High-resolution (10 km) Eta simulations of over 20 MCSs in the Upper Midwest are used to study variations in predictability of rainfall among the systems. Cases have been simulated using both the Betts-Miller-Janjic and Kain-Fritsch convective parameterizations, and using three different techniques to improve the initialization of mesoscale features important to later MCS evolution. These techniques include a cold pool initialization, vertical assimilation of surface mesonetwork observations, and relative humidity adjustments based on radar echo coverage. Simulations have been integrated for 24 hour periods over limited domains.

Although all cases occur in the warm season and appear to be influenced by mesoscale forcing mechanisms, precipitation skill scores for the cases vary widely. Some events are highly predictable, with good forecasts from both convective schemes, and additional improvements in the forecasts occurring when mesoscale adjustments are made to the initial conditions. Other cases appear to have little predictability with poor skill scores from all variants of the model. Although on average for all cases, the adjustments to the initial conditions to better represent mesoscale features improve skill scores somewhat, for some cases, the adjustments actually worsen the forecasts. For other cases, the improvements in skill scores are an order of magnitude larger than the average.

All of the events in the study will be classified based on synoptic forcing mechanisms, and thermodynamic characteristics of the larger-scale environment to determine if certain environments are better simulated with a specific model configuration, convective scheme, or adjustment to the initial conditions. Discussion will be given to similarities among the seemingly unpredictable convective systems, and those that were highly predictable.