Mo et al. (2026) Self-Supervised Reservoir Water Area Detection Across Multi-Source Optical Imagery

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-03-18
• Authors: Guiyan Mo, Qing Yang, Xu Zhou
• DOI: 10.3390/rs18060918

Research Groups

[Information not available in the provided text.]

Short Summary

This paper develops a geo-spectral feature-guided Self-Supervised Water Detection (SWD) framework for automated, multi-source optical imagery to monitor reservoir water extent. The SWD framework outperforms supervised methods, demonstrating high consistency and stable generalization across scales and regions, and accurately captures water-level fluctuations without manual labels or model training.

Objective

• To develop a geo-spectral feature-guided Self-Supervised Water Detection (SWD) framework, an automated algorithm for multi-source optical imagery, to accurately and timely monitor reservoir water extent, addressing challenges posed by dam operation and surface heterogeneity.

Study Configuration

• Spatial Scale: Large-area observations, validated across six reservoirs.
• Temporal Scale: Frequent and multi-temporal monitoring, capable of capturing water-level fluctuations and hydrological responses.

Methodology and Data

• Models used: Self-Supervised Water Detection (SWD) framework, consisting of pixel-level classification (using Gaussian mixture model parameterized by automatically derived high-confidence samples integrating spatial priors and spectral features) and object-level refinement (using superpixel-constrained region growing). Performance was compared against Random Forest and U-Net.
• Data sources: Multi-source optical imagery from three unspecified sensors.

Main Results

• SWD achieved the best performance compared to Random Forest and U-Net across 36 test cases.
• In cross-scale tests, SWD demonstrated high consistency with an Intersection over Union (IoU) ≥ 0.774.
• In cross-region transfers, SWD maintained stable generalization with a standard deviation (SD) of 0.010.
• In hydrological response assessments, SWD captured water-level fluctuations with minimal bias variation (relative error (ΔRE) < 1%).
• The SWD framework is computationally efficient, with processing times ranging from 0.49 seconds per megapixel to 1.29 seconds per megapixel on a standard CPU.

Contributions

• Introduction of a novel geo-spectral feature-guided Self-Supervised Water Detection (SWD) framework for automated reservoir water area detection.
• Effectively addresses spectral variability and surface complexity in multi-source optical imagery without requiring manual labels or model training.
• Enables automated, large-scale, and multi-temporal reservoir water monitoring, overcoming limitations of supervised methods in transferability and operational deployment.
• Demonstrates superior performance, high consistency, and stable generalization across different scales, regions, and hydrological conditions compared to existing supervised techniques.

Funding

[Information not available in the provided text.]

Citation

@article{Mo2026SelfSupervised,
  author = {Mo, Guiyan and Yang, Qing and Zhou, Xu},
  title = {Self-Supervised Reservoir Water Area Detection Across Multi-Source Optical Imagery},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18060918},
  url = {https://doi.org/10.3390/rs18060918}
}