Portraying the Need for Temporal Data in Flood Detection via Sentinel-1

Xavier Bou, Thibaud Ehret, Rafael Grompone, Jeremy Anger,

Centre Borelli, ENS Paris-Saclay and Kayrros.

---

This repository is the official implementation of the paper Portraying the Need for Temporal Data in Flood Detection via Sentinel-1.

Identifying flood affected areas in remote sensing data is a critical problem in earth observation to analyze flood impact and drive responses. While a number of methods have been proposed in the literature, there are two main limitations in available flood detection datasets: (1) a lack of region variability is commonly observed and/or (2) they require to distinguish permanent water bodies from flooded areas from a single image, which becomes an ill-posed setup. Consequently, we extend the globally diverse MMFlood dataset [1] to multidate by providing one year of Sentinel-1 observations around each flood event. To our surprise, we notice that the definition of flooded pixels in MMFlood is inconsistent when observing the entire image sequence. Hence, we re-frame the flood detection task as a temporal anomaly detection problem, where anomalous water bodies are segmented from a Sentinel-1 temporal sequence. From this definition, we provide a simple method inspired by the popular video change detector ViBe [2], results of which quantitatively align with the SAR image time series, providing a reasonable baseline for future works.

[1] F. Montello, E. Arnaudo and C. Rossi. "MMFlood: A Multimodal Dataset for Flood Delineation From Satellite Imagery," in IEEE Access, vol. 10, pp. 96774-96787, 2022, doi: 10.1109/ACCESS.2022.3205419.

[2] Barnich, Olivier, and Marc Van Droogenbroeck. "ViBe: A universal background subtraction algorithm for video sequences." IEEE Transactions on Image processing 20.6 (2010): 1709-1724.

Requirements:

numpy
torch
pillow
opencv-python
pytorch-lightning

Data

You can download the extended multi-date Temporal MMFlood from the following link. Timeseries are provided only for the test set EMSR events of MMFlood (except EMSR162).

Place the unzipped data in data/mmflood-multidate withing the project's directory. If you desire, you can place your data at a different location. In such case, you will have to update the data_root variable in the bash scripts provided in the scripts/ directory.

Algorithm

The approach is completely unsupervised, and it is an adaptation of the popular background subtractor ViBe, which builds a background model to detect anomalous events at differnt locations in the image. The high-level steps of the algorithm are:

• Initialize a background model of sample_num samples at each pixel with a temporal median across a set of images.
• For each image, the number of observed water events at each pixel is evaluated. A pixel is classified as flood if the SAR processing computes a water event, and less than min_c water events are seen in the model at that location
• Finally, all non-anomalous (non-flood) pixels are updated with the new information at a random position in the stack of collected observations

Usage :

You can just run the following command from the project's directory:

bash scripts/predict_mmflood_test.sh

You can also run the detection on specific individual scenes by running the scripts/run_flood_detection.bash bash file instead.

Contact

Xavier Bou
Email: xavier.bou_hernandez@ens-paris-saclay.fr

Citation

If you found our work useful, consider citing it as follows:

@article{Bou:2024,
      title={Portraying the Need for Temporal Data in Flood Detection via Sentinel-1}, 
      author={Xavier Bou and Thibaud Ehret and Rafael Grompone von Gioi and Jeremy Anger},
      year={2024},
      eprint={2403.03671},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2403.03671}, 
}

License and Acknowledgementgit a

This project is licensed under the GNU Affero General Public License v3.0 - see the LICENSE file for details.