Improving the Prediction of Annual Average Daily Traffic for Nonfreeway Facilities by Applying a Spatial Statistical Method

Abstract

Annual average daily traffic (AADT) data are important for various transportation research areas, including travel model calibration and validation, pavement design, roadway design, and air quality compliance. Specifically for model calibration and validation in long-range transportation planning, a base-year model requires numerous count locations across the study region. Sometimes count data for the lower classified roadways are not readily available. Detailed models require traffic counts for not only higher classifications of roadways such as freeways and arterials but also collector and, in some instances, local roadways. To predict AADT better for desired count locations on nonfreeway facilities, spatial dependency is considered. The theory behind the use of spatial dependency is that the traffic volume at one monitoring station is correlated with the volumes at neighboring stations. The spatial regression model takes into account both spatial trend (mean) and spatial correlation, which is modeled by a geostatistical approach called kriging. The spatial regression model is applied to AADT in Wake County, North Carolina. Results indicate that the overall predictive capability of the spatial regression model is much better than that of the ordinary regression model. In addition, the urban area has more reliable prediction than the rural area. Finally, the spatial regression model is expected to provide better predictions for desired count locations where no observed data currently exists due to budget limitations.