Problems to be solved
Seasonal prediction of climate has shown promise in recent years, in particular for the tropics (ENSO), but also for the extra tropics and Europe, with potential important socio-economic benefits. There is a need to further develop this capability and to involve user communities to maximize benefits.
Scientific objectives and approach
The overall objective is the development of a European multi-model ensemble system for seasonal to inter-annual climate prediction, to integrate specific user application models and to assess the economic value of the system. Six global coupled ocean-atmosphere models developed at different institutes in Europe will be installed on a common supercomputer. A set of multi-model ensemble hind casts will be produced using reanalysis data for initialisation and validation. By including independent models in the ensemble, the impact of model uncertainty on seasonal predictions can be quantified. The validation will include an assessment of the predictability of El Niño and the North Atlantic Oscillation (NAO), and seasonal weather elements over Europe. The project calls for about 30 years of ensemble integration using ERA-40 data (with existing ERA-15 data as a back up). Each integration will be 6 months long, and each model will be used to provide model-ensembles. Empirical correction techniques will be used to provide model-dependent bias corrected data. Thorough evaluation of the meteorological and oceanographic skill of the hind casts, using probabilistic validation tools, will be made. Evaluation of the predictability of El Niño, the NAO and seasonal weather elements over Europa and tropical Africa will be undertaken. Data from the hind casts will be made available to the research, user, and forecasting community. A number of sensitivity studies will be undertaken and the importance of using coupled models, and of using ocean and satellite altimeter data will be evaluated. Two methods for providing downscaled products will be assessed. Data from the hind casts will be input into quantitative user application models for predicting probability distributions of crop yield over Europe, and incidence of disease in the tropical Africa. This will be used for a quantitative assessment of the value of the forecast system in the two sectors.
The project paves the way for a fully operational seasonal climate prediction system, which would give important benefits for almost every sector of society in Europe and in regions of European interests.
A multi-model ensemble system for seasonal to interannual was developed, comprising 7 quasi-independent global coupled ocean-atmosphere models. This referred to as the DEMETER system. An extensive archive of ensemble hindcast data was created, running the DEMETER system, with 6 month integration lengths over the ERA-40 period. By including quasi-independent models in this way, the DEMETER ensemble incorporates a representation of model uncertainty. In addition, each model was run from a sub-ensemble of perturbed ocean-atmosphere initial states; in this way the DEMETER ensemble also includes a representation of initial uncertainty. A key result from an analysis of these hindcasts is the improved reliability and skill of the multi-model ensemble forecasts over the single-model ensemble forecasts. The notion of reliability relates to the (frequentist) analysis of the ensemble as a probability forecast. For example, if probability forecasts of some climate event E (such as temperature above normal) are reliable, then the frequency of occurrence of E, for a sample of ensemble forecasts, which predict E with probability P, must equal P. It has been shown that the DEMETER multi-model forecasts are substantially more skilful in this sense than forecasts from the component models (even when the ensemble size of a component model is increased to equal the DEMETER ensemble size). Corresponding probabilistic skill scores in which reliability is an important component (such as the Brier Skill Score or Ranked Probability Skill Score) are also more skilful for the multi-model ensemble, especially in the tropics. A system to produce monthly mean anomalies for a large set of variables, i.e. the bias-corrected dataset, has been installed.
This code runs automatically whenever there are new hindcasts available for any model. This procedure effectively removes the mean bias from the model output, but it has no effect on biases of the higher moments. Also, for non-negative variables, such as precipitation, this kind of bias correction can produce occasional negative values, which is not very satisfactory. Therefore, more advanced approaches have been tested on PROVOST and ERA-15 data. However, neither the predictive skill nor the reliability were improved for PROVOST simulated precipitation in the region that was tested (Europe). The Rossby Centre model (RCA) is the limited area model finally selected to run seasonal integrations with boundaries from global models. The feasability of using global seasonal hindcasts to produce dynamically downscaled data has been proven with eighteen 180-day integrations completed. Results are archived and available from MARS. However, a first evaluation of the runs (e.g. for the potential economical value) found only for some precipitation seasonal forecasts produced by the RCA model higher values than the global direct model outputs. Sensitivity studies to assess the relative importance of coupled models compared to uncoupled models have shown that in particular for longer lead times the coupled model seems to be more skilful. The better performance of the coupled model appears partially associated with better corresponding predictions of SST in the Niño3 region. However, improved predictions of Niño3 SST anomaly are not always associated with improved precipitation forecasts. The assessment of the potential economic benefit of seasonal forecasts to users has shown promising results for both the agriculture and health applications. The application of a simple cost/loss model demonstrated an increase in the potential economic value after forcing the end-user models with seasonal forecasts.
Funding SchemeCSC - Cost-sharing contracts
L3 5QA Liverpool
RG12 2SZ Bracknell