By: Laura Schewel on August 15th, 2018

Print/Save as PDF

New 2017 AADT Metrics: It’s Time to Ditch Temporary Counts for Big Data Analytics

Big Data | Software Updates | Transportation | Validation

To quote Shawn Turner of the Texas Transportation Institute, “Using road tubes to collect traffic volume data is a proven method, but it’s an old practice and puts people in harm’s way.” Well, what if you could get high quality annual average daily traffic (AADT) counts for any roadway in just a few minutes – without leaving your office? What if those estimates were as good or better than your typical two-day tube or modeled count?

We think on-demand AADT could improve the way public agencies allocate roadway maintenance funding. Your staff could spend less time installing and monitoring sensors on dangerous roadsides – and more time improving infrastructure. StreetLight Data is making this possible with the launch of our 2017 AADT Metrics in StreetLight InSight®, our on-demand transportation analytics platform. We believe they are the most accurate Big Data derived AADT to hit the industry to date. (And we’re happy to show you why.)

In this article, I’ll argue that it makes sense to use Big Data to estimate AADT and share how we improved our AADT Metrics over the past year. I also will provide a high-level introduction to our methodology and validation work, which shows that our 2017 AADT Metrics are as accurate as temporary counts. For a detailed report, click here. Note that we will continue to update this report as additional validation studies are made public. Want to go even deeper? Watch our recorded webinar with StreetLight Data’s lead AADT and normalization Data Scientist, Dr. Christy Willoughby.

The Backstory: About Our 2017 AADT Metrics

When we launched our first AADT Metrics in beta in June 2017, we broke new ground. But we also knew that additional refinement was critical. That’s why we worked with the Texas Transportation Institute and the Minnesota Department of Transportation on a rigorous validation study. They put our 2016 AADT to the test, and we took their feedback to heart.

However, we didn’t stop with that one study. We also incorporated additional feedback from other clients and read new papers in the field. This helped us to bring in new data and machine learning techniques to make our traffic counts even better. This September, our monthly StreetLight InSight update will include our brand-new 2017 AADT Metrics.

Why Estimate AADT From Big Data?

First, AADT is the number one requested Metric from our customers. AADT is a core input to a huge array of transportation processes. For example, it goes into most travel demand modeling efforts, it’s required reporting for many federal systems like the Highway Performance Monitoring System, and it’s a precursor to calculating vehicle-miles traveled which in turn affects greenhouse gas emissions and federal funding. When agencies underestimate AADT, they aren’t able to access the funding they need and deserve – this has real-world consequences, especially in rural areas where estimating AADT is particularly expensive and time consuming.

In addition, the status quo for estimating AADT at scale is not good enough. The gold standard is leaving a well-maintained, permanent counter on each road segment. But that is overwhelmingly cost-prohibitive.

Instead, practitioners do short term counts for a day or two, then use various seasonal factors to try to expand that 24- or 48-hour count to represent an entire year. The short comings of this approach are obvious, especially on roads with lots of unusual variation. On top of being inaccurate, these short-term counts are rather expensive (especially in rural areas with longer distances), and they put staff in harms way by roads.

In sum, the whole industry would benefit from an accurate (enough), low-cost, quick, safe, and scalable solution to estimating AADT. Our conversations with clients indicated that this need is especially sharp for rural and low volume roads.

Why Update Our AADT Metrics?

We decided to upgrade our AADT Metrics for several reasons:

  1. We’ve learned a lot from our clients who used our beta 2016 AADT Metrics and shared when and where they were most and least effective. We used this feedback to guide our product development.
  2. A year has passed, so we wanted to offer more up-to-date 2017 counts.
  3. We’ve learned from the progress in the field on this topic overall and incorporated the findings of others’ research into our product development.
  4. We’ve expanded our team, so our data scientists have had more time to test out alternative algorithmic approaches and develop a larger set of calibration data (see Figure 1 below).


Figure 1: We went on a hunt across the US for well-cleaned permanent counter data and found over 2,605 counter data points. We wanted our data to be spread across the US, between small and large roads, urban and rural. The biggest challenge was finding permanent counter data for small rural roads. This map and the following charts show the locations of the counters we used. Of the 2,605 counters, we used 2,441 to train the algorithm and 164 to test it.

How Accurate Are StreetLight InSight  2017 AADT Metrics?

Our goal was to develop a Metric that was as accurate (if not more so) than modeling AADT or expanding temporary counts. We also wanted a Metric that performed better than published results from other Big Data-derived attempts. In short, we set accuracy targets for Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). We developed these targets based on literature about the accuracy of modeling and expansion techniques, as well as papers about other Big Data-driven techniques.

We trained our algorithm on a set of 2,441 permanent AADT counters from across the country. Then we tested the algorithm on a set of 164 permanent AADT counters. This test was “blind” (i.e., the results we show you only include data that wasn’t included in the calibration or training of our algorithm). You can read a lot more about this process in our white paper.

The results were great: We came close to or outperformed all of our targets. We feel that these results are from “as accurate” to “more accurate” than modeling AADT or expanding a temporary counter, and they’re also are better than other published Big Data-derived AADT results we could find.


Figure 2– StreetLight 2017 AADT for Test Data compared to Permanent Counter AADT. R2 is 0.96. No outliers were removed.

How Can I Run an AADT Metric or Validate Your Results for Myself?

Starting the 2nd week of September, if you have a StreetLight InSight account with AADT 2017 enabled, go to the “Create Projects” tab. On Project Type, select “2017 AADT.” Then choose which Zones you want to estimate AADT for. Note – Zones need to be road segments (either gate or line geometries) with “pass through” flagged, and they need to be bidirectional. In coming releases, we will add unidirectional AADT, as well as hourly and seasonal AADTs. (If you try this before our September release, your only option will be 2016 AADT [BETA]).

Figure 3: How to Run AADT Metrics in your StreetLight InSight account.

If you don’t have StreetLight InSight account with AADT, you should get one! Email your account representative or contact us here to learn more. Also, please also use that form if you’re interested in doing further validation of this Metric with us. You can also use our 2017 AADT to scale any other Metric up to an estimated count, including select link, corridor studies, internal/external analyses, origin-destination matrices, and more.

Learn More About About Our 2017 AADT Metrics

You can learn more about our methodology and validation in our white paper here. In addition, you can watch our recorded webinar here.




Your email address will not be published. Required fields are marked *