18th Conference on Weather and Forecasting, 14th Conference on Numerical Weather Prediction, and Ninth Conference on Mesoscale Processes

Thursday, 2 August 2001
Ensemble methods applied to hurricane track forecasting
Brian F. Jewett, Univ. of Illinois, Urbana, IL; and M. K. Ramamurthy and H. Liu
Hurricane track forecasting remains a significant and costly problem. The potential damage incurred by hurricane landfall is well known. Damage costs from hurricane Andrew (1992) reached $27 billion, and Floyd (1999) totaled $6 billion. However, an additional cost is associated with hurricane forecast track errors as residents and tourists flee coastal cities, businesses suffer losses and communities prepare for a landfall that does not occur. This problem is particularly acute for tropical cyclones paralleling the U.S. east or Gulf coasts before landfall. The USWRP (2000) estimates that a 20% reduction in coastline warning area would save $80 million in preparation costs. Clearly, improved track forecasting has many potential benefits to coastal communities.

The coastal hurricane problem includes location and time of landfall and storm intensity when it moves onshore. The propagation speed is also of interest since severe inland flooding is now understood to produce damage comparable to that inflicted by storm surge and high winds along the immediate coastline. Many of the uncertainties in predicting these factors stem from errors in initial hurricane location, intensity and structure while offshore as well as uncertainties in the synoptic environment affecting hurricane intensification and movement. Ensemble modeling methods may be used to address these uncertainties in a systematic way.

We are applying ensemble modeling methods to tropical cyclone (TC) prediction. Most operational ensemble forecasting systems use initial condition perturbations and a "perfect-model" approach to generate ensembles. In addition, a broad spectrum of plausible solutions may be generated by varying parameters controlling different physical processes in a model, including empirical parameters (e.g., drag coefficients) and those controlling sub-grid scale processes (e.g., parameterization of cumulus convection and sensible and latent heat fluxes). Finally, the specification of the initial TC vortex used to augment large-scale analyses allows specification of the vortex size, tangential and radial velocity profiles, and the moisture distribution, both symmetrically about the TC center and asymmetrically (e.g. wave-one).

We investigating the advantages and limitations of ensemble versus deterministic forecasting of TC track and intensity predictions using the Penn State/NCAR MM5 non-hydrostatic model at high (under 20 km) resolution. We are exploring the impact of three boundary layer methods (Blackadar, Mellor-Yamada, and Hong-Pan) and Cumulus parameterization schemes (Anthes-Kuo, Betts-Miller, Kain-Fritsch, and Arakawa-Schubert). East-coast and Gulf cyclones will be considered, including but not limited to three recent hurricanes: hurricane Opal (1995), Georges (1998), and Floyd (1999). In addition, we are comparing three different approaches for constructing ensembles from initial perturbations: a) simple Monte-Carlo, b) the BGM method, and c) Perturbed Observations Method. Finally, the impact of the initial TC vortex specification on cyclone track is being explored. 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.

Supplementary URL: