Deep Learning for Subgrid‐Scale Turbulence Modeling in Large‐Eddy Simulations of the Convective Atmospheric Boundary Layer

In large‐eddy simulations, subgrid‐scale (SGS) processes are parameterized as a function of filtered grid‐scale variables. First‐order, algebraic SGS models are based on the eddy‐viscosity assumption, which does not always hold for turbulence. Here we apply supervised deep neural networks (DNNs) to learn SGS stresses from a set of neighboring coarse‐grained velocity from direct numerical simulations of the convective boundary layer at friction Reynolds numbers Reτ up to 1243 without invoking the eddy‐viscosity assumption. The DNN model was found to produce higher correlation between SGS stresses compared to the Smagorinsky model and the Smagorinsky‐Bardina mixed model in the surface and mixed layers and can be applied to different grid resolutions and various stability conditions ranging from near neutral to very unstable. The DNN model can capture key statistics of turbulence in a posteriori (online) tests when applied to large‐eddy simulations of the atmospheric boundary layer. Plain Language Summary Using grid‐scale state variables to represent subgrid‐scale (SGS) processes is a major step in large‐eddy simulations of turbulence, which typically rely on some physically based assumptions that do not always apply. Here we replace physically based assumptions with deep learning in SGS modeling and show the latter performs better in a priori (offline) tests of atmospheric turbulence. In addition, the deep neural networks model well captures key statistics of turbulence in a posteriori (online) tests when applied to in large‐eddy simulations of atmospheric boundary layers. Key Points Deep learning‐based subgrid‐scale models outperform traditional eddy‐viscosity models in a priori (offline) tests Deep learning‐based subgrid‐scale models can account for density stratification effects in boundary‐layer turbulence The DNN model can capture key turbulence statistics in a posteriori (online) tests when applied to large‐eddy simulations of the atmospheric boundary layer