OlmoEarth v1.3: State-of-the-art embeddings for the whole planet

We are thrilled to announce the release of OlmoEarth v1.3, a cutting-edge satellite embedding model that brings state-of-the-art capabilities to every corner of our planet. Along with this powerful new model, we're introducing the OlmoEarth Embeddings Explorer, an interactive tool for exploring and customizing maps based on these embeddings. The best part? You can start using OlmoEarth v1.3 right away without any sign-in or advanced technical knowledge required.

The key advantage of embedding models is that they compress satellite data into ready-to-use features that can be reused for multiple tasks. With OlmoEarth v1.3, users can easily explore and engage with planetary artifacts in whichever way best suits their needs. This model works by reading a year's worth of satellite data and producing a single 128-dimensional vector for every 10-meter pixel on Earth. The first 64 dimensions can also function as a smaller embedding on their own.

To evaluate the performance of OlmoEarth v1.3, we compared its embeddings against AlphaEarth Foundations (AEF) and Tessera v2 across 27 evaluations, covering six continents and various tasks. Our results show that OlmoEarth's 128-dimensional embeddings outperformed the others on 20 out of 27 evaluations, while the 64-dimensional version still led on 18 of those evaluations. We've also introduced a new crop-type benchmark called PLANTEUR, which shows that the embeddings generalize well to previously unseen locations.

Key features of OlmoEarth v1.3 include:

- Using a vision transformer, Attention Pooler, and linear layer to create embeddings

- Comparing embeddings against AlphaEarth Foundations (AEF) and Tessera v2

- Including PLANTEUR, a new crop-type benchmark for tropical farming

- Offering the model weights, pretraining data, and code for free

- Being more cost-effective and environmentally friendly than previous versions

With OlmoEarth v1.3, you can now easily classify smallholder crops in Kenya, detect brick kilns west of Dhaka, assess fire damage in Indonesia, and track gold mining in the Peruvian Amazon. The embeddings also help answer questions about land use changes, such as the impact of illegal gold mining in Peru's Madre de Dios region. The interactive tool, the OlmoEarth Embeddings Explorer, allows you to explore these changes without any setup or account requirements, making it easier than ever to understand and work with planetary data.