PHOTO BY CAMI KOONS/IOWA CAPITAL DISPATCH The Raccoon River as it flows past the Bill Riley trail in Des Moines June 4, 2026.

Researchers at the University of Iowa are building an artificial intelligence system that will find patterns between extensive NASA satellite data and Iowa water quality sensors to forecast nitrate fluctuations.

Several water treatment systems in Iowa are working with the researchers, who hope the AI system will deliver forecasts utilities can use to plan for changing nitrate concentrations in surface water sources.

Jesus Gomez-Velez, the project lead and an associate professor of civil and environmental engineering, said the nitrate modeling will help water treatment systems make decisions on blending water from different sources, operating nitrate removal systems or issuing water use reductions.

“Having a prediction beforehand of how the nitrate will look in the next seven days will allow them to actually plan better about which sources to use, but also situations where they are unlikely to meet the water quality standard,” Gomez-Velez said.

The project is led by IIHR, the university’s hydroscience and engineering unit, and is in collaboration with Des Moines Water Works and the cities of Cedar Rapids and Iowa City.

Amy Kahler, the CEO and general manager of Des Moines Water Works, said water operators “rely on data.”

“Because river conditions are influenced by both the weather and activities across the landscape, it can be difficult to predict what water quality challenges our treatment plants will face on any given day,” Kahler said in a news release. “This initiative will give water utilities a system that helps us anticipate changing conditions in water sources and make more informed operational decisions for treatment.”

Des Moines Water Works, as part of the Central Iowa Water Works regional authority, had to issue temporary lawn watering bans this summer and last summer due to high nitrate concentrations in the Des Moines and Raccoon rivers. The demand reductions allowed the utility to continue producing water that did not exceed the U.S. Environmental Protection Agency’s maximum nitrate concentration of 10 milligrams per liter, but highlighted the impact nitrate pollution can have on everyday Iowans.

Gomez-Velez said while the project is currently focused on the three utilities, he hopes in the future it could expand to provide similar information to all of Iowa’s public water supply systems. He also said the AI model will, eventually, be publicly available on the Iowa Water Quality Information System website.

“So anyone in the state can essentially just go and see how that forecasting product is changing over time,” he said.

The Nitrate Forecasting and Early Warning System will pull data from NASA Earth Observations, which catalogs things like soil moisture conditions, vegetation, atmospheric conditions and more, and combine it with the real-time data supplied by the 60 sensors in the Iowa Water Quality Information System. These sensors, some of which are operated by the U.S. Geological Survey, record nitrate, pH, dissolved oxygen concentrations, discharge rates and temperature.

“We take observations that we have now, and we have these NASA products, and we use deep learning, or artificial intelligence, to let the model learn the relationship between these variables that NASA is measuring and the concentrations of nitrate,” Gomez-Velez said.

Once the model has learned the connections between the NASA data and the nitrate concentrations, Gomez-Velez said they can shift the model to start making short term and seasonal nitrate forecasts. He said the time frame of the forecasts will depend on feedback the researchers receive from partnering utilities.

The project is funded through NASA, but Gomez-Velez said the AI model would not be possible without the data collected by IWQIS, a program that has been scrambling for a sustaining budget since its state funding was redirected in 2023.

“The reason this type of project is possible – that we can create these models – is because these networks exist,” Gomez-Velez said. “Otherwise, we will never be able to train these models.”