Evaluation of Model Guidance for Two High-Impact Flooding Events in July 2022

The Model Evaluation Group (MEG) of the NCEP/Environmental Modeling Center (EMC) evaluated the performance of operational convection-allowing models (CAMs) and ensembles (CAEs) during two high-impact flash flooding events in the Midwest United States in late July 2022. One event brought intense rainfall in a narrow band through the Saint Louis metro area, with local amounts exceeding 10 inches. The other event occurred over complex terrain in and around southeastern Kentucky, over already-saturated soils. Both events caused over one billion dollars in combined damages.

The MEG compared probabilistic quantitative precipitation forecasts (QPFs) and flash flood guidance from the High Resolution Ensemble Forecast system (HREF), as well as QPFs produced by the 3-km CONUS nest of the North American Mesoscale Model (NAM Nest), the High Resolution Rapid Refresh (HRRR), and three configurations of the High Resolution Windows, which comprise the HREF membership.

While both events had similar impacts, the Kentucky event was less predictable due to its more complex synoptic and mesoscale setup. Specifically, probabilistic guidance was less useful for the Kentucky event owing to the more subtle forcing and relative absence of distinct boundaries. Additionally, most CAMs tended to forecast too much scattered precipitation over Kentucky, and only the NAM Nest maximum QPF approached the observed maximum precipitation by the Day 1 lead time. Modeled QPFs tended to be too low overall for both cases, but they were more skillful at longer lead times for the Saint Louis case. Additionally, HREF probabilistic QPF and flash flood guidance were useful for the Saint Louis event with essentially two days of lead time on the potential for a heavy rainfall event. However, displacement errors were practically significant for the highly populated Saint Louis metropolitan area. In both the Saint Louis and Kentucky cases, the HRRR short-range forecast skill varied but often trended toward worse placements of key features in the shortest lead times. Additionally, HREF probability-matched and local probability-matched means were useful for correctly restoring peak values to the HREF mean in both cases.

The MEG suggests that events like these are ideal candidates for improved very-short-term guidance from the development of the Warn-on-Forecast System (WoFS), and the addition of 06 UTC and 18 UTC HREF cycles could improve the utility of CAE products.