Hurricane Track Prediction Using a Statistical Ensemble of Numerical Models

Abstract

A new statistical ensemble prediction system for tropical cyclone tracks is presented. The system is based on a statistical analysis of the annual performance of numerical track prediction models, assuming that their position errors are systematic and depend on storm structure, location, and motion. For a given tropical cyclone advisory and given model forecasts of a particular storm at any base date and time, the statistical analysis of the model performances in the year preceding the base date and time can be used to produce a track prediction and geographical maps of strike probability distributions at all prediction times. The statistical ensemble prediction system was developed using tropical cyclone advisories, model predictions, and best-track positions of all Atlantic hurricane seasons between 1996 and 2000. Track predictions were carried out in each individual year between 1997 and 2000. The 24-, 48-, and 72-h mean position errors, averaged over the whole time period 1997?2000, were found to be 120, 215, and 296 km, respectively, and showed positive skill (negative relative error) of about 20% relative to all high quality numerical models and approximately equal skill relative to the consensus model of Goerss at all prediction times. Equivalent experiments with an ?operational? ensemble, consisting only of models that were available at the issuing times of official forecasts, resulted in corresponding mean position errors of 128, 238, and 336 km and positive skill of about 5%?15% versus the official forecasts. A major characteristic of the new track prediction system lies in the automatic production of geographical strike probability maps. Mean 1997?2000 diameters of the 66.7% strike probability regions of 274, 535, and 749 km (304, 682, and 1033 km in the case of the operational ensemble) at 24-, 48-, and 72-h prediction time and good agreement between the observed percentages of storm positions inside regions of given strike probabilities with the corresponding predicted percentages, document the potential of the new system with regard to operational tropical cyclone track prediction.