Laboratory channel widening quantification using deep learning

Authors: WANG, ZIYI - Beijing Normal University; LIU, HAIFEI - Beijing Normal University; QIN, CHAO - Tsinghua University; Wells, Robert - Rob; CAO, LIEKAI - Tsinghua University; XU, XIMENG - Chinese Academy Of Sciences; MOMM, HENRIQUE - Middle Tennessee State University; ZHENG, FENLI - Northwest Agriculture And Forestry University

Submitted to: Geoderma
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/13/2024
Publication Date: 9/20/2024
Citation: Wang, Z., Liu, H., Qin, C., Wells, R.R., Cao, L., Xu, X., Momm, H., Zheng, F. 2024. Laboratory channel widening quantification using deep learning. Geoderma. 450(2024). Article 117034. https://doi.org/10.1016/j.geoderma.2024.117034.
DOI: https://doi.org/10.1016/j.geoderma.2024.117034

Interpretive Summary: Development of Technology to Measure Gully Evolution and Erosion Processes in Agricultural Fields: Gully erosion is the primary source of sediment in many agricultural watersheds. Agricultural management practices are often used to repair smaller gullies in fields through tillage. Although, repaired gullies often reappear following large storm events resulting in significant soil loss that degrades the productivity of these fields. An automatic measurement method to determine channel widening and soil erosion within gullies was developed. Utilizing this method, the widening of small gullies was found to produce soil erosion that is strongly dependent on the shape, slope, and soil type of the gully channel. Gullies were shown to have the largest number of sidewall banks collapse in the downstream sections of the channel while the total area of the downstream collapsed blocks decreased. Gully channels were shown to widen faster at the downstream end with increasing channel slope. This study builds upon previous studies that provides the framework for improved equations utilized by USDA erosion prediction technology by exploring automated detection and prediction of gully evolution and erosion. Advanced technology is needed by conservationist when developing management plans that applies effective conservation practices within agricultural watersheds.

Technical Abstract: Gully erosion devastates arable land and significantly contributes to soil loss in agricultural watersheds. In the presence of a less- or non-erodible layer, gully widening governs the erosion process once the gully bed incises to this layer, accompanied by failure block generation and transport. Current knowledge on channel widening, however, is limited due to the lack of robust methods to capture the rapid sidewall expansion process. Laboratory experiments were designed to simulate the channel widening process with an initial channel width of 10 cm. Two packed soil beds with a non-erodible layer and two slope gradients (5% and 11%) were subjected to the inflow rate of 0.67 L s-1. Images were captured by mounted digital cameras and automatically transformed into orthophotos. Channel edges and failure blocks were automatically detected by deep learning algorithm in a newly developed Channel-DeepLab network model based upon Deeplab V3+ platform. The procedure includes learning samples labelling, data augmentation, model construction, training, and validation. Sediment discharge and changes in channel width, geometry of channel edges, and failure blocks were measured. The results indicate strong correlation between sediment discharge and channel surface area. The slope section that witnessed the fastest channel widening rate migrated downwards as the slope gradient increased. In test 1, the total number and area of the failure blocks increased with time, while the collapse frequency of the sidewalls decreased. The initial period had the highest sidewall collapse frequency, total area and disaggregation of the failure blocks, and transport rate. Upstream had the highest sidewall collapse frequency and disaggregation and transport rate of failure block material while downstream had the highest total number of collapses. Time lag was found between sidewall collapse and sediment discharge, which increased as time progressed, attributing to the decreased runoff erosivity. Results of this study will provide methodological support for gully erosion and streambank retreat monitoring. Future work can be focused on a coupled failure and widening model to quantify the delayed response process between sidewall failure and sediment discharge.