CFD simulation of pollutant dispersion around buildings: comparison between RANS k-ε and LES approaches

Turbulence modeling is a key issue in computational wind engineering, particularly in the prediction of pollutant dispersion in cities. To be directly applicable, turbulence models need validation by comparison with experiments. This paper evaluates the performance of two different modeling approaches (RANS k-e and LES) for three test cases with varying complexity. For each case, wind tunnel experiments are used for validation. It is shown that the performance of the standard k-e model is very case-dependent and that it also depends on the turbulent Schmidt number, whose optimum value is a priori unknown. On the contrary, LES with the dynamic subgrid-scale model shows a good performance for all cases, without requiring any parameter input to solve the dispersion equation. For the test case of an actual urban environment, predicted concentration values with LES differ from experiments by less than a factor of 2, compared to less than a factor of 4 with the standard k-e model.