Technical Abstract: Spatial and temporal mismatches between coarse resolution output of General Circulation Models (GCMs) and fine resolution data requirements of ecosystems models are the major obstacles for assessing the site-specific climatic impacts of climate change on natural resources and ecosystems. The objectives of this study were to (i) present a simple but elegant method for statistically downscaling GCM output of climate change at the native GCM grid scale to station-scale, and (ii) further disaggregate spatially downscaled monthly GCM projections to daily weather series for input to hydrologic and crop models using a stochastic weather generator (CLIGEN), and (iii) demonstrate the site-specific impact assessment of climate change on natural resources at the Changwu station, Shaanxi, China using the Water Erosion Prediction Project (WEPP) model. Monthly precipitation and temperature projected by the U.K. Hadley Centre’s GCM under the A2, B2, and GGa emissions scenario were downloaded for the periods of 1900-1999 and 2010-2039 for the grid box containing the Changwu station. Univariate transfer functions were derived by matching probability distributions between station-measured and GCM-projected monthly precipitation and temperature for the 1950-1999 period. The derived functions were used to spatially downscale the GCM monthly projections of 2010-2039 to the Changwu station. The downscaled monthly data were further disaggregated to daily weather series using a stochastic weather generator (CLIGEN). HadCM3 predicted that average annual precipitation during 2010-2039 would increase by 4 to 18% at Changwu. Simulated wheat and maize yields would increase in response to increases in precipitation. Due to the warming effects, frequency and intensity of large storms would also increase in the area. Under conventional tillage, percent increases under climate change, compared with the present climate, would be 49-112% for runoff and 31-167% for soil loss. It should be pointed out that the predicted runoff and soil loss using the proposed spatial downscaling method were several times greater than those predicted without an explicit spatial downscaling.