OlmoEarth v1.1: A more efficient family of Earth observation models

Summary

OlmoEarth v1.1 is a new family of Earth observation models designed to reduce the compute required for large-scale satellite imagery predictions. Compute dominates the lifecycle from data export through preprocessing, inference, and post-processing, motivating a focus on transformer token sequence length. For Sentinel-2, the models collapse three resolution-specific tokens into one, reducing token counts threefold; this initially caused a 10 ppt drop on the m-eurosat kNN benchmark, so the pre-training regimen was modified. OlmoEarth v1.1 runs up to three times cheaper than v1 while maintaining similar performance across a mix of research benchmarks and partner-built tasks, although some regressions remain, and the same dataset is used to isolate methodological effects.

Takeaways

1. 1

   Sentinel-2 inputs previously produced one token per timestep at each of three resolutions; collapsing those resolutions produces three times fewer tokens.

2. 2

   Naively merging resolution tokens caused a 10 ppt drop on the m-eurosat kNN benchmark, requiring changes to the pre-training regimen.

3. 3

   OlmoEarth v1.1 is available in Base, Tiny, and Nano sizes and can run up to three times cheaper than OlmoEarth v1, with some reported regressions.