Nvidia’s research division recently published an AI weather model that scientists expect will improve weather forecasting at a regional level for short-term weather events. It’s a step forward in the evolving landscape of atmospheric science, where machine learning is playing a greater role.
While scientists around the world are still learning the complex physics of clouds and storms, AI is increasingly helping us fill in some of the blanks.
“AI is completely changing how we simulate the atmosphere,” said Nvidia researcher Mike Pritchard.
Solving physics equations about future weather events is usually done by huge, costly supercomputers, he said. “Which limits how high resolution forecasts can be and how many we can make. But now AI is proving to do as good, if not better, a better job than physics equations at similar predictions.”
And the new research released this week by Nvidia presents a new model for quickly predicting local weather events, like thunderstorms, down to the kilometer.
“It certainly does help with the timeliness of information,” said Maria Molina with the University of Maryland.
The machine learning model can compute more efficiently so you can run predictions more often, she said.
Compared to supercomputers used by richer countries, once machine learning models are trained “they are much cheaper to run, and that is what opens doors for access,” Molina added.
Access for countries with fewer resources to predict severe weather events like derechos and hurricanes.
This is a step toward “a very, very accurate weather forecast, a kilometer scale — weather forecast across the world, which will be game changing in terms of saving lives,” said technologist Peetak Mitra. “Like, you know, predicting drought sooner, predicting heat waves sooner, predicting cold snaps sooner.”
That could make better use of disaster response resources.
“Studies like this, or techniques like this, would open up a lot of opportunities where the relief could be sent to the right place at the right time — beforehand, instead of just spreading it too far, too thin,” said Mitra.
And down the road, he said we could see storm warning systems improve in different parts of the world.
