As wildfires have grown in numbers and intensity throughout the western United States, it’s caused a run on new kinds of technology that can help fight them. That includes machine learning for data analysis, drones, unmanned aerial vehicles, and satellite surveillance. California alone tracked 4.2 million acres burned in 2020, with five of the six largest fires in state history occurring simultaneously. That has led to multiple tech-driven firefighting solutions being approved in the state, including predictive analysis, fire-spotting from orbit, and AI-powered equipment inspections. “AI-enabled systems are already being used to coordinate disaster relief, conduct reconnaissance, and direct recovery efforts. Detecting patterns, trends, and anomalies in supply chains and for logistical support has also become a common task for Machine Learning algorithms,” said JT Kostman, the CEO of artificial intelligence firm ProtectedBy.AI, in an interview with Lifewire. “These capabilities can be configured to stock grocery shelves or to provide relief in the wake of natural disasters.”
Eyes in the Sky
There’s a surprising problem in wildfire management that isn’t covered much. Simply put: wildfires, especially new or smaller ones started by natural phenomena, can be hard to find. If a lightning strike hits a tree in the middle of nowhere or an isolated power line falls somewhere between towns, it could be a multi-acre blaze by the time any human spots it. As such, one of the most significant roles of an AI in firefighting at this point is in detection and analysis: finding isolated fires in distant locations, tracking them, and determining what provided the initial ignition. One high-profile cause comes from electrical wires, as demonstrated by California’s Pacific Gas and Electric disasters. Ordinarily, those wires are designed so they won’t contact each other and cause high-energy arcing. However, high winds or unusual dry spells can cause the lines to swing, which creates sparks and bits of hot metal to fall off the lines, potentially igniting dry vegetation. “As a potential solution, the aerial images collected using helicopter patrols and unmanned drone flight surveys are combined with the AI-based simulation models to assess the potential for wildfire incidents under various outlier conditions,” said David Cox, head of energy and utilities consulting at Cognizant, in an interview with Lifewire. “The output of the modeling is fed to various geospatial visual dashboards to identify the high-risk profile circuit lines. This approach has helped utility organizations to prioritize grid system maintenance in areas with the highest risk profiles. Machine learning technologies are currently being deployed on top of the already-existing AI-based models to improve the accuracies of prediction.” “The very same technology that is able to accurately distinguish a dog from a cat,” said Kostman, “can be attuned to find hotspots using traditional and thermal imaging through cameras, drones, and satellites.”
How to Play With Fire
Another Berkeley project, headed by Tarek Zohdi of its Fire Research Group, uses machine learning to produce a “digital twin”—a virtual duplicate of an existing fire—which is used by data scientists as a test case. Using the digital twin, data scientists can produce a reasonable model for a fire’s future behavior, which allows for more informed logistics for the firefighters. It’s easier to plot a flight plan around or above a wildfire, for example, if you have a good idea of where the wildfire’s going. Similar projects are at work in the same department for prevention effects and biosphere modeling, such as figuring out what days would be best to carry out “prescribed burns,” a deliberate fire started to manage and protect a natural environment. The most metal anti-wildfire tech in the field right now, however, is the use of drones for bombing runs. In previous decades, land managers would carry out their own prescribed burns from the air by dropping potassium-glycol charges—known as “dragon eggs”—via helicopter. Now, drones can do the same thing, cheaper and with greater precision, using the same dragon eggs to help create barriers against active wildfires by carefully depriving those fires of fuel that they could use to expand. “There’s a regrettable tendency to wait until disasters have occurred before developing capabilities to combat them,” said Kostman. “Given the existential threats humanity now finds itself having to contend with—climate change, global pandemics, unprecedented cyber threats, economic apartheid, political instability, and the rampant rise of authoritarianism—the time to evolve AI-enabled systems capable of keeping us safe is not tomorrow. It was yesterday.”
