GribStream Blog
DWD AICON Global AI forecasts are now on GribStream
Query DWD AICON Global in GribStream: an operational deterministic AI forecast on the approximately 13 km native ICON grid, with four current public cycles and forecasts through 180 hours.
GribStream now supports DWD AICON Global as aicon: an operational deterministic AI weather forecast spanning the globe at approximately 13 km, with surface weather fields and atmospheric variables on 13 model levels.
AICON adds a distinct forecast lineage to the catalog. Deutscher Wetterdienst built it as a data-driven counterpart to the physics-based ICON family, trained to learn atmospheric evolution from DWD's ICON-DREAM reanalysis. It produces a single global scenario every three forecast hours and complements—not replaces—physical models and ensembles.
A Data-Driven Member of the ICON Family
DWD describes AICON as a machine-learning forecast system for global operational weather prediction. Unlike the physics-based ICON model, it does not advance explicit equations of motion and physical parameterizations. It learns the temporal development of the atmospheric state from the ICON-DREAM reanalysis and advances that state autoregressively in three-hour steps.
That relationship gives AICON a familiar geographical foundation. Its approximately 13 km horizontal resolution corresponds to global ICON grid 26, the native unstructured triangular grid shown in the temperature map above. DWD also uses the global ICON initial state as AICON's starting atmosphere, so the two systems share grid-cell coordinates.
AICON became operational on 3 September 2025 at 06 UTC. DWD's original operational change notice documented a nominal schedule of eight runs per day. The current public feed is narrower, so GribStream follows what DWD actually publishes rather than assuming every nominal cycle is present.
Four Current Public Cycles Through Seven and a Half Days
Live inspection of DWD Open Data currently shows four AICON cycles each day:
- 00 and 12 UTC: forecast steps every 3 hours, from +0 through +180 hours for most fields
- 06 and 18 UTC: forecast steps every 3 hours, from +0 through +120 hours for most fields
Total precipitation begins at +3 hours rather than +0. DWD can also publish later forecast steps before earlier ones because AICON computes forecast times in parallel, so consumers should discover available valid times instead of treating file arrival as strictly chronological.
GribStream retains the latest five days of AICON runs. That window supports current forecasts, recent cycle comparisons, operational dashboards, feature generation, and short backtests while keeping the model's original run time visible.
Surface Weather And Thirteen Atmospheric Levels
The initial aicon inventory contains 12 source parameter names and 72 exact selector combinations:
- Near-surface state:
T_2M,RELHUM_2M,U_10M, andV_10Mprovide 2 m temperature and relative humidity plus 10 m wind components. - Pressure:
PSandPMSLprovide surface pressure and mean sea-level pressure. - Precipitation:
TOT_PRECprovides total precipitation accumulated from the beginning of the run. - Model levels 1–13:
T,QV,U,V, andPprovide temperature, specific humidity, wind components, and pressure aloft.
TOT_PREC uses info = "acc" and units of kg/m^2. A value at forecast hour 24 is the total since the model run began, not an independent three-hour amount. Calculate interval precipitation by differencing suitable cumulative lead times from the same run.
The model-level selectors use DWD's numbered generalized vertical coordinates. They are not fixed pressure levels: select the exact level from the AICON model inventory, and use the accompanying P field when an application needs the pressure associated with that model layer.
Where AICON Is Useful—and Where It Is Not Enough
AICON's global reach, three-hour forecast spacing, and compact atmospheric state make it useful for:
- global temperature, humidity, wind, and pressure guidance
- renewable-energy and electricity-demand features
- precipitation monitoring and agriculture workflows
- aviation and marine context across national forecast domains
- comparing data-driven and physics-based forecasts from similar starting states
- multi-model systems that combine independent forecast lineages
It remains a deterministic forecast. One AICON run provides one scenario and does not quantify the probability of alternatives. For decisions sensitive to forecast uncertainty, severe weather, or precipitation extremes, combine it with ensemble guidance, observations, and authoritative warnings.
Its field set is also deliberately compact. AICON does not currently provide the broad cloud, radiation, land-surface, ocean, or convective-diagnostic inventories found in some physical forecast systems. Pair it with ICON-EU, IFS Oper, AIFS Oper, or AIGFS Surface according to the required domain, variables, and forecast lineage.
Open Data with DWD Attribution
DWD publishes AICON through its Open Data service. DWD permits commercial and non-commercial reuse under CC BY 4.0, provided the source is attributed. GribStream identifies Deutscher Wetterdienst as the creator in the catalog and structured metadata.
Start Here
- DWD AICON Global model page and parameter inventory
- ECMWF AIFS Oper model page
- ECMWF IFS Oper model page
- NOAA AIGFS Surface model page
- DWD ICON-EU model page
- GribStream API documentation
Sources
- DWD AICON Open Data announcement and model overview: https://www.dwd.de/DE/leistungen/opendata/neuigkeiten/opendata_maerz2026_1.html
- DWD AICON operational change notice: https://www.dwd.de/DE/fachnutzer/forschung_lehre/numerische_wettervorhersage/nwv_aenderungen/_functions/DownloadBox_modellaenderungen/icon/pdf_2025/pdf_icon_03_09_2025.pdf?__blob=publicationFile&v=2
- DWD AICON live Open Data tree: https://opendata.dwd.de/weather/nwp/v1/m/aicon/
- DWD numerical weather-prediction data overview: https://www.dwd.de/EN/ourservices/nwp_forecast_data/nwp_forecast_data.html
- DWD Open Data terms and attribution guidance: https://www.dwd.de/DE/leistungen/opendata/faqs_opendata.html
- Creative Commons Attribution 4.0 licence: https://creativecommons.org/licenses/by/4.0/legalcode
