Evaluation of GFDL hurricane ensemble wind speed probability guidance

An ensemble forecast system based on the operational Geophysical Fluid Dynamics Laboratory (GFDL) hurricane model has been refined as part of the regional modeling effort for the Hurricane Forecast Improvement Project (HFIP). In this paper, we briefly describe the design of the ensemble forecast system and present results from the 2012 and 2013 Atlantic hurricane seasons. The primary focus of this paper is to introduce and evaluate new near-real-time GFDL ensemble guidance for tropical cyclone wind speed probability forecasts.

The GFDL hurricane ensemble has been running as part of the HFIP project since 2010 and it provides real-time guidance on track and intensity to the National Hurricane Center (NHC) during the Atlantic hurricane season. In 2012, graphical products showing the probabilities of 34-, 50-, and 64-kt 10-m winds occurring in the 5-day forecast were added to the suite of ensemble products delivered in near-real-time to the GFDL ensemble website.

As in the 2011 version of the ensemble, most perturbations are applied by modifying the observed input parameters that define the target wind profile for the spin-up of the bogus vortex. A new ensemble member pair was added in 2012 that perturbed the sea-surface temperatures (SSTs) in the core region. In 2013, an additional member pair was included that modified the model surface physics. Over the past few years, GFDL ensemble membership has been further refined to create maximum spread in the model's 5-day forecasts while also showing realistic possibilities of storm track and intensity evolution.

Verifications indicate that for 2012, the GFDL ensemble mean was very competitive with the top operational guidance for intensity forecasts. Ensemble mean track forecasts were slightly better than the operational GFDL, but were not competitive with the top operational guidance. Ensemble track and intensity verifications will also be shown for the 2013 season. Preliminary wind speed probability verifications show that for Hurricane Sandy (2012), Brier Skill scores for 64-kt wind probabilities were comparable between the GFDL ensemble and the NHC's Monte Carlo model. For all 2012 Atlantic hurricanes, Brier Skill scores for 64-kt wind probabilities were somewhat lower for the GFDL ensemble than the Monte Carlo model, but they do show enough skill to encourage the utility of this type of wind speed probability product that is based on dynamical ensemble model output.