6C.6
Probabilistic prediction of tropical-cyclone position and intensity
Harry C. Weber, University of Munich, Munich, Germany
A new method for the probabilistic prediction of track and intensity (maximum wind speed) of tropical cyclones is presented. During a training period (defined as the tropical-cyclone season preceding the season where forecasts are produced), the forecasts of all operationally-available statistical and numerical models are subjected to a simple statistical analysis. At any given base date and time during the forecast period, the results of this analysis can be used as a predictor for the probability of future storm positions and intensities.
The new method was developed using operationally-available tropical- cyclone advisory and best-track information and model predictions of all global tropical-cyclone events of the years 2000, 2001 and 2002 as provided by the Automated Tropical Cyclone Forecast System (ATCF). Forecasts have been produced for the years 2001 and 2002. Beside the prediction of probability distributions of future storm positions and intensities, the new method provides also deterministic forecasts: for example, mean position errors for all global storm events of the year 2002 at 24, 48, 72, 96 and 120 h prediction time are 136 km (1714 cases), 235 km (1413 cases), 331 km (1142 cases), 436 km (877 cases) and 556 km (652 cases). With that, the new method produces better track guidance than all operational statistical and numerical models and its quality of performance is equal to that of the dynamical-model consensus. The corresponding mean errors in maximum wind speed of 6.5 m/s (1602 cases), 10.6 m/s (1307 cases), 12.4 m/s (1053 cases), 15.3 m/s (783 cases) and 17.1 m/s (586 cases) are found to be of equal or slightly inferior quality relative to those of the intensity consensus, STIPS and STIFOR, but superior to those of most numerical prediction models.
The major advantage of the new method lies in the automatic production of probability distributions of future storm positions and intensities. The high quality of these predictions can be assessed in terms of the mean diameters of the 66% probability intervals, i.e. the mean size of geographical regions and intervals of maximum wind speed in which future storm positions and intensities are expected to be found with a probability of 66%. For example, statistical evaluation of the predictions of all global tropical cyclone events of the year 2002 at 24, 48, 72, 96 and 120 h prediction time resulted in mean diameters of 283, 479, 662, 782 and 1025 km (positions) and 13.3, 21.5, 24.8, 27.7 and 28.9 m/s (maximum wind speeds), respectively. The high quality of these results recommends operational application of the new method in the framework of the ATCF System.
Session 6C, Tropical Cyclone Prediction and Predictability II: Track and Intensity
Tuesday, 4 May 2004, 10:15 AM-11:30 AM, Napoleon II Room
Previous paper