Ensemble Performance

The ECMWF Ensemble Prediction System (EPS) became part of the ECMWF operational suite on 19 December 1992, following a one-year pre-operational trial. From 19 December 1992 to 30 April 1994, the EPS was run only three times a week, starting at 12 UTC on Fridays, Saturdays and Sundays. From 1 May 1994 to 25 March 2003 the EPS was run once a day (at 12UTC), and then since 25 March 2003 the EPS has been running twice a day, starting at 00 and 12 UTC.

From 12 September 2006, the introduction of the VAriable Resolution Ensemble Prediction System (VAREPS) extended the ensemble forecast lenght from 10 to 15 days. In March 2008, the 15-day VAREPS was merged with the monthly forecasting system, to provide a seamless ensemble forecasting system for up to 32 days. In the new VAREPS-monthly system, the 00 UTC ensemble forecasts are performed with a coupled ocean model, while the 12 UTC forecasts are uncoupled and use persisted sea surface temperature anomalies.

On the 26th of January 2010 (model cycle 36r1), the horizontal resolution of the ECMWF single high-resolution system and of the data-assimilation system will increase to TL1279, and the EPS horizontal resolution will be increased to TL639L62 (equivalent to about 35 km in physical space) from day 0 to day 10, and to TL319L62 (equivalent to about 70 km in physical space) from day 10 to day 15 (day 32 every Thursday at 00 UTC). The resolution of the coupled global wave model will also increase (see link below). In March 2010 (model cycle 36r2) the methodology used to define the EPS initial perturbations will be changed with the replacement of the evolved singular vectors with perturbations defined by an Ensemble Data Assimilation (EDA) system.

A list of some of the most recent changes to the ensemble system can be accessed from this page:

• List of some of the most recent EPS changes (updated in April 2010)

The following files show the EPS performance in predicting the 500 hPa geopotential height from 1 May 1994 to 31 July 2006 (forecasts started at 12UTC for the Northern Hemisphere and Europe are assessed using a wide range of accuracy measures):

• Ensemble performance over NH (May 1994 - July 2006)
• Ensemble performance over Europe (May 1994 - July 2006)

The following files show the EPS performance in MJJ (May-June-July) in predicting the 500 hPa geopotential height during the latest four years (2006, 2005, 2004 and 2003):

• ACC and RMS scores, Z500 over NH and EU, for MJJ 2006
• Probabilistic scores, Z500 over NH and EU, for MJJ 2006

• ACC and RMS scores, Z500 over NH and EU, for MJJ 06-05-04-03
• Probabilistic scores, Z500 over NH and EU, for MJJ 06-05-04-03

The following files show the EPS performance in FMA (February-March-April) in predicting the 500 hPa geopotential height during the latest four years (2006, 2005, 2004 and 2003):

• ACC and RMS scores, Z500 over NH and EU, for FMA 2006
• Probabilistic scores, Z500 over NH and EU, for FMA 2006

• ACC and RMS scores, Z500 over NH and EU, for FMA 06-05-04-03
• Probabilistic scores, Z500 over NH and EU, for FMA 06-05-04-03

For a 7-month period, from 1 April to 27 October 2006, the EPS has been compared to two lagged-ensemble systems generated using either the latest 6 high-resolution ECMWF forecasts (6LAG) or the latest 4 high-resolution ECMWF forecasts (4LAG). The following report discusses the 7-month average performance of these systems over Europe, and their performance in predicting two storms that hit the Scandinavian countries in January 2005 and in October 2006:

• Comparison of the ECMWF EPS and high-resolution, lagged-forecast ensemble systems (2008, Monthly Weather review, in press)

For a 3-month period, from 1 April to 30 June 2006, the EPS has been compared to two lagged-ensemble systems generated using either the latest 6 high-resolution ECMWF forecasts (6HHL) or the latest 4 high-resolution ECMWF forecasts (4HHL). Some key performance indices for 500 hPa geopotential height forecasts (started at 12UTC) verified over the Northern Hemisphere and Europe can be accessed from this page:

• EPS versus lagged-ensemble systems (1 April 2006 - 30 June 2006)

These results follow a similar comparison based on forecasts issued in winter 2005 (from 1 December 2004 to 28 February 2005):

• EPS versus lagged-ensemble systems (1 Dec 2004 - 28 Feb 2005)

For few limited periods, the ECMWF EPS has been compared over the Northern Hemisphere to the American system developed at the National Centers for Environmental Predictions (NCEP) and the Canadian system developed at the Meteorological Service of Canada (MSC), and over the Southern Hemisphere to the Australian system developed at the Bureau of Meteorology Research Center (BMRC). Results from these comparisons have been summarised in two publications:

• Comparison of the ECMWF, MSC and NCEP ensemble systems (Buizza et al 2005)
• Comparison of the ECMWF and BMRC ensemble systems (Bourke et al 2004)

(site map)