Recent advances in remote sensing technologies and open data efforts have made large volumes of data available to Earth Scientists. Owing to this, Supervised Machine Learning and Deep Learning techniques are being leveraged extensively to tackle difficult problems in Earth Science. However, these techniques require large volumes of labeled datasets, which is only sparsely available in Earth Science due to human hour limitations and volume of Earth observation data. This provides opportunities for applying deep learning algorithms that do not rely on labeled data. This session aims to foster exploration of deep generative and self-supervised techniques within scope of Earth Science. Recently, deep generative models and self-supervised learning techniques have been shown to effectively model high dimensional data to be used for image compression, super-resolution, temporal and spatial extrapolation, modality transfer and featurization, among other applications. We welcome submissions on any of the aforementioned tasks for advancing Earth Science research.