Evaluation of the extreme rainfall predictions and their impact on landslide susceptibility in a sub-catchment scale

Abstract

Geohazards induced by extreme-rainfall events have considerably affected Taiwan throughout the last decade. Because of differences in rainfall behavior, the hazards of rainfall-induced landslides differ according to the catchment. In addition to have detailed analyses, comparisons and discussions, a rainfall-frequency analysis was conducted and an atmospheric general circulation model (AGCM) was used to assess the temporal and spatial rainfall behavior in an adopted catchment in Central Taiwan. To assess the future spatial hazard of landslides in the study area, landslide susceptibility analysis methods, including logistic regression (LR) and support vector machines (SVMs), were trained before being applied. The predictions of rainfall extreme and their Impact on landslide susceptibility were evaluated, compared, and discussed. The results suggest the AGCM predictions are sensitive to the route of typhoon and the local topography. Due to uncertainties of rainfall prediction and the prediction limitation of landslide susceptibility models, a multi-method analysis is necessary for the rainfall-induce landslide susceptibility predictions.