Yang et al. (2025) Watershed boundary extraction from digital elevation models using RBM-SegNet

Identification

• Journal: Environmental Modelling & Software
• Year: 2025
• Date: 2025-11-29
• Authors: Huanyu Yang, Hongming Zhang, Yuwei Sun, Lu Du, Weilin Xu, NI Jincheng, Qiankun Chen, Chunmei Wang, Qinke Yang, Haijing Shi
• DOI: 10.1016/j.envsoft.2025.106805

Research Groups

• College Of Information Engineering, Northwest A&F University, Shaanxi, China
• College of Urban and Environmental Sciences, Northwest University, Shaanxi, China
• Institute of Soil and Water Conservation, Northwest A&F University, Shaanxi, China

Short Summary

This study developed RBM-SegNet, a deep learning framework, to overcome limitations in traditional watershed boundary extraction from Digital Elevation Models (DEMs), demonstrating superior accuracy compared to existing methods.

Objective

• To develop an accurate, adaptable, and efficient deep learning framework (RBM-SegNet) for watershed boundary extraction from Digital Elevation Models (DEMs), addressing the limitations of traditional methods that rely on manual thresholds and supplementary terrain features.

Study Configuration

• Spatial Scale: Focused on general watershed delineation from Digital Elevation Models (DEMs), applicable to various spatial scales, including "large-scale applications." No specific geographic area or resolution is provided in the text.
• Temporal Scale: The methodology processes static Digital Elevation Model (DEM) data; no temporal dynamics are explicitly considered in the extraction process.

Methodology and Data

• Models used:
  • RBM-SegNet (Residual Bottleneck Attention Multi-feature Fusion Network)
  • Semantic segmentation model (core component of RBM-SegNet)
  • Compared against traditional methods (e.g., those using D8 algorithm for flow direction) and existing deep learning methods.
• Data sources:
  • Digital Elevation Models (DEMs)
  • Derived terrain features: Slope, Hillshade, Aspect (used as optimal input combination)

Main Results

• RBM-SegNet significantly outperforms both traditional and existing deep learning methods in terms of accuracy for watershed boundary extraction.
• The optimal input combination for the model was identified as [DEM, Slope, Hillshade, and Aspect].
• The integration of residual connections and the Bottleneck Attention Module (BAM) effectively enhances feature transmission and suppresses irrelevant regions within the network.
• Multi-feature fusion within RBM-SegNet refines the prediction of structural details.
• A post-processing module improves the completeness and hydrological consistency of the extracted watershed boundaries.

Contributions

• Utilized [DEM, Slope, Hillshade, and Aspect] as the optimal input combination for watershed boundary extraction.
• Introduced residual connections and the Bottleneck Attention Module (BAM) to enhance feature transmission and suppress irrelevant regions in the semantic segmentation model.
• Incorporated multi-feature fusion to refine structural and detail prediction of watershed boundaries.
• Integrated a post-processing module to improve the output completeness and hydrological consistency of the extracted boundaries.

Funding

No funding information was provided in the excerpt.

Citation

@article{Yang2025Watershed,
  author = {Yang, Huanyu and Zhang, Hongming and Sun, Yuwei and Du, Lu and Xu, Weilin and Jincheng, NI and Chen, Qiankun and Wang, Chunmei and Yang, Qinke and Shi, Haijing},
  title = {Watershed boundary extraction from digital elevation models using RBM-SegNet},
  journal = {Environmental Modelling & Software},
  year = {2025},
  doi = {10.1016/j.envsoft.2025.106805},
  url = {https://doi.org/10.1016/j.envsoft.2025.106805}
}