Topics: Traffic Forecasting, Transportation Forecasting
Websites and Citations:
- Atlanta Regional Commission Travel Demand Model Documentation
- Theme Music: Five Star Fall, Mercurial Girl, Magnatune.com
Episode 45 – Traffic Forecasting
Hello and welcome to another episode of Talking Traffic. My name is Bill Ruhsam and I host this podcast and its sister website, talkingtraffic.org. Today is March 3 and this is episode 46 of Talking Traffic. Today’s episode topic is Traffic Forecasting.
I know what you’re thinking…Boooooorring. And I won’t lie, this is not the most engaging topic ever. Like, you’ll probably never sit down around a dining room table with friends, open a bottle of wine and say, “Let’s talk about the traffic analysis zones near home to better engage in the discussion of future land use with our local government!” Well, unless you hang out with a bunch of transportation planners. If that’s the case, bring more wine.
Before we get into some of the details, let me give you some definitions.
Traffic Forecasting, or Transportation Forecasting, according to the Wikitionary, “…is the process of estimating the number of vehicles or people that will use a specific transportation facility in the future.” In the general context, “vehicles” means literally any vehicle: Bikes, Autos, trucks, trains, ships, airplanes, etc. So transportation forecasting means any transportation. I’m going to concentrate on the specific example of forecasting that deals with roads and highways.
What this means on a day to day basis depends on what it’s being used for. I’ll give you some concrete examples to illustrate the differences.
Ok, your local municipality may have a problem with an intersection that is obviously related to the number of people turning right. The turning traffic is backing up into the through lane and causing a congestion problem. A solution is to make a turn lane, or extend the existing turn lane. But how long should that turn lane be? Nobody wants to spend more money than necessary, so you want to extend the turn lane as long as it needs to be, but no longer. And that “need” should extend into the future to meet the future traffic that will be coming through the intersection. So a traffic forecast here would simply be to predict the future number of vehicles using the intersection and turning right.
Now let’s say you’re Walmart and you want to install a new SuperDuperWalmartCenter. Obviously, as soon as you open you’ll have a direct impact on the street right next to the Walmart, but what about downstream at the next several signals, or the hypothetical interstate interchange that’s down that road? Now we’re talking about a whole system of roadways being affected by a large installation. That requires a much more detailed and in-depth analysis to forecast the traffic that will occur.
For a third example, let’s say you’re a metropolitan region. You need to plan for the future based on your best available information. You need to know where the traffic congestion will be worst 10, 20, 30 years from now. You have literally thousands of different variables and assumptions to incorporate into a forecast model that will look through its crystal ball out into the future. Based on that model, you’ll evaluate your transportation network and (hopefully) develop a plan to meet your future needs.
These are examples of the different scales of traffic forecasting. But they all generally use the same process. The process is called the four-step transportation forecasting process. Step one is Trip Generation, Step two is Trip Distribution, Step three is Mode Choice, and Step four is Assignment.
A “trip” is a journey from an origin to a destination by a particular vehicle type. These types are referred to as “modes”. So if you walk from your doorstep to the CVS, you’ve just made a trip by the pedestrian mode. The walk back is another trip by the pedestrian mode. If you drove, it would be a trip by an automobile mode. If you took the bus it would be a trip by a transit mode.
So for step one, trip generation, we need to figure out how many trips are new, due to whatever is going on. In my first example, where we’re just looking at an intersection, there wouldn’t be any new vehicles hitting the road, per se, but there would be additional ones based on an annual growth in traffic. But, what if we knew that the same corner was going to have a new gas station constructed there? That gas station would generate incoming and outgoing traffic, or, in the vernacular, it would generate trips. This becomes much more important the wider you cast your analysis area. If you’re that walmart in my second example, you’ll need to figure out how many people are going to come and go after the Walmart is opened. You’ll also need to have some idea, if you’re thorough, what other types of development will occur because of the Walmart. This sort of crystal-ball gazing is usually outside the scope of the type of traffic study I do, but lets think like a reasonable citizen: A walmart comes in, maybe with a couple restaurants in the outlying areas of it’s parking lot. Now there’s suddenly this huge draw that other businesses will want to take advantage of. Gas stations and fast food and tire shops and other places that will gather in some of the people who are already going to Walmart.
And just to add more complexity to the mix, at the regional scale, the trip generation is conducted using what’s called traffic analysis zones. Those zones contain data such as the number of households and the types of employment in the area and the number of automobiles owned. Basic, and not so basic, socioeconomic data. It’s from those traffic analysis zones, or TAZ, that the trip generation occurs based on data drawn from surveys and other sources.
So that was three levels of trip generation. What about step two of traffic forecasting, Trip Distribution. Well, here it’s pretty simple: You have a new trip, i.e. a new driver or pedestrian about to embark on a journey. Where are they going? Are they going home? Are they going to a shopping destination? We can make these assumptions based on the surveys and socioeconomic data I mentioned earlier. It’s all about statistics. I know that a certain average number of people will leave a certain TAZ and go to another TAZ. That’s trip distribution. Of course, on the very small scale, like for instance our first example of an intersection turning movement, the trip distribution often gets mixed in with step 4, Assignment. You don’t necessarily have to do both steps separately. I’ll talk about that in a second.
Step three of the process is Mode Choice. Mode if you recall refers to the method of moving about: by automobile, by foot, by bicycle, by bus, etc. This is another step that can get smooshed into the others on very small projects, but for large analyses becomes very important. How many people are getting to the walmart by bus? Is there a transit hub nearby? that will obviously drive up the number of people taking rail or bus. Is this new generator (that’s our fancy term for a facility that creates trips, a “generator”) within walking distance of a housing area? How many of those people will choose to walk and thus create a trip not on the road, but on the sidewalk? These modal choices are very important to take into account.
The last step is Assignment. Assignment is taking steps one, two, and three, then actually assigning those trips to certain segments of facility. For example, that new driver leaving her home to go to Walmart has to use a road to get there. Which road or roads, exactly? that’s what Assignment is all about. Assignment is quite easy with small projects, and computationally demanding with large regions.
Earlier I said we could smoosh together steps two, three, and four for small projects. Let’s go back to our intersection turning lane and the hypothetical new gas station on the corner.
I’ve done step one, trip generation, and concluded that (I’m making things up here) 200 vehicles will enter and 200 vehicles will exit in the PM peak hour. By the way, that’s about a medium-sized Quicktrip or Racetrac. Let’s just take the 200 vehicles exiting for my example. I know they have to go somewhere, and I know there’s only two driveways, so the trip distribution, step 2, and assignment, step 4, happen at the same time. A certain percentage of the vehicles go, say, north away from the gas station. A certain percentage exit on one driveway and the rest on the other. From these numbers, I calculate the right and left turns out of the gas station, and the turns (or throughs) on the adjacent intersection, just depending on where the traffic is going. This all gets added to the traffic that’s already there and voila, you have your existing plus generated traffic.
But…if you’re doing this with, say, an entire county, suddenly it’s not just what is the most direct route for a vehicle to get from A to B. There are delay factors such as congested routes and signalized intersections and high speed arterials that affect the equation. This becomes an iterative process requiring complex computer software to get an answer.
Here’s one easy description of an iterative travel assignment process. You take all the trips from every traffic analysis zone to every other traffic analysis zone, which could number in the millions, you let them go by the most direct route possible, then you count up all the traffic on each of the roads. You discover that there are some two lane roads that are exceeding 50,000 vehicles per day, an absurd number that just can’t happen. So you run the model again, but you put in a delay factor on those two lane roads that make them less attractive to the computer program because they are no longer the shortest route, at least in time. You look at the results again and discover that the two lanes roads are down to a more reasonable number, but now you have another set of roads that are being assigned more volume that they could reasonably handle.
What I’m describing here is a large regional Travel Demand Model and is the be-all and end-all of the traffic forecasting four step process. Entire careers can focus around travel demand models. Dissertations dealing with the minutia of the trip assignments or other mathematical formulas have propelled academic careers. Whole businesses are focused around creating, vetting, modifying, and running these large scale models. They. Are. A.
Because a lot of money depends on them. The Travel Demand Models are where the traffic numbers come from that future projects are based on. In the Atlanta Region, where I live, the 2040 Regional Transportation Plan, which discusses projects out ot year 2040, is 60 BILLION dollars. That’s Billion with a B. And that number is far short of what is actually needed. So getting those travel demand models right, or wrong!, can have an enormous financial impact.
Of course, every travel demand model has to be vetted carefully when it’s used at the local level. Going back to our small project example of the turning lane, the travel demand model may not be showing an average daily traffic volume that quite matches what’s actually present at the time of the project analysis. Lets say that the model shows 10,000 vehicles per day, but I go out there and measure 8,000. Is the model wrong? Not necessarily. It could be dead on for the projected growth, but we need to start from 8,000 and not 10,000. If we grew 10,000 vehicles per day at 2% for 20 years, we’d have 15,000. If we start at 8,000, we’d only end up with 12,000. That can be the difference between, say, installing a dedicated turn signal or not. Those volumes are important.
So, to summarize what I just told you, there are four steps to the traffic forecasting process. Trip Generation, where we decide what traffic is new, Trip Distribution, where we decide where it’s going, Mode choice where we decide how they’re getting there, and Assignment where we actually stick them on a route. ALL of those steps are critically important to a large regional travel demand model, and some of those steps can be smooshed together when working at a much lower level.
There’s a few things I want to send you away with that relate to various bits of what I just described. The very first one is a significant warning about travel demand models. I am not an expert. I know what I know because I use their outputs, I’m generally familiar with the methods that Travel Demand Modellers use, but I could not critique one technique over the other. I have given you a more glossed over version of this than I normally do when dealing with one of our complicated traffic engineering topics. While I haven’t told you anything that isn’t true, just consider it heavily simplified. I wasn’t kidding when I said that people have built entire careers surrounding travel demand models. If you want an idea about the complexity of a regional travel demand model, I will link to the model documentation for the Atlanta regional model in the show notes. Go ahead. read it. It’s fascinating.
Another note about travel demand models. The four step procedure I’ve just described is, appropriately, called the four step trip-based model. There is a relatively new type of model called the Activity based model, however I’m not going to go into it today for reasons of simplicity, and also because I’m not sure I understand it well enough to explain it.
Another takeaway is that these models are never and can never be perfect. They rely on data inputs that may be so much fantasy after a housing market crash, or an unknown development occurs, or some other similar unknown detail emerges. They are a perfect example of garbage in, garbage out and while I don’t know any traffic forecasting or travel demand that is actually faulty, by the time the “design” year of a project rolls around, the traffic volumes may have little or no resemblance to what was projected. However, we do the best we can.
And that has been a long-winded talking traffic episode. Honestly, when I sat down with this topic I did not expect to be talking this much. It turns out, this was a longer row to how when trying to explain the core concept to a person who doesn’t know the lingo. One reason I like doing these podcasts is because it gives me a chance to practice talking to people who aren’t engineers, who don’t know the acronyms and the shorthand, who just want to be better informed on the topic of traffic engineering. I hope this helps you, and I know it helps me.
Thanks for listening to talking traffic. If you like what you heard, or didn’t, be sure to let me know by leaving a comment on the show notes or sending an email to bill at talking traffic.org.
The music you’ve been listening to is by five star fall and can be found at magnatune .com. This episode is released under a creative commons attribution non-commercial share alike 3.0 license. Feel free to distribute and/or modify this podcast, but please link back to me and to talkingtraffic.org.
Until next time, have a great week.