EUROSIP operational history

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.

The EUROSIP multi model seasonal forecast system has produced and archived products since 2005. Here we provide information on the contents of the EUROSIP operational archive, as well as a brief summary of the evolution of the multi-model system.

Composition of the multi-model ensemble

The composition of the multi-model ensemble changes each time that one of the component models changes. For the user of either our graphical products or the EUROSIP multi-model data products, the changes are largely transparent. One exception is that where multi-model hindcast products are used, they are only available for the effective hindcast period. For users who have access to the raw data of the individual models, it is important to keep track of all of the changes and make sure that the correct data is accessed. Only one version of each model is operational at a given time, and the tables below shows the evolution of the operational system.

From September 2012, the spatial maps are formed by referencing each model to a common climate reference period, to the extent possible. Although models may have a larger set of re-forecast dates available (see later table), only dates from the common reference period are used to form the spatial maps.

The following table gives information on the ensemble sizes and hindcast dates available for each of the models that has been used in the EUROSIP system. Hindcasts are usually available for dates from the end of the hindcast period to the start of real-time forecasts for a given system, but not always.

Brief text summaries of the component models are as follows.

ECMWF

System 2 consists of CY23r4 of the IFS at TL95 resolution coupled with a 1 deg version of the HOPE ocean model. The IFS has 40 levels, extending up to 10 hPa. Ocean initial conditions come from an assimilation system based on an OI analysis. Atmosphere and land surface initial conditions come from a mixture of ERA15 and ECMWF operations. See also the ECMWF Seasonal Forecast User Guide (System 2)*

System 3 consists of Cy31r1 of the IFS at TL159 resolution coupled with a 1 deg version of the HOPE ocean model. The IFS has 62 levels, extending up to approximately 5 hPa. Ocean initial conditions come from an assimilation system based on an advanced multivariate OI analysis with bias adjustments. Atmosphere and land surface initial conditions come from a mixture of ERA40 and ECMWF operations. A more complete description is available in the ECMWF Seasonal Forecast User Guide (System 3)*

System 4 consists of Cy36r4 of the IFS at TL255 resolution (80 km grid point resolution) coupled with the ORCA1 configuration of the NEMO ocean model. The IFS has 91 levels and includes the whole stratosphere. Ocean initial conditions come from an assimilation system based on an advanced multivariate variational analysis with bias adjustments. Atmosphere and land surface initial conditions come from a mixture of ERA Interim and ECMWF operations, and an offline run of the HTESSEL surface model.

SEAS5 uses Cy43r1 of the IFS at a TCo319 resolution (36 km grid point resolution) coupled with the ORCA025 configuration of NEMO. Re-forecast period was increased to 1981-2016, and re-forecast ensemble size was increased to 25. A more complete description is available in the latest version of the ECMWF Seasonal Forecast User Guide.

Met Office

GLOSEA2 has an atmospheric component with a spatial resolution of 2.5x3.75 deg grid. The ocean model has a basic resolution of 1.25 deg, but meridian refinement to 0.3 deg at the equator. Ocean initial conditions are taken from the Met Office ocean analysis system.

GLOSEA3 uses the same forecast model to GLOSEA2, but includes a fix to the ocean currents in the ocean initial conditions for start dates between May and October. Other start dates are identical to GLOSEA2.

GLOSEA4 uses HadGEM3, the new Met Office Hadley Centre ( MOHC) coupled ocean-atmosphere climate model. This has both enhanced atmospheric horizontal resolution relative to GloSea3 (~120 km vs ~240 km) and enhanced atmospheric vertical resolution (38 levels vs 19 levels). It also uses a
different ocean model, NEMO (Nucleus for European Modelling of the Ocean), with a basic resolution of 1.0 deg and meridional refinement to 0.3 deg at the equator. As well as a new model, there are also important changes to the way in which forecasts and hindcasts are produced. A 'lagged start' approach is combined with stochastic perturbations generated during the model integration (both stochastic backscatter and perturbed physics) to generate ensemble members. Note that the data archived at ECMWF are processed to look as if they come from a single ensemble initialized at the start of each month - there is no change in the structure of the archive. A regular programme of system updates is envisaged, and for this reason the hindcasts will be freshly generated by the Met Office each month. They will be made available as part of the EUROSIP system together with the corresponding real-time forecasts, but will not be available in advance. The GLOSEA4 system runs in Exeter, and only a limited number of fields are transmitted to ECMWF for inclusion in EUROSIP.

System 5 is identical to System 4, and uses the same GLOSEA4 model. It is given a different system number because the hindcasts are repeated and stored separately.

System 6 has increased vertical resolution - 75 levels in the ocean and 85 levels in the atmosphere - giving a well resolved stratosphere. The hindcast dates also changed to a more recent 14 year period.

System 7 is identical to System 6, but given a different system number because the hindcasts are repeated and stored separately.

System 8 is identical to System 7, but given a different system number because the hindcasts are repeated and stored separately.

System 9 is GLOSEA5, a major new model version. NEMO ocean resolution is increased to 1/4 by 1/4 degree with 75 levels, and the atmosphere model is GEM3 GA3.0. Resolution is N216L85, which corresponds to approximately 50 km resolution.

Starting from September 2013, Met Office data is supplied to EUROSIP with a variable ensemble size, due to the fact that a significant number of model integrations fail to complete. Processing at ECMWF ensures that the ensemble size of the hindcasts remains the nominal value (12 at present) by duplicating available members to fill gaps. This is done in a way that does not bias the model climate.However, the real-time forecasts are not duplicated, and so the ensemble size remains variable. All EUROSIP products account for this correctly, using whatever data is available in an unbiased way.

System 10 is similar to System 9, but given a different system number because the hindcasts are repeated and stored separately.

System 11 is similar to System 10, but given a different system number because the hindcasts are repeated and stored separately.

System 12 is similar to System 11, but given a different system number because the hindcasts are repeated and stored separately. The period of the re-forecasts was increased to 1993-2015.

System 13 is similar to System 12, but given a different system number because the hindcasts are repeated and stored separately. The size of the re-forecast ensemble was increased to 28.

System 14 is similar to System 13, but given a different system number because the hindcasts are repeated and stored separately.

Météo-France

System 2. The Météo-France coupled model is Arpege/ORCA. Arpege (the atmospheric component) has 91 vertical levels and a spatial resolution of about 300Km. ORCA (the oceanic component) is an ocean model developed at LOCEAN in Paris. The ocean initial conditions are prepared by MERCATOR in Toulouse.

System 3. The Météo-France coupled model is Arpege/ORCA. Arpege is now run with 91 vertical levels, and reaches high into the stratosphere. Horizontal resolution is unchanged.

System 4. A new version of Arpege/ORCA. Horizontal resolution is increased to TL127 with a gaussian grid spacing of about 160km. Vertical resolution is reduced and the stratosphere is no longer fully resolved.

System 5. ARPEGE-Climat version moves to Arpege-IFS cycle 37 with a finer horizontal resolution of 0.75° (T_L255 truncation). The vertical resolution is increased to 91 levels allowing an explicit representation of the stratosphere (ozone, non-orographic gravity wave drag). Surface processes are managed by the dedicated SURFEX 7.3 modelling platform. Sea-ice is computed by a dedicated model: GELATO v5. Ensemble spread is generated by stochastic dynamics technique in addition to using a lagged initialization.

NCEP

System 2. CFSv2 became operational in 2011. It uses a custom reanalyis (CFSR) to produce the initial conditions for both the re-forecasts and in real-time. The resolution of the seasonal forecast model is TL126. Documentation is available from NCEP at http://cfs.ncep.noaa.gov/

JMA

System 2. Data from the Japanese Meteorological Agency seasonal forecast system was incorporated operationally into EUROSIP in March 2017. The data comes from JMA/MRI-CPS2. Atmospheric resolution is TL159  (approx 110km grid spacing), with 60 levels and a model top of 0.1 hPa. Atmosphere and land initial conditions come from JRA-55, ocean initial conditions come from MOVE/MRI.COM-G2. Documentation is available from http://ds.data.jma.go.jp/tcc/tcc/products/model/outline/cps2_description.html

Development of multi-model products

The multi-model graphical and data products from EUROSIP have been developed over time. There are many ways in which data from different models can be combined, and it is not possible to give a full definition of the methods used in the GRIB headers or the MARS definition of the data. Chapters 2 and 3 describe the data processing which is used to produce the current operational multi-model products, valid from September 2012. Here we summarize how these methods have changed over time. This information is given to ensure that the contents of our operational archives are adequately documented.

July 2013 -present

With the switch to GLOSEA5 in July 2013, there is a temporary limit of 5 months to the Met Office data. To cope with this, the spatial maps and multi-model numerical products are produced using data from all four models up to 5 months, but for products including month 6, only the three available models are used. The plots are labelled accordingly. The multi-model nino plumes are simply restricted to 5 months lead time. This is a temporary solution, until the Met Office are able to send data covering the full 6 months.

April 2005 - August 2012

From April 2005 to August 2012, three models were used to form multi-model products (ECMWF, Met Office, Météo-France).

During these years, the EUROSIP nino plume plots were calculated using bias correction for each model, but without any scaling of variance. Since some models systematically over/under estimated the variability of the SST indices, this meant that the range of values in the nino plumes was often unrealistically large. Calibrated pdf plots were not produced.

Also during this period, the multi-model spatial maps were formed by averaging the anomalies for the individual models without regard to the reference period. That is, each model anomaly was formed with respect to its own calibration period (given as "hindcast period" in Table 2 above), and the three values were then simply averaged.

The tropical storm graphical products used all three models up until August 2009; from September 2009 they use data from only the ECMWF and Météo-France models.