Numerical modeling of micro-scale wind-induced pollutant dispersion in the built environment

Despite recent efforts directed towards the development of cleaner and more efficient energy sources, air pollution remains a major problem in many large cities worldwide, with negative consequences for human health and comfort. If the transport of pollutants by wind in urban areas can be predicted in an accurate way, remedial measures can be implemented and the exposure of people and goods to pollution can be decreased to limit these negative effects. This prediction can be achieved by experimental techniques, on-site or in wind tunnels, but also numerically, with the use of Computational Fluid Dynamics (CFD). In this thesis CFD is used to simulate wind-induced pollutant dispersion in the built environment. The accuracy of this approach in terms of pollutant concentration prediction always needs to be assessed. The reason is twofold. First, the wind flow around buildings is turbulent and cannot be solved exactly with CFD. This type of flow must therefore be approximated with so-called turbulence models. Second, various types of errors are present in the numerical solution and can affect its accuracy. The Reynolds-Averaged Navier-Stokes (RANS) and Large-Eddy Simulation (LES) turbulence modeling approaches are the most widely used in computational wind engineering. They are compared in this thesis, and evaluated by comparison with reference wind-tunnel experiments. In the first part, several generic cases of simplified isolated buildings are considered and, in the second part, an applied case of pollutant dispersion in an actual urban area (part of downtown Montreal) is studied. In the computations, care is taken to accurately simulate three key aspects of urban pollutant dispersion: (1) the atmospheric boundary layer flow, (2) the wind flow around buildings, and (3) the dispersion process. On average, the transport of pollutants by wind can be seen as the combination of, on the one hand, the transport by the mean flow and, on the other hand, the transport by the turbulent fluctuations. This decomposition is used here to evaluate the RANS – with various turbulence models – and LES approaches. Overall, the better performance of LES in terms of flow and concentration field prediction is demonstrated. In addition, LES has the advantage to provide the time-resolved velocity and concentration fields. Given the good accuracy of LES, this approach is used to investigate the physical mechanism of pollutant dispersion for the case of a simplified isolated building. The vortical structures present in the shear layers developing from the roof and sides of the building are shown to play a crucial role in the turbulent mass transport process. LES used as a research tool also allows evaluating models employed with RANS for turbulent mass transport, which is often assumed to act as a diffusion mechanism. The results of this study show that this hypothesis is not always valid and in some cases the turbulent mass flux in the streamwise direction is directed from the low to high levels of mean concentration (counter-gradient diffusion)