MIT describes AI method for improving two-to-six-week weather forecasts
A research team including MIT scientists reports improved subseasonal forecast tests using a machine-learning layer that corrects model errors.
What the record says
MIT’s October 14 account describes a postprocessing technique that learns systematic errors in existing weather models from historical data. The researchers report improved probabilistic forecasts for the two-to-six-week range, including extreme-weather predictions, in comparisons discussed in a Nature paper. MIT says a competition entry combining forecasting models with this correction layer placed first. The approach could be relevant to agriculture, energy planning and emergency preparation if appropriately validated and adopted.
Source details and supporting locations.
Practical next step
Read the research description for its test conditions and continue to use official weather services for current forecasts and warnings.
Limits and what is not established
Reported research and competition performance do not establish guaranteed operational forecasts, a current severe-weather warning or a doubling of accuracy for every location, event and model. The linked paper was not separately retrieved for this record.
Sources checked October 19, 2026. Later notices may change instructions or arrangements.
Original sources
Massachusetts Institute of Technology: original information