This project has ended |
2000 - 2003

Development of a European Multimodel Ensemble system for seasonal to interannual prediction (DEMETER)

The objective of the project is to develop a well-validated European coupled multi-model ensemble forecast system for reliable seasonal to interannual prediction. A fundamental aspect is to establish the practical utility of such a system, particularly to the agriculture and health sectors.

The basic reference of the project is:

Palmer, T.N., A. Alessandri, U. Andersen, P. Cantelaube, M. Davey, P. Délécluse, M. Déqué, E. Díez, F.J. Doblas-Reyes, H. Feddersen, R. Graham, S. Gualdi, J.-F. Guérémy, R. Hagedorn, M. Hoshen, N. Keenlyside, M. Latif, A. Lazar, E. Maisonnave, V. Marletto, A. P. Morse, B. Orfila, P. Rogel, J.-M. Terres, M. C. Thomson, 2004.Development of a European multi-model ensemble system for seasonal to inter-annual prediction (DEMETER). Bulletin of the American Meteorological Society, 85, 853-872

 

Due to copyright issues, this paper can not be distributed. Get a copy of an ECMWF technical memorandum with a similar content

 

 

DEMETER is the acronym of the EU-funded project entitled "Development of a European Multimodel Ensemble system for seasonal to inTERannual prediction".  The objective of the project was to develop a well-validated European coupled multi-model ensemble forecast system for reliable seasonal to interannual prediction. A fundamental aspect was to establish the practical utility of such a system, particularly to the agriculture and health sectors.

The research of DEMETER was continued in the ENSEMBLES project.

Publications

The main reference for the project is:

Tellus A, Vol 57, issue 3,  published a special issue about DEMETER in May 2005 with sections on multi-model forecasting, sensitivity analyses, downscaling and user applications. View this special issue here.

Project partners

Methodology

Seven comprehensive European global coupled atmosphere-ocean models were installed at ECMWF. The use of a multi-model ensemble system allows for uncertainties in model formulation to be included in the estimation of seasonal forecast probabilities. Based on evidence from the earlier Fourth Framework project PROVOST, it was believed that such a multi-model system will produce the most reliable seasonal climate forecasts possible. The DEMETER system was tested on past cases using the ERA-40 database for initial and validation data.

An extensive set of 6-monthly hindcast ensemble integrations were generated using the DEMETER system. These were run four times a year over a period of about 40 years. Each model  produced nine members of the multi-model ensemble. Output from the individual members of the multi-model ensemble were linked with tropical disease prediction models and European crop-yield models. Sensitivity and downscaling studies were also undertaken.

In order to combine ensemble members from models with different systematic-error characteristics, it is desirable to apply a model correction technique to each individual-model ensemble. Datasets containing corrected multi-model ensembles were produced. The sensitivity of the coupled-model simulations to the inclusion of satellite altimeter data, to changes in the model parametrizations, and to the use of persisted sea surface temperature anomalies were investigated. The feasibility of applying the DEMETER output to scales smaller than can be resolved by the global models will also be assessed.

Data dissemination

The hindcasts produced in DEMETER can be retrieved from ECMWF's archive either via the MARS catalogue (restricted access for registered users) or through the Public Datasets interface.

The variable "expver" has been used to distinguish the different partners. The possible values of this variable are:

  • 'cnrm' for Météo-France
  • 'smpi' for Max-Planck Institute
  • 'ukmo' for the Met Office
  • 'scwf' for ECMWF
  • 'scnr' for INGV
  • 'lody' for LODyC
  • 'crfc' for CERFACS

Some partners have performed several experiments that are distinguished with the MARS variable "method". Users are strongly advised to retrieved data using the method=1 in their MARS requests. Both the time required for the requests and the load of the system are sensibly reduced. However, if you are interested in accessing data for particular experiments, the value taken by the variable method is described below.

ECMWF

The last two cyphers of the MARS command method indicate the ERA40 experiment used to initialise the model:

  • 20: from February 1958 to February 1972
  • 334: from May 1972 to August 1972
  • 30: from November 1972 until November 1984
  • 50: from February 1985 to November 1985
  • 40: from February 1986 to November 1988
  • 18: from February 1989 onwards.

Then, different numbers are added up to construct the final method:

  • 0: for the control run (1958-2001)
  • 600: for the control run with 60 levels in the vertical of the atmospheric model, while the control run has 40 levels (1989-2001)
  • 2000: for an experiment similar to the control run but using Reynolds OI SST instead of Reynolds 2DVar SST to relax the ocean model during the generation of the initial conditions (1989-1999)
  • 3000: for an ocean assimilation experiment with the same assimilation code as system 2 (1989-2001)
  • 4000: for a control experiment with a 48-member ensemble (1989-1999).

The official ECMWF version for DEMETER is the control run.

LODyC

The following list applies:

  • 30: from February 1974 to May 1985
  • 50: from August 1985 to November 1985
  • 40: from February 1986 to November 1986
  • 12: from February 1987 to August 1988
  • 17: for November 1988
  • 18: from February 1989 onwards

Météo-France

Method number 11.

MPI

Method number 13 from 1969 to 1986 and 12 afterwards.

INGV

Method number 112.

MetOffice

The number for the coupled experiment is 12 for the monthly means and 11 for the daily data. You add up 200 to obtain the data of the atmospheric model forced with persisted SSTs.

CERFACS

Method number 11.