LES Model Sensitivities to Domains, Grids, and Large-Eddy Timescales

Abstract

This study has addressed the sensitivity of large-eddy simulation (LES) models to domain size, grid resolution, and the number of large-eddy turnover times (t?) of integration, and their respective effects on the characteristic turbulence statistics computed from the simulation of convective flows in planetary boundary layers (PBLs). This includes the review of relevant results from past LES studies, as well as new findings using LES models for Type I cloud-topped boundary layers with ice and snow. Horizontal domains in this study were varied from 5 km ? 5 km to 10 km ? 10 km, grid length varied from 125 m ? 125 m to 78 m ? 78 m, and three t? times of 11, 15, and 19. Adequate results for horizontal variance were obtained for the 10 km ? 10 km domain, the 78 m ? 78 m grid, and all the t? values, based on the evaluation criteria introduced to replace the time-averaging methodology. Similar results were found for vertical velocity skewness; however, 19t? offered improvement over 11t? and 15t?, showing sensitivity of model simulations to length of time integration as well as domain and grid sizes (especially for higher-order moments). Results are also presented to explain the discrepancy between LES model turbulence statistics and those calculated from field observations of convective PBLs. Recognizing that field data, in essence, have no domain size and grid-resolution restrictions, Project Lake Effect Snow Studies data over Lake Michigan were subjected to a high-pass filter (ranging from 0.5 to 5 km), which corresponds to removing the effects of larger-scale components not adequately represented by the limited domain of LES models. Turbulence statistics (such as vertical velocity skewness) derived from the filtered field data are seen to be rendered more compatible with LES model statistical results.