Episode 62. What If? … How modelers envision the worst to enable the best response.

Bernie Fette (host) (00:14):
Hello and welcome to Thinking Transportation. Conversations about how we get ourselves and the things we need from one place to another. I’m Bernie Fette with the Texas A&M Transportation Institute. Extreme events take many different forms. With weather calamities — hurricanes and such — we have the benefit of advanced warning through accurate forecasting. Other catastrophic occurrences like earthquakes or terrorist attacks happen suddenly with little or no warning at all. What all such events share in common though, is that their impacts can be long-lasting and far-reaching. In this episode, we’re visiting with someone who works to model those events to better understand their impacts on transportation networks and help organizations develop contingency plans for how to recover and resume normal operations. Jeff Shelton is a research scientist at TTI. Welcome to Thinking Transportation, Jeff. It’s great that you could share some of your time with us.

Jeff Shelton (guest) (01:22):
Thank you, Bernie. It’s a pleasure to be here.

Bernie Fette (01:25):
I was hoping we could start maybe with some definitions. How would you explain your work? Just for someone who doesn’t have a clue about what traffic modeling is. Someone at your high school reunion, for instance.

Jeff Shelton (01:41):
So we all encounter traffic every day when we drive, right? Right. And sometimes we experience congestion. Sometimes we experience work zones. Sometimes we experience accidents. Modeling is trying to take existing conditions of the roadway network of traffic and put it into a computer simulation. Now, if you wanna do a prediction of something, then you can change the model to predict what’s gonna happen. For instance, if there is going to be an expansion of the freeway and you’re gonna have a work zone for six months, how is that gonna impact traffic? You know, if there’s an accident on the freeway and you’re delayed 20 minutes, you’ll probably still take the freeway tomorrow. But if there’s a work zone for six months, you may take a different route. So simulation modeling is trying to predict those changes in traffic patterns on what people might do, given certain conditions that happen on the roadway. What I do is I build a computer model and it could be something as simple as an intersection, or it could be something as large as the city of Houston. You know, I’ve built several cities around Texas, all the major cities I’ve built, and those are a little more cumbersome, tedious, time consuming to build. But you get a bigger picture of how traffic is running in your city. What I try to do is just recreate existing conditions and then change the model to predict some sort of future event. Whether it’s a work zone, an accident, an improvement in the roadway. Are you gonna charge a toll that you didn’t charge before? So I’m kind of trying to make predictions of traffic.

Bernie Fette (03:24):
I can’t help but ask you, does anybody ever ask you if you work with a crystal ball <laugh>?

Jeff Shelton (03:30):
My wife tells people I play video games for a living <laugh>.

Bernie Fette (03:35):
Okay.

Jeff Shelton (03:35):
I tell her it’s a little more elaborate than that. It’s no, it’s no crystal ball. It’s, it’s just a lot of school, a lot of knowledge, a lot of experience. Knowing how traffic behaves. And then you build the models and you put, I like to call ’em “what if” scenarios. What if okay, the freeway is expanded, or if it’s closed down for a certain time, what happens to traffic?

Bernie Fette (03:59):
Okay. Another way I was thinking of asking about this that might help people understand modeling. Are there any modeling parallels in other industries, you know, other than transportation where there are people like you whose job it is to predict outcomes or impacts, maybe financial industry, real estate, things like that?

Jeff Shelton (04:21):
Yeah, I mean, first thing that comes to mind to me is the weatherman. You know, meteorologists, they predict the weather based on certain data that they get. You know, so they look at existing conditions. They have computer models that kind of give ’em all this data and then they make a forecast for the next day or for the next week. So they can kind of tell you is it gonna rain or what’s the percentage of chance of rain? Um, what the temperature’s gonna be like. Is it gonna be sunny? Is it gonna be cloudy? Right. You know, so that’s kind of a parallel. They’re kind of making a prediction of what the weather’s gonna be like. And I’m trying to make a prediction of what traffic’s gonna be like. Another parallel that came to my mind is my wife works in a restaurant, she’s a restaurant manager. And so she needs to predict how many guests are gonna come to the restaurant that day or that week. And with that she can predict how much food to prepare, how many people to have on staff. So she’s taking all this data to make a prediction about the future.

Bernie Fette (05:17):
Okay. In your work, you make a lot of references to what you call extreme events. What exactly constitutes one of those? Maybe you could give us some examples.

Jeff Shelton (05:29):
There could be a lot of different kinds of extreme events, Bernie. You know, in Minneapolis, Minnesota, the bridge collapsed there.

Bernie Fette (05:37):
And that was the I-35 bridge over uh,

Jeff Shelton (05:39):
I-35 over the Mississippi River. Yeah, it was under construction. They had added weight on the bridge. It was old and it collapsed. We consider that an extreme event because how is traffic gonna operate now that that roadway, I mean, because that served, you know, a lot of traffic. A lot of cars cross that bridge every day and now that bridge isn’t there. So how does traffic gonna react? Where is it gonna go? You know, extreme events could be in all kinds of forms. It could be a tornado, it could be an earthquake, flood, forest fires. A refinery explosion.

Bernie Fette (06:16):
Yeah. And could be a man-made event as well, right? A terrorist attack.

Jeff Shelton (06:21):
Terrorist attack could also be considered an extreme event.

Bernie Fette (06:26):
I’m curious, the example you gave about the I-35 bridge in Minneapolis, did somebody model that disaster before it happened? Or was that an example of an event that surely could have used a bit of modeling before it happened?

Jeff Shelton (06:41):
Well, all cities model their traffic. So they have a model of that traffic. Now, they did not predict what was gonna happen. They didn’t know that bridge was gonna collapse, so they didn’t have that. So they modeled it afterwards. To see what the impact of traffic would be without that bridge in place. So, okay, where is traffic gonna reroute to basically, and how bad is it gonna be at these other routes?

Bernie Fette (07:02):
Yeah, that sounds like they really could have benefited from some of the work that you and your colleagues do.

Jeff Shelton (07:07):
Yes. If they had done some sort of advanced, what-if scenarios of these events happening, they could have maybe had some more contingency plans in place for rerouting traffic.

Bernie Fette (07:19):
Not to overlook the fact that in the wake of that disaster, the relevant operating agencies surely did move quickly to fix the situation and to replace that bridge.

Jeff Shelton (07:31):
Yeah, I think it was up and running in about a year. I mean, because of the amount of traffic that goes through that bridge, it’s heavily used. So they needed that thing up and running soon. So yeah, almost immediately they started the process of reconstruction.

Bernie Fette (07:45):
Let’s stay on what-if scenarios. You apply that phrase to when you tackle one of those projects, how do you go about it? I think you told me when we spoke about this before, that you start with a computer model of a particular city or a region. Can you walk us through the steps that you take? Maybe even using a specific example, how do you go about your work?

Jeff Shelton (08:11):
Okay, so I’ll, I’ll give you an example. We had our MPO ask us about an extreme event on the border. U.S.-Mexico border. I live in El Paso, so Juarez, Mexico is right across the street. And we have four points of entry. One of them is in Sants Teresa, New Mexico, but that’s right down the street. So we consider that part of our neighborhood. We have four ports of entry that have traffic going back and forth throughout the day. And of course, CBP has to inspect every vehicle that comes in. So of course there’s a long line of cars waiting to come in, so they’re all congested. So our MPO asked us, well what happened if the bridge in the center, it’s called Bridge of the Americas or BTOA for short — What if there was an extreme event and we lost that bridge? And you know, they’re interested in the economic impact, but before you can do that, you have to do it from a traffic perspective. So they wanted to know what would happen short term and long term. And so what I do is I actually have to build a computer simulation of El Paso and Juarez together in the computer. We have a planning model in El Paso that our MPO builds and then in Mexico, in Juarez, their MPO also has a planning model that they use. So what we did is we kind of married those two models together to have one big model. So traffic goes back and forth across the bridge cuz the, the models are basically truncated at the border. Okay, okay, you have the bridge, but you don’t have traffic on the other side for both models. So what we did is we actually married ’em, so they talked to each other.

Jeff Shelton (09:41):
So traffic flows back and forth. So once we built the model and we calibrated it, when calibration means your model kind of simulating actual traffic conditions and it looks realistic, that’s what calibration means. So once we do that, then we ran a scenario of shutting down the BOTA bridge northbound, and we did it for an extended period of time and they wanted to know what was the impact. So of course, short term, like the day that the incident happened, you know, traffic was just chaotic. It was a gridlock around the port of entry because it was closed, people couldn’t get across. It was basically just gridlock all around the port. But after about a month of being closed, people know it’s closed. So they start rerouting to other bridges. And that’s where it becomes interesting because people are looking for their shortest route, time dependent route to get to where they’re going. All people kind of do that. You want to get there as quickly as possible. So that’s what the model’s trying to do. It’s looking for a shortest path to get across the bridge. And since the border bridge is closed, they have to reroute to other bridges. And there we can see very apparently whether or not the bridges can handle the excess traffic. And we saw very clearly that it couldn’t.

Bernie Fette (10:46):
Right. And in that case, most of the traffic that is at the border crossing is trying to come into the United States? Or is it a pretty even split between the two?

Jeff Shelton (10:56):
Uh, it’s pretty even split going back and forth. And I just heard a stat today that 70 percent of the vehicles crossing are Mexican. They come into the U.S. to work and then they cross back to go home. And then 30 percent are U.S., so.

Bernie Fette (11:12):
That’s a good example. Did you have another one that you might like to share?

Jeff Shelton (11:16):
Yeah, we’ve been thinking about modeling something near the coast. Texas routinely gets hit by hurricanes, you know, in the Gulf. And those are very large extreme events. Those are not instantaneous, like an earthquake that may shut down the bridge cuz it collapsed. This is more, you have a little bit of advanced warning, but hurricanes move. So where you tell people to evacuate from and when you tell ’em to evacuate is, you know, it’s not an exact science, but we’re trying to use these models to give as realistic as possible, how people would react given this condition, this hurricane, especially people that live closest to the coast need to evacuate. The question becomes, do you evacuate them earlier than people further inland? Where do you evacuate ’em to? Do you evacuate ’em inside the city in someplace safe? Or do you evacuate ’em out of the city? Do you open up roadways? I don’t know if you’ve heard of contraflow lanes, Bernie. Contraflow lanes has basically opened up the opposite direction of traffic to the freeway. So you have northbound and southbound in Houston and I-45, you open up the southbound and you turn it around, you make everything northbound. So you have double the capacity to go out just to get everybody out of the city as quickly as possible. Right. So we can do that with modeling also.

Bernie Fette (12:35):
Right. To run from the storm. Okay. Right. Yeah, that’s a particularly interesting example because you had talked earlier about the parallel of weather forecasting, so it sounds like you would be making your predictions on top of another set of predictions. Correct.

Jeff Shelton (12:51):
<laugh>.

Bernie Fette (12:52):
So just to throw an added challenge in there, can you talk a little bit about the operational impacts that you’ve helped to predict? And I guess specifically what I’m asking about is what that means in terms of people having to change commute routes and also what those impacts might mean for shippers for people in commercial enterprises.

Jeff Shelton (13:15):
I live in El Paso and Interstate 10 runs right through the heart of El Paso. The Interstate goes coast to coast, actually from California to Florida, but it goes right through the heart of El Paso. And this section in El Paso is about 60 years old. I mean, the freeway, you know, everything has a shelf life, right? So they are about to reconstruct I-10 and they’re gonna do it in sections. And the first section they wanna do is around the downtown portion of El Paso and I-0, and they have three different design alternatives. They want to know which one performs the best. So we take their designs and we put ’em in the model and we run all three scenarios to see how traffic reacts, where people are going. If the roadways that they change, are they reconfigured? Can they handle the traffic now that it’s rerouting?

Jeff Shelton (14:15):
And also some of the construction may not be warranted. And let me explain that a little further. So they have the design drawings, and at one interchange they have those things, those Texas U-turns, you know, where you just, you can, yes. You don’t have to go through a light, just make a U-turn real quick. And so when I’m running this model for El Paso and I 10, I found actually the one right by our office right here. Nobody was taking those U-turns, there was virtually no traffic. I mean, zero on two of those scenarios. And so when I’m gonna present to the DOT Department of Transportation, they need to know that because why are you gonna spend millions of dollars on these U-turns? If nobody’s gonna use it, wouldn’t that money be better spent somewhere else?

Bernie Fette (15:00):
Right, right. Can you talk a little bit more about what your work is intended to lead to? You talked a little about contingency plans. Can you talk a little about what might be included in those contingency plans?

Jeff Shelton (15:15):
I guess contingency plans for regional stakeholders. So if there is an extreme event, you have to be able to inform the public on which routes to take.

Bernie Fette (15:28):
Uh-huh,

Jeff Shelton (15:28):
You know, what’s closed,

Bernie Fette (15:30):
Right.

Jeff Shelton (15:31):
So if we can use a simulation model to predict what would happen if an extreme event occurred, then we can give that to, for instance, the city of El Paso. They may be able to retime some signal timings that are not adequate for this rerouting. You know, if there’s an accident on the freeway and people are all diverting off of one ramp, they’re gonna have to go through this intersection. You may have to re-time and add additional green time on that phase just to get all that traffic exiting the ramp. You know, so it’s just, you know, helping them make decisions in an orderly fashion just to help people get to their destination quicker.

Bernie Fette (16:07):
It sounds like on top of a certain number of uncertainties about things that you can’t really control, you’ve also got the fact that we’re talking about people — whose actions can not always be as predictable as you might like. But I guess you in that case, kind of have to go on patterns, kind of like you mentioned with those Texas U-turns and how you noticed that even though the option was there, there were a lot of people who weren’t taking advantage of that particular option.

Jeff Shelton (16:35):
Yeah. One of the jokes that we have as modelers, we say the hardest thing to model is human behavior.

Bernie Fette (16:41):
Yeah. I’m curious about any surprises that you might have encountered in your work. Have you ever gone into a modeling task, a predictive task, expecting to see one outcome and instead seeing something entirely different?

Jeff Shelton (16:58):
Yeah, we did. This was actually a research project we did seven, eight years ago, maybe a little bit longer than that. And it had to do with the new autonomous vehicles coming out now, autonomous and connected vehicles. So the cars talk to each other, they inform each other at the distance, their speed. It’s called cooperative adaptive cruise control. Right now cars have adaptive cruise control, ACC and then you add that C on their cooperative. Adaptive cruise control is basically the cars talking to each other and not just, you know, reacting to traffic conditions. If it sees traffic ahead, it slows down by itself. No. This is actually cars talking to each other, telling ’em their speed, their location, their direction, everything. And so the question was, if we have this technology in the future, will it help alleviate congestion? And we thought, yeah, it probably could. Let’s model it and find out. We’ve modeled several different scenarios. Each scenario, we increased the number of connected vehicles in the model. And the more connected vehicles that we simulated, it seemed like traffic conditions got worse. Okay. And we thought they would get better. We thought the speeds would improve. The travel times would shorten, and it was the total opposite. We saw that the connected vehicles messaging would send a message back to another connected vehicle behind it that in turn would broadcast a message to a vehicle behind that one. And then the message just propagates back further and further. And the more vehicles that are connected sending this messages, the more vehicles that are slowing down. So it ended up slowing down traffic more, and that was the total opposite of what we thought would happen.

Bernie Fette (18:28):
Right. It’s a pretty sharp contrast. So you’ve got this mix, this mosh pit of different kinds of cars, latest technology in many cases, but you know, 10, 15 year old cars at the same time.

Jeff Shelton (18:40):
Yeah. And so I heard one person say that, well, we’ll just make all vehicles automated connected in the future. Well, I don’t think that’s ever gonna happen. You know, they’ve been trying to ban smoking for how long and people still smoke, you know? Yeah, yeah. If you drive a Ferrari or a 1968 Shelby Mustang, do you want it to drive itself or do you want to drive it?

Bernie Fette (19:00):
Yeah.

Jeff Shelton (19:01):
Right. <laugh>, when I’m talking about modeling like huge cities like the city of Houston, the one limiting factor right now that I think we’re encountering is model runtime. So to run Houston and you know, you have to run it over multiple iterations. And what I mean by that, you simulate it for a whole day and it looks at the travel patterns of the vehicles, and then you, you simulate it again and it’s like another day of learning. You know, like if you move to a new city and you don’t know your route to work, you try different routes until you find the one that’s best and gets you the quickest. That’s how the model works. Mm-hmm. <affirmative>. And so we’re simulating that and we do Houston about 15 iterations until we get what we call convergences. But that basically means the model’s stabilized. Okay. And that takes three days to run that.

Bernie Fette (19:52):
And that’s where you get to start noticing some of those patterns that you talked about earlier.

Jeff Shelton (19:57):
RIght. And when I say three days burning, I’m talking about three days on a very good computer. If you just use an average computer, I use a a different computer. Three days, took 12 days on that computer.

Bernie Fette (20:08):
Okay. Well, I’m glad for what you shared there because my next question had to do with what you hope to see in terms of future advancements. The technology has moved along quite a bit to help you do your work better, faster, et cetera. So then is it pure computer muscle, computer capacity that where you see the biggest needs for improvement or, or there other things that you think would improve the modeling that you and your colleagues do?

Jeff Shelton (20:37):
Well, in addition to the improvement of the model performance, but I think just the algorithms that are built into the models. You know, the algorithm that predicts shortest path from where you’re coming from to where you’re going. Okay. That’s origin and destination. That’s what we call the model. Okay. And there’s algorithms to predict that and they can always be better.

Bernie Fette (20:59):
So basically doing the work faster and doing it with more precision is what you hope to move toward in the future.

Jeff Shelton (21:06):
Correct.

Bernie Fette (21:06):
You know, for as long as I’ve known you, you’ve always appeared to be really excited about the work that you do. So what exactly is it that motivates you to show up to work every day?

Jeff Shelton (21:19):
Well, I, I love modeling. Yeah. Like I said before, my wife says I play video games for a living, <laugh>

Bernie Fette (21:25):
<laugh>.

Jeff Shelton (21:26):
Sometimes I do look at it that, I mean, um, it’s a lot more challenging than playing a video game because you’re actually predicting real-life conditions. But it is kind of neat to see your predictions come out to reality. And I’ll give you an example. So the Texas Department of Transportation asked us to do a little study out on the east side of El Paso where traffic was spilling back from the intersection onto the freeway in the afternoon. So I went out there, I looked at it, I built a model for it, and I gave them several alternative design changes they could do to help improve traffic. And they implemented those things. So it’s really cool to go out there and see your prediction and your recommendations get put to life in real-world conditions.

Bernie Fette (22:17):
Yeah. Really great example of public service, honestly. Jeff Shelton, research scientist at TTI. Jeff, thanks very much for spending some time with us, and thanks for your hard work and your enthusiasm.

Jeff Shelton (22:31):
Thanks, Bernie. I really appreciate it.

Bernie Fette (22:35):
Nearly all of us encounter roadway traffic every day, either as a driver or a passenger. Those experiences might involve congestion or roadway maintenance. At some point, it’s a safe bet that those things are going to happen, and we have at least some idea of when. But some extreme events like earthquakes and bridge failures occur without warning, and there’s no way to forecast those in either case. However, we can predict more accurately than ever before the nature and severity of their transportation impacts, along with how we might best respond to them. In the absence of a crystal ball, that might just be the next best thing. Thanks for listening. Please take just a minute to give us a review, subscribe and share this episode, and please join us again next time for a visit with Ben Edelman. And a close look at the latest trends in crashes involving pedestrians. Thinking Transportation is a production of the Texas A&M Transportation Institute, a member of the Texas A&M University System. The show is edited and produced by Chris Pourteau. I’m your writer and host, Bernie Fette. Thanks again for listening. We’ll see you next time.