Geological control and mobility of typhoon gaemi-induced landslides in Zixing City, Hunan Province, China

Typhoon Gaemi triggered widespread landslides in Zixing city, Hunan Province, China, in July 2024. To improve the identification and understanding of typhoon-induced landslides and investigate the landslide distribution patterns, mobility characteristics, and controlling factors associated with these events, an integrated approach combining deep learning techniques with Planet imagery, statistical analysis, and explainable machine learning (ML) modeling was employed. The deep learning model successfully identified 16,120 actual landslides across the 2746.79 km 2 study area. The total affected area was 32.16 km 2 . Spatial analysis revealed that landslides were predominantly concentrated in the eastern and northeastern regions surrounding the Dongjiang Reservoir, with the maximum number density and area density reaching 129 per km 2 and 37.58%, respectively. Most landslides were small in scale, with most occurring in the middle–lower slope segments (30–50% of the total slope length). The greatest number of landslides (63.37%) was recorded in granite areas, followed by sandstone areas (29.30%), demonstrating pronounced lithological control characteristics. SHAP analysis revealed that lithology was the primary factor contributing to landslide formation, and significant differences in the dominant factors controlling landslides across regions with different lithologies were observed. The analysis of landslide mobility through interpretable ML indicated that the slope gradient was a critical factor influencing mobility, accounting for approximately 45% of the SHAP value and exhibiting a significant negative correlation with mobility. These results demonstrated that geological conditions would form the basis of a fundamental susceptibility framework, whereas slope gradients would control mobility patterns through gravitational and hydrological processes. Therefore, this study could serve as a scientific foundation for landslide prediction and risk assessment measures in similar geological settings.