INTERNATIONAL SOCIETY FOR

SOIL MECHANICS AND

GEOTECHNICAL ENGINEERING

This paper was downloaded from the Online Library of

the International Society for Soil Mechanics and
Geotechnical Engineering (ISSMGE). The library is
available here:

https://www.issmge.org/publications/online-library

This is an open-access database that archives thousands

of papers published under the Auspices of the ISSMGE and
maintained by the Innovation and Development
Committee of ISSMGE.
X IIIIC S M F E , 1994. N ew Delhi. India / XIII C IM S T F , 1994. New Delhi, Inde

ASSESSMENT OF CHARACTERISTIC VALUES OF SOIL PARAMETERS FOR

DESIGN
DETERMINATION DE LA RESISTANCE CARACTERISTIQUE DES SOLS POUR
PROJECTS

Hans Denver1 N. Krebs Ovesen2

'Sen. Res. Engineer, 2M anaging Director

Danish Geotechnical Institute, Lyngby, Denmark
SYNOPSIS: The geotechnical design process involves a number of elements: Soil parameters, loads, calculation methods and safety factors constitute the
most important ones. The selection of soil parameters is today often the weak link in the design process. Engineering judgment, company traditions and
personal experience enter into this selection process and thereby contribute to an unnecessary large amount of the inherited uncertainty in the geotechnical
design process. Making use of Bayesian statistics the paper proposes a method Tor selection of soil parameters in a systematic and reproducible way. The
method is described and it is demonstrated by means of an example on how to use apriori knowledge in combination with the results of field and laboratory
tests to assess (he characteristic value of soil properties corresponding to a specific fractile in the statistical density function of the parameter. In addition the
paper treats the problems of identifying the soil influence zone and the soil material model for the geotechnical design property.

INTRODUCTION In geotechnical design it is general practice to distinguish between two main

classes of limit states:
The geotechnical design process consists normally of two main phases.
During the first phase a soil investigation involving soundings, borings and Ultimate limit states at which:
field and laboratory testing is carried out to determine the soil and ground - either a failure mechanism is formed in the ground
water profile including the soil parameters. During the second phase physical, - or a failure mechanism is formed in the superstructure or severe
numerical and mathematical modelling is employed to predict the behaviour structural damage occurs due to movements in the ground
of the structure and safety requirements are implemented.
Serviceability limit states at which deformations in the ground will
Both phases are important as they contribute to the final design result. In the cause loss of serviceability in the superstructure.
geotechnical literature there is a tendency to focus mainly on field and
laboratory testing and bn physical, numerical and mathematical modelling. The limit stale design procedure is often used in connection with partial
Rarely, however, the geotechnical literature focus on the very important factors of safety. According to this format the prescribed characteristic loads
question of how to select soil parameters on basis of the results of field and are multiplied by certain partial factors to obtain design values of the loads,
laboratory tests for use in the geotechnical design process. The interrelation and the characteristic material properties are divided by other partial factors
between the selection of soil parameters and the safety level introduced in the to obtain the design material properties. The design criterion then simply is
design is often disregarded too. to design for equilibrium in the limit state using design values of loads as
well as material properties.
In the opinion of the authors the selection of soil parameters for geotechnical
design purposes is consequently often the week link in the design process. A The limit slate design concept making use of partial factors is today’s
large amount of personal skills, company traditions and engineering judgment preferred format for codes of practice. The new European code of practice
is involved in this process but the choice of soil parameters is indeed rather for geotechnical design, Eurocode 7 (1993), represents an example of such
arbitrary. In addition experience where this selection process has failed is a code. In the code which is part of a set of harmonized codes for structural
scarce and not systematically collected. Consequently, this process of design values are given for loads, partial factors, etc. However, one of the
selecting soil parameters contributes to an unnecessary large amount of the most difficult problems encountered in this code - and this applies to
inherited uncertainty in the geotechnical design process. geotechnical design in general - is how to define the characteristic soil
parameters on basis of apriori knowledge, results of field and laboratory
LIM IT STATE DESIGN testing, etc. The present paper will illustrate some of the problems involved
in the selection of soil parameters for geotechnical design and propose a
In order to ensure an adequate, technical quality of a geotechnical structure method for assessment of characteristic values of soil parameters.
it is required that the structure as a whole and its various parts fulfils certain
fundamental requirements of stability, stiffness, etc. during construction and SOIL INFLUENCE ZONE
throughout the design life. The fundamental requirements are normally
expressed in specific terms as performance criteria. The soil conditions may vary substantially on a construction site. It is
consequently necessary to identify the specific volume of the soil which
Whenever a geotechnical structure fails to satisfy one of its performance governs the behaviour of the structure in the limit stale under consideration.
criteria it is said to have reached a limit state. According to the concept of In the following this volume is denoted the soil influence zone.
limit state design each limit state is considered separately in the design and
its occurrence is either eliminated or shown to be sufficiently improbable. The extent of the soil influence zone depends on

437
• The structure, its dimensions and the limit state considered CHARACTERISTIC VALUE
• The loads and their combination
• The soil conditions In Eurocode 7 the definition of the characteristic value of soil parameters is
supported by specifying a fractile of the statistical distribution of the
It is evident that the type and the size of the structure have a most significant parameter as follows:
effect on the extent of the soil influence zone. Having identified the type and
size of the structure a suitable calculation model is selected corresponding to .... the characteristic value should be derived such that the calculated
the limit stale considered and the soil volume involved in this state is probability o f a worse value governing the occurrence o f a limit state is
identified. This soil volume amounts to the soil influence zone. not greater than 5%.

The soil influence zone furthermore depends on the loads involved. For An argument for keeping a significant part of the reliability margin within the
instance the shape of the failure surface below a strip footing depends on the procedure for selection of characteristic soil parameters is that the parameters
ratio between the horizontal and the vertical components of the load. can be obtained by various testing methods with different inherent uncer­
tainties. Furthermore, the number of tests carried out for the single project
The variation of the soil conditions with depth and place may also influence may vary substantially. If a mean value definition was adopted an extensive
the extent of the soil influence zone. E.g. a failure surface in sand is list of partial safety factors would be needed in order to avoid too conserva­
dependent on the value of the angle of shearing resistance and a weak soil tive designs for a majority of geotechnical structures.
layer located beneath a firm layer may extend the influence zone to a greater
depth. It is within the context of the definition given above for the characteristic
value to perform a rational reduction of data from the soil model. It should
It should be emphasized, however, that distinction shall be made between the be recalled that soil model I can be used directly in connection with such
soil influence zone and the extent of the volume of soil which need to be advanced numerical calculation models as finite elements or where the spatial
investigated by soundings and borings as part of an appropriate soil data information is weighted proportionally with spatial dissipation of energy
investigation prior to the design. for an approximate solution for homogeneous soil.

In many cases a soil parameter profile can be used directly as input for a
SOIL MATERIAL MODEL calculation model. It is in this connection appropriate to define a characteris­
tic profile in a similar way as the characteristic value. Here it should be
Before discussing the principles for selecting a characteristic value for a soil recalled that weak layers in some connections may be compensated for by
parameter a soil material model valid for the soil influence zone shall be adjacent stronger layers. This should be kept in mind if different portions of
established. In this context the most important step is to identify the different a characteristic profile is used for design of structures of different types and
soil strata which constitute the soil influence zone. This is due to the fact that sizes, e.g. if column footings of different sizes are designed by use of the
completely different classes of soil parameters are assigned to different soil same characteristic profile.
types.
UPDATING PROCEDURE
A soil material model consists of a description of the variation of one or
more parameters within the soil influence zone both in terms of mean values It is assumed that the soil type is known prior to the time when test results
and variances. Otherwise, it will not be possible to derive a characteristic become available allowing an experienced geotechnical engineer to estimate
value defined by a specified tractile as demonstrated in the following section. the variation of a certain parameter 0 in terms of a mean value and a
standard deviation. As negative values are prohibitive for most soil
An advanced soil material model of this kind (denoted type I) can be derived parameters a logarithmic normal distribution of 0 for the site is usually
if a large number of measured values of the soil parameter is available. Such assumed.
a model consists of a complete description of the spatial parameter variation -
i.e. 0 = 0 (x, y, z). Reference may be made to the concept of stochastic The concept of updating knowledge has been treated extensively in the
interpolation where the parameter is assigned the values determined in the statistical literature. This paper is based on a procedure described by
measuring points. In other points the parameter varies in a systematic pattern Ditlevsen & Madsen (1990, 1993), where the probability density function of
converging to the known values in the measuring points. An important the pair (M, I ) is consecutively updated. M is the mean value Qi) interpreted
feature is that this method allows for an evaluation of the reliability of the as a stochastic variable and I 2 is the similar interpretation of the variance
parameter for each point in statistical terms. (o2) of the logarithm of the normalized parameter. The theory behind the
procedure will not be presented here, but a set of formulas is given which
A simpler type of soil material model (type II) is based on the assumption will allow a determination of characteristic values.
that the spatial variation of the soil parameter is solely a function of the
depth - i.e. 0 = 0 (z). This assumption is convenient in connection with The knowledge of the soil type is assumed to form the basis for providing
horizontal soil layers or for soil properties dependent on the vertical stress estimates for E(ML) and V(Ml ) where E is the mean value and V is the
level. Vertical profiling resulting from borehole tests or penetration tests ate coefficient of variation defined as V = D/E where D is the standard deviation.
excellent means to obtain data for a soil model of this kind.

Another type o f soil material model (type III) includes no information on the
spatial variation of the parameters. It simply consists of the local bulk values
Table I: Soil models
of the parameter 0 expressed in statistical terms as a probability density
function or described as a mean value and a variance. Soil Para­ N eed for
model meter
It should be noted that all models include a measure of reliability of the soil type varia­ Apriori Number of
parameters. In a case where only few measurements are available apriori tions knowledge measurements
knowledge of parameter variations may be extremely important for estimating
the variability. A consistent implementation of such apriori knowledge is to I 0 (x,y,z) - - 100
use the concepts of Bayesian statistics in a procedure where the apriori
II 0 (z) + - 10
information can be updated when more information as a result of more soil
testing becomes available. In table 1 the need for apriori information is in 0 +++ - 1,2,3
shown for the tluee soil material models mentioned.

438
Figure 1. A set of logarithmic normal distributions

The suffix L refers to the logarithmic normal probability function. Further­

more, an estimate of the mean value E fV J of the coefficient of variation
valid for the influence zone shall also be made.

IKMi) can be estimated as the mean of the highest and lowest site mean
values expected for the actual soil type. The width of this interval is
Figure 2. E(E)/D(M) and D(M)/p as functions of v
characterized by V (M J - examples of lognormal density functions is shown
in figure 1. The estimated coefficient of variation with respect to an actual
value of M l for a specific site is thus denoted E (V J.
n + v (7)
This means that a wide interval for the parameter 8 for a certain soil type can
be anticipated. On a certain site a smaller variation can be assumed - n n
n + v ( 8)
although the mean value of this smaller variation can fluctuate in the above-
mentioned wide interval for the entire soil type.

It is then necessary to transform the prior information as well as the results s* - (9)
n + v +n v^ n+vJ
from the testing into a normal distribution. A convenient normalization
constant is k = ECMJ and the following approximations are valid:
From the predictive, posterior density distribution a certain, lower X-fractile
E(M) - ln(E(M,J)/k) = 0 ( 1) of t i can be calculated by

D(M) « VCMJ (2) n* + 1

ill - T - s* ( 10)
E ff) = ECVJ ( 3) n’ - 1

where (M, Z) = (E(li), D(T|)) and (M,.. V J = (E(0), V,). where tl4 is found from a table as the (l-X)-fractile in the t-distribution with
(n‘ - 1) degrees of freedom.
This apriori knowledge can be interpreted as a fictive sample with the
parameters for sample size (v) and sample standard deviation (P) determined The posterior density of the expectation M = E(r|) is also related to the t-
by distribution and similarly X-fractile of this distribution Is

P E (Tl)x = ri- - s '

D(M) = (4) ( 11)
V^v-3 V/n 7 ^

v 2 Finally, the fractiles should be transformed back to the original logarithmic

E © =P (5)
distribution by using 0 j = k exp T)j and E (0)i = k exp E ( n ) \ for equation
2 r < « ) (10) and equation (11), respectively. The value 0* thus means the X-fractile
in the predictive density for 0. Correspondingly, the value E (0)^ means the
where T is the gamma function and T) is the parameter in the transformed X-fractile of predictive mean value of 0.
space - i.e. T| = In (9/k). To facilitate the computations E(E)/D(M) and
IXMVp are shown as functions of v in figure 2. Equation (11) is well-suited to calculate characteristic values defined by a
specified fractile if the problem can be related directly to the properties of the
The above-mentioned apriori knowledge is supplemented with the results of mean value of the soil material parameter within the Influence zone.
soil tests on n samples with a mean value of ii and a sample standard
deviation of s definal as EXAMPLE

In this example the characteristic value of the undrained mean shear strength
( 6) E(cll) of clay till should be determined in connection with the design of a
£ (n, - n)2/n
I- 1 strip footing with a width of about 0.5 m. The characteristic value shall be
The total sample has the following parameters for sample size, mean value used to determine the bearing resistance of the footing in the ultimate limit
and sample standard deviation, respectively: state.

439
Apriori it la estimated that the clay till is weak and has the following Table 2. Results from vane shear tests
statistical values of undralned shear strength
c, kN/m1 In M 0
E (M J = k = 100 kN/m1 50 -0.693
D (M J = 25 kN/m2
E (V J = 0.25 50 -0.693

60 -0.511
The corresponding parameters for the normal distribution are approximately
81 -0.211
E (M) = 0
89 -0.117
D (M) = D (M J/E (M J = 0.25

and furthermore E (X) = E (V J = 0.25 Calculations performed on basis of more test results gave considerably
greater mean value, and the procedure by incorporating the apriori knowledge
This yields E (E)/D (M) = 1 was thus highly justified in this case.

and from figure 2 the statistical parameters for the Active sample are obtained DISCUSSION AND CONCLUSIONS

v = 3.2 It may be worth mentioning that geotechnical structures very rarely fail due
to the fact that the calculation procedure used does not precisely model the
tXM) problem or due to the fact that a too low safety factor has been introduced.
~ r ~ - 2 Geotechnical structures fail due to such gross errors as the existence of
unidentified weak soil layers or unexpected pore water pressures. The adverse
effects of such errors cannot be prevented by any rational design process.
Only the knowledge and skills of the designer will play a role in avoiding
such gross errors.

The soil testing consists of 5 vane tests performed in the influence volume - The limit state design format with the use of partial factors of safety is at
i.e. to a depth of d = 0.7 b beneath the base. The results of the tests are given present considered the most rational design procedure for geotechnical routine
in table 2. problems. Within this framework the characteristic values of soil properties
have to be selected in a well defined way if a specific safety level against
According to Danish experience the vane strength (cv) may be interpreted failure in the various limit states shall be obtained. The experienced engineer
directly as a measure for the undrained shear strength ( c j for clay till without will always attempt to incorporate his apriori knowledge of the soils involved
application of any conversion factor. in this selection process. However, the apriori knowledge will have to be
quantified in such a way that it can be combined with the results of soil
With T| = in (cjk) the mean value and sample standard deviation are: testing when selecting characteristic values. The Bayesian statistical method
T)= -0.445 and s = 0.241, respectively. provides a tool for such a quantification.

According to equations (7), (8) and (9) the parameters for the total sample It should be noted, however, that not all aspects of the concept of structural
are reliability are included in the above mentioned procedure. At the present
stage of development where only a few partial factors are used design by
n‘ = 8.2 means of such factors is in itself an approximation when the intention is to
T|= -0.271 obtain a specified level of reliability for a large category of structures.
s' = 0.243
A great amount of research is being performed in the area of structural and
and the lower 5% fractile of the mean value is obtained from equation (11) geotechnical reliability and the numerical values of the partial factors which
as E (Ti)j, = -0.443. The characteristic value can now be determined as are at present used for geotechnical design will undoubtedly undergo
substantial changes in the future both with respect to a more precise
c,^ = k exp E(ii)j* = 64 kN/m1 definition and their numerical values.

If no apriori information had been applied the following values would have ACKNOW LEDGEMENT
been obtained
The authors would like to acknowledge the helpful assistance received from
Tl=-0.445 and s = 0.24! v M n-1) = 0.269 Professor Gunnar Mohr, the Engineering Academy of Denmark, with respect
to the implementation of the Bayesian method and reviewing the paper.
The characteristic value would then have been calculated as c,* = k exp
E(n)‘j* = 50 kN/m1. REFERENCES

Certainly, a lower value than when the apriori assumption with an anticipated Eurocode 7. Part I: Geotechnical Design, General Rules. Fourth draft
mean of c, = 100 kN/m1 was included. It should be added that the present produced by Comité Européen de Normalisation. CEN, 1993. - May be
example deals with values from a real site where the apriori knowledge was obtained from the Danish Geotechnical Institute.
not based on test results. More than 5 vane tests were actually performed, but Ditlevsen, O., Madsen, H.O. (1990). Bœrende konstruktioners sikkerhed.
the five results presented here were selected randomly. SB 1-Report no 211, Statens Byggefoiskningsinstitut (in Danish).
Ditlevsen, O., Madsen, H.O. (1993). Structural reliability methods. To be
published.

440