Stochastic parameterization of column physics using generative adversarial networks

We demonstrate the use of a probabilistic machine-learning technique to develop stochastic parameterizations of atmospheric column physics. After suitable preprocessing of NASA's Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA2) data to minimize the effects of high-frequency, high-wavenumber component of MERRA2 estimate of vertical velocity, we use generative adversarial networks to learn the probability distribution of vertical profiles of diabatic sources conditioned on vertical profiles of temperature and humidity. This may be viewed as an improvement over previous similar but deterministic approaches that seek to alleviate both, shortcomings of human-designed physics parameterizations, and the computational demand of the "physics" step in climate models. Impact Statement Global climate models can now be used to produce realistic simulations of climate. However, large uncertainties remain: for example, the estimated change in globally-averaged surface-temperature to a doubling of atmospheric CO2 varies between 2 degrees and 6 degrees C across leading models. Uncertainty in representing cumulus convection (think thunderstorm) is a major contributor to this spread: Scales at which they occur, 100m-10 km, are too small to be resolved in global climate models, requiring their effects on larger scales to be approximated with simple models. Improving such approximations using new probabilistic machine-learning techniques, initial steps toward which are successfully demonstrated in this work will likely lead to improvements in the modeling of climate through improvements in the representation of cumulus convection.