Climate Tech for Winter Driving – Heather Reeves, University of Oklahoma

This is texturung TV. Winter driving can be challenging for even the most experienced motorists but new science and technology Innovations are making it easier to navigate this season's harshest weather conditions joining me now is Dr. Heather Reeves a research scientist at the University of Oklahoma who is working collaboratively with the National Weather Service on these Innovations, Dr.

Reeves. Thank you so much for joining me. Thank you.

I'm glad to be here. We're happy to have you on Tech strong TV. If you could tell me a little bit about yourself and a general overview of the work that you're doing.

Great, so I'm a research scientist and meteorologist by trade and I've gotten very interested in how weather impacts the transportation sector and as a result started working on that and now I lead a team of researchers of other passionate people about this same topic and so we're interested in things like whether impacts on the aviation sector on road transportation, and I'm also on Marine Transport as well. So these are the types of things that we do and we develop decision support tools by decision support tools. I mean, these are tools that lead the meteorologist from the point of here's the weather to here's the impact the weather's likely to have on people and their day-to-day decisions.

That is so important because a lot of us do hear about the weather but people don't know how to apply in business and I understand this is a new technology that you've been working on determining road conditions and predicting them as well because as we know in a meteorology, there's a big difference when you have temperature equation whether a road is going to be freezing or just a little bit slick. Can you talk more about the technology that you've been working on to determine these changing factors? Yeah, Road temperatures really the critical linchpin and whether or not you're going to have snow or ice accumulate on the road or if it's snowing and the road temperature is above freezing it'll start to melt but sometimes it just turns into slush and you still lose traction that way sometimes it melts completely into rain.

So this is really important information for decision support and or should the roads be plowed what kinds of treatment should go on the roads. We started researching this topic by looking at actual Road temperature observations. So some State dots embed thermometers in the road pavement itself.

So we know what temperature those roads are at least in those places and we discovered that there's a lot of spatial and temporal variability in that and whether or not the roads are sub freezing this critical decision Point are they sub freezing you can have one lane of traffic where the road is above freezing and then in the opposite side of the road the opposite direction, maybe it's below freezing because there's different amounts of traffic. Or shading or there's something about that side of the road that's different. So we worked.

Use these observations in concert with other data that we had to create a probabilistic tool. So this gives a probability that the road temperatures are sub freezing or that they will be subfreezing as opposed to trying to make a deterministic declaration of it is or it isn't that way the forecaster can use their best judgment and say in this circumstance. I think we need to be more conservative.

I'm going to message a more conservative Outlook or I don't think we need to worry today. So I'm going to message a more optimistic Outlook to my stakeholders. Can you explain a little bit more about the technology behind these these tools and how developers really began to break it down to make these determinations.

Yeah, we use the like I said the road pavement thermometers we use data from that over several years to do what's called machine learning machine learning is just a big pot where you just dump in the relevant data and then you allow the computer to figure out make all those teleconnections of how these various things connect to each other and they results and a certain outcome. So the the particular tool that we use is called a random forest and a random Forest provides us with a certain number. I think we had a hundred members in our forest or 100 trees if you will on our forest and each Force gives a binary prediction of yes or no, the road is sub freezing and then across this entire Forest of yes or no decisions and we declare probability.

If 50% of the trees say the road is sub freezing and 50 say it's not then it's a 50% probability. That's really interesting. Were there any challenges that that you faced along the way when using machine learning to come up with these probabilities?

Yeah data curation is a big problem for machine learning because the old proverb of garbage in garbage out for computer science certainly holds when it's applied to machine learning so going through and figuring out whether or not those Road temperature observations are robust was a real challenge for us. It's common for these things to become uncalibrated and we needed to figure out whether that was a problem or just start reporting some bizarre observations and or to be damaged during snow plow snow removal efforts, and so just going through and figuring out when is this observation good and pulling them out so that we could train with robust data. It was a real challenge.

Was there anything that surprised you and this work? I think that as a meteorologist you're certainly experienced in forecasting these conditions. But as you took a deeper dive into this project and implemented the machine learning, is there anything that you found that you thought?

Well, I never thought that it would show this in terms of the modeling. Yeah. Yeah.

I really we started looking at the output from this predictive model and comparing it to what was happening and we were running it in what's called an analysis type capacity. So it wasn't really a forecast it was this is what it is right now with all the information we have and is what it's telling us consistent with what we see when we look at traffic camera. We love looking at traffic camera pictures and our team we'll look at every week.

We're looking at terrific camera pictures. What do we see with our eye versus what is the artificial intelligence telling us? And the thing that captured our attention is the frequency of events where we're still seeing something happening on the road in spite of the fact that the road or the probability of the road being subfreezing is near zero.

So we expect the road temperatures are above freezing but there's still some apparent accumulations. We actually had that here in Norman Oklahoma yesterday the road temperatures were quite warm, but we had the these snowfall here a snowstorm and these are enormous snowflakes. I've really seen snowflakes this big and I grew up in Michigan.

So these are just monsters some of them one to two inches in diameter these large Aggregates and we were starting to see slush accumulate even on the highway outside my office window. And I think that's simply because of the nature of what was falling there was a lot of it and these are really big flakes which take a little bit more time to melt they have a lot of air around them. So they're well insulated so the frequency of that and the importance of that was something I hadn't fully appreciated until we got into this project.

That is so fascinating. I wonder if machine learning can pick up those variations like large size snowflakes that are someone unusual that you wouldn't see them as large as you were saying in everyday storms. Well, that's an interesting idea Bonnie.

We haven't looked into that particular thing because this is just if the road temperature is above or below freezing and it's not telling us anything about what's falling on the road. But you know, I know that you mentioned I think that's I think that's a good idea for research is to see can we get some idea about the character of the snow? How big are these snowflakes?

How are they partially melted by the time that they're hitting the road though? That might be an interesting application. Yeah.

Absolutely. Especially if you're looking at the temperature at different levels of the atmosphere as a data point to put in there, but it's it's amazing a work that you've done so far. You mentioned that this is going to be looked at by Department of Transportation for different cities.

Probably where this this is a big issue in Northern cities where we have that freezing mix and those temperature variations can make such a difference in terms of not just the everyday commute but getting emergency vehicles where they need to be and not having the traffic slowdowns. Do you see this research? That you're doing being implemented not just into the public sector but into the private sector as well.

Yeah, right. Now we're targeting the National Weather Service and how they disseminate that information will be up to them. They're very likely going to be disseminating the information in the form of messages either telephone calls chats or Graphics that they create especially tune for their stakeholders, like State dots public schools and the like but certainly this is something that could be licensed and used in the private sector.

And we we talked about some of the tools that you're using. How do you see the API aspect of this? Well right now this data or this tool is targeted for national weather service operations and all of the National Weather.

They're responsible for disseminating the data that they produce whether or not this will be produced and available to the public is a choice that the National Weather Service will have to make I can't really comment on that but they do dissimulate a lot of things publicly. I just don't know if this one will be and right on that edge of the public private partnership with this particular tool. That's a really good point.

Now. Do you also use focus groups in your research? Yes, in fact last winter we conducted several focus groups.

These are with National Weather Service forecasters because this is their first exposure to this kind of a thing and so we presented them with the tool. We presented them with the package or the software package in which it'll be embedded and started asking them. What do you where do you see the implications for this or the points of application in your daily operations?

When would you use it? How would you use it? What do you need in order to use it intelligently and then we're hoping it out yours to start involving other stakeholders in these conversations like the partnering dots or emergency managers.

That is that is great. Now that we're in a new year and it's we're coming into the heart of Winter. Certainly.

What are you working on now going forward as we're into 2023 with this project and with road sensors and the technology you were describing. Well now we're starting to Target The Next Step. So the first step was is the road sub freezing or not.

Now the next step is is there something that's going to accumulate on it or not. And and if so, what what will it what will it be? Like if it's freezing rain will I be able to see it or will it be black ice if it's snow will it stay as snow?

Will it melt what will happen with it? Once it once it hits that road surface or if it's rain has it rained lately if it hasn't rained for a long time, you can have this accumulation of oil and silt on the road and does that little bit of water same thing as ice can be very slippery. And so now we're getting to that point of decision.

So now we have the the information we need to know what will happen. And then now that we can say that we can say what should you do? Well, Dr.

Heather Reeves, thank you so much for your Insight and your time as we navigate the roads this winter. We'll be thinking about this valuable research that you're doing with the University of Oklahoma and the National Weather Service and how they'll be putting all this great information out there for people to stay safe. I really appreciate you joining me on techstrong TV, my pleasure.

Thank you. Thank you. We'll stay with us.

We're gonna have a lot more news coming up.