Remote sensing and GIS based artificial neural network system for landslide suceptibility mapping

Abstract

Landslide susceptibility mapping is necessary in order to facilitate rational, systematic and efficient decisions concerning planning of development in mountainous regions and also for the mitigation and management of landslide disasters. Radial Basis Function Link Networks (RBFLN) was used as a landslide inventory-driven method for the identification of landslide susceptibility. Generation of input data for RBFLN involved the landslide causal factor (evidential theme) maps comprising geology, photo-lineament, land use land cover (LULC), soil, slope angle, aspect, relative relief, profile curvature, distance to drainage and distance to reservoir boundary. 116 landslide incidence and 116 no incidences were used to train the network. A unique condition grid map was prepared by the combination of each evidential theme. For each input training vector, weights in the form of fuzzy membership function were assigned. Based on fuzzy membership values, weights of each pixel of unique condition grid map were computed on the basis of RBFLN. The RBFLN weights were linked to the unique condition grid and a continuous landslide prediction map was created which was further classified into five relative susceptible zones.