AlphaEarth is a global geospatial embedding dataset released by Google DeepMind and Google Earth Engine team in 2025.AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data", which aims to compress multi-source remote sensing and geographic data into unified, reusable spatiotemporal embeddings, and complete mapping and monitoring more efficiently under conditions of scarce annotations.

This dataset consists of annual global embedding layers from 2017 to 2024, with each 10-meter pixel corresponding to a 64-dimensional unit-length vector. It covers all land and shallow water areas worldwide. The current public version is annotated as AEFv2.1. The dataset has global coverage, a unified coordinate system, is robust to clouds, bands, and missing values, supports linear aggregation in both spatial and temporal dimensions, and can replace traditional spectral indices or hand-crafted features. It is suitable for scenarios such as land cover and crop classification, ecological and risk monitoring, change detection, regression modeling, and large-scale mapping.