Coseismic landslides caused by the 2022 Luding earthquake in China: Insights from remote sensing interpretations and machine learning models

Zhiwen Ding^1,2Chun Wang^3*

• ¹State Key Laboratory of Geohazard Prevention and Geoenvironment Protection, Chengdu University of Technology, Chengdu, Sichuan Province, China
• ²College of Environment and Civil Engineering, Chengdu University of Technology, Chengdu, Sichuan Province, China
• ³Sichuan Institute of Nuclear Geological Survey, Chengdu, China

On September 5, 2022, an Ms 6.8 earthquake occurred in Luding County, Sichuan Province, China, triggering numerous landslides and causing extensive damage to buildings and casualties. A comprehensive study of the characteristics of coseismic landslide distribution in this area is highly important for postearthquake emergency response. In this paper, coseismic landslides in highintensity areas were interpreted through remote sensing images, and 5,386 landslides with a total area of 22.2 km 2 were identified. The spatial distribution of coseismic landslides was analyzed in relation to seismic, topographic, and geological factors to assess their susceptibility at the regional scale. The results revealed that the majority of coseismic landslides occurred on both sides of the Xianshuihe fault, which is the causative fault, and the landslides exhibited a linear distribution. These landslides were concentrated mainly at elevations between 1,000 and 1,800 m, with slopes of 30-50°, and they occurred in areas with hard intrusive rock masses. The spatial distribution of coseismic landslides in the study area was predicted using three models: Random Forest (RF), Gradient Boosting Decision Tree (GBDT) and eXtreme Gradient Boosting (XGBoost). Furthermore, SHapley Additive exPlanations (SHAP) theory was used to conduct a quantitative analysis of the main geomorphological factors controlling the landslides. This paper revealed that different topographic factors had varying degrees of nonlinear impacts on landslide formation and that the combined effects of multiple factors, such as the peak ground acceleration (PGA), slope, and lithology, controlled the formation of landslides. This paper highlights the significant advantages of machine learning-based intelligent identification and analytical techniques in landslide disaster emergency surveys and analysis of formation conditions. Rapid prediction of the spatial location and distribution pattern of coseismic landslides provides effective support and guidance for emergency response, risk mitigation, and reconstruction planning.