Queryable Earth

Queryable Earth transforms satellite imagery into a searchable database: use natural language to find and track specific objects and changes globally, leveraging AI and vector embeddings.

This is a fundamental shift in Earth observation: a searchable database of the planet's surface, operating like Google for the physical world. The core technology leverages Vision-Language Models (VLMs) like SkyCLIP and RemoteCLIP, converting both satellite images and natural language queries into vector embeddings for similarity search. Planet's fleet, which images the entire Earth nearly every day, supplies the massive data volume (30 terabytes daily) for this index. The system allows users to ask complex questions—for example, 'How many houses are in Pakistan?' or 'Show me all deforestation in the Amazon between this week and last week.' This capability moves beyond traditional geospatial analysis, delivering actionable insights and massive-scale change detection via simple text input.

https://www.planet.com/pulse/queryable-earth-combining-satellite-imagery-and-next-generation-ai/