Torrential rainfall associated with landfalling tropical cyclones can be a significant hazard to life and property. Within the core and spiral rainbands of a tropical cyclone, as well as near coastal boundaries and high topography, intense precipitation often occurs. During August, 1997, Supertyphoon Winnie seriously affected portions of Taiwan and China with winds of 75 knots and widespread rainfall amounts of over 200 mm. Steep topographic regions received over 400 mm of rain. To investigate this event, a high resolution nested regional spectral model is employed, along with an EOF-based perturbation method, to generate ensemble forecasts of the track, intensity, and precipitation.
A five day forecast of Supertyphoon Winnie is carried out using the Florida State University Global Spectral Model (FSUGSM) with triangular truncation at wavenumber 126, which corresponds to a transform grid resolution of 0.9375 degrees. Then, a seven member ensemble is generated in which initial temperature and wind fields throughout the tropics are randomly perturbed. The fastest growing modes are determined by performing EOF analysis on the difference between the control and perturbed forecasts through 36 hours. The regional model, containing the same dynamics and physics as the FSUGSM, utilizes a grid spacing of 0.5 degrees. Accordingly, the resolution of the input data for this model integration is increased from T126 to T170. Also, this initial data is physically initialized in which observed rainfall distributions from both rain gauge and satellite along with calculated surface fluxes of moisture are assimilated to produce a physically consistent and more realistic spin-up of the initial state.
The results show that the ensemble mean track and precipitation forecast errors are clearly reduced, especially in the high-resolution, physically initialized case, where orographic effects and air-sea interactions are much improved. Furthermore, useful products generated by the use of ensemble forecasting include the variance of the track and spatial distributions of maximum wind speed and precipitation probability. These statistics can then be used to validate the forecast.