Deep learning-based prediction of flood hazard and future streamflow changes in the Brahmaputra River Basin under CMIP6 climate change scenarios

The Brahmaputra River Basin (BRB), a large climate-sensitive transboundary river system in South Asia. This study develops a climate-driven deep learning framework to simulate and project streamflow in the basin by linking observed daily discharge with key hydro-meteorological variables. A hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model is implemented for spatiotemporal modeling, capturing both spatial variability in precipitation and long-term temporal dependencies in streamflow dynamics. The hybrid model is compared with standalone deep learning architectures to identify the best-performing approach. Additionally, Shapley Additive Explanations (SHAP) are applied to interpret the model and quantify the contribution of each climatic variable to the predicted streamflow. The CNN–LSTM model achieved Nash–Sutcliffe Efficiency (NSE) values of 0.92 during training and 0.86 during testing, outperforming the standalone deep learning models. SHAP analysis indicates that precipitation is the dominant driver of streamflow variability, while temperature provides a secondary influence, and evaporation and snow contribute minimally. Future projections using bias-corrected CMIP6 climate scenarios (SSP2–4.5 and SSP5–8.5) suggest substantial increases in mean and extreme flows, with the 100-year return period discharge projected to rise by about 25%. These findings demonstrate that climate-driven hybrid deep learning models can provide interpretable and reliable long-term streamflow projections, supporting improved flood-risk assessment and climate adaptation planning in the basin. • CNN–LSTM predicts Brahmaputra River Basin streamflow using only climate data. • CNN–LSTM model outperformed standalone deep learning models. • Precipitation is the key driver of streamflow variability. • Future CMIP6 projections indicate intensified high flows.