MIT’s AI predicts unseen extreme weather without historical data
Engineers at MIT have unveiled an AI that maps extreme weather events that have never occurred in recorded history. Unlike conventional models that rely on historical disaster data, the new tool, called Extreme Event Aware or η-learning, generates plausible worst-case scenarios even where no such event has ever been observed.
Developed by Kai Chang, a mechanical engineering graduate student, and Professor Themis Sapsis, the algorithm combines point statistics with spatial mapping to forecast rainfall intensity and possible impact areas without needing prior examples of extreme events. “These methods assume there are very disastrous events that we have seen in the dataset,” says Chang. “We’re trying to quantify the Katrina that happens every 100 years—not just every 30 to 40.”
Beyond the limits of historical risk
Most risk models for insurers, city planners, and grid operators depend on datasets that already contain extreme events. They learn the conditions that produced past disasters and project similar patterns forward. But this approach restricts forecasts to what has already happened, leaving planners unprepared for unprecedented extremes. Sapsis highlights the challenge: “What will be the Katrina that happens every 100 years? How bad will it be?”
How the algorithm redefines forecasting
The system uses two types of data. Point statistics capture how often a given rainfall intensity occurs across a dataset, while spatial maps show how an event’s impact varies across a region. By learning the statistical relationship between intensity and spatial detail, the algorithm can generate realistic worst-case maps even when its training data contains few or no examples of such extremes.
In testing, the researchers used 25 years of hourly rainfall data for the continental US. They trained the spatial model on low- and high-resolution maps from just the first six months—periods with relatively light rainfall. The algorithm then applied point statistics from the full record to constrain how extreme the generated patterns could become. For New York City, where the highest recorded rainfall is 200 millimetres, the model produced plausible maps showing storms delivering 300 millimetres.
Why it matters
This approach shifts extreme weather forecasting from reactive to proactive. By removing the dependence on historical disasters, it helps planners prepare for plausible but unprecedented events, improving infrastructure resilience and emergency response planning. For industries and municipalities, the ability to anticipate worst-case scenarios—even those never seen before—could mean the difference between preparedness and catastrophe.
Source: AI News. AI-assisted editorial synthesis — TechnoExpress.