On the suitability of general-purpose finite-volume-based solvers for the simulation of atmospheric-boundary-layer flow

Abstract

Abstract. In the present work, the quality and reliability of a general-purpose second-order-accurate Finite-Volume-based (FV) solver are assessed in wall-modeled large-eddy simulations of a neutrally-stratified Atmospheric-Boundary-Layer (ABL) flow with no Coriolis effects. The sensitivity of the solution to parameters such as grid resolution and aspect ratio is analyzed, and results are contrasted against those from a well-proven mixed Pseudo-Spectral and Finite-Difference (PSFD) code. Considered flow statistics include mean streamwise velocity, resolved Reynolds stress, turbulence intensities, skewness, kurtosis, spectra and spatial autocorrelations. It is found that first- and second-order velocity statistics are sensitive to the grid resolution and to the details of the near-wall numerical treatment, and a general improvement is observed with horizontal grid refinement. Higher-order statistics, spectra and autocorrelations of the streamwise velocity, on the contrary, are consistently mispredicted, regardless of the grid resolution. Skewness and kurtosis of the streamwise velocity, for instance, are overpredicted in the surface layer, whereas one-dimensional spectra feature a strong sensitivity to the grid resolution in the production range and a rapid decay of energy density at higher wavenumber. In addition, the typical signatures of Large-Scale Motions (LSMs) are absent in the premultiplied streamwise velocity spectra, the spatial autocorrelation functions rapidly decay along both the streamwise and spanwise coordinate directions, and instantaneous snapshots of the velocity field are populated by relatively short and thin streaks, confirming that the flow lacks LSMs. Further, the dominant mechanism supporting the tangential Reynolds stress in ABL flow – spanwise-paired sweeps and ejections– is much weaker than what commonly observed in ABL flows, ejections are severely underpredicted, and sweeps account for most of the tangential Reynolds stress in the surface layer, which is at odds with available measurements and with corresponing results from the PSFD-based solver. The inability of the solver to correctly capture the spatially-localized and relatively strong ejection events, in the authors’ opinion, is the root-canse of many of the observed mismatches and sensitivity of flow statistics to grid resolution. The present findings show that truncation errors have an overwhelming impact on the predictive capabilities of second-order-accurate FV-based solvers, introducing a degree of uncertainty in model results that may be difficult to quantify across applications involving boundary-layer flows. Although mean flow and second-order statistics become acceptable provided sufficient grid resolution, the use of said solvers might prove problematic for studies requiring accurate higher-order statistics, velocity spectra and turbulence topology.

Development of a large-eddy simulation subgrid model based on artificial neural networks: a case study of turbulent channel flow

“…These discretization errors, however, can strongly influence the resolved dynamics (e.g. Ghosal, 1996;Chow and Moin, 2003;Giacomini and Giometto, 2020), especially at the smallest resolved scales. Since the ANNs haves access to both the instantaneous DNS flow fields and corresponding coarse-grained field during training, the instantaneous discretization errors can in principle be taken into account.…”