Large-Eddy Simulations of Stratified Atmospheric Boundary Layers: Comparison of Different Subgrid Models

The development and assessment of subgrid-scale (SGS) models for large-eddy simulations of the atmospheric boundary layer is an active research area. In this study, we compare the performance of the classical Smagorinsky model, the Lagrangian-averaged scale-dependent (LASD) model, and the anisotropic minimum dissipation (AMD) model. The LASD model has been widely used in the literature for 15 years, while the AMD model was recently developed. Both the AMD and the LASD models allow three-dimensional variation of SGS coefficients and are therefore suitable to model heterogeneous flows over complex terrain or around a wind farm. We perform a one-to-one comparison of these SGS models for neutral, stable, and unstable atmospheric boundary layers. We find that the LASD and the AMD models capture the logarithmic velocity profile and the turbulence energy spectra better than the Smagorinsky model. In stable and unstable boundary-layer simulations, the AMD and LASD model results agree equally well with results from a high-resolution reference simulation. The performance analysis of the models reveals that the computational overhead of the AMD model and the LASD model compared to the Smagorinsky model is approximately 10% and 30% respectively. The LASD model has a higher computational and memory overhead because of the global filtering operations and Lagrangian tracking procedure, which can result in bottlenecks when the model is used in extensive simulations. These bottlenecks are absent in the AMD model, which makes it an attractive SGS model for large-scale simulations of turbulent boundary layers.

Automated summary

The study compares the performance of different SGS models for large-eddy simulations of the atmospheric boundary layer. It finds that the AMD and LASD models capture velocity profiles and turbulence energy spectra better than the Smagorinsky model. The computational overhead of the AMD and LASD models is slightly higher than the Smagorinsky model, with the AMD model being more efficient for large-scale simulations.