Deep learning based super-resolution for ICON-O

Data driven machine learning algorithms are a promising tool to improve the effective resolution of numerical circulation models. We integrate a neural network of U-net-type in the ocean general circulation Model ICON-O of the Max Planck Institute for Meteorology. The neural network is trained with high-resolution data from ICON-O simulations. Our analysis of numerical experiments of varying complexity demonstrates the potential of the ML enabled ICON-O model to improve low-resolution simulation in a physically sensible manner and at modest computational cost.