CS4Env

Title: OlmoEarth: From Foundation Model Research to an End-to-End EO Platform


Speakers: Yawen Zhang and Patrick Johnson, Allen Institute for AI

Earth observation data presents a unique challenge: it is spatial like images, sequential like video or text, and highly multimodal. We present OlmoEarth, a multimodal spatio-temporal foundation model that introduces a new self-supervised learning formulation, masking strategy, and loss function tailored specifically to the Earth observation domain. OlmoEarth achieves state-of-the-art performance compared to other foundation models across a wide range of research benchmarks and real-world partner tasks, ranking first on 15/24 tasks using embeddings and 19/29 tasks with full fine-tuning.

Building on this model, we deploy the OlmoEarth Platform, an end-to-end solution that integrates frontier foundation models with data management, labeling, fine-tuning, and inference tools. The platform removes the need for organizations to manage GPU infrastructure or deep learning pipelines, enabling non-profits, NGOs, and environmental practitioners to translate research advances into operational solutions for conservation, climate resilience, and food security.

In this talk, we will introduce the OlmoEarth model, walk through the platform’s capabilities, followed by demos showcasing real-world applications: wildfire risk assessment and crop type mapping.
Zoom link