NetGoodIndexSubmit a correction

Benefit Ledger · verified · Climate & Environment

GenCast beats ECMWF ENS on most 15-day ensemble weather targets

DeepMind’s GenCast, published in Nature on 4 December 2024, generated 0.25° 15-day ensemble forecasts that outperformed ECMWF ENS on 97.2 percent of 1,320 evaluated targets.

4 Dec 2024Tier 3 MajorMethodology 0.1

Current score

+2.46

10 base · Major (tier 3 of 5, 10 pts)
× 0.9000 attribution · Primary causal contribution
× 0.7000 evidence · Peer review or independent validation
× 0.6500 realization · Independently validated or deployed
× 0.6000 durability
Event-level product before credit split: 2.46

First widely reported ML ensemble to beat ENS on a broad scorecard (tier 3). Peer review. Realization is open model plus later productization, not full agency replacement. Medium durability.

What happened

GenCast is a diffusion-based probabilistic weather model trained on ERA5. The Nature paper reported better ensemble skill than ENS, including extremes, tropical-cyclone tracks, and wind-power applications, with a single 15-day member in about eight minutes on a TPU. Code and weights were released. Evaluation is on 2019 holdout data, not a multi-year operational campaign.

Model attribution

+2.46

GraphCast

Diffusion ensemble weather model compared with ECMWF ENS.

The published GenCast system is the artifact.

Attribution 0.9000 · Credit share 100% · Google DeepMind

Claims

  • GenCast outperformed ENS on 97.2 percent of 1,320 evaluation targets in the paper.

    outcome · supported

Sources

primary sources

Secondary domains: Energy, Physics

Revision history

  • 13 Sep 2026 · 0.00 → 2.46

    Initial adjudicated seed score under methodology 0.1.