Seasonal Time Series Models to Support Traffic Input Data for Mechanistic–Empirical Design Guide

Abstract

To predict the performance of pavements, traffic volume should be forecast for the design life. Two aspects associated with traffic volume forecasting are incorporated into the NCHRP Project 1–37A mechanistic–empirical (M-E) design guide: long-term annual growth and short-term seasonal variation. Annual traffic growth can be accounted for by either a linear or a compound trend model. Seasonal traffic variation is captured by monthly adjustment factors. Class-specific traffic forecasting improves design accuracy but implies increased efforts in traffic data input and computer running time. Long-term and short-term estimations are performed independently and separately. This study demonstrates that the characterization of traffic volume is better achieved by estimating growth trends and seasonality simultaneously. Because of the data-intensive nature of the M-E design guide, any efforts toward facilitating data processing and input should be pursued. In this paper, seasonal time series techniques are applied to integrate both the long- and short-term aspects into one model. The growth trend term in a time series model predicts traffic volume as a function of time in years, and trigonometric functions are used to capture seasonal variations. The results show that the seasonal time series technique is both effective and efficient. In addition, similarities and differences regarding traffic growth trends and seasonality among different vehicle classes are easily identified through the parameter estimates. Finally, the study statistically demonstrates that long-term information generating growth trends and short-term information generating seasonal variability can be combined to predict traffic for M-E design in a more efficient way.