Microsoft PBC improves sub-seasonal weather forecasts (Nature)
Result confirmed
The takeaway
On Oct 14, 2026 Microsoft Research announced Probabilistic Bias Correction (PBC), published in Nature: a machine-learning method that learns a forecast model’s systematic errors and corrects its probabilities for the 2-to-6-week (sub-seasonal) range.
Status
- Claim
Result confirmed
- Our reporting
- High confidence
- Verification
- Peer-reviewed
- Importance
- 2 of 5
- Last verified
- 15 October 2026
Your AI and this story
- GPT-6 Astra167 days after its cutoff
- Claude Opus 5.5106 days after its cutoff
- Gemini 4 Argon197 days after its cutoff
- Grok 4.7136 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 106 days before it.
Key facts
- Method: PBC ‘takes your favorite model as input and learns in what ways it tends to make errors’ (Lester Mackey, Microsoft Research New England)
- Gains on ECMWF’s sub-seasonal model: better on 98% of precipitation, 92% of temperature and 91% of pressure targets; it ‘doubled the accuracy’ of ECMWF’s AI Forecasting System (AIFS), per Microsoft
- AI Weather Quest (ECMWF real-time competition launched August 2025): team MicroEnsemble’s MicroDuet (PBC on ECMWF’s dynamical model and the hybrid PoET model) placed first for every variable and lead time in the inaugural three-month season, ahead of six operational centres and 34 teams
- Examples given: flagged the Blizzard of 2026 (on Jan 29), Winter Storm Fern (Jan 1, 2026) and a mid-December 2025 cold outbreak (Nov 20, 2025) 19–26 days ahead
- Deployment: with Rhiza Research, the Kenya Meteorological Service Authority, UC Santa Barbara’s Climate Hazards Center and KIT; KMSA’s Phillip Okello says it ‘improved our forecasts more than any other method we tested’
- Availability: code at github.com/mouatadid/pbc; a Microsoft Weather S2S model (six-week probabilistic guidance) is in private preview
What happened
Microsoft Research published a post-processing method for sub-seasonal forecasts, the weakest range between dependable two-week weather forecasts and seasonal climate outlooks. It corrects the forecast probabilities of an existing model rather than replacing it.
Why it matters
Two-to-six-week forecasts matter for farming, water and disaster planning. The method is model-agnostic, open source and already used operationally in Kenya. The Nature paper itself was not opened here; figures are from Microsoft’s story.
Sources
3 sources from 3 sites. Numbers match the chips in the text.
3 sources: 2 primary, 1 press
Primary
- Microsoft Research: New AI framework bridges a critical gap in mid-range weather forecasts (Oct 14, 2026)microsoft.com, official
- GitHub: mouatadid/pbcgithub.com, code
Press
Changes
- Filed