Software

GeoTessera Python library for accessing and working with Tessera geospatial foundation model embeddings. GeoTessera provides access to geospatial embeddings from the Tessera foundation model, which processes Sentinel-1 and Sentinel-2 satellite imagery to generate 128-channel representation maps at 10m resolution. These embeddings compress a full year of temporal-spectral features into dense representations optimized for downstream geospatial analysis tasks.
Interactive Tessera Embedding Classifier This repository contains a Jupyter notebook based tool for interactive, human-in-the-loop classification of geospatial data using the Tessera foundation model embeddings. The tool allows a user to define an area of interest, visualize the high-dimensional embedding data with PCA, and iteratively train a machine learning model by simply clicking on the map to label.

Papers

Z. Feng et al. TESSERA: Precomputed FAIR Global Pixel Embeddings for Earth Representation and Analysis, Preprint on ArXiv, July 2025.
C. Atzberger et al. A scalable, annual aboveground biomass product for monitoring carbon impacts of ecosystem restoration projects, Remote Sensing of Environment, Volume 327, 2025.
M. C. Lisaius, A. Blake, S. Keshav and C. Atzberger, Using Barlow Twins to Create Representations From Cloud-Corrupted Remote Sensing Time Series, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 13162-13168, 2024.

Please follow this link for papers in progress and under submission.

2-slide summary (PPTX) for CRI Flash Talks, S. Keshav, October 7, 2025
TESSERA overview presentation with a focus on ecological applications (PDF) University of Maryland, Frank Feng, October 1, 2025
TESSERA overview presentation (PPTX) James Cook University, S. Keshav, September 29, 2025
TESSERA overview presentation (PDF) University of Cambridge, DAB, Frank Feng, May 20, 2025
Self-supervised learning for earth observation, (PPTX) S. Keshav, Exeter, April 2025

Linked here.

First paper (2025) demonstrating the effectiveness of spectral-temporal representations for AGB monitoring in tropical countries
First paper (2024) demonstrating that spectral-temporal representations from optical EO time series can be learned fully self-supervised using the Barlow Twins loss function - demonstration on crop type classification
Master's report (2022) with the first implementation of the Barlow Twins at Cambridge University (link)
Original Barlow Twin paper by Zbontar et al. (2021) for classical still images which inspired the development of a spectral-temporal representation learning