Probabilistic Prediction of Tropical Cyclones. Part II: Intensity

Abstract

A new objective aid for operational probabilistic intensity (defined as maximum wind speed) prediction of tropical cyclones is presented. Based on statistical analyses of the performance of all operationally available numerical models (using datasets of the U.S. Navy?s Automated Tropical Cyclone Forecasting System) during training periods defined by the years 2000 and 2001, probabilistic and, as a by-product, deterministic intensity predictions were carried out for all global tropical-cyclone events during subsequent forecast periods defined by the years 2001 and 2002, respectively. The annual mean deterministic intensity errors of the years 2001 (2002) at 24-, 48-, 72-, 96-, and 120-h prediction time were found to be 6.2 (6.5), 9.6 (10.6), 11.7 (12.4), 15.4 (15.3), and 17.2 (17.1) m?s?1, respectively. On average, the deterministic forecasts were of approximately the same quality as those of all current consensus approaches and of superior quality than those of the majority of all operational dynamical models. The quality of the probabilistic forecasts, provided in the form of intensity probability intervals at given prediction times, was assessed by the annual mean sizes of given probability intervals. For example, in the years 2001 (2002) the annual mean sizes of the 66% confidence intervals at 24-, 48-, 72-, 96-, and 120-h prediction times were found to be 12.6 (13.3), 19.7 (21.5), 24.4 (24.8), 39.6 (27.8), and 40.4 (29.0) m?s?1, respectively. Postanalyses showed that the sizes of all intensity probability intervals represented conservative and reliable estimates of future storm intensities in that the observed percentages of storm intensities inside given intervals were larger than the corresponding expected percentages.