Remote sensing-enabled machine learning for modeling turbidity in regulating lakes and reservoirs across inter-basin under Google Earth Engine

Study region Two lakes and six reservoirs (Lake Nansi, Lake Dongping, Xiashan Reservoir, Menlou Reservoir, Mishan Reservoir, Datun Reservoir, Donghu Reservoir, and Shuangwangcheng Reservoir) along the Eastern Route of the South-to-North Water Diversion Project. Study focus A turbidity retrieval model was developed using Landsat OLI imagery and four machine learning models (RF; XGBoost; SVR; MLP) to analyze the spatiotemporal variations during 2013–2023. Furthermore, meteorological, land use, and socio-economic data were integrated to identify natural and anthropogenic factors driving long-term water quality evolution. New hydrological insights for the region (1) Machine learning models showed superior accuracy and generalizability in turbidity estimation, with RF achieving the best performance (R² = 0.89, RMSE = 0.10 NTU). (2) From 2013–2023, turbidity across all lakes and reservoirs exhibited a decreasing trend (-0.26 NTU/year). Spatially, the turbidity of mixed lakes (14.33–17.41 NTU) > mixed reservoirs (11.07–16.48 NTU) > regulating reservoirs (10.18–15.17 NTU). (3) During water diversion period, turbidity was lower in mixed reservoirs (11.69 vs. 13.58 NTU) but higher in mixed lakes (16.29 vs. 14.35 NTU) compared with non-diversion period, whereas regulating reservoirs exhibited no significant differences (12.88 vs. 12.50 NTU). (4) Turbidity variations across different lake and reservoir types were driven by the combined effects of natural and anthropogenic factors, although the dominant drivers varied significantly among waterbody types.