Wang et al. (2025) Flood inundation mapping with CYGNSS over CONUS: a two-step machine-learning-based framework

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-09-10
• Authors: H.T. Wang, Fangni Lei, Xinyi Shen, Qing Yang, Emmanouil N. Anagnostou, Wade T. Crow, Hyunglok Kim, Clara Chew
• DOI: 10.1016/j.jhydrol.2025.134224

Research Groups

• Department of Civil and Environmental Engineering, University of Connecticut, Storrs, CT, USA
• School of Freshwater Sciences, University of Wisconsin-Milwaukee, Milwaukee, WI, USA
• Hydrology and Remote Sensing Laboratory, USDA Agricultural Research Service, Beltsville, MD, USA
• Department of Environment and Energy Engineering, Gwangju Institute of Science and Technology, Gwangju, Republic of Korea
• Muon Space Inc, Mountain View, CA, USA

Short Summary

This study developed a two-step machine learning framework using CYGNSS bistatic reflectance observations and ancillary data to retrieve daily fractional flood inundation at a 3-kilometer resolution across the contiguous United States, demonstrating comparable performance to SAR-based flood maps and other inundation products.

Objective

• To develop and evaluate a machine learning framework for retrieving fractional flood inundation as the area proportionally covered by water, using bistatic reflectance observations from Cyclone Global Navigation Satellite System (CYGNSS) and ancillary land surface variables.

Study Configuration

• Spatial Scale: Contiguous United States (CONUS) at a 3-kilometer resolution.
• Temporal Scale: Daily.

Methodology and Data

• Models used: Random Forest (RF) model, implemented in a two-step framework (Sequential Two-Step - STS, and Parallel Two-Step - PTS).
• Data sources:
  • Cyclone Global Navigation Satellite System (CYGNSS) bistatic reflectance observations.
  • Ancillary land surface variables.
  • Sentinel-1 C-band Synthetic Aperture Radar (SAR) high-resolution flood maps (used as reference for training).
  • Official CYGNSS water mask product (for comparison).
  • Semi-empirical method-based CYGNSS product (for comparison).
  • Microwave remote sensing inundation product (for comparison).

Main Results

• The Sequential Two-Step (STS) Random Forest model significantly outperformed both the Parallel Two-Step (PTS) model and a single regressor for daily CYGNSS inundation retrievals.
• Cross-validation using a leave-one-year-out approach yielded a correlation coefficient of 0.762 and a root-mean-square-error of 0.039 between the CYGNSS inundation retrievals and the reference SAR-based water fractions.
• Consistent spatial variations were found between CYGNSS and Sentinel-1 inundated regions, indicating satisfactory model performance.
• The developed CYGNSS inundation product showed comparable performance in characterizing flood inundation at a 3-kilometer resolution and daily temporal frequency when compared against several other inundation products.

Contributions

• Development of a novel two-step machine learning framework (STS Random Forest) for robust fractional flood inundation mapping using CYGNSS data.
• Demonstration of CYGNSS's capability for daily, 3-kilometer resolution flood inundation mapping over the contiguous United States, addressing limitations of optical sensors (e.g., cloud cover, vegetation penetration).
• Validation of CYGNSS inundation retrievals against high-resolution SAR data and comparison with multiple existing inundation products, establishing its competitive performance for flood monitoring.

Funding

• Not specified in the provided text.

Citation

@article{Wang2025Flood,
  author = {Wang, H.T. and Lei, Fangni and Shen, Xinyi and Yang, Qing and Anagnostou, Emmanouil N. and Crow, Wade T. and Kim, Hyunglok and Chew, Clara},
  title = {Flood inundation mapping with CYGNSS over CONUS: a two-step machine-learning-based framework},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134224},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134224}
}