Microsoft's PBC Tool Improves Sub-Seasonal Forecasts — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Microsoft published open-source machine learning tool Probabilistic Bias Correction (PBC) in Nature, developed with MIT, ECMWF, and other institutions to improve 2-6 week weather forecasts
- PBC improved ECMWF's leading physics-based weather model's accuracy on 98% of precipitation and 92% of temperature targets, and doubled the accuracy of ECMWF's new AI forecasting system
- Lester Mackey of Microsoft Research New England said PBC runs on CPUs, unlike physics-based models requiring supercomputers or generative AI weather models needing powerful GPUs
- ECMWF strategic lead Matthew Chantry called the paper "a good example of new ideas being developed for the benefit and consideration of the community" but noted turning ideas into operational reality remains challenging
- Genevieve Flaspohler, a paper co-author and Rhiza Research executive director, is testing PBC with the Kenya Meteorological Service Authority (Gates Foundation-funded) ahead of an expected extreme El Niño rainy season
- PBC's methodology corrects systematic bias in existing models — such as cold-temperature bias from outdated historical records — rather than creating a standalone forecast model, making it more accessible for resource-strapped agencies
Why it matters: Weather agencies in developing countries face the highest exposure to extreme events but the lowest access to GPU infrastructure, and PBC's CPU-only design with 98% precipitation and 92% temperature accuracy gains over a leading physics-based model directly lowers that barrier — Kenya's meteorological service is already piloting it before an expected severe El Niño season.
Ask SkimNews
