Development of a proof-of-concept for an observation-based machine learning digital twin of the dynamics of the atmosphere for environmental applications that encapsulates spatio-temporal interactions. This will be done in two stages, (1) development of a machine learning based modeling core prototype and (2) Integration thereof into a digital twin architecture. The project is carried out in close collaboration with multiple external partners such as Climate21, Cs4OD, InterTwin and could be applied commercially for risk assessment, e.g. for floods, droughts, or fires. Insurance companies for example are facing significant challenges to adapt their risk assessment methodologies to the effects of climate change, making it a promising industry to collaborate with and license out the modeling core.