Hundreds of previously unreported releases of raw sewage into UK rivers have been detected thanks to artificial intelligence, researchers say.
Scientists identified 926 “spill events” from two wastewater treatment plants over an 11-year period by employing machine learning.
The UK Environment Agency said it was “impressed by the accuracy” of the approach.
“We welcome any tool that prevents pollution,” the agency said.
The researchers, who published their study in the journal Clean Water, trained a computer algorithm to recognise, through the pattern of flow through a treatment plant, when a spill was happening.
Algorithms are software-based instructions for solving a problem.
The research – led by Prof Peter Hammond, also from the UKCEH – used a pattern recognition algorithm originally developed for medical genetics.
“Previously, I was using machine learning to detect subtle differences in the shape of children’s faces to help diagnose certain genetic conditions,” he said.
“Instead of mapping the 3D shape of a face, here we have the shape of the flow through a wastewater treatment works.”
While that may seem like a departure, the pattern recognition and machine learning approach works in the same way.
As the name suggests, pattern recognition is a way of using computing to detect regular or repeating elements in data. Machine learning is an approach to detecting those patterns using algorithms that improve automatically, through experience and through the analysis of data.
The researchers spent years gathering data about flow rates in two treatment plants – teaching the algorithm to recognise the “shape of the flow” when a plant was operating normally and when it was spilling untreated wastewater.
“It builds up knowledge and then you test it,” said Prof Singer. “You basically give it all the data and say, ‘can you find the spills’?”
Using 11 years of flow data from two plants, which the researchers did not identify in their study, the algorithm “recognised” 926 cases where untreated sewage was being released for at least three hours.
The researchers say that water companies around the UK could put a similar approach in place at any plant to detect “spills that appear to be going unnoticed and unreported”.
The Environment Agency agreed there were “good opportunities should water companies wish to consider the model as a planning tool to help manage pollution and prevent incidents from occurring”.
Prof Singer added: “I have spent my career understanding the effects of pollution on our environment and to remedy the problems, or at least, understanding better to inform people who make decisions.
“And we need to sort out the raw sewage problem that we have in the UK.”
Ms Colvin added: “Fixing the problem is not going to be easy because it’s going to require extensive investment in both old and new sewerage infrastructure and rethinking how we manage rainfall runoff in our towns and cities.
“But we cannot afford the polluting impact on our waterways if we want to be able to use them safely for recreation, and if we want to enable a truly green recovery that brings back wildlife into our rivers.”
In its response to the study, the industry body, Water UK, told BBC News it would be investing £1.1bn over the next five years “in improving storm overflows and wastewater treatment works”.
“Many companies are already employing AI techniques in order to manage their assets,” a Water UK spokesperson added. “We will continue to explore new and innovative techniques to enhance the health of rivers further.”
