# TESSERA

> A geospatial foundation model providing satellite-derived embeddings for Earth observation.

This index links to full Markdown versions of published articles hosted on this site. External coverage is not reproduced. Interactive visualisations require JavaScript; their surrounding text and image descriptions are included in Markdown.

## Blog

- [Cloud Native Geospatial Forum 2026 Outtakes](https://geotessera.org/blog/2026-10-14-cng-forum-2026-outtakes.md): Notes from presenting global, wall-to-wall Tessera at the 2026 Cloud Native Geospatial Forum: why cloud-native embeddings matter to users, what they want next on provenance and multiscale outputs, and early excitement for v2.
- [Global 2017-2025 v1.1 embeddings now available in geotessera 0.11.0](https://geotessera.org/blog/2026-10-05-geotessera-0-11.md): GeoTessera 0.11 defaults to a complete global run of v1.1 embeddings, streamed straight from AWS. Open the dataset, point it at a region, and read.
- [Classifying Finnish forest and mire types with Tessera embeddings using tessera-eval](https://geotessera.org/blog/2026-09-29-tessera-eval-finnish-forest-mire.md): The Python library tessera-eval, from the University of Cambridge Tessera team lets users train a model on Tessera embeddings easily and quickly. In this demonstration, Silja Sormunen uses tessera-eval's command-line tool to classify forest and wetland types across a region of northern Finland.
- [Evaluating your downstream task with Tessera embeddings](https://geotessera.org/blog/2026-09-09-evaluating-downstream-tasks-with-tessera-embeddings.md): tessera-eval turns a ground-truth shapefile into an honest accuracy estimate for Tessera embeddings — cross-validation, learning curves, and spatial splits.
- [Tessera Postcards of Anywhere on Earth](https://geotessera.org/blog/2026-08-07-tessera-postcards.md): The Tessera Postcard generator renders any 10km x 6km patch of land on Earth straight from the model's embeddings, in a range of colour palettes. Pick a spot that means something to you and send someone a postcard of how Tessera sees it.
- [New Faces: Kyle Gao](https://geotessera.org/blog/2026-07-21-new-faces-kyle-gao.md): Kyle Gao joins Tessera as an academic visitor from Aalto University, working on 3D reconstruction, multi-modal remote sensing, and geoAI.
- [New Faces: Jingtao Li](https://geotessera.org/blog/2026-07-11-new-faces-jingtao-li.md): Jingtao Li joins the TESSERA team as a research associate, focused on anomaly detection and foundation model techniques for remote sensing imagery.
- [New Faces: Silja Sormunen](https://geotessera.org/blog/2026-07-11-new-faces-silja-sormunen.md): Silja Sormunen joins the TESSERA team as a research associate, bringing a background spanning psychology, medicine, physics and network science to the study of Earth embeddings.
- [TESSERA v1.1: more stable embeddings, fewer artefacts](https://geotessera.org/blog/2026-06-09-tessera-v1-1.md): v1.1 embeddings are more temporally stable year-on-year and have far fewer artefacts in low-observation areas, and nothing in your workflow has to change.
- [TESSERA at the FP Launchpad kickoff at IIT Madras](https://geotessera.org/blog/2026-05-01-fp-launchpad-iit-madras.md): A talk on functionally programming petabytes of Earth observations at the launch of the FP Launchpad Centre for Functional Systems Research and Education at IIT Madras.
- [Calibrated biomass and below-ground biodiversity with TESSERA](https://geotessera.org/blog/2026-04-17-gedi-fungi-preprints.md): Two new preprints use TESSERA embeddings for calibrated biomass gap-filling and below-ground fungal biodiversity prediction.
- [Agribound: agricultural field boundary delineation](https://geotessera.org/blog/2026-04-01-agribound-field-boundaries.md): Agribound is a new Python toolkit that uses TESSERA embeddings for unsupervised agricultural field boundary delineation across multiple satellite sources.
- [Planetary scale training and inference with AMD and Vultr](https://geotessera.org/blog/2026-03-30-training-and-inference-at-scale.md): How we used AMD Accelerator Cloud and Vultr's MI325X GPU clusters to train the TESSERA v1 model and generate 250TB of global embeddings.
- [Restoration Dialogue: AI and Habitat Monitoring in the UK](https://geotessera.org/blog/2026-03-23-restoration-dialogue-workshop.md): Over 30 researchers from universities, statutory agencies and NGOs gathered in Cambridge to explore whether geospatial foundation models like TESSERA can support mapping and monitoring nature across the UK.
- [GeoTessera R library now available](https://geotessera.org/blog/2026-03-16-geotessera-r-library.md): Simon Frost from Microsoft Research has contributed an R port of the geotessera library, bringing TESSERA embeddings to the R ecosystem.
- [TESSERA v1 model weights now available](https://geotessera.org/blog/2026-03-10-tessera-v1-weights.md): The TESSERA v1 model weights are now publicly available, including the QAT checkpoint for quantized int8 inference.
- [TESSERA accepted at CVPR 2026](https://geotessera.org/blog/2026-03-11-introducing-tessera.md): Our paper introducing TESSERA, a pixel-wise foundation model for multi-modal Earth observation time series, has been accepted at CVPR 2026.

## News

- [Mapping Senegal's small farms from space, with a fraction of the data](https://geotessera.org/news/2026-09-29-senegal-smallholder-crops.md): New Cambridge research shows Tessera can map crops on Senegal's smallholder farms more accurately than current methods, using a fraction of the computing power and ground data.
- [A year of Tessera embeddings, and GeoTessera 0.10](https://geotessera.org/news/2026-08-27-geotessera-a-year-on.md): The geotessera library turns one, and version 0.10 moves the embedding archive onto Source Cooperative with CloudFlare edge caching, adds UTM-native Zarr routing and matryoshka embedding support.
- [A faster, cheaper way to map the world's forests with AI](https://geotessera.org/news/2026-07-29-trentino-tree-species.md): Cambridge researchers used Tessera, a geospatial foundation model, to map tree species across the mountainous Trentino region in Italy, outperforming traditional satellite methods and opening the door to low-cost, continent-scale forest mapping.
- [Powering Tessera, the AI that sees Earth in 128 dimensions](https://geotessera.org/news/2026-07-22-vultr-amd-partnership.md): A substantial portion of TESSERA's global embeddings was produced on Vultr cloud infrastructure powered by AMD Instinct MI325X GPUs, part of a new partnership announced this week at AMD's Advancing AI Day summit.
- [TESSERA heads to Toronto to map the farms satellites can't see](https://geotessera.org/news/2026-07-02-smallholder-farms.md): New University of Cambridge research shows the TESSERA AI model can classify crops in the world's smallest fields more accurately than current methods, using a fraction of the computing power.
- [TESSERA arrives in Denver for its CVPR Debut](https://geotessera.org/news/2026-06-04-cvpr-earth-intelligence.md): A new AI model developed at the University of Cambridge distils millions of satellite images into a format that is easily used by researchers to reveal crop patterns, habitat loss, deforestation, and more.
- [Cambridge's TESSERA Helping to Protect UK Hedgehogs from Space](https://geotessera.org/news/2026-05-19-tessera-hedgehogs_1.md): Researchers at the University of Cambridge have teamed up with the Cambridge Conservation Initiative to use TESSERA's satellite AI to map hedgehog habitats across the UK.