Prioritizing Sanitary Sewers for Rehabilitation Using Least-Cost Classifiers

Abstract

Sewer rehabilitation is one control option for relieving wet-weather sanitary sewer overflows by reducing rainfall dependent inflow and infiltration. The prioritization of rehabilitation activities is based on the physical condition of a collection system, which is usually only partially known. Linear regression methods have been used to estimate the condition of the unobserved set of pipes based on relationships derived from the observed set of pipes. This method can provide unsatisfactory relationships between the sewer condition and measured independent variables, resulting in highly uncertain rehabilitation cost estimates. Discriminant analysis has been effective at weighing the costs of misclassifying pipes and deriving least-cost classification rules. Classification rules may be derived within a data-mining framework using evolutionary algorithms or with logistic regression methods. The estimated spatial distribution of deficient pipes may then be refined on an aggregated scale with a bootstrap estimate of classification error. The result is a screening and prioritization tool for providing cost estimates for rehabilitation and replacement activities. This method is demonstrated with an 8,919 pipe sanitary sewer system in Vallejo, Calif.