Environ Monit Assess. 2026 Aug 1;198(8):900. doi: 10.1007/s10661-026-15759-3.
ABSTRACT
The Danjiangkou Reservoir is the drinking-water source of the South-to-North Water Diversion Middle Route and a national Class I drinking-water protection area. Ensuring low turbidity is essential for safeguarding downstream public health and national water security. However, turbidity retrieval in large inland reservoirs using remote sensing is often constrained by limited in situ samples and weak model generalization across seasons and years. We integrated 49 GaoFen-1 (GF-1) satellite images with in situ measurements from the national water quality network to construct a matchup dataset for model development and validation from 2020 to 2024. Using this matchup dataset, five modeling approaches were compared under a time-ordered validation framework to identify a suitable model for predicting in situ turbidity from GF-1 spectral features. Among them, the sparrow search algorithm (SSA)-optimized extreme gradient boosting (XGBoost) approach achieved the best performance and was selected for retrospective quarterly mapping from 2013 to 2024. Across 45 quarterly clear-sky scenes, 95.5% of all retrieved turbidity values were below 15 nephelometric turbidity units (NTU), while persistent spatial heterogeneity and recurrent inflow-zone hotspots remained evident. Scene-wise diagnostics indicated stable performance across acquisition dates, with per-scene mean absolute error (MAE) < 4 NTU for all 49 scenes and 61.2% of scenes falling within 0-2 NTU. Precipitation showed the clearest association with turbidity and exhibited a region-dependent lagged response within the 52-h pre-acquisition window, with the strongest response observed in the precipitation-sensitive zone about 28-51 h before image acquisition (r = 0.36). Water temperature showed a moderate negative relationship with turbidity (r = – 0.47). These results provide spatial evidence for long-term turbidity surveillance, recurrent hotspot identification, and post-storm monitoring in large source-water reservoirs.
PMID:42542499 | DOI:10.1007/s10661-026-15759-3
