Hailcast was originally developed to forecast hail in South Africa and Alberta, where hail diameters larger than 2 are rare. In this study, a database of 382 severe (>= 0.75 diameter) hail proximity soundings within the contiguous U.S. was collected from 1997 to 2002. Unlike the Alberta database, this database includes 203 significant (>= 2 diameter) cases. Assuming severe hail of any size is likely, the model shows relatively high skill at delineating between low-end severe (< 2) and significant (>= 2) hail.
The correlation between observed and forecast maximum hail diameter ranges from 0.65 to 0.74 depending on how reports are filtered. These numbers are in general agreement with past evaluations by Brimelow. Dividing the database into < 2" (NON-SIG) and > 2" (SIG) cases, the model is able to achieve a very high detection rate near 0.89 while falsely detecting SIG hail only 23% of the time. It is noteworthy that the model produces accurate forecasts despite large variability in thermodynamic environments in the U.S. large CAPE ranges), unlike those commonly observed in Alberta.