In the western United States, 27% of electricity demand is met by hydropower, so power system planners have a key interest in predicting hydropower availability under changing climate conditions. Large-scale projections of hydropower generation are often simplified based on regression relationships with runoff and they are not always ready to inform power system models due to the coarse time scale (annual) or limited number of represented power plants. We developed an enhanced process-based hydropower model to predict future hydropower generation by addressing the commonly under-represented constraints, including (1) the ecological spills, (2) penstock constraints to provide flexibility in electricity operations, and (3) biases in hydro-meteorological simulations. We evaluated the model over the western United States under two emission scenarios (RCP4.5 and RCP8.5) and ten downscaled global circulation models. At the annual time scale, potential hydropower generation is not projected to change substantially, except in California. At the seasonal time scale, systematic shifting of the generation patterns can be observed in snowmelt-dominated regions. These projected annual and regional trends are comparable to other regression-based relationships. However, the representation of more complex operations and constraints tend to reduce the uncertainties inherent to climate projections at seasonal scale. In the Pacific Northwest region where hydropower is the dominant electricity source, our predicted future change of hydropower generation is about 10% less than the regression-based projections in spring and summer. The model can also capture the seasonal non-stationary in hydrologic changes. The spatio and temporal scales of the model, increased accuracy and quantification of uncertainty allow one to use the products to inform power system models toward supporting energy sector planning activities and water-energy trade-offs.