Hydropower is a significant renewable energy source for Nepal, contributing to its electricity generation and economic development. However, climate change-induced fluctuations in river discharge concerns long-term hydropower sustainability. This study investigates the effects of climate change on the hydropower potential of Nepal’s Sunkoshi basin using a physically based hydrological approach (SWAT) compared with a machine-learning method (Bi-directional LSTM). In the face of increasing global energy consumption and calls for sustainable development, the paper utilizes long historical hydrological records (1980–2015) and future climates in the simulation of river flow for the period of 2024–2050. While the physically realistic data-driven SWAT approach captures physical watershed processes, the BiLSTM exploits the pattern of historical flow for future flow forecast. Both models forecast a nearly 48% reduction in the average flow in the historical period, with significant rises in the duration of low flow with total hydrological variability. Although the correlation coefficient (r = 0.99) for the relationship between the two approaches in predicting the yields of hydropower energy is very high, the results in the forecast of events at extremes conflict: while the SWAT overestimates the peak flows, the BiLSTM offers smoother curves. The paper emphasizes the contribution of multi-model approaches towards hydrological forecasting and highlights the importance of planning for adaptation in the face of changing climatic conditions, for the requirement of adaptation measures, investment in resilient infrastructure, as well as updating policies to maximize the utilization of hydropower in the face of changing climatic conditions. The SWAT model achieved strong calibration performance with R2 = 0.91 and Nash-Sutcliffe Efficiency (NSE) = 0.82, while the BiLSTM model demonstrated superior short-term accuracy with Test MSE = 0.0006, MAE = 0.0099, and RMSE = 0.0246. Energy output projections suggest hydropower generation could decline significantly, emphasizing the need for resilient infrastructure, adaptive policy reforms, and hybrid renewable energy integration. This study underscores the necessity of multi-model approaches for hydrological forecasting and provides critical insights for climate-resilient hydropower planning in Nepal. The findings are instrumental for policymakers, engineers, and researchers aiming to enhance energy security and sustainable development. In this study, the rationale for selecting SWAT lies in its strength to interpret physical watershed processes, while BiLSTM was chosen for its ability to capture short-term temporal dependencies in hydrological time series. The regional significance of this study is emphasized by Nepal’s reliance on hydropower for over 90% of electricity, making accurate discharge forecasting vital for national energy security.



