, µmax]
and [σmin, σmax] of µ and σ are available

1 1 For µ∈ [µmin, µmax]

P(µ,σ) = µmax- µmin × σmax- σmin and σ∈ [σmin, σmax]
0 Others
 Posterior Distribution
• Bayes’ Theorem
P(µ,σ|DATA) = KP(DATA|µ,σ)P(µ,σ)
Posterior Likelihood Prior distribution
distribution function

where K is a normalizing constant

• Total Probability Theorem
P(Eu|DATA, PRIOR)
= ∫ P(Eu|µ,σ) P(µ,σ|DATA,PRIOR)dµdσ
=

LogNormal PDF P(µ,σ|DATA)=

 MCMCS
• Markov Chain Monte Carlo Simulation (MCMCS)
is a numerical process that simulates a sequence
of samples of a random variable as a Markov
Chain
Limiting stationary PDF PDF of the target
of the Markov Chain = random variable
(e.g., P(Eu|DATA, PRIOR) )

• A feasible way to generate samples from an

arbitrary PDF (e.g., P(Eu|DATA, PRIOR) )
 BEST EXCEL Add-In
(Bayesian Equivalent Sample Toolkit)
https://sites.google.com/sit
e/yuwangcityu/best/1
• 12 Built-in model • User-defined model
 Application Example
A clay site of US National Geotechnical
Experimentation Sites at Texas A&M University
SPT N -value Eu (MPa)

0
0 5 10 15 20 25 30 35
0
0 5 10 15 20 25 30 35 40
DATA = 5 SPT data
Stiff Clay Stiff Clay
1 1 Uniform PRIOR with
µ ∈ [5MPa, 15MPa]
σ ∈ [0.5MPa, 13.5MPa]
2 2
Depth (m)
Depth (m)

3 3
(Phoon and Kulhawy
1999a and 1999b)
4 4

42 Pressuremeter
5 5
test data
6
(Briaud 2000) 6
(Briaud 2000)
Application Example

Mean:
11.5
Stdev:
6.0
90%
 Application Example

5 SPT data &

Prior knowledge

50%
42 Pressuremeter
test data

Mean - stdev
5.5MPa 11.5MPa
BEST App in Android Phone
 Concluding Remarks
• Development of practical Bayesian statistical methods
and user-friendly tools for selection of characteristic
values for soil or rock properties
Bayesian equivalent sample method
Quantification of prior knowledge
User-friendly software (BEST) in EXCEL and
Android App
• BEST is applicable to direct or indirect measurements
• Random field modeling of spatial variability is not
covered
 References
Wang, Y., Cao, Z., and Li, D. (2016). “Bayesian perspective on geotechnical variability and site
characterization.” Engineering Geology, 203, 117-125.
Cao, Z., Wang, Y., and Li, D. (2016). “Quantification of prior knowledge in geotechnical site
characterization.” Engineering Geology, 203, 107-116.
Wang, Y., Akeju, O. V., and Cao, Z. (2016). "Bayesian Equivalent Sample Toolkit (BEST): an Excel VBA
program for probabilistic characterisation of geotechnical properties from limited observation
data." Georisk, 10(4), 251-268.
Cao, Z., Wang, Y., and Li, D. (2016). “Site-specific characterization of soil properties using multiple
measurements from different test procedures at different locations – A Bayesian sequential
updating approach.” Engineering Geology, 211, 150-161.
Wang, Y. and Aladejare, A. E. (2016). "Bayesian characterization of correlation between uniaxial
compressive strength and Young's modulus of rock." International Journal of Rock Mechanics &
Mining Sciences, 85, 10-19.
Wang, Y. and Aladejare, A. E. (2016). "Evaluating variability and uncertainty of Geological Strength
Index at a specific site." Rock Mechanics and Rock Engineering, 49(9), 3559–3573.
Wang, Y., Zhao, T., and Cao, Z. (2015). “Site-specific probability distribution of geotechnical
properties.” Computers and Geotechnics, 70, 159-168.
Wang, Y. and Aladejare, A. E. (2015). “Selection of site-specific regression model for characterization of
uniaxial compressive strength of rock.” International Journal of Rock Mechanics & Mining
Sciences, 75, 73-81.
Cao, Z. and Wang, Y. (2014). “Bayesian model comparison and characterization of undrained shear
strength.” Journal of Geotechnical & Geoenvironmental Engineering, 140(6), 1-9.
Wang, Y. and Cao, Z. (2013). “Probabilistic characterization of Young's modulus of soil using equivalent
samples.” Engineering Geology, 159, 106-118. (Highly Cited Research Award)
“Probability theory is nothing
but common sense reduced to
calculation.”
Pierre-Simon Laplace
(1749-1827)
Thank you!