Modeling Failures in Water Mains Using the Minimum Monthly Antecedent Precipitation Index

Abstract

This paper examines the performance of the minimum monthly antecedent precipitation index (MMAPI) and other covariates in modeling pipe failures using the nonhomogeneous Poisson process (NHPP). Monthly time lag is introduced into the MMAPI to allow predictions to be made using past values. The influence on pipe failure for most covariates was as expected, except for the ageing factor. It is possible that other significant time-dependent factors have been omitted in the model, which led to underestimations of the ageing factor for asbestos cement pipes and for models calibrated using a longer training period. The influence of time lag in the MMAPI was also investigated for practical use. The predictions for the total number of failures from models with 1-month (T1) and 2-month (T2) time lag in the MMAPI were not as accurate as the model with no time lag in the MMAPI (T). The T1 and T2 models allow predictions to be made for the next 1 and 2 months, respectively. This is useful for small-diameter pipes that are often repaired after reaching a certain threshold. The predictions can be considered as the number of repair jobs in the future and allow sufficient resources (e.g., workers and materials) to be allocated in time, especially for networks with failures influenced by climate factors.