Flood risk assessment and watershed resilience strategies in the Qinba Mountains based on Random Forest and frequency ratio-Borda count analysis

This study addresses the increasingly severe flood risks in the Qinba Mountains of southern Shaanxi Province, taking the catastrophic flood disaster of 6 August 2020, in Luonan County as the research background. We developed an integrated flood risk assessment method combining Random Forest machine learning and Frequency Ratio-Borda Count decision analysis. The study comprehensively considers 11 key factors across multiple categories including topography, hydrology, meteorology and environment, establishing a multi-dimensional risk assessment system. The Random Forest model demonstrated good overall performance, employing a conservative prediction strategy to ensure reliable identification of high-risk areas and identified aspect, drainage density and precipitation as the dominant risk factors. Through the integration of both methods to classify risk levels for 30 sub-watersheds, we found that 6 very high-risk sub-watersheds concentrated 49.5% of historical flood points. Multicollinearity analysis confirmed that all environmental variables maintained good independence, mainly distributed in low-lying areas along rivers. Based on the spatial distribution characteristics of risk, the study proposes differentiated management strategies of upstream ecological protection, midstream comprehensive management and downstream engineering protection, providing scientific support for mountain flood risk management.