MIT and Microsoft AI Wrapper Boosts Extreme Weather Forecasts Up to 6 Weeks Out

Researchers from the Massachusetts Institute of Technology and Microsoft $MSFT published a study in Nature detailing a machine learning post-processing wrapper that improves subseasonal weather forecasts 2 to 6 weeks in advance. When applied to the European Centre for Medium Range Weather Forecasts' leading systems, the probabilistic bias correction (PBC) tool doubled the subseasonal accuracy of the AI model and improved the numerical model's skill across 91% of pressure, 92% of temperature, and 98% of precipitation targets. In evaluations of real-world floods over the past decade, the wrapper increased general flood prediction skill by 76% to 92%, and boosted forecasting skill by 86% to 150% for severe floods that caused more than 100 deaths or displaced at least 80k people. In the 2025 ECMWF real-time forecasting competition, the model-agnostic system outperformed numerical models from 6 operational centers, an international ensemble, ECMWF's internal AI model, and 34 competing global teams across all weather variables and periods. The open-source PBC framework does not require a supercomputer to operate, and the meteorological office for the Republic of Kenya is already experimenting with the tool to improve monsoon forecasting. "It’s model agnostic, and the paper shows that whether it’s a more traditional model or a newer AI model, we can correct these errors that are systemic to the model." — Judah Cohen, Climatologist and Research Scientist at MIT.