Here’s a photograph that’s all too familiar: A red cloud of fire retardant pours from the belly of a propeller plane onto a forest below. As climate change supercharges wildfires, fire crews are increasingly relying on fire retardants to create chemical breaks and contain fires. Knowing where the spray lands helps crews and scientists manage and study its effects.
A new remote sensing tool may help crews and scientists pinpoint the exact location of fire retardant by taking advantage of satellite images.
The tool “should be faster, cheaper, and better” than current methods, said Jerry Tagestad, a Pacific Northwest National Laboratory geographer who developed the technique.
Machine Learning in Action
The exact coordinates where fire retardant lands depend on the wind and topography of the ground below. The red slurry loses its color within weeks under the Sun and washes away in the rain, making it difficult for scientists to study how it has affected the landscape.
The U.S. Forest Service records a GPS location each time a plane releases fire retardant from its hatch. In some cases, fire crews fly a second plane to take photographs of the ground and trace drop locations by hand. But restrictions to air space and other constraints make follow-up flights difficult during a fire.
To develop the tool, Tagestad and his colleagues trained a machine learning algorithm to locate retardant lines in images taken by the European Space Agency’s Sentinel-2 satellite. They first sat down with satellite images and denoted areas with and without retardant. Using three machine learning classification models, the group then trained the computer to recognize those patterns in new images.
“We’re taking the human out of the loop in terms of the mapping itself,” Tagestad said.
The team tested the tool on images from seven fires in the southwestern United States that burned between 2020 and 2021. Five of the seven study sites were in scrub and shrubland—a more accessible landscape for remote sensing—and two were in conifer forests.
The three machine learning models successfully identified fire retardant lines at the seven sites—the best-performing model captured 62% of fire retardant with 99% precision. “Using this method, you may be able to report [the fire retardant location] within a week after the fire,” Tagestad said. The team published the work in the journal Remote Sensing.
Applications for Aquatic Habitats
“This work offers a novel and promising way to map fire retardant more effectively,” said conservation ecologist Karen Hodges at the University of British Columbia, who was not involved in the research.
NOAA Fisheries expert Joseph Dietrich, who studies fire retardant toxicity in Chinook salmon, called the remote-sensing technique a positive step forward. He said the tool could be designed to detect accidental fire retardant drops over water automatically by identifying breaks in fire retardant lines captured in the satellite images. Previous research by the Forest Service found that fire retardants likely affect some threatened and endangered aquatic life, and other studies have suggested that fire retardant enhances weeds and harms Chinook salmon.
Dietrich is interested in future versions of the tool estimating the quantity and concentration of fire retardant across a landscape to understand better the amount entering salmon habitats. Tagestad said the team is excited to expand the tool’s capabilities, including remotely sensing the thickness of fire retardant across the drop zone, which could be used to calculate the amount of fire retardant entering sensitive habitats.
—Jenessa Duncombe (@jrdscience), Staff Writer