XGBoost-SHAP based multi-scale evaluation and nonlinear response of river health in Guangdong Province

Comprehensive assessment of river health at the regional scale is of great significance for coordinating cross-basin ecological protection and sustainable development. This study selected 46 typical rivers in Guangdong Province as research objects, constructing an evaluation system with four criterion layers (physical structure, hydrology, aquatic organisms, and social functions) and 15 indicators. A comprehensive assessment method was proposed based on the XGBoost-SHAP model and spatial interpolation techniques. By integrating subjective and objective weights through the AHP-EWM combined weighting method and utilizing data from 116 monitoring sites, this study revealed the spatial heterogeneity patterns and nonlinear decision-making mechanisms of river health in Guangdong Province. Results indicate that: (1) The XGBoost model demonstrated strong predictive performance in river health assessment (R2 = 0.86, RMSE = 2.27). SHAP interpretation identified water quality grade (C6), fish preservation index (C10), flood control compliance rate (C12), and ecological flow satisfaction rate (C5) as the key driving factors of river health status; (2) Significant threshold effects were observed among indicators. Interval decomposition based on SHAP values captured nonlinear regulatory responses of river health to thresholds, including Class III water quality standard, 90 % ecological flow restoration rate, and flood control compliance critical value; (3) Spatial analysis revealed a northeast-to-southwest decreasing gradient in the River Health Index (RHI) across Guangdong Province. The Pearl River Delta region exhibited significantly lower RHI values compared to mountainous northeastern areas, attributable to intensive human activities and other compounding factors. The developed multi-scale collaborative analytical framework provides quantitative decision-making support for watershed ecological management.