GEORISK, 2016

http://dx.doi.org/10.1080/17499518.2016.1235711

Defining and selecting characteristic values of geotechnical parameters for

designs to Eurocode 7
Trevor L. L. Orr
Department of Civil, Structural and Environmental Engineering, Trinity College Dublin, Dublin 2, Ireland

ABSTRACT ARTICLE HISTORY

Recent geotechnical design codes (such as Eurocode 7) aim to achieve geotechnical designs with an Received 15 June 2016
appropriate target reliability by applying partial factors to characteristic parameter values. Accepted 8 September 2016
Development in Eurocode 7 of the definition and selection of the characteristic values is traced.
KEYWORDS
A significant development occurred with the definition of the characteristic value as a cautious Eurocode 7; characteristic
estimate of the value affecting the occurrence of the limit state. Statistical equations to select value; correlation factor;
characteristic values are presented, including one proposed for the revised of Eurocode 7.An target reliability
alternative equation reduces the amount of subjective judgement and is more likely to achieve
the target reliability.

1. Introduction selection of characteristic geotechnical parameter values

is normally based on the results of a limited number of
In recent years, geotechnical design codes, such as Euro-
test results tempered with judgement and using previous
code 7 – that is, EN 1997-1 (CEN 2004) – based on the
relevant experience rather than through being selected
limit state design method have been adopted that aim to
statistically. The consequence of this is that the values
achieve geotechnical designs with an appropriate target
selected by different engineers for soil parameters for a
reliability through the application of partial safety factors
particular design situation, even when based on the
to characteristic parameter values. It should be noted
same test results, can vary greatly due to the subjective
that the term “partial safety factor” is not used in Euro-
nature of their judgment. Furthermore, due to the use
code 7 and the other Eurocodes; instead these codes refer
of subjective judgement, it is impossible to verify if the
to “partial factors” which are factors applied to actions,
target reliability of a design has been achieved. The pur-
material properties and/or resistances to provide safety
pose of this paper is to trace the changes that have
and also to account for model uncertainties and dimen-
occurred in how the parameter values for use in geotech-
sional variations. In principle the characteristic values of
nical designs have been defined and selected and what
geotechnical parameters are selected so as to take
changes have been proposed for the revision of Eurocode
account of the inherent variability of the ground, the
7 to make the selection of characteristic soil parameter
uncertainty in the determination of the soil parameters
values more objective and consistent so that geotechnical
and the extent of the relevant failure mechanism, while
designs are more reliable.
the magnitude of the partial factors, which are usually
set by national standards organisations, are chosen so
as to provide the level of reliability or safety required
by society and, ideally, so that the reliability level of 2. Start of Eurocode 7 and definition of
designs is as close as possible to the target reliability parameter values for use in design
level. In structural design using manufactured materials, In 1981, when a committee of the International Society
such as steel and concrete, the characteristic value of a for Soil Mechanics and Foundation Engineering
material parameter is defined statistically as a certain (ISSMFE), now called the International Society for Soil
fractile of an unlimited test series. Because structural Mechanics and Geotechnical Engineering (ISSMGE),
materials are manufactured, such a statistical definition started work on drafting a Model Code for Eurocode 7
is appropriate. However, due to soil being a natural (see Orr (2008) for details), most of the existing codes
material with variable properties and due to the limited of practice for geotechnical design provided little or no
number of test results that are usually available, the guidance on the selection of geotechnical parameter

CONTACT Trevor L. L. Orr [email protected]

© 2016 Informa UK Limited, trading as Taylor & Francis Group
2 T. L. L. ORR

values for use in design calculations. For example, the opinion of the Committee these differences make it at
British Code of Practice for Foundations, CP 2004 (BSI present impossible to write a detailed, prescriptive Euro-
1972) stated that, when calculating the ultimate bearing code for geotechnical design.
capacity of foundations on cohesive soils, “the shear In the context of the move towards adopting the limit
strength may be determined from the results of direct state method with characteristic values and partial for
shear tests or in-situ vane tests” without giving any geotechnical and structural designs, Simpson, Pappin,
further guidance on the selection of the parameter value. and Croft (1981) noted that it had been found difficult
While the committee preparing the Model Code was to develop an approach to design which is consistent
initially a sub-committee of the ISSMGE, during the and appears sensible to both structural and geotechnical
first year it was decided that it was not appropriate for engineers. They also noted that the degree of uncertainty
the ISSMGE to be involved in the preparation of stan- with which a given geotechnical parameter, such as
dards and hence the Model Code Committee became undrained shear strength, can be assessed varies signifi-
an Ad Hoc Committee. In 1987 the Ad Hoc Committee cantly from one project to another. They concluded that
completed its work on the preparation of the Model prescribed partial factors applied to mean or “character-
Code for Eurocode 7 and submitted this document (Ad istic” values of the parameters are therefore not adequate
Hoc Committee 1987) to the Commission of the Euro- for deriving design values intended to accommodate the
pean Communities in 1987, but it was not published. variability. They recommended that engineers con-
This Model Code provided common unified rules for sidered directly the worst conditions that might credibly
geotechnical design that aimed to serve as a model for occur, as well as the expected conditions, that is, the most
Eurocode 7, which was intended to be one of the suite likely conditions. However they pointed out that it would
of Eurocodes for the design of structures, including geo- sometimes be too conservative to set all independent
technical aspects, all based on the same limit state design variables to their worst values at the same time. Hence
method with partial factors applied to characteristic par- they proposed what they termed the λ method to select
ameter values defined statistically as a particular fractile the appropriate combination of the expected and worst
of an unlimited test series. credible values of the parameters and provide uniform
One of the key reference documents for the prep- reliability. It should be noted that Simpson, Pappin,
aration of the Model Code for Eurocode 7, and also and Croft (1981) were aiming at design values of par-
the other Eurocodes, was CIRIA Report 63 (CIRIA ameters, not characteristic values, but their suggestions
1977) Rationalisation of safety and serviceability factors for the design values of the non-leading variables
in structural codes. This document presented the results would be fairly similar to characteristic values.
of a comprehensive study and analysis of the use of the In 1981, the Danish Code of Practice for Foundation
limit state method with characteristic values and partial Engineering (DIF 1984) was the only geotechnical code
factors to achieve structural designs with an appropriate of practice in Western Europe which was based on the
reliability. However, the Model Code committee did not limit state design approach with partial factors applied
consider the focus on calculation models and the statisti- to characteristic parameter values (Orr 2008). Hence,
cal definition of the characteristic value in this docu- this code of practice formed a useful reference document
ment, which was intended for manufactured materials, for the preparation of the Model Code for Eurocode
to be appropriate for geotechnical designs and the deter- 7. With regard to the selection of geotechnical parameter
mination of the characteristic values of soil properties, values, the Danish Code stated that “characteristic values
since soil is a natural material, and anticipated it would for weight density, and for the strength and deformation
not be accepted by the European geotechnical commu- parameters for soil should therefore be established by a
nity. This view is reflected, though not stated explicitly, conservative estimate based on the results of relevant
in the Preface to the Model Code where it is noted measurements”. While the statement that characteristic
For conventional structural design it is a common fea- values should be “established by a conservative estimate”
ture that much of the safety evaluation is centred around provides some guidance, it is not specific and leaves the
calculation models. In contrast in geotechnical design designer to decide subjectively how conservative this
much more effort is devoted to identification and estimate should be.
characterization of the relevant ground mass and the Although the Danish Code was used for guidance
processes (physical, chemical and mechanical) taking
when drafting the Model Code, the term “characteristic
place in it, than to applying sophisticated calculations
models. … The Ad-Hoc Committee has also observed value”, which was adopted in the Danish Code, was
considerable differences in the professional geotechnical not used in the Model Code. In the short Annex to the
practice in the various European countries. In the Model Code, some partial factors were given and it was
 GEORISK 3

stated that, when lower bound design values are required small, the results of individual tests are representative
for calculations, “the values are found from the lower of the behaviour of the material in the section being
representative values by division by the partial factors”. designed so that the definition of the characteristic
However, it is not explained how the representative value as a particular fractile of the test results is appropri-
values should be selected, apart from stating in the ate. However, in geotechnical design, because of the par-
code text that the design values “must be appropriate ticular features of soil, the definition of the characteristic
for the particular limit state and its corresponding value as a particular fractile of the test results is not
model under consideration”. appropriate. These features include the fact that the
volume of soil involved in a geotechnical limit state is
usually larger than the volume involved in a structural
3. Eurocodes 7 and the definition of the limit state and is much larger than the volume tested
characteristic value (Orr 2012). The parameter value controlling soil behav-
iour is the mean value over the failure surface, which has
Once work on the Model Code had been completed and a lower variation than the local variation in the par-
it had been submitted in 1987 to the European Commis- ameter value (Vanmarcke 1977). Hence the character-
sion (and not published), work started on the next ver- istic value needs to be assessed as a cautious
sion of Eurocode 7, which was the ENV version, that assessment of the mean value over the relevant volume,
is, the European Prestandard or, in German, Euro- not as a particular fractile of the test results. Also,
päischer Vornorm, from which the abbreviation ENV being a natural material, it is more variable compared
is derived. ENVs are issued for provisional application to a manufactured material, such as steel and concrete.
and comment. The European Commission transferred Hence for increased soil variability and when the extent
the work on the Eurocodes to the European Committee of the failure zone is small, the characteristic value will be
for Standardization (CEN) in 1990 and the ENV version a lower and more cautious value. Furthermore, in a geo-
of Eurocode 7 was published in 1997 as ENV 1997-1: technical design situation, only a few tests are carried out
Eurocode 7: Geotechnical Design – Part 1: General on a very small portion of the total volume of soil
Rules (CEN 1997). Moving from the Model Code to a involved in the particular limit state so that, if a statistical
trial Eurocode, the ENV version had to adopt the limit method is used, account needs to be taken of the number
state method, with partial factors applied to character- of test results, the extent of the failure surface and rel-
istic values in order to achieve designs with the target evant experience of similar ground conditions.
reliability specified in the head Eurocode, that had In order to adopt the limit state method with
been adopted as the basis for all the Eurocodes and characteristic values in Eurocode 7, a definition for
was eventually published in 2002 as the European Stan- the characteristic value had to be found that was con-
dard version, EN 1990: Eurocode – Basis of Structural sistent with the EN 1990 definition and was also
Design (CEN 2002). appropriate for geotechnical design. After much dis-
Since the Eurocodes are mainly concerned with the cussion by the ENV Project Team with the Coordi-
design of structures involving manufactured materials, nation Committee for the Eurocodes, the following
such as concrete and steel, the definition of the charac- definition in §2.4.3(5) of ENV 1997-1 was agreed for
teristic value of a material property given in §1.5.4.1 of the characteristic value of a soil parameter: “The
EN 1990 is that it is the characteristic value of a soil or rock parameter shall
value of a material or product property having a pre- be selected as a cautious estimate of the value affecting
scribed probability of not being attained in a hypotheti- the occurrence of the limit state”. This definition
cal unlimited test series. This value generally remained unchanged in §2.4.5.2(2) of the EN, that is,
corresponds to a specified fractile of the assumed stat- European Standard version of Eurocode 7, that was
istical distribution of the particular property of the published in 2004. Examples of the selection of charac-
material or product.
teristic values in the case of a pile are shown in Figure
EN 1990 further states in §4.2(3) that “Unless other- 1 from Frank et al. (2004), wherein the characteristic cu
wise stated … where a low value of material or product values for the pile shaft and base (tip) resistances are
property is unfavourable, the characteristic value should shown. The characteristic cu value for the pile base
be defined as the 5% fractile value”. resistance is determined from the local and usually
In structural design involving manufactured few cu values in the small volume of soil around the
materials, where a large number of samples of the struc- pile base and thus needs to be a much more cautious
tural material can be tested and where the variability of estimate than the characteristic cu value over the
the material in the section being designed is usually length of the pile shaft, which is closer to the average
4 T. L. L. ORR

Figure 1. Characteristic cu values for determination of the shaft and base resistance of a pile (Frank et al. 2004).

value of a larger number of cu values. These values are been worded so as not to be prescriptive. How cautious
examined in Section 5.4. the selected estimate of the characteristic value should
The first requirement in Eurocode 7 for selecting the be has been left to the designer to decide, taking
characteristic value of a geotechnical parameter, given in account of the factors listed above and, as noted in
§2.4.5.2(1) of EN 1997-1, is that it “shall be based on 2.4.5.2(1), also taking account of well-established experi-
results and derived values from laboratory and field ence. This requirement indicates that characteristic
tests, complemented by well-established experience”. values should not be based on a purely statistical analysis
Further guidance is provided in the form of the following of the test results but involves the designer making sub-
six factors listed in §2.4.5.2(4) that need to be taken into jective judgements on a number of factors and using
account when selecting the characteristic value: experience.
Due to the subjective nature of the selection process,
(1) Geological and other background information, such when a number of geotechnical engineers are presented
as data from previous projects with the same data, they will select different character-
(2) The variability of the measured property values istic values, reflecting their different experiences, assess-
(3) The extent of the field and laboratory investigations ments of the design situation and interpretations of the
(4) The type and number of samples characteristic value. Bond and Harris (2008) found this
(5) The extent of the zone governing the behaviour of when they asked geotechnical engineers to assess the
the structure at the limit state being considered, and characteristic parameter values of London and Lambeth
(6) The ability of the structure to transfer loads from clays from the results of Standard Penetration Tests
weak to strong zones. (SPTs) carried out in these soils. The characteristic SPT
values selected by the engineers, shown as solid lines,
The wording in Eurocode 7 to define and select a are presented in Figure 2 and show a wide spread,
characteristic geotechnical parameter value is a major which demonstrates that engineers are not consistent
improvement on previous geotechnical codes of prac- when selecting characteristic geotechnical values, par-
tice, most of which provided no, or very little, guidance ticularly if the available data are scattered. This indicates
on how to select soil parameters values. The definition that more guidance is needed or this interpretation
of the characteristic value in Eurocode 7 has purposely uncertainty should be explicitly accounted for so as to
 GEORISK 5

situation involving a small zone of soil, that is, a local

failure, compared to a situation involving a large zone,
that is, a global failure, where averaging of parameter
values is assumed to occur and less conservative par-
ameter choices are justified. This is an example of
some guidance provided in Eurocode 7 with regard to
Factor 5 in Section 3 concerning the extent of the zone
governing the behaviour and is discussed further in Sec-
tion 5. This note also recognises that, in geotechnical
design, there is always a limited number of test results
from a very small proportion of the total volume of
soil involved in a limit state and hence the mean value
is an estimate of the relevant mean value.

4.2. Selection of characteristic value to Eurocode 7

using correlation factors
Apart from providing the guidance discussed above in
Section 4.1 with regard to the selection of characteristic
parameter values for a local failure as opposed to a global
failure, Eurocode 7 provides correlation factors, ξ to
Figure 2. Results of SPTs on London and Lambeth clays, with
engineers’ interpretations of the characteristic value (Bond and determine the characteristic values of pile resistances
Harris 2008). directly from test results, which take account of the
spatial variability of the pile resistance over the site due
to the variability of soil. Bauduin (2002) has provided
reduce the spread of the selected characteristic values
some of the theoretical background to the three sets of
and achieve designs with more consistent reliability.
ξ values given in Eurocode 7, which are based on statisti-
This is particularly important in view of the fact that
cal analyses for situations involving: (a) a number of sta-
the partial factors are precisely prescribed.
tic pile tests, (b) a number of resistances calculated from
profiles of tests and (c) a number of dynamically tested
piles. In the case of static pile load tests, the mean and
4. Selection of characteristic value according minimum measured pile resistances, (Rc;m)mean and
to Eurocode 7 using statistics and correlation (Rc;m)min, are divided, as indicated in Equation (1), by
factors the appropriate ξ1 and ξ2 values depending on the num-
ber of pile load tests. Eurocode 7 provides recommended
4.1. Selection of characteristic value according to
ξ1 and ξ2 values, which are shown in Table 1, with the
Eurocode 7 using statistics
lowest result giving the characteristic value. Since the
Although not promoting the use of statistics to deter- mean and minimum resistance values are divided by
mine the characteristic value of a geotechnical par- the correlation factors to obtain the characteristic resist-
ameter, Eurocode 7 states in application rule §2.4.5.2 ance, the correlation factors are all not less than unity.
(11) that, if statistical methods are used to select the

(Rc;m )mean (Rc;m )min
characteristic value, then it should be derived such that Rc;k = Min ; . (1)
“the calculated probability of a worse value governing j1 j2
the occurrence of the limit state under consideration is
not greater than 5%”. A note to this clause states that When using the recommended ξ1 and ξ2 values to
a cautious estimate of the mean value is a selection of the determine Rc;k from static pile load tests, the procedure
mean value of the limited set of geotechnical parameter
values, with a confidence level of 95%; where a local fail- Table 1. Correlation factors ξ to derive characteristic values from
ure is concerned, a cautious estimate of the low value is a static pile load tests.
5% fractile
ξ for n = 1 2 3 4 ≥5
This note is stating that a lower, that is, more cautious, ξ1 1.40 1.30 1.20 1.10 1.00
ξ2 1.40 1.20 1.05 1.00 1.00
characteristic value should be selected in a design
6 T. L. L. ORR

is fully prescribed, involving assumptions about the from piles where the ground is weaker to piles where
design situation, including the variability of the ground. the ground is stronger. If, however, the structure is stiff
It does not require any further judgement with regard to and strong enough to do so, then the ξ values may be
the six factors that need to be taken into account, which reduced by the factor 1.1, provided they do not become
are listed in §2.4.5.2(4) and given in Section 3, and there- less than 1.0. This is an example of a way in which Factor
fore all designers should select the same characteristic 6 noted in Section 3, that is, “the ability of a structure to
resistance if they are using the same set of static pile transfer loads from weak to strong zones”, may be taken
load test results. into account when selecting the characteristic value.
In the case of pile resistances calculated from profiles As may be seen in Tables 1 and 2, the recommended ξ
of soil test results, similarly to measured resistances values decrease towards unity as the number of static pile
obtained from pile load tests, the mean and minimum load tests and profiles of test results increase, reflecting
calculated pile resistances, (Rc;cal)mean and (Rc;cal)min are the fact that the statistical uncertainty in the pile resist-
divided by appropriate ξ3 and ξ4 values with, as indicated ance reduces and the reliability increases as more infor-
in Equation (2), the lowest result giving the characteristic mation is obtained about the ground properties so that
value. Eurocode 7 provides recommended ξ3 and ξ4 higher, that is, less cautious, Rc;k values and hence
values, which are shown in Table 2 and depend on the more economical pile designs may be selected to provide
number of profiles of test results. Eurocode 7 does not the required degree of reliability. This shows the advan-
define what exactly a profile of test results is but it is nor- tage of carrying out more loads tests and test profiles. It is
mally assumed to be a continuous or reasonably continu- noted that, except for when there is only one pile load
ous profile of soil test results such as is obtained, for test and one profile of test results, the recommended ξ1
example, from a cone penetration test or a standard pen- and ξ2 values are all smaller than the corresponding ξ3
etration test. and ξ4 values. This reflects the fact that these correlation

 factors take account of uncertainty in the calculation
(Rc;cal )mean (Rc;cal )min model used to calculate the resistance as well as uncer-
Rc;k = Min ; . (2)
j3 j4 tainty due to the spatial variability in the soil properties,
whereas the ξ1 and ξ2 values only account for spatial
variability as they are applied to resistances that are
The recommended ξ values given in Tables 1 and 2 measured directly in static pile load tests and hence do
have been determined so as to provide characteristic not involve calculation model uncertainty. This aspect
pile resistance values with a confidence of 95% for the is also examined in Section 5.4.
following situations:

. When the coefficient of variation (COV) of the pile 5. Use of statistical methods to determine the
resistance is less than about 0.1, the characteristic characteristic parameter value
resistance is governed by the mean of the measured
5.1. Characteristic value as the 5% fractile of an
pile resistances from static pile load tests or the
unlimited test series
mean of the pile resistances calculated from profiles
of tests. In the case of manufactured materials, such as concrete
. When the COV of the pile resistance is greater than and steel, whose properties are random variables with
about 0.1, the characteristic resistance is governed an inherent variability that can be represented by a nor-
by the lowest measured pile resistances from static mal probability Gaussian distribution, described by a
pile load tests or the lowest pile resistance calculated mean value and a COV, and where the characteristic
from profiles of tests. value is defined as the 5% fractile of an unlimited test
series, the characteristic value, Xk is given by
A further assumption in the derivation of the ξ values
Xk = Xmean − 1.645SD = Xmean (1 − 1.645COV), (3)
is that the structure to be supported by the piles is flexible
and hence not stiff and strong enough to transfer loads where Xmean is the mean value, SD the standard deviation,
COV the coefficient of variation and 1.645 is the Student
Table 2. Correlation factors ξ to derive characteristic values from t-correlation factor, t95;∞ corresponding to a 95% confi-
profiles of test results. dence in the occurrence of the 5% fractile value for an
ξ for n = 1 2 3 4 5 7 10 unlimited test series. This equation provides a too cau-
ξ3 1.40 1.35 1.33 1.31 1.29 1.27 1.25 tious characteristic value for a geotechnical parameter
ξ4 1.40 1.27 1.23 1.20 1.15 1.12 1.08 when, as in many geotechnical design situations, the
 GEORISK 7

geotechnical failure mode is larger than the scale of fluc- confidence level of 95% corresponding to 13 test results
tuation (SOF) of the relevant soil parameter and hence and yields a ξ value that is much less conservative than
the relevant value is the 95% confidence in the mean the ξ value for the 5% fractile obtained using Equation
value controlling the occurrence of a limit state rather (3). The inverse of (1–0.5 COV) gives a ξ value that is
than the 5% fractile. Examples of geotechnical design situ- similar to the ξ values in Tables 1 and 2 to determine
ations where the 5% fractile of a parameter value, or a characteristic pile resistances. The ξ values obtained
value close to it, is the characteristic value occur in the using Equation (4) with COV values ranging from 0 to
case of seepage or some shallow foundations. Choosing 0.5 are shown in Table 4.
the characteristic value as the 95% confidence in the Using Equation (4) and the information in Tables 3
mean value assumes that the soil profile is known, there and 4, the characteristic weight density of soil is equal
is continuous spatial variability and no local weak zones to the mean weight density since ξ = 1.0 for COV = 0;
that result is preferential failure paths. the characteristic friction angle is obtained by dividing
The inverse of (1–1.645 COV) gives a ξ factor to tanϕ′ mean by 1.05 since ξ = 1.05 for COV = 0.1; and the
determine the characteristic value of a soil parameter characteristic undrained shear strength is obtained by
from the mean value and hence is similar to the ξ values dividing the mean cu value by 1.25 since ξ = 1.25 for
in Tables 1 and 2 to determine characteristic pile resist- COV = 0.4.
ances. Ranges of typical COV values and recommended
COV values by Schneider and Schneider (2013) for
different soil parameters are given in Table 3. Using 5.3. Characteristic value as a function of the
Equation (3) and the recommended COV values of 0.1 number of test results
and 0.4 for tanϕ′ and cu gives ξ values of 1.20 and 2.92.
A further development in the use of statistics in geo-
As noted above, in many design situations, these ξ values
technical design codes to determine the characteristic
would result in characteristic geotechnical parameter
values was provided by the Japanese Geotechnical
values much smaller than normally selected and hence
Society (JGS 2004) in its document Principles for Foun-
too cautious.
dation Design Grounded on Performance based Design
Concept. This document includes the following statisti-
cal equation to calculate the characteristic value as the
5.2. Characteristic value determined using
value with a confidence level of α% in the mean value
Schneider’s (1997) equation
and a function of the number of test results, n with a
While ξ values have been provided in EN 1997-1 to standard deviation, SD:
determine the characteristic resistance from static pile ta;v SD
load tests, profiles of test results and dynamic impact Xk = Xmean − √ , (5)
n
tests, no ξ values have been provided to determine
characteristic soil parameter values from field or labora- where tα;v is the Student t factor for a confidence level
tory tests. Schneider (1997) proposed the following stat- of α% in the case of v degrees of freedom and v is equal
istical equation, which is a modification of Equation (3), to n−1, assuming a normal distribution. The reason
to calculate an approximate value for the characteristic why the statistical Equation (5) has been included in
parameter value: the Japanese document and not in Eurocode 7 may
be explained by the results of a survey of carried out
Xk ≈ Xmean (1 − 0.5 COV). (4)
by Shirato, Fukui, and Matsui (2002), who found that
This equation aims to provide a characteristic value that 56% of Japanese geotechnical designers thought the
is a cautious estimate of the mean value with a statistical approach was a useful tool. This contrasts
with the view of the Model Code committee, noted in
Table 3. Ranges of typical COV values for soil parameters and Section 2, who considered a statistical approach
recommended values (Schneider and Schneider 2013). would not be accepted by the European geotechnical
Range of typical Recommended
Soil parameter Symbol COV values COV values community. Many respondents to the survey provided
Weight density γ 0.01–0.10 0 positive comments which included that it eliminated
Angle of internal tanφ′ 0.05–0.15 0.1
friction
Cohesion c′ 0.30–0.50 0.4 Table 4. ξ to derive characteristic parameter values for different
Undrained shear cu 0.30–0.50 0.4 COV values.
strength
COV 0 0.1 0.2 0.3 0.4 0.5
Compressibility mv 0.20–0.70 0.4
modulus ξ 1.00 1.05 1.11 1.18 1.25 1.33
8 T. L. L. ORR

Table 5. ξ values for 95% confidence in the mean for n test results with COV = 0.1 and 0.4 derived from Equation (6).
Number of test results, n 2 3 4 5 6 7 8 9 10
Degree of freedom v = n−1 1 2 3 4 5 6 7 8 9
t95,v value 6.31 2.92 2.35 2.13 2.02 1.94 1.90 1.86 1.83
ξ when COV = 0.1 1.81 1.20 1.13 1.11 1.09 1.08 1.07 1.07 1.06
ξ when COV = 0.4 (−1.27) 3.07 1.89 1.62 1.49 1.42 1.37 1.33 1.30

subjectivity in the determination of parameter values plotted in Figures 3 and 4. Examining the graphs in
and was excellent in terms of accountability and objec- Figures 3 and 4 it can be seen that the ξ values obtained
tivity. However the respondents also commented that from Equation (6) for both COV = 0.1 and COV = 0.4
parameter values should be determined by considering fall between those given by Equations (3) and (4) and
other information as well as test results and expressed that Equation (3) for the 5% fractile gives a very large
the view that generally it is rare that a sufficient num- ξ value, and hence a very conservative Xk value for
ber of test results is available to apply a statistical more than three test results, while Equation (4) gives
approach. Also since the reliability of ground investi- a conservative Xk value for n less than 10. Which
gation data depends on the investigators, there is a ξ value should be used to determine the Xk value
great need for improvements in the quality of ground depends on the design situation and the extent of the
investigations. governing failure zone with respect to the soil spatial
Since SD = COV × Xmean, substitution for SD in variability, as noted in Section 5.1.
Equation (5) yields the following equation for Xk, like The test data in the pile example shown in Figure 1 are
Equations (3) and (4): used now to show the advantage of the statistical
  approach for a situation where only material parameters
ta;v COV are present. The mean and standard deviation of the 10
Xk = Xmean 1 − √ . (6)
n cu values over the pile shaft were calculated and the COV
was found to be 0.29. Using this COV value, the ξ value
The Student t-distribution factors for 95% confidence based on Equation (6) is calculated to be 1.17, which is
level, t95;v for v degrees of freedom ranging from 1 to 9 very similar to ξ = 1.11 corresponding to the line
are given in Table 5. The value of ξ to determine the shown in Figure 1 for the characteristic cu value over
characteristic value with a 95% confidence level for n
√ the length of the shaft that was selected from the test
test results is given by the inverse of (1 − t95;v COV/ n). results using judgement. Carrying out a similar calcu-
The ξ values based on Equation (6) for n ranging lation using the five tests results around the base of the
from 2 to 10 with COV values of 0.1 and 0.4 are also pile with COV = 0.41 gives ξ = 1.76. This is very similar
given in Table 5 and plotted in Figures 3 and 4. These to the ξ = 1.67 corresponding to the line shown in
show that for design situations where COV = 0.1 and Figure 1 for the characteristic cu value at the base of
the characteristic value is the 95% confidence in the the pile that was selected from the test results using
mean value based on the number of tests results between judgement.
2 and 10, the ξ values range from 1.81 to 1.06, while for
COV = 0.4 the ξ values range from 3.07 to 1.30 for
n between 3 and 10. For comparison, the graphs of the
ξ values obtained using Equations (3) and (4) are also

Figure 3. ξ values as a function of the number of test results for Figure 4. ξ values as a function of the number of test results for
COV = 0.1. COV = 0.4.
 GEORISK 9

Similar to the graphs in Figures 3 and 4 giving the ξ is the coefficient of variation of the transformation errors
values to determine the Xk values for different numbers and COVstat is the coefficient of variation of the statisti-
of test results, the Eurocode 7 ξ values in Tables 1 and cal parameters.
2 to determine the characteristic pile resistance, Rk for Essentially Equation (7) is stating that the inherent
different numbers of pile load tests and profiles of tests variability part of the total variability depends on the
are plotted in Figure 5. These graphs show how the ξ size of the governing failure mechanism relative to the
values decrease and hence a less cautious Rk is obtained SOF.
as the number of pile load tests and profiles of tests For a 1D slip line in a 2D situation and based on Van-
increases. They also show that the. ξ1 and ξ2 values to marcke (1977):
determine Rk from pile load tests are less than the ξ
values obtained using Equation (4) for number of soil G 2S = [2Lw /dw −1+exp( − 2Lw /dw )]/(2L2w /d2w ), (8)
tests when n exceeds 4, and that the ξ3 and ξ4 values to where δw is the SOF along the potential slip line direc-
determine Rk from profiles of test results embrace the tion, w, and Lw is the length of the potential slip line in
5% fractile value when COV = 0.1. the y–z plane. Schneider and Schneider (2013) adopted
the following simplification for the variance reduction
factor, which is G 2S = G 2x G 2y G 2z , where G 2x and G 2y are
5.4. Characteristic value as a function of the the variance reduction factors in the two horizontal
extent of the failure surface directions, x and y, and G 2z is the variance reduction fac-
As noted in Section 3, when determining the character- tor in the vertical direction, z together with the following
istic value, account should be taken of Factor 5 which two simplified equations by Vanmarcke (1983) for G 2i in
is “the extent of the zone governing the behaviour of a particular direction, i:
the structure at the limit state being considered”. The
G 2i = [di /Li (1 − di /3Li )] if Li .di (9)
spatial variation maybe taken into account using the
SOF, which is the distance over which the properties and
are relatively strongly correlated. To account for the
influence of the spatial variability and other uncertainties G 2i = [1 − Li /3di ] if Li .di , (10)
affecting the characteristic value, the COV in Equation where δi and Li are the SOF and the extent of the failure
(3) can be replaced by the following additive total coeffi- mechanism in the direction i, respectively.
cient of variation, COVtotal proposed by Phoon and Kul- If soil parameter values are derived using accurate
hawy (1999): equipment and strictly in accordance with the relevant
 testing standards, the measurement errors should be
COVtotal = GS2 COV 2inher +COV 2meas +COV 2trans +COV 2stat , small so that COV2meas  0. Similarly, if a well-estab-
(7) lished model is used to transform the measured test
results into the required parameter, then
where G 2S is the variance reduction function (Vanmarcke
COVtrans  0, and if the parameters required to
2
1977) considering the spatial extent of the governing fail-
describe the statistical distribution are known with
ure mechanism, COVinher is the coefficient of variation
reasonable accuracy from experience with similar soils,
of the soil’s inherent variability, COVmeas is the coeffi-
then COV2stat  0. Hence, assuming a normal distri-
cient of variation of the measurement errors, COVtrans
bution and taking account of the above comments
regarding the COV values, that is, assuming a perfect
transformation model and no measurement errors, sub-
stituting the COVtotal from Equation (7) for COV in
Equation (3) gives
 
Xk = Xmean 1 − 1.645 COVinher G 2S , (11)

while for a log-normal distribution, the characteristic

value is given by
√  √ 
Xk = Xmean 0.2 ln(1+G S COVinher )/ 1+G S COVinher ,
2 2 2 2
(12)

Figure 5. ξ values for no. of pile load tests and profiles of soil Schneider and Schneider (2013) recommend that a nor-
tests from Eurocode 7. mal distribution is assumed when COV < 0.3, for
10 T. L. L. ORR

example, in the case of tanϕ′ , and a log-normal distri-

bution is assumed when COV > 0.3, for example, in the
case of cu and c′ . Hence the ξ values to determine the
characteristic value from the mean value when COVinher-

< 0.3 is given by the inverse of (1 − 1.645 COVinher G2S )
while when COV > 0.3, the ξ value is given by the inverse
√ √
of 0.2 ln(1+GS COVinher )/ 1+GS COVinher .
2 2 2 2

Due to the way soil is normally stratified in horizontal

layers, the SOF is usually much smaller in the vertical
direction than in the horizontal direction with the conse-
Figure 7. Correlation factor ξ versus extent of failure mechanism
quence that the extent of the failure mechanism in the
for COV = 0.4.
vertical direction has more effect on the characteristic
value than the extent in the horizontal direction. To
examine the effect of the extent of the failure surface in 5.5. Characteristic value obtained using Bayesian
the vertical direction, Lz on the characteristic value, the analysis
ξ values obtained using Equations (11) and (12), assum-
The ENV version of Eurocode 7 stated in §2.4.3(6) that
ing δz = 2.0 m and G 2x = G 2y = 1.0, are plotted in Figures 6
“Statistical methods may be employed in the selection
and 7 for COV = 0.1 and 0.4, respectively. These graphs
of characteristic values for ground properties. Such
show that, as when the number of tests increases, the cor-
methods should allow a priori knowledge of comparable
relation factor, ξ approaches closer to unity when the
experience with ground properties to be taken into
extent of the failure zone increases, that is, the character-
account for example by means of Bayesian statistical
istic value becomes closer to the mean value. For com-
methods”. Denver and Ovesen (1994) have shown how
parison the graphs of ξ, obtained from Equations (3)
the assessment of the characteristic value can be
and (4) are also given in Figures 6 and 7. These graphs,
improved in a systematic way by using the a priori
with constant ξ values as Lz increases, show that
knowledge based on the well-established experience
Equation (3) for the 5% fractile gives a high ξ value,
referred to in §2.4.5.2(1) of the EN 1997-1. However,
and hence a very cautious Xk value, while Equation (4)
the reference to the use of Bayesian statistical methods
only provides a cautious ξ value when the extent of the
was not included in the EN version of Eurocode 7 EN,
failure zone exceeds about 20 m. While Equations (11)
although their use is not prevented.
and (12) and the assumptions upon which they are
based present a method for taking the extent of the fail-
ure surface into account, they are crude approximations, 6. Future development in selection of
as noted by Orr (2015), and more rigorous calibrations characteristic value
are needed to ascertain their associated errors.
6.1. Proposal for characteristic value by evolution
group 11
In preparation for a revised version of Eurocode 7,
which CEN plans to publish in 2020 along with revised
versions of the other Eurocodes, the CEN committee
responsible for Eurocode 7, TC250/SC7 established an
evolution group, EG11: Characterization. The aim of
this evolution group, given in its unpublished Final
Report by (EG11 2015), was to provide guidance on
determining the characteristic value of ground par-
ameters that was:

. Simple to understand
. Simple to use in practice
Figure 6. Correlation factor ξ versus extent of failure mechanism . Fulfilled the requirements of EN 1997-1 and EN 1990
for COV = 0.1. . Consistent with traditional practice and statistics.
 GEORISK 11

The aim was also to provide a more transparent sampling methods and levels of experience. The follow-
approach resulting in geotechnical engineers selecting ing suggested values are given for a:
more consistent characteristic values.
The first suggestion in the EG11 report was that the a = 0.5: In case of high quality test values and reliable,
definition of the derived value of a geotechnical par- good local ground investigation based on excellent
ameter, in §1.5.2.5 as the “value of a geotechnical par- comparable experience
ameter obtained by theory, correlation or empiricism a = 0.75 Average quality test values and ground
from test results” should be extended by the addition investigation
of the phrase “considering the uncertainties of the trans- a = 1.0 Values estimated from general experience or
formation model (e.g. from SPT or CPT to shear tabulated values and no local ground investigation.
strength) and measurement errors”. The addition of
this consideration is to provide some guidance with The ξ value to determine Xk from Equation (13) is the
√
regard to taking account of Factor 2 in Section 3 con- inverse of (1 − a(1 − Xextr /Xmean 1/Lv )).
cerning “the variability of the measured property values”. Equation (13) offers a method that is simple and easy
By considering and accounting for uncertainty in the to understand, providing guidance that could lead to the
transformation model and testing in the derived value, selection of more consistent characteristic values and
only the inherent variability, represented by COVinher geotechnical designs that are more likely to achieve the
needs to be taken into account when selecting the target reliability. According to the EG11 report, when
characteristic value and hence the subjective judgement using Equation (13) to determine Xk, the mean and
is reduced. extreme derived soil parameter values, Xmean and Xextr,
In order to take account of the Factor 3 in Section 3, are provided by the soil investigator, who should correct
which is “The extent and quality of field investigations” these values to account for the uncertainties in the testing
and also Factor 5, which is “The extent of the zone of method and the transformation model, while the geo-
ground governing the behaviour of the structure at technical design engineer is responsible for determining
the limit state being considered”, EG11 has proposed the Xk and Lv values for each design situation. This is
the following simple equation for the characteristic appropriate as the design engineer will know the nature
value: of the limit state and extent of the failure surface for each
design situation. However, in the author’s view, the soil
 investigator is not the appropriate person to provide
1
Xk = Xmean − a(Xmean − Xextr ) , (13) the Xmean and Xextr values. The site investigator may pro-
Lv vide derived parameter values but the design engineer
should be responsible for selecting the Xmean and Xextr
where Xmean = The mean of the derived values based on values since, as noted in Section 3, §2.4.5.2(1) requires
field or laboratory tests or the estimated mean value the characteristic value to be based on derived values
from comparable experience or the estimated mean that are complemented by well-established experience
value from tabulated soil properties, Xextr = The and §2.4.5.2(2) requires the design engineer to take
extreme value recorded or estimated corresponding account of data from previous projects and the ability
to an expected extreme for the hypothetical case of a of the structure to transfer loads from weak to strong
large number of tests (by means of tabulated coeffi- zones when selecting the characteristic value. The ENV
cients of variation Xextr could be estimated or con- version of Eurocode 7 stressed the importance of the
firmed), Lv = Represents the zone of ground involvement of the design engineer in the selection of
governing the behaviour of a geotechnical structure parameter values as it stated in §3.3.1(1) that the derived
at a limit state. Lv is the vertical dimension of the values of soil parameters “shall be interpreted in a man-
zone of influence (= vertical part of length of failure ner appropriate to the limit state being considered”.
surface or vertical zone where relevant settlements
occur, etc.). The horizontal length of influence Lh can
6.2. Further proposal for characteristic value
reasonably and conservatively be neglected for typical
horizontal zones of influences less than about 10– Equation (13) requires the designer to determine or esti-
20 m. It is remarked that the expression 1/Lv is dimen- mate the extreme parameter value Xextr from an actual or
sionless because 1 stands for a typical vertical SOF of hypothetical large number of test results. This requires
1 m. a = a factor to account for extent and quality of the designer to use judgement in order to estimate
field and laboratory investigations or estimation Xextr. To avoid having to use this subjective judgement,
method, type of tests for selecting derived values, the author proposes that, assuming a normal distribution
12 T. L. L. ORR

as used by Cherubini and Orr (1999), the extreme value of this definition remained unchanged in the EN version
is at a distance of 3SD from the mean; such a value cor- of Eurocode 7 published in 2004. In designs based on
responds to the 0.2% fractile and hence is the effective Eurocode 7, characteristic values are combined with pre-
limit of the Gaussian curve. Hence Xmean–Xextr = 3SD. scribed partial factors to achieve target reliability values
Substituting this in Equation (13) gives: set out in EN 1990. However the definition of the charac-
 teristic value in Eurocode 7 is purposely not prescriptive,
1
Xk = Xmean − a3SD . (14) except that characteristic values shall be based on values
Lv derived from test results, certain factors need to be taken
Hence: into account, including the number of tests and the
 extent of the failure surface, and also well-established
1 experience. The option of using a statistical approach
Xk = Xmean − a3Xmean COV , (15)
Lv to select the characteristic value is offered in Eurocode
The correlation factor is therefore: 7, but is not encouraged, except in the case of pile design
 where ξ values, referred to as correlation factors, are

1 given to evaluate the characteristic resistance. Since the
j = 1/ 1 − 3a COV . (16)
Lv present definition of the characteristic value provides lit-
tle specific guidance but involves subjective judgement, it
Using the pile example shown in Figure 1, and assuming can result in different designers selecting very different
good quality test values and ground investigation, a = characteristic values from the same data for a particular
0.5. The length of the pile shaft is Lv = 12 m and since design situation resulting in designs with very different
the COV of the data along the pile shaft is 0.29, the ξ reliabilities. A number of statistical methods are available
value calculated using Equation (16) is 1.14 compared to select the characteristic value but are not included in
to 1.17 calculated using Equation (6) and 1.11 as selected Eurocode 7. A new statistically based equation is being
by judgement in Figure 1. Similarly using the data points proposed for the next version of Eurocode 7 that is
in the zone around the base, assumed to be a vertical dis- simple, easy to understand and addresses objectively
tance of 3 m, where the COV of the data = 0.41, the ξ some of the uncertainties in the selection of characteristic
value calculated using Equation 16 is 1.55 compared to geotechnical parameters. The author has proposed a
1.76 calculated using Equation (6) and 1.67 as selected further simplification of this equation involving less sub-
by judgement in Figure 1. The advantage of Equation jective judgement so that selection of characteristic
(15) compared to Equation (13) is that the designer values is more objective and consistent and so that the
does not have to determine or estimate the extreme par- target reliability is more likely to be achieved for geotech-
ameter value and only needs to assess the quality of the nical designs. It should be noted that the author’s
investigations and tests and, if sufficient data are not equation and the other equations for selecting the
available, select a suitable COV value, for example characteristic value assume that the soil layering is
based on published values. Hence Equation (15) enables known and there is continuous spatial variability. The
the characteristic value to be selected using a method that statistical approaches in this paper should not be used
is easy to understand, involves less subjective judgement where there are, or may be, discrete changes, such as
and has been shown to give values similar to those thin, low strength layers, in the soil. Another aspect on
obtained using Equation (6) for the 95% confidence in which guidance is required when selecting the character-
the mean value. istic value, but has not been considered in this paper, is
how to take account of soil parameters that vary with
depth. This is an aspect that merits future research.
7. Conclusions
Prior to the introduction of Eurocode 7, the codes of
practice and standards for geotechnical design provided Acknowledgement
either no or very little guidance on the selection of par-
ameters values for use in geotechnical design. With the The author is very grateful for advice and comments received
from Dr Kenji Matsui and Dr Stephen Buttling during the
introduction of the limit state method in conjunction preparation of this paper.
with partial factors, a definition for the characteristic
value of a geotechnical parameter as the value selected
as a cautious estimate of the value affecting the occur-
rence of the limit state was introduced in the ENV ver- Disclosure statement
sion of Eurocode 7 published in 1997 and the wording No potential conflict of interest was reported by the author.
 GEORISK 13

References Guide to EN 1997-1 Eurocode 7: Geotechnical Design –

General Rules. London: Thomas Telford.
Ad Hoc Committee. 1987. “Draft Model for Eurocode 7.” JGS. 2004. (JGS, 2004) Principles for Foundation Design
Report Prepared for the European Communities by Grounded on a Performance-based Design Concept,
Representatives of the Geotechnical Societies within the (GeoGuide). Tokyo: Japanese Geotechnical Society.
European Communities. Unpublished. Orr, T. L. L. 2008. Story of Eurocode 7, Spirit of Krebs Ovesen
Bauduin, C. 2002. “Design of Axially Loaded Piles According Session – Challenges in Geotechnical Engineering, dgf
to Eurocode 7.” Paper presented at the proceedings 5th Bulletin 23. Danish Geotechnical Society, Lyngby.
international conference on Piling and Deep Foundations, Orr, T. L. L. 2012. “Codes and Standards and their Relevance.”
DFI, Nice, Presses de L’ENPC, Paris, June. In Manual of Geotechnical Engineering, Institution of Civil
Bond, A., and A. Harris. 2008. Decoding Eurocode 7. London: Engineers, Vol. 1, edited by J. Burland, W. Powrie, and T.
Taylor and Francis. Chapman, Chapter 10, 105–124. London: ICE Publishing.
BSI. 1972. CP2004: Code of Practice for Foundations. London: Orr, T. L. L. 2015. “Managing Risk and Achieving Reliable
British Standards Institution. Geotechnical Designs Using Eurocode 7.” In Risk and
CEN. 1997. ENV 1997-1:1997: Eurocode 7 - Geotechnical Reliability in Geotechnical Engineering, edited by K. K. Phoon
Design – Part 1: General Rules. Brussels: European and J. Ching, Chapter 10, 395–433. Boca Raton: CRC Press.
Committee for Standardization. Phoon, K-K., and F. H. Kulhawy. 1999. “Characterization of
CEN. 2002. EN 1990: Eurocode - Basis of Structural Design. Geotechnical Variability.” Canadian Geotechnical Journal,
Brussels: European Committee for Standardization. Canadian Science Publishing 36 (4): 612–624.
CEN. 2004. EN 1997-1:2004: Eurocode 7: Geotechnical Design – Schneider, H. R. 1997. “Definition and Characterization of Soil
Part 1: General Rules. Brussels: European Committee for Properties.” Paper presented at the proceedings XIV inter-
Standardization. national conference on soil mechanics and geotechnical
Cherubini, C., and T. L. L. Orr. 1999. “Considerations on the engineering, Hamburg, Balkema, Rotterdam, September.
Applicability of Semi-Probabilistic Bayesian Methods to Schneider, H. R., and M. A. Schneider. 2013. “Dealing with
Geotechnical Design.” Paper presented at the proceedings Uncertainties in EC7 with Emphasis on Determination of
XX Convegno Nazionale di Geotecnica, Parma, Characteristic Soil Properties.” In Modern Geotechnical
Associazione Geotechnica Italiana, September, 4221–426. Design Codes of Practice, edited by P. Arnold, G. A.
CIRIA. 1977. CIRIA Report 63: Rationalisation of Safety and Fenton, M. A. Hicks, T. Schweckendiek, and B. Simpson,
Serviceability Factors in Structural Codes. Construction 87–101. Amsterdam: IOS Press.
Industry and Research and Information Association, Shirato, M., J. Fukui, and K. Matsui. 2002. “Present Situation
London, ISBN: 0-86017-017-9. regarding Ground Investigations and the Determination
Denver, H., and N. K. Ovesen. 1994. “Assessment of of Geotechnical Parameters in Japan.” In Paper Presented
Characteristic Values of Soil Parameters for Design.” In at the Proceedings of Foundation Design Codes and Soil
Paper Presented at the Proceedings XIII International Investigation in View of Harmonization and Performance
Conference on Soil Mechanics and Foundation based Design, Kamakura, Japan, 237–243. Lisse, The
Engineering, New Delhi, 5-10 January 1994, Vol. 1, 437– Netherlands: A. A. Balkema.
440. Rotterdam: A.A. Balkema. Simpson, B., J. W. Pappin, and D. D. Croft. 1981. “An
DIF. 1984. Code of Practice for Foundation Engineering. Approach to Limit State Calculations in Geotechnics.”
Translated Edition December 1984, Dansk Ground Engineering, 14 (5): 21–28.
Ingeniørforening, ISBN 87-571-0829-3 Vanmarcke, E. H. 1977. “Probabilistic Modeling of Soil
EG11 2015. “Final Report of Evolution Group 11: Profiles.” ASCE Journal of the Geotechnical Engineering
Characterization of CEN TC250/SC7.” Unpublished. Division 103 (11): 1227–1246.
Frank, R., C. Bauduin, R. Driscoll, M. Kavvadas, N. Krebs Vanmarcke, E. H. 1983. Random Fields: Analysis and
Ovesen, T. Orr, and B. Schuppener. 2004. Designers’ Synthesis. Cambridge, MA: MIT Press.