Cluster Analysis for Optimal Sampling of Traffic Count Data: Air Quality Example

Abstract

Data collection is one of the most expensive tasks in many transportation-air quality studies. Finding effective ways to minimize data collection time and cost without losing vital information is crucial. This paper presents a new cost-efficient data sampling approach to solve an important transportation-air quality problem directly affecting estimates of mobile emissions. We use hierarchical, complete linkage cluster analysis on the correlation (similarity) matrices to statistically identify groups or clusters of count locations with similar hourly profiles. Once counts have been clustered, a single count within each cluster may be chosen to represent the hourly diurnal profile of count proportions. Our results indicate a savings of approximately $500,000 for data collection of 21 count-days over 319 count locations in the Central California Ozone Study. Without the sampling optimization, there would have been little way to determine the key counts needed to both minimize data costs as well as provide sufficient and robust information for subsequent modeling.