Hybrid SWAT-ANN Modeling of Climate-Driven Changes in Streamflow and Sediment Yield: Manjira River Basin, India

Accurate prediction of sediment yield and streamflow is essential for effective watershed management and climate change adaptation planning. This study develops and validates an innovative SWAT-ANN hybrid model that integrates the physically based Soil and Water Assessment Tool (SWAT) with Artificial Neural Networks (ANN) to improve hydrological predictions in the monsoon-dominated Manjira River Sub-Basin (MRSB), India.

The SWAT model was calibrated and validated using daily streamflow and sediment observations from three Central Water Commission gauging stations (1998-2019). Multi-site calibration achieved satisfactory performance with NSE = 0.75 and R² = 0.79 for streamflow, while sediment yield modeling yielded NSE = 0.56 and R² = 0.60. Building on these simulations, an ANN model was trained using SWAT-generated outputs combined with meteorological variables to capture nonlinear sediment transport relationships. The SWAT-ANN hybrid model demonstrated significant improvements, with streamflow predictions achieving NSE = 0.95 and R² = 0.98, compared to standalone SWAT. For sediment yield, the hybrid approach improved NSE from 0.56 to 0.72 and R² from 0.60 to 0.75, showcasing the complementary strengths of physics-based and data-driven modeling.

Climate change impact assessment was conducted using 13 CMIP6 models under SSP245 (moderate mitigation) and SSP585 (high emissions) scenarios. Under SSP245, ensemble mean streamflow increased by 47.5% (2015-2045), 68.5% (2046-2070), and 123.9% (2071-2100) relative to baseline (1998-2014). SSP585 projections were more severe, with streamflow increases of 41.3%, 137.4%, and 269.4% for the respective periods. Sediment yield responses were equally dramatic: SSP245 scenarios projected increases of 61.3% (near-future), 81.9% (mid-future), and 146.6% (far-future), while SSP585 showed 48.3%, 166.4%, and 331.9% increases. The most aggressive model (CanESM5) projected sediment yield increases exceeding 1,900% by 2100 under SSP585, while conservative models (INM-CM5-0) showed minimal changes.

The SWAT-ANN model successfully captured temporal variability in both streamflow and sediment responses across all climate scenarios. These projections indicate that the basin will experience unprecedented hydrological changes, with sediment yields rising 2.5-4.3 times baseline by 2100 depending on emission pathways. The developed hybrid methodology provides a powerful tool for water resource managers to quantify climate-driven changes in streamflow and sediment dynamics, enabling adaptive management strategies and sustainable planning in data-limited monsoon-dominated basins. The transferable methodology addresses critical gaps in sediment yield prediction for similar South Asian river systems.