One of the most prominent climate changes over the central United States during the late 20th and early 21st centuries is the increasing incidence of extreme precipitation (daily rates exceeding 102 mm) despite only a modest increase in mean annual precipitation. Here we evaluate the ability of an ensemble of regional climate models (RCMs) to reproduce observed frequency of extreme warm-season precipitation over the central U.S. and use the same models to project future changes in extreme precipitation. In support of the CORDEX-North America and DOE FACETS projects we have performed a suite of dynamically downscaled simulations over North America using two RCMs (WRF-ARW and RegCM4), each of which was run using three grid spacings (50 km, 25 km, and 12 km), and each driven by three global models (HadGEM2-ES, MPI-ESM-LR, and GFDL-ESM2M). This produces a systematic matrix of 18 simulations in a three-dimensional parameter space (RCM, resolution, and GCM) allowing clearer interpretation compared to uncontrolled RCM ensembles of opportunity. Results show complex interactions among the RCM, resolution, and the driving GCM; for example, RegCM4 is more sensitive to the choice of driving GCM than is WRF, with this sensitivity in turn varying with resolution. We also found some unexpected results such as a tendency for the frequency of heavy precipitation to decrease with decreasing grid spacing in RegCM4. Nevertheless, the trend in precipitation extremes is robust: warm-season extreme precipitation over the central U.S. becomes more common in future climates for every model configuration that we tested.

This research was sponsored in part by the U.S. Department of Energy and by the USDA National Institute of Food and Agriculture.