The Swiss Data Cube team now provides access to annual geombeddings for Switzerland based on the Temporal Embeddings of Surface Spectra for Earth Representation and Analysis (TESSERA) foundation model from Cambridge University (https://github.com/ucam-eo/tessera).
These geomebeddings are now available on SwissDataCube Catalog (https://explorer.swissdatacube.org/products/tesserach) for the whole Switzerland (2024 only) and partially for years 2017-2023 (the full dataset will be completed in 2026).
Geoembeddings are a mapping from a complex input (e.g., text, image, time-series, …) to a vector (or tensor) in a lower-dimensional space, such that the vector captures meaningful structure or features of the input. The objective is to turn noisy, high-dimensional data from time-series of satellite imagery (e.g., lots of bands, many dates, missing data due to clouds, noise) into a more manageable, dense representation that preserves the important structure relevant for downstream tasks (e.g., classification, mapping, change detection). Instead of manually computing indices (e.g., NDVI), complex features, or custom heuristics, geo-embeddings provide a general-purpose representation learned by a model, hence “foundation model for Earth observation”.
You can know more about TESSERA in the following publication – Feng Z. et al. (2025) “TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysis.” https://doi.org/10.48550/arXiv.2506.20380
