Identifying the Effects of Soil and Climate Types on Seasonal Variation of Pavement Roughness Using MML Inference

Abstract

Pavement roughness is a common measure of pavement distress and one regularly measured by road authorities. While permanent pavement deterioration that equates to increased roughness is commonly modeled, cyclical or seasonal variations are often not included. While these variations may be small, they may be important when alternate pavements are compared directly for performance. We propose that seasonal variation may be described by partitioning the data into groups that are modeled as a segmentation problem. We developed a minimum message length (MML) segmentation tree (MMLST) criterion for partitioning and segmentation of the data. We performed simulated comparisons comparing common segmentation criterion (MMLST, maximum likelihood, Akaike information criterion, and Bayesian information criterion) and conclude that MMLST is the preferred criterion. MMLST assists in answering the following questions. First, is the observed segmentation pattern due to seasonal variation or merely random scatter? Second, given evidence of seasonal variation, what type of segmentation pattern should model these trends? Furthermore, does the interaction of climatic and soil conditions appear to affect this variation?