Monitoring chlorophyll-a (Chl-a) dynamics in estuarine reservoirs is essential for assessing eutrophication and algal bloom risk under climate change. This study developed an integrated prediction framework combining Sentinel-2 derived Chl-a, the Soil and Water Assessment Tool (SWAT), and a Long Short-Term Memory (LSTM) network to evaluate future Chl-a dynamics in Namyang Reservoir, South Korea, under SSP1–2.6 and SSP5–8.5 scenarios. The SWAT model was calibrated and validated to simulate inflow and water quality variables, including suspended solids, total nitrogen, total phosphorus, and water temperature and parameter induced uncertainty was further assessed using 95% prediction uncertainty. These outputs, integrated with meteorological variables and Sentinel-2 derived Chl-a data, were used to train site-specific LSTM models for grid-based forecasting, and SHapley Additive exPlanations (SHAP) were applied to interpret predictor importance. The SWAT model satisfactorily reproduced hydrological and water quality conditions, while the LSTM framework captured the overall magnitude and temporal variability of Chl-a across the reservoir. SHAP analysis identified water and air temperature as dominant predictors, with nutrient related variables showing higher influence in the midstream region and meteorological factors dominating the downstream region. Under future climate scenarios, SSP5–8.5 produced higher Chl-a concentrations, intensified seasonal peaks, and expanded upstream hotspots, indicating increased eutrophication risk. The integrated framework provides a practical tool for climate responsive reservoir management and adaptive planning in vulnerable systems.
Jang et al. (Thu,) studied this question.