UrbanSARFloods

January 6, 2026 · View on GitHub

This repo contains the dataset, explanation, and some corrections of the paper:
UrbanSARFloods: Sentinel-1 SLC-Based Benchmark Dataset for Urban and Open-Area Flood Mapping
by Jie Zhao, Zhitong Xiong, and Xiao Xiang Zhu.

This dataset release focuses on data availability. All baseline models were trained using standard segmentation pipelines implemented in the segmentation_models_pytorch (SMP) library. Since no customized code was developed, training scripts are not included.

👉 Check out the UrbanSARFloods for details.

Introduction

UrbanSARFloods is a large-scale, SAR-based flood mapping dataset processed from Sentinel-1 Single Look Complex (SLC) data, designed to address the lack of urban flood data in deep learning research. While SAR is widely used for flood detection due to its all-weather capability, most existing datasets focus on open-area floods while neglecting urban environments.

UrbanSARFloods features Sentinel-1 intensity and interferometric coherence image chips (512 × 512), covering 807,500 km² across 5 continents, 20 land cover classes, and 18 flood events. We benchmarked state-of-the-art CNN models and found that imbalanced data and limited training samples remain major challenges, particularly for urban flood detection.

Expanding this dataset and exploring transfer learning and data balancing strategies could further improve SAR-based flood mapping. To ensure accurate geospatial analysis, all data is provided in GeoTIFF format, preserving both geolocation and projection information.

Usage

1️⃣ Download the training and validation dataset (urban_sar_floods.tar.gz) from here, then extract it to ./urban_sar_floods.
2️⃣ The extracted dataset will be organized as follows:

urban_sar_floods
├── 01_NF
│   ├── GT       # Ground truth TIFF files
│   │   ├── file_1.tif
│   │   ├── file_2.tif
│   │   └── ...
│   ├── SAR      # SAR TIFF files
│   │   ├── file_1.tif
│   │   ├── file_2.tif
│   │   └── ...
├── 02_FO
│   ├── GT
│   ├── SAR
├── 03_FU
│   ├── GT
│   ├── SAR
├── Train_dataset.txt   # List of training samples
├── Valid_dataset.txt   # List of validation samples

3️⃣ Download the testing dataset (testing_case_256 /testing_case_orig ) from here as follows:

testing_case_256  # Testing dataset (preprocessed SAR into 256×256 patches)
├── Event1
│   ├── SAR_patch_001.tif
│   ├── SAR_patch_002.tif
│   ├── ...
├── Event2
│   ├── SAR_patch_001.tif
│   ├── SAR_patch_002.tif
│   ├── ...
├── Event3
│   ├── SAR_patch_001.tif
│   ├── SAR_patch_002.tif
│   ├── ...

testing_case_orig  # Testing dataset (original full-size SAR and GT files)
├── Event1
│   ├── SAR.tif  # Full-size SAR image
│   ├── GT.tif   # Full-size ground truth
├── Event2
│   ├── SAR.tif
│   ├── GT.tif
├── Event3
│   ├── SAR.tif
│   ├── GT.tif

4️⃣ The original Sentinel-1 Single Look Complex (SLC) data used in the UrbanSARFloods dataset are provided here.

Sentinel-1_SLC_data  # Original Sentinel-1 SLC data for UrbanSARFloods
├── Event1
│   ├── S1A_*.zip  # Pre-flood SLC acquisition
│   ├── S1A_*.zip  # Pre-flood SLC acquisition
│   ├── S1A_*.zip  # Post-flood SLC acquisition
├── Event2
│   ├── S1A_*.zip
│   ├── S1A_*.zip
│   ├── S1A_*.zip
├── Event3
│   ├── S1A_*.zip
│   ├── S1A_*.zip
│   ├── S1A_*.zip
├── ...

For each flood event, the repository includes:

  1. Two pre-flood Sentinel-1 SLC acquisitions, and
  2. One post-flood Sentinel-1 SLC acquisition. All files are stored using their original Sentinel-1 product filenames. Users can directly retrieve detailed acquisition metadata, including exact sensing time, absolute orbit number, relative orbit (path), and track information, from the official ESA Sentinel-1 data hub by referencing these filenames.

🛠 Note: Cropping GeoTIFF Data

If you need to crop images in testing_case_orig to a specific size or align it with another geotif file, you can use GDAL's gdalwarp tool.

For more details, check out the GDAL Warp documentation.

Correction

📢 Note: This table is the corrected version of Table 2 from the CVPRW UrbanSARFloods paper, with missing information supplemented. Please refer to this version for accurate data.

ContinentLocationEvent DateImage SizeAbsolute OrbitPathNumber of NF tilesNumber of FO tilesNumber of FU tiles
North AmericaHouston, US19 April 201617766 × 1330610890143175209154
Houston, US30 August 201714918 × 129817169143274323129
Lumberton, US11 Oct 20169931 × 646513449771472012
Sainte-Marthe-sur-le-Lac, Canada02 May 201921638 × 1118427055332794312
AfricaBeledweyne, Somalia08 May 201814293 × 116562180735495734
Beira, Mozambique20 March 201914904 × 126081543261338914
Beledweyne, Somalia14 Nov 202315457 × 1263451207354555620
Jubba, Somalia*01 Dec 202315548 × 13078505801084005913
15454 × 12710514551084606510
Lokoja, Niger13 Oct 202215500 × 1258744902304271071
AsiaIwaki/Koriyama, Japan12 Oct 201911751 × 10096184474635315819
Weihui, China*27 July 202118927 × 122453896240293221135
Aqqala, Iran29 March 202119549 × 12580265545733335125
Zhuozhou, China05 August 202319906 × 1220749739142332204137
Langfang, China05 August 202319458 × 1222049739142279269117
OceaniaCoraki, Australia2 March 202218160 × 1335842146741055411
Sydney, Australia24 March 202119495 × 1358237144147468956
Sydney, Australia5 July 202219498 × 135844396914739115523
Port Macquarie, Australia19 March 202118774 × 1346037071741248926
EuropeNovaKakhovka, Ukraine*09 June 202322596 × 12226489111410361237

License

The dataset is are released under the CC-BY-4.0 license.

Citation

If you find this repository useful, please consider citing the following paper:

@INPROCEEDINGS{10678367,  
  author={Zhao, Jie and Xiong, Zhitong and Zhu, Xiao Xiang},  
  booktitle={2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},  
  title={UrbanSARFloods: Sentinel-1 SLC-Based Benchmark Dataset for Urban and Open-Area Flood Mapping},  
  year={2024},  
  pages={419-429},  
  keywords={Training;Satellites;Urban areas;Transfer learning;Sentinel-1;Land surface;Benchmark testing;Sentinel-1;flood mapping;benchmark dataset;urban flood},  
  doi={10.1109/CVPRW63382.2024.00047}  
}

Further Information

For a comprehensive review of urban flood mapping with satellite SAR data, we invite you to explore our recent related paper:

🔗 Urban Flood Mapping Using Satellite Synthetic Aperture Radar Data: A review of characteristics, approaches, and datasets

This paper provides an in-depth discussion on challenges, methodologies, and future directions in urban flood detection, complementing the UrbanSARFloods dataset.

@ARTICLE{10795465,
  author={Zhao, Jie and Li, Ming and Li, Yu and Matgen, Patrick and Chini, Marco},
  journal={IEEE Geoscience and Remote Sensing Magazine}, 
  title={Urban Flood Mapping Using Satellite Synthetic Aperture Radar Data: A review of characteristics, approaches, and datasets}, 
  year={2024},
  volume={},
  number={},
  pages={2-34},
  keywords={Floods;Buildings;Urban areas;Sensors;Synthetic aperture radar;Reviews;Backscatter;Sensor phenomena and characterization;Spatial resolution;Spaceborne radar},
  doi={10.1109/MGRS.2024.3496075}}