- Posted on 24 Apr 2025
- 5 minutes read
UTS data scientists have built cutting-edge machine learning tools that predict water quality in the dams, rivers and creeks that feed into our drinking water supply.
Key points
- An award-winning artificial intelligence solution is helping provide fresh and clean drinking water for around four million residents in Sydney and Melbourne.
- The tools use real-time monitoring, weather and upstream data to predict raw water quality from one day up to one month ahead, with approximately 90 per cent accuracy.
- They have been developed by the UTS Data Science Institute in partnership with Sydney Water, WaterNSW, Melbourne Water and TRILITY.
When Sydney faced torrential rain in 2022, water infrastructure copped a hammering. At Warragamba Dam in the western suburbs, only around 10 percent of water in the reservoir was safe to be treated and cleaned. Ironically, water restrictions were imminent.
As climate change increases the chances of extreme weather events like flooding and droughts, this scenario will repeat itself.
Having a lot more, or a lot less, water available in the catchments that source our drinking water has a dramatic impact. So too do the particles, silt and contaminants that wash in from flooding and bushfires.
Water utility operators rely on regular monitoring to assess the quality of water flowing into their pipes, and adjust their filtration settings and chemical treatments accordingly.
However, going a step further and being able to accurately forecast water quality in advance could be a game-changer.
"Currently, utilities continuously test and sample for water quality but it can take up to a day to get results back from the lab,” says Professor Yang Wang from the UTS Data Science Institute.
“We’ve been exploring prediction models that can give them a day’s notice ahead of time."
"If the utilities know water quality is going to be good tomorrow, they can produce more potable water. If quality is going to be bad, they can prepare for heavier filtration and treatment.”
“Previously, we looked at predicting water quality within the pipes in the network."
"But more recently, we’ve been collecting water quality information in catchments like creeks, rivers and lakes, and using that to predict the raw water quality there before it gets into the pump.”
The ability to use machine learning prediction for varying catchment and climate impacts gives us significant opportunity to enhance continuous water quality for our customers.
The use of artificial intelligence to predict water quality is opening up opportunities for utility companies like Sydney Water and Melbourne Water.
‘’The delivery of consistent water quality to our customer is of the highest priority for our industry,” says Dammika Vitanage, Project Manager at Sydney Water.
“The ability to use machine learning prediction for varying catchment and climate impacts gives us significant opportunity to enhance continuous water quality for our customers.”
AI-driven solution for raw water quality
A new suite of machine learning tools are being used to plan and treat the water supply for around 4 million residents in the nation’s two most populous cities.
The tools use real-time monitoring, weather and upstream water quality data to predict a range of water quality indicators from one day up to one month ahead, with approximately 90 per cent accuracy.
These indicators include dissolved organic matter, colour, turbidity (cloudiness) which can indicate contamination risks and natural organic matter which gives an indication of water safety.
The tools have been developed by scientists from the UTS Data Science Institute in partnership with Sydney Water, WaterNSW, Melbourne Water and TRILITY.
“When we started, neither us nor our industry partners were sure it was even possible to build this kind of reliable prediction model based on the data available in a water catchment,” says Dr Hongda Tian, who’s also a member of the team from the UTS Data Science Institute.
They started working with Sydney Water and private utility provider Trility by looking at data from two out of the city’s 13 water delivery systems servicing around 500,000 residents at MacArthur and Nepean in the western suburbs.
The models were then developed further by partnering with Melbourne Water around two catchments in Melbourne that provide drinking water to 80 percent of the city’s residents.
“We developed catchment-independent modelling for the four catchments we studied that looked at each separate from each other. But the overall modelling strategy is the same,” Dr Tian says.
“This means our framework could not be just confined to these four catchments. It can be easily extended to other catchments not just in Australia, but around the world.”
Technology keeps water clean
Already, the toolkit is being used to help treatment plants make decisions around chemical dosing, filtration rates and reservoir management.
This data-driven forward planning allows for more efficient use of treatment chemicals and reduces operating costs.
Most importantly, it’s helping to safeguard our water supply, keeping it consistent and safe even during extreme weather events.
“The adoption of the solution will enable operation and production teams to be more efficient in providing timely changes to the planning and treatment processes, particularly when extreme events occur due to climate change,” says Martin Harris, Water Quality Lead at Trility.
The solution has already been recognised as NSW winner in sustainability category at the 2024 iAwards, Australia’s most prestigious technology and innovation awards.
It has also been finalist for R&D excellence in the Victorian Water Awards.
However, the team isn’t resting on their laurels. They’re looking to further improve how machine learning can help utilities keep clean water flowing through to customers.
“Looking at the big picture, our project has three phases. First, we looked at predictions for supply water quality, that’s the water in the pipes."
"Second – this phase of the work – is to develop software to predict raw water quality, that’s the water in the catchments,” explains Professor Wang.
“The potential final phase will be to look at combining both the raw and supply water quality to produce a unified dashboard."
"This will help our industry partners keep the water flowing through to our taps fresh and clean, whatever the weather.”
UTS research team
Professor, Data Science Institute
Senior Lecturer, Data Science Institute