Climate simulations have projected an increase in extreme precipitation during the winter months in much of the continental United States. Determining the impact of model resolution on extreme precipitation is an important part of developing confidence in these projections. The winter months are well suited for this analysis as weather systems producing extreme precipitation events have synoptic scales that regional climate models can resolve. We focus on sub-daily (six-hourly) extreme precipitation as previous work shown that models with 50-km or greater grid spacing cannot produce well such extremes.

This study focuses on the Upper Mississippi River Valley during December, January, and February (DJF). We analyze six-hourly precipitation for 2002-2012 to evaluate the effectiveness of the RegCM4 and WRF regional climate models in replicating Stage IV observational extreme precipitation at grid spacings of 12, 25 and 50 km. We also evaluate 2-m temperature, 2-m specific humidity, 500 hPa geopotential heights, and 10-m winds to assess the physical conditions leading to extreme precipitation events. The 700 hPa vertical motion, 2-m temperature gradients, 2-m moisture gradients, and 10-m wind convergence are also analyzed to examine aspects of frontal strength and rising motion. The convective and resolved-scale precipitation are analyzed to illustrate the dynamical differences between WRF and RegCM4 that may aid or hinder extreme precipitation development.

As resolution increases, WRF tends to replicate better the observed behavior compared to RegCM4, especially for extreme precipitation. WRF produces strong temperature gradients, horizontal convergence and vertical motion during extreme events, especially at finer resolutions. WRF also produces substantially more convective precipitation during extreme events, which may be linked to the strong vertical motions. The outcome is extreme six-hourly precipitation that agrees well with the observed extremes and, more generally, the observed precipitation vs. frequency distribution, especially when using 12-km grid spacing.