Key points are not available for this paper at this time.
A Landslide susceptibility mapping along mountain highways is often based on static predisposing factors and historical inventories, which may fail to identify slopes currently destabilized by recent engineering activities. This study adapts and extends a deformation enhanced landslide susceptibility framework for the Ninglang to Yongsheng Expressway corridor in Yunnan Province, Southwest China. A landslide inventory comprising 256 mapped events was established from historical records and high resolution optical imagery. Susceptibility models were developed using Random Forest and XGBoost classifiers based on a suite of static conditioning factors describing topography, land cover, geology, hydrology, and human activity. To account for ongoing slope processes, multi temporal InSAR derived deformation information was integrated into the modeling framework to represent present day slope instability. The inclusion of deformation features leads to a clear improvement in predictive performance, with ROC AUC increasing from 0.764 to 0.826 for the Random Forest model and from 0.758 to 0.820 for the XGBoost model, while PR AUC reaches approximately 0.876–0.878 for the deformation enhanced models. Crucially, a multi-phase SHAP evaluation across the engineering lifecycle revealed that the predictive importance of InSAR deformation surged during active excavation, confirming that the models captured actual physical kinematics rather than statistical artifacts. SHAP based interpretation indicates that distance to road and elevation are the most influential predictors, while deformation information provides complementary dynamic evidence of active slope instability. The adapted framework enhances the operational value of landslide susceptibility mapping for inspection prioritization and geohazard management along newly constructed mountain highways.
Xu et al. (Thu,) studied this question.