AI-driven Global Model Output Availability
In partnership with the NOAA/NWS National Centers for Environmental Prediction (NCEP) through the Cooperative Opportunity for NCEP Data Using IDD Technology (CONDUIT) program, the NSF Unidata Program Center is pleased to announce the availability of artificial intelligence (AI)-driven global weather prediction model output in NSF Unidata datastreams. The three new datasets from NOAA are:
- AIGFS (Artificial Intelligence Global Forecast System)
- AIGEFS (Artificial Intelligence Global Ensemble Forecast System)
- HGEFS (Hybrid-GEFS - a hybrid ensemble of AIGEFS and Global Ensemble Forecast System output)
More information about these models can be found on the NOAA announcement page.
Dataset Structure and Sizes
These AI-driven models generate output four times daily (at 0, 6, 12, and 18Z). The output is provided on a global latitude/longitude grid at 0.25-degree grid spacing, covering 13 vertical pressure levels from 1000 to 50 hPa. Each model run begins with a 0-hour forecast, with subsequent forecasts every 6 hours, extending out to 16 days. The ensemble datasets include mean and standard deviation fields derived from the ensemble members, but not the ensemble members themselves. Regarding dataset size, a single run of either the AIGFS or HGEFS dataset is approximately 5.4 GB, while a single run of the AIGEFS dataset is larger, at approximately 8.7 GB.
Availability Timeframes
Grids from the AIGFS and AIGEFS begin transferring between three and four hours after the initial forecast time and take roughly forty minutes to arrive. Grids from the HGEFS begin transferring six to seven hours after the initial forecast time and take roughly ten minutes to arrive.
IDD Access
The three new datasets are available through the NSF Unidata Internet Data Distribution via the CONDUIT feed. The following are pqact patterns that could be used by those with access to the CONDUIT feed:
| pqact snippets |
|---|
# AI-GFS Global 0.25 degree forecast only (filename example: AIGFS_20260108_1800.grib2) CONDUIT ^data/nccf/com/aigfs/.*/aigfs\.(........)/(..)/model/atmos/grib2/aigfs FILE AIGFS_\1_\200.grib2 # AI-GEFS Global 0.25 degree forecast only (filename example: AIGFS_20260108_1800.grib2) CONDUIT ^data/nccf/com/aigefs/.*/aigefs\.(........)/(..)/ensstat/products/atmos/grib2/aigefs FILE AIGFS_\1_\200.grib2 # HGEFS Global 0.25 degree forecast only (filename example: AIGFS_20260108_1800.grib2) CONDUIT ^data/nccf/com/hgefs/.*/hgefs\.(........)/(..)/ensstat/products/atmos/grib2/hgefs FILE HGEFS_\1_\200.grib2 |
Remote access via the THREDDS Data Server
Remote access to the new datasets is provided alongside other IDD model datasets via the NSF Unidata THREDDS Data Server (TDS). Specifically:
These output are also available from the University of Wisconsin - Madison department of Atmospheric and Oceanic Sciences TDS.
Visualize the AI Forecast Models
If you are using the IDV, the new model output is available in the default catalog under “Artificial Intelligence GFS - Global Coverage.” IDV lead developer Yuan Ho has created a new bundle based on the IDV workshop Globe View bundle, which used output from a 2002 run of the NCEP Aviation (AVN) model. The new bundle can be accessed from within the IDV using the toolbar under “Bundles → Weather → Real Time Weather → AI-GFS Globe View Display“ or downloaded from the NSF Unidata's RAMADDA data repository instance.
This “refreshed” bundle loads a globe display showing color-shaded mean sea level pressure, a color-filled contour vertical cross section of temperature, 500 mb height contours, and an isosurface of wind speed (roughly depicting jet stream locations). For an example of how to use Siphon and MetPy to remotely access and subset output from the latest run of the AIGFS and visualize temperature forecasts using Python, see this Jupyter notebook.
Please review the NSF Unidata web-scraping policy prior to setting up automated access workflows to help us to maintain service quality.
Explore on the NSF Unidata Science Gateway
If you would like to try the IDV bundle or the Jypyter notebook without installing any software, you can see them in action on the NSF Unidata Science Gateway! To access the Science Gateway, you will be prompted to sign in via CILogon, a federated identity provider that allows you to access cyberinfrastructure resources using your institution's credentials. See the Science Gateway login page for more details on how to log in.
If your institution participates in CILogon, you will be able to access the NSF Unidata Science Gateway with your institutional credentials. Upon login, you will be asked which kind of server you'd like to launch: Standard Power or Medium Power. “Standard Power” will be more than enough to run and explore the Jupyter notebook. However, if you plan on exploring the IDV bundle, you will want to select “Medium Power”. You may always switch between the two by selecting “File → Hub Control Panel“ from the Jupyter Lab toolbar, selecting “Stop My Server“, then “Start My Server“.
Jupyter Notebook
You can open the example notebook inside a fully configure cloud-hosted environment by visiting the NSF Unidata Science Gateway JupyterHub. Once inside the jupyter-lab interface, the notebook will be located under “unidata-cellar/2026/01/aigfs_announcement/notebook/aigfs-announcement-example.ipynb”. If you are prompted to select a Juypter kernel when opening the notebook, please use "Python [conda env:unidata]".
IDV Bundle
You can also explore the AI GFS output (and bundle) in the IDV using NSF Unidata’s cloud-hosted desktop environment available in the Science Gateway. After logging in and reaching the JupyterLab interface, launch an NSF Unidata IDV desktop session directly in your web browser. This cloud desktop provides the same interactive visualization capabilities described above for the IDV without requiring any local software installation.
Add new comment