Analytical Hierarchy and Neural Network based Landslide Risk Assessment at Sikkim India: A Comparative Study

Abstract

In the Himalayan region, landslides are mainly triggered due to severe rainfall and earthquake events. Especially in the Sikkim Himalayas, landslides occur every year, resulting in casualties and huge property losses. Therefore, it is essential to study the landslide susceptibility and risk zonation of the Himalayan landscape. This study attempts to classify 2656 landslide events in the Sikkim in their hazard impact. The Sikkim landslide repository of Bhukosh-GIS is combined with the digital elevation model (DEM) to classify landslides using the analytical hierarchy process (AHP). The state of Sikkim is divided into four landslide risk zones, with North Sikkim being the highest risk zone (Zone-D) and South Sikkim at the least (Zone-A). Thereafter, a Levenberg Marquardt (LM)-based backpropagation neural network (BPNN) is proposed to predict the risk scores based on the landslide input parameters and AHP-based target scores. After performing several numerical experiments, the input-output dataset is divided into 60:15:25 as training, validation, and testing set. Finally, it is observed from the simulation runs of comparative study, the proposed neural network with 92.5 % testing accuracy could be a promising alternative for the landslide risk assessment in Sikkim, India.