Abstract
In mid-to-late April 2024, an extreme heavy rainfall event occurred in Shaoguan City, Guangdong Province, triggering a large number of landslide disasters in the Jiangwan Town area of Shaoguan, causing certain regions to remain disconnected for nearly 36 hours and attracting widespread social attention. Rapid and accurate identification of the basic characteristics, developmental distribution patterns, and formation conditions of landslides is crucial for disaster emergency decision-making and risk hazard elimination and disposal. Using post-disaster optical remote sensing imagery combined with deep learning models, rapid intelligent identification and manual verification of rainfall-induced landslides in Jiangwan Town, Shaoguan were conducted, interpreting a total of 1,192 landslides with a total area of approximately 3.14 km². The landslides were predominantly small-to-medium in scale, mainly distributed in clustered bands along rivers in a northeast-southwest direction, with significant clustering effects. Spatial statistical analysis indicates that landslides are mainly distributed within the elevation range of 200-300 m, on concave slopes with gradients of 10°-30°. Furthermore, the Random Forest model and SHAP theory were used to conduct quantitative analysis of the geomorphological controlling factors of landslides, revealing that different topographic and geomorphological factors have varying degrees of nonlinear influence on landslide formation, and that the coupled effects of multiple factors including elevation, slope, and water convergence conditions jointly control landslide formation. This study highlights the significant advantages of deep learning-based intelligent identification and analysis technologies in emergency investigation of landslide disasters and analysis of formation conditions, which can provide important technical support for rapid disaster loss assessment and risk hazard investigation.
Full Text
Preliminary Analysis of Extreme Rainfall-Induced Cluster Landslides in Jiangwan Township, Shaoguan, Guangdong, April 2024
XU Qiang, XU Fanshu, PU Chuanhao, LI Weile, FAN Xuanmei, DONG Xiujun, WANG Xiaochen, LI Zhigang
(1. State Key Laboratory of Geohazard Prevention and Geoenvironment Protection, Chengdu University of Technology, Chengdu 610059, China)
Abstract
In mid-to-late April 2024, an extreme rainfall event struck Shaoguan City, Guangdong Province, triggering numerous landslides in Jiangwan Township. The disaster left some areas isolated for nearly 36 hours, drawing widespread public attention. Rapid and accurate characterization of landslide features, distribution patterns, and formation conditions is essential for effective emergency response and risk mitigation.
Using post-disaster optical remote sensing imagery combined with a deep learning model, we conducted rapid intelligent identification and manual verification of rainfall-induced landslides in the study area. A total of 1,192 landslides were mapped, covering a combined area of approximately 3.14 km². The landslides were predominantly small-to-medium in scale and exhibited pronounced clustering effects, distributed in distinct northeast-southwest bands along river corridors.
Spatial statistical analysis revealed that landslides were primarily concentrated on concave slopes at elevations between 200-300 meters with gradients of 10°-30°. Furthermore, we employed a Random Forest model integrated with SHAP (SHapley Additive exPlanations) theory to quantitatively analyze the dominant geomorphological controlling factors. The results demonstrated that various topographic factors exert different degrees of nonlinear influence on landslide formation, with the coupled interactions of elevation, slope gradient, and water convergence conditions collectively controlling landslide development.
This study highlights the substantial advantages of deep learning-based intelligent identification and analysis techniques for emergency landslide investigation and formation condition analysis, providing critical technical support for rapid disaster loss assessment and risk hazard screening.
Keywords: Extreme rainfall; Cluster landslides; Intelligent identification; Distribution pattern; Formation conditions