The National Center for Atmospheric Research in Boulder is developing artificial intelligence tools that enable forecasters to predict severe weather threats more accurately and farther in advance.
While traditional computer models can predict tornadoes, and hail and wind events two to three days in advance, NCAR’s AI models can anticipate those events up to eight days in advance with higher accuracy.
“These tools are allowing us to look into the future further, to get an indication of where these events will happen,” NCAR scientist Ryan Sobash said.
Until 2019 or 2020, Sobash said, all forecasts were created by relying on big, complex and expensive computer models that take a lot of computing power. Now, AI has accelerated all of that. The AI tools are trained on how to analyze the atmosphere, and once they’re trained, they can create forecasts in minutes instead of hours and have the capacity to run a lot more models. And, Sobash has found, the AI tools provide better predictions.
“Some of these models are actually more accurate than the older computationally intensive techniques,” Sobash said.
NCAR is disseminating these forecasts online and to other labs and forecasters. The experimental forecasts will be evaluated in the coming weeks at NOAA’s Hazardous Weather Testbed in Oklahoma.
The models are trained to look at severe weather across the U.S., including in Colorado. Sobash anticipates these models will allow Boulder and Colorado to get a better picture of hail, wind and thunderstorms before they happen, especially entering late May and June, a particularly active time for that type of weather.
“Those (weather events) are affecting people in increased ways for a variety of reasons, and so we’re trying to develop these tools to give people additional lead time by improving the accuracy of the forecast,” Sobash said.
In the future, Sobash said, he hopes to adapt these AI tools to predict other types of weather events, including flash flooding and winter precipitation, which have big impacts in Colorado. He and his team are also striving to continue to see improvements in accuracy and lead time as their work continues, hoping to achieve accurate forecasts more than a week in advance.
There are funding limitations given the political climate, Sobash said, but the team has a solid path forward to make progress in this area. President Donald Trump made threats to dismantle NCAR in December, and by the National Science Foundation is aiming to restructure it and remove some of its capabilities.
Anyone can view the AI-powered forecasts by visiting .
.
