Bayesian estimation of classified mean daily traffic

Classified mean daily traffic (MDT) is an important input for modern highway and pavement design methods. Using Bayesian decision theory, an algorithm for computing the probability of a match between a short count site, and each of a set of permanent counting stations showing distinct trends, was developed. Using vehicle classification data collected at long term pavement performance project (LTPPP) sites in Minnesota, this algorithm was tested, and a very reliable assignment was found using a 12 day sample, consisting of a one-day classification count from each month of the year, while ambiguous assignment was seen for simple two-day samples. In order to reduce estimation error arising from uncertainty concerning the correct adjustment parameters, a Bayes estimator of classified MDT was developed as a weighted average of the expected MDT given the sample and a set of adjustment parameters, with the posterior classification probabilities providing the weights. The effects of sample size, age of adjustment factors, and the sampling time on the accuracy of the estimates were studied. An evaluation of two-day classification count samples indicated that for estimating MDT by vehicle class, these samples should be taken between May and October, and between Tuesday and Thursday. In this case, the estimation error of Bayes estimates of classified MDT are about 10–12% on average and within 26%, 95% of time.