Improvements to the GFDL hurricane forecast system

Beginning in 1995 the GFDL Hurricane Prediction System has provided operational guidance for more than 400 storm cases in the Atlantic. Relative to the other guidance of the National Weather Service, the mean GFDL track forecasts appear to be significantly superior after the first day. Nevertheless, the GFDL Hurricane Prediction system exhibits small track biases and rather large intensity biases. For the most part, the nested GFDL Hurricane Prediction System has been utilized with its standard three mesh model resolutions of 1, 1/3, and 1/6 deg. for both operational and research use. In the past several years, the resolution of global models have increased such that they are comparable if not finer than the outermost mesh of the GFDL model. It may be that finer resolution of the GFDL model in the outermost mesh will further improve track prediction. In addition, a lack of finer resolution in the storm region has caused maximum winds to be under forecast for hurricanes stronger than 90 knots even when model surface pressures fall below 930hPa. This can be seen by analyzing the GFDL model pressure-wind relationship relative to that observed. Therefore a set of sensitivity experiments have been initiated to study the impact of resolution on track and intensity forecasts of the GFDL model. An upgrade of the 3-nest system is underway in which the model resolution is increased to 1/2, 1/6, and 1/12 deg. Improvements in intensity forecasts due to the increased resolution near the storm will be shown. The computational overhead for this grid configuration should be ~ 8 times than that of the standard grid configuration. In addition experiments increasing the vertical resolution of the GFDL model from 18 levels to 42 levels have also been carried out. These grid configurations are now computationally feasible in research mode and should be feasible operationally in the near future. A suite of sensitivity experiments will be run with this new resolution configurations and the results will be presented.

In addition some defects in GFDL model simulations may be related to deficiencies in physical parameterizations such as boundary layer processes, convective processes and diffusion formulations. A second series of experiments have been run for various cases study experiments to test the sensitivity of the GFDL model in track and intensity to different physical parameterization packages. Especially promising has been the improved vertical profile of boundary layer winds when turbulent KE is predicted. This eliminates to some extent the low bias of surface winds in the GFDL hurricane model. Results from these experiments will be shown.