← All notes

Summary

GraphCast is a machine-learning weather model trained from reanalysis data. It predicts hundreds of weather variables, over 10 days at 0.25 degree resolution globally, in under one minute. The authors show that GraphCast significantly outperforms the most accurate operational deterministic systems on 90% of 1380 verification targets, and that its forecasts help with severe events including tropical cyclones, atmospheric rivers, and extreme temperatures. The method uses graph neural networks and is trained on ECMWF ERA5 reanalysis. Code and trained weights are pointed to a public GitHub repository. [S1]

Claims

• id: C1 claim: 'GraphCast predicts hundreds of weather variables over 10 days at 0.25 degree resolution, in under one minute.' source: S1 evidence: "over 10 days at 0.25 degree resolution globally, in under one minute"
• id: C2 claim: 'It outperforms the leading operational deterministic systems on 90% of 1380 verification targets.' source: S1 evidence: "significantly outperforms the most accurate operational deterministic systems on 90% of 1380 verification targets"
• id: C3 claim: 'Code and trained weights are pointed to a public GitHub repository.' source: S1 evidence: "https://github.com/deepmind/graphcast"

Verification

• Code available: Paper: https://github.com/deepmind/graphcast, which returned HTTP 200 and redirected to https://github.com/google-deepmind/weathernext on 2026-11-11. Not run.
• Reproduced: not reproduced (2026-11-11). Code not run.
• Related work: No live library note is a paper-verification page for GraphCast itself. A pointer note in this week's batch covers the Fourier Neural Operator.
• Limits: The comparison is against deterministic HRES, not a full ensemble. The authors discuss regions where HRES remains stronger.
• License: CC BY 4.0 on the abs page (2026-11-11; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 24 Dec 2022 (v1); last revised 4 Aug 2023 (v2). Sources: [S1] https://arxiv.org/abs/2212.12794, v2 (4 Aug 2023). checked_on: 2026-11-11.

When to apply

Use when asking how a learned global model compared with operational numerical weather prediction on medium-range forecasts.

Facts

• id: F1 claim: 'GraphCast predicts hundreds of weather variables, over 10 days at 0.25 degree resolution globally, in under one minute.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F2 claim: 'The authors report that GraphCast outperforms the most accurate operational deterministic systems on 90% of 1380 verification targets.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F3 claim: 'Code and trained weights were released at the GitHub URL given in the paper.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0

Sources

• [S1] arXiv 2212.12794, https://arxiv.org/abs/2212.12794. License: CC BY 4.0. v2 (4 Aug 2023). Accessed 2026-11-11.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

@misc{iil_note_bda6pydr,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{GraphCast: Learning skillful medium-range global weather forecasting}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2212-12794-graphcast/},
  note = {Permanent ID iil:note:bda6pydr. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2212-12794-graphcast.md.