‘Not a replacement’: How AI is transforming weather predictions

Amy McGovern of the University of Oklahoma explains how AI weather models can be used to create 'hyper-personalized forecasts.'

Artificial intelligence is reshaping how people predict the weather by learning from decades of forecasts – not by rewriting the laws of physics, according to one expert.

Amy McGovern, a professor with the School of Meteorology at the University of Oklahoma, told CTV’s Your Morning on Thursday that AI weather models mark a new era of prediction that’s faster, more adaptive and increasingly personal.

“The current ones are built on basically emulating the laws of physics,” McGovern explained. “They’re just cubes that are all talking to each other. Each one of them has laws of physics, and they’re having to pass the information from the different cubes to each other.”

In contrast, she said, “The AI weather models are trained on 50 or 60 years of historical data, and they just learn to mimic, given what they’ve seen in the past, what they’re going to see in the future.”

One standout model is Google’s DeepMind, which recently drew attention for its accuracy in predicting hurricanes, which McGovern said “helps you prepare more in advance.”

“If you know that the model is doing well and it’s predicting that the formation of the hurricane is going to happen a day or two before all the other models, you have that much more time to get that information out there and to get resources in place,” she said.

McGovern cited Hurricane Melissa as an example.

“(DeepMind) was able to predict that it was going to intensify to a Category 5 faster than all the other models,” she noted. “Then you can put the resources in place to help people survive.”

The catastrophic storm made landfall in western Jamaica It shredded Jamaica’s western region and then made landfall in eastern Cuba, where it destroyed homes and crops last month and was one of the strongest Atlantic hurricanes on record.

While these AI systems aren’t yet official integrated into national forecasts, McGovern said they’re already influencing them.

“(The U.S.-based) National Hurricane Center, when they put out their information on each hurricanes this year, they reference the Google’s DeepMind model,” she said.

For McGovern, who has spent two decades working on AI and weather, technology’s potential goes beyond storm tracking.

“This gives us the real possibility to be able to give us (a) personalized forecast – heavily hyper-personalized forecasts,” she said. “If people in that region have different needs … they’re farmers or have small business that needs to react to the weather in a specific way, they could get a forecast that’s personalized to them.”

Still, she cautioned AI won’t replace meteorologists anytime soon.

“It’s definitely not a replacement,” McGovern said. “All models are wrong. Some of them are useful. I think it’s going to be more useful.”

With files from The Associated Press