To study the impacts of climate change on water resources in the western U.S., global climate simulations were produced using the National Center for Atmospheric Research/Department of Energy (NCAR/DOE) Parallel Climate Model (PCM). The Penn State/NCAR Mesoscale Model (MM5) was used to downscale the PCM control (1995-2015) and three future (2040-2060) climate simulations to yield ensemble regional climate simulations at 40 km spatial resolution for the western U.S. This paper describes the regional simulations and focuses on the hydroclimate conditions in the Columbia River (CRB) and Sacramento-San Joaquin River (SSJ) Basins. Results based on global and regional simulations show that by mid-century, the average regional warming of 1-2.5^oC strongly affects snowpack in the western U.S. Along coastal mountains, reduction in annual snowpack was about 70% as indicated by the regional simulations. Besides changes in mean temperature, precipitation, and snowpack, cold season extreme daily precipitation increased by 5 to 15 mm/day (15-20%) along the Cascades and the Sierra. The warming resulted in increased rainfall over snowfall and reduced snow accumulation (or earlier snowmelt) during the cold season. In the Columbia River Basin, these changes were accompanied by more frequent rain-on-snow events. Overall, they induced higher likelihood of wintertime flooding and reduced runoff and soil moisture in the summer. Such changes could have serious impacts on water resources and agriculture in the western U.S. Changes in surface water and energy budgets in the Columbia River and Sacramento-San Joaquin basins were affected mainly by changes in surface temperature, which were statistically significant at the 0.95 confidence level. Changes in precipitation, while spatially incoherent, were not statistically significant except for the drying trend during summer.

Differences in climate signals derived from the downscaled ensemble simulations were 10-40% of the ensemble mean for temperature and 0-200% for precipitation. For precipitation, especially in the Sacramento-San Joaquin Basin, the interannual variation is too large to yield a high-level confidence in the signal. In addition to estimating uncertainty using ensemble simulations, comparison of the MM5 results with the National Center for Environmental Prediction (NCEP) Regional Spectral Model (RSM) simulations (which were driven by a different set of PCM climate change scenarios, but with the same CO₂ and aerosol forcings and for the same period as the PCM simulations used to drive MM5) indicates that different models can yield large differences in climate projections. This suggests that simulations with multiple models as well as larger ensemble size with single model are needed to reduce uncertainty in regional climate projections. Various regional ensemble techniques need to be further explored to provide better estimates of uncertainty in climate change signals. Because snow and runoff are highly sensitive to spatial distributions of temperature and precipitation, this study further shows that (1) downscaling provides more realistic estimates of hydrologic impacts in mountainous regions such as the western U.S., and (2) uncertainty in projecting climate changes can result in larger uncertainty in projecting water resources impacts.