Edinburgh & ESA build COP-GEN, a latent diffusion model that treats Earth observation as a distribution, not a single prediction

anselm · x · 2026-09-09

Researchers from Edinburgh and ESA present COP-GEN, a multimodal latent diffusion model addressing a flaw in Earth observation: most models are deterministic, mapping a DEM and land-cover map to a single "most likely" optical image. But the same terrain can plausibly look very different depending on cloud cover, season, soil moisture, and atmospheric scattering — the correct answer is a distribution, not one image.

COP-GEN is built to model that distribution of plausible outcomes rather than collapse it to its mean, and the authors argue this probabilistic shift matters far more than it sounds.

Original post →

More from Multimodal

Multimodal channel →