Yung et al. (2026) Flood monitoring: An innovative application of multisource image fusion and transfer learning

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-02-10
• Authors: Yang Hao Yung, Chen Xiuquan, Wang Yanting, Liu Yan, Yan Yi, Cheng Rongjie, Kaiyuan Yang, Xu Xinxin
• DOI: 10.1016/j.jhydrol.2026.135107

Research Groups

• College of Geological Engineering and Geomatics, Chang’an University, Xi’an 710054, China
• Institute of Desert Meteorology, China Meteorological Administration, Urumqi 830002, China
• Xinjiang Key Laboratory of Desert Meteorology and Sandstorm, Urumqi 830002, China
• Key Laboratory of Tree-ring Physical and Chemical Research, China Meteorological Administration, Urumqi 830002, China
• Field Scientific Experiment Base of Akdala Atmospheric Background, China Meteorological Administration, Urumqi 830002, China
• Power China Northwest Engineering Corporation Limited, Xi’an 710065, China

Short Summary

This study proposes a cross-sensor framework for robust water body mapping using multitemporal optical imagery, integrating relative radiometric normalization, spatiotemporally invariant feature extraction, a PSO-RF classifier, and cross-sensor sample transfer strategies to significantly improve flood monitoring accuracy and efficiency.

Objective

• To develop and validate a cross-sensor framework for robust water body mapping using multitemporal optical imagery, aiming to mitigate radiometric discrepancies and temporal drift for improved flood monitoring and dynamic water resource management.

Study Configuration

• Spatial Scale: Regional (validated in arid regions).
• Temporal Scale: Multitemporal, time series.

Methodology and Data

• Models used: Relative radiometric normalization approach, spatiotemporally invariant feature extraction mechanism, Particle Swarm Optimization-enhanced Random Forest (PSO-RF) classifier, cross-sensor sample transfer strategies.
• Data sources: Satellite imagery from Sentinel-2A, Landsat-8, Gaofen-1 (GF-1), and Huanjing-1A (HJ-1A). Spectral indices (e.g., Normalized Difference Water Index (NDWI)) and spatial-textural attributes were used as reference benchmarks.

Main Results

• The proposed framework significantly improved classification accuracy compared to traditional NDWI-based methods.
• Landsat-derived water body extractions exhibited a 0.62% to 2.10% increase in precision.
• Post-normalized Sentinel-2A and GF-1 imagery consistently yielded Kappa coefficients above 0.8.
• Classification accuracy for heterogeneous images improved by 1% to 2%.
• The framework demonstrated significant operational efficiency and detection accuracy improvements for flood monitoring, particularly in arid regions.

Contributions

• Development of a novel cross-sensor framework for robust water body mapping using multitemporal optical imagery.
• Introduction of a relative radiometric normalization approach for heterogeneous imagery to mitigate radiometric discrepancies and temporal drift.
• Design of a spatiotemporally invariant feature extraction mechanism.
• Implementation of a Particle Swarm Optimization-enhanced Random Forest (PSO-RF) classifier for enhanced model robustness.
• Development of cross-sensor sample transfer strategies to improve the interoperability of training data.
• Demonstrated significant improvements in operational efficiency and detection accuracy for flood monitoring, particularly in arid regions.

Funding

• Not specified in the provided text.

Citation

@article{Yung2026Flood,
  author = {Yung, Yang Hao and Xiuquan, Chen and Yanting, Wang and Yan, Liu and Yi, Yan and Rongjie, Cheng and Yang, Kaiyuan and Xinxin, Xu},
  title = {Flood monitoring: An innovative application of multisource image fusion and transfer learning},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135107},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135107}
}