Google Satellite Embedding V1 in Earth Engine

Google / DeepMind’s AlphaEarth Foundations model produces analysis-ready annual 10 m, 64-D embeddings from optical, radar, LiDAR, and other inputs. What is public is the Satellite Embedding V1 data product: an Earth Engine collection and GCS COGs. The training weights are not open.

A pass-through today is realistic as a data proxy over Earth Engine or GCS for customers who accept Google’s dataset terms. It is not “we run AlphaEarth weights.” Serving the published layers needs no GPU. It does need Earth Engine or GCS access, and billing awareness (the GCS bucket is noted as provider-pays as of mid-2026 in the Earth Engine docs).

The dataset is CC-BY 4.0. Required attribution: “The AlphaEarth Foundations Satellite Embedding dataset is produced by Google and Google DeepMind.” The customer supplies an AOI and year(s), plus an Earth Engine project or a GCS read path.

We do not own this dataset or the model. We do not host it today. We would only proxy published layers under those terms.