Unraveling long-term health dynamics of impounded lakes integrating water diversion-adapted planktonic index of biotic integrity and machine learning.

Assessing ecosystem integrity of impounded lakes subjected to sustained water diversion is fundamental for advancing adaptive water governance. Nevertheless, drivers of lake health variation under recurrent flow regulation, particularly their multiscale spatiotemporal manifestations and persistent trajectories, remain inadequately characterized. Using decade-long monitoring data from typical impounded lakes along the Eastern Route of China's South-to-North Water Diversion Project (SNWDP), this study developed a water diversion-adapted planktonic index of biotic integrity (P-IBI) to quantify spatiotemporal health heterogeneity and coupled it with supervised machine learning (SML) for predictive modeling and interpretable diagnostics of P-IBI dynamics. Results showed while water diversion improved water quality, it restructured plankton communities and identified twelve metrics spanning plankton density, composition, diversity, function, tolerance, and keystone species to construct the P-IBI framework. Spatially, diversions homogenized adjacent lakes but increased heterogeneity within individual lake sub-regions, while temporally, P-IBI initially declined but gradually recovered to stabilize. Comparative evaluation of various SML models identified eXtreme Gradient Boosting (XGBoost) as the optimal predictor for P-IBI estimation, exhibiting superior performance and stability across 100 randomized dataset partitions. The XGBoost model independent of biomonitoring features demonstrated robust predictive capability for assessing water diversion impacts on lake health (R2=0.89, accuracy=0.816). SHapley Additive exPlanations (SHAP) analysis highlighted suspended solids (35-70 mg/L) and Chlorophyll a (<8.25μg/L) as the key water quality factors influencing lake health, primarily by modulating light penetration, nutrient availability, and algal succession. This framework validates integrating P-IBI with SML to predict and interpret diversion effects, providing a scientific basis for sustainable management of large-scale water diversion projects.