by Weather forecasting, a practice that dates back to ancient times, is being given a high-tech twist. Artificial intelligence is now being used to predict the weather – and it can make forecasts more accurate.
Weather forecasting is already a difficult task. Giant supercomputers, some the size of school buses, typically do the heavy lifting for weather forecasting using physics simulations that take hours to compute and are mostly owned by government organizations such as the European Center for Medium-Range Weather Forecasts. (ECMWF).
But that could change soon. Artificial intelligence, a force that is disrupting many industries, may soon do the same for weather forecasting. Google announced its GraphCast program Tuesday, which uses machine learning to predict the weather and is part of the company's DeepMind artificial intelligence research lab. In a media release, Google said the app offers a 10-day weather forecast within one minute and with “unprecedented accuracy.”
“There was tremendous progress last year,” Stephen Eastbrook, professor of computer science and director of the University of Toronto's School of the Environment, told Global News.
AI is used for weather forecasting as well as in other applications such as personal assistant ChatGPT. Google's GraphCast takes terabytes worth of weather data from the past four decades and uses past trends to predict how current conditions will play out, Eastbrook said.
Just as ChatGPT doesn't know what it's actually saying but predicts which word will come next based on the literature it consumes, GraphCast does the same and is actually a “mimic” of what supercomputers spend hours doing. According to Eastbrook. And it can get results – fast.
“It's about remembering all the patterns you might see in a traditional weather forecast model,” Eastbrook said. “It can produce results almost instantly.”
That speed is what almost guarantees a role for programs like GraphCast in weather forecasting, Eastbrook said. He predicts that traditional forecasters will likely use both AI and supercomputers together to improve forecasting.
Anthony Farnell, chief meteorologist at Global News, said he was excited about AI's potential in weather forecasting, and even found it outperformed the vast majority of regular numerical models. Its speed can help meteorologists adapt on the fly, he said, and it can help with evacuations in case of weather emergencies and assess who may be at risk.
Eastbrook noted that AI has the potential to identify major weather systems better than supercomputers, and Google said GraphCast accurately predicted Hurricane Lee would make landfall in Nova Scotia nine days in advance, six days ahead of traditional weather forecasts.
However, Farnell isn't worried about being replaced by a robot.
“It doesn't tell the whole story,” he said. “You still need experts to basically say, ‘This is what's going to happen.'
Eastbrook cautioned that there are still limitations, as well as some risks, to the use of artificial intelligence in weather forecasting. For one, it is better to forecast weather in macro scenarios rather than at the local level. And because machine learning weather models are largely untested in operational settings, such as when there is an extreme weather event, this can leave questions about how accurate they can be.
“It's harder to use data from the past to predict what these extreme weather events will look like in the future,” he said. Weather records are being broken more and more often, such as the heat records last summer, with little comparison to what has been experienced in the past.
Eastbrook also said that AI weather models can sometimes break the laws of physics in their predictions because the technology doesn't actually know anything about the physics of weather events, just as ChatGPT doesn't actually speak but predicts words. (Googlers have said that the company's Bard personal assistant can sometimes be a pathological liar.)
“You have to watch out for things like that,” Eastbrook said.
The typical hyperbole behind such Google products could be off-putting to some, he noted, especially if predictions of extreme weather events are given an unwarranted quality.
The low cost of AI weather models—GraphCast is open source, meaning anyone can use it—also opens the door to more privatization of weather forecasting, which can have its own risks. Eastbrook said that in a tech world where things move in and out, publishing an inaccurate weather forecast that looks legitimate is a “huge risk.”
“We are going to privatize weather forecasting services and it will be a closed box,” he said. “That's a big threat in this industry.”
and copy 2023 Global News, a division of Corus Entertainment Inc.
