The field of remote sensing and geospatial analysis is rapidly advancing, driven by innovations in machine learning, computer vision, and data fusion. Recent research has focused on developing novel methods for learning representations of satellite images, estimating land surface temperature, and predicting land cover changes. Notably, the use of deep learning techniques such as convolutional autoencoders and generative adversarial networks has shown promising results in various applications, including weather event classification, land cover mapping, and object detection. Furthermore, the integration of physical models and domain knowledge into machine learning frameworks has led to more accurate and interpretable predictions. The development of new benchmarks and datasets has also facilitated the evaluation and comparison of different methods, driving progress in the field.

Some noteworthy papers include: WGAST, a weakly-supervised generative network for daily 10m land surface temperature estimation, which outperformed existing methods in both quantitative and qualitative evaluations. TerraMAE, a novel framework for learning spatial-spectral representations from hyperspectral earth observation data, which demonstrated superior performance in various downstream tasks. Deep Space Weather Model, a novel approach for predicting solar flares using multi-wavelength images, which achieved state-of-the-art performance on a new benchmark dataset.