EUROSIP Data products

Please note that Products from the EUROSIP Multi-model Seasonal Forecasting System has been discontinued on October 2019. 

The Copernicus Climate Change Service (C3S) has  similar multi-model seasonal data available in the Climate Data Store.

ECMWF produces a number of multi-model products which are created from the integrated output of the component models. They are officially referred to as "EUROSIP products". Most EUROSIP products are provided in graphical form only, although there are some numerical data products which we describe here. Further EUROSIP data products may be made available in future, according to demand and subject to policy considerations. These multi-model products can be accessed just like any other ECMWF products.

Data from the individual models is also archived at ECMWF, but these data can only be accessed subject to specific terms in the EUROSIP data policy. In particular, commercial use of the data from an individual model is only allowed once an agreement has been made with the producer of the data. Official duty use is, however, allowed for all Member States. MARS access to these data is restricted.

Real-time EUROSIP forecast product data becomes available at 12Z on the 15th of each month.

EUROSIP products

From September 2008 onwards, the MARS archive contains EUROSIP ensemble mean anomalies of a limited number of monthly mean fields such as 2 metre temperature and precipitation. These are in the EUROSIP monthly mean stream ESMM, and an example retrieval for August 2012 2 metre temperature anomaly predicted from 1st July 2012 would be:


The MARS archive also contains a corresponding set of EUROSIP hindcast monthly means, in stream EHMM. These contain the multi-model ensemble mean anomalies for the set of available common hindcast dates, and allow an assessment of the past performance of the corresponding real-time product. An example retrieval would be:


This retrieves the ensemble mean anomaly for each of the 5 specified hindcast years (hdate), where the multi-model combination for those years is made in the same way as it is for the real-time forecast. The range of available "hdates" is limited by the years for which all of the components of the real-time forecast are available. This is given as the "effective hindcast period" in the first table of section 4.1.

At the present time the ESMM data are included in the Catalogue of ECMWF products, but the EHMM data are not.

Individual model data

Data from individual models is archived in MARS using the same multi-model streams as are used for ECMWF seasonal forecast data.

stream = mmsf -> multi-model seasonal forecast
stream = msmm -> multi-model seasonal forecast monthly means
stream = mmsa -> multi-model seasonal forecast monthly mean anomalies

To define which model the data comes from, use the "origin" descriptor. ECMWF data has origin =ecmf, Met Office data has origin=egrr, Meteo-France data origin=lfpw and NCEP data origin=kwbc. Make sure that you access the correct version of each model by specifying the correct "system" number. Operational history contains the information on which systems were operational on which dates. Re-forecast data must always be retrieved with the same system number as the corresponding forecast data.

From September 2013, the EUROSIP archive allows real-time forecast ensemble sizes to be variable. That is, the actual number of ensemble forecasts processed and archived for a given model might vary from month to month. This only occurs for models listed with a "variable" ensemble size in the tables in Chapter 4, where the maximum possible size is also listed. MARS retrievals can be made for the maximum number, with "expect=any" added to the request, and all available members will be retrieved. The available ensemble members are always numbered from 0 to n-1.

Initial dates for lagged-average systems

Most inidividual model data is archived with an initial date corresponding to the nominal initial date of the forecast, that is 0Z on the first of the month. Some seasonal forecast systems, such as that at ECMWF, run in "burst mode", so this is the actual start date of all ensemble members. However, a number of systems run in "lagged average" mode, where the ensemble is made up of models with a range of start dates. For the purposes of EUROSIP, start dates up to and including the first of the month are allowed (even if the start time is eg 18Z on the first for some ensemble members). Some lagged average systems, for example Météo-France, are archived as if all ensemble members started at 0Z on the first of the month. However, since 2012 the EUROSIP processing system also supports models whose data are archived with the true initial date, for both forecasts and re-forecasts. At present, only NCEP is archived like this. The NCEP real-time forecasts consist of 4 members each day. In the ECMWF archive, these are treated as a 4 member ensemble valid at 0Z. The multi-model system uses a total of 52 members each month, 4 from the 1st of the month, and 48 from the 12 last days of the preceding month. The rest of the data are archived and are available for use, but are not used to derive EUROSIP multi-model products. The NCEP re-forecast data are produced at 5 day intervals starting on the 1st January for each of the re-forecast years, and are archived at ECMWF as a 4 member ensemble for each of the start dates. When calculating EUROSIP products, only 12 re-forecast members are taken from each re-forecast year, corresponding to the 3 closest start dates up to and including the 1st of the month. As with the real-time forecasts, the full re-forecast dataset is available in the archive.

Archiving data from lagged-average systems in this way gives a more complete and flexible data archive. However, it does require more complex retrieving and processing of the data to produce seasonal forecast products. Further, the concept of an "anomaly" relative to the re-forecast set is not so well-defined for the real-time forecast ensemble sets, since there are a range of strategies that could be used to define it, and valid times for which it could be defined. Thus we do not create or archive monthly mean anomalies for models whose data is archived in lagged-average mode.


Some examples of data retrieval from the multi-model system are given below. Note that although most models have a similar set of fields archived to those of the ECMWF model, there can be small differences. If you are trying to access less commonly used fields, use the MARS browser to confirm exactly which fields are archived for each model.

The following retrieval accesses the 2m temperature predicted at day 20 from the first three members of the Météo-France real-time ensemble forecast initiated on the 1st of May 2012:


The retrieval below accesses the monthly mean precipitation rate from the first month of the corresponding Met Office real-time forecasts from May 2012:


This example accesses monthly mean SST from the 12 ensemble members of the NCEP re-forecasts for the year 2001 that are used to support the real-time forecast from NCEP used in the EUROSIP forecast from 1st September 2012:


Finally, this example accesses the monthly mean ensemble mean mslp anomaly from the Met Office forecast from May 2012: