As wildfire season becomes more threatening, experts are turning to AI

With climate change driving hotter, drier summers across the West, the intensity of recent fire seasons is outpacing workers’ ability to track and analyze fires with the traditional methods.

“There simply aren’t enough boots on the ground, or eyes in pairs of binoculars to cover the base and the extent of what we’ve been dealing with and that’s been true across the western U.S. and across the world,” said Sarvesh Garimella, the chief scientist and chief technical officer at weather app MyRadar.

But government agencies and private sector companies like Garimella’s are adapting artificial intelligence technologies in their wildfire monitoring and fighting strategies.

Increase in number and reach of wildfires

The Environmental Protection Agency’s data shows that wildfires have become more common and spread further over the last few decades.

The 1990s was a “period of transition” for climate cycles that tend to change every few decades, and this shift may have contributed to “warmer, drier conditions,” the EPA reports, which make wildfires easier to spread in the Western US. Between 1981 and 2021, the amount of land involved in wildfires and suffered severe damage has risen from 5% to 22%.

Changing climate conditions are just one of the contributing factors, said Michael Pavolonis, a physical scientist at the National Oceanic and Atmospheric Administration’s (NOAA) Center for Satellite Applications and Research. He added that land management practices, a century of fire suppression that left forests overgrown and combustible, and human activities have greatly changed the fire landscape in the U.S.

The Western Fire Chiefs Association says that nearly 90% of all wildfires are caused by human activities, like discarding a cigarette, leaving campfires unattended or through an equipment malfunction.

Fire activity for 2024 is already above average from the last decade, the New York Times reported this week.

Oregon firefighters worked on what was at the time the biggest fire in the country, in late July which spread across nearly 270,000 acres and threatened evacuation for thousands of residents. A week after the initial response from EMS, residents of the state were still dealing with smokey, unhealthy air conditions. And late last week, Coloradans faced evacuation orders for a blaze along the state’s Front Range, which killed one person.

Zach Tolby, director and lead scientist at NOAA’s fire weather testbed said living with wildfires is “nerve wracking.” The now Colorado-based weather expert said while living in Reno, Nevada, there were a few years where the wildfires kept people from being outside nearly all summer.

Incoming threat of fire or the residual smoke often upends social gatherings, interrupts work and school schedules and can keep people from feeling safe in their communities.

“Once you do have wildfires, there’s a lot of, you know, kind of PTSD from the effects of being around them,” Tolby said.

AI in fire detection

Weather science is all about data, climate experts say, but viewing, logging and processing that data is an overwhelming task. It’s where the AI comes in: the bread and butter of AI models lies in the processing and sorting of mass amounts of information.

In the case of wildfire prevention, if you can automate the sorting of fire information, more humans are freed up to make decisions, call for resources and be deployed on a scene to stop a wildfire from spreading.

That’s the intention behind NOAA’s Next Generation Fire System, which uses an AI model to identify fires from the department’s geostationary satellites. The program, and a location to test fire prevention systems, was developed via funding from the Bipartisan Infrastructure Law passed by the Biden administration.

NOAA’s Geostationary Operational Environmental Satellites (GOES) are the “workhorses” of weather monitoring of the entire United States from space, Pavolonis told States Newsroom. The satellites are able to record an image as often as every 30 seconds, which generates an overwhelming amount of information, Pavalonis said.

“So humans stare at all of these images and look for the fires themselves,” he said. “They do that and they do catch some of them. But it’s impossible to stare at every image.”

The AI model, which was built through academic partnerships, automatically scans those images and uses heat detection to look for features of an emerging fire. It then pipes the information to a situational dashboard that allows those monitoring it — whether it’s the National Weather Service or a land management partner — to determine that a fire could use their attention.

The Next Gen Fire System has been in development for a few years, and NOAA piloted the technology at the Colorado-based fire testbed in June. Tolby, director of the center, said they tested the technology with a few existing systems, like the National Weather Service, to see if they could identify fires and weather phenomena with enough accuracy to be used in real-world situations to issue public alerts.

Tolby and Pavalonis stressed that humans are still at the helm of decision-making. The pilot at the fire testbed provided a realistic operational environment that allowed NOAA to understand how new capabilities would work when the agencies have to make real decisions.

“Science and technology are essential, but they’re not sufficient,” Pavalonis said. “You need to also work closely with decision makers throughout the development and testing process.”

Some states that experience a lot of wildfires, like California, have camera systems to monitor the spread of fires. But more rural and lower-population areas can be vulnerable to wildfires going unchecked. The satellite systems, developing AI tools and cross-department collaborations are valuable tools for those communities, Tolby said.

“If you can get to a fire when it’s … only a couple of trees, you’ve got a much better chance of putting it out than you do when it’s 10 acres or 100.”

Modeling future fires

A project from the University of Southern California aims to use AI to improve the effort to fight fires once they start, predicting their possible spread and behavior. The research team at the university, headed by Professor Assad Oberai, uses a physics-informed approach to predict the spread of wildfires.

The team uses a generative AI model called conditional Wasserstein Generative Adversarial Network (cWGAN) and trained it, informed by data from past wildfires, satellite images and from data assimilation, or a method of combining different data sources, to predict future fire spread.

They’ve spent months testing their algorithm with simulated wildfire data that was built upon traits of real fires that occurred from 2020 to 2022. They then compared their model’s predictions against how those fires actually spread to understand the accuracy of its prediction abilities.

The generative AI works similarly to a chatbot like ChatGPT, Oberai said. ChatGPT gives text responses based on a prompt provided by the user, and the USC model will show predictive imaging about wildfire spread.

“You can think of the satellite measurement as a prompt, right,” Oberai said. “And you can think of sort of the image of the spread of the wildfire as the response to that prompt from this generative algorithm.”

The USC team’s algorithm differs from other fire prediction tools in that it generates several predictions, and gives a user the likelihood of each outcome, similar to the graphs used to predict the possible course of a hurricane. Their model is also relying on data from these simulated wildfires to help piece together what will likely happen.

Bryan Shaddy, a Ph.D. student and researcher on the team, said possible next steps for the team include folding in more variables. Right now, the algorithm focuses on predicting the likelihood of fire progression, but they could train the cWGAN on others, like how terrain might affect spread. The team will continue training and adding variables, and could foresee the technology being adapted into existing fire prediction tools.

Private sector technologies

While many industries are just considering how AI can be incorporated into their work, climate scientists have been early adopters to supercomputing and big data processing in general, MyRadar’s Garimella told States Newsroom.

The weather monitoring app offers high-definition radar, NOAA weather alerts, temperatures, forecasts, flight tracking and disaster warnings. Some users depend on it just for their daily weather updates, while others have business reasons, like for wedding planners or sports coaches planning an event, Garimella said.

Activities are heavily affected by environmental events, he said, and climate change has made weather harder to pin down and predict over the last several decades.

The company’s Orbital Wildfire Resilience solution was just chosen to advance in the XPRIZE wildfire competition, which is seeking to revolutionize wildfire technologies.

MyRadar will launch its AI system via four satellites in February. The technology is “edge deployed,” meaning the AI processing happens in the satellites themselves rather than in computers back on the ground, and it takes less power to transmit those messages back down to Earth. It allows for smaller satellites and quicker message times.

The obvious attraction is more information for their app users, Garimella said, but eventually, they can feed the data into data streams that government agencies, like NOAA, would use to monitor and fight fires.

AI technologies are also being used for on-the-ground response to fires. Autonomous machines, often called drones, are being tested across the country as a resource for emergency responders.

One example is public safety technology company BRINC’s Responder drone. Right now, it’s being used for structure fires, but the company’s vice president of strategy and growth, Andrew Cote, said he believes the future of AI and machine learning will allow for more predictive analytic capabilities in firefighting.

The Responder drone is used by emergency response teams to cut significantly down on response times to a scene, and to monitor via cameras how a fire scene is developing. It can help firefighters identify safer places of entry into a structure and can drop resources like survival kits or flotation devices down onto a scene.

Cote said this method not only gets resources to a scene faster, but also helps EMS teams deploy them in a more efficient and cost-effective way.

“We’re long overdue for new types of techniques, at least to give them a try,” Cote said. “And they’re not that expensive, as opposed to new, new water tankers, new aircrafts and all the other things that we try and currently use.”

It will be several months before NOAA’s and MyRadar’s AI models are in regular use, but we’ll likely continue to see more AI technologies being adapted in weather prediction and fire monitoring in the coming years.

Firefighting methods haven’t changed much in the last 50 years or so, the climate scientists said. So new technologies that can streamline systems, provide early intervention and potentially provide more information to government agencies and everyday people are a win, they said.

“This is another tool in the toolkit,” Garimella said. “But it’s also one of the most powerful ones that have ever been invented.”