Many operational NWP centers such as NCEP and ECMWF are now routinely employing ensemble forecasting methods for medium range forecasting. This represents a shift from a strictly deterministic approach to a more probabilistic approach, taking into account uncertainties in the description of the actual state of the atmosphere at any time and the reality that even small differences in the initial condition will grow and lead to large forecast differences. While the bulk of the work to date has gone into the application of ensemble forecasting in medium range forecasting using initial condition perturbations, the potential usefulness of ensemble forecasting to tropical cyclone (TC) predictions is less clear. TC track prediction is perhaps one of the simplest and yet may become one of most important applications ensemble forecasting, because only one parameter, the location, is being forecast. Similarly, the prediction of TC intensity change (e.g. central pressure and peak tangential winds), one of the most difficult problems confronting the hurricane prediction community, can benefit from ensemble forecasting due to the many uncertainties in the description of processes influencing tropical cyclone intensity change.
In this study, we investigate the usefulness of ensemble versus deterministic forecasting of TC track and intensity predictions in mesoscale models at high resolutions (5-30 km grid spacing). To that end, we use the NCAR/Penn State Nonhydrostatic MM5 Model and conduct experiments on the Hurricane Opal case, which was one of the most intense tropical storms of the 1995 Atlantic/Gulf of Mexico hurricane season. The initial condition for this study is chosen at 0000 UTC on October 2, 1995, and an ensemble of forecasts is made for a 96-hour period. In this particular study, the impacts of uncertainties in initial conditions, cumulus parameterization and boundary layer schemes, and lateral boundary conditions are examined on Opal's track and intensity predictions. Specifically, sensitivity to four different cumulus parameterization schemes (Anthes-Kuo, Grell, Betts-Miller, and Kain-Fritsch), three planetary boundary layer schemes (Blackadar, Burk-Thompson, and Hong-Pan), and two sets of initial conditions (NCEP and ECMWF analyses) are explored. Since the MM5 model is a limited-area model, the lateral boundary conditions become important, especially beyond one day. Therefore, we will demonstrate the importance of lateral boundary condition sensitivities.
Preliminary results indicate that both the track and intensity forecasts of Opal exhibit considerable sensitivity to the choice of the PBL and convective schemes. The track forecast is particularly sensitive to the lateral boundary conditions. Moreover, considerable sensitivity is seen in the predicted rainfall amounts and its distribution within the storm. Local and global error growth, track and intensity forecasts, and other results from the MM5 ensemble will be presented at the conference. The results of the ensemble predictions will be compared with both observations as well as operational forecasts.