Numerical Investigations of Urban Pollutant Dispersion and Building Intake Fraction with Various 3D Building Configurations and Tree Plantings

Rapid urbanisation and rising vehicular emissions aggravate urban air pollution. Outdoor pollutants could diffuse indoors through infiltration or ventilation, leading to residents’ exposure. This study performed CFD simulations with a standard k-ε model to investigate the impacts of building configurations and tree planting on airflows, pollutant (CO) dispersion, and personal exposure in 3D urban micro-environments (aspect ratio = H/W = 30 m, building packing density λp = λf = 0.25) under neutral atmospheric conditions. The numerical models are well validated by wind tunnel data. The impacts of open space, central high-rise building and tree planting (leaf area density LAD= 1 m2/m3) with four approaching wind directions (parallel 0° and non-parallel 15°, 30°, 45°) are explored. Building intake fraction  is adopted for exposure assessment. The change rates of  demonstrate the impacts of different urban layouts on the traffic exhaust exposure on residents. The results show that open space increases the spatially-averaged velocity ratio (VR) for the whole area by 0.40–2.27%. Central high-rise building (2H) can increase wind speed by 4.73–23.36% and decrease the CO concentration by 4.39–23.00%. Central open space and high-rise building decrease  under all four wind directions, by 6.56–16.08% and 9.59–24.70%, respectively. Tree planting reduces wind speed in all cases, raising  by 14.89–50.19%. This work could provide helpful scientific references for public health and sustainable urban planning.


Introduction
Rapid urbanisation has aggravated urban environmental problems over the past several decades. The rapidly increasing vehicular emissions in street networks deteriorate urban air quality and have become one of the main pollutant sources in modern cities [1][2][3]. Urban air pollutant exposure has induced rising risks of respiratory and cardiovascular diseases, or even premature mortality [4,5]. People spend more than 90% of their lifetime indoors, on average. Moreover, outdoor air pollutants can diffuse into the indoor environment by infiltration or ventilation via windows, vents, and so on. The indoor pollutant exposure is closely influenced by the outdoor air quality, especially for buildings with natural ventilation. Therefore, near-road residents usually suffer from much higher air pollutant exposure than those in other regions [4,[6][7][8]. Special attention is required to develop sustainable urban designs to improve urban ventilation and reduce urban residents' exposure [9,10].
The urban canopy layer (UCL) represents the atmospheric layer from the ground to the building rooftops, where most urban residents live. To mitigate the pollutant exposure of residents in the UCL, improving the ventilation and pollutant dilution capacity is one of the major solutions [9,10]. Recently, field observations, computational fluid dynamic (CFD) simulations, and laboratory-scale physical modelling (wind tunnel or water channel experiments) have been widely employed to investigate the ventilation and pollutant dispersion at the street scale (~100 m) or neighbourhood scale (~1 km) [11][12][13][14][15][16][17][18]. Field observation can directly monitor the critical characteristics of air flow and dispersion in real cities, but is restricted by low spatial resolution, uncontrollable boundary conditions, and complicated building configurations [15]. Although laboratory-scale physical modelling techniques can control boundary conditions and building configurations well, and have been widely used to validate numerical models, they have to meet the similarity criteria requirements and the costs are relatively high [13,15,17,18]. Numerical modelling with a high temporal-spatial resolution turns out to be a more efficient and relatively low-cost tool to study the flow features and dispersion characteristics, but sometimes it has challenges in attaining satisfactory validation by experimental data [19][20][21][22][23][24][25]. In this work, CFD modelling is applied to simulate the flow field and pollutant dispersion at the street scale.
Key urban morphological parameters include street aspect ratio (AR = H/W, where H is the building height, W is the street width) [26,27], tree planting [28][29][30], the direction of approaching flow [31,32], building packing densities [33,34], building height variation [35,36], special building designs including open space [37][38][39] and high-rise building [40,41]. Previous studies have investigated their impacts on the urban ventilation and pollutant dispersion. Nevertheless, while most studies focused on the airflow and pollutant dispersion in the street canyon, the integrated impacts of different urban layouts on residents' exposure in three-dimensional (3D) urban models are still rare. Therefore, this study aims at evaluating the synthetic impacts of these urban parameters (urban open space, tree planting and central high-rise building in this work) on ventilation, pollutant dispersion and related human exposure. The work provides a scientific reference and effective methodologies for sustainable urban design and public health. Velocity ratio (VR) is used throughout the work to normalise and quantify the wind environment experienced by pedestrians [11], defined by Equation (1): where V p is the wind velocity at the pedestrian level (z = 2 m) and V δ is the wind velocity at the top of the boundary layer. Here, V δ = 4.34 m/s at z = 300 m [39].

Building Intake Fraction <P_IF>
The concept of the building intake fraction <P_IF>, derived from the personal intake fraction P_IF, is employed to evaluate the impacts of urban layouts on the residents' pollutant exposure. P_IF represents the total pollutant inhalation per person, which is widely adopted to quantify the indoor and street-scale (~100 m) vehicular pollutant exposure [42][43][44].
The intake fraction (IF) for a certain population is defined in Equation (2) [45][46][47][48]: (2) where N stands for the total number of population age groups considered in this research; M is the total number of micro-environment types; P i is the number of the population in the age group i; Br i,j (m 3 /s) is the volume-mean breathing rate for individuals of the age group i in the micro-environment group j; ∆t i,j is the time that group i stays in the microenvironment j; Ce j (kg/m 3 ) is the time-averaged concentration C of the certain vehicular pollutant in the micro-environment j; and m (kg) is the total emission of the vehicular pollutant over the research period. The population in the research is divided into three groups (N = 3) according to Luo et al. [46]. The composition of the target population is: children (<18, i = 1), adults (18-60, i = 2) and elders (>60, i = 3). Chau et al. [49] considered four types of microenvironment (M = 4) in their work, including indoor at home (j = 1), other indoor locations (j = 2), near vehicles (j = 3) and other outdoor locations (j = 4). To simplify the computation, only one micro-environment (j = 1, indoor at home) is considered in this study, and the buildings are assumed to be the residential type with natural ventilation. As Table A1 in the supplement shows, the percentage of children, adults and elders are 21.2%, 63.3% and 15.5%, respectively, in this paper. The breathing rate Br and time percentage spent indoors at home for different age groups is 12.5 and 61.70% (children), 13.8 and 59.50% (adults) and 13.1 and 71.60% (elders), respectively [46,49,50].
IF has been used to express the source-to-intake relationship for vehicular pollutants in realistic street canyons [48]. Since IF would change linearly with the variation of the population, it has been optimised by defining personal intake fraction P_IF in Equation (3) [26,42]. P_IF is independent of population size and density, and it represents the average IF for each person.
To estimate the influence of urban layouts on personal exposure, P_IF is employed in this work to quantitatively evaluate the pollutant inhalation of residents.
Previous researchers [6,7] found that the ratio of indoor and outdoor pollutant concentration (I/O) was approximately 1 for buildings with natural ventilation. Therefore, to reduce the number of grids and computational resources, the inner space of buildings is not considered in simulations. The pollutant concentrations on building surfaces are adopted as the indoor concentrations in this work, and the vehicle emission is assumed as the only source of the indoor environment [26,43,45]. Building intake fraction <P_IF> is the spatial mean of P_IF at all building surfaces. Throughout this work, <P_IF> is used to present the spatially-averaged personal intake exposure for the whole urban area. The change rate of <P_IF> could represent the varied exposure risks for the indoor residents due to the impacts of different urban layouts.

CFD Model Description
The governing equations of mass, momentum, turbulent kinetic energy (k) and its dissipation rate (ε) of the employed CFD model are shown in Equations (4)-(7), as follows: the mass conservation equation: the momentum equation: the transport equations of turbulent kinetic energy (k) and its dissipation rate (ε): where u j stands for time-averaged velocity components (u j = u, v, w as j = 1, 2, 3); v is the kinematic viscosity; and v t is the kinetic eddy viscosity The SIMPLE scheme is applied for coupling pressure and velocity. The underrelaxation factors for pressure term, momentum term, k and ε terms are 0.3, 0.7, 0.5 and 0.5, respectively. When all the absolute residuals are smaller than 10 −6 , the iteration is converged.

Model Set-Up and Boundary Conditions
The 3D idealised full-scale UCL model with neutral atmosphere conditions is adopted in this study. The whole UCL model has a 5 × 5 building matrix composed of 25 cubic models (H = B = W = 30 m) with moderate packing density (aspect ratio H/W = 1, building packing density λ p = λ f = 0.25). The designed UCL model is an idealised typical urban residential area in miniature, especially relating to the communities in small and mediumsized towns, or the communities in the old town of modern cities. To evaluate the impacts of approaching winds with different directions, θ (the included angle of the approaching wind and axis x) is set as 0 • , 15 • , 30 • and 45 • for every scenario. The setup of the simulation domain is shown in Figure 1a,b. Figure 1a depicts the simulation area of the cases with parallel approaching wind (θ = 0 • ), with the geometry of 1700 m (x) × 870 m (y) × 300 m (z). The distances from the UCL model to the domain inlet, outlet and lateral boundaries are 6.7H, 41H and 10H, respectively. The symmetry boundary condition is adopted at the domain top and the lateral boundaries, while the domain outlet takes the zero normal gradient boundary condition [39,69,70].
The domain geometry of the cases with non-parallel approaching wind (θ = 15 • , 30 • and 45 • ) is 1700 m (x) × 1700 m (y) × 300 m (z) (Figure 1b). In this condition, the distances from the UCL model to the domain inlets and outlets are 6.7H and 41H, respectively. The symmetry boundary condition is only adopted at the domain top. The boundary condition for the domain inlet is provided by Equations (8)- (10) [27,67,71,72]: where at the building height (H = 30 m), the reference velocity U re f is 3 m/s. The friction velocity u * is 0.24 m/s. The von Kármán constant κ v is 0.41. The empirical constant C µ is 0.09. Six building configurations are considered in the current study ( Table 1). The UCL model of base cases is a building matrix with 5-row and 5-column blocks (Figure 1c). For the study of the open-space effect on city ventilation and human exposure, we remove the central building (Building 3-3) to obtain the open space (Location 3-3). For highrise building scenarios, we double the height of Building 3-3. Each type of building configuration is combined with tree-planting and tree-free types. According to the settings above, the cases are named "Case [urban layout, wind direction]", as shown in Table 1. For the comprehensive data analysis, we present the results both in the entire UCL Region A1 (5 × 5 building matrix) and in the Region A2 (the central area of Region A1, including Building 3-3/Location 3-3 and the surroundings), as shown in Figure 1d. For all tested cases, the minimum size of the hexahedral cells near wall surfaces is 0.2 m. The total number of grids ranges from approximately 5 million and 15 million for the treeplanting and tree-free cases, respectively. Figure 1e illustrates the grid arrangement from the top view and side view. This grid arrangement is sufficient to ensure the requirement recommended by CFD guidelines [69,70].

Description of Pollutant Dispersion Modelling
Scientists have found that traffic emissions have critical negative impacts on respiratory and cardiovascular function [73,74]. The situation in Asian cities might be more severe due to the very high population density and more residents living in close proximity to road traffic compared with those in European cities. To model the dispersion of traffic-related pollutants in the street canyon, reactive gaseous as primary particles [75], inert gas like CO [76], as well as reactive gaseous pollutants such as NO x and VOCs [77] are adopted in the simulation for investigating the dispersion of traffic-related air pollutants. Unlike various studies focusing on the reactive compounds from traffic such as NO x , VOCs and particles, we adopted monoxide (CO) as an indicator of the traffic emissions. Although NO x , VOCs and particles have more significant health impacts than CO, these compounds are more or less chemically or photo-chemically reactive, which means they could not be used as a tracer for the variation of the traffic emissions. To numerically investigate the impacts of various urban layouts on the physical dispersion of the traffic-related pollutants and related human exposure, a stable indicator is needed. As one of the main inert pollutants, CO has been widely used as a tracer of traffic emissions [42,46,48]. This study mainly focuses on the dynamic dispersion of traffic pollutants influenced by different urban layouts. Deposition and chemical reactions are not considered. The CO emission source is settled from z = 0 m to 0.5 m, with a width of 16 m and is 7 m away from the kerbside building (marked with dark grey colour in Figure 1f).
The governing equation of time-averaged CO concentration C (kg/m 3 ) is applied as Equation (11): where u j is the time-averaged velocity component in the direction of j. D m and D t are molecular diffusivity and the turbulent diffusivity of the pollutant. D t = ν t /S Ct , while ν t is the kinematic eddy viscosity, and S Ct is the turbulent Schmidt number. It is a parameter describing an important property of the flow defined as the ratio of the eddy diffusivity of momentum to the eddy diffusivity of mass. Di Bernardino et al. [78] found that S Ct increased with the height above the canopy, with the maxima of about 0.6 in their waterchannel experiments and simulations. In our preliminary work, we found that modelling with S Ct = 0.7 had the best performance compared with the wind tunnel experiment results from Gromke and Blocken [63]. Thus, S Ct = 0.7 is used throughout this work. The CO emission rate S is set as 1.25 × 10 −6 kg/m 3 /s, derived from a field observation campaign in a real street of Hong Kong [79]. Such an emission source setting has been adopted in CFD simulations for many studies of urban pollutant dispersion [10,27,42,80].

Description of the Vegetation Modelling
Tree planting is simulated as a series of cubic blocks on both sides of the streets over the whole urban area, except the surroundings of Building 3-3 ( Figure 1f). Since the size of the tree trunk is much smaller than that of the crown, the impact of the trunk on the airflow is assumed to be negligible, therefore only the crowns are simulated in the tree-planting cases. According to Yang et al. [80], the y-density is set to 1, which means that the tree crown is continuous in the y-direction. As shown in Figure 1f, the scale of the crown cubic is designed as 4 m × 6 m × 30 m. The distance between the crown bottom and the ground surface is 4 m, and that between the crown and the adjacent building wall is 3 m.
Differing from solid obstacles such as buildings, the airflow can pass through the tree crown from spaces in between the branches and leaves. Previous studies found that vegetation models with the porous medium for airflow and pollutant dispersion were more consistent with the wind tunnel experimental results than those with a non-porous medium [81][82][83]. Accordingly, we adopt the porous fluid zones to simulate tree planting, and the governing equations are listed in Equations (12)- (14).
where S u i , S k , S ε are the additional source and sink terms of momentum, turbulent kinetic energy and turbulent dissipation rate for trees, respectively. ρ (kg/m 3 ) is the air density. C d is the leaf drag coefficient, ranging from 0.1 to 0.3, which is related to the tree species.
In this paper, we adopt the commonly used empirical value of C d = 0.2 to avoid species particularity [81]. LAD (m 2 /m 3 ) is the leaf area density, which represents the one-side leaf area per unit volume of the crown [80,84]. It is related to tree species and crown-height variations, and ranges between 0.5 and 2.0 m 2 /m 3 for deciduous trees. To simplify the model, we supposed that the trees in the simulation domain were all deciduous trees with a homogeneous crown height, and thus set the LAD value to 1 m 2 /m 3 for the simulation. u i is the time-averaged velocity component on direction i, and U is the magnitude of the velocity. β p is the portion of turbulent kinetic energy converted from mean kinetic energy under the influence of drag, and β d is the dimensionless coefficient of the Kolmogorov cascade. We adopt β p as 1.0, and β d as 5.1, according to [81,82], respectively. Both C ε4 and C ε5 are empirical constants of 0.9.

Validations for Flow, Dispersion and Vegetation Modelling
The direct validation for the CFD model of real urban areas is difficult due to the very limited field observation data. The uncontrollable boundary condition is another challenge for repetitive experiments [85]. However, the wind tunnel experiment is a credible solution for model validation if the Reynolds number (Re) independence is satisfied (Re >> 11,000) [86][87][88]. We have implemented a series of comprehensive validations for the flow, the dispersion and the vegetation model applied throughout this work, based on the published wind tunnel experiment datasets [86,89,90]. Similar validation methods have been employed and proven valid in the literature [43,80,88].

•
Flow validation by wind tunnel tests of cubic arrays Figure A1 presents the results of the flow validation. The UCL model (moderate building density) with a 7 × 11 building matrix was used in the employed wind tunnel dataset [86]. The size of each building model is H = B = W = 15 cm. The measuring points for the vertical profiles of the stream-wise velocity (u) and turbulence kinetic energy (k) are set in the centre of each street, named Pi (i = 1-6) ( Figure A1a Figure A1c-j tests the grid independence (with a minimum grid size of 0.4 m, 0.2 m and 0.1 m) and the performance of different turbulence models (standard k-ε, RNG k-ε and realisable k-ε models) with standard wall function. The results illustrate that differences generated by the mesh setting are negligible. Thus, the moderate grid size (0.2 m for minima) is applied for all cases to save computational resources. Furthermore, the depicted vertical profiles of u and k verify that modelling using the standard k-ε model has better agreement with the wind tunnel data than those using the RNG k-ε and realisable k-ε models. Important statistics are summarised in Table A2, including the normalised mean square error (NMSE), fractional bias (FB) and correlation coefficient (R). The results denote that the standard k-ε model is employed in the study throughout this work.

• Pollutant dispersion validation by wind tunnel tests without tree models
Wind tunnel experiment data of inert gas dispersion [89] is employed in our work for the validation of pollutant dispersion. The configurations of the experiment are illustrated in Figure A2a [89] are set for the domain inlet. The standard k-ε model and standard wall function are adopted in the simulation. Since the tracer gas concentration provided by the wind tunnel experiment is in a non-dimension form, the normalised concentration K [89] is derived referring to Equation (15), in convenience for comparing the experimental data and simulation results.
where C is the inert gas concentration, and the emission rate E is 0.01 m/s. Important statistics are summarised in Table A3, including the NMSE, FB and R. The good agreement between the results of the wind tunnel experiment and the CFD simulation confirm that the selected turbulence model and wall function is appropriate for evaluating the pollutant dispersion in our work.

• Pollutant dispersion validation by wind tunnel tests with tree models
The validation for vegetation modelling in this paper is performed on the basis of the wind tunnel experiment conducted by Gromke and Ruck [90]. The configurations and boundary conditions are set as in Figure A3a. The 2D street canyon is constructed by two parallel building models, with the same sizes for both in the wind tunnel experiment and the CFD simulation (L × W × H = 1.2 m × 0.12 m × 0.12 m). The standard k-ε model and standard wall function coupled with the porous crown model (details in Section 2.2.4) are adopted in the simulation. The vertical profiles of u, k and ε for the domain inlet are provided by Gromke and Ruck [90]. U ref = 4.7 m/s is applied; thus, the reference Re is 38,630 >> 11,000. The normalised concentration K of the inert gas (SF 6 ) is used to compare the results of the wind tunnel experiment and the simulation. Vertical profiles of K are presented in Figure A3b,c. Important statistics are summarised in Table A4, including the NMSE, FB and R. In general, the results satisfy the recommended criteria [91,92], except that of y/H = 2 at the leeward wall. Nevertheless, as we focus on the pollutant dispersion in the 2D street canyon, the results of y/H = 0 (central region of the canyon with fully developed turbulence) are more representative of the pollutant distribution features. Good agreements of these statistics at y/H = 0 also confirm the modelling accuracy and reliability of the simulations. These results verify that the porous crown model with the standard k-ε model and standard wall function has good performance, and is suitable for studying the tree-planting effects in this work.     (Figure 4a). On the contrary, with the high-rise building (Figure 4e), the wind speed is significantly enhanced, especially near the windward side of the high-rise building. Compared with the base case, the former vortex on the left side of Building 3-3 is concentrated to one-half of the original size and the right vortex has disappeared.  Figure 2b,d,f presents the flow field in the domain with tree planting. Whether for base cases (Figure 2a,b), open-space cases (Figure 2c,d) or high-rise building cases (Figure 2e,f), tree planting slightly changes the flow field in the entire domain (Region A1). Small vortices are generated over the whole domain, and the continuity of the flow is obstructed. However, the overall wind speed in Region A1 remains a similar value with the addition of vegetation under a parallel approaching wind.

Influence of Tree Planting on Airflow
This finding is further confirmed in Region A2, as presented in Figure 3. The wind vector around trees and street corners becomes denser and more complex. Tree planting slightly reduces the pedestrian-level wind speed in the central area of the base cases (Figure 3a,b) and open-space cases (Figure 3c,d). However, for high-rise building cases (Figure 3e,f), the VR value in Region A2 slightly increases with tree planting.
The vertical profiles of the flow field in Region A2 at y = 135 m are shown in Figure 4. We find that tree planting has a slight influence on the vertical airflow for all three building configurations. Compared with Case [Open, 0 • ], the two-vortices structure of the flow field is destroyed (Figure 4c,d). In the base cases (Figure 4a For Case [High, θ], the VR at z = 2 m has been significantly enhanced in both Region A1 and A2 compared with Case [Base, θ]. The ambient airflow of the high-rise building (<VR> A2 ) is enhanced by 52.78-119.05%, more strongly than that in the whole building matrix (<VR> A1 ) by 4.73-23.36%.
The results in Figure 5a,b and Table A5 show that tree planting reduces the VR in both Region A1 and A2 for different building configurations with all four approaching wind directions (θ = 0   To investigate the impacts of building configurations on pollutant diffusion, Figure 6a,c,e presents the distributions of CO concentration (C) at the pedestrian level (z = 2 m) in tree-free cases. Overall, the level of C in the three tree-free cases is similar, but the regions of high C are affected by different building configurations. In general, open space enhances the CO accumulation on the leeward side of the buildings in the central and downstream areas. Nevertheless, the high-rise building enhances the CO accumulation in the central and upstream area significantly. However, CO in the central area is diluted to a quite low level due to the strongly strengthened wind velocity surrounding the high-rise building. To better understand how building configurations affect the pollutant dispersion in the central region (Region A2), Figure 7 shows the detailed vertical distribution of C at the x-z plane (y = 135 m). Comparing Figure 7c with Figure 7a, the low wind speed weakens the dilution and leads to high C levels in the open area. Particularly at the near-ground level of the upwind area, C is higher than 13 mg/m 3 , while C in the same area of the base case is about 5 mg/m 3 . In contrast, the strong airflow in the upwind of the high-rise building evidently decreases the C (Figure 7e). The near-ground C is decreased to about 2 mg/m 3 . Additionally, the CO distribution in the upwind of the high-rise building is reduced to a very limited vertical range, while that in the downwind is expanded due to the existence of the high-rise building.  Figure 6b,d,f illustrates the C distribution in the urban area with tree planting, coupled with the basic design, open-space and high-rise-building design, respectively. Comparing them with Figure 6a,c,e, a significant increase of C is found in the whole urban area under the tree-planting design, no matter which type of building configuration is considered. The CO dispersion is significantly weakened by trees in the whole domain, and new hotspots with high C appear. Figure 7b,d,f. Compared with Figure 7a,c,e, tree planting evidently increases the near-ground C on the leeward side of all buildings. Particularly in Case [Open-tree, 0 • ], the near-ground C in the upwind of the open space increases to higher than 15 mg/m 3 . Trees around the high-rise building also have a significant influence on CO dispersion (Figure 7f). An area with a high C appears at the upper layer of the building wall on the upwind of Building 3-3, corresponding to the vortex of the flow field in Figure 4. At the downwind of Building 3-3, the ground-level C is higher than 15 mg/m 3 . Moreover, another hotspot with C higher than 15 mg/m 3 appears in the area of the tree crown. For the base cases, <CO> A1 slightly rises from 4.14 mg/m 3 to 5.24 mg/m 3 with the θ varying from 0 • to 45 • . <CO> A2 , with the range of 4.87-7.11 mg/m 3 , is higher than <CO> A1 with the same direction of approaching flows. Both <CO> A1 and <CO> A2 decrease in the cases with open space and high-rise buildings (Figure 8a,b). For the open-space cases, <CO> A1 and <CO> A2 decrease rapidly with all four wind directions compared with the base cases, by 7.83-20.54% and 0.08-24.43%, respectively. For the high-rise building cases, the decrement ranges from 4.39% to 23.00% for <CO> A1 , and ranges from 43.88% to 47.40% for <CO> A2 , in comparison with base cases. The wind direction influences C more significantly for the area around the high-rise building (Region A2) than for the whole domain (Region A1).

Detailed vertical distributions of C surrounding Building 3-3 in the central region (Region A2) with different building configurations are illustrated in
For cases with tree planting, the results show that both the <CO> A1 and <CO> A2 evidently increase in all conditions with increasing rates of 2.84-31.88% and 2.85-35.46%, respectively. Taking θ = 0 • as an example, <CO> A1 increases by 20.19%, 22.44% and 12.61% in Case [Base-tree, 0 • ], Case [Open-tree, 0 • ] and Case [High-tree, 0 • ], compared with the tree-free cases. For the base cases, the largest increasing ratios of <CO> A1 and <CO> A2 both appear when θ = 0 • . With tree planting, the CO concentration at the pedestrian level and in the central area (<CO> A2 ) are higher than in the entire urban area (<CO> A1 ) for both the base cases and open-space cases. The <CO> A2 is particularly high in open-space cases with tree planting, with the values ranging from 6.59 mg/m 3 to 7.89 mg/m 3 . For the high-building cases with tree planting, both <CO> A1 and <CO> A2 increase by 12.61-26.38% and 6.10-21.19%, respectively, compared with the tree-free cases. Regardless of the building configuration, the tree-planting design obviously weakens the dilution and dispersion capacity of pollutants and remarkably increases the CO concentration.

Impacts of Building Configurations and Tree Planting on <P_IF>
As illustrated in Section 2.1.2, we use <P_IF> to quantify the influence of urban layouts on personal exposure in UCL. As the buildings are assumed to be a residential type with natural ventilation, the pollutant concentration (C) on the building surfaces is adopted as the indoor concentration due to I/O ≈ 1 [6,7]. Figure 9a-f plots C on the building surfaces in different cases when θ = 0 • . The C distribution is apparently influenced by different building configurations and tree planting, especially in the central area. Figure 10 and Table A7 summarise    As displayed in Figure 6a,c,e, the open space and high-rise building change the C distribution on the building surfaces. Both layouts make CO accumulate in the centre of the building matrix, while regions of high C in the base cases are in the downstream area of the approaching flow (Figure 9a-f). Comparing Figure 6b,d,f with Figure 6a,c,e, tree planting significantly increases the C on building surfaces, especially of the central 3 × 3 building matrix.
Similar results can also be found in Figure 10 and  . Moreover, since the variation of <P_IF> is closely related to CO concentration, the <P_IF> of Case [Open] will increase and that of Case [Open-tree] will decrease in Region A1−A2. It is also the opposite of that in Region A2.

Discussion
As critical determinants for urban ventilation and pollutant dispersion, the impacts of tree-planting plans and varied aspect ratios of 2D street canyons have been investigated in previous studies through both field experiments [93] and numerical simulations [80]. Chen et al. investigated the effect of different tree-planting parameters on the urban thermal and wind environment by conducting scaled outdoor field experiments [93]. Tree planting was found to reduce the pedestrian-level wind velocity in street canyons with all investigated AR values. The decreasing rate ranged 29-70%. Although the experiments are conducted in 2D idealised street canyon models, we also find that tree planting has a restraining effect on the urban wind in our work as well. Yang et al. [80] evaluated the integrated impact of tree planting and various AR values in a full-scale street canyon by CFD modelling (standard k-ε model) with the same emission settings as ours, and concluded that tree planting can lead to the reduction of velocity by various magnitude and an increase in CO exposure. In the canyon with AR = 1, the tree-induced CO increment is almost 70% (from 9.63 mg/m 3 to 16.30 mg/m 3 ). However, in 2D street canyon models, only the condition with perpendicular approaching wind to the street axis is considered, which corresponds to the worst ventilation situation, since only air exchange across the street roof contributes to pollutant removal. In this work with a 3D urban canopy, the ventilation can be better than in 2D models and is closer to that of the real urban community. With our 3D UCL model, the largest decreasing rate contributed by the tree planting is about 22% in spatial mean (Case [Open-tree, 0 • ] vs Case [Open, 0 • ]). Meanwhile, the tree-induced CO increment ranges 2.84-35.46% in this work (Table A6). We can conclude that even with the same AR (AR = 1) and tree-planting plan (tree planting on both sides of the street), the natural ventilation and dispersion conditions in the 3D building matrix are much better than that in 2D street canyons.
Using dimensional variables to evaluate the variance of the residents' exposure owing to the varied impactors in different cities has huge challenges, because the emission strengths of different sources are not the same and may even be at different orders. Meanwhile, the size of the target population in different studies may vary significantly. As mentioned in Section 2.1.2, the variable IF has been used to express the source-to-intake relationship for vehicular pollutants in realistic street canyons [48], but this would be strongly affected by the population size and the spatial scale. For example, Habilomatis and Chaloulakou [45] found that the IF of vehicular ultrafine particles is 371 ppm in a 2D street canyon in the central area of Athens. The IF at the city scale is relatively small. Marshall et al. [94] reported that the IF of particles in US cities ranges from 1 to 10 ppm and the IF of CO is 270 ppm in Hong Kong, with a huge population size [46]. The IF of particles at the regional scale are reported to range from 0.12 to 25 ppm in the entire United States [95].
Consequently, the variable <P_IF> is derived and applied for the exposure assessment in this work. This normalised exposure index <P_IF> is more suitable for evaluating and comparing the exposure risks in different areas, since the influence caused by different orders of the population size and pollutant emission rates is avoided. Hang et al. found the <P_IF> of CO in the tree-free idealised 2D street canyon (AR = 1) was 5.21 ppm [42]. Yang et al. found the <P_IF> of CO in tree-planted 2D street canyons (AR = 1, LAD = 1) were 5.60 and 5.58 ppm, and the values raised with increased AR. When AR is raised to 5, the <P_IF> is an order of magnitude larger than that with AR = 0.5, 1 and 3. Comparing these works in 2D street canyons, the <P_IF> of CO in the 3D UCL model ranges from 1.54 to 2.87 with various tree-planting plans, building configurations and wind directions. The values in all scenarios are much lower than those in 2D street canyons.
According to the comparison above, an important suggestion for the urban designer is to avoid building 2D street canyons either too deep or too long in urban districts. If it cannot be avoided, more leakages and a wider roadway could improve the ventilation conditions in 2D street canyons. Moreover, if the street canyon is designed longer than 8H [17], the axis of the street should be approximately parallel to the prevailing wind direction.
To simplify the calculation process, we adopt idealised 3D UCL models in this paper. The building models are all assumed to be residential-type and are highly simplified with the same configuration (in a 5 × 5 building array) for the case study. The trees are treated as cubes of porous media. Neutral atmospheric conditions are adopted, and inert gas (CO) is considered as the tracer pollutant from the traffic emissions. Only four wind directions (θ = 0 • , 15 • , 30 • , 45 • ) are considered in this research. Nevertheless, the real urban environment is affected by various parameters. Thus, it is worth mentioning that the results may be significantly different if urban morphologies, atmospheric conditions or other parameters are changed.
The impacts of urban morphological parameters in realistic urban areas are much more complicated than in such an idealised model. The study of different building coating plans and the direct radiation effect of the aerosol within urban canopy, as well as their impacts on the urban thermal environment and human outdoor thermal comfort (Figures 3 and 4), is being implemented now. In future work, more kinds of realistic factors and conditions will be carefully considered and evaluated, including non-neutral atmospheric conditions and radiation impacts, the chemical reactions and composition of air pollutants, and more complicated urban morphological arrangements. Furthermore, the different tree species and the pollutant deposition on trees will also be considered in the ongoing work. CFD simulations coupling turbulence and radiation models will be validated by our scaled outdoor experiments (H = 1.2 m), as reported by Chen et al. [96,97]. These works will be adopted in numerical studies for full-scale realistic or idealised urban models. Our work is a step-by-step approximation of the real urban situation using the idealised model, and we are constantly improving our work on the way towards approaching the final target.

Conclusions
This paper is novel in that it numerically investigates the integrated impacts of open space, high-rise buildings and tree planting on urban airflow, pollutant dispersion and related human exposure in 3D idealised UCL models (5-row and 5-column, aspect ratio H/W = 1, building plan area fraction λ p = frontal area aspect ratio λ f = 0.25) under neutral atmospheric conditions. Four approaching wind directions (parallel 0 • and non-parallel 15 • , 30 • , 45 • ) are considered. The computational fluid dynamics (CFD) simulations with the standard k-ε model are well validated by the wind tunnel data from the literature. The personal intake fraction P_IF and its spatially-averaged value for the entire UCL building surfaces <P_IF> are adopted to quantify the pollutant exposure on residents.
The CFD simulation results show that open space, high-rise building and tree planting all have strong effects on the flow structure, pollutant dispersion and residents' exposure. Some meaningful findings are concluded as follows:  This work provides a practical and efficient method to investigate the impacts of synthetic urban layouts on urban ventilation and pollutant dispersion. This work also extends the application of the CFD methodology to the assessment of exposure, and consequently connects to the area of public health. The method is applicable for further study coupling with more kinds of urban configurations under various atmospheric conditions. The results can provide helpful references for urban designers developing the sustainability of the city.           3. Quantitative investigations of the flow, pollutant concentration and traffic-related exposure