AI in weather forecasting: How the technology can improve predictions in life-threatening situations
Researchers at Colorado State University say that AI could eventually be a game-changer when tracking weather conditions around a fast-moving wildfire
Researchers at Colorado State University say that AI could eventually be a game-changer when tracking weather conditions around a fast-moving wildfire
Researchers at Colorado State University say that AI could eventually be a game-changer when tracking weather conditions around a fast-moving wildfire
Artificial intelligence could someday help meteorologists make more accurate and precise weather forecasts.
The trick is figuring out when AI is most useful and when it can lead forecasters astray.
Researchers with Colorado State University's Cooperative Institute for Research in the Atmosphere (CIRA) are working on figuring out some best practices by connecting AI with more traditional forecast models.
“I’m using artificial intelligence to connect satellite observations with these models." said Kyle Hilburn, "That’s helping us be able to make better forecasts of severe weather.”
Meteorologists have used computer forecast models or "numerical weather prediction" as a tool for making predictions for decades.
These traditional models take current weather observations, run them through a complex set of equations, and produce answers that depict possible future weather conditions.
Artificial Intelligence software makes predictions based on huge data sets gathered from past weather events. Essentially, the technology learns from what has already happened — something traditional models cannot do.
“Since AI is learning from the data, it can avoid some of the errors that you see in weather models. For example, we’re seeing that AI is making smaller errors in forecasting the speed of movement of low-pressure systems," said Hilburn.
While AI can avoid some known biases with traditional forecast models, it can also create new ones.
Models run with AI can only predict scenarios that are represented in past data. That means they would not be able to forecast record temperatures or unusually strong storms, even if those are the most likely scenarios.
“Because it’s learning from data, as the climate changes, AI models will need to be retrained to make accurate predictions in new climates,” adds Hilburn.
That retraining process takes a lot of time and computing power, resources that most groups don't have.
AI also cannot currently produce forecasts in a way that is practical for a meteorologist.
Dr. Imme Ebert-Uphoff, also with CIRA, explained that an AI model forecasts pixel by pixel, but a meteorologist is more interested in knowing about entire features like storm systems, wind fields, or air masses.
Forecasters themselves would also need to fully understand how the AI model works in order to figure out what it can predict well and what it is likely missing.
“I think the next step in developing forecasts with AI is to bring in forecasters and social scientists to really evaluate them because right now they’re mainly being evaluated by computer scientists,” said Ebert-Uphoff.
AI modeling could also one day be used in life-threatening scenarios like rapidly spreading wildfires. This is currently part of Hilburn's research at CIRA.
"AI offers the promise that we can keep up with [wildfires] and make high-resolution forecasts of that fire spread. It’s going to take a while until we trust it enough to use it in those high-risk situations, but I think it could be a very powerful tool to help us better respond to wildfires.”
Both Hilburn and Ebert-Uphoff believe that the future of more accurate forecasts will involve a combination of traditional weather models and AI rather than a complete replacement.
National Stories Below
Hunter Biden pleads not guilty after agreement with prosecutors falls through
Mitch McConnell says he's 'fine' after freezing mid-news conference