Widespread decrease of phosphorus and the potential driving mechanisms in Taihu basin’s lakes

• A robust machine learning model was applied to TP and PP monitoring based on Sentinel-2 MSI data. • Widespread decrease trends in TP and PP in Taihu Basin’s lakes were observed. • The major ecosystem processes driving phosphorus dynamics were elucidated. A comprehensive monitoring of the spatiotemporal dynamics of total phosphorus (TP) and particulate phosphorus (PP) is vital for mitigating algal blooms and improving lake management. However, it is challenging for characterizing large-scale patterns with traditional in situ methods. As an effective supplement, remote sensing could provide accurate estimation of non-optically active phosphorus, while remains difficult due to limited satellite-ground synchronization data. To address this gap, we established a predictive framework utilizing the Extreme Gradient Boosting (XGBoost) algorithm to retrieve TP and PP using a large in situ dataset and Sentinel-2 MSI imagery (2016−2023) across lakes in the Taihu Basin. The model achieved a mean absolute percentage error of 28.2% for TP and 28.2% for PP on an independent validation dataset. Meanwhile, major ecosystem processes driving phosphorus dynamics were elucidated through three representative situations: riverine input, wind-driven resuspension, and algal blooms. Distinct spatial and seasonal patterns were observed for TP and PP in these lakes with higher values in the west, upstream, and summer. Notably, a widespread significant decrease trend in phosphorus concentration was observed in the lakes ( P < 0.05). Our findings highlight that the joint controls of external phosphorus loads and in-lake phytoplankton biomass could be vital for algal bloom mitigation. These results indicate a remarkable improvement in phosphorus control and eutrophication management.