TESSERA Earth Embeddings
TESSERA (Temporal Embeddings of Surface Spectra for Earth Representation and Analysis) is an open Earth observation foundation model and dataset from the University of Cambridge, built by the Energy and Environment Group in the Department of Computer Science and Technology and published with dClimate on the AWS Registry of Open Data. For every 10 metre land pixel it produces one annual 128 dimensional embedding that fuses a full year of Sentinel-1 radar and Sentinel-2 optical time series into a single compact vector, so downstream work can start from a learned descriptor rather than raw scenes. Global annual layers are provided for 2017 to 2025 and are updated as new Sentinel data lands.
You can use the ready-made embeddings without running the model. The simplest route is the geotessera Python client (pip install geotessera, source on the ucam-eo GitHub organisation), which fetches tiles for a region of interest. The full product is also hosted as an Icechunk Zarr store in a public AWS S3 bucket that needs no account, so you can list and pull it with the AWS CLI in no-sign-request mode or browse it through the registry's web explorer.
For RAG and AI the appeal is that the heavy feature engineering is already done. The vectors are analysis ready inputs for land cover and crop mapping, biomass and carbon estimation, habitat monitoring and change detection, and they drop straight into a vector index for geospatial similarity search and nearest-neighbour lookups over places. This is numeric, imagery derived data rather than text, so pair it with tabular or documentary sources when you need grounded textual answers.
On licensing, the embeddings carry CC BY-SA 4.0, so you must credit the project and release any derived or combined products under the same share-alike terms, which can be restrictive if you plan to fold them into a proprietary pipeline. The model weights are separately CC0 and the client is MIT, and a TESSERA Acceptable Use Policy also applies. It is the open, reproducible counterpart to Google's AlphaEarth satellite embeddings, which we also list.
Related sources
Cohere Wikipedia 22-12 Embeddings (per-language)
Wikipedia encoded with the Cohere multilingual-22-12 embedding model, with embeddings (numeric vectors that capture meaning) computed on each article's title plus text. Published per language, covering many including Arabic, German, English, Spanish, French, Hindi, Italian, Japanese, Korean, Simple English, and Chinese. You can stream it rather than download it in full, which matters given the size.
Cohere Wikipedia Multilingual Embeddings (2023-11)
The November 2023 Wikipedia dump across 300+ languages, split into passages and turned into embeddings (numeric vectors that capture meaning) with the Cohere Embed V3 multilingual model. Close to 250 million paragraph embeddings in total. Drop it into a vector database for semantic search over all of Wikipedia, or use it as a knowledge source for multilingual RAG, including cross-lingual search where a query in one language returns relevant results from another.
Google Satellite Embedding (AlphaEarth Foundations)
Global annual satellite embeddings produced by Google and Google DeepMind's AlphaEarth Foundations model. Every 10 metre pixel holds a 64 dimensional vector summarising a year of multi-sensor Earth observation, available yearly from 2017.
LAION-400M with CLIP Embeddings
400 million image-text pairs filtered from Common Crawl using CLIP, a model that scores how well an image matches a caption, and released alongside their pre-computed CLIP embeddings and kNN indices for fast similarity search. The ready-made embeddings make it unusually easy to work with. Note that it ships image URLs, not the images themselves, so some links have decayed.