MIT AI tool creates realistic maps of unprecedented extreme weather scenarios

Scientists at the Massachusetts Institute of Technology have introduced a machine-learning framework called Extreme Event Aware (η-learning) that can synthesize realistic representations of extreme weather events that have never been recorded. By extracting point statistics and spatial patterns from 25 years of hourly precipitation data across the continental United States, the algorithm learns how low-resolution maps relate to high-resolution outcomes and then generates storms that stay within statistically plausible limits.

The system can answer queries such as “What would a once-in-a-century storm look like in New York City?” and output maps detailing the storm’s extent and rainfall intensity. Graduate student Kai Chang notes the tool’s value for designing infrastructure resilient to rare events, while professor Themis Sapsis highlights its broader relevance to any system with limited slack. Beyond rainfall, the method could be applied to extreme floods, wildfires, robotic navigation, and financial market crashes.