Statistical downscaling is widely used to improve spatial and or temporal distributions of meteorological variables from regional and global climate models. This downscaling is important because climate models are spatially coarse (50-200km), and often misrepresent extremes in important meteorological variables such as temperature and precipitation. However, these downscaling methods rely on current estimates of the spatial distributions of these variables, and assume that the small-scale spatial distribution will not change significantly in a modified climate. Here we compare data typically used to derive spatial distributions of precipitation (PRISM) to a high-resolution (2km) weather model (WRF) under current climate. We show that there are regions of significant difference in November-May precipitation totals (>300mm) between the two, and discuss possible causes for these differences. We then present a simple statistical downscaling based on the 2km WRF data applied to a series of regional climate models (NARCCAP), and compare the downscaled precipitation data to observations at 65 SNOw TELemetry (SNOTEL) sites throughout Colorado for the winter seasons from 1988 to 2000. Finally, we compare statistically downscaled precipitation from a 36km model under an imposed Pseudo Global Warming (PGW) scenario to dynamically downscaled data from a 2km model using the same forcing data. While the statistical downscaling improved the domain average precipitation compared to the original 36km model, the changes in the spatial pattern of precipitation (PGW-current) did not match those of the changes in the dynamically downscaled 2km model. This study illustrates some of the uncertainties in applying statistical downscaling to future climate.
