Utility Week

30 | SEPTEMBER 2022 | UTILITY WEEK Analysis Building a real-time picture of bathing water quality Having scooped Digitalisation Project of the Year at the 2022 Water Industry Awards, we talked to Wessex Water about its use of AI to pore through data from Warleigh Wei. A popular wild swimming destination on the River Avon near Bath – though not yet designated an inland bathing water – Warleigh Weir has hosted an initia- tive to explore which factors in its upstream river catchment in uence water quality and the role Wessex Water assets play. While bathing water quality is measured by faecal indicator bacteria concentration, there is currently no technology capable of continuously tracking these levels in real time. As such, Wessex's director of environ- mental strategy, Ruth Barden, says gauging whether or not a site will pass muster is a "great unknown". "The Environment Agency is responsible for taking samples, but they sample only once a site has been designated – taking 20 between May and September each year, which go through laboratory analysis and their quality assurance system," she tells Utility Week Innovate. "This gives people a false sense of secu- rity because they think if it's designated, it's … ne – but actually it might not be," she adds. "If you've been swimming in May or June, you don't really want to wait to … nd out whether it's good or bad quality." However, with readily available sensors able to provide real-time insight into quality indicators such as temperature, pH, conduc- tivity, dissolved oxygen and river ow, Wes- sex Water and technology specialist UnifAI are deep into a trial to understand relation- ships between currently measurable char- acteristics and the concentration of bacteria inŠwater. Taking place between 2021 and 2023, the project has already seen Wessex install more than 50 sensors, collect water samples from 31 locations within a 10km upstream radius of Warleigh Weir, and analyse 16 di" erent determinants in the hope that as more data is collected, the company will be able to arm the public with real-time water quality noti… cations. "It's all about linking complex datasets to provide people with information when they want it," Barden explains. "The idea is that by using an array of datasets and including real-time water quality sensors and putting that into an AI platform, you can understand the relationships between di" erent datasets, … nd out which are most closely aligned with bacteriological quality, and use those as a proxy for water condition." An 'evolutionary' approach to tech The project currently uses … ve di" erent types of sensor, which feed in numbers alongside data from other Wessex assets, such as sew- age treatment works and storm over ows. Third party sources provide some 5.5 mil- lion data points – a number Barden says is increasing by 80,000 a day. "Daily we're getting more data from a greater number of sensors as they're coming online, plus from lab research and the likes of the Met O— ce, all of which is managed by UnifAI as data is automatically uploaded into its AI platform to process and learn from," she explains. Yet even before Wessex and UnifAI tackle the challenge of managing this rising reser- voir of data, Barden explains that both pur- chasing and deploying sensors – coming in an array of di" erent shapes and sizes – has been "quite tricky" and necessitated project exibility. "You need to get landowner permission to put them in – then they're all battery oper- ated and on 4G, so establishing communica- tion from the individual sensor to the base station and relaying that data on a 15-minute basis to the AI platform has all been quite complicated," she says. "Once that's hap- pened, it's a question of giving the platform time to process data and understanding where there are relationships. "We've been trialling di" erent bits of kit as they become available, meaning it's been di— cult to budget because we want to try everything to understand the robustness and applicability of it," she adds. "It has taken a lot longer to procure some of these bits of equipment than we had anticipated – some of the probes have taken upwards of six months to get. Our approach has had to be evolutionary because we've been constantly updating the type and range of information we're collecting on the AI platform as kit has become available and been deployed." Publicly sharing 'very complex' results However, aœ er wading through this chal- lenge, the project team then faces the pros- pect of sharing the results with public stakeholders who are increasingly exposed and sensitive to the issue of pollution and river quality. "Bathing water status and the sources of contamination are very complex – but they "The idea is that by using an array of datasets and including real-time water quality sensors and putting that into an AI platform, you can understand the relationships between different datasets… and use those as a proxy for water condition." RUTH BARDEN, DIRECTOR OF ENVIRONMENTAL STRATEGY, WESSEX WATER "The idea is that by using an array of datasets and including real-time water quality sensors and putting that into an AI platform, you can understand the relationships between different datasets… and use

• Cover