Multi-modelEnsembles,PoorMansEnsemblesandSuper-.docVIP

Multi-model Ensembles, Poor Mans Ensembles and Super-Ensembles - a brief review Prepared for the WMO Workshop on Applications of Ensemble Forecasting, 16-20 October 2000, Beijing, China Ken Mylne The Met Office, Bracknell, UK Introduction This is intended as a brief overview of some of the different types of ensemble forecasting systems currently being developed or tested in the World. It is not intended to be a comprehensive literature review, but to give some references for further study if required. Current Operational Ensemble Prediction Systems (EPS) for medium-range forecasting mostly use a single model with a set of perturbed initial conditions to take account of the analysis uncertainty eg ECMWF - Molteni et al (1996), Buizza et al (1998); NCEP - Toth and Kalnay (1993). This approach essentially assumes that the model is perfect, and that forecast uncertainty is due only to initial condition errors. Some attempt to allow for model errors is now made in the ECMWF ensemble, by the use of stochastic perturbations of the tendency due to model physics (stochastic physics), Buizza et al (1999). A more comprehensive approach is followed by Houtekamer et al (1996) who use a number of different versions of their model with different physical parametrization schemes as well as initial condition perturbations. All these systems find that the spread of the ensemble is too small, indicating that the simulation of the error sources is incomplete. Some of the new approaches to ensemble forecasting, such as multi-model ensembles, are attempting to address some of this additional uncertainty. Multi-model and Multi-analysis Ensembles One approach to taking account of model uncertainties is to include other model(s) in the ensemble system. Ideally the models used should be as independent of each other as possible so that their errors and weaknesses are as independent as possible. Evans et al (2000), Mylne et al (1999) and Mylne et al (2000) describe experiments combining the E