An innovative method for landslide susceptibility mapping supported by fractal theory, GeoDetector, and random forest: a case study in Sichuan Province, SW China

Abstract

Globally, but especially in Sichuan Province (Southwest China), landslides are considered to be one of the most common geological hazards. The purpose of the current study is to develop a novel model based on fractal theory, GeoDetector, and random forest, namely, the “GD-FFR-RF” model, for landslide susceptibility modeling and identifying areas prone to landslides with a case study of Sichuan Province, China. On this basis, fourteen landslide conditioning factors, including slope angle, slope aspect, relief amplitude, surface roughness, cutting depth, elevation, river density, rainfall, stream power index, road density, plan curvature, profile curvature, peak ground acceleration, and lithology, are selected. The information gain ratio is used to determine the significance of different conditioning factors. The overall performance of the three resulting models is assessed using the area under the receiver operating characteristic (ROC) curve (AUC) and several statistical indices. The GD-FFR-RF model yields better overall performance and more accurate results than the FFR-RF and FR-RF models. Therefore, in conclusion, the GD-FFR-RF model can be applied as a promising comprehensive model for the spatial prediction of landslide occurrence in subsequent studies. The generated landslide susceptibility maps will be useful for developing strategies for landslide hazard prevention and mitigation.