Skip to Content
Artificial intelligence

How AI watches wildfires

October 26, 2022
Stephanie Arnett/MITTR

A look at how AI and other tech is being used to help predict, detect, and pinpoint the location of wildfires in the first of a two-part series.

We meet: 

  • Dustin Tetrault, Deputy Fire Chief, Big Sky Fire Department
  • Sankar Narayanan, Chief Practice Officer, Fractal Analytics

Sounds from: 

  • Early Animated Smokey Bear Commercial, via YouTube. 
  • Smokey Bear PSA: Please only you can prevent forest fires (1965 - faded color), via YouTube. 
  • Grilling | Wildfire Prevention | Ad Council, via YouTube. 
  • Australia's Wildlife Emergency | 101 East, via YouTube.
  • At Least Two Killed In California Wildfire, CBS Mornings, via YouTube.
  • CAL FIRE Home Hardening 30 Sec PSA, via YouTube.

Credits:

This episode was reported and produced by Jennifer Strong, Anthony Green and Emma Cillekens. It was edited by Mat Honan and contains original music from Garret Lang and Jacob Gorski. Our mix engineer is Garret Lang and our artwork is by Stephanie Arnett.

Full transcript:

[TR ID]

Smokey Bear: “You have so many reasons to protect your forests.” 

Jennifer: Smokey Bear has been part of American culture since the 1940s… 

Smokey Bear: “Remember, only you can prevent forest fires.”  

Jennifer: For generations, he’s been warning via cartoons, radio shows and comic strips about the role people play in starting fires. 

Smokey Bear: “Every tree is a family tree… with families much like yours. So please be careful with fire.

Jennifer: These days, the character has a new lease on life… as a digital voice assistant… carrying those same messages… and educating the public. 

Grillmaster: “SO.. what should I do with all of these coals?”

Voice Assistant: “Don’t just toss them out. Put them in a metal container because those embers can start a wildfire.”

Jennifer: But there’s a reason for that phrase “fight fire with fire”...  Fire suppression isn’t necessarily a good thing for forests… it can raise the risk of much larger fires… and we’ve been doing it for decades. The world also faces growing threats from climate change… with a dangerous mix of extreme heat and drought.

News Anchor 1: “For more than 6 months…Australia has burned.”

News Anchor 2: “Now to Northern California and a devastating development in a fight to control the state’s largest wildfire this year.”

News Anchor 3: “Seattle’s air was choked with smoke for a second day on Thursday. Wildfires have left Oregon and Washington State with what official data shows as the worst air quality in the nation.” 

Jennifer: The United Nations predicts a global increase of extreme fire by perhaps 50 per cent by the end of the century… and those who live in the path will need to do much more than put out their campfires.

Announcer: “Wildfire can destroy your home and property. Prepare your home with fire ignition resistant materials before wildfire strikes.”

Jennifer: Researchers are working on algorithms to improve forecasting, more quickly detect fires when they start, and pinpoint their locations before they spread.

[beat]

Jennifer: I’m Jennifer Strong and in this first of a two-part series, we head to the woods for a better understanding of the challenges firefighters face… and how technology might help. 

[Show ID]

[NAT SOT: Truck door, leaves, wind.]

Dustin Tetrault: Wanna walk in here a little bit. 

Jennifer: This is Big Sky, Montana… next to Yellowstone National Park… and I’m here to learn firsthand how firefighters are starting to use AI. There’s a lot of wind…and a whole lot of tree branches and other things on the ground…which to a fire means fuel. 

Dustin Tetrault: When you look at just the amount of fuel that's in here… I mean, it's pretty staggering of how much dead and down material on the ground. And it’s like this throughout our entire fire district.

Jennifer: I’m taking a walk with the town’s Deputy Fire Chief, Dustin Tetrault. He enjoys experimenting with tech and figuring out how it can be useful to his team. Among the tools they’re experimenting with are high definition cameras that can detect smoke and pinpoint its location.

Dustin Tetrault: When you look at the density of the forest that we have to work in, a lot of times don't know where a fire's at. So before the camera, we would get a call and say, Hey, somebody sees smoke from the highway. And so that, that was what we would get. And our engines would just be driving around aimlessly trying to figure out where this smoke's at. So that smoke could be in something like this in the middle of nowhere. Well, you know, a small fire that started in here and we put an engine down here trying to figure out where it's at. Next thing you know, it gets a little bit of wind on it. Our guys would essentially lose their escape route getting outta here. And so having that situational awareness from that camera to look at it from above and be like, Hey, this thing's still just smoking. You guys can keep going and looking for it. Or Hey, you know, this thing's running towards the road, need to get you guys outta there right now. 

Jennifer: Real time data is hard to come by … and we’ll dig into that a bit later in the series… but much of the information fire districts like this one have access to is refreshed daily at best. 

Dustin Tetrault: So for me, I can, I can pull it up and I can get just a static snapshot real quick if I want to kind of see what's going on. Or I can grab that camera and I can manipulate it and spin it around and take a 360 degree, look at kind of what's going on in the surrounding. And then I can also, you know, go up, down and pan around as much as I want to on it. And so I can zoom very, very far in detail, um, into certain areas to, to try to figure out what's going on. And then there's also the piece of, you know, I can mark a fire on there as well. So if I know, hey, there's a fire there and you know, for some reason the AI's not picking it up, I can mark a fire or, you know, vice versa where the AI picks up a fire and then it goes through, um, their dispatching center and they're like, No, I don't think that's a fire. And then we can say, No, that is a fire and we can mark it too. 

Jennifer: Fire is a natural part of a forest ecosystem… it clears away fuel from the forest floor and recycles nutrients in the soil… it’s why forest managers sometimes prescribe fire—or a fire they manage. 

Dustin Tetrault: This place up here, there's little small portions of it that in the last hundred something years have been logged, but for the most part, I mean, it's just hasn't been, nothing's been touched up here and, you know, any sort of fire's been aggressively suppressed up here, so there hasn't been any fire on the landscape and it's just led to this, you know, accumulation of fuel. So…

Jennifer: Yeah this really brings it home, alright.

Dustin Tetrault: Yeah. And you'll come back next year and there'll be a, you know, $4 million house sitting in this little spot right here.  

Jennifer: Big Sky is in the midst of a housing boom. Home prices have risen nearly 80% in the past three years… and more people are living here year round, instead of just during ski season, when the fire risk is lower. All of this has added urgency to the fire problem… and helped Tetrault more quickly adopt new tools meant to protect people and their properties… including those cameras.

Dustin Tetrault: As it’s there and it's seeing all these things like, you know, dust from a heavy piece of machinery, you know, on a gravel lot or all these different things. It's learning from those and then it's making the AI that much better. So like this year, I don't think we've gotten any false alarms this year off of it. I think everything that we've seen off of it have been legitimate or they've been, you know, somebody's actually burning something and it's catching it or picking up fires from a long ways away. It's picking up fires from one valley over, which is like 20 miles as a bird flies away. 

Jennifer: Wow. 

Dustin Tetrault: So it's picking those fires up too, which is just crazy. 

Jennifer: It’s wildfire season and the constant din of helicopters draws my attention to the sky. He lets me know though it’s actually construction work we’re hearing. 

Dustin Tetrault: Pouring concrete with helicopters up there.

Jennifer: [Laughs.]

Dustin Tetrault: So they've got a big bucket, you know, like a water bucket you'd see like on a wildland firefight but it's concrete. And so they're going down and they're getting concrete from the batch plant and they're flying it up and they're dumping it into the forms up top. That's pretty crazy. 

Jennifer: That is pretty crazy. Wow. Yeah. 

Dustin Tetrault: And, um, yeah, the, the problem just keeps getting worse and worse. And it's left us, you know, as, as fire organizations trying to figure out ways to maybe get the upper hand and, and tools like wildfire detection cameras and technology that's out there is amazing. It's just you need people in positions like mine who can figure out a way to adapt that technology to make it useful for our jobs and to make our jobs safer and, and to make it more efficient.

[sounds of walking back to truck, starting, road noise]

Dustin Tetrault: You know, if you get a big fire up here, you get all these people that are up here in VRBOs and who, you know, maybe got a taxi or an Uber, or they have these mountain shuttles that bring people up here. But that adds a whole new complexity to it cuz now they don't have an avenue to get out. They don't have a vehicle to get out and they, half of 'em don't know how to get out or where to go. And so you're sitting there trying to, you know, figure out one who's here, how many people are here, you know, and then two, like, do they have transportation? How do we get 'em out of there? You know, where can we put 'em? And things like that. And so that's been a big effort. We gotta grant this last year and we've been working hard on some evacuation modeling and stuff like that to be able to make some better decisions. 

Jennifer: It’s a problem with many moving parts… and alot more complexity than I think most of us realize. Many decades of fire suppression and other factors have led to a build up of fuel.

Population, development and tourism are all booming and adding new complexities. Plus, construction is happening deeper into the forest… sometimes where roads don’t support fire services, like tanker trucks… and cell service is spotty at best.

Dustin Tetrault: But we've pushed really hard, like on the developers up here for non combustible construction. So you'll see like most of these houses will have some masonry block, stuff like that all the way down to the ground. We've done away with bark mulch next to the house. So there shouldn't be any bark mulch unless it's an older house that maybe grandfathered in. Boxing the eves. So the underside of the soffits there and they’re all boxed in so there's no holes or anything there that embers can get into.

Jennifer: Assessing the potential fire risk of a property is an important but very time consuming task. As part of this reporting, I met a fire chief who told me it would take his team a hundred years to complete this process for his community… and the thing no one wants to talk about… is that high risk areas face a real possibility of being erased within decades.

Dustin Tetrault: There's only two insurers left that are, that are writing insurance up here because of the risk. The new software that they're using to do risk modeling, the base risk for Big Sky almost puts it over insurable property. And then they have to pretty much through mitigation measures and utilizing ignition resistant material and things like that, they have to pretty much earn their way into getting insured. I mean that's, that's kind of the climate that we're living in with the insurance industry right now up here. It's pretty crazy. And so we get, we get probably at least two calls a week of a homeowner saying, Hey, my insurance company's gonna drop me unless I do this, this, or this. And it's like, okay, well, you know, we come up with a plan, the insurance adjusters, you know, are, they look at us and are like, Oh, now these guys know what they're doing and, and we'll go by whatever they say for you guys to do. So that's kind of how it's been working a lot up here. It's pretty wild. 

Jennifer: After the break, we meet someone helping the insurance industry design this process… using image analytics to plan for future wildfires. 

And later in the series, we’ll learn about a fire district in California that’s automating that risk assessment process for its homeowners.

[music full]

Jennifer: You can find links to our reporting in the show notes... and you can support our journalism by going to tech review dot com slash subscribe.

We’ll be back… right after this.

[MIDROLL]

Jennifer: Before a disaster strikes, insurance collects small amounts of money from many different groups - like groups of homeowners or businesses - so that when the unthinkable happens, they’re protected. That only works though when no more than a set portion of the group is affected. 

Sankar Narayanan: So, simple example is climate change—that is affecting everybody. So the original purpose of that equation to leverage the power of many to address the possible few that are going to be affected is no longer applicable because climate change affects everybody. So in such a scenario, what does the future of insurance look like? My name is Sankar Narayanan. I lead insurance practice at Fractal.

Jennifer: Fractal Analytics uses AI to help insurance companies design the claims process for natural disasters… including wildfires.  

Sankar Narayanan: Over the past few years, there's been tremendous technological advances, which allows for us to leverage satellite imagery, drones, et cetera, to understand the extent of damage. Much faster, more effectively, but if you, if you go back in time, that used to be one of the biggest challenges. So how do we create an effective data framework that allows us to capture all of the key information that is required to understand how wildfire spreads.

Jennifer: To do that you need to understand the history of wildfires in a given area. And so his first step is to put together a big picture understanding of the wildfire ecosystem…  allowing him to better predict and prepare for the next one. 

Sankar Narayanan: So if you take a geographical location and study through drone images and satellite imagery, How is the geographical area set up? You know, how many buildings are there, how much ranch space is there, How many commercial versus personal properties are there? So there's a lot of data that goes into understanding how is it structured, how is an area structured right? But there is also a lot of unstructured data, which is the image of how they're organized, right? And these images give us a real good understanding of what may be the, uh, spark points, right? How can something spread very quickly versus how can you control or contain the extent of damage.  

Jennifer: It means image analytics are key…both before, and after, a fire.   

Sankar Narayanan: If I go back about three, four years, our ability to understand the extent of damage, uh, was actually limited by how quickly we could make an accurate assessment of the damaged area. Now that has become substantially advanced in terms of the amount of post image data that we can get. One, as well as the speed at which we can get that. Today, we are able to get it at t plus three - within three days we are able to get much detailed, granular image data to allow us to understand the extent of damage.

Jennifer: He’s using deep learning algorithms to much more quickly analyze the damage after a fire as part of the insurance claims process… and he’s also using it to make better predictions about future damage. 

Sankar Narayanan: The number of such events, if I take accumulative, um, you know, set of, uh, events has not been that many for us to make, you know, extremely accurate predictions. That's where we use techniques such as generative adversarial networks, right? So these GANs, as we refer to them, are very advanced algorithms that allow us to actually simulate the likely, you know, occurrence of wildfire. So, so we can actually change various parameters, you know, if over the next 10 years temperature, overall temperature increases by half a degree, then what's the likely change in the  probability of occurrence of wildfire is something we are able to do now because we are able to simulate the occurrence of wildfire as opposed to actually waiting for it to happen and then modeling it. 

Jennifer: And he believes AI can help expedite the claims process for policy holders in other ways too… so people impacted by fires and other natural disasters can spend less time on paperwork.

Sankar Narayanan: How can AI be used to produce the right kind of data? Instead of making this an even more convoluted process for policy holders. So that's where we are doing a lot of work today in seeing how can, how can we use the, you know, the power of generative adversarial networks to test, experiment, create more simulations, create more models that can improve the overall accuracy and efficiency of the, of the models that we are already using.

[music]

Jennifer: We’ll take a closer look at the tools and technologies being used in the field… from satellites and drones… to those smoke detecting cameras we heard about earlier… in part two of our reporting on wildfires.

This episode was reported and produced by me, Anthony Green and Emma Cillekens. It was edited by Mat Honan and mixed by Garret Lang… with original music from Jacob Gorski.   

If you have an idea for a story or something you’d like to hear, please drop a note to podcasts at technology review dot com.

Thanks for listening… I’m Jennifer Strong.

[TR ID]

Deep Dive

Artificial intelligence

A Roomba recorded a woman on the toilet. How did screenshots end up on Facebook?

Robot vacuum companies say your images are safe, but a sprawling global supply chain for data from our devices creates risk.

The viral AI avatar app Lensa undressed me—without my consent

My avatars were cartoonishly pornified, while my male colleagues got to be astronauts, explorers, and inventors.

Roomba testers feel misled after intimate images ended up on Facebook

An MIT Technology Review investigation recently revealed how images of a minor and a tester on the toilet ended up on social media. iRobot said it had consent to collect this kind of data from inside homes—but participants say otherwise.

How to spot AI-generated text

The internet is increasingly awash with text written by AI software. We need new tools to detect it.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.