Representing model uncertainty : stochastic parametrizations at ECMWF
Shutts, Glenn ; Leutbecher, Martin ; Weisheimer, Antje ; Stockdale, Tim ; Isaksen, Lars ; Bonavita, Massimo
When it was introduced, the ECMWF Ensemble Prediction System (EPS) was based on the assumption that errors in medium-range forecasts are mainly associated with errors in initial conditions. Later it was recognised that uncertainties in the model formulation may also be a significant factor. In particular, the physical parametrizations can be a significant source of random error. This led to the development of a stochastic representation of physical parametrization uncertainty (now known as the Stochastic Perturbed Parametrization Tendency (SPPT) scheme). The SPPT scheme has been used in the operational EPS since October 1998 this version will be referred to as SPPT-98. Through the judicious application of random number multipliers to forecast tendencies, there has been an increase in ensemble spread in the EPS and improved probability skill scores (Buizza et al., 1999). Investigation of the performance of the EPS has shown that there would be benefits in enhancing the representation of model errors. This resulted in recent operational changes. The impacts of those changes are outlined in this article. The Ensemble of Data Assimilations (EDA) was introduced in June 2010 to generate initial perturbations for the ensemble from ten independent 4D-Var assimilations with representations of observation and model errors (see, for example, Bonavita, 2011 and Isaksen et al., 2010). EDA-based perturbations replaced the evolved singular vectors and the initial singular vectors were retained to ensure sufficient spread in the medium-range. In the EDA, SPPT plays an essential role in providing different realizations of the physics tendencies within each ensemble member. Importantly, EDA provides a means by which stochastic model error representations may be confronted with observational reality.
Accès à la notice sur le site du portail documentaire de Météo-France