For the hexahedral meshes, snappyHexMesh, the mesh generator of OpenFOAM was used. More details on the quality of the meshes and its generation can be found in Rakai et al.’s [11]; here only the typical surface meshes are shown in Figure 3 and the resulting cell numbers can be found in Table 1.Figure 3Coarsest surface meshes on Building 33; see Figure 1.Table 1Cell numbers (million cells) http://www.selleckchem.com/products/BAY-73-4506.html of the investigated meshes.The Eulerian approach is considered for the calculation of passive scalar dispersion, using the results of the flowfield described in Rakai et al. [11]. The transport equation of the mean passive scalar concentration c can be seen in (1), with uj mean velocity vector, D molecular diffusion, Q source, and uj��c�䡥 turbulent scalar flux:?tc+?j(uj?c)=?j(D??jc)??j(uj��c�䡥)+Q.
(1)The closure of the turbulent scalar flux in general isuj��c�䡥=?Djk?kc.(2)In most of the cases in CWE, Djk is defined as a scalar field computed from the turbulent viscosity ��t divided by the turbulent Schmidt number Sct:Djk=��tSct.(3)But it can also be defined as a tensor using an anisotropic approach, see Yee et al. [8], as in (4) with turbulent kinetic energy k, its dissipation , and the mean velocity gradient tensor uj,k:Djk=Cs1k2?��jk+Cs2k3?2(uj,k+uk,j).(4)The constants are taken asCs1=0.134Cs2=?0.032.(5)For the comparison of experimental and simulation results a dimensionless concentration, c, is defined in (6), with the reference velocity Uref, a reference length L, and the source strength Qsource:c?=c?Uref?L2Qsource.(6)For the comparison three different statistical metrics are used.
Using matrix norms for comparison is the simplest approach; here the L2 norm is used; see (7). This metric can be seen as a normalized relative error of the whole investigated dataset:L2=��i=1n(Ei?Si)2��i=1nEi2.(7)Ei and Si are the corresponding experimental data and simulation results in the ith experimental point, with a total of n experimental points.The factor of two (FAC2, see (8)) metrics often used in air quality model evaluations, see Chang and Hanna [17], is also used to avoid judgment by only one, probably biased metric:FAC2=Nn=1n��i=1nNi??withNi={1,for??0.5��SiEi��2.00,for??else.(8)As in air quality modelling the results may differ in several orders of magnitude, an additional metric from Chang and Hanna [17] is used which is more sensitive to the changes in the order of the results, so the small differences are not hidden by the large order values.
The metric chosen is MG, geometric mean bias:MG=exp?(ln?Ei��?ln?Si��).(9)The boundary conditions, Dacomitinib model and numerical setting for the flow field calculations were described in Rakai et al. [11]. For the dispersion calculations zero gradient conditions were defined at all boundaries except for the inflow, where 0 value was given.