Quantification of human vulnerability to earthquake-induced landslides using Bayesian network

Abstract

Devastating landslides were frequently triggered by earthquakes and caused severe damage and human losses. More than 25,000 people were killed by the Wenchuan earthquake-induced landslides. A proper model for estimating loss of lives is important in quantifying the human risks posed by earthquake-induced landslides. This paper presents a new model for quantitatively estimating the human vulnerability to earthquake-induced landslides using Bayesian networks. A database incorporating information of fatal landslides induced by the Wenchuan earthquake is established. A Bayesian network is constructed according to a logic structure of loss-of-life mechanisms by considering four types of key factors, i.e., deposit thickness, building damage, evacuation, and resistance capacity. The probabilities of the nodes (factors) and the arcs (inter-relationships) of the network are quantified according to statistical data, judgment based on experiences, and existing physical or empirical models. The human vulnerability to earthquake-induced landslides is quantitatively estimated through constructing a systematic structure considering the uncertainties of the control factors and their inter-relationships. Comparison between the recorded fatality rate and predicted fatality is conducted to verify the proposed model. Finally, sensitivity analysis is conducted to identify the most important parameters that lead to loss of lives.