With a historic drought deepening across the Western United States, many residents, officials, and first responders are nervously awaiting an inevitable severe wildfire season. When it comes to wildfire, quick detection is key and even a few minutes can make the difference between homes saved and lost. Last year, we shared a story about Descartes Lab which is working on developing AI to detect wildfires as soon as they start. Now, a similar system called ALERTWildfire has been installed in Northern California and is ready for testing as the hot summer season arrives.
While Descartes uses AI to scan images from space and detect heat abnormalities, ALERTWildfire uses a series of cameras on towers placed throughout fire-prone regions. Images from these cameras are fed through an AI detection system to immediately identify a developing fire. The system’s 21 initial testing sites are all located throughout Sonoma County where the 37,000-acre Tubbs Fire in 2017 was the most destructive in California history.
Although the system uses AI to scan the tower images, the system also delivers image updates from each tower to dispatchers at the county’s fire emergency center every ten minutes to add a layer of human confirmation. So far, AI fire notifications from ALERTWildfire have preceded 911 calls from residents by as much as ten minutes.
Developed by Graham Kent, director of the Nevada Seismological Laboratory at the University of Nevada, Reno, the program is becoming more accurate over time. The AI system is still learning what fires look like and how to, for example, differentiate between smoke from a fire and steam from a geyser field. Researchers estimate that the system needs to “see” at least 70 real-world blazes to reach a full understanding of the region’s fire landscape.
Eventually, the program’s developers hope to install 850 ALERTWildfire cameras across six Western states to significantly decrease wildfire severity. They believe that with time, AI can even be trained to analyze fire-prone landscapes and burn scars to predict which areas are most likely to burn that year.