The impact of soils, weather and trees on water infrastructure failure.

The uninterrupted supply and reliable distribution of drinking water is fundamental in a modern society; however, water pipelines are subject to a range of operational and environmental factors which can lead to asset failure. For the privatised water-sector in the UK, utility companies are moving towards the deployment of statistical models for proactive asset management. For some companies, statistical models have facilitated the migration away from static annual burst targets, to targets which are dynamic and adjusted to observed environmental conditions. There is an increasing need for the development of accurate pipeline failure prediction models to support asset management and regulatory reporting. This thesis evaluates several quantitative measures to improve current methods of pipeline failure prediction. The aim of this thesis is to establish the impact of soils, weather and trees on water infrastructure failure and to develop a series of material-specific drinking water pipeline failure models for an entire distribution network. A quantitative assessment investigating the impact of data cleaning on the attained model error of a series of previously developed models was conducted. Material-specific variable selection and step-wise modelling approaches was used to construct a series of improved statistical models, which have an increased representation of the environmental factors leading to pipeline failure. A detailed national tree inventory was investigated for its use in enhancing pipeline failure predictions and for calculating failure rates of different pipe materials under varying soil shrink swell and tree density conditions. The value in creating separate winter and summer pipeline failure models was also evaluated, to increase representation of the highly seasonal nature of pipeline failure. Finally, a satellite approach was used to generate soil-related land surface deformation measurements across a regional area was investigated. The result is a series of enhanced statistical models for the prediction of water pipeline failure and a greater understanding into the environmental drivers leading to asset failure.