Anticipating Nutrient Risk in Lakes and Reservoirs: A Data-Driven Framework for Sustainable Water Quality Management

Sustainable management of lakes and reservoirs increasingly requires timely knowledge of how nutrient conditions evolve across both space and time. Total phosphorus is a key indicator of nutrient enrichment and eutrophication risk, yet its spatial distribution is highly heterogeneous and changes continuously under complex environmental influences. To provide a more informative basis for water quality monitoring and ecological management, this study develops a data-driven forecasting framework using monthly high-resolution water quality records from Chinese lakes and reservoirs spanning 2000–2023. The historical total phosphorus observations are organized as continuous spatiotemporal image sequences, from which future spatial distributions are predicted through a deep learning architecture incorporating Temporal Difference Injection, Regional Graph Token Reasoning, and Edge-Aware Residual Reconstruction. Temporal Difference Injection enhances sensitivity to month-to-month concentration changes, Regional Graph Token Reasoning captures non-local relationships among geographically separated regions, and Edge-Aware Residual Reconstruction improves the preservation of spatial gradients and local boundaries in areas with elevated total phosphorus concentrations. Across three independent runs, the proposed framework achieves an RMSE of 0.0317±0.0076, an MAE of 0.0147±0.0040, an SSIM of 0.9684±0.0146, and a PSNR of 30.3027±2.6113. Comparative experiments, component ablations, and hyperparameter sensitivity analyses further show that temporal-change representation, regional dependency modeling, and structural reconstruction contribute complementary improvements to forecasting performance. The resulting continuous spatial forecasts provide dynamic information on nutrient distributions that can assist in locating areas of elevated ecological risk, prioritizing monitoring efforts, supporting risk-warning activities, and improving differentiated management of lakes and reservoirs.