AI2 Launches OlmoEarth Platform for Geospatial AI Inference at Planetary Scale

Introducing the OlmoEarth Platform

AI2 has announced the launch of the OlmoEarth Platform, an infrastructure built to support large-scale geospatial model inference. This platform is designed to take geospatial models from fine-tuning and evaluation stages through to large-scale deployment, specifically for Earth observation applications.

Addressing Deployment Challenges

The platform aims to assist organizations, particularly those in environmental sectors, that may lack the extensive infrastructure or engineering teams required to manage the full lifecycle of AI models. This includes data labeling, model fine-tuning, and running large-scale inference, which are common hurdles for deploying Earth observation AI effectively.

Planetary-Scale Inference Capabilities

The OlmoEarth Platform is capable of running inference across continent-scale areas within approximately one day. It processes dozens of terabytes of imagery at a cost of fractions of a penny per square kilometer. This capability is crucial for applications such as deforestation monitoring, food security, and wildfire risk assessment, which rely on extensive satellite data analysis.

Technical Underpinnings

Developing the platform involved overcoming significant engineering challenges inherent in large-scale geospatial systems. These include accessing and aligning satellite imagery from multiple providers, efficient processing, stitching results into consistent maps, and managing distributed computing failures. The platform leverages AI2's experience operating platforms like Skylight and EarthRanger.