MIT built a model that forecasts the flood that has never happened — 300mm of rain on New York, where the record is about 200
Kai Chang, a mechanical engineering graduate student, and Professor Themis Sapsis at MIT's Center for Computational Science and Engineering have published a method for generating extreme weather scenarios that appear nowhere in the historical record. Machine-learning weather models learn from what has happened, which means the rarest and most damaging events are precisely the ones they have seen least — often not at all. Their approach, called Extreme Event Aware, or η-learning, combines two kinds of data: detailed spatial maps from a short window of observations, and long-run point statistics on how often extremes of a given size occur. The statistics constrain the maps, so the model can draw a rainfall pattern that has never been observed but is neither physically nor statistically absurd, and attach estimates of how long it would last, how intense it would be and how much ground it would cover. Trained on six months of maps plus 25 years of statistics on US precipitation, it produced plausible rainfall maps for New York City reaching 300 millimetres, against a historical maximum of about 200. The work appeared in Nature Communications on 20 August.
Related stories
A third of a new ultraviolet sky map was predicted, not measured.2026-10-10Only 31 of China's AI model releases came with a safety report.
2026-10-10Mathematicians asked their field to stop working with OpenAI.
2026-10-10A new scoreboard grades how often agents do what you never asked.
2026-10-09Google's medical AI interviewed patients before their doctor did.
2026-10-09