TY - GEN
AU - Thomas Haiden
AU - Martin Janousek
AU - Frédéric Vitart
AU - Fernando Prates
AU - Michael Maier-Gerber
AU - Cathy Wing Yi Li
AU - Matthieu Chevallier
AB - This report provides a summary of ECMWF’s forecast performance, covering medium, sub-seasonal, and seasonal forecast ranges. Headline scores adopted by ECMWF in collaboration with its member states monitor the evolution of various aspects of forecast skill. The report gives updates on these scores, as well as supplementary scores to provide a more complete assessment of forecast skill. The implementation of IFS Cycle 49r1 in November 2024 has led to clear improvements in forecast skill. The primary focus of this summary is the medium range, specifically the forecast performance for upperair variables. It is shown that in this respect, ECMWF continues to have an overall lead among centres. For surface variables, other centres have partially taken the lead, especially in the short range, but substantial improvements of ECMWF forecasts due to model cycle 49r1 can be seen, such as a reduction of large 2-m temperature and 10-m wind speed errors in the ensemble forecast. In the sub-seasonal forecast range, increasing skill above persistence is being observed. On the seasonal timescale, the change in 2024 of Pacific SSTs from El Niño to La Niña and subsequently neutral conditions was well predicted. This report also includes scores from machine-learning forecasts (AIFS and others), higherresolution forecasts (DestinE Continuous Extremes Digital Twin), and atmospheric composition forecasts from the Copernicus Atmosphere Monitoring Service (CAMS).
BT - Technical Memoranda
CY - Reading
DA - 09/2025
DO - 10.21957/51e665e5c8
M1 - 931
N2 - This report provides a summary of ECMWF’s forecast performance, covering medium, sub-seasonal, and seasonal forecast ranges. Headline scores adopted by ECMWF in collaboration with its member states monitor the evolution of various aspects of forecast skill. The report gives updates on these scores, as well as supplementary scores to provide a more complete assessment of forecast skill. The implementation of IFS Cycle 49r1 in November 2024 has led to clear improvements in forecast skill. The primary focus of this summary is the medium range, specifically the forecast performance for upperair variables. It is shown that in this respect, ECMWF continues to have an overall lead among centres. For surface variables, other centres have partially taken the lead, especially in the short range, but substantial improvements of ECMWF forecasts due to model cycle 49r1 can be seen, such as a reduction of large 2-m temperature and 10-m wind speed errors in the ensemble forecast. In the sub-seasonal forecast range, increasing skill above persistence is being observed. On the seasonal timescale, the change in 2024 of Pacific SSTs from El Niño to La Niña and subsequently neutral conditions was well predicted. This report also includes scores from machine-learning forecasts (AIFS and others), higherresolution forecasts (DestinE Continuous Extremes Digital Twin), and atmospheric composition forecasts from the Copernicus Atmosphere Monitoring Service (CAMS).
PB - ECMWF
PP - Reading
PY - 2025
T2 - Technical Memoranda
TI - Evaluation of ECMWF forecasts
UR -  
ER -