Hurricane ensemble prediction using EOF-based perturbations

In this study, a method to generate perturbations for hurricane ensemble prediction is proposed and examined on five hurricane cases with different kinds of tracks. The model used is the Florida State University Global Spectral Model (FSUGSM) with horizontal spectral resolution of T63 and 14 vertical levels.; The method proposed here is based on the premise that (a) model perturbation grows linearly during the first few days of model integration; and (b) in order to make a complete set of ensemble perturbations of hurricane forecasts, both initial intensity and position of the hurricane need to be perturbed. The initial position of the hurricane is perturbed by displacing its original position 50 km equally toward the north, south, east and west directions. The fast growing perturbations can be generated by implementing EOF analysis to the differences between forecasts starting from regular analysis and randomly perturbed analysis. The eigenmode with the largest eigenvalue is then considered as the fast growing perturbation.; The proposed perturbation method has been examined through five hurricane case studies. The results show that EOF based perturbations are indeed the optimal perturbations for hurricane ensemble forecasting compared to Monte Carlo Forecasting method. A comparison has also been made between the control experiment (single forecast from regular analysis) and the ensemble experiment. The results show that the predicted hurricane position errors are largely reduced by the ensemble prediction for most of the hurricane cases that have been tested, compared with the control experiment.; A higher horizontal resolution model T106 is performed on one hurricane case (Andrew) to provide a comparison between different resolution models.