A Stochastic Physics Scheme in Representing Model-Related Errors in Global Ensemble Forecast

Ensemble forecast using GFS with breeding technique significantly improves global numerical weather forecast, especially for medium range, compared with single forecast using the same resolution. Further improvement may be achieved by including model-related uncertainty in the ensemble forecast system, and one approach is to add a stochastic forcing term to the tendency of model variable. A scheme in formulating the stochastic physics is proposed and tested, based on the ensemble perturbations. Experiments suggests that the scheme can significantly increase the ensemble spread while achieving noticeable improvement in terms of ensemble mean forecast and ensemble-based probabilistic forecast. In this paper, the scheme is described and results of some preliminary experiments are presented.