A chance encounter at a Utah State University Ecology Center seminar led to a research collaboration that resulted in a comprehensive year-long AI learning project for undergraduate and graduate students. A continent-wide, publicly available river imagery dataset, a peer-reviewed paper published in a top journal, and a presentation at USU’s upcoming Spring 2026 runoff conference.
“In discussions with a group of researchers from USU, the National Park Service, and the U.S. Geological Survey, including hydrologist Christy Leonard Stegman, at a seminar, we began developing a real-world problem. It was then presented to five USU undergraduates enrolled in applied research in a machine learning and AI class,” says USU statistician Brennan Bean. “With guidance from two doctoral students in the same class, the scholars approach the problem as a hands-on semester project.”
The challenge, he said, was to use AI tools to determine whether specific types of rapids could be identified from satellite images of rivers.
“These rapids are important because there is ongoing research by the USGS on using rapids to infer river flow,” says Bean, an associate professor in USU’s Department of Mathematics and Statistics. “With this information, water managers could potentially estimate river flows remotely in locations without physical flow meters.”
The class project started as a small challenge to collect about 3,000 images and try to build a machine learning model, he says.
“We never imagined that this would develop into a sophisticated year-long machine learning project that we could share with river managers and the scientific community.”
The project team also included Bean. Leonard Stegman, adjunct assistant professor in the USU School of Watershed Sciences, and Julie Bahr of the National Park Service. Also participating were Carl Regreiter, a hydrologist with the U.S. Geological Survey, and Kevin Moon, director of the USU Center for Data Science and AI, a faculty member in USU’s Department of Mathematics and Statistics. But experts credit the students with driving the project, detailed in a paper published in the journal Jan. 22. remote sensing.
The first author is an undergraduate researcher at USU. Nicholas Brimhall. Co-authors include Bean, Moon, Leonard Stegman, Bahr, and Legleiter, as well as current and former USU students.
- Kelvin Braden, PhD student.
- Dr. Thomas Kirby, 25 years old, assistant professor at Brigham Young University.
- Cameron Swap, undergraduate student.
- Hannah Fluckiger, undergraduate student.
- Makenna Roberts, BS, 25, master’s student at Duke University.
- Kayden Hart, undergraduate student.
“Students approached this project with great energy and creativity, starting by exploring ways to automatically collect images from Google Earth, and ended up collecting more than 280,000 images,” Bean said.
From there, the students trained an AI neural network to isolate rivers in satellite images and predict the presence or absence of rapids.
“In this class, we refined an image segmentation model that can separate rivers in images and a neural network that identifies rapids in those images with fairly high accuracy,” Bean says. “The resulting dataset provides a framework to support a range of future hydrological applications, including flow estimation, habitat assessment, resource management, and recreation planning.”
Moon says that to the team’s knowledge, no one has developed a river imagery dataset of this scale.
“The images span the continental United States and Alaska,” he says. “This is one of the first researchers to look specifically at rapids.”
Bean said the undergraduate and graduate student mentors exceeded his expectations for the course.
“I thought the students did a great job of adapting to the realities of real-world data analysis,” he says. “When something doesn’t go as expected, they ask themselves, ‘What’s going wrong and what should I do?’ We didn’t set these challenges for them. As teachers, we had no idea how things were going to go. The challenges they faced and the solutions they came up with were really organic, and that made the learning process fun.”
Moon said the class developed into an effective experiential learning project for both undergraduate and graduate students.
“At the undergraduate level, students identified problems and developed solutions,” he says. At the graduate level, PhD students did an excellent job of mentoring undergraduate students, including teaching students machine learning tools and providing project guidance. It was also important to have actual NPS and USGS personnel as advisors. Students faced real-world challenges with unknown outcomes. ”
Bean said he acted as a student rather than a teacher in class.
“We didn’t teach our students, ‘Here’s the problem, implement this solution,'” he says. “Instead, we asked questions and asked students to defend their findings. If we had scripted this lesson, it would not have been as informative, meaningful, and stimulating for our students.”
Statistics doctoral student Kelvin Braden will present his team’s research during his time at USU. Spring 2026 final election meetingMarch 24 and 25 at the Cache County Event Center in Logan.
