Evaluation of uncertainties in regional climate change simulations

We have run two regional climate models (RCMs) forced by three sets of init ial and boundary conditions to form a 2x3 suite of 10-year climate simulati ons for the continental United States at approximately 50 km horizontal res olution. The three sets of driving boundary conditions are a reanalysis, an atmosphere-ocean coupled general circulation model (GCM) current climate, and a future scenario of transient climate change. Common precipitation cli matology features simulated by both models included realistic orographic pr ecipitation, east-west transcontinental gradients, and reasonable annual cy cles over different geographic locations. However, both models missed heavy cool-season precipitation in the lower Mississippi River basin, a seemingl y common model defect. Various simulation biases (differences) produced by the RCMs are evaluated based on the 2x3 experiment set in addition to compa risons with the GCM simulation. The RCM performance bias is smallest, where as the GCM-RCM downscaling bias (difference between GCM and RCM) is largest . The boundary forcing bias (difference between GCM current climate driven run and reanalysis-driven run) and intermodel bias are both largest in summ er, possibly due to different subgrid scale processes in individual models. The ratio of climate change to biases, which we use as one measure of conf idence in projected climate changes, is substantially larger than I in seve ral seasons and regions while the ratios are always less than I in summer. The largest ratios among all regions are in California. Spatial correlation coefficients of precipitation were computed between simulation pairs in th e 2x3 set. The climate change correlation is highest and the RCM performanc e correlation is lowest while boundary forcing and intermodel correlations are intermediate. The high spatial correlation for climate change suggests that even though future precipitation is projected to increase, its overall continental-scale spatial pattern is expected to remain relatively constan t. The low RCM performance correlation shows a modeling challenge to reprod uce observed spatial precipitation patterns.