AI’s new eyes in the sky helping Colorado get a jump as wildfires spark every few hours (2024)

With Colorado wildfires now cropping up daily between lunch and dinner, a new satellite-based fire spotting system driven by artificial intelligence is already hard at work looking for fresh trouble spots and helping weather forecasters point out likely breakouts of existing fires.

The Boulder office of the National Oceanic and Atmospheric Administration watches satellites in geostationary orbit 22,000 miles above North America. Those satellites take detailed ground photos every 30 seconds. No human reviewer could possibly keep up.

An AI software layer, however, can instantly alert NOAA and weather officials to a change in infrared wavelength sensors on an area as small as an acre. An alert goes out to local National Weather Service offices, who in turn take advantage of recently updated NOAA communication protocols to loop in state and local firefighting leaders.

“It’s the force multiplier,” said Michael Pavolonis, wildland fire program manager for NOAA in Boulder. “Humans are very good at picking out new heat signatures in these satellite images associated with emerging fires. But humans can’t digest all of that imagery coming in at that rate. It’s just a … fire hose of data.”

A human reviewer confirms the heat signatures before alerts go out, Pavolonis added. “AI tools in this space are used by human expert decision-makers. They’re not replacements for humans. They give humans the ability to make more efficient and informed decisions.”

The new AI-driven tool, Next Generation Fire System, was not necessary in early spotting of this week’s biggest fire so far, the Alexander Mountain fire. Nor did it take artificial intelligence to spot the dangerous Quarry or Stone Canyon fires up and down the heavily populated Front Range, where commuters, hikers, law enforcement, open space rangers and TV traffic helicopters can be early detectors.

Though the AI scans are still in experimental mode, they are being used in real time to look for more remote hot spots throughout the West, Pavolonis said. And on existing fires like Alexander Mountain or Stone Canyon or Quarry, AI could identify new hot spots that blow ahead of the main fire, or other conditions that firefighters and evacuation planners need to look out for.

The AI scans and communication tools perfected at NOAA’s Boulder test bed in June turn meteorologists into “scientific first responders,” said Todd Lindley, the National Weather Service science and operations officer in Norman, Oklahoma. “These tools will help us provide timely and life-saving warnings of particularly dangerous wildfires.”

NOAA’s satellites have extremely sophisticated cameras recording on multiple wavelengths, Pavolonis said — not just what is visible to the human eye in a photograph, but also various levels of infrared and other wavelengths that show a contrast in temperature between one spot and another. It’s the contrast that catches the “eye” of the AI review and flags it for humans to take another look.

The new tools were developed in cooperation with researchers from Colorado State University and University of Wisconsin, Pavolonis said.

NOAA is decidedly not selling the AI tool as an endgame, all-powerful firefighter.

“AI does spot some fires earlier than other sources. Not every time,” Pavolonis said. “It’s a piece of the puzzle. All of these technologies have to come together and be interoperable, and that gives you the best intelligence. So what we’re trying to do is maximize the value of this particular information source so that it’s complementary with other technologies.”