OlmoEarth v1.3 creates compact remote sensing data vectors efficiently
OlmoEarth v1.3: Embeddings
Computer Vision and Pattern Recognition
Summary
Satellites produce a lot of data about Earth over time, but it is hard to use because it's large and complex. The authors created OlmoEarth v1.3, a system that squeezes this data into small, easy-to-use summaries called embeddings. These embeddings keep important information from images and time changes while being faster and cheaper to make. The new model matches or beats older versions while using much less computer power. They also share global yearly summaries made from satellite images at a fine scale to help others use this data.
What this means in practice
- •For environmental monitoring teams: Use small yearly satellite data summaries to detect environmental changes at fine spatial scales.
- •For urban planning teams: Integrate efficient Earth embeddings to analyze land use changes and urban growth over time with reduced computing costs.
Authors
Gabriel Tseng, Favyen Bastani, Patrick Alan Johnson, Yawen Zhang, Piper Wolters, Ando Shah, Hadrien Sablon, Joseph Redmon, Heather DeHaven, Rachel Ratner, Sasha Levy, Jordan Steward, Katrina Midgley, Andrew Howe, Hunter Pitelka, Ted Schmitt
Abstract
We make some OlmoEarth embeddings. To do this, we train a new model: OlmoEarth v1.3. OlmoEarth v1.3 is designed to condense multimodal, multitemporal remote sensing data into a single digestible 64 or 128-dimensional vector. OlmoEarth v1.3 produces state-of-the-art embeddings. In addition, OlmoEarth v1.3 is a strong encoder backbone -- competitive with OlmoEarth v1 and v1.2 -- while requiring 18% and 53% of the GPU hours to train, respectively. We use OlmoEarth v1.3 to compute global, annualized 10m embeddings, which we release under the OlmoEarth artifact license.