URBANA, Ill. – Predicting how water and nutrients move through agricultural landscapes is essential for protecting water quality, managing fertilizers, and preparing for extreme weather. Yet scientists have long struggled to forecast these dynamics across entire watersheds. Monitoring networks are sparse, and traditional models often struggle to capture the complex interactions among climate, land use, and river systems.
A new study from researchers at the University of Illinois Urbana-Champaign shows that artificial intelligence can significantly improve predictions of both streamflow and nitrogen pollution across agricultural watersheds. The study was recently published in the journal Water Research.
The research team developed a new AI system, HydroGraphNet, that combines machine learning with scientific knowledge of how water and nutrients move through river networks. The method represents a watershed as a connected system of upstream and downstream areas, allowing the model to better capture how water and pollutants move through the landscape.
“Our goal is to bridge the gap between traditional hydrologic models and modern artificial intelligence,” said first author Jie Yang, a researcher in the Agroecosystem Sustainability Center at Illinois. “By embedding scientific knowledge into AI models, we can make predictions that are both more accurate and more consistent with how real watersheds behave.”
Unlike many previous AI models that treat each watershed as a single unit, the new approach can generate daily predictions for individual sub-watersheds across an entire river network. This allows scientists to identify where water and nitrogen pollution originate and how they move downstream.
The Illinois team tested their model in the Upper Sangamon River Basin, a largely agricultural watershed in central Illinois dominated by corn and soybean production. The AI system was trained using both existing monitoring data and results from hydrologic simulations, allowing it to learn patterns from observations while also incorporating scientific understanding of water movement.
The results showed that the new approach substantially improved predictions of both streamflow and nitrogen export, especially in areas with limited monitoring data. The model also revealed clear seasonal patterns in nitrogen losses, with large pulses of nitrogen leaving fields during wet periods when runoff and drainage rapidly carry nutrients into streams.
“Hydrologic systems are inherently connected,” said Bin Peng, a professor of crop sciences at Illinois. “By explicitly representing those connections in an AI framework, we can better understand how upstream conditions influence downstream water quality.”
Beyond improving predictions, the researchers say the approach reflects a broader shift in environmental science. “Artificial intelligence is rapidly transforming Earth system science,” said Kaiyu Guan, founding director of the Agroecosystem Sustainability Center. “By combining physical understanding with AI, we can generate high-resolution predictions of water and nutrient dynamics across large landscapes.”
Such capabilities could help scientists and policymakers design more targeted strategies to reduce nutrient pollution, manage water resources, and adapt agriculture to a changing climate. By providing detailed daily predictions across entire river networks, the new system could support next-generation monitoring and decision tools for sustainable watershed management.
Why this matters: Nitrogen runoff from agricultural landscapes is a major contributor to the Gulf “Dead Zone.” Better information about when and where nitrogen leaves fields could help farmers and policymakers target conservation practices more effectively.
The paper, “Knowledge-guided graph machine learning for spatially distributed prediction of daily discharge and nitrogen export dynamics,” is published in Water Research [DOI: 10.1016/j.watres.2026.125613]. The research was supported by the U.S. Department of Energy, the National Science Foundation, and the U.S. Department of Agriculture.
For more information, contact:
Bin Peng, Assistant Professor
Department of Crop Sciences
University of Illinois Urbana-Champaign
binpeng@illinois.edu
Jie Yang, PhD
Postdoctoral Research Associate
Department of Crop Sciences
University of Illinois Urbana-Champaign
jieyang7@illinois.edu
Kaiyu Guan, Professor
Department of Natural Resources and Environmental Sciences
University of Illinois Urbana-Champaign
kaiyug@illinois.edu
