Everybody Goes Home: How technology is helping prevent wildfires from spreading

0:00Zachary Wells:These fires just devastate people, devastate communities. If we can prevent this anxiety and this fear and deep-rooted just concern from people that have gone through this, that's a success.

0:15Brad Smith:First priority is life. And the more that I can do to engage in technology that helps protect my communities and protects my firefighters, those are no-brainers.

0:26Neal Driscoll:That's Deputy Fire Chief Zachary Wells and Dr. Neil Driscoll, a professor at the University of California, San Diego. Together, they're two of the bright minds behind Alert California, one of the most ambitious early warning systems for wildfires anywhere in the world. The Microsoft AI for Good team has worked with Zach and Neil to use AI and cloud computing to detect fires earlier, share intelligence faster, and give firefighters the real-time data they need when every minute counts.

1:00Brad Smith:Alert California is an enhanced situational awareness platform. We have cameras strategically placed on mountaintops to allow us to visually identify and confirm fire in its incipient phase.

1:12Neal Driscoll:We talk about how this technology is already starting to change the way firefighters protect lives and communities, and what it will take to scale its life-saving potential around the world. Up next on Tools and Weapons. Now, when did the two of you meet?

1:31Brad Smith:2017, a Ventura County firefighter said, you have to hear about what they're doing at UC San Diego and meet this guy named Neil. And we've hit it off. We've learned how to work with each other from the fire side, the university side, and develop tools that can be turned over to firefighters for them to use at no cost to them.

1:51Neal Driscoll:And Neil, what was the inspiration for you that got you going down this path in the first place?

1:56Zachary Wells:2003.

1:58Neal Driscoll:What happened in 2003?

2:07Zachary Wells:impacted a lot of people was the largest fire at the time in california yeah skirted up my neighborhood oh really people there was like 23 24 people lost their lives oh my gosh we didn't know where to go so you called 911 and said i can't see the whole sky was orange where do i go They couldn't tell me. And this fire jumped all the lanes at Miramar. My kids were so frustrated and scared. They kept saying, are we going to be safe? I'm still scarred by fire. Sure. I had to take a pause there.

2:45Neal Driscoll:Yeah.

2:45Zachary Wells:And I didn't lose my house. I didn't lose a loved one. If we can prevent this anxiety and this fear and deep-rooted just concern from people that have gone through this, that's a success. These fires just devastate people, devastate communities, compromise biohabitats, changing the vegetation. We need to use every bit of technology to try to hold them in check. And I've been building this network. I build instruments to bring signals from out in the wilderness back to laboratories. We decided to put a couple of cameras and actually use the infrastructure to bring in information from remote areas.

3:32Zachary Wells:I can have a firefighter from Oregon run my system if I need to import. They don't have to know anything about the mountain ranges, the camera names. The AI says, look at these cameras. Something has changed since I've rotated or I've taken the 60-degree image. So the AI is just saying, go look at the change. Tell me if it's a fire or it's the marine layer. It's fog. It's a dust devil. So all of a sudden, we have a change detector that allows watchstander fatigue to be removed. It reduces the noise. If fire or an incident comes up, it just hits that screen and you can see it. There's nothing else there but that fire or that event.

4:14Zachary Wells:And then you can interrogate it and figure out what it is.

4:17Neal Driscoll:Can you say a little bit about how much territory you cover and then compare the size of that territory with the number of fire engines or pieces of equipment and people that you have to cover that territory?

4:28Brad Smith:Yeah. So in Kern County, we cover 8 ,141 square miles. It's the rough size of the three smallest states in the continental U.S. And I protect that area with 47 fire stations. So each station has a fire engine with three firefighters. About 170 firefighters in our department is on every single day. Okay. But 170 spread throughout 8 ,000 square miles is a big area to cover.

4:53Neal Driscoll:We live in a time when I think most of us are used to having a smoke detector in our homes. And the obvious principle is if you detect the fire early, that's how you best put it out. This is like, in effect, an AI-driven smoke detector for this entire region. What is the impact of early detection for somebody like you and your team as you're responsible for putting out these kinds of fires?

5:21Brad Smith:So I've experienced already the ability to detect fire and confirm it in an incipient phase, respond an effective firefighting force to first alarm of engines and chiefs and all the crews that we need, and extinguish that fire without ever receiving a 911 call. Going back to the protecting life, property, and environment, first priority is life. And the more that I can do to engage in technology that helps protect my communities and protects my firefighters, those are no-brainers. Running a fire department, we don't have an R &D budget. We don't have time other than our priority is responding to emergencies.

6:04Brad Smith:We're not going to stop every single fire. You know, it's there's over 10 ,000 wildfires that happen in California every year. So seeing a program like Alert California and figuring out that we can invest our people in our effort, opening up our radio towers to add a camera on a tower that exists for our daily use. We feel that we can make a dent in the universe and make an impact. We already have a action plan in a system that works really well, but we want to make it better. We want to be able to engage those that are the boots on the ground that are doing the hard work.

6:40Neal Driscoll:And obviously a big part of your success is sending the right people and the right equipment to the right place at the right time.

6:48Brad Smith:It's hard to predict where an incident is going to occur. But using AI, using these cameras that are always looking and finding and observing, and then getting that signal through all the noise to the right people, the dispatchers, the firefighters, so that they can make those decisions. That's where we've found success. That's where we'll continue to find success as technology drives this to more advanced levels.

7:12Neal Driscoll:One of the things I find fascinating about the progress the two of you are making together is that you are making the technology better and cheaper. Yes. At the same time. Which is, of course, the story of computing.

7:23Zachary Wells:Right. We also, with the technology we're developing, we share that with other people. And so they cost them nothing. So here we share that R &D, which is huge. Yeah. Look at this. Look at this area. Look at how much area is covered by this one camera. Every day I look at all cameras. And then I look at cameras that have been recently moved, active cameras. Because that tells me somebody that's been trained by us, onboarded, deliberately moved a camera, so there had to be some reason. And so you look there first and you go, oh, that doesn't look good. And then you see them triangulate. And so here the public can see that.

8:04Zachary Wells:You can go on and say, click the button that says active cameras and see where firefighters are looking. That's such a benefit. And we have layers with maps from Esri that all the street maps are in there. So you can look at where your house is. You want to know where your house is, and then you can see the perimeter of the fire. And you can say, you know, I don't need a firefighter or a sheriff to knock on my door. I'm right in front of that fire that's coming at me, and it's moving north, and I live right here. Wow.

8:36Brad Smith:It's a benefit of having an open system and having the imagery available to the public. And we connect with other technology providers that provide information, including Cal Fire that hosts the imagery through our system on their site when they're posting official information. When you're dealing with emergencies to this magnitude, you don't have the time to dust off username and pass codes. You just need to be able to see it to make decisions very quickly for your department, for your community, for yourself. and that's where being in an open system has greatly benefited us.

9:11Zachary Wells:You have to use it because in an event, when fire's at your back door, trust me, you're so anxious and you're not making good decisions.

9:29When you think about the decade over which this work has been going forward,

9:33Neal Driscoll:I mean, a lot has changed. Cameras have gotten sharper. Connectivity has gotten faster. Batteries have gotten smaller. Computers have improved. What has been the impact in the last year or two of AI?

9:47Zachary Wells:I look at the metric that the AI is providing reducing noise that allows the dispatcher to have situational awareness and observe how things are changing. And it's telling me where I have change. And now I want to have it tell me where I have changed faster with better clarity. And what Zach has taught me, he says, Neil, when I can look at a fire in almost real time, like video, I gain so much knowledge in that first four to five minutes. It's critical. I can tell so much from that little time slice, especially if I get frame rates higher. We're never going to take the subject matter expert out of the loop.

10:31Zachary Wells:So when people ask me, well, is this going to take jobs? No, it's going to create hybrid jobs.

10:36Brad Smith:AI is not meant to replace firefighters. It's meant to enhance firefighters to allow them to do their job. But through AI being able to see the signal of an incipient fire and be able to respond, that's something that I can do something about. We can contain and confine those fires when we know about them. What AI does is reduces the potential for large complex fires and allows us to mitigate them at the smallest level, which means my firefighters get to return home to the areas that they respond to medical aids and traffic accidents and structure fires. And those resources are then prepared for the next fire.

11:17Brad Smith:AI has a way to help keep us safer and notify us so that we can do our job.

11:22Neal Driscoll:and one is the person who's brought the AI technology to all of this. It's sort of a story of the firefighter, the professor, and the data scientist. From the data science side, what's the key to working effectively in this kind of team?

11:38Zachary Wells:For us, what is critical is the fact that we need to get very good quality data. So these cameras not only provide amazing data, but you have great people that understand what is a fire, what is not a fire to label the data and indicate to the AI models. Like that's the key for our success. Firefighters labeling data, that's critical for us. So the next part is the fact that you have these models to have the ability to control the cameras and actually zoom in because that will provide much better quality data. That's another thing that is key for us. So I've had one dream fulfilled with this project.

12:16Zachary Wells:But what's that dream? That dream is that I wanted to have, I could drop a pin on the map, and I could see every camera that could see that pin illuminated, and I can go look at it. That's what I wanted. That was the dream. So I can know what cameras have unobstructed view of a fire. And that was hard to do. But now with the data quality and what Juan's talking about is, how do we take these data and get them so they go into the model?

12:51Neal Driscoll:All of a sudden, we're reaching what we believe is a better targeting on the fire than we had before.

12:59Zachary Wells:And I think that we're in a total agreement that we have to have awesome data and we have to have enough of it that we can populate the model. This is very promising. The next step is, in places where we have multiple cameras, is that the model will detect the camera. As long as you have two cameras that can look at the same fire, you actually know the location of the fire, because then the model will basically indicate the location of the fire. That's critical for you.

13:26Neal Driscoll:Well, think about the decade that you all have been on this journey. You started 10 years ago. Think about eight years. How far you've come. Think about 2035. what would you hope would be the state of the art in 2035 not just in terms of what technology can do but what technology actually is doing not just here in california but everywhere where these

13:51Zachary Wells:wildfires are such a hazard right my dream is that the camera after it sees what's a potential fire it moves all the cameras around it automatically points it at it and they zoom in automatically So I don't have to do all that. All that work is done because that's kind of tedious. It takes time. And then the firefighter just can look and say, yes, that this is an issue. Is it a small issue? Is it a big issue? Is it a controlled burn? Do I have to address this? All of these things are running through the dispatcher's mind. And if we can give them high-quality data and enhanced, you know, situational awareness, the dispatcher is going to be able to make a data-driven decision that is more complete.

14:37Zachary Wells:Every fire is different. And every fire has the chance to get out of the box. And that's what we're trying to solve is let's not let the fire out of the box. And the good thing here is also because you have so many cameras and you have, unfortunately, given the amount of fires that you have every year, you have every fire you can learn from the fire. Right now, the focus has been on detection. The next, I think the next also phase is prediction, is what's going to happen with that fire, correct? And that's something that, like, yes, every fire is different, but as long as, if you have enough of these fires, you can train models to see what's going to happen in the next hour, in the next 30 minutes.