Earth Embeddings: Learning “Mental Maps” for Open, Interoperable GeoAI (Marc Rußwurm)

The Open-Earth Monitor Global Workshop 2026 "Next-Generation EO & GeoAI Built on Openness and Distributed Data" 7-9 October 2026, Barcelona, Spain https://earthmonitor.org/global-workshop-2026/ Earth Observation is entering an era of abundance—global satellite archives, growing in-situ networks, and an expanding open-source ecosystem—yet turning this distributed wealth into decision-ready environmental information remains difficult because data are heterogeneous, incomplete, and hard to combine across sensors, resolutions, and regions. This keynote outlines Earth Embeddings: compact, AI-native “mental maps” that summarize what makes a location unique by learning directly from imagery and context. Using an intuitive “Satellite GeoGuessr” contrastive training setup, neural networks learn place-specific visual and contextual signatures and distill them into dense vectors that can serve as portable location tokens in downstream models, enabling reuse across tasks and regions. This talk will give an overview over different strategies to generate embeddings and outline research gaps and steps forward towards global interoperable FAIR embeddings.