Edge Rewrite
pun Source code View original
Jump to content

// Workers AI · dad joke modeWhat did the Integrated Forecasting System say to the weather? You're always forecasting our future together.

From Wikipedia, the free encyclopedia

The Integrated Forecasting System (IFS) is a global numerical weather prediction system jointly developed and maintained by the European Centre for Medium-Range Weather Forecasts (ECMWF) based in Reading, England, and Météo-France based in Toulouse.[1] The version of the IFS run at ECMWF is often referred to as the "ECMWF" or the "European model" in North America, to distinguish it from the American Global Forecast System. Since February 2025, ECMWF has also run a complementary machine-learning-based system, the Artificial Intelligence Forecasting System (AIFS), alongside the physics-based IFS (§ Artificial Intelligence Forecasting System (AIFS)).[2]

Mechanism

[edit]

The IFS comprises a spectral atmospheric model with a terrain-following vertical coordinate system coupled to a 4D-Var data assimilation system. In 1997 the IFS became the first operational forecasting system to use 4D-Var.[3] With IFS Cycle 50r1 (12 May 2026), the assimilation system became fully coupled across the atmosphere, ocean, and sea ice, allowing the atmospheric model to draw sea surface temperatures directly from the coupled ocean model rather than through the earlier partial-coupling scheme.[4]

Both ECMWF and Météo-France use the IFS to make operational weather forecasts, but using a different configuration and resolution (the Météo-France configuration is referred to as ARPEGE). It is one of the predominant global medium-range models in general use worldwide; its most prominent rivals in the 6–10 day medium range include the American Global Forecast System (GFS), the Canadian Global Environmental Multiscale Model (GEM and GDPS) and the UK Met Office's Unified Model.

Variants

[edit]

ECMWF runs the IFS in several configurations. Until 2026, the highest-resolution deterministic run ("HRES") was produced separately from the ensemble, but successive cycles merged the two: IFS Cycle 48r1 (27 June 2023) raised the ensemble's horizontal resolution from 18 km to 9 km, matching HRES,[5] Cycle 49r1 (12 November 2024) made the HRES run and the ensemble control member computationally identical, and Cycle 50r1 (12 May 2026) stopped running them separately, producing a single run known as "ENS Control" or "Control forecast".[4] This run is produced four times daily (00Z, 06Z, 12Z, 18Z), with the 00Z/12Z runs extending to 15 days and the 06Z/18Z runs to 6 days, at a horizontal resolution of 9 km using 137 layers in the vertical.[4] The wider 51-member ensemble ("ENS") runs at the same resolution and schedule.

Since Cycle 48r1, a separate sub-seasonal range ensemble (renamed from "extended-range" at Cycle 49r1) has run daily from 00 UTC out to 46 days with 101 members, at 36 km horizontal resolution and 137 vertical layers.[5]

ECMWF also produces seasonal forecasts through SEAS5, operational since 5 November 2017. SEAS5 runs as a 51-member ensemble at 36 km resolution out to 7 months, and is extended four times a year to a 13-month "annual-range" forecast using a 15-member subset.[6][7]

Artificial Intelligence Forecasting System (AIFS)

[edit]

Since 2025, ECMWF has run a parallel machine-learning-based forecasting system alongside the physics-based IFS: the Artificial Intelligence/Integrated Forecasting System (AIFS), named as a nod to the IFS itself.[8] AIFS is a data-driven model built on a graph neural network encoder and decoder with a transformer-based processor, developed with ECMWF's Anemoi toolkit and trained on the ERA5 reanalysis together with operational IFS analyses.[9]

The deterministic version, AIFS Single, has run operationally since 25 February 2025 — the first operational use of a global data-driven forecasting model at a major international weather centre — generating a complete 15-day forecast within minutes on ECMWF's computing infrastructure.[10] An ensemble version, AIFS ENS, has run operationally since 1 July 2025.[11] Both were upgraded to version 2 alongside IFS Cycle 50r1 on 12 May 2026, adding ECMWF's first data-driven wave and snow-cover forecasts.[4]

Usage

[edit]

Many ECMWF member states use ECMWF global forecasts to provide boundary conditions for their own higher resolution, limited domain forecasts.[12]

ECMWF forecasts remain free to the national weather services of its member states. Public access to ECMWF's real-time forecast data has broadened substantially since the early 2020s: following a phased transition approved by the ECMWF Council in December 2020, ECMWF reached full open-data status on 1 October 2025, and its entire Real-time Catalogue — covering both IFS and AIFS output — is now licensed under Creative Commons Attribution 4.0 (CC-BY-4.0), permitting reuse, redistribution, and commercial use subject to attribution.[13][14] A free subset is published at 25 km resolution with no cost; access to the full higher-resolution catalogue (down to 9 km) may still involve delivery or service charges to cover distribution costs, rather than a licensing fee.[14] This replaced ECMWF's earlier model, under which limited public data was available only under a more restrictive non-commercial, no-derivatives licence (CC-BY-ND-NC).

The full IFS source code — including its data assimilation system — remains available only to the national weather services of ECMWF member states.[1] The OpenIFS project, a version of the forecast model without the data assimilation system, was previously available only under a bespoke, non-open-source licence from ECMWF, but was made fully open source under the Apache 2.0 licence on 5 March 2026 and published on GitHub.[15] The EC-Earth climate model is based on the IFS.[16]

History

[edit]

The European Centre for Medium-Range Weather Forecasts was established when the Convention creating it as an independent intergovernmental organisation entered into force on 1 November 1975.[17] Its first operational medium-range forecast was produced on 1 August 1979, using a grid-point model with a horizontal grid spacing of 210 km and 15 vertical levels.[18] This was replaced in April 1983 by ECMWF's first operational spectral model (T63, 16 levels), with horizontal resolution subsequently increased to T106 (19 levels) in 1985–86 and to T213 (31 levels) in September 1991.[19] A 32-member Ensemble Prediction System and the WAM ocean wave model both became operational in 1992.[19]

The system's current name dates to 1987: Mats Hamrud, tasked with rewriting the forecasting model's code so it could work with a new variational data assimilation system, proposed calling it the "Integrated Forecasting System" to reflect the close integration of the forecast model with the assimilation system.[20] In 1997, the IFS became the first operational forecasting system to use 4D-Var data assimilation.[21]

Since 2016, ECMWF has published detailed records of each operational upgrade, identified by a "Cycle" number.[22]

Cycle / systemDateNotes
Cycle 41r28 March 2016Introduced the octahedral reduced Gaussian grid, increasing HRES horizontal resolution from 16 km to 9 km and ENS from 32 km to 18 km.[23]
Cycle 43r122 November 2016Major upgrade to the NEMO dynamical ocean model (0.25°, 75 layers) and its interactive sea-ice component.[22]
Cycle 43r311 July 2017Introduced the ecRad radiation scheme, replacing the older McRad scheme, alongside improvements to convection modelling and tropical cyclone data assimilation.[24]
SEAS55 November 2017New seasonal forecasting system (§ Variants).[22]
Cycle 45r15 June 2018Wide-ranging upgrade to data assimilation, observation use, and modelling, including enhanced dynamic coupling between the ocean, sea ice, and atmosphere.[25]
Cycle 46r111 June 2019Introduced a continuous data assimilation scheme and upgraded the Ensemble of Data Assimilations to 50 members; no resolution changes.[26]
Cycle 47r130 June 2020Improved treatment of observations and data assimilation; introduced quintic vertical interpolation in the semi-Lagrangian advection scheme and an improved MODIS-based surface albedo climatology.[27]
Cycle 47r211 May 2021Ensemble vertical resolution increased from 91 to 137 levels, matching HRES.[19]
Cycle 47r312 October 2021Re-implemented on new Atos HPC hardware on 18 October 2022.[22]
Cycle 48r127 June 2023Ensemble horizontal resolution raised to 9 km, matching HRES; sub-seasonal range restructured to a daily, 46-day, 101-member forecast (§ Variants).[22]
Cycle 49r112 November 2024HRES and the ensemble control member made computationally identical (§ Variants).[22]
AIFS Single v125 February 2025First operational data-driven global forecasting model at a major international weather centre (§ Artificial Intelligence Forecasting System (AIFS)).[22]
AIFS ENS v11 July 2025[22]
Cycle 50r1 / AIFS v212 May 2026HRES and ENS Control formally merged into a single run; fully coupled atmosphere–ocean–sea-ice data assimilation introduced; AIFS upgraded to v2 with first data-driven wave and snow-cover forecasts.[4]

See also

[edit]
  • ICON – DWD's global weather forecasting system
  • Global Forecast System – the United States National Weather Service's global weather forecasting system

References

[edit]
  1. 1 2 "AGREEMENT BETWEEN ECMWF & METEO-FRANCE FOR THE ACCESS AND USE OF THE JOINTLY DEVELOPED AND MAINTAINED NWP SOFTWARE IFS/ARPEGE" (PDF). Retrieved 2017-10-22.
  2. ↑ "About our forecasts". ECMWF. 11 May 2022. Retrieved 2026-08-03.
  3. ↑ Andersson, Erik; Thépaut, Jean-Noël. "ECMWF's 4D-Var data assimilation system - the genesis and ten years in operations" (PDF).
  4. 1 2 3 4 5 "Implementation of IFS Cycle 50r1". ECMWF. Retrieved 2026-08-03.
  5. 1 2 "Implementation of IFS Cycle 48r1". ECMWF. Retrieved 2026-08-03.
  6. ↑ Johnson, S. J.; et al. (2019). "SEAS5: the new ECMWF seasonal forecast system". Geoscientific Model Development. 12 (3): 1087–1117. Bibcode:2019GMD....12.1087J. doi:10.5194/gmd-12-1087-2019.
  7. ↑ "ECMWF's new long-range forecasting system SEAS5". ECMWF. 17 January 2018.
  8. ↑ "AIFS: a new ECMWF forecasting system". ECMWF. 17 January 2024.
  9. ↑ Lang, Simon; et al. (2024). "AIFS -- ECMWF's data-driven forecasting system". arXiv:2406.01465 [physics.ao-ph].
  10. ↑ "Machine learning services at ECMWF: new capabilities for users". ECMWF. 3 February 2026.
  11. ↑ "AIFS Machine Learning data". ECMWF. 12 October 2023.
  12. ↑ "Serving meteorology". Retrieved 2017-10-22.
  13. ↑ "Open Data Roadmap". ECMWF. 20 March 2024. Retrieved 2026-08-03.
  14. 1 2 "ECMWF makes its entire Real-time Catalogue open to all". ECMWF. 2025-10-01.
  15. ↑ "ECMWF's portable global forecasting model OpenIFS now available for all". ECMWF. 2026-03-05.
  16. ↑ "EC-Earth Home". Archived from the original on 2018-11-11. Retrieved 2017-10-22.
  17. ↑ "50 years of ECMWF" (PDF). ECMWF. 2025.
  18. ↑ "Forty years of medium-range forecasting". ECMWF. 22 October 2019.
  19. 1 2 3 "MARS content". ECMWF Confluence. Retrieved 2026-08-03.
  20. ↑ "How did the Integrated Forecasting System get its name?". ECMWF. 2018.
  21. ↑ Andersson, Erik; Thépaut, Jean-Noël. "ECMWF's 4D-Var data assimilation system - the genesis and ten years in operations" (PDF).
  22. 1 2 3 4 5 6 7 8 "Changes to the forecasting system". ECMWF Confluence. Retrieved 2026-08-03.
  23. ↑ "New model cycle brings higher resolution". ECMWF. 4 August 2016.
  24. ↑ "Cycle 43r3". ECMWF. 11 May 2022. Retrieved 2026-08-03.
  25. ↑ "Summary of cycle 45r1". ECMWF. 13 July 2018. Retrieved 2026-08-03.
  26. ↑ "IFS cycle 46r1 planning update". ECMWF Confluence. Retrieved 2026-08-03.
  27. ↑ "IFS Cycle 47r1 implementation notice". ECMWF Confluence. Retrieved 2026-08-03.
[edit]
Retrieved from "https://en.wikipedia.org/w/index.php?title=Integrated_Forecasting_System&oldid=1368055920"