TESSERA is an Earth observation foundation model developed at the University of Cambridge. Its name stands for Temporal Embeddings of Surface Spectra for Earth Representation and Analysis. The project transforms a year of Sentinel-1 radar and Sentinel-2 optical satellite observations into 128-dimensional embeddings for each 10-metre pixel, capturing patterns in landscapes over time rather than a single snapshot. A self-supervised training approach learns from satellite data without human labels, helping users achieve strong results on downstream tasks with relatively few labelled examples.

TESSERA provides open data, model weights and precomputed embeddings for research and practical applications. Its embeddings can support tasks such as land-cover classification, segmentation and regression, including identifying crops, forests, water, buildings and solar panels. The project also offers documentation, tutorials, papers and tools such as an Embeddings Explorer to help users explore coverage and apply the data. Current releases include global terrestrial coverage for 2017–2025, with newer weights and streaming formats in development.