Short term simulations using MM5 V3 and WRF were performed for 40 12h heavy precipitation periods in the Great Plains that occurred during the Spring of 2003. The MM5 and WRF were each run with a coarse grid mesh (36 km grid separation) covering much of the conterminous US surrounding a nested fine-grid mesh (12 km grid separation) over a portion of the Great Plains. The precipitation output at 03, 06, 09, and 12 h was compared to that available from the 20 km RUC run at FSL and the 12 km Eta models run at NCEP using a conventional mean-square error approach as well as a newly developed acuity-fidelity approach (to be discussed in a companion paper).

Both approaches indicated that MM5 performed better than WRF, RUC and Eta. In general, the RUC underpredicted precipitation and the Eta and WRF overpredicted precipitation. MM5 showed skill at all hours evaluated – although the skill decreased with the hour of forecast. The results are compelling – particularly the high skill at 03 and 06 h – because they suggest that the MM5 suffered less from model spin up error than did either RUC or Eta - even though the MM5 was initialized from the Eta analysis with no warm start (e.g., pre-forecast period nudging) – and even though RUC and Eta were initialized with sophisticated model-tailored warm start procedures. Antecedant synoptic and mesoscale conditions and factors as they account for the better performance of MM5 will be discussed.