Simulation of urban-scale dispersion using a Lagrangian stochastic dispersion model

Based on recent knowledge concerning the vertical structure of turbulence statistics within the roughness sublayer (i.e., the layer directly influenced by individual roughness elements) over urban surfaces the problem of urban-scale dispersion is studied. On this scale it is impossible to resolve each roughness element so that the rough character of the surface has to be taken into account by introducing a roughness sublayer, which is generally not present in dispersion models even when employed in urban environments. Two types of simulations are presented here: one, called 'urban', takes into account the roughness sublayer's turbulence structure, while the other, 'non-urban', does not. A brief overview is given on what changes are required for an 'urban' simulation as compared to a 'non-urban' or standard dispersion simulation. In particular, a parameterisation is proposed for the vertical profile of Reynolds stress within the roughness sublayer. Using a Lagrangian stochastic particle dispersion model 'urban' and 'non-urban' simulations are compared for a variety of boundary-layer states and different source configurations. It is found that neglecting the roughness sublayer results in the largest errors for low source heights and under conditions of mechanically dominated turbulence. This is of particular importance due to the fact that urban surfaces tend to increase the mechanical portion of turbulence and, in addition, low sources, such as traffic and domestic heating, are predominant in urban environments. On the basis of three tracer data sets from urban release experiments it is shown that, in general, the 'urban' simulation improves the model performance yielding smaller fractional bias at the same time as the normalised mean square error is reduced and the correlation to the observations is increased. This indicates that indeed the physical description of the dispersion process is better taken into account in the 'urban' simulation. For stable stratification the above statement does not hold true either due to other processes masking the roughness-sublayer influence in this regime or, alternatively, due to a failure of the similarity relations for the turbulence statistics under extremely stable stratification.