Simulating Atmospheric Processes in Earth System Models and Quantifying Uncertainties with Deep Learning Multi-Member and Stochastic Parameterizations (2402.03079v2)

Abstract: Deep learning is a powerful tool to represent subgrid processes in climate models, but many application cases have so far used idealized settings and deterministic approaches. Here, we develop stochastic parameterizations with calibrated uncertainty quantification to learn subgrid convective and turbulent processes and surface radiative fluxes of a superparameterization embedded in an Earth System Model (ESM). We explore three methods to construct stochastic parameterizations: 1) a single Deep Neural Network (DNN) with Monte Carlo Dropout; 2) a multi-member parameterization; and 3) a Variational Encoder Decoder with latent space perturbation. We show that the multi-member parameterization improves the representation of convective processes, especially in the planetary boundary layer, compared to individual DNNs. The respective uncertainty quantification illustrates that methods 2) and 3) are advantageous compared to a dropout-based DNN parameterization regarding the spread of convective processes. Hybrid simulations with our best-performing multi-member parameterizations remained challenging and crash within the first days. Therefore, we develop a pragmatic partial coupling strategy to sidestep issues in condensate emulation. We conduct Earth-like stable runs over 5 months with the multi-member parameterizations, while single DNN-based simulations fail within days. Even though our hybrid simulations exhibit biases in thermodynamic fields and precipitation patterns, they reduce biases seen in the traditional convective parameterization of extreme precipitation and its diurnal cycle over tropical continents compared to a superparameterization and observations. Our results show the potential of a new generation of parameterizations using machine learning with realistic uncertainty quantification that improve the representation of subgrid effects and their stochasticity.