Item

Abstract

Data-driven approaches using machine learning to parameterizing model physical processes in Earth System Models have been actively explored in recent years. Deep-learning-based convection parameterization is one such example. While tremendous progress has been made in emulating convection using neural networks, serious roadblocks remain, including generalization of the emulators trained on model data from current climate to a warmer climate and model instability when the emulators are implemented online in model integrations. This study uses an ensemble of deep convolutional residual neural networks to emulate convection simulated by a superparameterized global climate model (GCM). The neural networks (NN) use the current environmental state variables and advection tendencies, as well as the history of convection to predict the GCM grid-scale temperature and moisture tendencies, cloud liquid and ice water contents from moist physics processes. The independent offline test shows that the NN-based emulator has extremely high prediction accuracy for all output variables considered. In addition, the emulator trained on data in the current climate generalizes well to a warmer climate with +4K sea surface temperature in an offline test, with high prediction accuracy as well. Further tests on different aspects of the NN architecture are performed to understand what factors are responsible for the good generalizability of the emulator to a warmer climate. The details will be presented at the meeting.