Chen et al. (2026) Sen2GF3Floods: A Benchmark Multi-Source Flood Dataset with Dual-Temporal and Active Learning Annotation

Identification

• Journal: Scientific Data
• Year: 2026
• Date: 2026-02-26
• Authors: Wenting Chen, Yueqin Zhu, Wenlong Han, D. Z. Liu, Guiquan Mo, Ziyao Xing
• DOI: 10.1038/s41597-026-06929-6

Research Groups

• National Institute of Natural Hazards, Ministry of Emergency Management of China, Beijing, China
• Key Laboratory of Investigation on Disaster and Accident, Ministry of Emergency Management, Beijing, China
• Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China

Short Summary

This paper introduces Sen2GF3Floods, a novel multi-source flood dataset integrating pre-disaster Sentinel-2 optical and post-disaster Gaofen-3 SAR imagery, annotated using a dual-temporal and active learning framework to enhance flood detection algorithm development.

Objective

• To address limitations in existing flood datasets by creating a high-quality, diverse, and precisely annotated multi-source benchmark dataset (Sen2GF3Floods) for advancing flood detection algorithms and supporting operational disaster response.

Study Configuration

• Spatial Scale: 21,483 standardized samples from nine major flood events covering diverse geomorphological settings.
• Temporal Scale: Dual-temporal, integrating pre-disaster (Sentinel-2) and post-disaster (Gaofen-3) imagery.

Methodology and Data

• Models used: U-Net++, U-Net, DeepLabV3+, DANet, SegFormer (for benchmark evaluations and active learning).
• Data sources: Sentinel-2 optical imagery (RGB and NIR bands), Gaofen-3 Synthetic Aperture Radar (SAR) imagery (HH and HV polarizations), Sentinel-1 SAR imagery (for transferability testing).

Main Results

• The Sen2GF3Floods dataset comprises 21,483 standardized samples from nine major flood events.
• Multi-source fusion of Sentinel-2 RGB and NIR data with Gaofen-3 HH and HV SAR data demonstrates robust flood mapping performance across varied scenarios.
• The dual-temporal collaborative annotation framework, incorporating semi-automatic labeling and active learning based on U-Net++, effectively balances annotation accuracy and efficiency, reducing annotation costs while maintaining quality.
• Models trained on Gaofen-3 SAR data exhibit good transferability to Sentinel-1 SAR data.

Contributions

• Introduction of Sen2GF3Floods, the first dataset to integrate pre-disaster Sentinel-2 optical imagery with post-disaster Gaofen-3 SAR imagery for flood detection.
• Development of a novel dual-temporal collaborative annotation framework combining semi-automatic labeling with active learning, optimizing annotation efficiency and precision.
• Provides a valuable benchmark dataset and methodology to advance flood detection algorithms and support operational and real-time disaster response.

Funding

• Department of Science and Technology of Ningxia Hui Autonomous Region (Grant number 2024BEG01005).

Citation

@article{Chen2026Sen2GF3Floods,
  author = {Chen, Wenting and Zhu, Yueqin and Han, Wenlong and Liu, D. Z. and Mo, Guiquan and Xing, Ziyao},
  title = {Sen2GF3Floods: A Benchmark Multi-Source Flood Dataset with Dual-Temporal and Active Learning Annotation},
  journal = {Scientific Data},
  year = {2026},
  doi = {10.1038/s41597-026-06929-6},
  url = {https://doi.org/10.1038/s41597-026-06929-6}
}