
Smoothing Rough Waters: USU Data Scientists Use Deep Learning to Identify River Rapids - Utah State University
Smoothing Rough Waters: USU Data Scientists Leverage AI to Identify River Rapids
Researchers at Utah State University (USU) have successfully employed deep learning to create a continental-scale river image dataset, enhancing hydrologic research. This project involved both undergraduate and graduate students who collaborated on a year-long initiative that culminated in a peer-reviewed publication and a presentation slated for the 2026 Spring Runoff Conference.
The initiative originated from a seminar that facilitated a partnership between USU, the National Park Service, and the U.S. Geological Survey. Spearheaded by USU's statistician Brennan Bean, students were challenged to determine if AI could identify specific rapids in satellite images. This breakthrough could aid water managers by enabling remote inferences of river flow in locations lacking physical measurements.
Initially focusing on the collection of 3,000 images, the project grew tremendously, resulting in over 280,000 satellite images collected. Students utilized AI to train neural networks capable of isolating rivers and accurately identifying rapids. The creation of this extensive dataset supports diverse hydrologic applications such as discharge estimation, habitat assessment, and resource management.
The interdisciplinary effort included contributions from professionals and students, emphasizing the vital role of hands-on learning in practical problem-solving. The skills gained from this project have equipped students to tackle complex challenges in real-world contexts, significantly elevating their educational experience.
USU statistics doctoral student Kelvyn Bladen will present the findings at the upcoming conference, showcasing how AI can effectively transform river management practices and contribute to ecological sustainability.

