GIS-based bivariate statistical modelling for earthquake-triggered landslides susceptibility mapping related to the 2008 Wenchuan earthquake, China

The main purpose of this research is to evaluate the modelling capability and predictive power of a bivariate statistical method for earthquake-triggered landslide susceptibility mapping. A weight index (W-i) model was developed for the 2008 Wenchuan earthquake region in Sichuan Province, China, using a wide range of optical remote sensing data, and carried out on the basis of a geographic information system (GIS) platform. The 2008 Wenchuan earthquake triggered 196007 landslides, with a total area of 1150.43 km(2), in an approximately oblong area around the Yingxiu-Beichuan coseismic surface fault-rupture (the Yingxiu-Beichuan fault). The landslides of the study area were mapped using visual interpretation of high-resolution satellite images and aerial photographs, both pre-and post-earthquake, and checked in the field at various locations. As a consequence, a nearly complete inventory of landslides triggered by the Wenchuan earthquake was constructed. Topographic and geological data and earthquake-related information were collected, processed and constructed into a spatial database using GIS and image processing technologies. A total of 10 controlling parameters associated with the earthquake-triggered landslides were selected, including elevation, slope angle, slope aspect, slope curvature, slope position, lithology, seismic intensity, peak ground acceleration (PGA), distance from the Yingxiu-Beichuan fault, and distance along this fault. To assist with the development of the model, the complete dataset of 196007 landslides was randomly partitioned into two subsets; a training dataset, which contains 70% of the data (137204 landslides, with a total area of 809.96 km(2)), and a testing dataset accounting for 30% of the data (58803 landslides, with a total area of 340.47 km(2)). A landslide susceptibility index map was generated using the training dataset, the 10 impact factors, and the W-i model. In addition, for a conditionally dependent factor analysis, seven other factor-combination cases were also used to construct landslide susceptibility index maps. Finally, these eight landslide susceptibility maps were compared with the training data and testing data to obtain model capability (success rate) and predictive power (predictive rate) information. The validation results show that the success and predictive rates of the W-i modelling exceeded 90% for the approaches that include the use of seismic factors. The final landslide susceptibility map can be used to identify and delineate unstable susceptibility-prone areas, and help planners to choose favourable locations for development schemes, such as infrastructure, construction and environmental protection schemes. The generic component of this research would allow application in other regions affected by high-intensity earthquakes and unstable terrain covering very large areas.