An application of logit modeling to the classification of network links for hourly traffic patterns in emission inventories

Hourly-gridded emissions inventories are required in photochemical air quality modeling. Travel demand model network and traffic count data from monitoring sites have been used to estimate hourly peaking patterns of link-running activities for network links. Principal components and cluster analysis procedures are employed in order to identify major peaking patterns of the traffic data, and a multivariate multiple regression is performed for each pattern to determine allocation factors (i.e., 24-hr factors that represent the ratios of hourly traffic volumes to the daily total). To accommodate model links that do not have traffic count data, as an alternative to the current D-Squared method, we present a new classification technique based on Logit Modeling. The Logit Modeling technique estimates the probabilities of each link being classified to each major pattern, and then the link is assigned to the pattern with the highest probability. From the implementation of the methodology on the Central California Ozone Study (CCOS) network of year 2000, we found two major patterns in weekend data and four patterns in weekday data. Compared to the D-Squared method, the Logit Modeling classification results are more accurate in most cases.; The Logit Modeling classification technique was implemented for the San Joaquin Valley Air Basin. Based on the new classification results, corresponding link-activity data were prepared for DTIM runs for weekend and weekday scenarios. The new emissions results were compared to the CCOS inventory, for which the D-Squared technique was applied. We found that the estimated daily total emissions are virtually equal for both inventories, while hourly emissions are considerably different. During peak hours, the differences between the two inventories are bigger than 5% in most areas and even larger than 20% in some areas for weekend and weekday scenarios. With the Logit Modeling classification technique, hourly traffic volumes can be more correctly estimated for most network links, which might lead to more realistic gridded hourly emissions inventories for photochemical air quality models.