Landslide susceptibility assessment based on an interpretable coupled FR-RF model: A case study of Longyan City, Fujian Province, Southeast China

Abstract

To enhance the prediction accuracy of landslides in in Longyan City, China, this study developed a methodology for geologic hazard susceptibility assessment based on a coupled model composed of a Geographic Information System (GIS) with integrated spatial data, a frequency ratio (FR) model, and a random forest (RF) model (also referred to as the coupled FR-RF model). The coupled FR-RF model was constructed based on the analysis of nine influential factors, including distance from roads, normalized difference vegetation index (NDVI), and slope. The performance of the coupled FR-RF model was assessed using metrics such as Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves, yielding Area Under the Curve (AUC) values of 0.93 and 0.95, which indicate high predictive accuracy and reliability for geological hazard forecasting. Based on the model predictions, five susceptibility levels were determined in the study area, providing crucial spatial information for geologic hazard prevention and control. The contributions of various influential factors to landslide susceptibility were determined using SHapley Additive exPlanations (SHAP) analysis and the Gini index, enhancing the model interpretability and transparency. Additionally, this study discussed the limitations of the coupled FR-RF model and the prospects for its improvement using new technologies. This study provides an innovative method and theoretical support for geologic hazard prediction and management, holding promising prospects for application.