Tech-Talk: HPC - Climate / Weather / Ocean Modeling

The next generation of climate models will stream the simulated state of the global atmosphere, ocean and land surface at km scales. Streaming small ensembles with a temporal compression of 100-1000, will make it possible to 'experience’ virtual Earths simulating the past, present, and future. The simulations envisioned pose grand, but surmountable, challenges for high-performance computing and data analytics. Learning from the simulations using AI will be necessary for their acceleration, for compression and analysis of their output, and for interactivity. Interactivity is necessary to develop confidence in the models and to effectively extract their information content. These qualities -- which will help us to understand climate changes on the one hand, and to concretely plan for what they forebode on the other -- are what transforms the simulation systems into Digital Twins. In this talk I will sketch the architecture of these next generation systems, which we envision being supported by machines like Earth-2, possibly as a contribution to new international initiatives, such as EVE.