Unraveling future reservoir health: a hybrid SWAT-CE-QUAL-W2 and machine learning framework under climate and LULC change

This study presents an integrated modelling framework for evaluating future reservoir conditions in the Rengali reservoir, India, by coupling machine learning techniques with the SWAT and CE-QUAL-W2 models. The performance of selected CMIP6 climate models was evaluated using a compromise programming approach, and the best-performing models were combined through a machine learning-based multi-model ensemble. Future LULC was projected using hybrid machine learning models, with the MLP-CA-MC model selected for subsequent analysis. Climate and LULC projections under SSP2-4.5 and SSP5-8.5 for 2026-2050 and 2051-2075 were used to simulate future streamflow, sediment yield, and nutrient loadings using SWAT, and these outputs served as boundary conditions for the CE-QUAL-W2 reservoir model. The simulations project increasing precipitation and air temperature under both scenarios, resulting in higher streamflow, sediment yield, and nutrient loading. Reservoir water temperature and total phosphorus concentration are projected to increase by up to 6% and 27%, respectively, under SSP5-8.5. The simulations also indicate stronger thermal stratification and greater phosphorus accumulation in deeper reservoir layers, with the highest temperature and TP concentrations occurring during late spring and summer. The projected TP concentrations shows that the reservoir is expected to shift from eutrophic towards hypereutrophic conditions under future scenarios.