Identification of Spatial Patterns in Water Distribution Pipe Failure Data Using Spatial Autocorrelation Analysis

Identifying spatial patterns of water distribution pipe breaks can improve the understanding of the significant factors that promote the deterioration of water infrastructure systems. Such understanding can be translated into better failure prediction models or leveraged by water municipalities to improve their maintenance operations and asset management. Herein, spatial autocorrelation analysis, based on global and local Moran’s I, is proposed for the identification of failure clusters in water distribution networks. The proposed approach extends traditional spatial autocorrelation analysis by accounting for network structure in the formulation of the spatial weights and is adaptable to different levels of spatial resolution. For the studied system, the results revealed that pipe failures exhibit a significant degree of spatial clustering. The locations of statistically significant hot- and coldspots of pipe failures were identified. Characteristics of the underlying pipe network, including pipe density, material, and age, were found to be the main drivers for spatial clustering in the failure data. Beyond water distribution networks, the proposed computational approach can be applied to detect and locate patterns of spatial events in other networked infrastructure systems and to identify local network characteristics that give rise to such patterns.