Improving Prediction Accuracy in Mechanistic–Empirical Pavement Design Guide

Abstract

Model calibration plays a fundamental role in the implementation of the mechanistic–empirical pavement design guide. The data used in the default calibration effort, which were afforded by the Long Term Pavement Performance (LTPP) database, have a network-level inference space. As implementation proceeds, state highway agencies may be inclined to calibrate at a local network level. However, with a focus on the calibration data set to local project-level conditions, model prediction error can be reduced further. Under this study, Nebraska Department of Roads Pavement Management System data were used to calibrate two design guide smoothness models at the local project level. The focused data set was categorized by annual daily truck traffic and surface layer thickness. Results showed that project-level calibrations reduced default model prediction error by nearly twice that of network-level calibration. This study offers a window into the accuracy that can be achieved with local focus calibrations of design guide prediction models.