Statistical Inference of Sewer Pipe Deterioration Using Bayesian Geoadditive Regression Model

Abstract

Several deterioration models have been developed for prediction of the actual and future condition states of individual sewer pipes. However, most tools that have been developed assume a linear dependency between the predictor and the structural condition response. Moreover, unobserved variables are not included in the models. Physical processes, such as the deterioration of pipes, are complex, and a nonlinear dependency between the covariates and the condition of the pipes is more realistic. This study applied a Bayesian geoadditive regression model to predict sewer pipe deterioration scores from a set of predictors categorized as physical, maintenance, and environmental data. The first and second groups of covariates were allowed to affect the response variables linearly and nonlinearly. However, the third group of data was represented by a surrogate variable to account for unobserved covariates and their interactions. Data uncertainty was captured by the Bayesian representation of the P-splines smooth functions. Additionally, the effects of unobserved covariates are analyzed at two levels including the structured level that globally considers a possible dependency between the deterioration pattern of pipes in the neighborhood and the unstructured level that account for local heterogeneities. The model formulation is general and is applicable to both inspected and uninspected pipes. The tool developed is an important decision support tool for urban water utility managers in their prioritization of inspection, maintenance, and replacement.