Improvements in tropical cyclone track and intensity forecasts using the GFDL initialization system

The initialization scheme designed at GFDL to specify a more realistic initial storm structure of tropical cyclones was tested on four real data cases using the GFDL high-resolution multiply nested movable mesh hurricane model. Three of the test cases involved Hurricane Gloria (1985) in the Atlantic basin; the fourth involved Hurricane Gilbert (1988) in the Gulf of Mexico. The initialization scheme produced an initial vortex that was well adapted to the forecast model and was much more realistic in size and intensity than the storm structure obtained from the NMC T80 global analysis. As a result, the erratic storm motion seen in previous integrations of the GFDL model has been nearly eliminated with dramatic improvements in track forecasts during the first 48 h of the prediction. Using the new scheme, the average 24-h and 48-h forecast error for the four test cases was 58 and 94 km, respectively, compared with 143 and 191 km for the noninitialized forecasts starting from the global analysis. The average National Hurricane Center operational forecast error at 24 and 48 h was 118 and 212 km for the same four cases. After 48 h the difference in the average track error became small between the integrations starting from the global analysis and the forecasts starting from the fields obtained by the initialization scheme. With accurate specification of the initial vortex structure, changes in the storm intensity were also well predicted in these cases. The model correctly forecasted the rapid intensification of Hurricane Gloria just after the system was first upgraded to a hurricane. The model storm intensification also ceased at approximately the same time as observed, with gradual weakening as the storm moved north and approached the east coast of the United States. In the forecast of Hurricane Gilbert, the model storm initially weakened as it moved over the Yucatan Peninsula and underwent only moderate reintensification after moving over the Gulf of Mexico, in good agreement with observations. Finally, in the case where the track of Hurricane Gloria was well forecast, the distribution of the maximum low-level winds was accurately predicted as the storm moved up the east coast of the United States. During this period the model successfully reproduced many observed features such as large asymmetries in the wind field, with strongest winds occurring well east of the storm center, and a sharp decrease of the wind speed at the coast. Although the asymmetry in the