The Influence of Computational Parameterization on Mean Flow and Turbulence Statistic in 2D Idealized Street Canyon: Computational Domain

Abstract

The use of Computational Fluid Dynamics (CFD) as a research tool has been utilized in many ways to predict the turbulent flow nature and pollution dispersion in an urban environment such as street canyon. Apart from the advance advantages offered by this method, some problems can be arising on respect to accurate prediction of data and uncertainties of the assumption that are made in numerical modelling. The execution and evaluation of the CFD studies need to undergo validation and generic sensitivity analyses to minimize the computational error and have an accurate prediction of data. This study performs a series of large eddy simulation (LES) to investigate the flow fields within and above a two-dimensional idealized street canyon with a unity aspect ratio (ratio of street width, W to building height, H). The effect of domain size on turbulent statistics is evaluated through the implementation of three different domain sizes i.e. varied streamwise lengths with fixed spanwise length and vertical height for different grid resolution (coarse, medium and fine). Comparison with the available experimental results shown a good agreement for mean velocity profiles. However, current LES underestimates the higher-order turbulent statistics. For the mean flow, the LES predicts the streamwise flow in the upper part and reversed flow in the lower part of the canyon, indicating the well-known flow regime of skimming flow. The value of mean velocities for all cases of the run have almost matching profile inside and above the canyon. For all cases, the maximum values of standard deviation in the streamwise and vertical directions are located approximately near the roof-level windward wall represent the occurrence of velocities fluctuation. In contrast, the Reynold shear stress for all cases shows some discrepancy where the peaked of turbulence intensity near the roof height is not smoothly captured in the current LES. Study shows that the LES cannot properly simulate those statistics accurately when the streamwise domain is limited to 10H even with the increasing of grid resolution sizes. Nevertheless, these results are expected to provide additional information relevant to uncertainty estimation upon execution of computational fluid dynamics (CFD) simulation.