🛰️ Daily AI Frontier
‹ back to 2026-08-09

Timestep-Conditioned Transformers for Global Weather Forecasting

Research AI Weather Forecasting

Ranking

Overall 58
Content 65
Popularity 42

Observed public metrics from 1 member.

Representative image for Timestep-Conditioned Transformers for Global Weather Forecasting

Merged summary

TL;DR - GEM-3 is a ~134M-parameter probabilistic global weather model whose autoregressive timestep can be chosen at inference time from a single set of weights, removing the usual fixed-timestep trade-off between sub-daily detail and error accumulation. It matters because one model can serve both short-range and extended-range forecasting without retraining specialists.

  • Fixed timesteps force a trade-off: short steps (1–6h) resolve diurnal dynamics but accumulate more error over a horizon, while 24h steps reduce accumulation but lose sub-daily usability.
  • Explicit multi-timestep inference lets users configure the step at run time; mixed-timestep training also consistently improved rollout stability versus timestep-specialist models.
  • Architecture is a lightweight neighborhood-attention transformer on an equirectangular grid, extending the earlier GEM-2 with further architectural changes.
  • Claimed outcome is near-SOTA medium-range probabilistic skill plus stable extended-range rollouts, efficient training/inference, and decision-relevant diagnostics; no specific metrics are given in the provided abstract.

Sources (1)

Timestep-Conditioned Transformers for Global Weather Forecasting

arXiv cs.LG Sam Levang, Fran Bartolic, Ty Dickinson, Chase Dwelle, Paulius Rauba, Viktor Cikojevic 2026-08-06 arXiv:2608.06241
Public signals Semantic Scholar citations 0 · Semantic Scholar influential citations 0
Providers: Hugging Face · N/A OpenAlex · N/A Publisher · N/A Semantic Scholar · Citations 0 · Influential citations 0 X · N/A Fetched 2026-09-02 14:24:56.522021 UTC

TL;DR - GEM-3 is a ~134M-parameter probabilistic global weather model whose autoregressive timestep can be chosen at inference time from a single set of weights, removing the usual fixed-timestep trade-off between sub-daily detail and error accumulation. It matters because one model can serve both short-range and extended-range forecasting without retraining specialists.

  • Fixed timesteps force a trade-off: short steps (1–6h) resolve diurnal dynamics but accumulate more error over a horizon, while 24h steps reduce accumulation but lose sub-daily usability.
  • Explicit multi-timestep inference lets users configure the step at run time; mixed-timestep training also consistently improved rollout stability versus timestep-specialist models.
  • Architecture is a lightweight neighborhood-attention transformer on an equirectangular grid, extending the earlier GEM-2 with further architectural changes.
  • Claimed outcome is near-SOTA medium-range probabilistic skill plus stable extended-range rollouts, efficient training/inference, and decision-relevant diagnostics; no specific metrics are given in the provided abstract.
item →