Self-supervised pretraining using masked autoencoding on large-scale Synthetic Aperture Radar (SAR) data significantly reduces the labeling requirements for downstream tasks crucial to climate change monitoring, such as vegetation cover prediction and land cover classification.
