Project by Ellie Madsen
My research explores the novel application of foundation models to Earth observation using methods of experimental data analysis, theoretical and technical writing, critical cartography, elemental media studies, topology, machine learning, and speculation. I conducted a range of representational experiments to explore various properties of Earth embedding data, including change detection, dimensionality reduction, equidistant projection, and false-color composites.
I argue that the high-dimensional latent space created by embedding Earth data is a not-for-human world with its own topology and texture. My research exists in the irony between calling for a human understanding of Earth embeddings and acknowledging that attempting to see like machines is a losing game.
My ongoing work attempts to illustrate the properties, limits, edge cases, opportunities, and form associated with Earth embedding data. I explore the high-dimensional turn of planetary mediation across themes of sparsity, dimensionality, scale, distance, and temporality.
