GeoTessera skill
GeoTessera is a Python library and skill teaching agents to read Tessera satellite embeddings for points and regions.
- 365 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The library reads a public store and writes files to a chosen directory.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/geotesseraDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The library reads a public store and writes files to a chosen directory.
This is third-party code. Review the repository files before installing.
What it does
The `geotessera` library reads embeddings from the Tessera model, which compresses a year of Sentinel-1 and Sentinel-2 imagery into 128 numbers per 10 m pixel. `GeoTesseraZarr` pulls values from a public Zarr store with no upfront download: points through `sample_points`, areas through `read_region`, patches through `read_patch`. The `skills/geotessera/SKILL.md` skill explains the rules to an agent: do not mix dataset versions, do not reproject embeddings before analysis, read large areas in strips. The `geotessera download` CLI saves a region as GeoTIFF and `geotessera webmap` builds a web map.
Who it is for. Remote sensing and GIS specialists who classify land use or detect change and delegate the routine to an agent.
Good fit when
- Embeddings at thousands of points for 2024 are needed for a training set.
- A region is exported to GeoTIFF and opened in QGIS.
- A land-use classifier is trained on 128-dimensional pixel vectors.
- A quick embedding web map for a district bounding box is wanted.
Not a fit when
- You need raw Sentinel imagery rather than embeddings.
- The work runs on Python older than 3.12.
- The analysis must combine embeddings from different dataset versions.
Example request
Take 2024 embeddings for these 500 points, train a forest classifier and report accuracy.Limitations
Python 3.12 or newer is required. Dataset v2 is still produced slowly with no guaranteed timeline, so take v1.1 covering 2017–2025 for work today. Embeddings from different versions cannot be compared or mixed because they live in incompatible spaces. Each dequantized pixel weighs 512 bytes, so large areas are read in strips. NPY tile export is deprecated.
How to disable. Remove the skill from the agent skills directory and the package with `pip uninstall geotessera`. Exported GeoTIFFs and web maps stay in the directories you chose and are deleted manually.
FAQ
How do I get an embedding for a point?
Call `gt.probe(lon, lat, year)`. The status `valid`, `water`, `nodata` or `outside` explains why a vector may be empty (NaN).
How do I export a region?
Run `geotessera download --bbox '-3.0,53.4,-2.9,53.5' --year 2024 --output region/`. GeoTIFF files are written one per UTM zone.
Which dataset version should I use?
v1.1 suits everyone: `v1.1-dclimate` covering 2017–2025 is read by default. Version v2 is released slowly and its embeddings can be pre-requested.
What does NaN in a vector mean?
No embedding exists: water or a pixel not yet produced. Drop such rows before fitting a model.
How do v1.1 and v2 differ?
v1.1 has wall-to-wall global coverage for 2017–2025. V2 is produced slowly with no guaranteed timeline, and embeddings of the two versions cannot be compared.
Discussion
No questions or reviews yet.
Related
SkillsA collection of 39 skills and 8 workflows for geoscience: well logs, seismic, 3D models, inversion, geostatistics and climate.
opengis-skills gives an agent 120 skills on GIS, CAD, AI, IoT, 3D and C Sharp: GDAL, QGIS, GeoServer, PostGIS.
Скилл Claude Code учит агента библиотеке tangermeme для анализа геномных нейросетей: DeepLIFT/SHAP, мутагенез, дизайн.
The Zvec skill guides an agent through the embedded vector database: collection schema, hybrid search, RAG and multimodal queries
A read-only MCP for the GeoLens geospatial data catalog: datasets, schemas, features, maps, stories and a safe SELECT.
A Python and R skill for publication figures with a contract, layout QA and independent export, plus the FigureForge editor
README in short
A Claude Code plugin sits in `.claude-plugin`, described as teaching agents to read embeddings through the Zarr interface. Exports record the dataset version in `tessera_metadata.json`.