Detection of untreated sewage discharges to watercourses using machine learning

• Published in: npj clean water Vol. 4; no. 1; pp. 1 - 10
• Main Authors: Hammond, Peter, Suttie, Michael, Lewis, Vaughan T., Smith, Ashley P., Singer, Andrew C.
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 11.03.2021; Nature Publishing Group; Nature Portfolio
• Subjects: 704/242; 706/2805; Aquatic Pollution; Discharge; Earth and Environmental Science; Effluents; Environment; Flow distribution; Flow pattern; Learning algorithms; Machine learning; Nanotechnology; Operators; Pollution monitoring; Sewage; Sewage disposal; Spills; Waste Water Technology; Wastewater discharges; Wastewater pollution; Wastewater treatment; Wastewater treatment plants; Water Industry/Water Technologies; Water Management; Water pollution; Water Pollution Control; Water Quality/Water Pollution; Water treatment; Watercourses; United Kingdom > UK; England
• DOI: 10.1038/s41545-021-00108-3
• ISSN: 2059-7037

Monitoring and regulating discharges of wastewater pollution in water bodies in England is the duty of the Environment Agency. Identification and reporting of pollution events from wastewater treatment plants is the duty of operators. Nevertheless, in 2018, over 400 sewage pollution incidents in England were reported by the public. We present novel pollution event reporting methodologies to identify likely untreated sewage spills from wastewater treatment plants. Daily effluent flow patterns at two wastewater treatment plants were supplemented by operator-reported incidents of untreated sewage discharges. Using machine learning, known spill events served as training data. The probability of correctly classifying a randomly selected pair of ‘spill’ and ‘no-spill’ effluent patterns was above 96%. Of 7160 days without operator-reported spills, 926 were classified as involving a ‘spill’. The analysis also suggests that both wastewater treatment plants made non-compliant discharges of untreated sewage between 2009 and 2020. This proof-of-principle use of machine learning to detect untreated wastewater discharges can help water companies identify malfunctioning treatment plants and inform agencies of unsatisfactory regulatory oversight. Real-time, open access flow and alarm data and analytical approaches will empower professional and citizen scientific scrutiny of the frequency and impact of untreated wastewater discharges, particularly those unreported by operators.