Improving US extreme precipitation simulation: sensitivity to physics parameterizations

Climate models tend to underestimate rainfall intensity while producing more frequent light events, leading to significant bias in extreme precipitation simulation. To reduce this bias and better understand its underlying causes, we tested an ensemble of 25 physics configurations in the regional Climate-Weather Research and Forecasting model (CWRF). All configurations were driven by the ECMWF-Interim reanalysis and continuously integrated during 1980-2015 over the contiguous United States with 30-km grid spacing. Together they represent CWRF's ability to simulate characteristics of US extreme precipitation, and their spread depicts the structural uncertainty from alternate physics parameterizations. The US extreme precipitation simulation was most sensitive to cumulus parameterization among all physics configurations. The ensemble cumulus parameterization (ECP) was overall the most skilled at reproducing seasonal mean spatial patterns of daily 95th percentile precipitation (P95). Other cumulus schemes severely underestimated P95, especially over the Gulf States and the Central-Midwest States in convective prevailing seasons. CWRF with ECP outperformed the driving reanalysis, which substantially underestimated P95 despite its daily atmospheric moisture data assimilation. The CWRF improvement over ERI is much larger in warm than cold seasons. Changing alone ECP closure assumptions produced two distinct clusters of convective heating/drying effects: one altered P95 mainly by changing total precipitation intensity and another by changing rainy-day frequency. Microphysics, radiation, boundary layer, and land surface processes also impacted the result, especially under mixed synoptic and convective forcings, and some of their parameterization schemes worked with ECP to further improve P95.