Image-based classification of stream stage to support ephemeral stream monitoring

February 17, 2026

A paper titled “Image-based classification of stream stage to support ephemeral stream monitoring” by Sarah E. Ogle, Garrett McGurk, Anahita Jensen, F. Martin Ralph, and Morgan C. Levy was recently published in Hydrologic and Earth System Sciences. This paper was also featured in the European Geophysical Union blog.

This research was funded by a Hellman Fellows Program (faculty) award, the National Science Foundation, and the U.S. Army Corps of Engineers Forecast Informed Reservoir Operations program. This work supports the “Novel Observations” goal of the Center for Western Weather & Water Extremes (CW3E) 2025-2029 Strategic Plan by creating a new method to classify CW3E field camera photos of intermittent stream levels.

Intermittent streams—waterways that flow seasonally rather than year-round—play a critical role in ecosystems and our water supply. Yet, these streams are vulnerable to climate change, mining, and urban development. In addition, intermittent streams lost federal protection under the Clean Water Act in 2023. Unfortunately, these streams are difficult to monitor, and traditional monitoring approaches are resource-intensive, making comprehensive observation networks unfeasible for many regions.

The method from this study was developed and validated at Perry Creek (Figure 1) in northern California, using a streamgage and a field camera operated by the Center for Western Weather and Water Extremes (CW3E).

Figure 1. Waterfall flowing over mossy sandstone on Perry Creek, an intermittent stream in northern California, United States (Photo taken by Sarah Ogle in January 2022). Figure A2 in Ogle et al. (2026).

The authors employ a logistic regression model to categorize stream photos as no water, low water levels, or high water levels. When compared to direct stream level measurements from August 2017 through September 2019, the image classifications generally agree (Figure 2). The image classifications were also helpful for quality control of stream level data, including indicating that stream level observations were likely artificially low from late 2017 through early 2018.

Figure 2. Observations of stream level and concurrent image classifications from field camera photos at Perry Creek from August 2017 to September 2019. Four example field camera images are shown. Field camera images were not available in 2019. From Figure 9 in Ogle et al. (2026).

To demonstrate broader applicability, the team successfully tested their approach on two additional sites from the U.S. Geological Survey (USGS) Flow Photo Explorer, suggesting that the method is transferable to other locations.

Looking forward, the researchers hope that their method could help similar efforts like the USGS Flow Photo Explorer to further the development of a low-cost, large-scale observation network for intermittent streams.

For water managers and practitioners interested in using field camera imagery, the team recommends strategic camera placement at streams critical for management objectives such as fish passage or drought planning, ensuring stable installation locations with unobstructed streambed views, maintaining consistency in camera equipment and angles, and allocating resources for long-term maintenance. The complete code and workflow for image classification are publicly available here.

Citation:

Ogle, S. E., McGurk, G., Jensen, A., Ralph, F. M, & Levy, M. C. (2026). Image-based classification of stream stage to support ephemeral stream monitoring. Hydrology and Earth System Sciences, 30(3), 709-742. https://doi.org/10.5194/hess-30-709-2026