virtual-tiff 0.3.0

A parser intended for use with VirtualiZarr to create virtual Zarr stores from TIFFs

Virtual TIFF

A Parser for creating Virtual Zarr stores from TIFF files using VirtualiZarr 2.0 and async-tiff.

Background

First, some thoughts on why we should virtualize GeoTIFFs and/or COGS:

1. Provide faster access to non-cloud-optimized GeoTIFFS that contain some form of internal tiling without any data duplication see notebook #1.
2. Provide fully async I/O for both GeoTIFFs and COGs using Zarr-Python
3. Allow loading a stack of GeoTIFFS/COGS into a data cube while minimizing the number of GET requests relative to using stackstac/odc-stac, thereby decreasing cost and increasing performance
4. Provide users access to a lazily loaded DataTree providing both the data and the overviews, allowing scientists to use the overviews not only for tile-based visualization but also quickly iterating on analytics
5. Include etags in the virtualized datasets to support reproducibility
6. A motivation that's less clear to me, but maybe possible, is using the virtualization layer to access COGs with disparate CRSs as a single dataset (https://github.com/zarr-developers/geozarr-spec/issues/53)

Getting started

The library can be installed from PyPI:

python -m pip install virtual-tiff

You can use Virtual TIFF to load data directly:

import obstore
from virtualizarr.registry import ObjectStoreRegistry
from virtual_tiff import VirtualTIFF
import xarray as xr

# Configuration
bucket_url = "s3://e84-earth-search-sentinel-data/"
file_url = f"{bucket_url}sentinel-2-c1-l2a/10/T/FR/2023/12/S2B_T10TFR_20231223T190950_L2A/B04.tif"

# Setup and open dataset
s3_store = obstore.store.from_url(bucket_url, region="us-west-2", skip_signature=True)
registry = ObjectStoreRegistry({bucket_url: s3_store})

parser = VirtualTIFF(ifd=0)
manifest_store = parser(url=file_url, registry=registry)
ds = xr.open_zarr(manifest_store, zarr_format=3, consolidated=False)
ds.load()

or create a virtual dataset:

import obstore
from virtualizarr import open_virtual_dataset
from virtualizarr.registry import ObjectStoreRegistry
from virtual_tiff import VirtualTIFF

# Configuration
bucket_url = "s3://e84-earth-search-sentinel-data/"
file_url = f"{bucket_url}sentinel-2-c1-l2a/10/T/FR/2023/12/S2B_T10TFR_20231223T190950_L2A/B04.tif"

# Setup and open dataset
s3_store = obstore.store.from_url(bucket_url, region="us-west-2", skip_signature=True)
registry = ObjectStoreRegistry({bucket_url: s3_store})

ds = open_virtual_dataset(
    url=file_url,
    registry=registry,
    parser=VirtualTIFF(ifd=0)
)

Contributing

1. Clone the repository: git clone https://github.com/virtual-zarr/virtual-tiff.git.
2. Download test data from S3: pixi run -e test download-test-images. WARNING: This will download ~1.4GB of TIFFs for testing to your machine.
3. Run the test suite using pixi run -e test run-tests WARNING: Some tests will fail due to incomplete status of the implementation.
4. Start a shell if needed in the development environment using pixi run -e test zsh.

Test data is populated from three upstream sources via sync scripts (run these before upload_test_data.py to refresh S3):

• uv run scripts/sync_gdal_tiffs.py — GDAL autotest TIFFs
• uv run scripts/sync_external_tiffs.py — external TIFFs from various URLs
• uv run scripts/sync_geotiff_test_data.py — synthetic and real-world fixtures from geotiff-test-data

License

virtual-tiff is distributed under the terms of the MIT license.