Earth Observation is producing massive amounts of data that are currently often processed with simple methods such as heuristically found rule-based systems. Our vision are end-to-end automated processing chains involving newest methods inspired from the domain of deep learning for earth observation data beyond the generation of common first-level products. In contrast to many perceptual problems in computer vision, we know many physical properties of the system under observation. Therefore, our newly developed precision learning approaches are ideally suited for this topic.

With this project we aim at combining FAU expertise of different fields in order to tackle above challenges. We want to initiate a strong, inter-disciplinary collaboration between different science fields at FAU in order to facilitate such processing chains from acquisition to data assimilation into models or to direct user products. Our goal is to transfer methodologies from e.g. medical imaging to earth observation and from mathematics to geophysical data analysis and error assessment.

Our vision is that biogeophysical models can be considerably improved by regular information with improved EO products and by providing tools or methodologies to user that handle large data quantities in different domains of environmental monitoring. At a first step we address well defined show cases from glaciology as well as nature conservation.