Cambridge’s Tessera AI maps UK hedgehog habitats from 20 petabytes of satellite data

TL;DR:

• The University of Cambridge has developed Tessera (Temporal Embeddings of Surface Spectra for Earth Representation and Analysis), a foundation model trained on a full year of Sentinel-1 and Sentinel-2 satellite imagery, and is using it to map UK hedgehog habitats.
• Tessera was trained on roughly 20 petabytes of data; researchers initially installed extra processors under their desks to meet compute demand before securing additional capacity through a deal with chipmaker AMD and AI firm Vultr Solutions.
• More than 100 research groups are using Tessera, which produces easy-to-use UK maps that correct for day-night variation and cloud cover, and links to companion data such as backpack-GPS trackers attached to individual hedgehogs.

Hedgehog populations have declined sharply in rural Britain since 1995, and the Wildlife Trusts and People’s Trust for Endangered Species have warned that the trend continues. Cambridge’s contribution is a specific UK research-and-conservation application of a general Earth-observation foundation model — the kind of practical, sector-specific deployment that the UK research base has been good at delivering, but rarely been good at publicising at scale.

A useful counterpoint to the frontier-AI news cycle

Tessera lands the same week the Alan Turing Institute named George Williamson as CEO with a sovereign-capability brief, Anthropic disclosed it will pay SpaceX roughly £15bn a year for compute, and OpenAI claimed an internal model had disproved Erdős’s unit-distance conjecture. The headline AI conversation is increasingly about frontier-lab unit economics and reasoning-model milestones; Tessera is a reminder that UK university research is producing AI models with concrete applied outcomes that the larger conversation tends to skip past.

Prof Silviu Petrovan of the People’s Trust for Endangered Species told the BBC that powerful AI models could identify “the very specific barriers for hedgehogs to find food and find their mates”. The compute scaffolding behind the project is itself a UK research story: a team that started by installing processors under their desks, then moved to a commercial AMD-Vultr arrangement, is a microcosm of the UK research-computing access problem that the AI Opportunities Action Plan and the Turing’s incoming CEO will need to address. A foundation model designed for Earth-observation tasks — land-cover classification, solar-panel identification, habitat mapping — is also the kind of dual-use tool that has applications in agriculture, planning, flood-risk assessment and infrastructure monitoring beyond conservation. UK SMEs working in agritech, planning consultancy and renewables siting are the natural commercial audience.

Looking forward

Watch for whether the Cambridge team formalises Tessera as a public-research utility or commercialises it as a spin-out, and how UKRI handles funding requests for similar applied foundation models in the next call cycle. For the People’s Trust for Endangered Species and other UK conservation bodies, the model’s value is now operational: where Tessera flags habitat threats — new housing developments, flooding, hedgerow loss — interventions can be prioritised more efficiently than with manual mapping alone. For UK businesses, the broader read is that “AI” need not mean LLMs from US labs; UK research groups are producing applied foundation models with concrete outputs in sectors that matter.