Titel:
"ClimSat - Climate-conditional Satellite Image Editing with Diffusion Autoencoders"
Description:
Author: Johannes Leonhardt
This short video introduces ClimSat, a deep learning model for simulating satellite imagery under varying climatic conditions. ClimSat is a multi-conditional diffusion autoencoder trained on Sentinel-2 imagery, land cover maps, and climate data. By conditioning the generation process on climate variables, the model can produce realistic satellite images that reflect different climatic scenarios while preserving the underlying landscape structure.
As an example application, ClimSat is used to augment training datasets for land cover classification, which improves model robustness by reducing domain shift between regions and climates.
In the series “My paper in 140s”, scientists from the Collaborative Research Center 1502 'DETECT' present syntheses of their project work that have been published in peer-reviewed papers.
Read the full paper here: https://doi.org/10.1016/j.srs.2025.100235.
Check also the webpage of the author, Johannes Leonhardt.
