89th American Meteorological Society Annual Meeting

Tuesday, 13 January 2009: 4:30 PM
Short range ensemble forecast (SREF) calibrated thunderstorm probability forecasts: 2007–2008 verification and recent enhancements
Room 131A (Phoenix Convention Center)
David R. Bright, NOAA/NWS/NCEP/SPC, Norman, OK; and J. S. Grams
Poster PDF (286.1 kB)
Since 2003, the Storm Prediction Center (SPC) has post-processed the National Centers for Environmental Prediction (NCEP) Short-Range Ensemble Forecast (SREF) to provide operational guidance over the contiguous United States (CONUS) on hazardous mesoscale weather such as severe convective storms and lightning. The National Lightning Detection Network (NLDN) cloud-to-ground lightning database is used as the predictand in the production of calibrated (i.e., statistically reliable) probabilistic forecasts of thunderstorms over the CONUS and its near-shore waters. The SREF runs four times each day (0300, 0900, 1500, and 2100 UTC), allowing new thunderstorm guidance through three and a half days to be created every six hours. This guidance is available to SPC forecasters and is distributed on the SPC web page for use by forecasters in the public, private, and military sectors.

An extensive independent verification of the SREF-based calibrated thunder probabilities commenced on 1 March 2007 and continued into 2008. The SREF guidance was matched to relevant operational forecast periods for the first 24 hours, valid from the issuance time until 0300 UTC and from 0300 UTC to 1200 UTC. The NLDN data served as truth in the verification study. Several metrics were used to verify the forecasts including the Brier score, the ROC score, and a measure of reliability. The forecasts were also verified across the CONUS for a geographical depiction of skill. Overall, the results show the SREF provides useful and skillful guidance for thunderstorm forecasting. Recent upgrades to the post-processing algorithm have been undertaken and are designed to reduce the high-bias over water and adjacent coastlines. These upgrades to the algorithm will be discussed and their potential usefulness shown.

Supplementary URL: