A report this week from The Washington Post examined a persistent and consequential gap in artificial intelligence weather modeling: while AI has made striking advances in predicting where hurricanes will go, predicting how strong they will be when they arrive remains a fundamentally harder problem that current models have not solved.
The core difficulty, as The Washington Post explains, is rapid intensification—the phenomenon where a storm's sustained winds increase by 35 miles per hour or more within 24 hours. This is not a rare edge case. Hurricane Ian in 2022 underwent rapid intensification before striking Florida's southwest coast as a Category 4 storm. Michael in 2018 did the same before making landfall as a near-Category 5. AI systems trained on historical storm data struggle with rapid intensification partly because the physical processes driving it—warm ocean eddies below the surface, subtle atmospheric wind shear patterns, inner core convective dynamics—are either poorly observed or chaotic in ways that resist pattern-matching at the scale AI models currently operate.
Traditional numerical weather prediction models built on physical equations have the same weakness, but the hope inside the meteorological community was that AI's ability to find non-obvious correlations in massive datasets would close the gap. So far, that has not happened in any operationally reliable way. The National Hurricane Center and international agencies still lean heavily on an ensemble of both approaches, and forecasters note that even the best AI-assisted outlooks carry substantial intensity uncertainty windows measured in tens of miles per hour.
The reason this matters specifically to people preparing for hurricane season goes beyond the obvious point that Category 2 and Category 5 are different threats. Preparedness timelines are calibrated to intensity forecasts in ways that are rarely discussed publicly. Evacuation zone triggers, shelter-in-place versus leave decisions, and critically, the point at which fuel, water, and medical resupply chains into a region get preemptively severed by state emergency managers—all of these administrative decisions are made against a forecast cone that includes an intensity number people tend to treat as more reliable than it actually is. A household that waits until a confirmed Category 4 forecast before acting may be waiting for information that arrives inside the window where rational options have already closed. The intensity forecast is, by the admission of the scientists building these systems, the number with the widest real-world error bar, and it is the number that most directly determines structural damage, storm surge height, and post-landfall access to the area.
