Weather 2.0: How AI and Balloons Are Shaping the Future of Forecasting

View guidelines for use of these broadcast-ready news packages.

Click here for the full media package

BY BEN TRACY, SENIOR CLIMATE CORRESPONDENT ON ASSIGNMENT FOR CLIMATE CENTRAL

We all check the weather — some of us, religiously. But even with dozens of apps and real-time radar maps at our fingertips, forecasts can still feel frustratingly uncertain.

Now, as extreme weather events grow more frequent and intense, a new crop of private companies is stepping in — armed with artificial intelligence (AI) and high-tech balloons — aiming to make weather predictions faster, smarter, and more accurate than ever before.

Forecasting Takes Flight in Oklahoma

In a quiet backyard outside Oklahoma City, Sean Rieb is preparing for an unusual part-time job. The task? Launching a weather balloon.

“Yes I did yes,” Rieb says with a smile when asked if he warned his neighbors first.

The startup behind this backyard balloon is Windborne Systems, and Rieb is now one of their trained “launch operators.” As the balloon takes off, his excitement is clear.  “Oh it was great, yeah!” he laughs. “No, that was a lotta fun!”

These balloons are not your average weather tools. They’re part of a global effort by Windborne to revolutionize how meteorologists gather critical data on wind, temperature, pressure, and humidity.

A Global Network of Citizen Scientists

Windborne’s Jehan Godrej trains citizen scientists like Rieb around the world to help launch these lightweight but powerful data collectors.

“We have one that’s been flying for 77 and a half days,” says Godrej. “But our average flight time right now is around 12 days.”

That’s a major improvement over traditional weather balloons used by the National Weather Service, which typically last just a few hours. Most weather data comes from the federal government, but agencies such as the National Oceanic and Atmospheric Administration, which includes the National Weather Service, are facing significant budget cuts from the Trump Administration. 

Each Windborne balloon costs about $1,000 and can be remotely steered as it circles the globe—collecting 30 times more data than government balloons and covering the 85% of Earth, mostly oceans, that are rarely monitored.

“You really need this level of monitoring to do accurate weather forecasting,” says John Dean, Windborne’s 28-year-old CEO.

Dean helped launch the company in 2019 with fellow Stanford classmates. Today, from their Silicon Valley headquarters, the team tracks balloon launches around the world—from Korea to Guam to… his parents’ house in New York.

“That’s my parents' backyard,” Dean confirms with a grin.

A High-Tech Weather Revolution

Currently, Windborne has around 100 balloons in the air. Their ambitious goal? 10,000.

With climate change driving a surge in billion dollar weather disasters — events that now strike the U.S. every 19 days, compared to every 82 days in the 1980s, according to Climate Central — better forecasting isn’t just nice to have. It’s essential.

Windborne’s data is already being used by the federal government, which says it has helped improve forecasts. The company is also among several working on AI-powered models that can generate high-resolution forecasts every six hours. These could extend warning times for hail and tornadoes from just 15 minutes to nearly an hour—and predict hurricane landfalls with more precision, days in advance.

“That would be fantastic because you could do more preparation,” says Amy McGovern, a meteorology professor at the University of Oklahoma. “You’re not gonna move the hurricane but you could prepare people better.”

The Promise—and Limits—of AI

AI thrives on data, and these models are trained on decades of publicly funded weather information.

“AI is really good at pattern recognition,” McGovern explains. “So if you can give it tons of data it can find patterns that humans aren’t going to be able to do and it can do it really quickly.”

But it’s not foolproof — especially when weather turns extreme.

“It doesn’t know when it is wrong and that’s something the AI is not very good at yet,” she says.
“Humans know when we’re uncertain. We’re like ‘no I’m not sure about this I need to talk to some more people.’ The AI is like ‘no let’s go. I got you a forecast let’s go.’”

That’s where companies like Windborne see opportunity: using ever more real-time data to keep improving accuracy.

“The more data we have the better the forecasts are going to get,” says Godrej.