Representing model uncertainty in the Global Ensemble Forecast System (GEFS)

Recently, a suite of stochastic physics parameterizations developed at NOAA's Earth System Research Laboratory (ESRL) has been implemented in the Global Forecast System (GFS). The representation of model error in the GFS has improved the global ensemble used in the data assimilation cycling, and has been included in the operational GFS upgrade this past January. Currently, NCEP is testing the stochastic physics in medium range ensemble. Initial tests at ESRL showed a substantial improvement in calibration of the ensemble forecasts in week-1. Subsequent testing at NCEP found that week-2 forecasts are not as well calibrated.

Investigation into the forecasts shows that the drop in the forecast model resolution at day-8 is altering the behavior of the ensemble. In addition, the stochastic physics parameterizations, which inject perturbations at the small scales, is not sufficiently projecting onto the larger scale perturbations that are prevalent in week-2. The results of these experiments, along with a proposed mechanism to address the model resolution change will be presented.