Ai2 releases OlmoEarth v1.3 state-of-the-art AI-ready satellite embeddings for the whole planet

The Allen Institute for Artificial Intelligence (Ai2) recently launched an advanced satellite image analysis tool called OlmoEarth v1.3, along with precomputed embeddings for the entire planet at various resolutions. This innovative model compresses a year's worth of satellite data into numerical fingerprints for each 10-by-10-meter ground patch, allowing for the identification of different land cover types like crops, forests, and mines.

Previously, users needed to fine-tune OlmoEarth using sophisticated computing resources to create maps for specific tasks, typically requiring access to powerful graphics processing units (GPUs) and large amounts of imagery. However, with the release of OlmoEarth Embeddings Explorer, a browser-based mapping tool, anyone can now utilize these embeddings for a range of applications without needing specialized hardware or extensive technical expertise.

OlmoEarth Embeddings Explorer allows users to create maps from the embeddings, enabling tasks such as tracking environmental changes over time or identifying areas where land use has been altered, like the deforestation in Peru's Madre de Dios region due to illegal gold mining activities. The model also facilitates agricultural analysis, helping analysts pinpoint regions where oil palm plantations have encroached upon forested areas.

By providing these powerful AI-ready satellite embeddings, Ai2 aims to democratize access to advanced satellite data analysis, allowing researchers, environmentalists, and other stakeholders to gain valuable insights into Earth's changing landscapes without the prohibitive costs and technical barriers associated with previous methods.