---
language:
- en
license: mit
pretty_name: Ocean Floor Bathymetry Enhancement Dataset
size_categories:
- 10K<n<100K
tags:
- bathymetry
- ocean-mapping
- climate-change
- uncertainty-quantification
- earth-observation
task_categories:
- image-segmentation
- other
task_ids:
- super-resolution
- semantic-segmentation
---

--

# Ocean Floor Bathymetry Enhancement Dataset

This dataset contains bathymetric data supporting the research presented in "Learning Enhanced Structural Representations with Block-Based Uncertainties for Ocean Floor Mapping" (ICLR 2025). It provides paired low-resolution and high-resolution bathymetric samples from various ocean regions worldwide.

## Overview

Accurate ocean modeling and coastal hazard prediction depend on high-resolution bathymetric data; yet, current worldwide datasets are too coarse for exact numerical simulations. This dataset addresses this gap by providing paired samples that can be used to train deep learning models for bathymetric enhancement with uncertainty quantification.

## Dataset Structure

The dataset is organized as follows:
- `source.tar.gz`: Contains low-resolution (32×32) bathymetry samples
  - NumPy (.npy) files containing single-channel depth measurements
- `target.tar.gz`: Contains high-resolution bathymetry samples
  - NumPy (.npy) files containing corresponding higher resolution measurements
- `data.csv`: Contains metadata for each sample including location, coordinates, etc.

## Data Coverage

The dataset includes samples from six major oceanic regions:
1. Western Pacific Region - Contains complex underwater ridge systems and notable bathymetric variation
2. Indian Ocean Basin - Notable for tsunami risk and tectonic activity
3. Eastern Atlantic Coast - Characterized by tsunami-prone areas and coastal flooding
4. South Pacific Region - Features cyclones and wave-driven inundation patterns
5. Eastern Pacific Basin - Contains frequent tsunamis and submarine volcanism
6. North Atlantic Basin - Known for hurricanes and storm surges

## Usage

The dataset includes a custom PyTorch dataset loader class that handles loading, normalization, and preprocessing:

```python
from dataset import BathyGEBCOSuperResolutionDataset

# Initialize the dataset
dataset = BathyGEBCOSuperResolutionDataset(
    base_dir="path/to/extracted/data",
    split_type="train"  # or "test"
)

# Access a sample
[low_res_16x16, low_res_32x32, high_res_64x64], metadata = dataset[0]

# Metadata contains information about the sample
print(metadata['location_name'])  # e.g., "Western Pacific Region"
print(metadata['latitude'], metadata['longitude'])  # Geographical coordinates
```

The dataset loader automatically handles normalization and can be configured with specific statistics:

```python
# Initialize with custom normalization parameters
cfg = {
    'mean': -3911.3894,
    'std': 1172.8374,
    'max': 0,
    'min': -10994
}

dataset = BathyGEBCOSuperResolutionDataset(
    base_dir="path/to/extracted/data",
    split_type="train",
    cfg=cfg
)
```

## Related Research

This dataset was developed to support research on uncertainty-aware deep learning for bathymetric enhancement. The associated paper introduces a block-based uncertainty mechanism for capturing local bathymetric complexity with spatially adaptive confidence estimates.

## Citation

If you use this dataset in your research, please cite:

```
@inproceedings{minoza2025learning,
  title={Learning Enhanced Structural Representations with Block-Based Uncertainties for Ocean Floor Mapping},
  author={Jose Marie Antonio Minoza},
  booktitle={Tackling Climate Change with Machine Learning Workshop at ICLR},
  year={2025}
}

@misc{ocean-floor-mapping2025_modelzoo,
  doi = {10.5281/ZENODO.15272540},
  author = {Minoza, Jose Marie Antonio},
  title = {Model Zoo: Learning Enhanced Structural Representations with Block-Based Uncertainties for Ocean Floor Mapping},
  publisher = {Zenodo},
  year = {2025},
  copyright = {Creative Commons Attribution 4.0 International}
}
```

## License

This dataset is licensed under the MIT License.

## Acknowledgments

- GEBCO for providing the original bathymetric data
- The research was presented at the ICLR 2025 Workshop on Tackling Climate Change with Machine Learning