Technical note: A low-cost approach to monitoring relative streamflow dynamics in small headwater streams using time lapse imagery and a deep learning model

Goodling, Phillip J.; Fair, Jennifer H.; Gupta, Amrita; Walker, Jeffrey D.; Dubreuil, Todd; Hayden, Michael; Letcher, Benjamin H.

doi:10.5194/hess-29-6445-2025

Articles | Volume 29, issue 22

https://doi.org/10.5194/hess-29-6445-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-29-6445-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 29, issue 22

Research article

|

19 Nov 2025

Research article |

| 19 Nov 2025

Technical note: A low-cost approach to monitoring relative streamflow dynamics in small headwater streams using time lapse imagery and a deep learning model

Phillip J. Goodling, Jennifer H. Fair, Amrita Gupta, Jeffrey D. Walker, Todd Dubreuil, Michael Hayden, and Benjamin H. Letcher

Supplement

https://doi.org/10.5194/hess-29-6445-2025-supplement

Data sets

Model Predictions, Observations, and Annotation Data for Deep Learning Models Developed To Estimate Relative Flow at 11 Massachusetts Streamflow Sites P. J. Goodling et al. https://doi.org/10.5066/P14LU6CQ

U.S. Geological Survey EcoDrought Stream Discharge, Gage Height and Water Temperature Data in Massachusetts (ver. 2.0, February 2025) J. B. Fair et al. https://doi.org/10.5066/P9ES4RQS

U.S. Geological Survey Flow Photo Explorer U.S. Geological Survey https://www.usgs.gov/apps/ecosheds/fpe

Model code and software

fpe-model v0.9.0 Jeffrey Walker https://github.com/EcoSHEDS/fpe-model

Short summary

We describe a stream monitoring method using low-cost cameras and a deep learning model. It produces a relative hydrograph (0–100%). We applied the method to 11 cameras at 8 sites and found model performance sufficient to describe floods and droughts. The models were trained on image pairs annotated by people. We examined how well people performed annotations and how many annotations were needed. We concluded this method can be used to gain new insights into under-monitored small streams.