A landslide susceptibility assessment method using SBAS-InSAR to optimize Bayesian network

Abstract

Landslide susceptibility assessment is crucial to mitigate the severe impacts of landslides. Although Bayesian network (BN) has been widely used in landslide susceptibility assessment, no study has compared the accuracy of different BN structure construction methods for this purpose. SBAS-InSAR technology plays a vital role in landslide research, but its advantages combined with BN to further improve prediction accuracy still need to be studied. This paper takes Hanyuan County as the study area. First, 20 traditional landslide impact factors were extracted from data such as topography and meteorology. A new method GDSP was designed to fuse GeoDetector and SHAP for dominant factor screening. Then, 8 different BN structure learning methods were compared using the AUC value of the ROC curve, among which Tabu&K2 method showed the highest accuracy. The deformation factor calculated by SBAS-InSAR is then incorporated into the BN model. The optimized Bayesian network (OPT-BN) outperformed the unoptimized version (ORI-BN) in accuracy, and the landslide susceptibility mapping was more reasonable. The reverse inference highlighted that areas with lower elevation, plow land, impervious cover, and higher rainfall are more prone to landslides. This method provides valuable insights into landslide hazard prevention and control and provides a new method for future landslide research.