The Performance of NCEP GEFS in Past 20 Years

NCEP’s Global Ensemble Forecast System (GEFS) has been in operation since December 1992, using the NCEP Global Forecast System (GFS) model for integration and the Breeding Vector (BV) technique to generate perturbations as the initial conditions. In additional to upgrades in data assimilation (DA) and the underlying forecast model, there have been several significant milestones in GEFS development over the last 20 years, including 1) addition of perturbed tropical storm (TS) vortex relocation in 2005 to improve TS track forecast; 2) introduction of Ensemble Transform (ET) to traditional BV in 2006 to enhance the quality of initial perturbations; 3) application of a Stochastic Total Tendency Perturbation (STTP) scheme to produce better representation of model uncertainties in 2010; 4) replacement of ET-BV initial perturbations in with Ensemble Kalman Filter (EnKF) 6-hour-forecast of DA in 2015, as part of the unification of DA and ensemble forecast.

The performance of the NCEP GEFS will be presented for past years in terms of common verification metrics, such as anomaly correlation of 500hPa geopotential height for ensemble mean, continuous ranked probability score (CRPS) skill of 500 hPa geopotential height and 850 hPa temperature. All these skill scores indicate that skill levels have been extended by about 4-days skill extended in past 20 years through the combined improvements in DA, models, and ensemble methods. Results from two additional performance metrics will be presented as well; they are the performance of Northern American Ensemble Forecast System (NAEFS) from combined NCEP and CMC’s ensembles; and the performance of tropical storm tracks.