Landslide susceptibility mapping in tropical cyclone-affected areas of Central Vietnam using bivariate statistical models

Abstract Landslides are a serious issue in mountainous Central Vietnam, a region characterized by frequent tropical cyclones and highly vulnerable terrain. Despite this, there has been limited research on effective methodologies for building landslide susceptibility maps to guide risk mitigation policies. This study aims to map landslide susceptibility using frequency ratio (FR), statistical index (SI), weights of evidence (WOE) models, and comparative assessment of their performance. The significance of this study is to determine the appropriate scale for mapping landslide susceptibility using bivariate statistical models. This is based on a GIS database on landslides in Phuoc Son district, Quang Nam province, Vietnam, which is often intensely affected by tropical cyclones, including a landslide inventory map and ten landslide-related factors, i.e., geology, distance to fault, elevation, relief amplitude, slope, aspect, rainfall, soil type, land use, and distance to road. The landslide inventory data consists of 858 landslide points constructed from different data sources such as field surveys, analysis of Google Earth satellite images, and inherited data from previous studies. They were randomly divided into two parts for modeling and validating, with 70% (601) and 30% (257), respectively. The result of the multicollinear test confirmed the independence of 10 input variables and their influence on landslide occurrence in the study area. The weights were calculated for the three models using 70% of data for modeling and related factors. Afterward, the landslide susceptibility index (LSI) was computed from weights using the spatial overlay method in GIS. The performance of the models was evaluated by receiver operating characteristic (ROC) curve analysis using 30% of the data for validation. The results show that the spatial prediction of landslides of the FR, SI, and WOE models all have high and approximately equal performance with the area under the curve (AUC) parameters of 0.890, 0.890, and 0.887, respectively.