You are on page 1of 12

OTC-25160-MS

Estimation of Temporal Probability in Offshore Geohazards Assessment


Farrokh Nadim, and Suzanne Lacasse, Norwegian Geotechnical Institute (NGI); Young Jae Choi, NGI Inc.;
Christopher Hadley, Shell Global Solutions US Inc.

Copyright 2014, Offshore Technology Conference

This paper was prepared for presentation at the Offshore Technology Conference held in Houston, Texas, USA, 5– 8 May 2014.

This paper was selected for presentation by an OTC program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Offshore Technology Conference and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Offshore Technology Conference, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the
written consent of the Offshore Technology Conference is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words;
illustrations may not be copied. The abstract must contain conspicuous acknowledgment of OTC copyright.

Abstract
Risk assessment for sea floor installations exposed to offshore geohazards requires an estimation of the
occurrence probability of a hazardous event during a reference time period, for example the annual
probability of the installation being impacted by a submarine mass gravity flow. The paper describes the
procedures developed in recent offshore geohazards studies for the estimation of the temporal probability
in two situations: 1) When there is a clear trigger for initiating the slide that would develop into a mass
gravity flow and 2) When there is evidence of slide activity, but no obvious trigger for slide initiation. In
the first situation, the assessment of temporal probability requires a probabilistic description of the
frequency and intensity of the trigger(s) releasing the submarine slide, a probabilistic model for calcu-
lating the response of the slope to the trigger, and a probabilistic model for evaluating the runout of the
released mass gravity flow. Using these models, the probability of a mass gravity flow impacting the
seafloor installation(s) should then be computed for all relevant scenarios and return periods in order to
derive the annual or lifetime probability. However, analyzing all possible scenarios and return periods
could be time-consuming and impractical. The paper presents a simplified procedure and demonstrates its
application through a case study for earthquake-triggered slides. In the second situation, where it is
difficult to identify the trigger(s) initiating the slide, one must rely on the identification and dating of
recent (in the geological sense) slide events in the area. The dating results and other relevant geological
evidence are then integrated into a Bayesian framework to establish the annual probability of slope
instability. The paper also presents the theoretical framework and example calculations for this situation.

Introduction
The exploitation of offshore resources, the development of transportation corridors, the preservation of
fishing habitat and the protection of coastal communities contribute to a growing interest in the
quantification of the risk posed by offshore geohazards. The offshore petroleum industry develops oil and
gas fields in ever-increasing water depths and ensuring adequate reliability is a necessary consideration
for seabed installations. The greatest threat to these installations is posed by seafloor mass movements,
which makes the assessment of their probability of occurrence and subsequent consequences a necessity.
Submarine slides can be the result of natural on-going processes or can be triggered by human activities
or external processes. Assessing the hazard of landslides is a challenge and requires information about the
2 OTC-25160-MS

Figure 1—Triggering mechanisms for 534 submarine slides(Hance 2003).

frequency of the sliding events, the soil conditions, the potential triggering factors and their timing or
frequency. The potential for reactivation should also be considered in areas where slope failures have
taken place earlier. Earthquakes have triggered major submarine mass movements all over the world (NGI
1997; Hance 2003). During one single event, enormous sediment volumes can be transported on very
gentle slopes over distances exceeding hundreds of kilometers. Hance (2003) reported that most of the
submarine slides can have multiple triggers. Among the 534 events in his database, over 40 % were
attributed to earthquake and faulting mechanisms (Fig. 1).
The first step in the assessment of the risk posed by mass gravity flow events on sea floor installations
is hazard analysis. The aim of the hazard analysis is to estimate the annual probability for critical facilities
on the seafloor to be impacted by mass gravity flow event(s). To assess the hazard quantitatively, the
problem is split into two sub-problems (see Fig. 2): (a) estimate the annual probability of a slide being
mobilized at a critical location on the escarpment, and b) once the slide has been mobilized, estimate the
probability of the resulting mass gravity flow to reach a certain location. The annual probability of interest
(or hazard) is the product of these two probabilities. This paper focuses on methodologies for answering
the first question.
Analysis approach
In general, two situations may be encountered when one attempts to estimate the annual probability of
submarine slide initiation. In the first situation, there is strong evidence of past sliding activity in the area
of interest, but there are no obvious triggers for the initiation of the slides. In the second situation, it is
straightforward to identify the dominant trigger for slide initiation.
In the first situation, the annual probability for a slope instability can only be estimated from the
geological evidence, e.g. observed slide frequency, geological history, the results of geophysical inves-
tigations and the radiocarbon dating of sediments. When the triggering mechanism is not obvious,
probabilistic slope stability calculations provide an estimate of failure probability for static conditions. It
is not straightforward to relate the calculated “timeless” failure probability to a failure frequency. Nadim
et al (2003) and Lacasse and Nadim (2007) developed several ideas for quantifying the annual probability
of slope instability:
OTC-25160-MS 3

Figure 2—Schematic representation of submarine landslide evolution (modified from Bryn et al 2005).

● Bayesian approach with Bernoulli sequence


● Statistical model for failure frequency
● Interpretation of static failure probability as the instantaneous hazard function
● Interpretation of computed static failure probability in the Bayesian framework
The first two approaches are purely statistical and do not involve any geotechnical calculations. Their
input is the frequency of slide events (or lack thereof), which may be based on observations or inferred
from geological evidence, for example dating of slide sediments. The third approach (Nadim et al 2003
and Lacasse and Nadim 2007) combines the calculated probability of static slope failure with the slide
frequency estimated from the geological evidence.
The last approach forms the starting point for the calculation procedure suggested for the second
situation, where it is straightforward to identify the dominant trigger for slide initiation. For example, there
is high seismic activity in the area, or the slopes of interest are close to a river delta with active
sedimentation. The annual probability of slope instability can be established by evaluating the conditional
probability of failure for different return periods of the trigger. The conditional probabilities are then
integrated over all return periods to obtain the unconditional failure probability. Calculations can be
simplified by using the approximation suggested by Cornell (1996) or a similar approach. The hazard
analysis (i.e. estimation of the annual probability of slide initiation) involves the following main steps:
● Identification of triggering mechanisms for slope instability,
● Probabilistic description of frequency and intensity of trigger(s), and
● Probabilistic assessment of slope stability under the action of trigger(s) for all relevant scenarios
and return periods.
Depending on the level of sophistication and detail of the hazard analysis, each of these steps will
involve a number of sub-tasks, as described in the case study later in the paper. Ideally, the calculated
annual probability for slide initiation in this situation should be compared to the annual probability
estimated on the basis of geological evidence and dating of earlier slide events to make sure that the two
independent approaches give consistent results.
Two case studies are presented below. The first case study presents a statistical model for the
estimation of the annual frequency of slide occurrence on the basis of geological evidence and dating of
sediments. The second case study, with a strong earthquake as the dominant trigger, presents a state-of-
the-art approach for the hazard analysis of slope instability under seismic loading.
4 OTC-25160-MS

Figure 3—Sea floor installations exposed to mass gravity flow impact, Case Study I.

Case Study I: Sigsbee Escarpment in Gulf of Mexico


Figure 3 shows an overview of the area of interest for Case Study I at the Sigsbee Escarpment in the Gulf
of Mexico. Critical sea floor installations might be impacted by mass gravity flows initiating on
Escarpment 2 or Escarpment 3.
Geophysical investigations and general geological knowledge about the area indicated evidence of
sliding activity during the past 100 ka. However, there is very little seismic activity and virtually no active
sedimentation in the area. The annual probability for the initiation of a critical slide was therefore
estimated on the basis of the number of mass gravity flow deposits (MGFD) on the seafloor identified in
the geophysical site investigations and the dating of these deposits.
A simplified Bayesian model with diffuse prior information developed by Ang and Tang (1984; 2007)
was used. Diffuse information means that the annual probability of a slide event initially could be any
value between zero and one, with a uniform distribution. One can then derive the following probability
distribution function for the annual probability of slide occurrence:
(1)

where Pslide is the annual probability of slide occurrence, r is the number of slide events (i.e. MGFDs)
identified on the escarpment of interest, n is the number of years during which observations are made, and
k is a normalizing coefficient. With the probability distribution above, the expected value Pslide on the
escarpment of interest is:
(2)

If no sliding has taken place over the past n years, then the expected value of Pslide (annual) is:
(3)

This Bayesian framework provides a simple estimate of the annual probability of slide occurrence
based on the age constraints derived from the identification and dating of recent slide events in the study
area (from the past 20 to 100 ka).
OTC-25160-MS 5

In the area below Escarpment 2, 17 MGFDs with volumes between 0.06 and 2.1 mill. m3 were
identified in the high-resolution shallow seismic profiles above a layer that was known to have an age of
about 20 ka before present (B.P.). The age estimates of the MGFDs varied between 7.8 ka and 17 ka B.P.
This gives roughly an annual probability of:

for a significant slide event (volume ⬎ 0.06 mill. m3) anywhere along Escarpment 2 during the past
17 ka years. The age estimates further implied that there is no evidence of significant sliding activity
during the past 7 800 years. This negative evidence gives an expected value of:

for the past 8 ka, which is about one order of magnitude lower than the landslide rate estimated for the
period 17 ka to 8 ka B.P. The reason for the low landslide rate during the last 8 ka compared with the
previous period is related to the stable high sea-level stand reached at about 8 ka B.P. The sea level
reached a low-level stand (about 120 to 130 m below the present sea level) during the last glacial
maximum, about 20 ka B.P. The sea level change during the glacial melting period between about 18 ka
and 8 ka B.P. caused significant change in the coastline with massive wave- and current-induced erosion
of the shallow water areas. This generated turbid waters with suspended fines and hyperpycnal flows
leading to high deposition rates both in the shallow and deepwater areas. The termination of the sea level
rise with stable coastline position and water depths at about 8 ka B.P. caused a significant reduction in
sedimentation rate on the shelf, on the Continental Slope and in the deepwater areas, and a corresponding
reduction in landslide activity.
The estimated values above are for a slide anywhere on Escarpment 2. In the forward modelling of a
slide event on Escarpment 2 that might impact the critical sea floor facilities, a specific sliding event will
have an annual probability of occurrence that is 4 to 10 times lower than the values estimated above,
because the specific scenario only affects a fraction of the escarpment. It seems therefore reasonable to
use an annual probability of slide initiation of 1 · 10-5 to 3 · 10-4/year, with a best estimate of 3 · 10-5 /year,
for the forward modelling scenarios on Escarpment 2.
In the area below Escarpment 3, 14 MGFDs with volumes between 0.23 and 11.3 mill. m3 were
identified. All of the MGFDs were estimated to be younger than 19 ka, and the largest one was estimated
to be younger than 12.8 ka. Using Eq. 2, the observations imply roughly an annual probability of:

for a significant slide event (volume ⬎ 0.23 mill. m3) anywhere along Escarpment 3 during the past
19 ka.
Using the same arguments for Escarpment 2, the annual probability of a specific sliding scenario to be
considered in forward modeling of an event on Escarpment 3 will be 4 to 10 times lower than the value
estimated above because the specific scenario only affects a fraction of the whole escarpment. It would
therefore seem reasonable to use an annual probability of slide initiation of 1 · 10-4 /year (range: 8 · 10-5
to 2 · 10-4 /year) for the forward modelling scenarios on Escarpment 3.
Because of the arguments presented earlier for Escarpment 2, it is expected that most of the MGFDs
were also results of slide events that occurred between 19 ka and 8 ka B.P. However, the constraints based
on the age estimates of the MGFDs below Escarpment 3 did not allow an assessment of the change of
sliding activity for the past 8 ka.
It should be noted that the estimated annual probabilities above were for the initiation of a slide event
that could potentially impact the critical facilities. The probability that they will actually impact the
facilities will be (much) lower than these values. The assessment of that probability requires the answer
6 OTC-25160-MS

Figure 4 —Sea floor installations exposed to mass gravity flow impact, and potential paths for mass gravity flows, Case Study II.

to also Question 2 in Figure 2, which is not addressed in this paper. The annual probabilities obtained in
this paper are therefore upper bound values, assuming certainty that the sliding masses will reach the sea
floor installations.
Case Study II: Estimation of annual probability of a seismically-triggered slide
Figure 4 shows an overview of the area of interest for Case Study II in Southeast Asia. In the field
development area, there is a feature referred to as the Bulge, which rises about 120m above the
surrounding terrain. The steepest part is located on the eastern side of the Bulge adjacent to one of the drill
centers and the export pipeline corridor with inclinations of more than 16° and locally approaching 20°.
On the northwestern side of the Bulge, adjacent to another drill center, the steepest slopes have
inclinations exceeding 14°.
The main soils of interest for the stability evaluations are the top clay drape materials, which were
deposited over debris and eroded surface from past slide events in the area. The field development area
has moderate seismic activity and earthquakes are identified as the main triggering mechanism for shallow
slides leading to mass gravity flows.
A procedure for estimating the annual probability of earthquake-induced submarine slides in marine
clays was developed by NGI through a number of joint-industry research projects and offshore geohazards
studies in the North Sea, the Caspian Sea, the Black Sea, offshore Malaysia, and the Gulf of Mexico. The
procedure accounts for the uncertainties in all steps of the assessment and utilizes the available
information to come up with a rational estimate of the annual probability of earthquake-induced slope
OTC-25160-MS 7

failure. The analyses are based on calculated probability of slope failure for the earthquake-induced failure
mechanisms outlined in Nadim et al (2007) using probabilistic slope stability assessment tools.
Nadim et al (2005) and Lacasse and Nadim (2007) described the principles of probabilistic slope
stability analysis with a first-order reliability method (Hasofer and Lind 1974). To do the probabilistic
analyses, one needs a precise mathematical description for the failure state. This is achieved by defining
a performance function G(X), such that G(X) ⱖ 0 means satisfactory performance and G(X) ⬍ 0 means
“failure”. The parameter X is a vector of basic random variables including resistance parameters, load
effects, geometry parameters and modeling uncertainty. The performance function for slope stability can
be described as:
(4)

whereFS is the factor of safety for the slope. If the joint probability density function of all random
variables fX(x) is known, then the probability of failure Pf is given by:
(5)

where L is the domain of X where G(X) ⬍ 0. In general, the above integral cannot be solved
analytically. In the first-order (FORM) approximation, the vector of random variables X is transformed
to the standard normal space U, where U is a vector of independent Gaussian variables with zero mean
and unit standard deviation, and where G(U) is a linear function. The probability of failure Pf is then:
(6)

where P[. . .] means probability that . . .; ␣i is the vector of direction cosines of the random variable
Ui, ␤ is the distance between the origin and the hyperplane G(U) ⫽ 0, n is the number of basic random
variables X, and ⌽ is the standard normal distribution function. The vector of the direction cosines of the
random variables (␣i) is called the vector of sensitivity factors, and ␤ is the (Hasofer- Lind) reliability
index.
The interpretation of the probability of static slope failure computed with FORM or other methods like
the Monte Carlo simulation, is not straightforward. The fact that the slope is standing today implies that
the current factor of safety is greater than one. The annual probability of failure becomes the likelihood
that the current factor of safety will fall below one during next year. The current factor of safety is
unknown, but its distribution can be computed (distribution from FORM analysis, but truncated to reflect
the fact that the slope is stable today). This interpretation is basically a Bayesian updating procedure where
the a-priori information is that FS ⱖ 1. The updated (or posterior) distribution of the factor of safety is:
(7)

The slope will fail during the next year only if its current value of safety factor is such that, with the
given rate of deterioration, it will fall below unity during one year.
Using the probabilistic slope assessment approach outlined above, the different steps of the analyses
are as follows (Lacasse et al 2011):
1. Identify the critical slopes and establish the geometry and mechanical soil properties for the slope
in a probabilistic format.
2. Using Monte Carlo simulation or FORM, establish the cumulative distribution function (CDF) of
the static, undrained safety factor for the slope.
3. Using Eq. 7, update the CDF of the static safety factor.
8 OTC-25160-MS

Figure 5—Results of probabilistic stability analyses prior to (in black) and after 3,000 and 10,000-yr earthquakes (in red), Case Study II.

4. Do a probabilistic seismic hazard assessment for the site of interest and identify representative
acceleration time histories for return periods of interest.
5. Establish a dynamic response model for the slope and do earthquake response analyses for at least
two return periods. Use a Monte Carlo simulation to account for the uncertainties in the soil
properties and earthquake motion characteristics. The main output parameter is the maximum
earthquake-induced shear strain along the potential failure surface.
6. Through a special laboratory testing program or literature survey establish the range of reduction
in the post-earthquake undrained shear strength as function of the maximum earthquake-induced
shear strain.
7. Using the results of Steps 5 and 6, establish the distribution function for the shear strength reduction
factor.
8. Using results of Steps 3 and 7, establish the CDF for the post-earthquake static safety factor. The
conditional probability of failure (given that the earthquake with the specified return period has
happened) is the CDF value at FS equal to 1.
9. The annual failure probability is the sum (integral) of all conditional failure probabilities for a
specific return period, divided by that return period.
10. The analyses above must be done for at least two return periods, one above and one below the
return period that contributes most to the annual failure probability (an iteration may be required).
11. With the results of Step 10, establish a model with load and resistance that matches the computed
failure probabilities at the return periods of interest. The most usual load parameter is the annual
peak ground acceleration (PGA), with typically an exponential or Pareto distribution; if the PGA
is the representative load parameter, the slope resistance should be specified as an acceleration
parameter; a lognormal distribution is then often assumed.
12. Using the simplified analogue in Step 9, estimate the probability that the resistance of the slope is
less that the applied load (e.g. the annual PGA). This value is the estimate of the annual probability
of earthquake-induced slope failure.
The computed and the updated CDF function for the static, undrained safety factor for the critical slope
on the Bulge are shown on Figure 5. The CDF function before updating was estimated numerically using
OTC-25160-MS 9

Figure 6 —Calibrated Pareto distribution of annual PGA Amax, with mean 0.0077g and standard deviation 0.011g, Case Study II (distribution is
calibrated to PSHA results for 1,000- to 10,000-yr return periods).

Figure 7—Dynamic response simulations for 3,000- and 10,000-yr earthquakes, Case Study II.

the FORM approximation. Figure 6 presents the results of the probabilistic seismic hazard study for the
site. A linear relationship in the log-log plot of PGA vs Annual Exceedence Probability implies a Pareto
distribution for the annual PGA at the site. For the example slope considered, the earthquake events with
return periods between 1,000 and 10,000 years contribute the most to the annual failure probability. The
dynamic response analyses were therefore done for earthquake events with return periods of 3,000 and
10,000 years. Each of these events was represented by four (4) sets of properly scaled acceleration time
histories.
Figure 7 shows the simulations of the dynamic response in terms of the maximum earthquake-induced
shear strain and shear strength reduction factor. Figure 8 shows the histograms of the shear strength
reduction factors and a fitted distribution function to the data, as obtained from the simulations for the
10 OTC-25160-MS

Figure 8 —Earthquake-induced undrained shear strength reduction factors, Case Study II.

Table 1—Parameters for best-fit probability distribution of shear strength reduction factors, Case Study II.
Best fit Lower bound Upper bound Mean Standard deviation

Beta 0.72 1.0 0.94 0.04


Beta 0.42 1.0 0.81 0.12

3,000 and 10,000-year earthquakes. The fitted distributions for shear strength reduction factor obtained
from the simulations are listed in Table 1.
Using the shear strength reduction factors from Table 1, the conditional CDF of the post-earthquake
safety factor (FS) can be obtained. The post-earthquake FS is the product of the shear strength reduction
factor and the (corrected) pre-earthquake static FS. The CDFs for the post-earthquake undrained safety
factor following the 3,000-year and 10,000-year earthquake events are also shown on Figure 5.
To estimate the annual probability of slope failure, a simplified model similar to that suggested by
Cornell (1996) was developed. A limit state function with the following format was defined for the
seismic resistance of the slope:
(8)

In the limit state function defined by the above equation, Amax is the annual peak ground acceleration
representing the earthquake load and Aresist is the resistance of the slope to earthquake loading described
in terms of the peak acceleration that would cause slope failure. The parameter ␧ describes the variability
of the peak ground acceleration at a given return period. The resistance parameter Aresist was assumed to
have a lognormal distribution. The parameters of the lognormal probability distribution for Aresist were
calibrated to match the conditional failure probabilities obtained for the 3,000-year and the 10,000-year
earthquake events. Aresist was found to a mean value of ␮ ⫽ 0.22g and standard deviation of ␴ ⫽ 0.11g.
The limit state function in Eq. 8 with the parameters from Table 2 and the FORM approximation were
used to estimate the annual failure probability. The following results were obtained:
Similar to Case Study I, it should be noted that the estimated annual probability above is for the
initiation of a slide event that could potentially impact the critical facilities. The probability that the critical
OTC-25160-MS 11

Table 2—Random variables for the evaluation of annual failure probability in the limit state function (Eq. 8).
Parameter Assumed distribution Mean value Standard deviation

Amax Pareto 0.0077g 0.011g


␧ Normal 1.0 0.12
Aresist Lognormal 0.22g 0.11g

facilities will actually be impacted is less than this value and depends on the results of probabilistic mass
gravity flow run-out analyses.
Summary and Conclusions
A submarine slide might develop into a mass gravity flow and impact the critical sea floor installations.
Estimation of the annual probability of occurrence of a submarine slide is an important consideration in
the geohazards risk assessment for many offshore field development projects. The assessment of temporal
probability for submarine slope failure requires a probabilistic description of the frequency and intensity
of the trigger(s) releasing the submarine slide, a probabilistic model for calculating the response of the
slope to the trigger, and the identification and dating of recent (in the geological sense) sliding events in
the area.
The paper described procedures developed in recent offshore geohazards studies for estimating the
temporal probability in two situations: 1) When there is a clear trigger initiating the slide that would
develop into a mass gravity flow and 2) When there is evidence of slide activity, but no obvious trigger
for slide initiation. The theoretical framework and example calculations for the two situations were
presented in the paper. For both situations, the approaches developed are believed to provide a realistic
estimate of the annual probability of the initiation of a slide event that might severely affect critical
facilities. The annual probabilities obtained in this paper are upper bound values, assuming certainty that
the sliding masses will reach the sea floor installations. The annual probability estimates can be improved
by doing probabilistic run-out calculations.

Acknowledgments
The authors would like to thank the management of Shell Global Solutions US Inc., Sabah Shell
Petroleum Company and PETRONAS for their permission to present this work. The authors would also
like to thank Mr Tore J. Kvalstad from NGI, as well as their colleagues from GEMS and URS, for their
assistance and valuable contributions.

References
Norwegian Geotechnical Institute (1997). Earthquake Hazard and Submarine Slide – A literature
survey, NGI Report 963014-1.
Hance, J.J. (2003). Development of a database and assessment of seafloor slope stability based on
published literature, MS Thesis, University of Texas at Austin. USA.
Bryn, P., Berg, K., Solheim, K., Kvalstad, T.J. and Forsberg, C.F. (2005). Explaining the Storegga
Slide. Marine and Petroleum Geology 22, 11–19.
Nadim, F., Krunic, D. and Jeanjean, P. (2003). Probabilistic slope stability analyses of the Sigsbee
Escarpment, Offshore Technology Conference ’03, OTC Paper15203, Houstonc, Texas.
Lacasse, S. and Nadim, F. (2007). Probabilistic geotechnical analyses for offshore facilities, Georisk,
Vol. 1, No. 1: 21–42.
Cornell, C.A. (1996). Calculating building seismic performance reliability: A basis for multi-level
design norms, Proc. 11th World Conf. on Earthquake Eng., Acapulco, Mexico.
Ang, A.H-S. and Tang, W.H. (1984). Probability Concepts in Engineering Planning and Design I &
II, John Wiley & Sons, New York.
12 OTC-25160-MS

Ang, A.H-S. and Tang, W.H. (2007). Probability Concepts in Engineering. Emphasis on Applications
to Civil and Environmental Engineering, 2nd edition, John Wiley & Sons, New York.
Nadim, F., Biscontin, G. and Kaynia, A.M. (2007). Seismic Triggering of Submarine Slides, Offshore
Technology Conference ’07, OTC Paper 18911, Houston, Texas.
Nadim, F., Einstein, H. and Roberds, W. (2005). Probabilistic stability analysis for individual slopes,
State of the Art Paper 3, International Conference on Landslide Risk Management, Vancouver, Canada,
31 May-2 June.
Hasofer, A.M. and Lind, N.C. (1974). An exact and invariant first order reliability format, J Eng.
Mech. Div., ASCE, 100(EM1): pp. 111–121.
Lacasse, S., Nadim, F. and Høeg, K. (2011). Hazard and Risk Assessment – State of Practice. ASCE
GeoCongress 2011. Oakland, CA, USA. V 1 pp. 146 –186.

You might also like