The large amount of data, available thanks to the recent sensors, have made possible the use of deep learning for Earth Observation. Yet, actual approaches tend to tackle one problem at a time, e.g. classification or image generation. We propose a new framework for Earth Observation images processing which learns an energy-based model to estimate the underlying distribution, possibly estimated using non-annotated images. On the varied image types of the EuroSAT benchmark, we show this model obtains classification results on par with state-of-the-art and moreover allows to tackle a high range of high-potential applications, from image synthesis to high performance semi-supervised learning.