Based on DeepMind’s AlphaEarth Foundations model, the product transforms the static annual dataset into on-demand representations, by area and by period.
Google Maps Platform announced on July 29, 2026 the private preversion of Custom Satellite Embeddings, a data product backed by Google DeepMind's AlphaEarth Foundations model. Where the existing data set offered frozen annual global coverage, organizations can now order representations calculated for an area of interest and a time window defined by them.
Technically, the model synthesizes optical imaging, elevation data, radar signals and 3D laser surveys (LiDAR) in a unified 64-band representation, allowing for a consistent view including over regions that are long-term cloud-covered. The frequency of updating falls to five days, with quarterly, monthly, weekly or non-calendar cuts.
The target use is agriculture, the public sector, supply chain monitoring, public services and telecommunications. Registrations for private preversion are open; a separate section makes free samples available to academic researchers via Google Earth Engine, with applications to be submitted before 15 October 2026. The existing annual data set remains available in Earth Engine and Google Cloud Storage.
Read the full article on the original source
Go to the source →