Xing et al. (2026) Large-scale flood monitoring based on time-series Sentinel-1 images and Z -score

Identification

• Journal: Natural hazards and earth system sciences
• Year: 2026
• Date: 2026-10-28
• Authors: Runmei Xing, Fengyuan Lu, Jun Zhao
• DOI: 10.5194/nhess-26-5299-2026

Research Groups

• School of Civil Engineering, Zhengzhou University, Zhengzhou 450001, China
• Logistics School, Henan College of Transportation, Zhengzhou 451460, China
• School of Computer Science and Artificial Intelligence, Zhengzhou University, Zhengzhou 450001, China

Short Summary

This study developed a time-series Sentinel-1 Synthetic Aperture Radar (SAR) anomaly-detection method, integrating Z-score analysis and a classification tree with auxiliary datasets, to accurately monitor large-scale floods and urban inundation. The approach demonstrated strong performance in classifying flood-affected areas in both natural and building environments during the 2021 Weihui flood event.

Objective

• To develop a robust and accurate time-series Sentinel-1 SAR anomaly-detection method capable of simultaneously monitoring large-scale floods and urban inundation, addressing the limitations of existing techniques in achieving both accurate urban inundation detection and extensive large-scale monitoring.

Study Configuration

• Spatial Scale: Regional (Weihui City, China, 862 km²) for application, with the method designed for large-scale monitoring (Sentinel-1 Interferometric Wide Swath mode provides a 250 km swath width). Data resolutions range from 5 m × 20 m (Sentinel-1) to 0.1° (GPM).
• Temporal Scale: Annual time series (Sentinel-1 for 2021) for baseline statistics, with specific flood event monitoring in July-August 2021. Data resolutions range from 30 minutes (GPM) to 6-12 days (Sentinel-1 revisit cycle).

Methodology and Data

• Models used:
  • Z-score method for pixel-wise temporal anomaly detection.
  • Rule-based flood classification tree integrating VV and VH polarization Z-scores.
  • Pixel aggregation for post-processing speckle noise removal.
  • Modified Normalized Difference Water Index (MNDWI) for optical image validation.
• Data sources:
  • Satellite: Sentinel-1 Ground Range Detected (GRD) products (VV and VH polarizations, Interferometric Wide Swath mode, 5 m × 20 m spatial resolution).
  • Satellite: Global Precipitation Measurement (GPM) Integrated Multi-satellite Retrievals for GPM (IMERG) precipitation product (0.1° spatial, 30 minutes temporal resolution).
  • Auxiliary: Global Surface Water (GSW) dataset (v1.4, 30 m spatial resolution, water occurrence band).
  • Auxiliary: Copernicus Digital Elevation Model (DEM) (30 m spatial resolution).
  • Validation: Sentinel-2 optical images (Green and Short-Wave Infrared bands, 8-9 hours time difference from SAR).
  • Validation: Unmanned Aerial Vehicle (UAV) optical images (ultra-high spatial resolution, 2 hours time difference from SAR).

Main Results

• The proposed method demonstrated strong performance during the 2021 Weihui flood event.
• In natural areas, the method achieved a Critical Success Index (CSI) of 60 % and an Overall Accuracy (OA) of 90 %.
• In building areas, the method achieved a CSI of 62 % and an OA of 73 %.
• Analysis of SAR backscatter characteristics revealed that in natural areas, vegetation growth significantly affects intensity consistency, while in building areas, rainfall and floodwater can intensify or suppress double-bounce scattering, disrupting consistency. Urban backscatter intensity increased sharply (by more than 6 dB) during flooding.
• A Z-score threshold of -2 was found appropriate for flood detection in natural areas, corresponding to a 95 % confidence level.
• The proposed dual-polarization Z-score classification tree provided the most balanced performance across both natural and urban environments, with mean CSI and OA of 61 % and 81.5 %, respectively, outperforming baseline methods in building areas and showing comparable, more reliable performance in natural areas.

Contributions

• Introduction of pixel-wise temporal standardization using Z-scores to reduce dependence on event-specific global thresholds and improve transferability across heterogeneous natural and built-up surfaces.
• Integration of VV and VH polarization anomalies within a rule-based classification tree, allowing for the consideration of open water, vegetation-related changes, and urban double-bounce effects in a unified framework.
• Enhanced operational scalability due to reliance solely on Sentinel-1 GRD time series, auxiliary water-occurrence data, and simple statistical operations, facilitating rapid large-area flood response while maintaining sensitivity to urban inundation.
• Detailed analysis of C-band SAR backscatter characteristics in time-series images for different land-cover types (natural and building areas) to inform the anomaly detection approach.

Funding

Not specified in the paper.

Citation

@article{Xing2026Largescale,
  author = {Xing, Runmei and Lu, Fengyuan and Zhao, Jun},
  title = {Large-scale flood monitoring based on time-series Sentinel-1 images and Z -score},
  journal = {Natural hazards and earth system sciences},
  year = {2026},
  doi = {10.5194/nhess-26-5299-2026},
  url = {https://doi.org/10.5194/nhess-26-5299-2026}
}