Investigating long-term performance of flexible pavement using Bayesian multilevel models

Many factors affect the performance of rehabilitative treatments for asphalt concrete pavements. However, which factors have been causing the deterioration of their long-term performance is still unclear. We considered the nested and panelled structure of pavement performance data and the unobserved heterogeneity among sites using multilevel Bayesian regression models. We incorporated alligator cracking, rutting and roughness represented by the international roughness index (IRI) as performance indicators. To verify the existence of unobserved heterogeneity, we adopted an iterative modelling path by adding the per-state or per-climatic region random parameters into the models. We then based our inference of the significant factors on the chosen models that are predictive and causally sound. The results show that there existed considerable heterogeneity among different sites and climatic regions. Including per-state or per-region intercepts improved the models' predictive capacity and accounted for the unobserved heterogeneity.

Automated summary

Many factors impact the performance of rehabilitative treatments for asphalt concrete pavements. However, the specific factors causing long-term deterioration are still unclear. By using multilevel Bayesian regression models and incorporating performance indicators such as alligator cracking, rutting, and roughness, this study explores the nested and panelled structure of pavement performance data and the unobserved heterogeneity among different sites and climatic regions. The inclusion of random parameters at the state or climatic region level improves predictive capacity and accounts for the unobserved heterogeneity.