Generating scenarios for extreme events, without extreme data

MIT News reports that engineers have developed a machine-learning tool that can generate plausible extreme weather scenarios without needing historical records of such events. The algorithm learns from everyday weather data, like daily precipitation maps, to project what a once-in-a-century storm might look like in terms of size, intensity, and duration. The team, led by graduate student Kai Chang and professor Themis Sapsis, says the method could help planners prepare for unprecedented events, and it could also apply to financial market crashes. The work appears in the journal Nature Communications. The researchers add that extreme events have become a strategic concern for national and economic resilience.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton