Susceptibility assessment of colluvium landslides along the Changyang section of Qingjiang River using Logistic regression and random forest methods

Abstract

The geological and environmental background conditions of the Qingjiang River Basin are highly complex, particularly with frequent geological disasters along the Qingjiang reservoir bank. Previous susceptibility assessment for geological disasters was mostly focused on administrative areas and seldom specialized evaluations for the reservoir bank zone. Furthermore, there is still room for improvement in the evaluation index system, as well as in the pertinence and reliability of the evaluation method. To address these shortcomings, a more suitable susceptibility evaluation index system was constructed to obtain accurate and applicable susceptibility zoning results. The Yuxiakou to Ziqiu section of the Qingjiang River Basin was chosen as the research area, with the wading slope body on both sides of the river selected as the research object and the slope unit chosen as the evaluation unit. A susceptibility evaluation system composed of ten indicators, including slope, aspect, elevation range, slope type, NDVI, TWI, slope structure type, engineering geological rock formation, accumulation thickness, and valley evolution, was constructed.The logistic regression and random forest methods were used to construct the evaluation model based on the normalized certain factors, and different susceptibility zoning results were obtained. According to the evaluation results, the high-prone areas were mainly distributed in the middle to lower water wading areas of the left bank from the east of Yuxiakou to the east of Ziqiu, along the main stream of the Qingjiang River. The logistic regression model showed better applicability in the reservoir-bank section with complex topography and landforms. The research revealed that the accumulation thickness and valley evolution indicators were effective in representing the unique geological background conditions of the Qingjiang reservoir bank. The logistic regression model was able to learn the developmental law of disasters and has a reliable susceptibility prediction ability.

Keywords

• accumulation thickness
• logistic regression
• random forest
• susceptibility assessment