Evaluation of WRF Forecasts of Severe Weather Environments Against Mobile Upsonde Observations from the Mesoscale Predictability Experiment (MPEX)

Improving numerical weather forecast accuracy continues to be a focal point in better understanding and predicting severe weather environments. Coniglio et al. (2013) conducted a study of thermodynamic behavior using radiosonde observations collected by the National Weather Service at standard launch times of 0000 and 1200 UTC. However, convection often develops at hours besides the standard launch times, and observations at these intermediate times are limited. The Mesoscale Predictability Experiment (MPEX) was conducted from 15 May to 15 June 2013 with the goal of determining the locations and times of additional sounding observations to most improve short-term (6-24h) forecasts of convective storms. Using the mobile ground based upsonde data obtained from MPEX, various thermodynamic properties of the observed environment were compared to five Weather Research and Forecasting (WRF) model ensemble members using different planetary boundary layer (PBL) parameterizations [Mellor-Yamada-Janjic (MYJ), Yonsei University (YSU), Asymmetric Convective Model, version 2 (ACM2), Quasi Normal Scale Elimination (QNSE), and the Mellor-Yamada-Nakanishi-Niino (MYNN)] from the Center for Analysis and Prediction of Storms. From the estimated 290 mobile upsondes conducted by the four MPEX teams [Colorado State University, National Severe Storms Laboratory, Purdue University, and Texas A&M University], a sample of 35 pre-convective soundings were selected and compared with model-predicted thermodynamic profiles. This sample was analyzed for differences in the calculated heights of the PBL, mean-layer convective potential available energy (MLCAPE), as well as a qualitative analysis of the accuracy of upsonde profiles between the observations and WRF model ensemble predictions. PBL height errors varied slightly across the five WRF model ensemble members but remained generally consistent with previous studies. The MYJ and QNSE median PBL heights were calculated to be too shallow while the ACM2, MYNN, and YSU were closer to the observed values, similar findings to the Coniglio et al. study. MLCAPE values, contrary to the findings of the Coniglio et al. study, were found to be above the median values for all five WRF model ensembles, suggestive of differences in instability between the afternoon and the standard 0000 UTC sounding time.