Digital twins in climate science: challenges and opportunities

In recent years, advancement in computational infrastructures has made possible to start exploiting the implementation and use of digital twins in climate science. A growing number of studies and prototypes have already appeared, aiming at modelling single or multiple components of the Earth system. Among them, it is worth mentioning the European Commission's Destination Earth initiative with the ambition of realizing a digital replica of the Earth. While the development of digital twins seems straightforward and is proceeding at fast pace, there are still some key conceptual issues and challenges to overcome and go beyond classic numerical models and digital shadows. Realising a continuous bidirectional data flow between the virtual system and the real one is among them. Together with innovative approaches in data assimilation and the integration of physics-consistent machine learning, there is the need to conceptualize what a continuous data loop means at time scales covering the coming years and decades. Furthermore, the need to address the "human-in-the-loop" requirement remains central to allow for actionable "what-if" scenario testing. In this contribution, we discuss these open issues as well as the minimum requirements twins should have. We conclude by proposing pathways to fulfil the ambition of having a digital twin of the Earth system.