Capturing variability in pavement performance models from sufficient time-series predictors: a case study of the New Brunswick road network

Abstract

This paper proposes the use of multi-level Bayesian modeling for calibrating mechanistic model parameters from historical data while capturing reliability by estimating a desired confidence interval of the predictions. The model is capable of estimating the parameters from the observed data and expert criteria even in cases of missing data points. This approach allows rapid generation of several deterioration models without the need to partition the data into pavement families. It estimates posterior distributions for model coefficients and predicts values of the response for unobserved levels of the causal factors. A case study from the New Brunswick Department of Transportation is used to calibrate a simplified mechanistic pavement roughness progression model based on 6-year international roughness index (IRI) observations. The model incorporates the effects of pavement structural capacity in terms of deflection basin parameter (AREA) in place of the modified structural number, traffic loading (ESAL) and environmental factors. The results of the model showed that, as expected, chipseal roads have higher as built roughness and deteriorate faster than asphalt roads. Sensitivity analysis of the deterministic (the mean predictions) part of the model showed that in New Brunswick where traffic is relatively low the environment is the most important factor.