In recent decades machine learning (ML) has become an intriguing tool for atmospheric scientists. It provides the unique ability to bridge data science with the physical sciences in order to improve our understanding of the Earth system (Boukabara et al., 2021;Reichstein et al., 2019). While ML is still a relatively novel approach to applications in climate science, there is already an abundance of research utilizing these techniques. Some examples include identifying mixed layer depths in the ocean via observations (Foster et al., 2021), attributing model biases from physics-dynamics coupling in climate models (Yorgun & Rood, 2016), improving severe hail predictions over the US high plains (Gagne et al., 2017), post-processing bias corrections of weather forecasts (Chapman et al., 2019), and implementing corrective schemes like "nudging" physics tendencies via coarse-graining or hindcasting (Bretherton et al., 2022;Watt-Meyer et al., 2021).General Circulation Models (GCMs) are made up of a dynamical core, responsible for the geophysical fluid flow calculations, and physical parameterization schemes. The latter estimate subgrid-scale processes that are generally not resolved by the dynamical core's computational grid. These processes include aspects of the Earth system such as radiation, convection, turbulence, and microphysical processes, among others. They are a source of significant bias and model uncertainty due to the heuristic nature of their development (Held, 2005;Hourdin