Simulation of mean flow and turbulence over a 2D building array using high-resolution CFD and a distributed drag force approach

A numerical simulation has been performed of the disturbed flow through and over a two-dimensional array of rectangular buildings immersed in a neutrally stratified deep rough-walled turbulent boundary-layer flow. The model used for the simulation was the steady-state Reynolds-averaged Navier-Stokes equations with linear and non-linear eddy viscosity formulations for the Reynolds stresses. The eddy viscosity was determined using a high-Reynolds number form of the k-epsilon turbulence-closure model with the boundary conditions at the wall obtained with a standard wall-function approach. The resulting system of partial differential equations was solved using the SIMPLE algorithm in conjunction with a non-orthogonal, colocated, cell-centered, finite volume procedure. The predictive capabilities of the high-resolution computational fluid dynamics (CFD) simulations of urban flow are validated against a very detailed and comprehensive wind tunnel data set. Vertical profiles of the mean streamwise velocity and the turbulence kinetic energy are presented and compared to those measured in the wind tunnel simulation. It is found that the performance of all the turbulence models investigated is generally good-most of the qualitative features in the disturbed turbulent flow field through and over the building array are correctly reproduced. The quantitative agreement is also fairly good (especially for the mean velocity field). Overall, the non-linear k-epsilon model gave the best performance among four different turbulence closure models examined. The turbulence energy levels within the street canyons and in the exit region downstream of the last building were underestimated by all four turbulence closure models. This appears to contradict the 'stagnation point anomaly' associated with the standard k-e model which is a result of the excessive turbulence energy production due to normal straining. A possible explanation for this is the inability of the present models to account properly for the effects of secondary strains on the turbulence and/or for the effects of large-scale flapping of the strong shear layer at the canopy top. The results of the high-resolution CFD simulations have been used to diagnose values of the drag coefficient to be used in a distributed drag force representation of the obstacles in the array. Comparisons of the measured spatially-averaged time-mean mean velocity and turbulence kinetic energy in the array with predictions of the disturbed flow using the distributed drag force approach have been made. (C) 2003 Elsevier Ltd. All rights reserved.