Edge Rewrite
everything Source code View original
// HTMLRewriter · presentation

This page was redesigned at the edge.

Cloudflare fetched the original article and streamed it through HTMLRewriter to apply an entirely new visual system without rebuilding the source page.

// request.cf · coarse context

A page that knows where it met you.

Only coarse request metadata is shown. This demo does not display or persist visitor IP addresses.

Country
US
Cloudflare location
CMH
Connection
HTTP/2
Language
Not provided

Ray ID: a295aa24af0707d6

Jump to content

// Workers AI · dad joke modeDoes the ICON weather model have an icon? Yes it's a forecast idol.

Add links
From Wikipedia, the free encyclopedia

ICON (short for ICOsahedral Nonhydrostatic) is a global numerical weather prediction and climate modeling framework jointly developed by the German Weather Service (DWD), the Max Planck Institute for Meteorology, and other research institutions in Germany and Switzerland.[1] The framework is used for operational weather forecasting and for regional and global climate simulations.

ICON has been DWD's operational global weather forecasting model since January 20, 2015, when it replaced the hydrostatic GME model.[2][3] DWD operates the model in global and higher-resolution regional configurations. More than 30 national weather services use ICON forecasts as boundary conditions for their own regional models.[4]

Operation

[edit source]

DWD runs the global ICON forecast four times each day, with forecasts initialized at 00:00, 06:00, 12:00, and 18:00 UTC. The 00:00 and 12:00 UTC runs extend to 180 hours, while the 06:00 and 18:00 UTC runs extend to 120 hours.[4]

The forecasts are initialized from analyses produced every three hours. DWD's global data-assimilation system uses a hybrid ensemble-variational method that combines a local ensemble transform Kalman filter with three-dimensional variational analysis.[5]

Principles

[edit source]

Unlike GME, ICON uses a fully nonhydrostatic dynamical core formulated on an icosahedral-triangular Arakawa C grid. The grid is created by repeatedly subdividing the 20 faces of an icosahedron projected onto a sphere.[6] This grid structure avoids the convergence of meridians near the poles that occurs in latitude–longitude grids and provides nearly uniform horizontal resolution across the globe.[4]

The dynamical core uses a flux-form continuity equation with air density as a prognostic variable. It provides local mass conservation and mass-consistent transport of atmospheric tracers and supports two-way grid nesting.[6]

Variants

[edit source]

The global deterministic configuration of ICON has a horizontal grid spacing of about 13 km and uses 120 vertical levels. ICON-EU is a two-way nest covering Europe and adjacent regions at a grid spacing of about 6.5 km with 74 vertical levels.[7][8]

ICON-EPS is the 40-member ensemble version of the global model. It has a horizontal grid spacing of about 26 km, with a nested ICON-EU-EPS configuration at about 13 km over Europe. The global and European domains use 120 and 74 vertical levels, respectively.[7][9]

ICON-D2 is a regional configuration covering Germany and nearby countries at a horizontal grid spacing of about 2 km with 65 vertical levels. It produces 48-hour forecasts every three hours. ICON-D2-EPS is a 20-member ensemble that uses the same grid spacing, model domain, and forecast period.[10]

ICON-D2-RUC is a rapid-update configuration that uses the same model domain and approximate grid spacing as ICON-D2. New forecasts are produced every hour. Both the deterministic model and its 20-member ensemble, ICON-D2-RUC-EPS, extend to 27 hours.[11][12]

DWD also operates ICON-D05, a deterministic configuration with a horizontal grid spacing of about 500 m that primarily covers Germany. It is run every three hours and produces forecasts extending to 48 hours. Unlike ICON-D2 and ICON-D2-RUC, it does not have an operational ensemble version.[2]

Other configurations extend ICON beyond conventional weather forecasting. ICON-ART ("Aerosols and Reactive Trace gases") adds the simulation of gases, aerosols, and their interactions with the atmosphere.[13] ICON-CLM is used for regional climate simulations, while ICON-Sapphire is designed for high-resolution Earth-system simulations at horizontal grid spacings finer than 10 km.[14][15]

DWD also operates AICON-Global, a deterministic machine-learning weather forecasting system developed using the Anemoi framework. AICON uses ICON's icosahedral grid and is run alongside the physics-based ICON model, but it is a separate forecasting system rather than a configuration of ICON's dynamical core.[16]

Usage

[edit source]

ICON forecast fields are used in several specialized forecasting systems in Germany. They provide meteorological input to DWD's Lagrangian dispersion model, which produces forecasts on demand following nuclear or chemical accidents, and to its ocean wave model. German state hydrological offices use ICON precipitation fields for flood forecasting, while the Federal Maritime and Hydrographic Agency uses its wind fields for storm-surge prediction.[4]

More than 30 national weather services use ICON forecasts as lateral boundary conditions for their own limited-area models.[4] The COSMO consortium also uses ICON as its main modeling framework for operational weather forecasting and research.[17]

DWD distributes ICON forecast data in GRIB2 format through its free Open Data Server.[10] Since January 31, 2024, the ICON source code has been available under the permissive BSD 3-clause license, which allows commercial use.[18][19]

Accuracy

[edit source]

The accuracy of ICON varies with forecast lead time, region, weather variable, and the method used to evaluate the forecast.[20] DWD compares ICON with the Integrated Forecast System (IFS) in its reports on the application and verification of ECMWF forecast products.[21][22]

DWD's 2021 and 2024 reports evaluated different aspects of ICON and IFS forecast performance. In its 2021 report to ECMWF, DWD stated that IFS upper-air forecasts continued to have smaller errors than ICON forecasts, although the difference between the models had decreased.[21] In its 2024 report, DWD reported that ICON had lower root-mean-square errors than the IFS for many surface variables over the Northern Hemisphere during the first forecast days. The same report noted that significant trends were difficult to identify over DWD's smaller German verification domain.[22]

An international model-comparison study published in 2022 evaluated seven global forecasting models, including ICON, using the same initial conditions. The study found that all of the models produced high-quality medium-range forecasts but differed substantially in their temperature and precipitation biases.[23] A subsequent comparison found that, at forecast lead times from 120 to 168 hours, the two IFS configurations had the lowest tropical-cyclone track errors, followed by ICON and SHiELD.[24]

The two DWD reports covered different variables, geographic domains, forecast lead times, and verification periods. Their results therefore describe particular aspects of forecast performance rather than a single overall comparison of the two forecasting systems.[21][22]

Development history

[edit source]

Development of ICON began in 2004 with research on the model grid and the numerical formulation of its dynamical core using an idealized shallow-water framework.[25]

Transition from GME and COSMO

[edit source]

ICON became DWD's operational global weather forecasting model on January 20, 2015, replacing GME.[2] A parallel-testing phase for the higher-resolution ICON-EU nest began in July 2015. ICON-EU replaced COSMO-EU in December 2016.[26]

DWD introduced the 40-member global ICON-EPS ensemble in January 2018.[2] On February 10, 2021, ICON-D2 and ICON-D2-EPS replaced COSMO-D2 and COSMO-D2-EPS for DWD's high-resolution regional forecasts.[27]

Recent developments

[edit source]

On November 23, 2022, DWD increased the number of vertical levels in global ICON from 90 to 120 and in ICON-EU from 60 to 74. The upgrade also reduced the horizontal grid spacing of ICON-EPS from about 40 to 26 km globally and from about 20 to 13 km over Europe.[9]

ICON's source code was released publicly under the BSD 3-clause open-source license on January 31, 2024.[18]

DWD added AICON-Global, a deterministic machine-learning weather forecasting system, to its operational numerical weather prediction system on September 3, 2025. It was initially used for evaluation, research, and training alongside ICON rather than as a replacement for the physics-based model.[28][16]

AICON-Global is run four times daily, with forecasts initialized at 00:00, 06:00, 12:00, and 18:00 UTC.[29] It uses a global icosahedral grid with a horizontal spacing of about 13 km and predicts atmospheric fields on 13 selected ICON vertical levels. The model was trained using the ICON-DREAM reanalysis and uses a GraphCast-like encoder–processor–decoder architecture constructed directly on ICON's triangular grid.[16]

AICON was developed using Anemoi, an open-source Python framework created jointly by the European Centre for Medium-Range Weather Forecasts and several European national weather services. The framework provides tools for preparing training data, training machine-learning models, and deploying them in operational forecasting systems.[30]

See also

[edit source]

References

[edit source]
  1. ↑ "ICON Partners". ICON partnership. Retrieved August 3, 2026.
  2. 1 2 3 4 "Database Reference for the Global and Regional ICON and ICON-EPS Forecasting System" (PDF). Deutscher Wetterdienst. Retrieved August 3, 2026.
  3. ↑ "ICON – Modell zur Wettervorhersage und Klimasimulation". Earth System Knowledge Platform. Retrieved August 3, 2026.
  4. 1 2 3 4 5 "Numerical weather prediction models – ICON (Icosahedral Nonhydrostatic) Model". Deutscher Wetterdienst. Retrieved August 3, 2026.
  5. ↑ "Data Assimilation". Deutscher Wetterdienst. Retrieved August 3, 2026.
  6. 1 2 Zängl, Günther; et al. (2015). "The ICON (ICOsahedral Non-hydrostatic) modelling framework of DWD and MPI-M: Description of the non-hydrostatic dynamical core". Quarterly Journal of the Royal Meteorological Society. 141 (687): 563–579. Bibcode:2015QJRMS.141..563Z. doi:10.1002/qj.2378.
  7. 1 2 "Modelldokumentation des Global- und Regionalmodells ICON, ICON-EPS" (in German). Deutscher Wetterdienst. Retrieved August 3, 2026.
  8. ↑ "Numerische Modellvorhersagedaten" (in German). Deutscher Wetterdienst. Retrieved August 3, 2026.
  9. 1 2 "Operational NWP System: Resolution upgrade in global ICON / ICON-EPS" (PDF). Deutscher Wetterdienst. November 23, 2022. Retrieved August 3, 2026.
  10. 1 2 "NWP forecast data". Deutscher Wetterdienst. Retrieved August 3, 2026.
  11. ↑ "New ICON-D2-RUC/ICON-D2-RUC-EPS NWP data available". Deutscher Wetterdienst. December 2, 2025. Retrieved August 3, 2026.
  12. ↑ "Extension of the forecast horizon for ICON-D2-RUC(-EPS) to +27h". Deutscher Wetterdienst. June 2026. Retrieved August 3, 2026.
  13. ↑ "The ICON-ART Model System". Karlsruhe Institute of Technology. Retrieved August 3, 2026.
  14. ↑ Pham, Trang Van; et al. (2021). "ICON in Climate Limited-area Mode (ICON release version 2.6.1): a new regional climate model". Geoscientific Model Development. 14 (2): 985–1005. Bibcode:2021GMD....14..985V. doi:10.5194/gmd-14-985-2021.
  15. ↑ Hohenegger, Cathy; et al. (2023). "ICON-Sapphire: simulating the components of the Earth system and their interactions at kilometer and subkilometer scales". Geoscientific Model Development. 16 (2): 779–811. Bibcode:2023GMD....16..779H. doi:10.5194/gmd-16-779-2023.
  16. 1 2 3 Prill, Florian; Jacob, Marek (September 17, 2025). "AICON – Introducing ML-based weather forecasting at DWD" (PDF). Deutscher Wetterdienst, presented at the ECMWF Workshop on HPC in Meteorology. Retrieved August 3, 2026.
  17. ↑ "Collaborating communities". ICON partnership. Retrieved August 3, 2026.
  18. 1 2 "Milestone in Climate and Weather Research: Weather and Climate Model ICON published under Open Source License". German Climate Computing Centre. January 31, 2024. Retrieved August 3, 2026.
  19. ↑ "Getting started with ICON". ICON partnership. Retrieved August 3, 2026.
  20. ↑ "Verification". Deutscher Wetterdienst. Retrieved August 3, 2026.
  21. 1 2 3 Deutscher Wetterdienst (2021). Application and Verification of ECMWF Products 2021 (PDF) (Report). European Centre for Medium-Range Weather Forecasts. Retrieved August 3, 2026.
  22. 1 2 3 Deutscher Wetterdienst (September 2024). Application and Verification of ECMWF Products 2024 – Germany (PDF) (Report). European Centre for Medium-Range Weather Forecasts. Retrieved August 3, 2026.
  23. ↑ Magnusson, Linus; et al. (2022). "Skill of Medium-Range Forecast Models Using the Same Initial Conditions". Bulletin of the American Meteorological Society. 103 (9): E2050–E2068. Bibcode:2022BAMS..103E2050M. doi:10.1175/BAMS-D-21-0234.1.
  24. ↑ Chen, Jan-Huey; et al. (2023). "Tropical Cyclone Forecasts in the DIMOSIC Project—Medium-Range Forecast Models with Common Initial Conditions". Earth and Space Science. 10 (7) e2023EA002821. Bibcode:2023E&SS...1002821C. doi:10.1029/2023EA002821.
  25. ↑ Zängl, Günther; Reinert, Daniel; Prill, Florian (2022). "Grid refinement in ICON v2.6.4". Geoscientific Model Development. 15 (18): 7153–7176. Bibcode:2022GMD....15.7153Z. doi:10.5194/gmd-15-7153-2022.
  26. ↑ "Priority Project "C2I": Transition of COSMO to ICON-LAM". Consortium for Small-scale Modelling. Retrieved August 3, 2026.
  27. ↑ "Operational NWP System: Replacement of COSMO-D2 / COSMO-D2-EPS with ICON-D2 / ICON-D2-EPS". Deutscher Wetterdienst. February 10, 2021. Retrieved August 3, 2026.
  28. ↑ "Operational NWP System: Introduction of AICON-Global". Deutscher Wetterdienst. September 2, 2025. Retrieved August 3, 2026.
  29. ↑ "Index of AICON forecast runs". Deutscher Wetterdienst. Retrieved August 3, 2026.
  30. ↑ "Introducing Anemoi: a new collaborative framework for ML weather forecasting". European Centre for Medium-Range Weather Forecasts. October 2024. Retrieved August 3, 2026.
[edit source]
Retrieved from "https://en.wikipedia.org/w/index.php?title=ICON_(weather_model)&oldid=1368055376"