A large network of nitrate sensors like this one, currently deployed in a small stream in an agricultural area, that provide water chemistry readings at a high frequency — monitored by computers using artificial intelligence to spot complex patterns and anomalies that would signal pollution — are the backbone of the system. Credit: Penn State.
USDA grant to fund project developing AI-powered database on water quality
Aug 26, 2024
Editor's note: This article originally appeared on Penn State News.
By Jeff Mulhollem
UNIVERSITY PARK, Pa. — Nitrate, a common chemical compound that occurs naturally and is found in plants, water and soil, can break down into molecules harmful to human, animal and ecological health and accumulate as a pollutant. Nitrate contamination in streams, lakes and estuaries is a critical problem in many agricultural watersheds, but water-quality data is limited, making monitoring stream health and making management decisions difficult, according to researchers at Penn State. To enhance available data, the U.S. Department of Agriculture (USDA) has awarded a four-year, $650,000 grant to a research team at Penn State.
The study will focus on the Upper Mississippi River Basin, the Ohio River Basin and the Chesapeake Bay watershed. The award, administered by USDA’s National Institute for Food and Agriculture, funds a new approach to understanding nitrate concentration dynamics. The proposed system will use deep learning — a subset of machine learning and computer science, and a form of artificial intelligence (AI) — to make sense of the huge volume of nitrate data collected.
The AI will be designed to detect complex patterns and anomalies in the deluge of data and generate a comprehensive nitrate database that allows resource managers to quantify hotspots of nutrient pollution, according to team leader Cibin Raj, associate professor of agricultural and biological engineering in the College of Agricultural Sciences. That knowledge will enable the resource managers to know precisely where to implement conservation practices.
“By advancing modeling, mapping and measurement via deep learning models and high-frequency sensors, managers can identify where sources and sinks of nitrogen exist, and they can determine locations where the implementation of conservation practice can provide the best return on investments,” he said.
High-frequency sensors are changing the way scientists monitor and manage water quality, noted Jonathan Duncan, associate professor in ecosystems science and management, who is co-principal investigator on the project. Those sensors, he explained, provide unparalleled insights on the time series of nitrate concentrations, enabling better understanding of fertilizer and manure applications, seasonal and precipitation events, storm event size and intensity.
“Our team will develop a hybrid, integrated modeling framework that includes field data collection, stream sensor data, and machine learning interpretation for understanding the nitrate dynamics of a region and generating continuous daily nitrate concentration data,” Duncan said. “The information it generates can be used for evaluating ecosystem health and designing more effective and targeted watershed management strategies.”
The researchers will make the data publicly available, and results will be presented to local decision makers, watershed planners and conservation district staff through well-developed collaborations with Penn State Extension.
Chaopeng Shen, professor of civil and environmental engineering, also will contribute to the project. The research was initially funded by the Penn State Institute of Energy and the Environment through its seed grant initiative.