OlmoEarth Studio Introduces Custom Embedding Exports for Earth Observation Analysis

OlmoEarth Studio, a platform for building Earth observation models, now enables users to compute and export embedding vectors—compact numerical representations of Earth-observation data produced by its open source OlmoEarth foundation models. The source code and model weights are publicly available, allowing the community to inspect how embeddings are generated. Embeddings provide a fast, cost-effective entry point for tasks such as similarity search, segmentation, and unsupervised exploration. Users can configure parameters like area of interest, time span (1-12 monthly periods), encoder variant (Nano, Tiny, Base), spatial resolution (10 to 80 meters per pixel), and imagery sources (Sentinel-2 L2A, Sentinel-1 RTC, or both). The output is a Cloud-Optimized GeoTIFF (COG) with one band per embedding dimension, stored as signed 8-bit integers. The blog demonstrates use cases including similarity search, few-shot segmentation, change detection, and unsupervised exploration using PCA. For higher performance, Studio also supports supervised fine-tuning (SFT). The embeddings are computed on demand, reflecting specific conditions, and can capture seasonal dynamics.