J3.2 Extended Range Assessment of Forecast Skill for Severe Convective Storm Environments in GEFSv12 Reforecasts

Monday, 29 January 2024: 2:00 PM
Holiday 6 (Hilton Baltimore Inner Harbor)
Andrew Berrington, Cooperative Institute for Severe and High-Impact Weather Research and Operations (CIWRO), Norman, OK; and A. J. Clark and K. A. Hoogewind

While prediction of severe convective storms (SCSs) on sub-seasonal timescales has become a topic of increasing interest to the research and forecast communities in recent years, estimates of the skill in predicting variables relevant to SCSs in forecast models have been less explored. In this study, 20 years of weekly Global Ensemble Forecast System Version 12 (GEFSv12) reforecasts over the CONUS are scrutinized using a variety of established and novel metrics. As this data is provided for lead times out to 35 days, it allows us to establish a climatology of forecast skill in the extended range for these variables.

Common variables used in the prediction of SCSs at shorter lead times such as convective available potential energy and composite SCS parameters struggle at longer lead times, with skill dropping below climatology before week 2. Synoptic variables such as 500 hPa geopotential height are more skillful, with skill exceeding climatology extending well into week 2 and even week 3 at times. With the motivation of minimizing the role of timing differences concerning key features between ensemble members at longer leads, generalization of forecasts using rolling means into 3-, 5-, or 7-day windows improves objective verification using deterministic and probabilistic metrics. Additionally, a novel method of using weekly aggregations of composite parameters to probabilistically analyze the potential for patterns favorable for SCSs in the extended range is employed. It is found through reliability, receiver-operator, and Brier tests that forecast improvements for composite parameters can be achieved through aggregation over longer time scales. These weekly summations may be used as predictors for weekly SCS activity through practically perfect hindcasts and subsequent machine learning experiments may be informed by the results of this study in order to develop SCS forecasts beyond current operational capabilities.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner