Science1 publisher2 min readPublished
Scarce storm-scale data limits AI forecasts of hurricane intensity, one researcher argues
AI weather models that rival physics systems globally lack the storm data to foresee jumps like Hurricane Polo's 24-hour rise to Category 5, a researcher says. The author's own research suggests chaos may still cap long-range forecasts even with better data.
The Scientist · Science desk
What happened
- Global AI weather models now produce forecasts that rival some of the best physics-based prediction systems, according to a researcher writing on phys.org.
- Hurricane Polo went from a tropical storm on Sept. 21, 2026, to a Category 5 hurricane within 24 hours off Mexico's Pacific coast, with winds reaching 180 mph.
- Hurricane intensity is a regional problem of fast-developing extremes, and the author says it needs far finer detail than the global datasets behind AI's gains typically provide.
- The author's recent research with colleagues suggests an element of chaos in hurricanes may prevent accurate AI intensity forecasts at long lead times.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- constraint AI intensity models will keep learning from a partial view of storms until something observes their full 3D structure over open ocean, whatever the model design or hardware.
- decision Funders weighing AI forecasting upgrades have a case for paying for open-ocean storm observation, where the author locates the shortfall, alongside work on models and chips.
- contradiction Even complete observations may not yield accurate intensity forecasts at long lead times, so more data narrows the problem without ending it.
Global skill scores reward a model for tracking the large-scale atmosphere. For that job AI has had plenty to learn from: decades of weather and climate records covering the whole planet, with millions of examples of how conditions evolve [3]. Nothing comparable exists for the inside of a hurricane, according to the researcher who wrote the phys.org piece [10]. A strong global score says little about a storm like Hurricane Michael, which grew to Category 5 in 2018 just before it hit Tyndall Air Force Base and Mexico Beach, Florida [6]. Surprises of that kind can leave communities too little time to evacuate and prepare [6].
The author describes two sources of training data, and each falls short [7][9]. Observations from weather stations, radars, buoys and satellites can be detailed. But they are mostly near coasts and unevenly spread, while many of the most important stages of hurricane development happen over open ocean [7]. Satellites help there, up to a point. They estimate only part of the rainfall, surface winds or cloud-top temperatures, and they cannot scan the full three-dimensional structure of every hurricane at the same time [8].
The second source is physics-based simulation. It gives the most complete three-dimensional picture of the atmosphere at high resolution, but every such model carries approximations, and some fine-scale processes always escape it [9]. "So, we simply don't have a good, full three-dimensional dataset to train an AI model for hurricane intensity prediction at present," the author wrote [10].
The argument is aimed at where the field's attention sits. The author credits AI forecasting's progress to three things: large amounts of weather data, better AI models and much greater computing power. Most current discussion of improvements, the essay notes, concerns models or new hardware [2]. For intensity, the essay places the shortage in the first of the three [2][10].
The author also names a second limit. Even if scientists could measure every part of thousands of storms every second, the essay says, AI still might not predict intensity perfectly, because of chaos [13]. "Tiny differences in the initial state of a hurricane can quickly grow over time," the author wrote [12].
The case is one researcher's, set out in a single essay. It does not include error figures for AI intensity forecasts, and it describes the chaos research without its methods or the lead time at which forecast skill runs out [11].
What to watch
- Publication of the author's chaos research with its methods and the forecast lead time at which intensity skill breaks down.
- Verification studies comparing AI and physics-based intensity forecasts for Hurricane Polo's 24-hour intensification.
- Any observing system that captures the full 3D structure of hurricanes over open ocean, a direct test of the data argument.