Python library for accessing and working with Tessera geospatial foundation model embeddings.

GeoTessera provides access to geospatial embeddings from the Tessera foundation model, which processes Sentinel-1 and Sentinel-2 satellite imagery to generate 128-channel representation maps at 10m resolution. These embeddings compress a full year of temporal-spectral features into dense representations optimized for downstream geospatial analysis tasks. Read more details about the model.

This repo provides precomputed embeddings for multiple years and regions. Embeddings are generated by randomly sampling tiles within each region to ensure broad spatial coverage.

If some years (2017–2024) / areas are still missing for your use case, please submit an Embedding Request:

• 👉 Open an Embedding Request
• Please include: your organization, intended use, ROI as a bounding box with four points (lon,lat, 4 decimals), and the year(s).

After you submit the request, we will prioritize your ROI and notify you via a comment in the issue once the embeddings are ready.

A request for support Due to limited compute resources, we're unable to fulfill embedding requests covering large geographic areas or requiring substantial processing time. To help us serve the community better, we kindly ask requesters—especially those from commercial organizations or those requiring large-scale processing—to sponsor their requests by providing us with Azure credits. Importantly, the resulting outputs will be contributed to our global embeddings database, making them freely available for the entire research and user community. This approach allows us to scale our service while building a shared resource that benefits everyone. If you are in a position to support us in this way, please contact Prof. S.Keshav at [email protected]. We greatly appreciate your understanding and support in making Tessera more accessible to all.

• Installation
• Architecture
• Quick Start
• Python API
• CLI Reference
• Complete Workflows
• Registry System
• Data Organization
• Contributing

pip install geotessera

For development:

git clone https://github.com/ucam-eo/geotessera
cd geotessera
pip install -e .

GeoTessera is built around a simple two-step workflow:

1. Retrieve embeddings: Fetch raw numpy arrays for a geographic bounding box
2. Export to desired format: Save as raw numpy arrays or convert to georeferenced GeoTIFF files

The Tessera embeddings use a 0.1-degree grid system:

• Tile size: Each tile covers 0.1° × 0.1° (approximately 11km × 11km at the equator)
• Tile naming: Tiles are named by their center coordinates (e.g., grid_0.15_52.05)
• Tile bounds: A tile at center (lon, lat) covers:
  • Longitude: [lon - 0.05°, lon + 0.05°]
  • Latitude: [lat - 0.05°, lat + 0.05°]
• Resolution: 10m per pixel (variable number of pixels per tile depending on latitude)

When you request embeddings, GeoTessera downloads several files via Pooch:

1. Quantized embeddings (grid_X.XX_Y.YY.npy):

   • Shape: (height, width, 128)
   • Data type: int8 (quantized for storage efficiency)
   • Contains the compressed embedding values

2. Scale files (grid_X.XX_Y.YY_scales.npy):

   • Shape: (height, width) or (height, width, 128)
   • Data type: float32
   • Contains scale factors for dequantization

3. Dequantization: final_embedding = quantized_embedding * scales

When exporting to GeoTIFF, additional landmask files are fetched:

• Landmask tiles (grid_X.XX_Y.YY.tiff):
  • Provide UTM projection information
  • Define precise geospatial transforms
  • Contain land/water masks

User Request (lat/lon bbox)
    ↓
Registry Lookup (find available tiles)
    ↓
Download Files (via Pooch with caching)
    ├── embedding.npy (quantized)
    └── embedding_scales.npy
    ↓
Dequantization (multiply arrays)
    ↓
Output Format
    ├── NumPy arrays → Direct analysis
    └── GeoTIFF → GIS integration

Before downloading, check what data is available:

# Generate a coverage map showing all available tiles
geotessera coverage --output coverage_map.png

# Generate a coverage map for the UK
geotessera coverage --country uk

# View coverage for a specific year
geotessera coverage --year 2024 --output coverage_2024.png

# Customize the visualization
geotessera coverage --year 2024 --tile-color blue --tile-alpha 0.3 --dpi 150

Download embeddings as either numpy arrays or GeoTIFF files:

# Download as GeoTIFF (default, with georeferencing)
geotessera download \
  --bbox "-0.2,51.4,0.1,51.6" \
  --year 2024 \
  --output ./london_tiffs

# Download as raw numpy arrays (with metadata JSON)
geotessera download \
  --bbox "-0.2,51.4,0.1,51.6" \
  --format npy \
  --year 2024 \
  --output ./london_arrays

# Download using a GeoJSON/Shapefile region
geotessera download \
  --region-file cambridge.geojson \
  --format tiff \
  --year 2024 \
  --output ./cambridge_tiles

# Download specific bands only
geotessera download \
  --bbox "-0.2,51.4,0.1,51.6" \
  --bands "0,1,2" \
  --year 2024 \
  --output ./london_rgb

Generate web maps from downloaded GeoTIFFs:

# Create an interactive web map
geotessera visualize \
  ./london_tiffs \
  --type web \
  --output ./london_web

# Create an RGB mosaic
geotessera visualize \
  ./london_tiffs \
  --type rgb \
  --bands "30,60,90" \
  --output ./london_rgb

# Serve the web map locally
geotessera serve ./london_web --open

The library provides two main methods for retrieving embeddings:

from geotessera import GeoTessera

# Initialize the client
gt = GeoTessera()

# Method 1: Fetch a single tile
embedding, crs, transform = gt.fetch_embedding(lon=0.15, lat=52.05, year=2024)
print(f"Shape: {embedding.shape}")  # e.g., (1200, 1200, 128)
print(f"CRS: {crs}")  # Coordinate reference system from landmask

# Method 2: Fetch all tiles in a bounding box
bbox = (-0.2, 51.4, 0.1, 51.6)  # (min_lon, min_lat, max_lon, max_lat)
embeddings = gt.fetch_embeddings(bbox, year=2024)

for tile_lon, tile_lat, embedding_array, crs, transform in embeddings:
    print(f"Tile ({tile_lat}, {tile_lon}): {embedding_array.shape}")

# Export embeddings for a region as individual GeoTIFF files
files = gt.export_embedding_geotiffs(
    bbox=(-0.2, 51.4, 0.1, 51.6),
    output_dir="./output",
    year=2024,
    bands=None,  # Export all 128 bands (default)
    compress="lzw"  # Compression method
)

print(f"Created {len(files)} GeoTIFF files")

# Export specific bands only (e.g., first 3 for RGB visualization)
files = gt.export_embedding_geotiffs(
    bbox=(-0.2, 51.4, 0.1, 51.6),
    output_dir="./rgb_output",
    year=2024,
    bands=[0, 1, 2]  # Only export first 3 bands
)

# Fetch and process embeddings directly
embeddings = gt.fetch_embeddings(bbox, year=2024)

for lon, lat, embedding, crs, transform in embeddings:
    # Compute statistics
    mean_values = np.mean(embedding, axis=(0, 1))  # Mean per channel
    std_values = np.std(embedding, axis=(0, 1))    # Std per channel
    
    # Extract specific pixels
    center_pixel = embedding[embedding.shape[0]//2, embedding.shape[1]//2, :]
    
    # Apply custom processing
    processed = your_analysis_function(embedding)

from geotessera.visualization import (
    create_rgb_mosaic,
    visualize_global_coverage
)
from geotessera.web import (
    create_coverage_summary_map,
    geotiff_to_web_tiles
)

# Create an RGB mosaic from multiple GeoTIFF files
create_rgb_mosaic(
    geotiff_paths=["tile1.tif", "tile2.tif"],
    output_path="mosaic.tif",
    bands=(0, 1, 2)  # RGB bands
)

# Generate web tiles for interactive maps
geotiff_to_web_tiles(
    geotiff_path="mosaic.tif",
    output_dir="./web_tiles",
    zoom_levels=(8, 15)
)

# Create a global coverage visualization
visualize_global_coverage(
    tessera_client=gt,
    output_path="global_coverage.png",
    year=2024,  # Or None for all years
    width_pixels=2000,
    tile_color="red",
    tile_alpha=0.6
)

Download embeddings for a region in your preferred format:

geotessera download [OPTIONS]

Options:
  -o, --output PATH         Output directory [required]
  --bbox TEXT              Bounding box: 'min_lon,min_lat,max_lon,max_lat'
  --region-file PATH       GeoJSON/Shapefile to define region
  -f, --format TEXT        Output format: 'tiff' or 'npy' (default: tiff)
  --year INT               Year of embeddings (default: 2024)
  --bands TEXT             Comma-separated band indices (default: all 128)
  --compress TEXT          Compression for TIFF format (default: lzw)
  --list-files             List all created files with details
  -v, --verbose            Verbose output

Output formats:

• tiff: Georeferenced GeoTIFF files with UTM projection
• npy: Raw numpy arrays with metadata.json file

Create visualizations from GeoTIFF files:

geotessera visualize INPUT_PATH [OPTIONS]

Options:
  -o, --output PATH        Output directory [required]
  --type TEXT              Visualization type: rgb, web, coverage
  --bands TEXT             Comma-separated band indices for RGB
  --normalize              Normalize bands
  --min-zoom INT           Min zoom for web tiles (default: 8)
  --max-zoom INT           Max zoom for web tiles (default: 15)
  --force                  Force regeneration of tiles

Generate a world map showing data availability:

geotessera coverage [OPTIONS]

Options:
  -o, --output PATH        Output PNG file (default: tessera_coverage.png)
  --year INT               Specific year to visualize
  --tile-color TEXT        Color for tiles (default: red)
  --tile-alpha FLOAT       Transparency 0-1 (default: 0.6)
  --tile-size FLOAT        Size multiplier (default: 1.0)
  --dpi INT                Output resolution (default: 100)
  --width INT              Figure width in inches (default: 20)
  --height INT             Figure height in inches (default: 10)
  --no-countries           Don't show country boundaries

Serve web visualizations locally:

geotessera serve DIRECTORY [OPTIONS]

Options:
  -p, --port INT           Port number (default: 8000)
  --open/--no-open         Auto-open browser (default: open)
  --html TEXT              Specific HTML file to serve

Display information about GeoTIFF files or the library:

geotessera info [OPTIONS]

Options:
  --geotiffs PATH          Analyze GeoTIFF files/directory
  --dataset-version TEXT   Tessera dataset version
  -v, --verbose            Verbose output

GeoTessera uses a registry system to efficiently manage and access the large Tessera dataset:

• Block-based organization: Registry divided into 5×5 degree geographic blocks
• Lazy loading: Only loads registry blocks for the region you're accessing
• Automatic caching: Downloads are cached locally using Pooch
• Integrity checking: SHA256 checksums ensure data integrity

The registry can be loaded from multiple sources (in priority order):

1. Local directory (via --registry-dir or registry_dir parameter)
2. Environment variable (TESSERA_REGISTRY_DIR)
3. Auto-cloned repository (default, from GitHub)

# Use local registry
gt = GeoTessera(registry_dir="/path/to/tessera-manifests")

# Use auto-updating registry
gt = GeoTessera(auto_update=True)

# Use custom manifest repository
gt = GeoTessera(
    manifests_repo_url="https://github.com/your-org/custom-manifests.git"
)

tessera-manifests/
└── registry/
    ├── embeddings/
    │   ├── embeddings_2024_lon-5_lat50.txt    # 5×5° block
    │   ├── embeddings_2024_lon0_lat50.txt
    │   └── ...
    └── landmasks/
        ├── landmasks_lon-5_lat50.txt
        ├── landmasks_lon0_lat50.txt
        └── ...

Each registry file contains:

# Pooch registry format
filepath SHA256checksum
2024/grid_0.15_52.05/grid_0.15_52.05.npy sha256:abc123...
2024/grid_0.15_52.05/grid_0.15_52.05_scales.npy sha256:def456...

1. Request tiles for bbox → Determine which 5×5° blocks overlap
2. Load block registries → Parse only the needed registry files
3. Find available tiles → List tiles within the requested region
4. Fetch via Pooch → Download with caching and integrity checks

Remote Server (dl-2.tessera.wiki)
├── v1/                              # Dataset version
│   ├── 2024/                        # Year
│   │   ├── grid_0.15_52.05/         # Tile (named by center coords)
│   │   │   ├── grid_0.15_52.05.npy              # Quantized embeddings
│   │   │   └── grid_0.15_52.05_scales.npy       # Scale factors
│   │   └── ...
│   └── landmasks/
│       ├── grid_0.15_52.05.tiff     # Landmask with projection info
│       └── ...

~/.cache/geotessera/                 # Default cache location
├── tessera-manifests/                # Auto-cloned registry
│   └── registry/
├── pooch/                            # Downloaded data files
│   ├── grid_0.15_52.05.npy
│   ├── grid_0.15_52.05_scales.npy
│   └── ...

• Embeddings: Stored in simple arrays, referenced by center coordinates
• GeoTIFF exports: Use UTM projection from corresponding landmask tiles
• Web visualizations: Reprojected to Web Mercator (EPSG:3857)

# Set custom cache directory for downloaded files
export TESSERA_DATA_DIR=/path/to/cache

# Use local registry directory
export TESSERA_REGISTRY_DIR=/path/to/tessera-manifests

# Configure per-command
TESSERA_DATA_DIR=/tmp/cache geotessera download ...

Contributions are welcome! Please see our Contributing Guide for details. This project is licensed under the MIT License - see the LICENSE file for details.

If you use Tessera in your research, please cite the arXiv paper:

@misc{feng2025tesseratemporalembeddingssurface,
      title={TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis}, 
      author={Zhengpeng Feng and Clement Atzberger and Sadiq Jaffer and Jovana Knezevic and Silja Sormunen and Robin Young and Madeline C Lisaius and Markus Immitzer and David A. Coomes and Anil Madhavapeddy and Andrew Blake and Srinivasan Keshav},
      year={2025},
      eprint={2506.20380},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2506.20380}, 
}

• Tessera Foundation Model
• Tessera Interactive Notebook
• Documentation
• PyPI Package
• Issue Tracker