Improvements to the Navy's version of the GFDL Tropical Cyclone Model (GFDN) and evaluation of its Prediction Skill

During the past two years, major upgrades to the National Weather Service's (NWS) version of the GFDL Hurricane Prediction System were made in close collaboration with scientists at NCEP (National Centers for Environmental Prediction) and URI (University of Rhode Island). These upgrades included doubling of the finest resolution from 1/6 to 1/12 degree, upgrade of the moist physics with NCEP's Ferrier microphysics, and improved momentum flux parameterization. Tests results indicated that these upgrades significantly improved the reliability of both the model's track and intensity prediction. Indeed, in 2006 the GFDL model for the first time exhibited superior performance compared to the statistical models in most forecast time periods, beating even the official forecast in the Atlantic..

In 1996 a version of the GFDL model (called GFDN) was transferred to the United States Navy in order to provide forecast guidance in the western Pacific. In the following year, the model began to also provide forecasts for the other ocean basins including the Atlantic, eastern Pacific and the Southern Hemisphere. Prior to 2003, it had been the practice to transfer the upgrades of the GFDL model to the Navy in time for operational implementation the following year. However, the major upgrades to the GFDL model that were made operational in 2005 and 2006 had still not been transferred to the Navy by 2007. In addition, the GFDN model remained uncoupled compared to the NWS version, which has been operationally coupled to a three-dimensional version of the Princeton Ocean Model (POM) in the Atlantic since 2001 and a 1D version of POM in the eastern and central Pacific basin since 2004.

It is believed that these factors have resulted in a steady decline in the reliability of the GFDN model, compared both to the global models and the version of the GFDL model run by the National Weather Service. To rectify these deficiencies, a major effort was undertaken at GFDL and URI in the summer of 2007, to transfer the 2005 and 2006 upgrades of the GFDL model to GFDN in time for the 2008 season. In addition, this new model will be coupled to a three-dimensional version of POM in the Atlantic with the 1D coupling extended beyond the eastern and central Pacific, to include all other basins including the western Pacific and ocean basins in the Southern Hemisphere.

The new upgraded GFDN model has been tested so far on 63 cases in the western Pacific from the previous 3 typhoon seasons, as well as a limited number of cases in the Atlantic, including Hurricanes Dean and Katrina. Preliminary results for both track and intensity prediction are quite encouraging. For example in the 3-5 day time period, the average track errors for the cases tested in the western Pacific were reduced about 13% compared to the current operational GFDN model. In addition, the average intensity error for the same cases was reduced about 12% at all time levels beyond 12h. For the case of Katrina, the west bias in the later forecasts of GFDN was reduced and the north bias noted in GFDN for Hurricane Dean was virtually removed. Indeed, the new model displayed the largest improvement for Dean during the critical period that the hurricane potentially threatened the USA. Examples of the improved track and intensity prediction with the new GFDN model will be shown and the average improvement in skill will be summarized in detail.

Finally, during 2008, an effort will be initiated to extend the one-dimensional ocean coupling in the western Pacific to a full three-dimensional coupling, using similar techniques as has been successfully employed in the GFDL model in the Atlantic. Details of this upgrade will be presented.