Podcast: From Prevention to Recovery: AI in Disaster Management

In this episode, we explore a critical AI developed by Texas A&M researchers that aims to play a major role in all phases of disaster management, from prevention to recovery, by changing the way stakeholders prepare for, respond to, and recover from disasters, enhancing public safety and improving outcomes in crisis situations.

This podcast is sponsored by Mouser Electronics. 

Episode Notes

(4:17) - AI For All Phases of Disaster Management

This episode was brought to you by Mouser, our favorite place to get electronics parts for any project, whether it be a hobby at home or a prototype for work. Click HERE to learn more about the history of machine vision, its prevalence in our daily lives, current bottlenecks, and what the future holds for this critical technology!

Become a founding reader of our newsletter: http://read.thenextbyte.com/

Transcript

What's going on, folks? Welcome back to the Next Byte Podcast. And in this episode, we're talking all about natural disasters. Well, to be specific, we're talking about hurricanes, Texas A&M, and the research group there that is trying to come up with an AI model which helps communities become better prepared and better react to hurricane disasters. If that's got you excited or scared, depending on how you feel about hurricanes, then I don't know, let's start chasing some storms.

I'm Daniel, and I'm Farbod. And this is the NextByte Podcast. Every week, we explore interesting and impactful tech and engineering content from Wevolver.com and deliver it to you in bite sized episodes that are easy to understand, regardless of your background. 

Farbod: All right, folks, like you heard, we're gonna be talking about the weather today, which is quite timely because that movie with Glen Powell, Twisters, I think it just did box office numbers. So, we really lined this up quite well. And-

Daniel: Is that how we lined it up?

Farbod: Sure, yeah. We definitely intended this to happen. We're part of the press release, you know?

Daniel: I know you're a big Glenn Powell fan, so.

Farbod: You know me well. You know since I saw him in Top Gun, he had me hooked. But with that said, let's quickly talk about today's sponsor before we jump into the episode. And today's sponsor is Mouser Electronics. Now folks, you know, if you're a fan of the podcast, that we love working with Mouser. And the big reason for that is because they're pretty, pretty in sync with us. They wanna share their knowledge about what's going on in various industries in a way that's easy to understand for the average person. And they do that with videos, and occasionally they make these technical resources, articles. And we have one link in the show notes today that talks about machine vision. You've probably heard about machine vision. It's all over the automotive industry, specifically with Tesla that relies heavily on it for the self-driving system that they got on board. But machine vision has actually a very interesting history. And the article goes over this history. It talks about the origins with the first weather satellites and how it started to evolve from this process that was machines taking the image and then the human doing the analysis to something that is pretty much all machine now because computing got so much better. It talks about the role of machine vision in our everyday lives from the scanning that happens in the factories that are looking for, ripe tomatoes or specifically not ripe tomatoes so they can kick it out and it doesn't make its way to the shelves in your grocery stores to imperfections for creating panels for a car. All these various ways that machine vision is impacting our lives and we basically don't even know about it. And my favorite part, it starts to highlight challenges for what the future of machine vision looks like. It talks about how our cameras are getting better and better and we need faster and faster computing to just manage the amount of output that's coming out from them. It gives a nice little shout out to the graphics processing units or GPUs. NVIDIA, if you've been following the stock market or the gaming industry, you're already hip to them.

Daniel: Yeah, you know, Nvidia stocks through the roof for this reason exactly, right? Which is the trend of machine vision. And by the way, I like this terminology, machine vision as opposed to computer vision, because I feel like it's a little bit more intuitive to like all the potential applications when, when I hear computer vision and I feel like when a lot of lay people hear computer vision, it's hard for them to tie parallels to something like a self-driving car because they don't view a car as a computer. Right, in this case, machine vision, we're using computers as a method of allowing these machines, right? I would say a car is a machine colloquially, allowing machines to use cameras and use computers to see. So, I like the terminology here with machine vision, but yeah, we love Mouser, we love this technical resource specifically, which is why we think it's worth sharing with you guys. And that's why we've got it linked in the show notes and you should check it out.

Farbod: Absolutely, that's a really good observation. I honestly hadn't thought about it, but now that you say it out loud, I get it, I see your point. But with that said, let's segue into today's article. And folks, we're going out to the Midwest, we're going to Texas, specifically Texas A&M. There's a professor, Dr. Ali Mostafavi, who's working in the civil engineering department, and they've been tackling disaster management, which again, you could probably tell by the intro that we did and talking about the movie Twisters. If you've been watching, The Weather Channel, you notice that there's been like a lot of hurricanes going on. Like a lot. And well, it looks like...

Daniel: Do people watch the Weather Channel?

Farbod: I love the hosts. They always got the best dad jokes. That's how I got inspired to do my dad jokes.

Daniel: And I guess I will say they probably are the least biased out of any news source in the world, right? They're just, they're all about the weather.

Farbod: Yeah. How could you be biased about the weather? Well, I have my feelings about humidity in Virginia, so let's not go there. But you've probably been noticing that there's been a lot of hurricanes and you're right on the money with that observation if you have been observing that because NOAA is predicting a record number of name storms in the Atlantic this year. So, it's been hurricane-y. And all of this leads to us and the government thinking, well, how do we make sure we're as prepared as we can be? How do we make sure that as these hazardous events come and go that we can respond to it as best as we can, both during and after the fact? And it turns out that Dr. Mostafavi's team has been focused on this very problem for the past couple of years. It's interesting, they provide some really interesting insight in the article where they talk about, we know that doing flood risk assessment, for example, is a great way to gather the right type of data so that we can understand what neighborhoods can get damaged. But we simply don't do it because it's very time intensive and it's very cost intensive. The same thing applies to understanding the lowest floor elevation of every building in a given area, right? Like understanding if the ground floor is 10 feet off the ground or 30 feet off the ground can allow the surveyors and the government to know if an area is gonna be severely impacted or if they're likely gonna be fine. But when you look at somewhere like Austin, because they're in Texas, there's a hundred of thousands of different buildings that you just have to survey and it doesn't work, right? Like there is a cap to the amount of manual labor that you can feasibly do for a task like this.

Daniel: Well, and let's talk about why this is so important, right? If you have all this data ahead of time, you can have very targeted guidance for people, depending on the type of natural disaster that's inbound, give very direct, very clear guidance on who needs to evacuate, when they need to evacuate, who needs to evacuate from which type of building, which type of zip code, et cetera. Right now, we usually get broad sweeping guidance saying, everyone in this county has to evacuate their homes. And that's helpful, but to some extent, there's maybe some people in buildings that are depending on a certain flash flooding event actually might be safe in their building. And by putting them out on the road, you're actually putting them at risk. You're congesting traffic, you're creating an inefficient evacuation pattern as an example, as one of the ways of managing a disaster. That's just one example of the way that the lack of data causes are the powers that be our governments to do slow and inefficient disaster management that actually leads to increased harm. So sometimes you tell too many people to evacuate, the road gets clogged, and then they get smacked by a hurricane and the people that were on the road, they get in trouble, they get hurt. And then it also leads to delayed recovery for affected communities. Imagine if there were some kind of preventative measures you can put up, I know that there's like flood barriers and stuff like that, that you can put up to try and protect certain buildings. If you can let building owners know which types of buildings are at risk and actually which exact buildings are at risk due to any incoming hypothetical disaster, you can help decrease the amount of damage, decrease the amount of harm to people, and then also speed up the amount of recovery for affected communities. So, the idea here is like, how do we leverage state of the art AI tools to try and enhance our understanding of the situation, understand different areas, understand structures, elevation, and then also how can we prepare better plans to react? So, like, one, understand ahead of time what the situation looks like, have a lot of data, and then two, develop better, faster, more accurate, more effective recovery plans using, again, in this case, cutting edge AI technologies.

Farbod: Yeah, you know what? I'm actually gonna reference back to the Mouser article here for a second, where they talked about the genesis of machine vision and how at first it was humans that were in the loop to do a lot of the analysis and a lot of the hard work. And over time, it was the machines that took over. So now I'm looking at this landscape of surveying where it was very manual for the longest time. And now this team is thinking about the problem and how it could be automated. And it seems like AI could do some of the heavy lifting of the analysis of readily available data to come to the conclusions that you would typically get from a survey going somewhere. And you're right, they kind of broke down their solution into two buckets. The first bucket being mitigation and preparedness. Like what is the best that we can do to just make sure our communities are ready for whatever's to come? And the second part is response and recovery. So, right as the hazardous event has passed, how can we make sure that we can respond as quickly as possible and get the right kind of information so that we're reaching people, satisfying people's needs, right? And it's interesting. I'll start with the mitigation and preparedness. Let's talk about the flood data, right? These folks realize that you can use hydrological, topographic and built environment features in a given area, so think about like a neighborhood or a city, to understand its flood risk. And the way the analogy they use, which I'm a big fan of, is if you were to look at someone's DNA composition, you could predict quite well what kind of diseases they're at risk for. And this system, this AI model they developed is quite well named flood genome because it aims to do just that. It takes those markers and it's like, okay, you know what? Daniel's house, you guys are probably fine for this flood that we have coming, but Farbod's house, due to your elevation or due to the manmade structures around you, you're actually kind of at risk. So, you should probably be the one that's evacuating. And then they-..

Daniel: It makes a ton of sense. And I love the analogy here to genome, right? Like it takes me back to like when we first sequenced the human genome. Right? We took in a bunch of data and we use it as an opportunity to unlock understanding about how certain predispositions in our DNA might manifest in the real life. And it doesn't mean if you're predisposed to this certain disease or you're predisposed to having red hair, that it's absolutely going to happen, based off of your DNA, but it vastly increased the likelihood that those characteristics will appear in your life.

Farbod: Right.

Daniel: Very similar here, right? You can't look at the hydrological, which is just the pattern of how water flows throughout certain areas, the topographic, the elevation and the shape of the land that they're on. And then the built environment features, the buildings that are on that land. You can't just look at those and absolutely determine which building is gonna flood and which one won't, but you can gain an understanding from a probability perspective, which buildings are more likely to flood and which ones are less likely to flood. You can't make 100% deterministic answers right now, but compared to the alternative right now, we have no understanding of which building might flood and why without maybe having a history of like, oh, last time I was in this building, it got trashed. Like this allows us to have a proactive approach as opposed to a reactive approach to try and understand what potential failure modes might occur before they even happen.

Farbod: Well, I don't know how we're so in sync today, but that is such a good segue into the next thing that I had in mind. You're saying, you know, the best way you could think about it is, well, last time I was in this building, I got hit, so maybe it'll happen again. Well, the folks took Flood Genome and a different model that they came up with, which can determine the lowest floor elevation by looking at Google Street View data. It's called Elev-vision, which by the way, the naming game, this team has been knocking it out of the park. So, I was like eight and a half out of 10.

Daniel: I was gonna say 8.9. You know, we like to give people credit when they have good names. Love the team here. Respect, Flood genome and Elev-vision. Great job.

Farbod: Yeah. So yeah, they took this and they were like, all right, now we have this model that should tell us where we should be worried about flooding with a certain amount of accuracy and how do we test this? Well, it just so happens that the one group besides you, the homeowner who really cares about flooding, AKA the insurance companies, have a lot of historical data about floods. So, they took that historical data and they tested their model against the areas that did have flooding. And I'm gonna take the quote from the article. “It performed beautifully.” So, this model has been proven to work as expected against historical events where flooding has impacted different regions, which is quite impressive.

Daniel: I'm with you, man. And it kind of, I don't wanna jump too quick to the end here, but one of the notes I had here around the so what, right? The potential impact here is in addition to disaster management and doing a better job in responding to potential crises. I do think there's a huge impact here for homeowners in flood areas. I've noticed a lot of headlines recently about like areas of Florida being almost completely, a quote here, “uninsurable”.

Farbod: Unable to, yep, yep.

Daniel: Because a lot of homeowners insurance or flood insurance companies are just unwilling to take on the risk of insuring certain homes in certain flood prone areas and I think having a better understanding of exactly which type of home might be flooded, which one might not, it might give insurance companies enough of a foothold to say, yes, actually we do understand how flooding may or may not occur in this certain type of neighborhood, in this certain type of, again, flood genome, right? And really that's what insurance companies need to be profitable is just enough data to understand what's going on. And again, I'm not vying for huge insurance companies to make tons and tons of money. But what I am looking for is like people who live in these flood prone areas, that they have some level of safeguard from a financial perspective against vast amounts of risk. And really that's all insurance is, is you and me and the other guy down the street, we all chip in a small amount of money every single month so that when one of our houses gets struck, struck by lightning, it's able to, you know, the insurance company is able to help cover that expense. And I'm hoping that with additional information from this Texas A&M team, Flood Genome and Elev-vision. Maybe this will help insurance companies dip their toes back into the water, so to speak, in certain areas of, again, Florida is the one where I've seen the most headlines about, where people literally can't get flood insurance for their homes, and when their home gets smacked by a hurricane, they have no way of recovering financially.

Farbod: Right, well, again, I don't know what's in the water. We're very in sync, because that's exactly what I wanted to talk about, but slightly more pessimistic than yours, because I was thinking, what happens if insurance companies can get their hands on this and start applying it to all the regions that they think might even slightly be a risk, and then start spiking up insurance costs for everyone to just kind of make up for the difference. But again, exciting nonetheless, if it helps people be safe, that's really what matters, which takes us to the next section, by the way. So, we already talked about mitigation and preparedness. Let's talk about response and recovery. As the event is happening, right, the folks as a part of Dr. Mostafavi's team said they can actually use location data to provide in real time information to stakeholders and decision makers about how people are evacuating, like what routes that they're using, which can help the authorities provide feedback to the general public of, well, this is getting backed up, maybe now try taking this route, et cetera, et cetera. It also helps them understand what communities are evacuating and which ones haven't yet, or maybe have only partially evacuated.

Daniel: That's what I was going to say is there's traditionally during these disasters, there's a bunch of holdouts who are like, no, my house isn't going to get destroyed by this hurricane. And maybe their house ends up getting destroyed and having some level of understanding during the evacuation process ahead of time. Say like, oh, actually this neighborhood here has a lot of people in it. Maybe it'll encourage, you know, folks to either go out and knock doors and say like, hey, are you prepared for this type of event to occur? And then also might help inform response faster if that neighborhood does end up getting smacked by a hurricane. And I'm sorry for saying smacked by a hurricane. I'm sure there's better terminology.

Farbod: Impacted.

Daniel: Yeah, if this neighborhood ends up getting impacted by a hurricane, right, they can know using data to understand, actually the vast majority of people have evacuated from here versus another neighborhood where a vast majority of people haven't evacuated and that might help them make sure people are accounted for, understand where they need to do disaster response, etc. The last thing you want to do is send a bunch of rescue folks to one pile of rubble that has no people in it, as opposed to the other pile of rubble that does when there's a finite amount of resources.

Farbod: For sure, for sure. There was another tidbit in this section that really stood out to me. It was called preparedness. And it was about how they have a model that can track the number of people that went to grocery stores over a period of time before the event hit to kind of understand how prepared a community is if they're to hold out in their area during the hazardous event. Which I think is quite neat. The last time I had heard about mass surveillance of grocery store data was for a private equity firm, or sorry, what are the stock trading places called?

Daniel: Hedge funds.

Farbod: Hedge funds, there we go. It was about a hedge fund that was using satellite data of Walmart over the course of like an entire quarter to determine their revenue and then make projections on their stock prices. So, this is a much better use case of that same technology and same idea.

Daniel: I'm surprised, dude. You're becoming increasingly authoritarian. I don't know. I feel like two years ago recording this podcast, you'd say, I don't care. I don't want the government to know whether I'm well-prepared or not. That's too big brother for me. And now you're saying?

Farbod: Well, I saved that portion for you where you tell our audience the potential downsides of technologies like this, because we made a promise to them that we give the pros and cons.

Daniel: Okay, I'm with you here, but like, I'm just trying to play devil's advocate because I'm so used to you being the, oh no, don't surveil me person. And now you're all happy about the government knowing whether or not you've got enough groceries before the upcoming storm.

Farbod: I said it was interesting, right? And I still think it's interesting. I just think that if the technology's out there, right? And you can have someone using it to predict the price of a stock and whether or not to buy or dump it or tell people, you know, get out because you're actually not prepared or get resources to them, there could be a silver lining. You never know.

Daniel: I guess this is certainly the lesser of those two evils, if I were to compare them. I'm with you.

Farbod: Right? There it is. But yeah, all of this put together is impressive. If you've been rocking with us for some time, you know we love talking about cutting edge research, but what we love more than that is for it to become a reality. And I'm very happy to report that this team is actually spinning up a company to commercialize this technology package for the public sector. So, it's quite likely that we'll see utilization of this technology by the local and maybe even federal governments that are involved in a flood and hazardous weather event. Preparedness. Mitigation. Yeah, do your spiel. Do your pros and cons.

Daniel: My biggest pet peeve about tech is when awesome, incredible academic teams come up with a new piece of technology and then it sits on a shelf and collects dust.

Farbod: That's a spiel.

Daniel: Yeah, if you've listened to the podcast for more than a couple episodes, you know, I say this often but real kudos to this team from Texas A&M, Dr. Mostafavi's team. Appreciate you developing an awesome technology that can go impact the real world and then going and making sure that this technology has legs to make it out into the real world. This is potentially life-saving technology in the orders of millions of people sometimes depending on the size of the national or natural disaster, so greatly appreciate this team here. One thing I did want to mention, the last part of their secret sauce, which I don't think we've covered so far yet. Which is they also try and get access to satellite images and use computer vision, the same computer vision that they're using before for their flood genome. They try and do this again after a disaster to quickly classify where there's property damage. And then they can probably overlay that with other data, including like which ones were more likely to be heavily impacted alongside with where people have and haven't evacuated or which ones do and don't have their groceries, right? This can help authorities respond quickly with recovery efforts in the areas where they're most needed. I kind of alluded to it earlier when I said like, if there's one pile of rubble that has no people in it and another pile of rubble that does, which one do you go send a bunch of rescue people into? Obviously, you wanna send it into the one where you can save some people's lives. This actually makes that a reality, which again, just goes to show how important it is that this team's doing the best that they can to spin this out into a commercial entity and make it into the real world.

Farbod: Yeah, yeah, for sure. I'm gonna do a quick recap and then we can go ahead and wrap the episode up.

Daniel: Yes, sir.

Farbod: So, folks, hurricane preparedness is very, very important. Unfortunately, even though that we know flood risk assessments can help us a lot, we don't do it because it costs a lot of money and it takes a lot of time. Well, that's okay because there's a very talented team of researchers at Texas A&M that are leveraging AI to make that happen for us at scale and in a cost-effective matter. See, they realize that you can use hydrological data, which is how water flows, topographic, the ups and downs and elevations of a landscape, and the built environments around it, so man-made structures and whatnot, to determine the flood risk quite well. How well? Well, they used that in combination with the lowest elevation of the first floor, which they pulled from Google Street View, by the way, insane, very crazy, to test against historical flood data and realize that they were perfectly matching the expected outcomes. That means they were perfectly predicting where flooding was happening. And that's a great save already, but they didn't stop there. They said, this is great for understanding how we can be prepared. But what happens as the disaster is happening and how do we respond to it? Well, they came up with even more models, models that can predict how people are evacuating out of a certain region, if certain communities have just not evacuated at all, or if they've partially evacuated. That helps first responders, stakeholders, and decision makers understand where they should be allocating resources. And after the disaster's over, they can use satellite images and computer vision to assess the damages, allocate the right amount of funds, and send people to the right rubbles. Well, all that to really say, I think these folks at Texas A&M have developed something that can drastically change our lives and potentially save them.

Daniel: Love it.

Farbod: I try. I do what I can. But yeah, I think that's it.

Daniel: Two things. One of them being Austria. The other one being Slovakia.

Farbod: Oh my God, we were trending.

Daniel: In both of those countries. So, to our friends in Austria, I'm going to say “Danke”. And to our friends in Slovakia, I'm going to say “dakujem vam”. I don't know if I said that correctly, but both of those mean thank you. I'm trying my best to say thank you in your native tongue, to appreciate and share our appreciation for you, helping us become one of the top trending podcasts in your country.

Farbod: To our Slovakian friends, if he said it wrong, please roast him on Twitter, Instagram, or your favorite social media platform.

Daniel: Yeah, and friends in Austria, I know I said Danke correctly, so. No worries.

Farbod: All right, folks, thank you so much for listening. And as always, we'll catch you in the next one.

Daniel: Peace.

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.

To learn more about this show, please visit our shows page. By following the page, you will get automatic updates by email when a new show is published. Be sure to give us a follow and review on Apple podcasts, Spotify, and most of your favorite podcast platforms!

--

The Next Byte: We're two engineers on a mission to simplify complex science & technology, making it easy to understand. In each episode of our show, we dive into world-changing tech (such as AI, robotics, 3D printing, IoT, & much more), all while keeping it entertaining & engaging along the way.