This repository contains the data registry files for the GeoTessera Python library, which provides access to geospatial embeddings from the Tessera geospatial foundation model.
The registry is organized into two main categories:
Contains checksum files for geospatial embedding tiles organized by year (2017-2024) and geographic regions. Each file lists the available embedding tiles for a specific year and geographic region with their SHA256 checksums.
File format: embeddings_{year}_lon{longitude}_lat{latitude}.txt
Each registry file contains entries in the format:
{year}/grid_{lon:.2f}_{lat:.2f}/grid_{lon:.2f}_{lat:.2f}.npy {sha256_hash}
{year}/grid_{lon:.2f}_{lat:.2f}/grid_{lon:.2f}_{lat:.2f}_scales.npy {sha256_hash}
The embeddings are organized in a 0.1-degree grid system where:
.npyfiles contain quantized embeddings (128 channels)_scales.npyfiles contain dequantization scales- Final embeddings are computed as:
embedding * scales
Contains checksum files for landmask TIFF files that define land/water boundaries for each grid tile, organized by geographic regions.
File format: landmasks_lon{longitude}_lat{latitude}.txt
Each registry file contains entries in the format:
grid_{lon:.2f}_{lat:.2f}.tiff {sha256_hash}
These registry files are automatically used by the GeoTessera Python library to:
- Discover available embedding tiles for a region of interest
- Verify data integrity through checksum validation
- Cache downloaded data locally using Pooch
The actual embedding and landmask data files are hosted at https://dl-2.tessera.wiki/ and automatically downloaded by the GeoTessera library as needed.
To use these embeddings in your projects, install the GeoTessera Python library:
pip install geotesseraSee the main repository for documentation and examples.