This study presents novel methodologies for identifying untreated sewage spills from wastewater treatment plants (WWTPs) using machine learning (ML). Daily effluent flow patterns from two WWTPs, supplemented by operator-reported incidents, served as training data. The ML model achieved over 96% accuracy in classifying spill and no-spill events. Retrospective analysis identified 926 days without reported spills that were classified as potential spills, suggesting non-compliant discharges at both WWTPs. This ML approach can assist water companies and regulatory agencies in improving WWTP management and regulatory oversight.

Peter Hammond, Michael Suttie, Vaughan T. Lewis, Ashley P. Smith, Andrew C. Singer