Could crowdsourcing hold the key to early wildfire detection?
USC researchers develop low-cost wildfire detection system to accurately detect wildfires minutes—even seconds—after they ignite
University of Southern California
The 2023 blaze in Lahaina, Hawaii, which claimed more than 100 lives and burned 6,500 acres of land across Maui, is a tragic example of how rapid wildfire spread can make effective response efforts impossible, resulting in the loss of life and property.
What if technology could help people detect wildfires earlier? The solution could already be in your pocket: a mobile phone.
USC computer science researchers have developed a new crowdsourcing system that dramatically slashes wildfire mapping time from hours to seconds using a network of low-cost mobile phones mounted on properties in high fire threat areas. In computer simulations, the system, FireLoc, detected blazes igniting up to 3,000 feet away and successfully mapped wilderness fires to within 180 feet of their origin.
Detecting wildfires within seconds of ignition
Presented at ACM SenSys on Nov. 5, the paper, titled “FireLoc: Low-latency Multi-modal Wildfire Geolocation,” serves as a proof of concept, according to the researchers. But how would it function in the real world?
“It’s a stepping stone towards broader wildfire mitigation efforts in the future.” Xiao Fu
For the user, it’s simple. Residents and businesses near high-risk areas would install an affordable, weatherproof mobile phone in their backyard or on their building, connect it to a power source, and point the camera toward nearby trees and brush.
Behind the scenes, complex multi-modal analysis and computer vision models process the data, gathered from the phone’s basic cameras and sensors, rapidly detect wildfires, often within minutes—even seconds—of ignition.
The system prioritizes privacy by focusing on areas with minimal human activity and primarily captures images of vegetation and wilderness. Adapted object localization techniques also ensure the system zeroes in on fire risks without inadvertently capturing images of people or homes.
Sustainable co-existence with extreme climate
For people who live and work on the periphery of open spaces that traditionally teem with parched fuel sources such as grass, shrubs, and timber, such a rapid response could mean the difference between life and death—or having a home or losing it.
In Southern California, the technology could serve as a model for how to best protect people and homes in wildland-urban interface (WUI) locations such as the Hollywood Hills, the Santa Monica Mountains, and the San Gabriel Valley. What’s more, the entire set-up would cost less than $100, said lead author Xiao Fu, a computer science PhD student.
“FireLoc envisions a future that we will provide a more effective wildfire response, providing better support in the WUI, and more sustainable co-existence with an extreme climate,” Fu said. “It’s a stepping stone towards broader wildfire mitigation efforts in the future.”
The paper is co-authored by Barath Raghavan, Fu’s advisor and an assistant professor of computer science, Peter Bereel, a professor of electrical and computer engineering, and students Yue Hu and Prashanth Sutrave.
Robust testing of wildfire environments
Traditional wildfire detection methods—such as lookouts, satellites, and drones—each have their drawbacks, including high costs, inconvenience, slow response times, and limited battery life. Consequently, firefighters often depend on human observation to spot new fires, which makes it difficult to pinpoint a fire’s exact location.
“This is also very overwhelming for the fire departments, especially in rapidly developing fires like the one in Paradise,” said Fu, referring to the deadly 2018 Camp Fire in Northern California, which killed 85 people.
The team evaluated the effectiveness of their mapping tool by running a simulator based on data from the 2019 Getty Fire, which burned 745 acres in Los Angeles. By adopting a real-world 3D model of the terrain and simulating realistic wildfire scenarios, they assessed the system’s overall performance, including its ability to accurately localize wildfires and its scalability.
Each camera was positioned to mimic the typical height of a residential second story or rooftop, approximately 30 feet above ground level. The results were clear: By adopting FireLoc, the researchers successfully detected more than 40% of wildfires in the target area with only four cameras.
“The simulator allows us to have robust testing of wildfire environments. We’re able to control the scalability – like increasing the number of cameras—is accuracy going to improve? Is coverage going to improve?” Fu said.
Reframing the problem and coming up with a solution
While the location information from the cameras is incredibly important, crowdsourcing plays an equally pivotal role. Requiring only electricity, an Internet connection and the phone (in a weatherproof holder), the software would automatically take pictures every, say, 30 seconds.
“Given several locations, the system is able to optimize where would be the best location to set up additional cameras for wildfire monitoring," said Fu.
When several cameras detect possible smoke or a fire, they would transmit that information to a cloud server which stitches the multiple images together using digital elevation models, computer visions techniques, and other sophisticated computing tools. This is a complex and critical process, Raghavan said, but you don’t need high-quality images. An algorithm would determine where the cameras should be placed to optimize coverage.
“We’re combining all the information from the images in a way that solves the problem,” Raghavan said. “That’s the solution part of our paper. But we also reframed the problem – that is, how can we map fires as quickly as possible? This paper does both: reframing the problem and coming up with a solution.”
As far as the researchers know, this is the first smart, low-cost crowdsourcing system specifically designed for wildfire detection.
Testing the system in real-world conditions would require community members to mount smartphones on their properties to act as wildfire sensors—the team plans future participatory studies to understand how people would engage with the technology. If deployed, would the researchers themselves join in?
For Fu, an outdoor enthusiast with a deep love for nature, it’s a no-brainer.
“My whole life, I have worked for green unions and environmental events,” said Fu, who grew up on her family’s fruit farm in the tropical region of Hainan, China. “Even when I can’t get outside because I’m working, I can still look at the photos of the trees and the vegetation, and that makes me happy. I hope this technology will help to protect our natural landscapes in the face of extreme climate change.”
Subject of Research
Not applicable
Article Title
FireLoc: Low-latency Multi-modal Wildfire Geolocation