Research
AI model doubles subseasonal weather forecast accuracy
MIT News AIhigh confidence
Why it matters
- →Breaks the historical limit where subseasonal forecasts were no better than random chance.
- →Provides a low-cost, open-source post-processing layer applicable to any existing weather model.
- →Enables earlier preparation for extreme weather events critical for energy and agricultural sectors.
Researchers from MIT, Microsoft, and other institutions have published a technique in Nature that significantly improves weather forecasts for the two-to-six-week window. The team developed an AI-based error correction layer that post-processes outputs from standard dynamical models, effectively learning and removing systematic biases. When applied to the European Centre for Medium-Range Weather Forecasts (ECMWF) model, this approach doubled predictive accuracy and outperformed climatological averages for the first time. The open-source solution is lightweight enough to run without supercomputers and has already attracted interest from meteorological offices in Kenya.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
