Microsoft's PBC Runs AI Weather Forecasts on CPUs — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Microsoft published Probabilistic Bias Correction (PBC) in Nature, a machine learning tool that corrects systematic bias in existing physics-based and AI weather models to extend reliable forecasts to 2-6 weeks out, when both approaches typically degrade.
- PBC runs on standard CPUs, not supercomputers or powerful GPUs, making AI-enhanced forecasting accessible to resource-strapped weather agencies — a shift Microsoft Research senior principal researcher Lester Mackey called a major efficiency gain.
- PBC improved the European Centre for Medium-Range Weather Forecasts' leading physics-based model's accuracy on 98% of precipitation and 92% of temperature targets, and doubled the accuracy of ECMWF's new AI forecasting system in the study's analysis.
- ECMWF had already awarded the PBC research in its AI Weather Quest competition before the Nature paper was published; strategic lead for machine learning Matthew Chantry called the work 'a good example of new ideas being developed for the benefit and consideration of the community,' while cautioning that turning research into an operational process remains 'a most challenging step.'
- Genevieve Flaspohler, a PBC co-author who interned at Microsoft, now heads nonprofit Rhiza Research, which is testing the tool with the Kenya Meteorological Service Authority — funded by the Gates Foundation — ahead of an expected extreme rainy season tied to a record-breaking El Niño.
- PBC is now released as open-source software, developed by researchers at Microsoft, MIT, ECMWF, and other institutions.
Why it matters: Weather agencies in economically developing countries gain a low-cost AI tool that runs on CPUs rather than expensive supercomputers, and Kenya's tests ahead of a forecast severe El Niño season provide a real-world proving ground for extending reliable forecasts to 2-6 weeks for farmers and emergency responders.
Ask SkimNews
