If you zoom in on a new map of California, you’ll start to see that the fields of green that represent the forest are actually made up of individual green points, and each point represents a real, individual tree. The tool, called the California Forest Observatory, uses AI and satellite images to create an ultradetailed view of the state’s forests—aiding work to prevent the type of catastrophic megafires that the state is experiencing now.
Scientists at Salo Sciences, a startup that works on technology for natural climate solutions, began creating the tool after interviewing dozens of experts in California about the state’s challenges with wildfires: They need more detailed, up-to-date information about the forests so they can better predict how fast and in what direction fires will spread, and remove the most hazardous fuels. Even the rough satellite maps that exist now are often three years out of date, making it hard for agencies to plan their work.
The new tool will be updated annually after the fire season ends, if not more often. Firefighters can use the tool to predict how current fires may spread as they’re burning. But just as critically, the state can also use the map to plan forest management to prevent future megafires. “What we really found was California more than anything has a vegetation and fuel load problem,” says David Marvin, cofounder and CEO of Salo Sciences. “This has occurred because, for the last century, we’ve been suppressing wildfire, and we’ve gotten really good at doing so. CalFire, the state fire agency, puts out something like 96% of fires, and we have thousands of them every year.”
The problem is that California’s forests are adapted for fire. Historically, low-intensity fires commonly burned through smaller vegetation, leaving forests with wide spaces between larger, fire-resistant trees. When firefighters quickly extinguish fires, the amount of vegetation continues to grow, meaning that when another fire happens later, there’s far more to burn. As climate change makes the state drier and hotter, the problem has intensified. Wildfires can quickly explode. “We need large-scale restoration efforts to pull those small fuels out of the forest, so that when a fire does occur, it stays low to the ground,” says Marvin. “It only burns the small trees and doesn’t ladder up into the canopy and become catastrophic.”
“If these fires didn’t have continuously dense fuels to burn, it would be much more difficult for them to get out of control,” Marvin adds. “So this is what we first saw was a major problem. What we wanted to tackle was, how do we get at identifying the areas to begin treating and removing these overgrown forests?
Using satellites to map forests is not new, but the new tool does it differently. First, it pulls in detailed data from Planet, a satellite company founded by NASA scientists that controls more than 100 satellites that spin in a line around the Earth, collecting images at a resolution an order of magnitude higher than older satellites. “Each one is about the size of a loaf of bread, with a very specialized remote sensing camera, and they collectively act as a line scanner for the planet,” says Andrew Zolli, vice president of global impact initiatives at Planet. “So as the earth turns sideways underneath, they image the entire terrestrial surface of the earth every day in high resolution, which for us is about three meters per pixel. So that’s not enough to read your newspaper or to spy on you, but it’s enough to see every tree and every field.”
Salo Sciences also turned to lidar, a laser technology, for data about trees that can be gathered from forests as planes equipped with the tech fly overhead. Then they fed both sets of images into a deep learning algorithm so that the AI could begin to recognize details such as tree height—typically measured with lidar—from satellite photos. “This does not exist anywhere else right now using satellite imagery,” says Marvin. “What you would traditionally have to do to get this type of detail would be hire a company to fly an airplane with one of those laser sensors on the airplane, and they would go and do that over a small area. It costs a lot of money and takes a lot of time to process. And you can’t do that on a regular basis.”
The team hopes to provide similar maps for other areas. “When we set out to build this, we built it in such a way where it was agnostic to the geographic location,” he says. “California we really view as the test case, but we are already beginning to talk with some organizations on trying to expand it out from California.”