Work

InSAR-based landslide susceptibility analysis in karst erosion landforms using a non-landslide sampling strategy

Year: 2025

Type: article

Abstract: • Propose an LSA framework with non-landslide sampling in low InSAR deformation areas. • Conduct LSA in areas with high vegetation coverage and karst erosion landforms. • Apply SHAP algorithm to explain global and local mechanisms in LSA. Landslide Susceptibility Analysis (LSA) is essential for effective geological hazard management, particularly in karst regions where complex terrain and dense vegetation present significant challenges. This study introduces an innovative LSA framework that integrates a low-InSAR deformation sampling strategy (BL-InSAR) to mitigate InSAR errors in surface deformation measurements over densely vegetated areas and reduce uncertainty in non-landslide sample selection. Focusing on Bijie District in Guizhou Province, a karst-dominated region, we incorporate karst hydrogeological data into the LSA process. Utilizing a C-Support Vector Machine (C-SVM) model, we analyze 14 factors, including karst hydrogeology, to generate four landslide susceptibility maps employing BL-InSAR and three conventional sampling methods. The SHapley Additive exPlanations (SHAP) algorithm is applied to elucidate both global and local landslide mechanisms. Our findings identify karst hydrogeology, precipitation, lithology, and slope as key factors influencing landslide susceptibility, with high-risk areas concentrated in the northwest, central, and northeastern regions. The BL-InSAR method achieved an Area Under the Curve (AUC) of 0.8885, outperforming other methods by 2.69 %, 3.66 %, and 8.45 %, respectively. This framework enhances the interpretability of landslide susceptibility models by considering landslide creep dynamics through InSAR and integrating karst hydrogeological data. It offers a more effective approach to capturing and interpreting landslide risks in karst environments, thereby contributing to more reliable LSA in complex terrains.

Source: International Journal of Applied Earth Observation and Geoinformation

DOI: 10.1016/j.jag.2025.104997

Authors: Zhixin Liu, Zhijie Zhang, Wanchang Zhang, Jingtong Ren, Qunli Chen