harshinde/hls-burn-scars-zarr

HLS Burn Scars Dataset (Zarr Format)

Dataset Summary

This dataset contains Harmonized Landsat and Sentinel-2 (HLS) satellite imagery of wildfire burn scars and associated segmentation masks for the years 2018-2021 over the contiguous United States. The dataset includes 804 scenes of 512×512 pixels, optimized for training geospatial machine learning models.

Key Features:

• 6 spectral bands per image (Blue, Green, Red, NIR, SWIR1, SWIR2)
• Binary segmentation masks for burn scar detection
• Zarr format for efficient cloud storage and streaming access
• 540 training samples + 264 validation samples

Dataset Structure

hls_burn_scars_zarr/
├── training.zarr/
│   ├── images/      # Shape: (540, 6, 512, 512), dtype: float32
│   └── masks/       # Shape: (540, 512, 512), dtype: int16
└── validation.zarr/
    ├── images/      # Shape: (264, 6, 512, 512), dtype: float32
    └── masks/       # Shape: (264, 512, 512), dtype: int16

Data Format

Images:

• Shape: (6, 512, 512) - 6 spectral bands, 512×512 pixels
• Data type: float32
• Values: Normalized surface reflectance (0-1)
• Resolution: 30m per pixel

Masks:

• Shape: (512, 512)
• Data type: int16
• Values:
  • 1 = Burn scar
  • 0 = Not burned
  • -1 = No data/missing

Band Information

Each scene contains six bands from HLS S30:

Channel	Name	HLS Band	Wavelength (μm)
1	Blue	B02	0.45-0.51
2	Green	B03	0.53-0.59
3	Red	B04	0.64-0.67
4	NIR	B8A	0.85-0.88
5	SWIR1	B11	1.57-1.65
6	SWIR2	B12	2.11-2.29

Class Distribution

• Burn Scar: 11%
• Not Burned: 88%
• No Data: 1%

Data Splits

• Training: 540 samples (67%)
• Validation: 264 samples (33%)

Usage

Using with ml-data Library

import ml_data

# Load training data
train_dataset = ml_data.load('hls_burn_scars', split='train')
val_dataset = ml_data.load('hls_burn_scars', split='val')

# Get samples
image, mask = train_dataset[0]
print(f"Image shape: {image.shape}")  # (6, 512, 512)
print(f"Mask shape: {mask.shape}")    # (512, 512)

# Batch loading
images, masks = train_dataset.get_batch([0, 1, 2, 3, 4])

Processing

• All bands converted to surface reflectance
• Normalized to 0-1 range
• Compressed using Blosc/zstd (compression level 3)
• Stored in Zarr v2 format for efficient access

Citation

If this dataset helped your research, please cite:

@software{HLS_Foundation_2023,
    author = {Phillips, Christopher and Roy, Sujit and Ankur, Kumar and Ramachandran, Rahul},
    doi    = {10.57967/hf/0956},
    month  = aug,
    title  = {{HLS Foundation Burnscars Dataset}},
    url    = {https://huggingface.co/datasets/nasa-impact/hls_burn_scars},
    year   = {2023}
}

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license.