AStochasticParameterizationSchemewithin.ppt

A Stochastic Parameterization Scheme withinNCEP Global Ensemble Forecast SystemDingchen Hou, Zoltan Toth and Yuejian Zhu  Acknowledgements: Mark Iredell, Henry Juang, Richard Wobus, Bo Cui, Cecile Penland and Prashant SardeshmukhWeiyu Yang, James Purser and Mozheng Wei  OUTLINE INTRODUCTION A STOCHASTIC PHYSICS SCHEME Formulation Characteristics of the Perturbations RESULTS OF EXPERIMENTS Spaghetti diagrams Outliers Ensemble mean and spread Probability forecasts Comparison with a bias-correction procedure SUMMARY AND ONGOING RESEARCH Introduction Some Shortcomings of the Current GEFS forecasts Cold bias and insufficient spread Two Major Sources of Forecast Uncertainty Uncertainty due to INITIAL CONDITIONS Uncertainty related to the MODEL, not considered in the system Representation of Model Uncertainties Perturbation rescaling (Toth and Kalnay, 1995) Multi-model ensemble (e. g. Houtekamer et al., 1996; Stensrud et al.,2000) Multi-model + multi-center ensemble (THORPEX, NAEFS) Multi-model version ensembles (e. g. Hou et al., 2004) Stochastic parameterization (e.g., Buizza et al., 1999) Nature of Stochastic Parameterization (Palmer, 2001, 2003) Using stochastic noise to represent unpredictable small-scale variability Stochastically sampling the probability distribution of the sub-grid tendency, rather than always sampling the mode of the distribution. Types of Stochastic Schemes Spatio-temporal stochastic forcing (e.g. Perez-Munuzuri et al, 2003) Stochastic backscatter (e.g. Frederisksen and Davies, 1997; Shutts 2004) Stochastic physics, ECMWF EPS (Buizza et al., 1999) A Stochastic Parameterization Scheme: Formulation Use combinations of the perturbation tendenciesStochastically determine the combination coefficients The combination coefficients (i=1,N; j=1,N) forms a square matrix C