CCWI2017: F145 'A Multivariate Geospatial Data-driven Approach to Descriptive Modelling of Burst Behaviour in a Small Island Context'
journal contributionposted on 01.09.2017 by T. Mackey, Stephen Mounce, Joseph Boxall, A. Cashman
Any type of content formally published in an academic journal, usually following a peer-review process.
The efficient management of water resources in the Caribbean is challenging for several reasons: weak institutional and regulatory frameworks, limited financial and technical resources, ageing pipe assets, average unaccounted-for water along with leakage rates that are approximately 50%, and limited data resources. This paper presented a methodology that used geographic information systems (GIS) to geo-spatially model 8 factors that contribute to pipe bursts (i.e. material, diameter, length, installation era, pressure, connection density, soil type, and rainfall) and incorporated these factors as input variables for data-mining using Self Organising Maps (SOM). SOMs is a data- driven technique used to explore complex multi-dimensional data even though data may be sparse or limited. The combination of GIS and SOMs allowed for the efficient use of available data and promoted knowledge discovery. Key findings revealed that pipe bursts in Barbados are mainly driven by pressure, clay soils, high rainfall, pipe age, and small pipe diameters. Although there were data limitations, results showed that this methodology can allow utilities, especially in small island states, to narrow their focus on susceptible areas of the pipeline, reduce futile efforts, maximise resources, and implement a technical tool that may be used before making any major investment decision.