Analysis of GFDL intensity forecasting performance for Gulf/Caribbean major hurricanes from 1995-2008

While both computer model and human track forecasts of Atlantic Basin hurricanes have improved significantly over the past several decades, forecasting of tropical cyclone intensity continues to prove difficult. This study investigates the intensity forecasting performance of the GFDL Hurricane Model for 19 major hurricanes over the Gulf of Mexico and Caribbean Sea over the time period from 1995 to 2008. For each storm, the forecast errors at 0, 24, 48, 72, 96, and 120 hours are analyzed, and these errors are broken down into five categories: average absolute error, average overestimate, average underestimate, maximum overestimate, and maximum underestimate. It is found that, for these systems as a whole, the overall average error ranged from a minimum of 15.1 knots at 24 hours to a maximum of 34.5 knots at 120 hours. Overall, the average underestimates were 5-15 knots worse than the average overestimates, and the maximum underestimates were 15-30 knots worse than the maximum overestimates. Comparing the GFDL to the statistical SHIPS intensity guidance, it is found that the GFDL slightly outperformed SHIPS on some of the overall and underestimate forecast times, but had significantly higher overestimates than SHIPS at all forecast times. By comparing the GFDL errors to the maximum intensity of the storm, it seems that there is a possible correlation between higher intensity and higher errors, particularly for the error category of maximum underestimates.

After looking at the statistics, 8 of the 19 hurricanes were identified as unusually challenging storms, based on an average 24-hour underestimate error of over 20 knots or a maximum 24-hour underestimate of over 45 knots. One of the key apparent factors in these poorly forecasted storms was rapid intensification (RI), previously defined as a 24-hour intensity change of more than 30 knots. The reasons for this RI, as well as its effects on the intensity errors, are explored. It is hoped that this and further analysis of these systems will illuminate some of the reasons for the model's difficulty with the intensity forecasts, and provide a basis for improving model forecasts and official forecasts of tropical cyclone intensity.