2221
Quantitative risk assessment for earthquake-triggered landslides using Bayesian
network
Évaluation quantitative du risque associé aux glissements de terrain déclenchés par séisme en
utilisant un réseau Bayésien
Nadim F., Liu Z.Q.
Norwegian Geotechnical Institute (NGI) / International Centre for Geohazards (ICG), Norway
ABSTRACT: Strong earthquakes in mountainous regions usually trigger many landslides that lead to damage and destruction. Sepa-
rate investigations of single hazard processes (earthquake and landslide) might lead to a misjudgement of the risks associated with this
type of cascading hazards. The assessment and mitigation of the risks require a multi-risk analysis approach that can account for the
interactions among the threats and among the vulnerabilities to these threats. In this paper, a quantitative risk assessment model using
Bayesian network is proposed to estimate the risk for the buildings exposed to the threat of earthquake-triggered landslides. A sensi-
tivity analysis was done to identify the optimum and appropriate risk reduction strategy in a multi-hazard perspective.
RÉSUMÉ : De forts séismes dans les régions montagneuses déclenchent habituellement des nombreux glissements de terrain qui mè-
nent à dommages et destruction. Si l’on tra
ite les aléas singuliers séparément, par exemple un tremblement de terre et un glissement
de terrain, une estimation erronée des risques associés à ce type d’aléas cascadés peut être obtenue. L'évaluation et l'atténuation des
risques nécessitent une approche multi-risques qui doit tenir compte des interactions entre les dangers et les vulnérabilités à ces dan-
gers L’article propose un modèle d'évaluation quantitative des risques en utilisant un réseau Bayésien pour estimer le risque aux bât
i-
ments exposés aux glissements de terrain déclenchés par un séisme. Une analyse paramétrique a été réalisée pour identifier une straté-
gie optimale et appropriée pour réduire le risque dans une perspective multi-aléas.
KEYWORDS: Landslides, Earthquake, Quantitative risk assessment, Bayesian network
1 INTRODUCTION
Earthquake-triggered landslides are one of the most common
secondary disasters caused by earthquake in mountainous areas.
In the Wenchuan earthquake of May 2008, more than 15 000
landslides of various types were triggered in the steep mountain
slopes (Huang 2008). The landslides caused more than 20 000
fatalities (Yin
et al
2009) and caused extensive damage to hous-
ing settlements and irrigation channels (Tang
et al
2011).
In earthquake-triggered landslide risk assessment, complex
interactions are present between the earthquake and landslide
threats. The vulnerabilities of the elements at risk are sometimes
also correlated to the threats. Amplified risk resulting from haz-
ard and vulnerability interactions has to be considered. Unfortu-
nately, to date, the risk assessment involving multiple hazards is
commonly done with independent analyses neglecting possible
cascade effects (Marzocchi
et al
2012) and standard approaches
for dealing with the multi-risk situations are not available (Kap-
pes
et al
2012). In this paper, a quantitative risk assessment
model using Bayesian network is proposed to estimate the risk
for the buildings exposed to the threat of earthquake-triggered
landslides.
2 BAYESIAN NETWORKS
A Bayesian network (BN), also called a belief network, Bayes
net or casual network, is an increasingly popular method for
reasoning under conditions of uncertainty and modelling uncer-
tain domains. It has been applied to a number of civil and envi-
ronmental engineering problems, ranging from avalanche risk
assessment (Grêt-Regamey and Straub 2006), dam risk analysis
(Smith 2006), earthquake risk management (Bayraktarli
et al
2005; Bensi
et al
2011), design of early warning system for
landslide hazard mitigation (Medina-Cetina
et al
2007) and en-
vironmental modelling and management (Aguilera
et al
2011).
A BN is a probabilistic model based on directed acyclic
graph
B
s
= G(Z, E)
(1)
where
B
s
represents the structure of the network,
Z
is the set of
random variables (
Z
1
, Z
2
,
… Z
n
)
, and
E
Z×Z
is the set of di-
rected arcs, representing the probabilistically conditional de-
pendency relationships among random variables. Each variable
Z
i
can be defined in a discrete and finite outcome space (discrete
random variable) or as a continuous outcome space (continuous
random variable).
One important property of the Bayesian network is that the
joint probability function of all random variables in the network
can be factorized into conditional and unconditional probabili-
ties implied in the network (Jensen 2007). Thus, the joint distri-
bution can be expressed in the compact form as
1 2
1
( , , ...,
)
(2)
n
n
i
i
i
P z z
z
P z pa Z
where
pa(Z
i
)
is the parent set of
z
i
. It should be noted that if
child node
z
i
has no parents, then the equation reduces to the
unconditional probability of
p(z
i
)
.
A simple Bayesian network with five nodes and five arcs is
illustrated in Fig.1. These nodes are: Magnitude (
M
), Distance
(
D
), Seismic severity (
S
), Landslide severity (
L
), and Building
damage (
B
). These nodes are connected via the arcs:
M-S
,
D-S
,
S-L
,
S-B
and
L-B
. The prior probability of
B
,
P(B = B
1
)
can be
calculated by
2 2 2 2
1
1
1 1 1 1
,
,
,
,
(3)
i
j
m
k
i
j k m
P B B
P B B M M D D S S L L
where
P
= probability,
B
1
= no damage,
M
1
= small magnitude,
M
2
= large magnitude,
D
1
= small distance,
D
2
= large distance,
S
1
= low seismic severity,
S
2
= high seismic severity,
L
1
= low
landslide severity,
L
2
= high landslide severity.
