First-of-its-Kind Monitoring Tool Uses AI to Forecast Water Contamination

It takes an armada of sentries to keep contaminants out of your drinking water. When storms stir up sediments and push them downstream, water managers have to be ready to shut off one source of water and pivot to others. This creates major inefficiencies in these complex municipal systems, especially when there isn’t much lead time.

Now, thanks to innovative work from USU researcher John Kemper, that complex task will become increasingly efficient and predictable.

A new, AI-powered tool created by Kemper and a national team of scientists uses machine-learning technology that’s been trained on data from existing water monitoring systems to predict contamination upstream of municipal water sources.

“This new predictive technology can make a major difference to the efficiency and safety of these systems across the U.S.,” said Kemper, who is in the Quinney College of Natural Resources and is lead author of the study.

The team built the tool using data from the National Water Model — a federally managed system that predicts up to a few days in advance the volume of water flowing through any given section of river or stream in the U.S.

They combined those streamflow predictions with actual measurements of water cloudiness from sensors in the river system that provides New York City’s daily drinking water. With an extensive network of sensors to monitor water flow and sediment, and episodic water-quality problems stemming from sediment, this major system was an ideal testing ground for the new technology.

“When too much sediment comes into the reservoir during or after big storms, New York City has to limit supply and modify their operations,” said Andrew Schroth, from the University of Vermont and the study’s co-lead along with Kemper.

The waterway is prone to high turbidity due to high amounts of fine grain sediment from the glacial clays, silts and gravels in the area, he said. When storms occur, stream banks erode, cutting into the glacial sediment and creating an elevated cloudiness which may linger for months, complicating forecasts and reservoir management.

The new model offers improved accuracy and more flexibility, especially during storm events.

“It can be implemented across the country and broadly utilized by folks that could use water quality forecasts in any number of applications,” Schroth said. “With the first-ever application of the National Water Model to predict water quality, we've opened a new window that can really benefit the country as a whole moving forward.”

The tool has widespread application, allowing places that face water quality issues a way to better predict threats, Kemper said.

If a water treatment plant has existing water quality data, they now have the potential to understand the impacts of an upcoming storm, providing more lead time for operational closures. Or, if an algal bloom is predicted to coincide with an upcoming storm, managers can now take informed action to close beaches or issue warnings. Farmers could optimize fertilizer application practices to minimize nutrient runoff from fields, and plan for turbidity during irrigation.

“Turning a streamflow forecasting tool into a water quality forecasting tool paves the way for increasingly available forecasts to serve community needs,” Kemper said, “and informs similar strategies for managing turbidity in basins worldwide.”

It has the potential to transform how water quality can be anticipated and managed across the nation, he said.