Susceptibility assessment of earthquake-induced landslides using Bayesian network: A case study in Beichuan, China

Because of the uncertainties and complexities of the factors involved in causing landslides, it is generally difficult to analyze their influences quantitatively and to predict the probability of landslide occurrence. In this work, a hybrid method based on Bayesian network (BN) is proposed to analyz...

Full description

Saved in:
Bibliographic Details
Published in:Computers & geosciences 2012-05, Vol.42, p.189-199
Main Authors: Song, Yiquan, Gong, Jianhua, Gao, Sheng, Wang, Dongchuan, Cui, Tiejun, Li, Yi, Wei, Baoquan
Format: Article
Language:English
Subjects:
Assessments
Bayesian analysis
Bayesian network
Bayesian theory
Boron nitride
case studies
China
computers
earthquakes
Landslide
Landslides
Mathematical models
Networks
probability
probability distribution
Seismic phenomena
Susceptibility assessment
Wenchuan earthquake
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Because of the uncertainties and complexities of the factors involved in causing landslides, it is generally difficult to analyze their influences quantitatively and to predict the probability of landslide occurrence. In this work, a hybrid method based on Bayesian network (BN) is proposed to analyze earthquake-induced landslide-causing factors and assess their effects. Our study area is Beichuan, China, where landslides have occurred in recent years, including mass landslides triggered by the 2008 Wenchuan earthquake. To provide a robust assessment of landslide probability, key techniques from landslide susceptibility assessment (LSA) modeling with BN are explored, including data acquisition and processing, BN modeling, and validation. In the study, eight landslide-causing factors were chosen as the independent variables for BN modeling. And this study shows that lithology and Arias intensity are the major factors affecting landslides in the study area. On the basis of the a posteriori probability distribution, the occurrence of a landslide is highly sensitive to relief amplitudes above 116.5m. Using a 10-fold cross-validation and a receiver operating characteristic (ROC) curve, the resulting accuracy of the BN model was determined to be 93%, which demonstrates that the model achieves a high probability of landslide detection and is a good alternative tool for landslide assessment.
ISSN:0098-3004
1873-7803
DOI:10.1016/j.cageo.2011.09.011