FourCastNet v2-small GFS Reforecast API

GribStream Code: aiwpfourcastnet

Official source: The CIRA and NOAA GSL AI Weather Prediction archive runs FourCastNet v2-small from NOAA GFS analyses for reproducible community evaluation.

Current GribStream coverage: aiwpfourcastnet begins with the September 30, 2020 12 UTC cycle and grows as new cycles arrive. That timestamp is the defined start of this GribStream series.

Experimental research availability: The registry warns that data may be missing and availability is not guaranteed; the public README says the data are provided as-is and documents earlier publication-timing and directory-structure changes. This evaluation dataset is not an official operational NOAA or ECMWF forecast. Do not make it the sole dependency for availability-sensitive systems: monitor forecasted_at freshness and keep an operational fallback dataset.

Version continuity: Each GribStream code is pinned to the current source-model version. A materially different upstream model version will receive a separate dataset code rather than being silently joined to this history.

Description

FourCastNet v2-small is a data-driven global model. AIWP runs this variant with GFS from NOAA GFS analyses and selects a period beginning in October 2020 to avoid overlap with model training or fine-tuning.

The FourCastNet v2-small paper describes its design. FourCastNet v2-small uses a spherical Fourier neural operator, performing global spectral operations on the sphere to support stable autoregressive weather prediction. Published results describe specific experiments and do not guarantee performance for this current archive.

The FourCastNet v2-small inventory covers 2 metre temperature, 10 and 100 metre wind, surface and mean sea-level pressure, total column water vapour, and five upper-air variables on 13 pressure levels. These fields support analysis of large-scale circulation, thermal structure, moisture transport, and weather-system evolution.

The aiwpfourcastnetifs variant uses the other initial-condition system. Comparing them helps study sensitivity to the starting state, but does not isolate model quality by itself: GFS and IFS are different analysis systems. This page covers GFS.

This deterministic product does not quantify forecast uncertainty and does not include precipitation, vertical velocity, clouds, radiation, or surface fluxes. For operational decisions, compare it with current authoritative guidance. The team asks users to cite the AIWP archive paper.

Detail

• Archive begins: 2020-09-30
• Official AIWP dataset registry: https://registry.opendata.aws/aiwp/
• Key features:
  • Complete 1,440 by 721 global grid: 1,038,240 cells at 0.25 degree spacing.
  • 00 and 12 UTC cycles with forecast hours 0 through 240 at 6-hour intervals.
  • 13 source variable names and 73 exact selectors: 2 metre temperature, 10 and 100 metre wind, surface and mean sea-level pressure, total column water vapour, and five upper-air variables on 13 pressure levels.
  • Published names, levels, float32 values, and units remain unchanged; the generated inventory documents every selector.
• Resolution: Regular global 0.25 degree grid with 1,440 longitudes by 721 latitudes.
• Temporal coverage: 00 and 12 UTC cycles; 6-hour steps through 240 hours, or 10 days.
• Primary uses: AI-model evaluation, initial-condition comparison, forecast verification, synoptic analysis, wind and temperature studies, cyclone research, and machine-learning benchmarks.

Related datasets

• The same model with the other initial-condition system
• GFS for NOAA operational global guidance
• IFS Oper for ECMWF operational global guidance

Parameters