CAE forecasts are generated by the NCAR’s Model for Prediction Across Scales (MPAS) using NCEP’s Global Ensemble Forecast System (GEFS) as initial conditions. The ensemble has 10 members that are initialized at 0000 UTC each day starting 5 days before the derecho date. The MPAS mesh has 60 km horizontal spacing over the globe decreasing to 3 km over an area centered in the observed derecho wind reports swath. The derecho cases were classified based on their operational predictability as determined using the Storm Prediction Center’s Convective Outlooks. If a Slight Risk (or equivalent) is introduced at least five (three) before the derecho and the forecaster’s discussion contains wording such as “derecho”, “bow echo MCS”, “widespread wind damage”, etc, the predictability is categorized as high (moderate). Otherwise, the predictability is low. There were no high predictability derechos in our 2010-2022 dataset. Twenty four derecho cases were selected for this assessment. Half of the cases had low predictability and half had moderate predictability. The MPAS CAE forecasts of 24-h maximum wind speed at the lowest model level were evaluated objectively using the fractions skill score (FSS) and area under the relative operating curve (AUC). Additionally, the convective mode, displacement of the derecho major axis, timing of derecho initiation, and the wind threat coverage were subjectively evaluated for each ensemble member. Composite maps of ensemble members with more successful forecasts were compared with composites of members with less successful forecasts.
The objective metrics indicate that the MPAS CAE has skill in forecasting strong wind gusts produced by simulated convective storms in the medium range. The skill is lower for the sample of low predictability cases and suggests that these cases have nearly one day less predictive lead time than the moderate predictability cases. The percentages of members producing bow echo MCSs range between 30% and 20% in the 3-5 days forecasts. However, only one third of the members that produce bow echoes (less than 10% of total members) are able to generate a sustained bow echo similar to that observed in the derecho cases. Though the percentage of sustained bow echo members is small in the medium range, it suggests at least a low probability of a derecho-producing MCS, which may be useful to increase awareness of operational forecasters regarding potential for a significant severe weather event in the coming days. Of note, in low predictability cases, medium-range forecasts are more likely to produce discrete supercells (also suggestive of severe storm potential), while in moderate predictability cases more ensemble members produce multicell clusters. Additionally, forecasts of low predictability cases are less likely to produce convection at approximately the right time and location compared to moderate predictability cases. This suggests the low predictability cases may be associated with atmospheric conditions that are more sensitive to synoptic and mesoscale features in triggering convective initiation.

