Implementation date8 Dec 2014
Detailed description of changesSee below
Data set affected


Meteorological changesAddition of new clusters


  • t+120 - t+168


  • t+72 - t+96
  • t+192 - t+240
  • t+264 - t+360
Meteorological impacts
Technical changesAvailability in dissemination and MARS
New disseminated model outputAdditional clusterings
Discontinued disseminated model outputNone

ECMWF cluster products have been updated on 8 December 2014

The ECMWF clustering is one of a range of products that summarise the large amount of information in the Ensemble forecasting system.

The clustering gives an overview of the different synoptic flow patterns in the Ensemble. Based on the similarity between their 500 hPa geopotential fields over the North Atlantic and Europe, the members are grouped together.

Cluster products have been produced operationally since 1992. However in November 2010 new cluster products were introduced in order to provide an effective framework to construct a flow dependent verification, and to extend the products to different forecast ranges (for more details see ECMWF Newsletter N.127 (Spring 2011, pp 6-11)

The clustering algorithm takes the 51 forecasts (50 perturbed plus 1 control forecast) and groups together those that show a similar evolution of the 500 hPa geopotential over the North Atlantic and Europe (75°N–30°N, 20°W–40°E). For two Ensemble members to join the same cluster they must display similar synoptic development at 500 hPa throughout a given time window. Clustering in this way, rather than on individual forecast days, has the advantage that temporal continuity and synoptic consistency are retained. The clustering is made independently for four time windows:

  • t+72 - t+96
  • t+120 - t+168
  • t+192 - t+240
  • t+264 - t+360

Since 2011, time windows 120–168, was archived in Mars and became available for dissemination. The other additional products (72–96, 192–240 and 264–360 hour) were only made available in graphical form on the ECMWF web site. After 4 years of positive feedback from users, and several Member state’s requests for the dissemination of the additional clustering products (72–96, 192–240 and 264–360 hour), on Monday the 8th of December 2014 ECMWF started to disseminate and archive the whole range of clustering products. Cluster products can be requested in dissemination and are also stored in MARS.

The extension of the clustering to cover the full time range of the monthly forecast considering, for example, time windows of 2 to 4 weeks ahead has been tested and the results are quite encouraging. The software used for the operational clustering has been adapted to produce additional products appropriate to the needs of individual users (e.g. different domains, variables and forecast ranges). Currently several Member States are using the same software, with the help of the user support section, to create tailored products.

The figure below shows the Ensemble members (scenarios) that better represents the cluster centroids for time window 264–360 hours for the forecast initiated 16 December 2014. Currently there are two contrasting scenarios, almost equiprobable, for the new year eve 2014. The first scenario indicates a reinforced westerly flow across the Atlantic and Europe while the second indicates strong Northerly flow with advection of cold air from the Artic regions.

Ensemble scenarios for time window 264–360 hours for the forecast initiated 16 December 2014. Maps of geopotential at 500 hpa and anomalies from a 29-year reanalysis climate (colour shading: red positive, blue negative). The geopotential field is scaled by 100

Extensions to the GRIB code enable cluster products to contain details of the cluster number and the numbers of the ensemble members from which a cluster was formed (see full description of those extensions).

The number of clusters generated varies from day to day. A maximum of 6 clusters are produced, depending on the overall spread of the ensemble. Note that all requested clusters will be disseminated even if they are "empty" clusters (i.e. if the number of requested clusters is 6 and only 4 are generated on the day, clusters 5 and 6 will be disseminated as dummy constant fields of a short length).

The following request language parameters are specific for the cluster mean and cluster representative requirements:

TYPE=CLUSTER MEAN [CM] or CLUSTER REPRESENTATIVE [CR]  NUMBER=1/2/3/4/5/6  LEVTYPE=PRESSURE LEVEL [PL]  LEVELIST=500  PARAM=Z only  GRID=a/b (latitude-longitude grid)   a= grid resolution (deg.) along a latitude line (i.e. East/West)   b=  grid resolution (deg.) along a longitude line (i.e. North/South) 1.5/1.5 or multiples thereof;   STEP are in hours, NB ALL is not permitted, only the 4 time windows are available   since the  clustering is computed independently for each time window it is advisable to retrieve the cluster  for each window in a separate file.  STEP=72/to/96/by/12 or 120/132/144/156/168 or 192/to/240/by/12 or 264/to/360/by/12

See also Graphical representations of cluster products

For questions related to the Cluster products, please contact us (Subject="Scientific or technical question about data")