Tests of a perturbed physics ensemble approach for regional climate modeling

Ensemble simulations were performed with the regional climate model RegCM2 using ranges of plausible values for two parameters in the deep convection scheme. The timescale for release of convective instability was varied through a range of five values from 600 to 7200 s, and the maximum stable-layer depth between updraft origin and the level of free convection was varied over five values from 50 to 150 hPa, resulting in 25-member ensembles. Simulations were integrated for two 60-day periods covering extremes of regional climate for the central United States, the severe drought in late spring/early summer of 1988 and the record flood of summer 1993. The mean-square error for precipitation was dominated by the term resulting from spatial position correlation. The relative operating characteristic showed that for the flood case the ensemble had skill for 60-day excess precipitation up to one interannual standard deviation above climatology, and for the drought case the ensemble had skill to simulate 60-day accumulated precipitation less than climatology. In both cases, the ensemble mean had superior skill to the reference forecast (i.e., the forecast using the default values of the closure parameters). For the flood case, improved skill was obtained through improvement in probability of detection, while for the drought case the improvement resulted from a decrease in the false alarm rate.