I've been studying, forecasting, and writing about hurricanes for more than two decades, way back when we drew weather maps by hand (seriously, it's a thing). Helene and Milton were the first two hurricanes where I leaned hard on AI-generated weather forecasts. It feels like a turning point.
In our era of escalating climate emergency, the warming atmosphere is helping to make weather more extreme and more dangerous -- putting more people into harm's way each year. More confident predictions of which hurricanes will go beast mode and which ones will safely fizzle out gives people more time to prepare.
During Helene and Milton, the AI tool I used most -- AI RI -- was developed by researchers at the University of Wisconsin. It gives updated hourly odds on the chances of a nascent hurricane going through a bout of rapid intensification. At one point, AI RI was giving a nearly 100% chance that Milton would strengthen from Category 1 to Category 5 within the next 24 hours. And of course, that prediction turned out to be correct. No Atlantic hurricane in our 175 years of recordkeeping strengthened faster than Milton.
Such an accurate prediction would have been unthinkable even five years ago using traditional computer weather models. Just a century ago, it was nearly impossible to reliably anticipate adverse weather conditions on any timescale. To give a sense of the scale of progress, a four-day weather forecast is as accurate now as a one-day forecast was in 1995. And AI promises to extend those gains days, weeks, and months into the future and at finer and finer geographic scales, even down to the microclimate and neighborhood level.