Wang et al. (2026) A novel dual-polarization SAR-based method for high-accuracy 768 km river mapping in steep mountainous regions

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2026
• Date: 2026-02-28
• Authors: Xin Wang, Keren Dai, Jianming Xiang, Jin Deng, Devin Wu, Ningling Wen, Rubing Liang, Khuzaima Saeed
• DOI: 10.1016/j.jag.2026.105190

Research Groups

• State Key Laboratory of Geological Disaster Prevention and Geological Environmental Protection, Chengdu University of Technology, Chengdu 610059, China
• College of Earth and Planetary Sciences, Chengdu University of Technology, Chengdu 610059, China

Short Summary

This study proposes a novel dual-polarization SAR-based method for high-accuracy river mapping in steep mountainous regions, effectively overcoming challenges of backscatter heterogeneity, layover, and shadow distortions, achieving superior accuracy (Kappa coefficient 95.53%) compared to traditional and machine learning methods.

Objective

• To develop an optimized water extraction method using dual-polarization SAR data to achieve high-accuracy river mapping in steep mountainous regions, specifically addressing challenges of backscatter heterogeneity, SAR layover, and SAR shadow distortions.

Study Configuration

• Spatial Scale: A 768 km reach of the lower Jinsha River mainstream (25.90°N – 28.70°N, 101.56°E – 104.56°E), with elevations ranging from 250 m to 2500 m above sea level. Also tested on an 87 km reach of the Lancang River.
• Temporal Scale: Sentinel-1 GRD data (2024-03-12, 2024-03-14, 2024-03-19), Sentinel-2 Level-2A data (2024-03-11, 2024-03-14), SRTM DEM (February 2000).

Methodology and Data

• Models used:
  • Proposed method: Window-adaptive Otsu thresholding based on Sentinel-1 Dual-Polarized Water Index (SDWI), synergistic use of ascending and descending orbit SAR data, DEM-based edge detection combined with blocking elevation thresholding.
  • Comparative methods: Threshold segmentation, Otsu's method, Random Forest (RF), Support Vector Machine (SVM).
  • Water indices: SDWI = ln(10VVVH) - 8, NDWI = (B3-B8)/(B3+B8).
• Data sources:
  • Satellite: Sentinel-1A (C-band SAR, IW mode, Level-1 GRD, VV and VH polarization, 10 m spatial resolution), Sentinel-2A/B (Multispectral Imager, Level-2A BOA surface reflectance, B3 and B8 bands, 10 m spatial resolution).
  • Observation/Reanalysis: Shuttle Radar Topography Mission (SRTM) 1 Arc-Second Global DEM (approximately 30 m resolution, resampled to 10 m).

Main Results

• Achieved high-precision water body extraction for a 768 km reach of the lower Jinsha River mainstream, covering an area of 414.58 km².
• The proposed method demonstrated superior accuracy compared to traditional methods (threshold segmentation, Otsu) and machine learning models (Random Forest, SVM).
• Achieved a User’s Accuracy of 96.99%, a Producer's Accuracy of 94.36%, a False Alarm Rate of 0.51%, and a Kappa coefficient of 95.53% for the Jinsha River.
• Effectively mitigated spatial backscatter heterogeneity using window-adaptive Otsu thresholding.
• Significantly reduced layover effects by synergistically combining ascending and descending orbit SAR data (reducing omission rate from approximately 20% for single orbits to 4.06% in a test area).
• Successfully eliminated shadow interference using DEM-based edge detection combined with blocking elevation thresholding.
• Demonstrated generalizability with good accuracy (Kappa 92.84%, UA 94.82%, PA 91.50%, FAR 0.21%) on an 87 km reach of the Lancang River.

Contributions

• Developed a novel integrated technical framework for high-precision automatic extraction of river water bodies in steep mountainous regions using dual-polarization SAR data.
• Addressed and effectively mitigated the three major challenges in mountainous SAR river mapping: spatial heterogeneity of backscattering coefficients, SAR layover, and SAR shadow distortions.
• Demonstrated superior performance over existing thresholding and machine learning methods in complex, high-relief terrain.
• Provided a transferable SAR-based solution for river monitoring in complex terrains, valuable for flood risk assessment, hydropower impact assessments, and integrated watershed management.

Funding

• National Natural Science Foundation of China (Grant No. 42371462)
• Sichuan Province Science Fund for Distinguished Young Scholars (2023NSFSC1909)
• China Postdoctoral Science Foundation (2020 M673322)
• National Key Research and Development Program of China (2021YFB3901403)

Citation

@article{Wang2026novel,
  author = {Wang, Xin and Dai, Keren and Xiang, Jianming and Deng, Jin and Wu, Devin and Wen, Ningling and Liang, Rubing and Saeed, Khuzaima},
  title = {A novel dual-polarization SAR-based method for high-accuracy 768 km river mapping in steep mountainous regions},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2026},
  doi = {10.1016/j.jag.2026.105190},
  url = {https://doi.org/10.1016/j.jag.2026.105190}
}