Refinements to Atlantic basin seasonal hurricane prediction from 1 December

Abstract

Atlantic basin seasonal hurricane predictions have been issued by the Tropical Meteorology Project at Colorado State University since 1984, with early December forecasts being issued every year since early December 1991. These forecasts have yet to show real‐time forecast skill, despite several statistical models that have shown considerable hindcast skill. In an effort to improve both hindcast skill and hopefully real‐time forecast skill, a modified forecast scheme has been developed using data from 1950 to 2007. Predictors were selected based upon how much variance was explained over the 1950–1989 subperiod. These predictors were then required to explain similar amounts of variance over a latter subperiod from 1990 to 2007. Similar amounts of skill were demonstrated for each of the three predictors selected over the 1950–1989 period, the 1990–2007 period, and the full 1950–2007 period. In addition, significant correlations between individual predictors and physical features known to affect hurricanes during the following August–October (i.e., tropical Atlantic wind shear and sea level pressure changes, ENSO phase changes) were obtained. This scheme uses a new methodology where hindcasts were obtained using linear regression and then ranked to generate final hindcast values. Fifty‐four percent of the variance was explained for seasonal Net Tropical Cyclone (NTC) activity over the 1950–2007 period. These hindcasts show considerable differences in landfalling U.S. tropical cyclones, especially for the Florida Peninsula and East Coast. Seven major hurricanes made Florida Peninsula and East Coast landfall during the top 15 largest NTC hindcasts compared with only two major hurricane landfalls in the bottom 15 smallest NTC hindcasts.