Cambridge releases open AI map of Earth's land surface

University of Cambridge researchers have released a global dataset mapping every 10-metre square of Earth's land surface from 2017 to 2025. The work was supported by cloud computing infrastructure from Vultr and AMD.

The dataset was generated through TESSERA, an artificial intelligence model developed by Cambridge's Energy & Environment Group using imagery from the European Space Agency's Sentinel satellites. The model creates compressed representations of the Earth's surface, allowing large volumes of satellite data to be processed and shared more efficiently.

The university has made both the embeddings dataset and the training pipeline openly available, allowing governments, non-governmental organisations, and research groups to reproduce the model on standard cloud infrastructure rather than rely on proprietary systems. The project is intended to broaden access to environmental monitoring tools that can be adapted for local and regional use.

The mapping covers the planet's land surface over an eight-year period and can be used to track changes in agriculture, biodiversity, and renewable energy development. In the UK, potential uses include monitoring hedgerow loss, assessing crop health at field level, and mapping the spread of solar and wind installations.

Open Model

TESSERA was developed by the university's Energy & Environment Group as a foundation model for Earth observation. It processes Sentinel data at 10-metre resolution and produces pixel-level embeddings for downstream analysis, including land-use tracking and environmental change detection.

The full training process can be reproduced from scratch on Vultr infrastructure. The computing set-up included AMD Instinct MI325X graphics processors, with configurations designed to support both model training and storage of the resulting dataset.

That combination of open access and cloud-based deployment is central to the project's design. It means organisations without specialist computing estates can still build or adapt large-scale environmental models.

Kevin Cochrane, Chief Marketing Officer at Vultr, linked the infrastructure support to wider public access to artificial intelligence tools. "This collaboration positions Vultr at the intersection of cutting-edge AI infrastructure and planetary conservation," Cochrane said. "By providing the computational resources required for TESSERA and embracing open source technology, Vultr is enabling research that has immediate real-world impact, from helping smallholder farmers optimize crop yields to tracking biodiversity loss in critical ecosystems."

AMD also described the project as an example of how academic researchers can process very large environmental datasets using commercially available systems. Stephanie Dismore, Senior Vice President, EMEA at AMD, said the work depended on both computing hardware and open software.

"TESSERA shows what becomes possible when world-class research teams can scale AI with the right compute and an open software foundation. By running on Vultr cloud infrastructure powered by AMD Instinct MI325X GPUs and the AMD ROCm open software stack, University of Cambridge researchers can efficiently process petabytes of satellite data and generate planetary-scale embeddings that help turn environmental observation into actionable insight," Dismore said.

Research Uses

The university identified three main areas for early use of the model: agriculture, biodiversity monitoring, and renewable energy infrastructure. In farming, the data could support crop health analysis, yield prediction, and risk assessment at field level. For conservation work, it can be used to monitor habitat change and ecosystem conditions across a range of landscapes. In energy, it can help identify and map solar farms, wind sites, and related infrastructure.

For UK users, those applications are likely to be especially relevant to land management and environmental planning. Hedgerow networks, for example, are an important habitat feature but can be difficult to monitor consistently over large areas, while field-level crop observations may help farmers and local authorities respond more quickly to changes in land condition.

Renewable energy mapping could also support public-sector planning by showing how deployment changes over time across different regions. Because the data is drawn from a consistent satellite record, it can be used to compare places and time periods on the same basis.

Professor Anil Madhavapeddy, Professor of Planetary Computing at Cambridge's Department of Computer Science & Technology and Co-Director of the Cambridge Centre for Earth Observation, said open access was a core aim of the project. "Our goal is to democratize access to planetary-scale environmental monitoring. By making TESSERA's embeddings freely available under a CC-BY license and publishing the complete training pipeline, we're ensuring that any researcher, government, or organization worldwide can deploy this technology for their specific conservation needs," Madhavapeddy said.

The collaboration began in late 2025, and the first global embeddings have been publicly available since early 2026. "This partnership exemplifies how cloud infrastructure can accelerate scientific research with genuine planetary impact," Madhavapeddy said. "We're enabling a new paradigm for open, accessible environmental monitoring that can inform policy decisions and conservation actions in the UK and worldwide."