Hurricane track forecasting remains a significant and costly problem. The potential damage incurred by hurricane landfall is well known. However, an additional cost is associated with hurricane forecast track errors as residents and tourists evacuate and businesses suffer losses, preparing for a landfall that does not occur. Improvements in weather prediction models, coupled with rapid strides in our observing capability and advances in assimilation techniques, have led to significantly improved tropical cyclone (TC) track guidance in recent years. Despite these improvements, however, there still exist many uncertainties in the dynamical prediction of TC tracks. The uncertainties stem from not only poor initialization of tropical cyclone structure, intensity and location in dynamic models, but also inaccurate specification of the large-scale environment in the vicinity of tropical cyclones. To account for these uncertainties and exploit them to improve hurricane track, intensity and landfall predictions, we have developed a comprehensive mesoscale ensemble predictions system using the Penn State / NCAR MM5V3 non-hydrostatic modeling system. Simulations incorporating variations in model input data, physical processes and parameters make up the ensemble set. The mesoscale ensemble prediction system spans the uncertainties that are particularly important to TC prediction.

Models such as MM5 offer several choices of physics (e.g. microphysical scheme and planetary boundary layer and cumulus parameterizations) and their use may significantly alter the solution. In addition, numerous empirical parameters are employed whose values are more properly considered to be a range than a single number, such as drag coefficients and those controlling sub-grid scale processes (e.g. parameterization of cumulus convection, sensible, and latent heat fluxes and microphysical processes). A range of model physics choices will make up the ensemble forecast set. Three boundary layer methods, four cumulus parameterization schemes and five microphysical approximations are being used.

Model initialization includes use of first-guess fields (e.g. NCEP or ECMWF) as a basis for an analysis incorporating rawinsonde and specialized observations where available. Ensembles will be constructed by varying these first-guess grids and via perturbations to the observations or grid values (Monte Carlo, Breeding of Growing Modes - BGM) in a systematic way. The aforementioned initial condition perturbations address the hurricane environment rather than the tropical cyclone structure. Ensembles will also be developed through implementation of a synthetic vortex. Specification of and modifications to the TC vortex will address uncertainties in vortex location, structure and intensity. We will use modifications to the inserted TC location, intensity and structur. Our goal is producing a suite of predictions based on these initial conditions that exhibits adequate spread in storm track and which encompasses the actual storm path, landfall and intensity.

The ensemble approach outlined herein is being applied to East-coast and Gulf cyclones, including but not limited to three recent cyclones: hurricanes Opal (1995), Georges (1998), and Floyd (1999). The combination of perturbations to the initial cyclone structure, its environment and the model physical parameterizations are being used to provide a range of plausible solutions with which to explore and improve hurricane track forecasting.