New MIT Method Predicts Extreme Storms Without Past Extreme Data

MIT engineers have developed a machine-learning method that generates plausible extreme events — such as how far a wildfire might spread, or how bad a once-in-a-century storm could be — without needing training data that already contains past extreme events. Kai Chang, an MIT graduate student in mechanical engineering, and Themis Sapsis, the William I. Koch Professor of Mechanical and Ocean Engineering at MIT, describe the method, called Extreme Event Aware, or "η-learning," in a paper published Aug. 20 in the open-access journal Nature Communications.

Most existing risk-assessment tools depend on records that already contain extreme events to project worse scenarios. "We are trying to model extreme, unprecedented events that no one has seen before, that are not in the dataset," Chang said. Citing Hurricane Katrina, which recurs every 30 to 40 years, Sapsis said the team wants to know: "What will be the Katrina that happens every 100 years? How bad will it be?"

To demonstrate the method, the team generated maps of future extreme precipitation events over the continental United States. Using 25 years of hourly precipitation data pooled into daily maps, they computed rainfall statistics from the full record but trained the algorithm on spatial map patterns from only the first six months, which had few or no extreme examples. The algorithm learned how low-resolution rainfall patterns correspond to detailed, high-resolution maps, then used the statistics to constrain the extremes in the maps it generated.

For example, if the most extreme rainfall ever recorded in New York City is 200 millimeters, the method can generate plausible spatial patterns for a storm producing an even more extreme 300 millimeters, showing where such a storm might strike and how large an area it could cover.

Beyond weather, the researchers said the same approach could apply to fields such as robotic navigation and financial markets. "Financial market crashes are extreme events that are a complicated combination of things, involving many different sectors," Chang said. "What is the interaction that leads to a market crash?"