Daily Traffic Counts in the Puget Sound Region

Estimating daily traffic counts is a practical application of operational data produced by traffic management centers.

To satisfy federal data reporting requirements, state highway departments measure the usage of state and federal highways using the Annual Average Daily Traffic (AADT) metric. For an example, see the 1997 WSDOT Traffic Report. This document states that the AADT for each given location is estimated with a formula that combines "mechanical counts of 48 or 72 hour duration" with a measurement of seasonal variation.

Given the long-term storage of operational data produced by WSDOT in the TDAD database, the opportunity arises to obtain traffic counts every day, enabling a finer level of analysis and a more accurate annual average.

How does it work?

Each night at midnight, a program analyzes the previous day's data to sum the volume measurements from each of the "station" sensors. These values represent the total number of vehicles, in all mainline lanes, that have passed a particular location within a given 20-second interval. The resulting daily counts are then placed in a database table.

A 24-hour day contains 4320 20-second intervals. Ideally, that same number of 20-second counts from each sensor will be found in the database for a particular day, and summing them will result in a perfect daily count. However, that is not always the case.

One irregularity that may occur is that not all of the 20-second counts may have been received and stored in the database. Another is that the sensor itself may report that it is not functioning correctly, and therefore its measurements are not to be trusted. For each of these cases, the solution is the same. If a span of time exists for which no 20-second counts are available for either of the above reasons, a "fake" 20-second count is generated by averaging the 20-second counts before and after the gap. The fake value is multiplied by the number of 20-second intervals in the gap, and the result is added to the sensor's daily count. In this way, a meaningful daily count can be generated even if small amounts of sensor data are missing or inaccurate.

Three separate measurements are stored along with each daily count, to explain the state of the underlying 20-second data. For each sensor, let n = the number of available samples in a day, and v = the number of those samples that are valid, with v ≤ n and n ≤ 4320. Then the following useful ratios exist:

• count_quality is v/4320
• data_completeness is n/4320
• sensor_validity is v/n

TDAD Project | ITS UW Home

webmaster@its.washington.edu
(August 27, 1999)