The DEMETER project aims to develop a European multimodel ensemble system for seasonal-to-interannual prediction. Leading European global coupled climate models show good reliability for seasonal climate prediction, including useful output for probabilistic prediction of malaria incidence and crop yield. Seasonal forecasts are now routinely made using comprehensive coupled models of the atmosphere, oceans, and land surface. These forecasts are supported by the successful prediction of El Niño by prototype coupled ocean–atmosphere models and the Tropical Ocean Global Atmosphere (TOGA) program. Seasonal forecasts are valuable for society, but the atmosphere is intrinsically chaotic, making day-to-day weather predictions sensitive to initial conditions. Ensemble forecasts of coupled ocean–atmosphere models, with small perturbations to initial conditions, provide a measure of predictability. However, uncertainties in initial conditions alone are not sufficient for reliable predictability, as model equations are also uncertain. A pragmatic approach is needed, and one such approach is the use of multimodel ensembles, which have been shown to produce more reliable probability forecasts than single-model ensembles. The PROVOST project, funded by the European Union, demonstrated that multimodel ensembles can provide more reliable forecasts than single-model ensembles. Based on these results, the DEMETER project was conceived and funded under the European Union Vth Framework Environment Programme. The principal aim of DEMETER was to advance the concept of multimodel ensemble prediction by installing a number of state-of-the-art global coupled ocean–atmosphere models on a single supercomputer, and to produce a series of 6-month multimodel ensemble hindcasts with common archiving and diagnostic software. The DEMETER system comprises seven global coupled ocean–atmosphere models. Each model is run in ensemble mode with nine different initial conditions. The system uses a common archiving strategy and diagnostic software to evaluate the hindcasts. The DEMETER project has applications in agronomy and tropical disease prediction. The system has been used to produce seasonal forecasts of crop yield and malaria incidence. The DEMETER system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The systemThe DEMETER project aims to develop a European multimodel ensemble system for seasonal-to-interannual prediction. Leading European global coupled climate models show good reliability for seasonal climate prediction, including useful output for probabilistic prediction of malaria incidence and crop yield. Seasonal forecasts are now routinely made using comprehensive coupled models of the atmosphere, oceans, and land surface. These forecasts are supported by the successful prediction of El Niño by prototype coupled ocean–atmosphere models and the Tropical Ocean Global Atmosphere (TOGA) program. Seasonal forecasts are valuable for society, but the atmosphere is intrinsically chaotic, making day-to-day weather predictions sensitive to initial conditions. Ensemble forecasts of coupled ocean–atmosphere models, with small perturbations to initial conditions, provide a measure of predictability. However, uncertainties in initial conditions alone are not sufficient for reliable predictability, as model equations are also uncertain. A pragmatic approach is needed, and one such approach is the use of multimodel ensembles, which have been shown to produce more reliable probability forecasts than single-model ensembles. The PROVOST project, funded by the European Union, demonstrated that multimodel ensembles can provide more reliable forecasts than single-model ensembles. Based on these results, the DEMETER project was conceived and funded under the European Union Vth Framework Environment Programme. The principal aim of DEMETER was to advance the concept of multimodel ensemble prediction by installing a number of state-of-the-art global coupled ocean–atmosphere models on a single supercomputer, and to produce a series of 6-month multimodel ensemble hindcasts with common archiving and diagnostic software. The DEMETER system comprises seven global coupled ocean–atmosphere models. Each model is run in ensemble mode with nine different initial conditions. The system uses a common archiving strategy and diagnostic software to evaluate the hindcasts. The DEMETER project has applications in agronomy and tropical disease prediction. The system has been used to produce seasonal forecasts of crop yield and malaria incidence. The DEMETER system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system has been used to produce seasonal forecasts of wheat yield and malaria incidence. The system has been shown to provide more reliable forecasts than single-model ensembles, particularly for large regions and long time series. The system