Rainfall and earthquake-induced landslide susceptibility assessment using GIS and Artificial Neural Network

Y Li,L Zheng,C Tang,G Zhou,G Chen

doi:10.5194/nhess-12-2719-2012

Abstract

Abstract. A GIS-based method for the assessment of landslide susceptibility in a selected area of Qingchuan County in China is proposed by using the back-propagation Artificial Neural Network model (ANN). Landslide inventory was derived from field investigation and aerial photo interpretation. 473 landslides occurred before the Wenchuan earthquake (which were thought as rainfall-induced landslides (RIL) in this study), and 885 earthquake-induced landslides (EIL) were recorded into the landslide inventory map. To understand the different impacts of rainfall and earthquake on landslide occurrence, we first compared the variations between landslide spatial distribution and conditioning factors. Then, we compared the weight variation of each conditioning factor derived by adjusting ANN structure and factors combination respectively. Last, the weight of each factor derived from the best prediction model was applied to the entire study area to produce landslide susceptibility maps. Results show that slope gradient has the highest weight for landslide susceptibility mapping for both RIL and EIL. The RIL model built with four different factors (slope gradient, elevation, slope height and distance to the stream) shows the best success rate of 93%; the EIL model built with five different factors (slope gradient, elevation, slope height, distance to the stream and distance to the fault) has the best success rate of 98%. Furthermore, the EIL data was used to verify the RIL model and the success rate is 92%; the RIL data was used to verify the EIL model and the success rate is 53%.

Highlights

The 12 May 2008 Wenchuan earthquake, with the characteristics of high magnitude (Ms = 8.0), shallow hypocenter, long fracture zone, great rupture, large energy release and long duration (Xu et al, 2011), triggered as many as 56 000 landslides of various types in the mountainous terrain in Sichuan Province, China, throughout an area of about 50 000 km2
Landslide susceptibility assessment is an important tool for land use planning and special development activity within a given area, which is based on past landslides, geology, topography, hydrology and other pertinent data
Comparison between landslide spatial distribution and conditioning factors was calculated from landslide frequency analysis based on slope unit

Summary

Introduction

The 12 May 2008 Wenchuan earthquake, with the characteristics of high magnitude (Ms = 8.0), shallow hypocenter (the depth of the hypocenter was less than 20 km), long fracture zone (approximately 300 km), great rupture (the largest rupture was about 7 m), large energy release (three times that of the 1976 Tangshan earthquake) and long duration (the main shock duration was about 120 s) (Xu et al, 2011), triggered as many as 56 000 landslides of various types in the mountainous terrain in Sichuan Province, China, throughout an area of about 50 000 km. Landslide susceptibility assessment is an important tool for land use planning and special development activity within a given area, which is based on past landslides, geology, topography, hydrology and other pertinent data. It is widely performed in two steps. The second step is the preparation of a landslide susceptibility map showing the “likelihood that a phenomenon will occur in an area on the basis of the local terrain conditions”

Objectives

Methods

Results