Loading...
Quantification of Non-Perennial Stream Flow using Time-lapse Photography and Machine Learning
Wilhelm, Jessica
Wilhelm, Jessica
Citations
Altmetric:
Abstract
While methods for measuring perennial streamflow are established in ecohydrological studies, methods for measuring non-perennial flow are still in development, and remain vastly underexplored. Understanding flow patterns in non-perennial streams requires new methods including quantifying sustained wet-up, long periods of low flow, and intense bursts of stormflow. In Kansas, over half of the streams are non-perennial due to the stark contrasts between wet- and dry- seasons. Here, we assessed the accuracy and suitability of an ecohydrologic method that combines time-lapse photography with machine learning techniques (i.e., gaugeCam) to quantify non-perennial streamflow. Our overarching question is, What conditions optimize the utility of ground-based time-lapse imagery and machine learning for quantifying non-perennial streamflow rates and connectivity? We installed trail cameras at the Konza Prairie Biological Station near Manhattan, Kansas, and translated their images into stream connectivity metrics using GaugeCam Remote Image Manager-Educational AI (GRIME AI), and built rating curves using GaugeCam Remote Image Manager-Educational 2 (GRIME2) and discharge data. We found that hydrologic metrics derived from gaugeCam images were useful for quantifying stream connectivity, providing a data-intensive technique to address the technical hurdles with assessing water quantity through space and time. Thus, time-lapse photography and machine learning serve as a novel application for quantifying long term changes in the physical dynamics of streams, including non-perennial systems.
Description
These are the slides from a presentation given at ASLO on 03/29/2025.
Date
2025-03-29
Journal Title
Journal ISSN
Volume Title
Publisher
University of Kansas
Research Projects
Organizational Units
Journal Issue
Keywords
Machine learning, Artificial intelligence, Streams, Streamflow