sm_unet_s2 Model trained on the Sentinel-2 L1C bands ["B02", "B03", "B04", "B08", "B11", "B12"] from the S1S2Water and WorldFloods datasets.
sm_unet_s1 Model trained on the Sentinel-1 GRD data ([VV, VH] channels) from the S1S2Water and Kuro Siwo datasets.
mm_unet_s1s2 Dual stream with modality token model, trained on the S1S2Water (Sentinel-1 GRD and Sentinel-2 L1C data), WorldFloods (Sentinel-2 L1C) and Kuro Siwo Sentinel-1 GRD data.
In order to run any of these models in Sentinel-1 and/or Sentinel-2 data see the tutorial Run model in the udl4fl package.
If you find this work useful please cite:
@article{PORTALESJULIA2025114882,
title = {Understanding flood detection models across Sentinel-1 and Sentinel-2 modalities and benchmark datasets},
journal = {Remote Sensing of Environment},
volume = {328},
pages = {114882},
year = {2025},
issn = {0034-4257},
doi = {https://doi.org/10.1016/j.rse.2025.114882},
url = {https://www.sciencedirect.com/science/article/pii/S003442572500286X},
author = {Enrique Portalés-Julià and Gonzalo Mateo-García and Luis Gómez-Chova},
keywords = {Flood detection, Deep learning, Multimodal fusion, Multispectral, SAR, Sentinel-1, Sentinel-2},
}
This research has been supported by the DEEPCLOUD project (PID2019-109026RB-I00, University of Valencia) funded by the Spanish Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033) and the European Union (NextGenerationEU).
DEEPCLOUD project (PID2019-109026RB-I00, University of Valencia) funded by MCIN/AEI/10.13039/501100011033.
