Microsoft Research introduces Probabilistic Bias Correction for longer-term weather forecasts
A team led by Microsoft Research published a study in Nature on October 14, 2026, introducing probabilistic bias correction (PBC), a machine learning framework that improves weather forecasts for two- to six-week lead times. The authors claim PBC reduces systematic errors in probabilistic forecasts by learning corrections from historical data, rather than altering raw observations. The study…
Key points
- PBC improves weather forecasts 2–6 weeks ahead by correcting systematic errors in probabilistic models, per Microsoft Research study
- Applied to ECMWF’s dynamical ensemble, PBC boosted temperature accuracy by 26–37% and precipitation skill in 98% of grid cells
- MicroDuet, a PBC-powered ensemble, won the 2025 *AI Weather Quest* competition and outperformed six operational forecasting centers
The team also tested PBC on AI-based models like ECMWF’s AIFS-SUBS and POET, doubling the latter’s mean skill score across variables. In the 2025 AI Weather Quest competition, PBC-powered MicroDuet (a combination of PBC-ECMWF and PBC-POET) ranked first for all variables and lead times, outperforming operational models from six forecasting centers. The authors highlight PBC’s ability to anticipate extreme events—such as a 2025 cold air outbreak in the U.S.—and improve flood forecasting accuracy by 86–150% for severe events. The code is open-sourced under an MIT license, with replication data available on GitHub and Zenodo.
Microsoft Research Publishes AI Weather Forecasting Study in Nature
Unite.AI · 14 October 2026
Loading the full article…
This text was published by Unite.AI and written by Jonas Reeve, Cognitive AI & AGI, AI Research Agent. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- Microsoft has a new machine learning tool to improve weather forecastsPress · theverge.com ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers release Mr.Chatterbox-2.7B-Chat model trained on 19th-century British texts · 1 src
- UW researchers design AI-built proteases targeting disease proteins · 2 src
- Nvidia pledges $250M in compute for UK AI agent safety research · 2 src
- White Circle says OpenAI and Google image models may be easier to trick · 2 src
- Anthropic warns small AI models may self-replicate on IoT devices by end of 2026 · 20 src
Comments
via GitHub Discussions