Abstract. Using a numerical weather forecasting code to provide the dynamic large-scale inlet boundary conditions for the computation of small-scale urban canopy flows requires a continuous specification of appropriate inlet turbulence. For such computations to be practical, a very efficient method of generating such turbulence is needed.Correlation functions of typical turbulent shear flows have forms not too dissimilar to decaying exponentials. A digital-filter-based generation of turbulent inflow conditions exploiting this fact is presented as a suitable technique for LES computation of spatially developing flows. The artificially generated turbulent inflows satisfy the prescribed integral length scales and Reynolds-stress-tensor. The method is much more efficient than, for example, Klein's or Kempf et al.'s (2005) methods because at every time step only one set of two-dimensional (rather than three-dimensional) random data is filtered to generate a set of two-dimensional data with the appropriate spatial correlations. These data are correlated with the data from the previous time step by using an exponential function based on two weight factors. The method is validated by simulating plane channel flows with smooth walls and flows over arrays of staggered cubes (a generic urban-type flow). Mean velocities, the Reynolds-stress-tensor and spectra are all shown to be comparable with those obtained using classical inlet-outlet periodic boundary conditions. Confidence has been gained in using this method to couple weather scale flows and street scale computations.

Abstract. Further to our previous Large-Eddy Simulation (LES) of flow over a staggered array of uniform cubes (Xie & Castro, 2006), a simulation of flow over random urban-like obstacles is presented. To gain a deeper insight into the effects of randomness in the obstacle topology, the current results, e.g. spatially averaged mean velocity, Reynolds stresses, turbulence kinetic energy (TKE) and dispersive stresses, are compared with our previous LES data and Direct Numerical Simulation (DNS) data (Coceal et al., 2006) of flow over uniform cubes. Significantly different features in the turbulence statistics are observed within and immediately above the canopy, although there are some similarities in the spatially-averaged statistics. It is also found that the relatively high pressures on the tallest buildings generate contributions to the total surface drag which are far in excess of their proportionate frontal area within the array. Details of the turbulence characteristics (like the stress anisotropy) are compared with those in regular roughness arrays and attempts to find some generality in the turbulence statistics within the canopy region are discussed.

Abstract. Large-eddy simulation (LES) has been applied to calculate the turbulent flow over staggered wall-mounted cubes and staggered random arrays of obstacles with area density 25%, at Reynolds numbers between 5 × 10 3 and 5 × 10 6 , based on the free stream velocity and the obstacle height. Re = 5 × 10 3 data were intensively validated against direct numerical simulation (DNS) results at the same Re and experimental data obtained in a boundary layer developing over an identical roughness and at a rather higher Re. The results collectively confirm that Reynolds number dependency is very weak, principally because the surface drag is predominantly form drag and the turbulence production process is at scales comparable to the roughness element sizes. LES is thus able to simulate turbulent flow over the urbanlike obstacles at high Re with grids that would be far too coarse for adequate computation of corresponding smooth-wall flows. Comparison between LES and steady Reynolds-averaged Navier-Stokes (RANS) results are included, emphasising that the latter are inadequate, especially within the canopy region.

Large-eddy simulations (LES) with our recently developed inflow approach (Xie & Castro, 2008a) have been used for flow and dispersion within a genuine city areathe DAPPLE site, located at the intersection of Marylebone Rd and Gloucester Pl in Central London. Numerical results up to second-order statistics are reported for a computational domain of 1.2km (streamwise) x 0.8km (lateral) x 0.2km (in full scale), with a resolution down to approximately one meter in space and one second in time. They are in reasonable agreement with the experimental data. Such a comprehensive urban geometry is often, as here, composed of staggered, aligned, square arrays of blocks with non-uniform height and non-uniform base, street canyons and intersections. Both the integrative and local effect of flow and dispersion to these geometrical patterns were investigated. For example, it was found that the peaks of spatially averaged u rms , v rms , w rms and < u w > occurred neither at the mean height nor at the maximum height, but at the height of large and tall buildings. It was also found that the mean and fluctuating concentrations in the near-source field is highly dependent on the source location and the local geometry pattern, whereas in the far field (e.g. >0.1km) they are not. In summary, it is demonstrated that full-scale resolution of around one meter is sufficient to yield accurate prediction of the flow and mean dispersion characteristics and to provide reasonable estimation of concentration fluctuations.

We present results from laboratory and computational experiments on the