|Title||Forecast quality assessment of the ENSEMBLES seasonal-to-decadal Stream 2 hindcasts|
|Publication Type||Technical memorandum|
|Secondary Title||ECMWF Technical Memoranda|
|Authors||Doblas-Reyes, F, Weisheimer, A, Palmer, TN, Murphy, JM, Smith, DM|
The ENSEMBLES system for climate prediction on seasonal, interannual and decadal (s2d) time scales includes three different methods to address the problem of model uncertainty: the multi-model ensemble technique (MME), the perturbation of physical parameters (PPE) and stochastic parameterizations of sub-grid processes (SPE). To compare the methodologically different approaches and to assess their relative merits in an s2d framework, a set of common simulations was defined, performed and analyzed. MME and SPE perform similarly well for the seasonal predictions, while PPE has a similar performance to MME in longer time scales. Benefits of the stochastic physics approach on seasonal time scales are found in reducing the ensemble-mean RMSE and increasing the ensemble spread when compared to the control version of the corresponding model. The main benefit of the MME lies in an increase of the reliability. The MME also shows a larger discriminatory ability between the occurrence of the events and non-events. Most of the MME superiority is linked to its larger ensemble size. The annual integrations have shown a relatively high level of skill in all systems over many areas that is above the level obtained with climatological and persistence forecasts. The preliminary ENSEMBLES decadal hindcasts have shown that, apart from the existence of substantial skill in interannual predictions, the PPE and MME systems offer a similar level and spatial distribution of skill and spread. In all cases, the decadal predictions suggest that the large uncertainty in the ocean initial state, especially for the XXth Century is a major problem for the assessment of decadal skill. The manuscript also discusses the utility of seamless prediction methods across time scales, where shorter time scales are known to be important in their feedback on the longer time scales. As most of the data used to obtain these results are publicly available, more detailed descriptions will follow soon.