Using Machine Learning for Model Physics: an Overview

The scientific and practical significance of interdisciplinary complex numerical models and prediction systems has increased tremendously during the last few decades, due to improvements in their quality via better understanding of the basic processes and their relationships and developments in numerical modeling. The past several decades have revealed a well pronounced trend in weather and climate numerical modeling. Th is trend marks a transition from investigating simpler linear or weakly nonlinear single - disciplinary systems like simplified atmospheric or oceanic systems that include a limited description of the physical processes, to studying complex nonlinear multidi sciplinary systems like coupled atmospheric - oceanic systems that take into account atmospheric physics, chemistry, land - surface interactions, etc. The most important property of a complex interdisciplinary system is that it consists of subsystems that, b y themselves, are complex systems. Thus, due to their increasing complexity, global and regional modeling activities consume a tremendous amount of computing resources, which presents a significant challenge despite growing computing capabilities.