This field has drawn a great interest with the development of multiple types of sensors, leading to a huge heterogeneous amount of data. To overcome the curse of dimensionality while providing a powerful tool for environment monitoring, planning, and decision-making, several techniques have been employed. Convolutional neural network or the deep learning7–9 represents a tendency in this field thanks to its ability to automatically discover relevant contextual features in image categorization problems.10 In the same context, sparse representation (SR),11–13 total variation,14,15 and machine learning16,17 techniques, to name a few, are also investigated for different purposes, such as enhancing spatial resolution, generating explanations, and extracting knowledge from the images. MGD, the scope of our paper, have been extensively used in several domains, namely pattern recognition18,19 and computer vision.20,21 Their use was also extended to the RS field, where we need to locate edges of roads, building, rivers, and forest to detect a potential change. Adding to that, MGD provide an analytical treatment of a scene by decomposing it in high frequencies and low frequencies. This essentially helps in fusing information from different sensors and in revealing hidden characteristics indiscriminately using only one sensor image. Therefore, we aim, in this work, at drawing attention to the MGD importance and contributions in the RS field. To the authors’ knowledge, several reviews were elaborated to describe MGD, but few of them focused on their particular use in RS.