Recurrent neural networks and the temporal component

Earth observation data is inherently time-dependent and captures phenomena governed by a variety of dynamic processes that are mostly complex and not yet fully understood. Thus, there is an evident demand for deep learning models that are able to process such data streams and to capture its temporal dynamics. This chapter introduces the concept of recurrent neural networks together with their gated variants and elaborates on their advantages over feed-forward neural networks in handling sequential data. Experiments show how such models can handle Earth observation data and motivate further application scenarios in the domain of Earth Sciences.