Graphcast Hindcast – 1979

Name: Graphcast Hindcast – 1979
Published: 2025-08-12
License: https://creativecommons.org/licenses/by/4.0/

Sudharsan, Naveen; Singh, Manmeet; Kamath, Harsh; Dashtian, Hassan; Dawson, Clint; Yang, Zong-Liang; Niyogi, Dev

Graphcast Hindcast – 1979

Sudharsan, Naveen et al.

Dataset

Please note:

This entry is metadata only. Metadata may change in the future. No data is available for this entry.

Summary: [ Derived from parent entry - See data hierarchy tab ]

45-year hindcast dataset (1979 to 2024) created from Google Deepmind's Graphcast-operational model. Developed by researchers at The University of Texas at Austin, this dataset provides daily 15‑day deterministic forecasts at 00 UTC on a 0.25°×0.25° global grid (≈25 km) for a 45-year period. GraphCast is a graph neural network that was trained on ECMWF’s ERA5 reanalysis. It predicts more than a dozen key atmospheric and surface variables on 13 vertical levels, delivering a full medium-range forecast in under one minute on modern hardware.

Hindcast generation: The UT Austin team used the open-source GraphCast model to produce the hindcast. Each hindcast day begins with ERA5 atmospheric analysis fields at 00 UTC as the initial state. GraphCast is then iteratively applied forward in time to produce a 15-day (360 h) forecast, with outputs every 6 hours. By rolling GraphCast’s own predictions forward, a continuous global forecast trajectory is obtained. In practice, this process was automated on high-performance cloud hardware.

Key aspects of the workflow include:
Initialization: Input state (0 h) from ERA5 00 UTC global analysis; also incorporates the 6 h prior state as GraphCast input.

Forecast integration: GraphCast operational advances the state in 6 h increments up to 360 h lead, without further data assimilation.

Output: At each lead time (6 h, 12 h, …, 360 h), the model outputs fields of interest. Results are saved in NetCDF4 format for easy access.
Project: UT Austin Graphcast Hindcast (UT-GraphCast Hindcast Dataset: A Global AI Forecast Archive from UT Austin for Weather and Climate Applications)
Spatial Coverage: Longitude 0 to 360 Latitude -90 to 90
Temporal Coverage: 1979-01-01 to 1979-12-31 (calendrical)
Use constraints: Creative Commons Attribution 4.0 International (https://creativecommons.org/licenses/by/4.0/)
Data Catalog: World Data Center for Climate
Access constraints: registered users
Format: NetCDF
Status: metadata only
Creation Date: 2025-08-12
Future Review Date: 2035-08-12
Download Permission: No - Metadata only.

Cite as: [ Derived from parent entry - See data hierarchy tab ]
Sudharsan, Naveen; Singh, Manmeet; Kamath, Harsh; Dashtian, Hassan; Dawson, Clint; Yang, Zong-Liang; Niyogi, Dev (2025). UT-GraphCast Hindcast Dataset (1979–2024): A Global AI Forecast Archive from UT Austin for Weather and Climate Applications. World Data Center for Climate (WDCC) at DKRZ. https://www.wdc-climate.de/ui/entry?acronym=graphcast

BibTeX RIS

Variable	Aggregation	Unit
air_pressure_at_mean_sea_level CF	6hr	Pa
air_temperature CF	6hr	K
air_temperature-at2m CF	6hr	K
eastward_wind CF	6hr	m s-1
eastward_wind-at10m CF	6hr	m s-1
geopotential_height CF	6hr	m
northward_wind CF	6hr	m s-1
northward_wind-at10m CF	6hr	m s-1
precipitation_amount CF	6hr	m
specific_humidity CF	6hr	kg kg-1

Parent

Graphcast Hindcast

Details

Parent project(s)

UT-GraphCast Hindcast Dataset: A Global AI Forecast Archive from UT Austin for Weather and Climate Applications

[Entry acronym: graphcast_1979] [Entry id: 5309933]