Adhikari et al. (2025) Development of flood detection framework integrating Synthetic Aperture Radar polarimetry and machine learning for semi-urban vegetation systems

Identification

• Journal: Journal of Environmental Management
• Year: 2025
• Date: 2025-12-06
• Authors: Ruma Adhikari, Alok Bhardwaj
• DOI: 10.1016/j.jenvman.2025.128208

Research Groups

• Department of Civil Engineering, Indian Institute of Technology Roorkee, India

Short Summary

This study proposes a novel flood detection methodology for semi-urban vegetation systems by integrating Synthetic Aperture Radar (SAR) polarimetry and machine learning. A Random Forest model, trained on a new Flood Index (FI) derived from Sentinel-1 SAR data, accurately identifies flood extents across diverse global flood events, outperforming existing methods and demonstrating high transferability.

Objective

• To develop an accurate and transferable flood detection framework for semi-urban vegetation systems by integrating Synthetic Aperture Radar (SAR) polarimetry and machine learning, specifically using a novel Flood Index (FI) with a Random Forest model.

Study Configuration

• Spatial Scale: Regional to national scale, covering flood events in Japan (Typhoon Hagibis), India (Delhi flood), and Greece (Larissa flood).
• Temporal Scale: Specific flood events in 2019 (Japan) and 2023 (India, Greece).

Methodology and Data

• Models used: Random Forest model, proposed Flood Index (FI) integrating Degree of Polarization (DOP), Linear Polarization Ratio (LPR), and Eigenvalues of the SAR covariance matrix. Compared against Otsu thresholding and Normalized Difference Flood Index (NDFI).
• Data sources: Sentinel-1 Synthetic Aperture Radar (SAR) satellite data.

Main Results

• The proposed Random Forest model achieved F1 scores between 0.81 and 0.86 and Intersection over Union (IoU) scores between 0.70 and 0.76 across the study sites.
• It outperformed Otsu and NDFI methods in all study areas, demonstrating lower False Negative Rates (0.09–0.17) and moderate False Positive Rates (0.19–0.39).
• The model exhibited better transferability across different flooded areas, indicating its potential for scalable flood management.

Contributions

• Introduction of a novel Flood Index (FI) that integrates both amplitude and phase information from SAR polarimetry, which is crucial for distinguishing smooth flooded surfaces from rough land or vegetation, unlike intensity-only indices (e.g., NDFI, VH/VV ratio).
• Development of a robust flood detection framework combining SAR polarimetry and machine learning specifically tailored for complex semi-urban vegetation systems.
• Demonstrated high accuracy and transferability of the trained model across geographically diverse flood events, addressing a critical need for scalable flood monitoring.

Funding

• Not explicitly mentioned in the provided text.

Citation

@article{Adhikari2025Development,
  author = {Adhikari, Ruma and Bhardwaj, Alok},
  title = {Development of flood detection framework integrating Synthetic Aperture Radar polarimetry and machine learning for semi-urban vegetation systems},
  journal = {Journal of Environmental Management},
  year = {2025},
  doi = {10.1016/j.jenvman.2025.128208},
  url = {https://doi.org/10.1016/j.jenvman.2025.128208}
}