“The accelerating application of machine learning foundation models to remotely sensed data transforms traditional spatial analysis by encoding the Earth’s features as universal, high-dimensional embedding representations. I argue that the construction of an embedded Earth is not merely a technological development, but an epistemological shift in planetary computation. I historicize Earth embeddings within lineages of cartography, computation, and aesthetics, asking not what these data tell us about the world, but what they tell us about ourselves—our desires and sensibilities, our ways of knowing and making sense.
This critical analysis reveals Earth embeddings as neither a neutral computational tool nor a reflection of the Earth, but a human-constructed space with no ground truth. I apply computational data analysis and visualization alongside conceptual frameworks from geography, science and technology studies, and media theory in order to question not how well Earth embeddings represent the planet, but what kind of world they construct.”
