A simulation of 20 warm season MCS events in the Upper Midwest was performed using a 10 km grid spacing version of the Eta model, with both the Betts-Miller-Janjic (BMJ) and Kain-Fritsch (KF) convective parameterizations. Analysis of the skill scores for these cases indicates large variability in predictability, considering both the different magnitudes of large scale forcing (e.g., frontogenesis, vorticity advection) and the impacts of using different convective parameterizations.

Taking into consideration the extreme cases (those with unusually ''low" and "high" skill scores), some general conclusions about the behavior of different convective parameterizations can be drawn. To improve the analysis of some representative cases, one-hour model outputs were used.

For one case (June 28, 1998 at 00Z) with high skill scores for both convective schemes, the synoptic situation was characterized by a cold front moving from west to east into the domain of integration. Also, it was found, based on hourly precipitation fields, that the BMJ convective scheme did a much better job in detecting the MCS present at the time of initialization. This can be explained by the fact that the initialization for all cases used 40 km grid spacing Eta model output from NCEP, whose assimilation system also uses the BMJ scheme. On the other hand, application of a "relative humidity" adjustment (elimination of dry layers where radar echo was present) to the initial conditions of the model, greatly improved the forecast using the KF convective parametrization.

A good example of a "poorly" predicted event was June 04, 1999, initialized at 12Z. The synoptic situation for this event was characterized by both a surface low over southern Canada with its associated fronts over the integration domain, and a stationary front oriented from north-west to south-east over the southwestern part of the domain (across Kansas and Nebraska). An additional area of thunderstorms formed around 12Z during the morning behind a complex which was propagating in an east-south-east direction through central and southern Iowa. Both convective schemes failed to detect this secondary MCS. The applied "relative humidity" adjustment to the BMJ convective schemes produced an improvement in detecting the "second" part of the system, but still the model was not able to sustain it. Finally, the agreement between observed and initialized wind fields was examined. Due to disagreement between these two fields, perturbations in the height and wind fields were added to force the initialization to better resemble the observed data. Unfortunately, this also did not improve the rainfall forecast.

Work is ongoing to similarly analyze several additional poorly predicted and well-predicted cases.