This an official Pytorch implementation of our paper "A Spatially Masked Adaptive Gated Network for Multimodal Post-Flood Water Extent Mapping using SAR and Incomplete Multispectral Data". The specific details of the proposed model are as follows.
Abstract: We propose SMAGNet, a multimodal deep learning model for post-flood water extent mapping that adaptively fuses SAR with incomplete MSI data. Addressing the challenge of missing pixels in optical imagery due to limited temporal resolution and coregistration process, SMAGNet achieves state-of-the-art performance in terms of IoU with a score of 86.47% on the C2S-MS Floods dataset. Furthermore, it demonstrates strong robustness by maintaining high accuracy even when MSI data is completely unavailable, which makes it highly practical for real-world disaster response.
- 2025.12.24: Our paper has been accepted for publication in the ISPRS Journal of Photogrammetry and Remote Sensing!
- 2025.12.11: Source code and trained model checkpoints were released.
This project is entirely written in Python, primarily using the PyTorch library. All required dependencies can be installed using:
# Create a virtual environment
conda env create -f environment.yaml
# Activate the environment
conda activate smagnet- Download: https://registry.opendata.aws/c2smsfloods/
- After downloading the dataset, matching Sentinel-1 and Sentinel-2 patches using metadata, along with preprocessing, is required.
We provide the trained model for quick reproduction. download.
Qualitative examples using these trained models are presented in examples.ipynb.
Comparison with state-of-the-art methods on the C2S-MS Floods dataset:
| Model | IoU (%) | Precision (%) | Recall (%) | OA (%) |
|---|---|---|---|---|
| U-Net (SAR) | 79.65 (±0.96) | 90.81 (±0.83) | 86.64 (±1.03) | 96.52 (±0.18) |
| PSPNet | 82.65 (±0.85) | 90.83 (±0.93) | 90.19 (±1.29) | 97.02 (±0.15) |
| VFuseNet | 83.33 (±1.00) | 92.98 (±0.62) | 88.92 (±0.89) | 97.20 (±0.18) |
| FuseNet | 83.40 (±1.13) | 92.95 (±0.71) | 89.03 (±0.87) | 97.21 (±0.20) |
| FTransUNet | 83.93 (±2.64) | 92.19 (±1.08) | 90.34 (±2.46) | 97.28 (±0.47) |
| FPN | 84.25 (±0.96) | 91.10 (±1.03) | 91.80 (±0.54) | 97.30 (±0.19) |
| U-Net++ | 84.41 (±1.54) | 92.75 (±0.69) | 90.36 (±1.32) | 97.37 (±0.27) |
| DeepLabV3+ | 84.48 (±1.19) | 92.04 (±0.68) | 91.14 (±1.26) | 97.37 (±0.21) |
| CMGFNet | 84.70 (±0.59) | 94.85 (±0.46) | 88.78 (±0.71) | 97.48 (±0.10) |
| CMFNet | 84.95 (±0.87) | 92.31 (±0.93) | 91.43 (±0.95) | 97.45 (±0.16) |
| U-Net | 84.96 (±0.97) | 92.88 (±0.60) | 90.88 (±0.92) | 97.47 (±0.17) |
| MCANet | 85.48 (±0.99) | 92.47 (±0.78) | 91.87 (±0.82) | 97.54 (±0.18) |
| MFGFUNet | 85.96 (±0.57) | 92.84 (±0.98) | 92.07 (±0.73) | 97.63 (±0.11) |
| SMAGNet (Ours) | 86.47 (±0.61) | 93.05 (±0.76) | 92.45 (±0.83) | 97.73 (±0.11) |
If you find this work useful, please consider citing the following paper:
@article{lee2026spatially,
title={A Spatially Masked Adaptive Gated Network for multimodal post-flood water extent mapping using SAR and incomplete multispectral data},
author={Lee, Hyunho and Li, Wenwen},
journal={ISPRS Journal of Photogrammetry and Remote Sensing},
volume={232},
pages={492--508},
year={2026},
publisher={Elsevier}
}