Microsoft open-sources a tool to improve weather forecasts

The method is meant to make medium-range forecasts easier to refine without large GPU systems.

Microsoft has published an open-source machine learning method, Probabilistic Bias Correction (PBC), that aims to make medium-range weather forecasts more accurate and cheaper to improve, especially for agencies that cannot rely on large supercomputers or GPU-heavy AI systems. The company says the approach can work on CPUs by correcting systematic bias in existing forecast models rather than replacing them, and it is meant to extend useful prediction quality into the two-to-six-week window. In tests reported in Nature, PBC improved a leading ECMWF physics-based model for 98 percent of precipitation targets and 92 percent of temperature targets, and it also doubled the accuracy of ECMWF’s newer AI forecasting system in the study analysis. The work was done by Microsoft researchers with collaborators from MIT, ECMWF, and other institutions, and ECMWF had already recognized the research in its AI Weather Quest competition before publication. The team is also testing PBC in a Kenya Meteorological Service Authority project with Rhiza Research and Gates Foundation funding ahead of a severe rainy season tied to El Niño.