Model Card for OceanSAR-1-wave

Model Details

Model Description

OceanSAR-1-wave is a linear probing head for significant wave height (SWH) prediction built on top of the OceanSAR-1 foundation model. It leverages the powerful features extracted by OceanSAR-1 to accurately predict ocean wave heights from Synthetic Aperture Radar (SAR) imagery.

• Developed by: Thomas Kerdreux, Alexandre Tuel @ Galeio
• Deployed by: Antoine Audras @ Galeio
• Model type: Linear Regression Head on Vision Foundation Model
• License: Apache License 2.0
• Base model: OceanSAR-1 (ResNet50/ViT variants)
• Training data: Sentinel-1 Wave Mode (WV) SAR images with collocated wave height measurements

Uses

Direct Use

This model is designed for significant wave height prediction from SAR imagery over ocean surfaces. It can be used for:

• Near-real-time wave height estimation from SAR images
• Marine weather forecasting
• Ocean state monitoring
• Maritime safety applications
• Wave climate studies

Performance Results

The model achieves state-of-the-art performance in linear probing on significant wave height prediction, with performance varying by backbone architecture:

How to Use

import torch
from transformers import AutoModelForImageClassification

# Load the foundation model and the linear probing head
oceansar = AutoModelForImageClassification.from_pretrained("galeio-research/OceanSAR-1")

# Prepare your SAR image (should be single-channel VV polarization)
# Here using random data as example
dummy_image = torch.randn(1, 1, 256, 256)  # (C, H, W)

# Extract features
with torch.no_grad():
    outputs = oceansar(dummy_image)
    # For regression, use the single output value as the wave height prediction
    wave_height = outputs.logits.item()  # Output in meters

Training Details

Training Data

• Dataset: Sentinel-1 Wave Mode (WV) SAR images with collocated wave height measurements
• Source: Wave height measurements from altimeters, buoys, and wave models
• Preprocessing: Same as base OceanSAR-1 model

Evaluation

Metrics

Significant wave height prediction performance is evaluated using Root Mean Square Error (RMSE), achieving:

• 0.63 m RMSE with ResNet50 backbone
• 0.57 m RMSE with ViT-S/16 backbone
• 0.55 m RMSE with ViT-S/8 backbone
• 0.54 m RMSE with ViT-B/8 backbone

Comparison to Other Models

The model outperforms existing approaches:

• MoCo: 0.77 m RMSE
• DeCUR: 0.82 m RMSE
• SoftCon (ViT-S/14): 0.78 m RMSE
• SoftCon (ViT-B/14): 0.79 m RMSE

Technical Specifications

Hardware Requirements

• Same as base model
• Minimal additional computational cost for inference

Dependencies

• PyTorch >= 1.8.0
• Transformers >= 4.30.0
• Base OceanSAR-1 model

Input Specifications

• Same as base OceanSAR-1 model
• Single channel (VV polarization) SAR images
• 256x256 pixel resolution

Citation

BibTeX:

@article{kerdreux2025efficientselfsupervisedlearningearth,
  title={Efficient Self-Supervised Learning for Earth Observation via Dynamic Dataset Curation},
  author={Kerdreux, Thomas and Tuel, Alexandre and Febvre, Quentin and Mouche, Alexis and Chapron, Bertrand},
  journal={arXiv preprint arXiv:2504.06962},
  year={2025},
  eprint={2504.06962},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2504.06962},
}

Acknowledgements

This work was granted access to the HPC resources of IDRIS and TGCC under the allocation 2025-[A0171015666] made by GENCI.