This study approaches the problem of downscaling global climate model simulations with an emphasis on validating and selecting global models. The downscaling method makes minimal, physically based corrections to the global simulation while preserving much of the statistics of interannual variability in the climate model. Differences among the downscaled results for simulations of present-day climate form a basis for model evaluation. The downscaled results are used to simulate streamflow in the Yakima River, a mountainous basin in Washington, USA, to illustrate how model differences affect streamflow simulations. The downscaling is applied to the output of three models (ECHAM4, HADCM3, and NCAR-PCM) for simulations of historic conditions and two future emissions scenarios (A2 and B2 for 2000-2100) from the IPCC assessment. The ECHAM4 simulation closely reproduces the observed statistics of temperature and precipitation for the 42 year period 1949-90. Streamflow computed from this climate simulation likewise produces similar statistics to streamflow computed from the observed data.
Downscaled climate-change scenarios from these models are examined in light of the differences in the present-day simulations. Streamflows simulated from the ECHAM4 results show the greatest sensitivity to climate change, with the peak in summertime flow occurring 2 months earlier by the end of the 21st century.