Evaluating Needs of Road Maintenance in Sweden with the Mixed Proportional Hazards Model

Abstract

National road databases often lack important information for long-term maintenance planning of paved roads. In the Swedish case, latent variables of which there are no recordings in the pavement management systems database are, for example, underlying road construction, subsoil conditions, and amount of heavy traffic measured by the equivalent single-axle load. The mixed proportional hazards model with random effects was used to capture the effect of these latent variables on a road’s risk of needing maintenance. Estimation of random effects makes it possible to identify sections that have shorter or longer lifetimes than could be expected from the observed explanatory variables (traffic load, pavement type, road type, climate zone, road width, speed limit, and bearing capacity restrictions). The results indicate that the mixed proportional hazards model is useful for maintenance planning because the weakest and strongest sections in a road network can be identified. The effect of the latent variables was visualized by plotting the random effect of each section in a map of the road network. In addition, the spatial correlation between road sections was evaluated by fitting the random effects in an intrinsic conditional autoregressive model. The spatial correlation was estimated to explain 17% of the variation in lifetimes of roads that occur because of the latent variables. The Swedish example shows that the mixed proportional hazards and intrinsic conditional autoregressive models are suitable for analyzing the effect of latent variables in national road databases.

Lifetime Performance of Polymer-Modified Bitumen in Swedish Roads Evaluated With Survival Analysis

“…This means that most sections reoccur several times in the data and the total carriageway length thus corresponds to more than today's total paved road network, which is approximately 84,000 km (52,000 mi). A previous study (19) found that the between-segment variation resulting from repeated observations was significant, that is, factors not included in the model affect the lifetime of individual road segments. However, these unobserved factors did not affect the parameter estimates, therefore, between-segment variation was not included in this study.…”

“…The survival models for estimation of expected lifetimes in this study were based on previous research using Swedish Pavement Management System data ( 17, 19, 24 ). The lifetime of a road is defined as the time between two maintenance treatments, where a maintenance treatment refers to a full-coverage repaving treatment on a more than 100-m-long continuous road section.…”