{"slug": "tessera-v1-1-released-with-smoother-and-temporally-stable-embeddings", "title": "Tessera v1.1 released, with smoother and temporally stable embeddings", "summary": "The Tessera team released v1.1 of their geospatial embedding model, improving temporal stability and reducing noise in regions with sparse satellite observations. The update eliminates grid-like seams and stabilizes embeddings over time, benefiting large-scale analyses and change detection tasks.", "body_md": "[Frank Feng](https://www.cst.cam.ac.uk/people/zf281) and I [announced TESSERA v1.1](https://geotessera.org/blog/2026-06-09-tessera-v1-1)\non behalf of [the team](https://geotessera.org/about#:~:text=for%2520Science%2520%C2%B7%2520Isambard-,People,-Lead%2520Faculty) earlier this week, and I wanted to follow up here with a more\nvisual explanation of what changed as I got quite a few questions about it!\n\nv1.1 is a retrained successor to the [original v1.0 model](https://anil.recoil.org/papers/2025-tessera) that\n[Frank Feng](https://www.cst.cam.ac.uk/people/zf281) and the team have been hammering on for months. Crucially, since we\npre-generate embedding 'map tiles', the new release is a drop-in replacement if\nyou just swap tiles; the basic format of 128 dimensions is unchanged. Accuracy\nof your tasks should improve in all cases (a trend which will continue as we\ntrain better models with more data and training FLOPS).\n\nTessera v1.0 could sometimes produce noisier tiles in regions with few clear satellite observations (e.g. due to persistent cloud or satellite sensor gaps). This exhibited as boundary-like seams in the tiles where the inferred embeddings didn't quite align; e.g. along Sentinel-1 ascending/descending coverage edges where one side of the line might have ~50 valid observations and the other ~150.\n\nTessera v1.1 now handles both sparse and imbalanced observation patterns gracefully! If your region of interest was small and didn't straddle a problematic tile you'll see no difference, but large-scale analyses should get cleaner.\n\nThe easiest way to see all this is to look at the embeddings themselves in the\n[TZE explorer](https://tze.geotessera.org). In this video I flip between the\nv1.0 and v1.1 embeddings over the same regions, visualised in false colour:\n\nWhat you're looking at in the v1.0 layer are the grid-like seams running through an otherwise homogeneous landscape (Ireland doesn't really have those jagged lines, you can confirm by visiting my lovely home).\n\nWhat's happening is that the number of valid observations jumps across the line, and the old v1.0 model showed that difference up into the embeddings. The speckly patches are areas where persistent cloud left the model with too few clean observations to produce a stable representation.\n\nWe then switch to the v1.1 layer, and the seams are gone and the noisy patches\nresolve into a smooth structure that follows the actual land cover. It's *very*\nsatisfying to click around the 10m² pixels and watch embeddings that used to\nflicker between years settle down into stable trajectories in [the explorer](https://tze.geotessera.org)!\n\nIf you're doing analysis over a long period of time, then the 128-dimensional embeddings are now much more consistent year-on-year for the same location. This is a big deal for tasks like change detection, trend analysis, and even just convenience since training a classifier on one year and applying it to another is now much more accurate.\n\nThis feature won't affect most users, but we're pretty pleased with how well change detection now works.\n\nThe v1.0 land mask we used to mask out ocean areas was too aggressive, and\ndropped legitimate land pixels along coastlines or on small islands. We've\nlistened to our [mangrove-loving friends](https://www.plantsci.cam.ac.uk/staff/dr-thomas-worthington) and extended the inference\nbuffer to 20km, which brings coastlines and remote islands properly into\ncoverage.\n\nI updated the [live coverage map](https://ucam-eo.github.io/tessera-coverage-map/) to now\ntrack both generations side-by-side, so you can see exactly which tiles exist\nfor v1.0 and v1.1 in any given year:\n\nThis is all updated via a [GitHub Action on ucam-eo/tessera-coverage-map](https://github.com/ucam-eo/tessera-coverage-map/blob/main/.github/workflows/map.yml)\nthat also updates an index Parquet file of all available manifests.\n\nTo get the new embeddings, grab the [geotessera 0.9.0+ release](https://github.com/ucam-eo/geotessera/releases/tag/v0.9.0) of the\n[Python library](https://anil.recoil.org/notes/geotessera-python) which went out alongside v1.1. It has a new\n`--dataset-version`\n\nflag to pick v1.0 or v1.1, and a `--dataset-variant`\n\nflag\nnow that multiple parties are generating embeddings for the community:\n\n`vultr`\n\nis the original `cambridge`\n\nis our Use [uvx](https://docs.astral.sh/uv/) to try this without any installation:\n\n```\nuvx geotessera download \\\n  --country \"United Kingdom\" \\\n  --year 2024 \\\n  --dataset-version v1.1 \\\n  --dataset-variant cambridge \\\n  --format npy \\\n  --output ./uk-v1.1\n```\n\nAll the embeddings (both versions) are also now in the `s3://tessera-embeddings`\n\npublic bucket on AWS Open Data, which geotessera 0.9 switches to by default.\nSpare a kind thought for \"okavango\", our single overworked Cambridge server that served every\nTESSERA embedding for the first six months without falling over (much)!\nBut seriously, at some point, we're going to have to turn off `okavanago' as it's\ntaking up a significant amount of the egress bandwidth for Cambridge, so I encourage\nusers to upgrade to geotessera 0.9 as soon as possible just to change the source\nof your embeddings download. Let me know if you have any problems!\n\nWe're also now on [Hugging Face](https://huggingface.co/geotessera/TESSERA-V-1.1)\nwith the full v1.1 (and [v1.0](https://huggingface.co/geotessera/TESSERA-V-1.0))\nmodel weights, with checkpoints for both the Microsoft Planetary Computer and\nAWS Open Data preprocessing backends. If you'd rather run inference yourself\nor fine-tune on your own data, everything you need is there, all under CC0 as\nusual. Do [let us know](https://eeg.zulipchat.com) if you fine-tune a model as\nwe'd love to see how it goes.\n\nIf there's a region of the world you need for your own research urgently,\nplease do [request an ROI](https://github.com/ucam-eo/geotessera/issues) on the\ngeotessera issue tracker and we'll prioritise it in the generation queue.\nOtherwise, sit tight as we'll have full global 2017-2025 coverage within a few\nmonths!\n\nSee also coverage from the [Meteorological Technology trade magazine](https://www.meteorologicaltechnologyinternational.com/news/satellites/cambridge-ai-tool-converts-satellite-archives-into-accessible-earth-intelligence.html) about the release.", "url": "https://wpnews.pro/news/tessera-v1-1-released-with-smoother-and-temporally-stable-embeddings", "canonical_source": "https://anil.recoil.org/notes/tessera-v11-out", "published_at": "2026-06-12 00:00:00+00:00", "updated_at": "2026-06-16 06:24:04.896369+00:00", "lang": "en", "topics": ["machine-learning", "computer-vision", "ai-products"], "entities": ["Tessera", "Frank Feng", "University of Cambridge", "Sentinel-1"], "alternates": {"html": "https://wpnews.pro/news/tessera-v1-1-released-with-smoother-and-temporally-stable-embeddings", "markdown": "https://wpnews.pro/news/tessera-v1-1-released-with-smoother-and-temporally-stable-embeddings.md", "text": "https://wpnews.pro/news/tessera-v1-1-released-with-smoother-and-temporally-stable-embeddings.txt", "jsonld": "https://wpnews.pro/news/tessera-v1-1-released-with-smoother-and-temporally-stable-embeddings.jsonld"}}