MICHEL MARTIN, HOST:
Hurricane forecasters are increasingly using artificial intelligence to make predictions. When the Atlantic season's first hurricane formed this month, AI beat out traditional models at predicting where it would go, but it didn't get everything right. Jessica Meszaros with member Station WUSF explains.
JESSICA MESZAROS, BYLINE: Hurricane Isaias made landfall near Destin, Florida, on October 9 as a Category 2 storm. Conventional forecasting models initially predicted it was heading further west towards southeastern Louisiana. But Google DeepMind's AI model called it right out of the gate - Isaias was going to hit the western Florida Panhandle.
MICHAEL LOWRY: Google DeepMind, in particular, has really proven itself the last few years as the gold standard model.
MESZAROS: Michael Lowry is a meteorologist and hurricane expert in Miami. He says, overall, the introduction of AI modeling has been revolutionary to Hurricane forecasting, but it's really different. Traditional hurricane models take data like temperature and wind shear. Then use complicated physics equations to predict what a storm will do next. But AI relies more on history, Lowry says. It identifies patterns by looking at past storms and atmospheric data.
LOWRY: The folks that are designing and developing these models, by and large, they're not meteorologists. They're software engineers, and they're developing neural networks.
MESZAROS: A neural network is a set of computer algorithms modeled on the human brain and trained to find patterns in massive amounts of data. Lowry says it can be unsettling for meteorologists who operate in emergencies to rely on a system they don't totally understand.
LOWRY: The AI models are a black box, and we don't know why they might get a forecast right. We also don't know why they might get a forecast wrong.
MESZAROS: That's one reason why forecasters don't rely on a single model. The National Hurricane Center looks at several models, both AI and traditional, but it's still humans who interpret those predictions and decide what to tell the public. Michael Brennan directs the center.
MICHAEL BRENNAN: Every storm has different forecast challenges or air characteristics. So that's why we kind of take that consensus approach and use that blend of the best models.
MESZAROS: The Hurricane Center started incorporating Google DeepMind into its official forecasts just last year. While it topped other models in tracking the path of Isaias, Brennan says it didn't do as well as the Hurricane Center forecast in determining the intensity or strength of the storm.
BRENNAN: The NHC forecast had a lot more skill, 20- to 40% more skill than Google DeepMind did for the intensity of Isaias, and there were other models that were competitive with or better.
MESZAROS: And he says Google's model didn't do well in tracking Hurricane Simon in the eastern Pacific. Brennan's team gives regular feedback to Google DeepMind. The great thing about AI is that it's improving really fast, says meteorologist Michael Lowry.
LOWRY: They are making advancements in skill in hurricane predictions over the course of a year that would have taken us 10 years, a decade or more, to make.
MESZAROS: AI updates take only months or even weeks, unlike traditional models, which are updated annually at best. Even with AI coming in fast, it's just one tool in the kit right now. Lowry says he's not worried about artificial intelligence replacing people.
LOWRY: For me, I don't feel some sort of existential crisis with AI. If anything, I think it is a boon for our business.
MESZAROS: He says the human side of forecasting isn't going anywhere anytime soon.
For NPR News, I'm Jessica Meszaros in Tampa.
(SOUNDBITE OF SONG, "SHELTER FROM THE STORM")
BOB DYLAN: (Singing) Come in, she said, I'll give you shelter from the storm.
MARTIN: Let me mention that Google is a financial supporter of NPR.
(SOUNDBITE OF BOB DYLAN SONG, "SHELTER FROM THE STORM") Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
