The balmy airmass across much of the eastern U.S. provided the fuel for severe storms across the Great Lakes and Mid-Atlantic regions for a while now.

ProbSevere v4 (PSv4) uses spatial AI methods and MRMS, lightning, and HRRR environmental information to predict the probability of severe weather in the next 45 minutes. NSSL is working on extending probabilistic predictions out to 6 hours using these data sources and more.

These storms in Michigan and Ohio formed along and traversed a stationary boundary of high theta-e gradient. The PSv4 model did a commendable job highlighting the severe-wind producing segments, particularly in Michigan and eastern Ohio and southwest Pennsylvania. Once storms weakened in reflectivity and lightning activity in northern Ohio, probabilities also weakened (< 25%), but there were sporadic severe wind reports.

Later on, the AI model correctly captured numerous severe wind reports (plotted 45 minutes after image time) in the main storm segments in West Virginia and Virginia/Maryland.

On the Pennsylvania / New Jersey border, we see a small cluster of storms that had high probabilities ahead of it. The tree-based PSv3 never achieved a probability greater than 17%, whereas PSv4 had > 75%. At this time, 4 minutes before a severe thunderstorm warning was issued, all of the reports in the next 45 minutes were within the ≥ 50% contour.

On September 5th, moderate shear, excellent CAPE, and a triggering cold front produced numerous damaging wind gusts. PSv4 again did well with these convective clusters, particularly in the Carolinas.