A Stochastic Parameterization Scheme within NCEP Global Ensemble Forecast System

A scheme of stochastic physics, representing model related errors, is designed and tested with the NCEP Global Ensemble Forecast System (GEFS). The strategy used is to add an additional stochastic term with structured noise in the tendencies of each model variable. For each member of the ensemble, the stochastic perturbation is constructed as a random linear combination of tendencies of other ensemble members and the control. Perturbations constructed this way have two desirable features: they are both balanced and flow dependent.

Given these features, the perturbation size can be rescaled to a size that ensures that perturbation growth in the ensemble matches error growth in the ensemble mean forecast, a necessary condition for statistically reliable forecast performance, without degrading forecast skill. Preliminary experiments show that the scheme can increase the ensemble spread, while reducing the number of outliers and also reducing some model related biases. The impact of the stochastic physics scheme on the performance of probabilistic forecasts based on the NCEP ensemble system is presented in this paper. Various performance scores, including Relative Operation Characteristics, Economic Value, Brier Skill Score and Ranked Probability Score, are evaluated for atmospheric variables. The effect of the scheme on the reliability and resolution of the probabilistic forecasts will also be addressed.