Numerical considerations for simulations of flow and dispersion around buildings

This paper presents some aspects of computational work undertaken as part of a multiprotein European Union project on flow and dispersion around buildings. Attention is concentrated on those features of the numerical methods which need particular care if adequate predictions are to be obtained. Some results from a few of the 15 or so test cases being computed by all the partners using a commercially available CFD code are used to illustrate the dangers that attend such calculations. It is shown that typical numerical solutions obtainable in an industrial context are likely to be strongly dependent not only on the turbulence model but also, and often more importantly, on mesh design and the numerical method. For example, a solution obtained with, say, the standard k-epsilon turbulence model on a course grid can give results closer to experimental laboratory data than would be obtained with improved gridding and/or numerical schemes. Statements concerning apparent accuracy can therefore be misleading.