Recent changes in the NCEP Global Ensemble Forecast System

Recently, an NCEP Global Ensemble Forecast System (GEFS) has been implemented to improve probabilistic forecast skills. This new system has been upgraded by 1) increasing the horizontal resolution from T62 to T126 for the 180 to 360-hr time range; 2) using 6-hour instead of 24-hour breeding cycles, associated with an adjustment in the geographical mask describing analysis uncertainty; and 3) applying a tropical storm relocation method to better control the initial perturbations associated with tropical storms. Off-line and real-time experiments show improved performance in terms of probabilistic forecasts for week 2 (associated with the 1st implementation item above), for the first few days (2nd item above), and in tropical storm track forecasts (3rd item). The most notable result is a 10% RMS error reduction in the ensemble mean tropical storm track prediction error. With this reduction, the ensemble mean track error for the 2004 season was lower than the higher resolution single control forecast (GFS) error at all lead times. This is due to a more realistic sampling of initial condition-related uncertainty that comes from the coupling of the breeding technique with the tropical relocation relocation algorithm, developed by Q. Liu and his colleagues, already used in the SSI data assimilation procedure.