MIT AI maps storms absent from the record

MIT researchers’ Extreme Event Aware learning method generates statistically plausible 300-millimetre rainfall maps for New York City, even though the city’s recorded maximum is 200 millimetres, using 25 years of hourly precipitation data without requiring the exact extreme in training, according to the peer-reviewed Nature Communications paper. The method is still a scenario generator, not a forecast of a specific storm; infrastructure teams should file it as a way to stress-test grids, seawalls, and supply chains against tail events that historical records underrepresent, complementing the archive’s analysis of AI infrastructure resilience.