Sediment dynamics under historical and future climate projection scenarios in the Tapi River basin, India

Abstract

Suspended Sediment Load (SSL) plays a crucial role in water resources management, agriculture, infrastructure development, river morphology, and ecological balance. SSL also affects the estuary and marine ecosystem as sediment is a habitat for invertebrates. Furthermore, excessive SSL poses significant challenges upstream of dams by reducing their water storage capacity. A warming climate is expected to influence the streamflow and, subsequently, the SSL of Indian river basins. While extensive research has been conducted to estimate streamflow under historical and future climate projection scenarios, further studies addressing streamflow and SSL dynamics need to be investigated. Recently, Physics Informed Machine Learning (PIML) has shown better performance over individual Physics-based hydrological (PBH) and Machine Learning (ML) models. We employed PBH, ML, and PIML models to predict streamflow and SSL in the Tapi River basin. Our study focused on a ~56,000 km² area to evaluate the impact of SSL on the Ukai dam, the largest dam located approximately 600 km downstream from the river's origin. The Ukai dam features an area of ~612 million m² and a total storage capacity of ~7,414 million m³. We used the Soil Water Assessment Tool (SWAT) as PBH, Long-Short-Term Memory (LSTM) as ML, and SWAT-informed LSTM as the PIML model. Our results show that the PIML model performs best for the historical streamflow and SSL simulation. We then used the generated PIML model to predict streamflow and SSL under future climate scenarios for SSP126 and SSP585. Bias-corrected climate data for future scenarios were derived from the four General Circulation Models (BCC-CSM2-MR, CMCC-ESM2, INM-CM5-0, and NorESM2-MM) included in the Coupled Model Intercomparison Project-6 (CMIP6). These datasets provided projections for precipitation, maximum and minimum temperatures, and wind speed. The models were applied to simulate historical (1951–2014) and future (2015–2100) streamflow and SSL under SSP126 and SSP585 scenarios. Our analysis indicates that SSL and streamflow will increase under the SSP126 and SSP585 scenarios. This increase in SSL will reduce the water storage capacity of the Ukai dam to 54% and 56% under the SSP126 and SSP585 scenarios, respectively. Such reductions in dam capacity and increased streamflow by 39% and 51% for SSP126 and SSP585, respectively, will pose significant challenges in managing extreme flood events in the future. Our findings hold critical implications for water resource management, flood risk mitigation, and the sustainability of river ecosystems.