05 May 2023
 | 05 May 2023

Technical Note: Monitoring discharge of mountain streams by retrieving image features with deep learning

Chenqi Fang, Genyu Yuan, Ziying Zheng, Qirui Zhong, and Kai Duan

Abstract. Traditional discharge monitoring usually relies on measuring flow velocity and cross-section area with various velocimeters or remote-sensing approaches. However, the topography of mountain streams in remote sites largely hinders the applicability of velocity-area methods. We here present a method to continuously monitor mountain stream discharge using a low-cost commercial camera and deep learning algorithm. A procedure of automated image categorization and discharge classification was developed to extract information on flow patterns and volumes from high-frequency red–green–blue (RGB) images with deep convolutional neural networks (CNNs). The method was tested at a small, steep, natural stream reach in southern China. Reference discharge data was acquired from a V-shaped weir and ultrasonic flowmeter installed a few meters downstream of the camera system. Results show that the discharge-relevant stream features implicitly embedded in RGB information can be effectively recognized and retrieved by CNN to achieve satisfactory accuracy in discharge measurement. Coupling CNN and traditional machine learning models (e.g., support vector machine and random forest) can potentially synthesize individual models’ diverse merits and improve generalization performance. Besides, proper image pre-processing and categorization are critical for enhancing the robustness and applicability of the method under environmental disturbances (e.g., weather and vegetation on river banks). Our study highlights the usefulness of deep learning in analyzing complex flow images and tracking flow changes over time, which provides a reliable and flexible alternative apparatus for continuous discharge monitoring of rocky mountain streams.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Chenqi Fang, Genyu Yuan, Ziying Zheng, Qirui Zhong, and Kai Duan

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on egusphere-2023-659', Anonymous Referee #1, 26 May 2023
    • AC1: 'Reply on RC1', Chenqi Fang, 14 Oct 2023
  • RC2: 'Comment on egusphere-2023-659', Anonymous Referee #2, 07 Sep 2023
    • AC2: 'Reply on RC2', Chenqi Fang, 14 Oct 2023
    • AC1: 'Reply on RC1', Chenqi Fang, 14 Oct 2023
Chenqi Fang, Genyu Yuan, Ziying Zheng, Qirui Zhong, and Kai Duan
Chenqi Fang, Genyu Yuan, Ziying Zheng, Qirui Zhong, and Kai Duan


Total article views: 627 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
441 148 38 627 29 28
  • HTML: 441
  • PDF: 148
  • XML: 38
  • Total: 627
  • BibTeX: 29
  • EndNote: 28
Views and downloads (calculated since 05 May 2023)
Cumulative views and downloads (calculated since 05 May 2023)

Viewed (geographical distribution)

Total article views: 633 (including HTML, PDF, and XML) Thereof 633 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
Latest update: 18 May 2024
Short summary
Measuring discharge at steep, rocky mountain streams is challenging due to the difficulties in identifying cross-section characteristics and establishing stable stage-discharge relationships. We present a novel method using only a low-cost commercial camera and deep learning algorithms. Our study shows that deep convolutional neural networks can automatically recognize and retrieve complex stream features embedded in RGB images to achieve continuous discharge monitoring.