OlmoEarth Platform: Geospatial Inference at Planetary Scale
Allen Institute for AI
Ai2 launched the OlmoEarth Platform, a geospatial inference infrastructure for large-scale Earth observation. It uses OlmoEarth foundation models pretrained on ~10 TB of satellite data, enabling applications like deforestation monitoring, food security, and wildfire risk. The platform runs continent-scale inference in about a day at sub-penny per square kilometer cost.
The OlmoEarth Platform, built by Ai2, addresses the infrastructure gap for organizations using Earth observation models. It processes satellite imagery from multiple providers, handling data acquisition, reprojection, and large-scale inference. The platform divides jobs into three stages: data preprocessing on CPU, inference on GPU, and postprocessing on CPU. It can run continent-scale tasks in roughly a day, processing dozens of terabytes at fractions of a penny per square kilometer. The platform manages metadata via its own index, supports automatic failure recovery, and plans to add automated runs, change detection, agentic tools, faster models, more modalities, embeddings, and multi-cloud support.
- Сокращения
- CPU = Central Processing Unit — центральный процессор
- GPU = Graphics Processing Unit — графический процессор
- STAC —
- SNS = Simple Notification Service
- COG = Cloud Optimized GeoTIFF
- ERA-5 = ECMWF Reanalysis v5
Source: Hugging Face blog —
original