Tests of stochastic perturbation schemes in the NCEP Global Ensemble Forecast System

The operational NCEP Global Ensemble Forecast System (GEFS) contains an additive stochastic perturbation scheme that uses random linear combinations of ensemble tendency perturbations added to each ensemble member. This stochastic method is included as a method for simulating the uncertainty in the forecast due to model imperfections. The NCEP scheme can only operate within an ensemble forecast, and requires the entire ensemble be run concurrently in one executable (so that tendency perturbations can be communicated from one ensemble member to another). This makes the method potentially vulnerable to the failure of any of multiple processors used in the integration of the ensemble. We have implemented a set of alternative stochastic perturbation schemes which do not have this limitation. These include the ECMWF stochastically perturbed physics tendency scheme, an additive noise perturbation scheme, and a scheme similar to the ECMWF scheme that operates on the dynamics tendencies as well as the physics tendencies. Ensemble forecasts with these schemes are compared to ensemble forecasts with no stochastic perturbations and the operational scheme. In addition, the impact of the stochastic schemes on the analysis through the use of a Ensemble Kalman Filter data assimilation system is investigated.