EMBED-CD
Change maps from satellite embeddings. Draw an area, pick two years, then label a few objects and a live classifier labels the rest on the fly.
Generates year-over-year land change maps from AlphaEarth satellite embeddings (Google Satellite Embedding V1), which cover the globe for every year 2017-2025 at 10 m. Works tile by tile and fills the canvas as each tile completes, so memory stays flat and large areas are possible. Missing data is shown as its own class, never as "no change". Label a few change objects and a live one-vs-rest classifier labels the rest on the fly. CPU-only, nothing to train, no account or API key - data is read directly from public cloud-optimized GeoTIFFs, and the one-time tile index is a small pre-built file needing only numpy. Tested on Windows (QGIS 3.44 and 4.0). macOS is untested; Linux support is planned for a later release. The AlphaEarth Foundations Satellite Embedding dataset is produced by Google and Google DeepMind (CC BY 4.0).