https://geotessera.org/blog Updates, tutorials, and research from the TESSERA community 2026-04-17T00:00:00.000Z https://geotessera.org/blog/2026-04-17-gedi-fungi-preprints 2026-04-17T23:00:00.000Z Robin Young

Two new preprints use TESSERA embeddings for calibrated biomass gap-filling and below-ground fungal biodiversity prediction.

https://www.tunbury.org/2026/04/08/intel-amx/ 2026-04-08T23:00:00.000Z Mark Elvers

Benchmarking the TESSERA pipeline on Intel Xeon with AMX bfloat16 acceleration, finding an 8-core AMX CPU achieves ~$0.29/tile versus $0.23/tile on a T4 GPU at similar price points.

https://github.com/ucam-eo/geotessera/releases/tag/v0.8.0 2026-04-05T23:00:00.000Z geotessera

Adds cloud-native Zarr v3 access via GeoTesseraZarr with automatic UTM zone routing and point sampling, fixes GeoTIFF memory leaks for large regions, adds download resume for GeoTIFFs, and improves registry scanning with truncated file detection and selective parquet generation. Next up: extended registry support for multiple mirrors, versions, and formats.

https://geotessera.org/blog/2026-04-01-agribound-field-boundaries 2026-04-01T23:00:00.000Z Anil Madhavapeddy

Agribound is a new Python toolkit that uses TESSERA embeddings for unsupervised agricultural field boundary delineation across multiple satellite sources.

https://geotessera.org/blog/2026-03-30-training-and-inference-at-scale 2026-03-30T23:00:00.000Z Anil Madhavapeddy

How we used AMD Accelerator Cloud and Vultr's MI325X GPU clusters to train the TESSERA v1 model and generate 250TB of global embeddings.

https://anil.recoil.org/notes/tessera-embeddings-convention 2026-03-27T23:00:00.000Z Anil Madhavapeddy

TESSERA will migrate from separate annual stores to a unified MegaZarr with time as a primary dimension, adopting the emerging geo-embeddings Zarr convention for broader compatibility across embedding models.

https://www.linkedin.com/pulse/from-pixels-embeddings-tessera-earth-observation-gijs-van-den-dool-uti7e/ 2026-03-25T23:00:00.000Z Gijs van den Dool

A contributed walkthrough of a custom Master-Worker pipeline for running TESSERA inference in Google Colab, with analysis of optimal cloud cover thresholds for West Africa.

https://github.com/ucam-eo/TEE/releases/tag/v3.5.0 2026-03-24T23:00:00.000Z Srinivasan Keshav

Two releases of the TESSERA Embedding Explorer: v3.4 adds manual labelling with pin and polygon tools, per-class similarity sliders, and spatial classifiers. v3.5 adds a multi-user login system to the demo site — request access via the TESSERA channel on eeg.zulipchat.com. See also the walkthroughs of TEE in the videos section.

https://geotessera.org/blog/2026-03-23-restoration-dialogue-workshop 2026-03-23T23:00:00.000Z David Coomes

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.

https://isaac.earth/geospatial-skills 2026-03-20T23:00:00.000Z Isaac Corley

A set of Claude Code skills for working with TESSERA embeddings via the Python and R libraries, enabling AI-assisted geospatial analysis workflows.

https://geotessera.org/blog/2026-03-16-geotessera-r-library 2026-03-16T23:00:00.000Z Anil Madhavapeddy

Simon Frost from Microsoft Research has contributed an R port of the geotessera library, bringing TESSERA embeddings to the R ecosystem.

https://anil.recoil.org/notes/tessera-zarr-v3-layout 2026-03-14T23:00:00.000Z Anil Madhavapeddy

How TESSERA restructured its geospatial embeddings from millions of individual files into sharded Zarr v3 stores, enabling efficient HTTP streaming via range requests for both single-pixel mobile queries and large-scale regional analysis.

https://www.tunbury.org/2026/03/13/oxcaml-inference/ 2026-03-13T23:00:00.000Z Mark Elvers

Building a zero-allocation ONNX inference engine for the TESSERA model using OxCaml's unboxed float32 types and SIMD intrinsics. Achieves ~200ms single-threaded latency, within 2.2x of ONNX Runtime.

https://github.com/ucam-eo/TEE/releases/tag/v3.2.0 2026-03-13T22:00:00.000Z Srinivasan Keshav

Adds a GPU-accelerated U-Net classifier operating on full 2D embedding grids via PyTorch with CUDA, MPS, and CPU support. Introduces 4× flip augmentation for Spatial MLP 3×3 and 5×5 variants, plus improved error handling for non-JSON server responses.

https://www.tunbury.org/2026/03/11/gpu-vs-cpu/ 2026-03-11T23:00:00.000Z Mark Elvers

Benchmarking TESSERA inference on GPU vs CPU, finding that NUMA-aware parallel CPU jobs can achieve 2x the aggregate throughput of a single GPU for satellite tile processing.

https://geotessera.org/blog/2026-03-10-tessera-v1-weights 2026-03-10T23:00:00.000Z Zhengpeng Feng

The TESSERA v1 model weights are now publicly available, including the QAT checkpoint for quantized int8 inference.

https://github.com/ucam-eo/TEE/releases/tag/v3.0.0 2026-03-09T23:00:00.000Z Srinivasan Keshav

New validation panel with learning-curve evaluation and tunable hyperparameters. Moved UMAP computation to a client-side Web Worker, replaced GeoTIFF pyramids with PNG+JSON, added granular pipeline progress, and improved label management with a split-view panel.

https://jon.recoil.org/blog/2026/03/weeknotes-2026-10.html 2026-03-08T23:00:00.000Z Jon Ludlam

An interactive TESSERA notebook running in the browser using OCaml, js_top_worker and odoc plugins — enabling geospatial mapping and ML inference with ONNX web runtime bindings.

https://anil.recoil.org/notes/2026w10 2026-03-08T22:00:00.000Z Anil Madhavapeddy

A browser-based streaming interface for TESSERA satellite embeddings performing client-side analysis using WebGPU and WebAssembly.

https://fplaunchpad.org/2026/03/06/applications-open-post-bacc-fellowship.html 2026-03-06T23:00:00.000Z FP Launchpad https://arxiv.org/abs/2603.02080 2026-03-02T23:00:00.000Z Isaac Corley et al.

Introduces EuroSAT-Embed, a dataset of 81k embedding GeoTIFFs from three models, and evaluates 13 pooling approaches. Recommends Generalized Mean Pooling as an efficient alternative to mean pooling, reducing geographic generalization gap by up to 40%.

https://anil.recoil.org/notes/2026w9 2026-03-01T23:00:00.000Z Anil Madhavapeddy

Progress on TESSERA with Zarr streaming support enabling browser-based ML pipelines, plus a Docker paper making the CACM cover.

https://cloudnativegeo.org/blog/2026/02/the-technical-debt-of-earth-embedding-products/ 2026-02-28T23:00:00.000Z Isaac Corley https://www.tunbury.org/2026/02/25/teserra-pipeline/ 2026-02-25T23:00:00.000Z Mark Elvers

A technical walkthrough of the Tessera pipeline, processing Sentinel-1 and Sentinel-2 satellite imagery to generate machine learning embeddings.

https://core-stack.org/first-round-of-innovation-challenge-advances/ 2026-02-23T23:00:00.000Z CoRE Stack https://geotessera.org/blog/2026-03-11-introducing-tessera 2026-02-21T23:00:00.000Z Zhengpeng Feng

Our paper introducing TESSERA, a pixel-wise foundation model for multi-modal Earth observation time series, has been accepted at CVPR 2026.

https://github.com/ucam-eo/TEE/releases/tag/v2.0.0 2026-02-20T23:00:00.000Z Srinivasan Keshav

Major release adding per-user authentication with bcrypt, disk quotas, read-only demo mode, high-res map export as JPG, label export with 128-dim embeddings, satellite source selector, and Gunicorn/Apache production deployment.

https://core-stack.org/tessera-and-core-stack-hackathon-on-geospatial-ai/ 2026-02-19T23:00:00.000Z CoRE Stack https://anil.recoil.org/notes/first-tessera-hackathon 2026-02-19T22:00:00.000Z Anil Madhavapeddy

The inaugural TESSERA hackathon at the Indian AI Impact Summit in Delhi, exploring integration with IIT-Delhi's CoRE Stack for environmental monitoring.

https://www.tunbury.org/2026/02/15/ocaml-tessera/ 2026-02-15T23:00:00.000Z Mark Elvers

An OCaml implementation of the Tessera pipeline achieving 9.5x faster inference than the Python CPU baseline with GPU acceleration.

https://ancazugo.github.io/posts/2026-02-15-weekly-notes.html 2026-02-15T22:00:00.000Z Andres C. Zuniga-Gonzalez

Work on Local Climate Zone classification using satellite imagery, challenges with dataset labeling, and progress on OpenStreetMap road mapping.

https://terrabytes.substack.com/p/tessera-a-blueprint-for-earth-observation 2026-02-15T21:00:00.000Z Rakshith Sathish

An independent review of TESSERA's architecture, training strategy, and open-source accessibility, with analysis of its strengths and areas for improvement.

https://github.com/ucam-eo/geotessera/releases/tag/v0.7.5 2026-02-15T20:00:00.000Z geotessera

Reduced startup time, auto-snap coordinates to valid tile centers, switched to pandas MultiIndex lookups, and split coverage data into per-year files for the globe viewer.

https://anil.recoil.org/notes/2026w6 2026-02-08T23:00:00.000Z Anil Madhavapeddy

A doctoral viva, presenting research projects at ARIA, media coverage of conservation evidence work, and a CACM video interview.

https://github.com/ucam-eo/TEE/releases/tag/v1.4.0-stable 2026-02-04T23:00:00.000Z Srinivasan Keshav

Added PCA as a fast alternative to UMAP for 3D embedding projection, enabling quicker dimensionality reduction for interactive exploration.

https://anil.recoil.org/notes/red-pill-conservation 2026-02-03T23:00:00.000Z Anil Madhavapeddy

Hosting UK nature conservation body leaders at Pembroke to present TESSERA, focusing on evidence-based approaches and transparent AI for conservation.

https://www.youtube.com/watch?v=I4sBoCKMidA 2026-01-28T23:00:00.000Z GIS & Research Group

Overview of the TESSERA foundation model and its approach to encoding the Earth's surface as a new language for geospatial analysis.

https://github.com/ucam-eo/geotessera/releases/tag/v0.7.4 2026-01-27T23:00:00.000Z geotessera

New --tile option for single-tile download and coverage queries, enhanced --bbox supporting both formats, and resolved README/LICENSE mismatch.

https://anil.recoil.org/notes/2026w4 2026-01-25T23:00:00.000Z Anil Madhavapeddy

Porting TESSERA to Zarr, a new preprint on standardising geoembedding access, SIDL podcast coverage, and TESSERA support merged into TorchGeo.

https://youtu.be/9yrpwFrwbGY?t=645 2026-01-23T23:00:00.000Z Zhengpeng Feng

Presentation at the AI for Good seminar introducing TESSERA's temporal embedding approach for satellite imagery analysis.

https://www.satellite-image-deep-learning.com/p/tessera-a-temporal-foundation-model 2026-01-21T23:00:00.000Z Sadiq Jaffer, Zhengpeng Feng

Podcast with Robin Cole covering temporal embeddings, self-supervised training at petabyte scale, wildfire detection, and plans for TESSERA v2.

https://arxiv.org/abs/2601.13134 2026-01-19T23:00:00.000Z Heng Fang et al. https://about.conservationevidence.com/2026/01/16/geospatial-foundation-models/ 2026-01-16T23:00:00.000Z Julia P. G. Jones

How geospatial foundation models like TESSERA could transform conservation by integrating satellite data with ecological information to improve habitat mapping and policy decisions.

https://ai4nature.org/the-alliance-journal/usot739onqmh3n43tgt3r8b9m6j3ye 2026-01-14T23:00:00.000Z Anastasia Filatova

The Ai4Nature Alliance launch event bringing together ecological and technology leaders to establish frameworks for using AI responsibly in nature recovery.

https://open.spotify.com/episode/3ZlkQoPsO3XRHxA3Hdc7Sr 2026-01-14T22:00:00.000Z AMD TechTalk https://toao.com/blog/earth-observation-budget-solar-farms-tiny-model 2026-01-14T21:00:00.000Z Sadiq Jaffer

Detecting and mapping solar farms across the UK using satellite imagery and a compact 42k-parameter neural network, leveraging pre-trained TESSERA embeddings.

https://www.linkedin.com/posts/anilmadhavapeddy_team-tessera-has-open-planetary-embeddings-activity-7413185711761170432-imfr 2026-01-03T23:00:00.000Z Anil Madhavapeddy https://github.com/ucam-eo/geotessera/releases/tag/v0.7.3 2025-12-17T23:00:00.000Z geotessera

Retired Pooch text manifests in favour of Parquet, added tolerance for incomplete embedding directories, and improved warning diagnostics.

https://www.linkedin.com/posts/clement-atzberger-8abb8065_fun-question-what-is-the-meaninginterpretation-activity-7403754770894835712-cMfR 2025-12-08T23:00:00.000Z Clement Atzberger https://www.linkedin.com/posts/ieee-grss_ieeegrss-grsswebinar-esi-activity-7402335630942527489-R8qn 2025-12-04T23:00:00.000Z IEEE GRSS https://www.swissdatacube.org/index.php/2025/12/03/geoembeddings-for-switzerland/ 2025-12-03T23:00:00.000Z Swiss Data Cube https://anil.recoil.org/notes/foundational-ecosystem-workshop 2025-12-03T22:00:00.000Z Anil Madhavapeddy

Report from two Cambridge Conservation Initiative workshops exploring how TESSERA and LLM-based simulations can enhance ecosystem resilience forecasting and conservation policy.

https://www.tunbury.org/2025/12/02/tessera-stac/ 2025-12-02T23:00:00.000Z Mark Elvers

A scalable tile server using Parquet files and Arrow libraries to efficiently query geographic tiles through a simple API.

https://github.com/ucam-eo/geotessera/releases/tag/v0.7.2 2025-12-02T22:00:00.000Z geotessera

Added Windows platform support with conda CI and PowerShell tests, fixed coordinate printing order, and resolved tile discovery false negatives.

https://www.tunbury.org/2025/11/30/tessera-zarr/ 2025-11-30T23:00:00.000Z Mark Elvers

Exploring TESSERA's NPY tile format and converting to Zarr for cloud-native chunked storage, with a survey of STAC and CF Conventions for standardised metadata.

https://www.youtube.com/watch?v=u3_HxRJE8Ss 2025-11-30T22:00:00.000Z Geoawesome

Interview covering TESSERA's open-source philosophy, the engineering of a global-scale geospatial foundation model, and plans for the future.

https://watch.eeg.cl.cam.ac.uk/w/hanDLh2nzjXY4yL1akVEYh 2025-11-24T23:00:00.000Z Srinivasan Keshav, David A. Coomes

Workshop presentation at Pembroke College on applying foundational AI methods to predict ecosystem health and resilience.

https://github.com/ucam-eo/geotessera/releases/tag/v0.7.1 2025-11-19T23:00:00.000Z geotessera

New --format zarr option for cloud-native downloads with xarray integration, built-in compression, and georeferencing metadata. Added registry scan command.

https://anil.recoil.org/notes/geotessera-python-0-7 2025-11-17T23:00:00.000Z Anil Madhavapeddy

GeoTessera 0.7 transitions to GeoParquet databases for faster initialization, adds Zarr format support for distributed storage, and introduces new sampling APIs for downstream tasks like solar panel detection.

https://ancazugo.github.io/posts/2025-11-16-tessera_example.html 2025-11-16T23:00:00.000Z Andres C. Zuniga-Gonzalez https://github.com/ucam-eo/geotessera/releases/tag/v0.7.0 2025-11-11T23:00:00.000Z geotessera

Major release transitioning to Parquet-based registry, adding point sampling and mosaic APIs, SHA256 verification, lazy iterators, 3D globe coverage viewer, and CRAM test suite.

https://watch.eeg.cl.cam.ac.uk/w/trq77gFwc8ZbUeGqyDpzN7 2025-10-05T23:00:00.000Z Sadiq Jaffer

PROPL workshop talk at ICFP/SPLASH demonstrating how GeoTessera makes planetary-scale remote sensing accessible, including a live solar farm classification demo.

https://anil.recoil.org/notes/icfp25-propl 2025-10-05T22:00:00.000Z Anil Madhavapeddy

Report from the second PROPL workshop in Singapore, featuring Sadiq Jaffer's talk on TESSERA and building classifiers for solar farms in minutes on a laptop.

https://arstechnica.com/ai/2025/09/can-ai-detect-hedgehogs-from-space-maybe-if-you-find-brambles-first/ 2025-09-26T23:00:00.000Z Benj Edwards https://toao.com/blog/can-we-really-see-brambles-from-space 2025-09-24T23:00:00.000Z Sadiq Jaffer

A field validation study testing whether a machine learning model using TESSERA embeddings can accurately predict bramble locations from satellite imagery.

https://www.linkedin.com/posts/samuel-barrett-b86b85171_finding-silage-storage-with-tessera-embeddings-ugcPost-7369734079144955908-kdzM 2025-09-05T23:00:00.000Z Samuel Barrett https://anil.recoil.org/notes/geotessera-python 2025-08-31T23:00:00.000Z Anil Madhavapeddy

Announcing geotessera, a Python library and CLI for downloading and visualising TESSERA embeddings, with coverage tools and false-colour rendering.