Google has released a new Satellite Embedding dataset that uses AI to compress a year's worth of multi-source satellite data into 64-dimensional vectors for every 10-meter pixel on Earth. Created with Google DeepMind's AlphaEarth Foundations model, these embeddings combine optical imagery, radar data, elevation models, climate information, and text descriptions into analysis-ready features. The dataset enables similarity searches to find locations with similar environmental conditions, change detection by comparing embeddings across years, automatic clustering to discover hidden patterns, and more accurate classification with less training data. Available through Google Earth Engine, the dataset covers global terrestrial surfaces from 2017 onwards and eliminates the need for traditional preprocessing steps like atmospheric correction and cloud masking.
Table of contents
AI-powered pixels: Introducing Google’s Satellite Embedding datasetWhat’s embedded in an embedding?Earth in 64 dimensions: Coordinates versus bandsSo what can you do with the Satellite Embedding dataset?Find other places like this: Similarity searchDetecting change: Tracking shifts in embedding spaceDiscover hidden patterns: Automatic clusteringCreate detailed maps with less manual labelingBringing AI to Earth EngineSort: