Predicting Day and Night Traffic Volumes on Rural Roads for Statistical Road Safety Modeling

Abstract

Statistical road safety modelers have commonly used some combination of segment length and traffic volume as measures of exposure. Traffic volume is usually represented in statistical road safety models with annual average daily traffic (AADT), which turns out to be a highly influential right-hand-side variable for regression models of expected crash frequency. Models that use AADT alone do not explicitly capture differences in traffic volume patterns throughout the 24-h day; this factor can have significant effects on safety performance. This study adds to the existing literature by developing more disaggregated estimates of traffic volumes for day and night conditions in rural areas and modeling road safety using those estimates. The proposed approach is demonstrated with the data from all automatic traffic recorder stations in Utah, with subsequent safety analysis focused on rural two-lane horizontal curve segments. Universal kriging, along with multiple covariates, proved to be an effective spatial technique for predicting day and night traffic volumes at unmeasured locations using data from permanent traffic-recording stations. Predicted day and night traffic volume estimates were incorporated into statistical road safety models of the expected number of crashes on rural two-lane horizontal curves to determine how this new information affected safety model estimation results. The parameter estimate for the predicted ratio of night-to-day traffic volume was positive and statistically significant and verified the hypothesis that horizontal curves with higher proportions of traffic at night were expected to experience more crashes than similar curves with higher proportions of traffic during the day.