Pavement degradation: a city-scale model for San Francisco, USA

Abstract

Data from long-term systematic pavement condition surveys provide the opportunities to better understand the pavement degradation process. To provide more accurate predictions on future pavement conditions, spatial conditions are incorporated into degradation models of pavements in this paper. Long-term, city-scale pavement condition data from the San Francisco open data portal are used to test and guide model development. Spatial and non-spatial degradation models are developed and compared, with parameter estimations carried out using the Bayesian approach. Specifically, the Integrated Nested Laplace Approximation (INLA) method is used for the Bayesian regression. It was found that: (1) the non-spatial model including only coarse categories of pavement types is too simple to provide a good fit to the data; (2) for models with fine categories (individual street segments), the spatial model is more preferable than the non-spatial model due to its lower Deviance Information Criterion (DIC) and slightly smaller fitting and testing errors; (3) only the spatial model can reveal the spatial clustering of streets where high/low degradation rates concentrate.