MIT AI forecasts extreme weather without historical data
MIT engineers built an AI that forecasts extreme weather events that have never happened before — without historical training data.
Why it matters
A novel machine learning approach to predicting tail-risk events expands what AI can forecast beyond the patterns it has seen, with implications for climate resilience planning and rare-event modeling across domains.
The key facts
10 to knowDeveloped by Kai Chang (mechanical engineering graduate student) and Professor Themis Sapsis at MIT
Generates probability maps for statistically-possible extreme weather events not in historical records
Trains without disaster data — a departure from standard forecasting approaches
Published August 25, 2026
Capability focus: extrapolation beyond training distribution for rare/tail events
Developed by Kai Chang (grad student) and Professor Themis Sapsis at MIT
Forecasts extreme weather without historical disaster data
Generates maps of statistically-possible but historically-unobserved events
Includes uncertainty estimates for each forecast
Addresses the 'black swan' problem in climate prediction
Go to the source
AI Newsartificialintelligence-news.com
Publisher excerpt: MIT engineers have built an AI tool that forecasts extreme weather without training on historical disaster data. Kai Chang, a mechanical engineering graduate student, and Professor Themis Sapsis developed the tool. It produces maps of events that have not appeared in a region’s historical record…