Examination of Convection-Allowing Model Forecasts of Convective Initiation in Cases with Missed and False Alarm Mesoscale Convective Systems

Convection-allowing models often have difficulty predicting some details of Mesoscale Convective Systems (MCSs), with convective initiation (CI) being an area of particular interest. For this study, the 06, 12, and 18 UTC runs of the High-Resolution Rapid Refresh (HRRR) model were investigated for occurrences of both missed and false alarm MCS predictions from June through September 2018-2021. A preliminary analysis comparing HRRR forecasts to observed MCS events found that 10, 12, and 16 of the 40 HRRR runs examined for each time (06, 12, and 18 UTC) generated convection that did not grow upscale, respectively, and were thus missed MCS forecasts. When analyzing HRRR runs on days with no observed MCSs, it was determined that there were 12, 10, and 12 of the 240 days examined for each time (06, 12, and 18 UTC) when HRRR incorrectly generated an MCS, respectively.

To further examine the problems in these simulations, five missed and five false alarm MCS cases were selected, in addition to five cases where the HRRR correctly predicted an MCS. Each case was re-run for more detailed analysis using the Weather Research and Forecasting (WRF) model with four different sets of initial conditions (ICs). These included the reanalysis products of the Global Forecast System (GFS), the North American Mesoscale Forecast System (NAM), the Rapid Refresh (RAP), and the European Center for Medium-Range Forecasts Reanalysis v5 (ERA5). The WRF simulations were then compared to radar reflectivity to determine the model accuracy of predicting MCSs. Furthermore, numerous environmental parameters regarding moisture, instability, lifting, and shear were analyzed and compared to RAP analysis (used to represent observations) the hour prior to CI to determine causes for the errors in the MCS forecasts. For the cases where HRRR correctly showed an MCS, all but one of the WRF simulations produced an MCS. Timing errors were not particularly large, with CI in the WRF runs occurring on average 0.65 hours prior to observed CI. However, location errors were more substantial, with an average distance of 226 km between the simulated CI and that observed. For false alarm and missed MCS cases, under half the simulations produced the correct type of event at 35% and 45%, respectively. The average error of the timing of CI varied greatly depending on initialization for false alarm cases but location errors were less, only 151 km from the observed location. For missed MCS cases, timing of CI in the WRF was on average 0.75 hours after that observed, and displacement errors were 125 km. Environmental parameters were not found to differ noticeably between the WRF output and RAP analyses for correctly simulated MCS cases. However, bigger differences were present in instability and lifting-related parameters for missed and false alarm cases. For false alarm cases, when MUCAPE values were greater in the WRF simulations than observed, and MUCIN values were under 20 J kg^-1, convection tended to grow upscale and produce MCSs, contrary to what was observed. For missed events, simulations had an underproduction of surface convergence compared to observations, which appeared to be the primary reason for the failure to simulate an MCS.