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

On characteristic values and the reliability-based

assessment of dykes

Michael A. Hicks, Divya Varkey, Abraham P. van den Eijnden, Tom de Gast &
Philip J. Vardon

To cite this article: Michael A. Hicks, Divya Varkey, Abraham P. van den Eijnden, Tom de Gast
& Philip J. Vardon (2019) On characteristic values and the reliability-based assessment of dykes,
Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 13:4,
313-319, DOI: 10.1080/17499518.2019.1652918

To link to this article: https://doi.org/10.1080/17499518.2019.1652918

© 2019 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group

Published online: 18 Aug 2019.

Submit your article to this journal

Article views: 1145

View related articles

View Crossmark data

Citing articles: 9 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ngrk20
GEORISK
2019, VOL. 13, NO. 4, 313–319
https://doi.org/10.1080/17499518.2019.1652918

On characteristic values and the reliability-based assessment of dykes

Michael A. Hicks, Divya Varkey, Abraham P. van den Eijnden, Tom de Gast and Philip J. Vardon
Section of Geo-Engineering, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, The Netherlands

ABSTRACT ARTICLE HISTORY

A case study involving the assessment and re-design of an existing dyke, founded on a layered soil, Received 30 November 2018
has compared deterministic analysis based on 5-percentile property values and a reliability-based Accepted 4 August 2019
random finite element analysis consistent with the requirements of Eurocode 7. The results show
KEYWORDS
that a consideration of the spatial nature of soil variability generally leads to higher computed Characteristic values; dykes;
factors of safety and, for those dyke sections requiring remedial action, to more economic Eurocode 7; slope reliability;
designs. Back-figured characteristic values are shown to be considerably higher than the 5- spatial variability
percentile soil properties; hence, a reduction in over-conservatism is achieved.

1. Introduction Hicks (2012) and Hicks and Nuttall (2012). In particular,

they highlighted Clause (11), which gives guidelines for
Around 1 billion euros per year are required to maintain
when statistical methods are used (see Table 1). It infers
and upgrade the Dutch dyke network, which protects
that characteristic values should be selected so as to give
around 40% of the Netherlands from inundation. This
a structure reliability (relative to the limit state) of at
includes 14,000 km of rural dykes, which are currently
least 95%. Although this appears to be contradicted by
maintained and upgraded using rules mainly derived
the two parts of the footnote, the first part of which refers
from research on primary dykes (a very different type of
to mean values and the second part which refers to the soil
structure). The current strategy for determining when
property distribution, it was demonstrated that the clause
maintenance and/or upgrading are needed is based on
and both parts of the footnote are entirely consistent, and
assessment using partial factors and reliability-based
explained by a consideration of the scale of fluctuation θ
characteristic values derived only from the point statistics
(the distance over which soil properties are significantly
of the material properties. This paper reports a recent
correlated) and the size of the problem domain D (e.g.
reliability-based assessment of a dyke ring in the west of
the extent of the failure surface), relative to a modified
the Netherlands, based on statistics derived from labora-
“effective” property distribution governing the limit
tory and site investigation data. In particular, for a selected
state. In Figure 1, which, for simplicity, shows a single
dyke cross-section, deterministic solutions for the factor
soil property X represented by a normal distribution, 3
of safety are compared with probability distributions of
scenarios are possible for the characteristic value Xk:
factor of safety based on reliability analyses using (a)
only the point statistics, and (b) random fields.
(1) For very small values of θ/D, there is considerable
averaging of soil property values over the potential
2. Background
failure surface. This leads to a narrow “effective”
Dutch stability assessments of rural dykes are based on property distribution centred about the mean (Xm)
the Eurocode 7 (EC7) philosophy of partial factors and of the underlying distribution. In this case, the 5 per-
characteristic values of soil properties, in which the par- centile of the modified distribution represents a cau-
tial factors are defined by the code and the characteristic tious estimate of the mean (cf. part 1 of footnote);
values are chosen by the engineer (CEN 2004). In par- (2) For very large values of θ/D, failures tend to be local
ticular, they adopt a statistical approach to deriving and there is a large range of possible solutions. This
characteristic values. leads to the “effective” distribution tending towards
Extracts from Section 2.4.5.2 of EC7, “Characteristic the underlying distribution, from which the charac-
values of geotechnical parameters”, were reviewed by teristic value is the 5 percentile (cf. part 2 of footnote).

CONTACT Michael A. Hicks [email protected]

© 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License (http://creativecommons.org/licenses/by-nc-nd/
4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in
any way.
314 M. A. HICKS ET AL.

Table 1. Clause (11): Extract from Section 2.4.5.2 of Eurocode 7 However, for reasons of simplicity, engineering practice
(CEN 2004). often uses the 5 percentile of the underlying distribution
(11) If statistical methods are used, the characteristic value should be as the characteristic value, regardless of the value of θ/D
derived such that the calculated probability of a worse value governing
the occurrence of the limit state under consideration is not greater or the geotechnical application. The implications of this
than 5%. simplification are demonstrated below, through use of a
NOTE: In this respect, a cautious estimate of the mean value is a
selection of the mean value of the limited set of geotechnical reliability-based random finite element approach con-
parameter values, with a confidence level of 95%; where local failure is sistent with the requirements of Eurocode 7.
concerned, a cautious estimate of the low value is a 5% fractile.

(3) For intermediate values of θ/D (i.e. the usual scen- 3. Case history
ario), Xk is problem-dependent and there are 2 com-
The Starnmeer polder is situated in the province of
peting factors: (a) the averaging of soil properties
North Holland and is managed by the water board Hoo-
over the potential failure surface leads to a narrower
gheemraadschap Hollands Noorderkwartier (HHNK). It
“effective” property distribution; (b) the tendency
was originally drained in 1643, covers an area of 580 hec-
for failure to be attracted to semi-continuous weaker
tares, and is contained within a ring dyke of around
zones leads to a reduced mean (Xm*) relative to the
13 km in length. Recently, HHNK initiated a stability
underlying distribution.
assessment of the dyke. This was performed by dividing
the dyke into 10 sections and, for each section, the factor
of safety (F ) against slope failure was determined for a
Note that Scenario 3, as illustrated in Figure 1, is the representative cross-section using the limit equilibrium
general case, whereas Scenarios 1 and 2 (not shown) software D-Geo Stability (Deltares 2018). This revealed
are limiting cases. Moreover, although the mean is that 5 of the 10 sections do not comply with current
reduced in Scenario 3, because the modified distribution safety requirements. Indeed, not only did they return fac-
is narrower than the underlying distribution, the 5-per- tors of safety below the required F; in some cases, factors
centile characteristic value is generally greater than in of safety as low as 0.5 were reported even though the
Scenario 2; that is, relative to the underlying distribution, dyke has remained stable for hundreds of years.
Xk corresponds to a percentile greater than 5%. In this paper, the authors investigate the assumptions
Various approaches have been proposed for selecting made in analysing a dyke cross-section which returned a
the characteristic values of soil properties; for example, factor of safety of 0.59 based on design property values.
as reported by Orr (2017) and Shen et al. (2018). Figure 2 shows that the 3.8 m high dyke is composed

Figure 1. Derivation of characteristic property value satisfying EC7: underlying distribution of X, and “effective” distribution accounting
for influence of spatial correlation and problem being analysed.
 GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS 315

Figure 2. Dyke cross-section (scale in metres).

Table 2. Unit weights and shear strength parameter values used in analysis of dyke cross-section. (Layers 1–6 refer to Figure 2; layers 7–
8 refer to Figure 5.)
c′ tan f′
γ (kN/ Mean 5-percentile value Partial Design value 5-percentile Partial Design
Layer m3) (kPa) (kPa) COV factor (kPa) Mean value COV factor value
(a) Layers 1–6
1 13.9* 4.4 1.1 0.773 1.20 0.917 0.580 0.506 0.081 1.15 0.429
2 9.8 3.2 1.0 0.656 1.20 0.833 0.398 0.361 0.058 1.15 0.310
3 9.9 2.0 0.5 0.775 1.20 0.417 0.358 0.279 0.145 1.15 0.241
4 15.0 4.5 1.7 0.544 1.20 1.417 0.559 0.547 0.012 1.15 0.465
5 15.0 5.4 2.9 0.352 1.20 2.417 0.601 0.594 0.007 1.15 0.503
6 20.0 0.0 0.0 0.000 – 0.000 0.637 0.637 0.000 1.20 0.531
(b) Layers 7–8
7 17.0 6.2 1.6 0.773 1.20 1.333 0.531 0.463 0.081 1.15 0.403
8 20.0 0.0 0.0 0.000 – 0.000 0.637 0.637 0.000 1.20 0.531
*γ = 6.9 kN/m3 above phreatic surface.

of clay, and is founded on a peat layer underlain by a thin Table 3. Factors of safety F for dyke cross-section based on
clay layer and a thick sand layer. Table 2(a) summarises deterministic and stochastic analyses.
the unit weights and shear strength properties used in the Deterministic analysis Stochastic analysis
original assessment, based on the results of extensive lab- F corresponding to CDF of 0.05
Property without partial with partial
oratory (triaxial and direct simple shear) tests on soils values F θh (m) factors factors
from Starnmeer (Kames 2015). In this table, the mean Mean 1.31 0.5 1.10 –
and 5-percentile values for the cohesion (c′ ) and tangent 5-percentile 0.66 6.0 0.98 0.85
–
of the friction angle (tan f′ ), for each material zone indi- Design 0.54 12.0 0.98

cated in Figure 2, are reported, as well as the respective

partial factors and design property values used in the
stability analysis (in which the design value is equal to have been obtained using an in-house finite element soft-
the characteristic value divided by the partial factor). ware, developed at TU Delft, that computes the factor of
Also shown in the table are the coefficients of variation safety using the strength reduction method, and they are
(COV = standard deviation/mean) of c′ and tan f′ , based on the same cross-sectional geometry and material
which have been back-figured from the respective properties used previously. Moreover, the authors have
mean and 5-percentile values assuming a lognormal dis- assumed the same external water level and phreatic sur-
tribution, and are on the conservative (high) side due to faces (represented by the blue lines in Figure 2) as in the
soil samples coming from the Starnmeer area as a whole original assessment, in which the higher phreatic surface
rather than from the specific cross-section being ana- relates to all soil layers except for layer 6, for which the
lysed. Note that no test results were reported for the bot- lower phreatic surface is used. Figure 3 illustrates the sig-
tom (sand) layer, and that the 5-percentile value of tan f′ nificance of the underlying peat layer, by showing the
adopted for this layer is the value suggested by NEN computed failure mechanism based on homogenous
9997-1 (2011) for a moderately packed sand. soils.
Firstly, Table 3 lists the deterministic factors of safety
obtained using the mean, 5-percentile and design prop-
erty values for the different material zones (from Table
3.1. Re-analysis of dyke stability
2(a)). Based on the design properties, F = 0.54, which
Table 3 summarises the results of a re-evaluation of the compares favourably with the D-Geo Stability solution
stability of the dyke section by the authors. These results of 0.59, as well as with an F of 0.56 obtained by the
316 M. A. HICKS ET AL.

Figure 3. Plastic shear strain contours at failure based on homogeneous soil layers.

authors using the commercial finite element code material zone, and the factor of safety of the dyke then
PLAXIS. Each of these solutions takes account of the computed using the strength reduction method. This
uncertainty in the design property values by basing gives 500 factors of safety, from which a cumulative dis-
them on characteristic values representing the 5 percen- tribution function (cdf) of F can be plotted.
tile of the property distribution; that is, by adopting the Figure 4 shows the cdf of F computed using RFEM for
approach called Scenario 2 in Section 2. However, as dis- each value of θh (as a solid curve), based on the soil prop-
cussed, this is not consistent with the intention of EC7, as erty statistics given in Table 2(a). Also indicated in the
illustrated in Figure 1, in that the characteristic values figure are the factors of safety obtained from determinis-
take no account of the spatial nature of the soil variability tic analyses based only on the mean, median and 5-per-
nor of the problem being analysed. centile values, as well as that obtained based on the
Hence, Table 3 also shows stochastic results account- design property values. The cdf of F from a stochastic
ing for the spatial variability of soil property values analysis based only on the point statistics (i.e. with no
within the material zones. These have been computed spatial averaging) is included, to highlight the signifi-
with the same in-house finite element software, but cance of spatial averaging in the RFEM analyses.
now implemented within a Monte Carlo simulation in Figure 4 shows that for a reliability (R) of 95%, a con-
which each realisation uses different random fields of servative estimate of F = 0.98 is obtained when θh =
soil property values for each material zone, a procedure 6.0 m. In order to determine the value of F correspond-
often referred to as the random finite element method ing to the design property values, for each material zone
(RFEM) (Fenton and Griffiths 2008). The random the property distribution for c′ has been scaled down by a
fields have here been generated by covariance matrix partial factor of 1.20 and the property distribution for tan
decomposition using a Markov autocorrelation function; f′ has been scaled down by a partial factor of 1.15 (or
see van den Eijnden and Hicks (2017) for details. The 1.20, in the case of the sand layer). These “design” prop-
RFEM process uses the same point statistics as listed in erty distributions have then been used in a further RFEM
Table 2, but additionally, for each soil property and analysis (with θh = 6.0 m), to give a new cdf (shown as a
each material zone, vertical and horizontal scales of broken curve) and a value of F = 0.85 corresponding to R
fluctuation are specified to quantify the distance over = 95% (Table 3). This value represents a significant
which property values are significantly correlated. As (57%) increase in F when accounting for the spatial
insufficient data are available for the cross-section, the nature of the soil variability, although, as it is still less
vertical scale of fluctuation (θv) has been taken as than the F = 0.95 safety requirement (based on the
0.5 m for all properties and all material zones. This is a IPO-class, i.e. design class, this dyke section belongs to
conservative (high) estimate based on a range of 0.2– (Kames 2015)), some upgrading of the dyke section is
0.5 m reported by de Gast, Vardon, and Hicks (2017) needed.
for similar soils found at the Leendert de Boerspolder
site in South Holland. Three values for the horizontal
3.2. Re-design of the dyke section
scale of fluctuation (θh) have initially been considered;
0.5, 6.0 and 12.0 m, to investigate the sensitivity of the Figure 5 shows an initial proposal for the re-design of the
solution to this statistical measure. For each value of dyke section, following on from the original stability
θh, an RFEM analysis involving 500 realisations has assessment of F = 0.59 (using D-Geo Stability). This
been conducted, in which, for each realisation, the involves moving the ditch further away, infilling the orig-
point and spatial statistics have been used to generate inal ditch with sand, and constructing a clay berm over
uncorrelated random fields of c′ and tan f′ for each the sloping face to increase the resistance against failure.
 GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS 317

Figure 4. Comparison of deterministic and stochastic solutions for factor of safety.

Figure 5. Initial re-design for dyke cross-section.

The unit weights and shear strength properties for the Table 4 shows the results of further RFEM analyses,
sand infill and clay fill are summarised in Table 2(b). corresponding to a range of berm heights and berm
This led to increased deterministic factors of safety, widths (as quantified by the distance between old and
based on the design property values, of F = 1.33 using new ditches), see Figure 5. These results show how F cor-
D-Geo Stability and F = 1.21 using the in-house software. responding to R = 95%, with and without partial factors,
However, an RFEM analysis based on the design prop- varies as a function of the berm geometry. In particular,
erty distributions, θv = 0.5 m, and θh = 6.0 m, for the it highlights how a berm with a height of H/2 and inter-
cross-section in Figure 5, gave F = 1.531 for R = 95%, ditch spacing of W/3 gives a factor of safety (with partial
an increase of 27% relative to the deterministic in- factors) satisfying the safety requirement (i.e. F = 1.015 >
house solution. 0.95). This represents a significant saving relative to the

Table 4. Factors of safety corresponding to R = 95% for various berm designs.

H 2H/3 3H/5 H/2 0H
without partial without partial without partial with partial without partial with partial without partial with partial
factors factors factors factors factors factors factors factors
W 1.789 1.461 1.268 1.083 1.197 1.027 0.968 0.826
2W/3 1.736 1.377 1.265 1.080 1.193 1.021 – –
W/2 1.724 1.375 1.259 1.079 1.186 1.016 – –
W/3 1.647 1.360 1.249 1.071 1.181 1.015 – –
318 M. A. HICKS ET AL.

original re-design (Figure 5), both in terms of volume of Specifically, a single characteristic percentile has been
fill required and impact on neighbouring property. back-figured, which, when applied to the distributions of
c′ and tan f′ for each material zone, gives characteristic
values that return the correct factor of safety for R = 95%.
3.3. Characteristic values This percentile has been determined by conducting mul-
tiple deterministic analyses, in which, for any given
The above analysis and re-design of the dyke section realisation, the shear strength parameters for all material
using RFEM is fully consistent with EC7, in that it is zones are sampled from the same percentile location in
based on characteristic soil property values giving a the respective property distributions (i.e. each material
95% reliability of the structure, factored down by the zone is treated as homogeneous). Thus, in realisation r,
required partial factors. Note that, even though the the input (Xi)r for a parameter Xi (i.e. either c′ or tan
characteristic soil properties have not been calculated f′ ) is calculated using
explicitly during the analyses (i.e. the 5 percentile of
the “effective” distribution), it is the reliability-based fac- (Xi )r = exp (mln X i + sln X i × kr ) (1)
tor of safety that is needed in the safety assessment.
Moreover, calculating characteristic values for a problem where mln X i and sln X i are the mean and standard devi-
ation of the natural logarithm of Xi, respectively, and kr is
in which there are two soil properties and multiple soil
layers is not straightforward. In contrast to the simple the standard score, computed using
illustration given in Figure 1, in which the characteristic kr = F−1 (r/N) (2)
property is a single value, for this dyke section the
characteristic values for each material zone are rep- where Φ−1 is the inverse of the standard normal cumu-
resented by a surface in c′ –tan f′ space; in other lative distribution function and N is the total number
words, there are many combinations of c′ and tan f′ of realisations. Figure 6 compares the cdf of F obtained
that give the same reliability for the structure. Nonethe- using this approach with the cdf of F obtained using
less, it is informative to back-calculate percentiles (of the RFEM with θv = 0.5 m and θh = 6.0 m, for the original
underlying property distributions) representing the dyke cross-section in Figure 2. For F = 0.98, correspond-
characteristic values and, for illustrative purposes, a ing to R = 95% in the RFEM analysis, the value of r/N is
simple approach has here been adopted. 0.34. Hence, for this particular dyke section and loading

Figure 6. Comparison of factor of safety distribution obtained using RFEM with deterministic analyses based on same distribution
percentile.
 GEORISK: ASSESSMENT AND MANAGEMENT OF RISK FOR ENGINEERED SYSTEMS AND GEOHAZARDS 319

conditions, the characteristic percentile that may be used Funding

for both the c′ and tan f′ distributions (for all material This work is part of the research programme Reliable Dykes
zones) is 34%. with project number 13864 which is financed by the Neder-
Note that no correlation has been assumed between c′ landse Organisatie voor Wetenschappelijk Onderzoek
and tan f′ in this research, although previous studies (NWO), and was carried out on the Dutch National e-infra-
have mainly suggested a negative correlation between structure with the support of SURF Foundation.
these two parameters, which would result in a narrower
cdf of F (Vardon, Liu, and Hicks 2016) and thereby to a
higher characteristic percentile. Thus, the characteristic References
percentile of 34% for this particular dyke section is likely CEN. 2004. Eurocode 7: Geotechnical Design. Part 1: General
to be a conservative estimate. Rules, EN 1997-1. Brussels: European Committee for
Standardisation.
de Gast, T., P. J. Vardon, and M. A. Hicks. 2017. “Estimating
4. Conclusions Spatial Correlations Under Man-Made Structures on Soft
Soils.” Proceedings of 6th International Symposium on
A comparison has been made between using a determi- Geotechnical Safety and Risk, Denver, Colorado, 382–389.
nistic assessment method and the random finite element Deltares. 2018. D-Geo Stability. https://www.deltares.nl/en/
method to assess the stability and re-design of an historic software/d-geo-stability-2/.
dyke in the Netherlands, based on a reliability-based fra- Fenton, G. A., and D. V. Griffiths. 2008. Risk Assessment in
Geotechnical Engineering. New York: John Wiley and Sons.
mework consistent with Eurocode 7. It has been shown
Hicks, M. A. 2012. “An Explanation of Characteristic Values of
that a proper consideration of spatial variability, such Soil Properties in Eurocode 7.” In Modern Geotechnical
as with the random finite element method, can lead to Design Codes of Practice: Development, Calibration and
higher factors of safety and, for those structures requir- Experiences, edited by P. Arnold, G. A. Fenton, M. A.
ing attention, to less costly and less intrusive mitigation Hicks, T. Schweckendiek, and B. Simpson, 36–45.
measures. The advantage of the proposed approach is Amsterdam: IOS Press.
Hicks, M. A., and J. D. Nuttall. 2012. “Influence of Soil
that it satisfies the requirements of Eurocode 7 without Heterogeneity on Geotechnical Performance and
the need to explicitly select or calculate the characteristic Uncertainty: a Stochastic View on EC7.” Proceedings of
property values. Nevertheless, for the particular dyke sec- 10th International Probabilistic Workshop, Stuttgart,
tion analysed in this paper, and for illustrative purposes Germany, 215–227.
only, characteristic soil property values consistent with Kames, J. 2015. Veiligheidstoets Boezemkaden. Technical
Report 14.0046944, Hoogheemraadschap Hollands
Eurocode 7 were back-calculated and found to be the
Noorderkwartier.
34 percentiles of the respective property distributions. NEN 9997-1. 2011. Geotechnical Design of Structures – Part 1:
This represents a significant increase in strength capacity General Rules, Dutch Standard (Eurocode 7).
over simpler interpretations of Eurocode 7 based only on Orr, T. L. L. 2017. “Defining and Selecting Characteristic
the point statistics. However, given the problem-depen- Values of Geotechnical Parameters for Designs to
dent nature of characteristic values, as well as the desire Eurocode 7.” Georisk: Assessment and Management of
Risk for Engineered Systems and Geohazards 11 (1): 103–
for simpler validated approaches amenable to practice, 115.
further studies are recommended for a more general Shen, M., S. Khoshnevisan, X. Tan, Y. Zhang, and C. H. Juang.
insight. 2018. “Assessing Characteristic Value Selection Methods for
Design with LRFD – A Design Robustness Perspective.”
Canadian Geotechnical Journal, 1–11. doi:10.1139/cgj-
Acknowledgements 2018-0038.
van den Eijnden, A. P., and M. A. Hicks. 2017. “Efficient Subset
The authors are grateful for the assistance of Henk van Hemert
Simulation for Evaluating the Modes of Improbable Slope
(STOWA) and Niels Tenhage (Hoogheemraadschap Hollands
Failure.” Computers and Geotechnics 88: 267–280.
Noorderkwartier).
Vardon, P. J., K. Liu, and M. A. Hicks. 2016. “Reduction of
Slope Stability Uncertainty Based on Hydraulic
Disclosure statement Measurement via Inverse Analysis.” Georisk: Assessment
and Management of Risk for Engineered Systems and
No potential conflict of interest was reported by the authors. Geohazards 10 (3): 223–240.