Evaluation of the ventilation and pollutant exposure risk level inside 3D street canyon with void deck under different wind directions

With continuous global warming, growing urban population density, and increasing compactness of urban buildings, VD (void deck) street design has become increasingly popular in city planning, especially in tropical countries. However, understanding on traffic pollutant dispersion inside the street canyons with VDs is still at early stage. This paper evaluates quantitatively the effects of VD location and wind direction on the ventilation and traffic pollutant exposure inside the street canyon with VDs. The results show that under seven wind directions (0°, 15°, 30°, 45°, 60°, 75°, and 90°), the VD provides higher ACH than that of the regular canyon, especially at high α (angle between the approaching wind and the canyon axis). Also, mean K (dimensionless pollutant concentration) values of the canyon wall and pedestrian respiration plane on one side where VD is located are significantly reduced compared to the regular canyon. Therefore, when VDs are at both buildings, both pedestrian respiration planes and walls have the lowest K values, thus providing the best living environment for pedestrians and near-road residents. In addition, as α increases, the K values on both respiration planes significantly decrease except for the leeward respiration plane of the canyon with the windward VD. These findings can help to design urban street canyons for mitigating traffic pollution risk and improving ventilation in tropical cities with frequently changing wind directions.


Introduction
Nowadays, with the increase of urban population density, the urban structure is becoming more and more compact (Issakhov and Tursynzhanova 2022;. At the same time, due to rapid industrial development and worsening global warming, urban residents live in extreme thermal discomfort and traffic gas damage, especially in tropical Southeast Asian countries (Chen and Norford (2017); Santiago et al. (2021); Yim et al. (2010); Zhao et al. (2021)). Therefore, many studies have been conducted over the past 30 years to improve air quality in the urban street canyons (Baik et al. 2012;Ricci et al. 2019;Antoniou et al. 2019).
According to the results of previous studies, wind direction has a significant influence on the airflow field and pollutant dispersion inside the street canyon (Jeanjean et al. 2015;Plate 1999). However, most of the studies have been conducted under the perpendicular approaching wind condition (α = 90°), as the pollutant dispersion process is considered to be the worst in this case (Soulhac and Salizzoni 2010). Some studies considering different wind directions are summarized as follows. Plate (1999) studied the influence of the different wind directions on the pollutant distribution inside the regular symmetrical street canyons with the length-to-height ratio of L/H = 5 and L/H = 10 through WT experiments. For the canyon with L/H = 10, when the wind direction changes from perpendicular to parallel, the pollutant concentration decreases at the measurement point (y = 0, z/H = 0.083) of the upwind building. Huang et al. (2019) Responsible Editor: Marcus Schulz performed CFD numerical simulation for 3D street canyons with L/H = 10 to investigate the effect of wind direction on the pollutant dispersion inside the canyon. According to their results, the air exchange rate (ACH) of the canyon has maximum and minimum values at α = 30° and 90°, respectively. The mean dimensionless pollutant concentration (K) on the leeward wall increases significantly with increasing α, reaching a maximum value at α = 75°. The K on the windward wall has a maximum value at α = 0°. Gromke and Ruck (2012) conducted WT tests on street canyons with different aspect ratios under three wind directions (α = 90°, 45°, and 0°) and different tree-avenue model conditions. The experimental results show that K on both walls of the canyon increases at 45° and 90°, while the effect of street trees is greatest at 45°.
In addition, the street canyon structure (symmetric or asymmetric structure Reiminger et al. 2020), VD (Huang et al. 2020), viaduct (Hao et al. 2019), building roof ) also significantly affects the airflow and pollutant dispersion inside the canyon. Among them, VDs can provide very beneficial effects in removing traffic pollutants and ameliorating thermal discomfort within the street canyons by substantially changing the flow fields (Ma et al. 2021;Huang et al. 2021). VD is an architectural structure widely used in urban design of tropical countries, which refers to the elevated ground floor under public apartments (Chew and Norford 2019;Sin et al. 2022). Unlike temperate regions, the seasonal division is not obvious in the tropical zone, and the outdoor temperature is very hot. Therefore, the residents living in tropical regions experience extreme heat stress, especially inside the street canyon. The VD design is welcomed in tropical areas because it can relieve the thermal stress of urban residents and provide beneficial and comfortable living space (Li et al. 2018;Roth and Chow 2012). Moreover, VDs can improve the effective self-ventilation capacity within street canyons, thereby reducing infection rates of infectious diseases such as influenza, especially the Covid-2019 pandemic (Xu et al. 2020;Pani et al. 2020;Ahmadi et al. 2020). Therefore, in recent years, many studies have been carried out to investigate the airflow and pollutant dispersion inside the street canyons with VDs to improve the wind speed at pedestrian respiration level and to reduce damage from traffic pollutants (Weerasuriya et al. 2020;Huang et al. 2020Huang et al. , 2021Chen and Mak 2021). Huang et al. (2021) conducted 2D CFD numerical simulations to analyze the effects of various street structures and found that VDs have a significant impact on the pollutant distribution inside the canyon. Under the perpendicular approaching wind condition, when VDs are at both buildings, the K values on the leeward and windward walls are reduced by 1.6% and 21.6%, respectively, compared to the regular canyon without VDs. Chew and Norford (2018) investigated the pedestrian-level wind environment inside 2D street canyon under different VD configurations and different building heights through water-channel experiments and numerical simulations. The results showed that the mean wind speed at the pedestrian level can be increased by two times. After that, Chew and Norford (2019) revealed that the VD height significantly affects the wind enhancement, while building height has a minor effect. Increasing the building height from 24 to 48 m increases pedestrian-level wind speed by only 10%. Zhang et al. (2019) found that the 1st floor VD reduces NO 2 and NO concentrations within deep street canyon (H/W = 5) by one or two orders. Chen and Mak (2021) evaluated the pedestrian-level wind comfort around various "lift-up" building models with non-traditional configurations. VDs can significantly improve wind comfort around the buildings, and the effect varies greatly depending on the wind direction (0°, 90°, 180°) and building shape.
As mentioned above, most previous studies have focused on improving thermal comfort and wind speed at the pedestrian level, while studies on the traffic pollutant dispersion inside the street canyons with VDs have considered only the perpendicular approaching wind condition. Although wind direction varies seasonally and temporally, each region has its own dominant wind direction. Therefore, consideration of the wind direction in urban street design is very important to improve the canyon ventilation and mitigate the pollutant exposure risk to residents in street canyons. However, the effect of wind direction on the traffic pollutant dispersion within street canyons with VDs has not been studied. Therefore, this study explores the effects of wind direction on the flow structure and pollutant distribution inside canyons with VDs. The structure of the paper is illustrated as a flowchart in Fig. 1. In this study, 21 numerical models validated based on the WT experimental dataset reported by Kastner-Klein (1999) were adopted. The ventilation capacity of each street canyon was quantified by the ACH, a key factor in estimating the ventilation capacity of the street canyon, and the pollutant concentrations at both walls and pedestrian respiration level were quantitatively evaluated from the health perspective of near-road residents and pedestrians.

Governing equations for flow and pollutant dispersion
In this study, two assumptions are given: firstly, the airflow inside and outside the street canyon is regarded as an incompressible steady-state turbulent flow; secondly, the turbulence caused by solar radiation and traffic is neglected. According to previous studies, large eddy simulation (LES) outperform Reynolds-Averaged Navier-Stokes (RANS) in turbulence prediction (Tominaga and Stathopoulos 2011;Gousseau et al. 2011;Blocken 2015). Nevertheless, RANS models have been widely used to study airflow and pollutant dispersion in 2D/3D street canyons (Chen 2009;Cui et al. 2016;Allegrini et al. 2014;Sanchez et al. 2017;He et al. 2017). This is because RANS has a relatively short computation time compared to LES and provides good results for the mean and spatial mean flow characteristics (Lin et al. (2014); Salim et al. (2011b)). LES is also more difficult than RANS to generate the appropriate time-dependent wall and inlet boundary conditions (Sha et al. (2018); Tabor and Baba-Ahmadi (2010)). Among the RANS approaches, the standard k-ε turbulence model has been recommended as one of the best choices because it provides good agreement between CFD simulation and WT measurement results on the airflow and pollutant distribution within the canyon (Ai et al. (2013); Huang et al. (2009);Jin et al. (2016); Sha et al. (2018)). Therefore, in this study, the incompressible and isothermal 3D steadystate flow fields inside the street canyons with VDs are analyzed using RANS standard k-ε turbulence model. The equations are given in Eqs. (1)-(2)  where k is the turbulent kinetic energy, and ɛ is its dissipation rate, v t = C k 2 is the turbulent eddy viscosity, the constants of the standard k-ε turbulence model are (C μ , C ε1 , C ε2 , σ k , σ ε ) = (0.09, 1.44, 1.92, 1.0, 1.3). The steady-state species transport equation for estimating the dispersion of gaseous pollutants is given as follows by Eq. (3). where u, v, and w are the mean velocity components in x, y, and z directions, respectively. C is the molar concentration of species β, D m is the molecular diffusivity of species β, and S p is the pollutant source term. The turbulent Schmidt

Street canyon configurations
In this paper, the canyon model employed in the WT experiment performed by Kastner-Klein (1999) is selected as the reference canyon model for model validation. Figure 2 sketches the selected regular canyon configuration. As shown in Fig. 2, the regular street canyon is set as a space between two identical buildings placed in parallel. Four line sources are placed at the bottom of the canyon to mimic emission of traffic pollutants. The dimensions (H × W × L) of each building are H (height) × H (width) × 10H (length).
The distance between two buildings is H, namely the canyon width is equal to the width of building. Therefore, a long street canyon of H (height) × H (width) × 10H (length) forms in the space between the two buildings. Here, H is fixed to 0.12 m (18 m at the full scale). The line sources (0.042H (height) × 0.042H (width)) are located 0.23H and 0.35H away from the nearby canyon walls (Fig. 2b). In addition, to account for vehicle exhaust emissions at the road intersection on both sides of the canyon, the four line sources on both sides of the canyon are extended by 0.92H each (Salim et al. 2011a). In other words, the length of each line source is 11.84H. Also, to assess the exposure risk of traffic pollutants for pedestrians in street canyons, pedestrian respiration planes are set near both walls of the canyon. The pedestrian respiration plane is set at a height 0.083H (1.5 m at the full scale) above the ground and a width 0.17H (3 m at the full scale) from the canyon walls (Fig. 2a). Based on the regular street canyon configured above, the canyons with VDs are constructed as shown in Fig. 3: the canyon with VD at the upwind building ( Fig. 3a), the canyon with VD at the downwind building (Fig. 3b), the canyon with VDs at both buildings (Fig. 3c). In addition, buildings with VD (height H/6, 3 m at the full scale) are modeled as ideal blocks with empty space below (Chew and Norford (2019); Huang et al. (2021); Sin et al. (2022)). For the three street canyons constructed above, which are constructed according to the VD location, the influence of seven wind directions are considered; therefore, twentyone numerical cases are adopted in this study.

Computational domain and grid generation
In this study, the computational domain is adopted according to the guideline by Tominaga and Stathopoulos (2011). The computational domain is fixed by introducing a 3D rectangular coordinate system, where the coordinate origin is located at the center of the canyon bottom. Figure 4 represents the computational domain when the wind approaches perpendicular to the canyon (α = 90°). Here, the inlet plane is set at a distance 9.5H from the coordinate origin, the outlet plane is 20.5H, the top plane is 8H, and both lateral planes are 12H. That is, the dimensions of the computational domain are 30H (length) × 24H (width) × 8H (height).
The computational models are built and meshed using Gambit 2.4.6. In the canyon and VDs, hexahedral grid cells of size δx = H/72, δy = H/12, and δz = H/72 are generated. The remaining computational area is meshed by tetrahedral grid cells to maintain stable simulation under oblique wind condition. Here, the maximum grid expansion ratio is limited to 1.08. For the oblique wind cases, the computational domain does not change; only the canyon model rotates by the corresponding angle with respect to the coordinate origin. See Subsection 2.5 for details.

Boundary conditions and numerical scheme
The commercial CFD software ANSYS-Fluent 14.5 is used for CFD numerical simulation. The boundary conditions for all simulation cases are as follows. On the inlet plane, the atmospheric inlet boundary conditions are imposed. For model validation, the inlet profiles of the horizontal wind velocity (U), turbulent kinetic energy (k), and its dissipation rate (ɛ) are set as follows, identical to the experiments performed by Kastner-Klein (1999). where u * is the friction velocity (0.54 m/s), δ is the depth of the boundary layer (0.5 m), and c vk is von Karman constant (0.4). For the outlet plane, the zero-gradient boundary condition is given. The top and both lateral planes have symmetric boundary conditions, while the ground and building walls have no-slip boundary conditions. The volumetric flow rate of the mixture of air and SF 6 is provided to the line sources. For CFD simulation, SIMPLE algorithm is adopted to deal with the pressure-velocity coupling. The second-order upwind scheme is employed for the spatial discretization of the governing equations, and the residuals of all variables are kept at 10 −6 .

Dimensionless pollutant concentration (K)
In this study, SF 6 concentrations were expressed in dimensionless as follows (Kastner-Klein 1999): where K is the dimensionless pollutant concentration, C is the measured volume fraction of SF 6 , U ref (4.7 m/s) is the free-stream wind speed at a reference height z ref (0.12 m), L is the line source length (1.42 m), and Q s is the source strength.

Air exchange rate
This paper utilizes the ACH to evaluate the self-ventilation capability of canyons with VDs under the different wind directions. The ACH represents the amount of air exchanged per unit time between street canyon and freeatmosphere, which has been widely used as an important index to evaluate the ventilation capacity of the canyons  2022)). For 3D street canyons with VDs, air exchange can occur at the interfaces between the canyon and VDs as well as at both sides and top plane of the canyon. Therefore, the ACH of the canyons with VDs can be evaluated as follows .

Grid independence analysis
Since the accuracy and calculation cost of numerical simulation are determined by the number of grids, the independence analysis is performed for three cases (including the fine, basic, and coarse grids) under the perpendicular approaching wind condition (α = 90°). For the three mesh resolution cases, the grid size of the canyon and VDs are H/96, H/72, and H/48, respectively. Also, the grid expansion ratio of 1.02, 1.05, and 1.08 are applied outside the canyon, respectively. Therefore, fine, basic, and coarse grid models have 7.2 million, 3.7 million, and 2.4 million cells, respectively. Figure 5 depicts the K profiles obtained along three vertical lines (y/H = 0, 1.26, and 3.79, respectively) near both walls of the canyon for the three mesh cases under the perpendicular wind direction (α = 90°). Each vertical line is 5 mm away from the wall. On the leeward wall (Fig. 5a), the difference in K between the basic and coarse grids is relatively large (> 20%); however, the increase in mesh resolution from "basic" to "fine" results in a much smaller K deviation (< 2%). For the windward wall (Fig. 5b), the differences in K profiles between the three mesh resolutions is negligible. Therefore, when α = 90°, the basic grid of 3.7 million cells is determined as the independent grid, and all subsequent models for other wind directions are meshed around this grid number, 3.65-4.09 million.

Model validation
The validation study is performed by using the results of 3D street canyon (model scale 1:150) used in WT experiments conducted at the University of Karlsruhe, Germany (see the internet database CODASC, https:// www. umwel taero dynam ik. de/ bilder-origi nale/ CODA/ CODASC. html). In their experiments, a mixture of air and tracer gas (SF 6 ) with a flow rate of 1000 ppm was continuously emitted from the line sources. SF 6 concentrations were measured at specific points on both walls of the canyon. Figure 6 shows that the CFD simulation results are in good agreement with the WT experimental data at α = 0°, 45°, and 90°. This means that the employed RANS standard k-ε model (Sc t = 0.3) can ensure the accuracy of CFD numerical simulations. In addition, the performance of the CFD model was statistically validated. As shown in Table 1, various statistical indicators (the fraction of predictions within a factor of two of the observations (FAC2), the fractional bias (FB), the normalized mean square error (NMSE), and the correlation coefficient (R)) were evaluated quantitatively using WT and CFD results. From Table 1, it can be found that all statistical metrics are within the required acceptable range (Moonen et al. 2013). Therefore, the numerical models configured in this study are reliable for simulating the airflow and pollutant dispersion inside the street canyons with VDs under different wind directions.

Results
Based on the above validated numerical models, we evaluated the airflow and pollutant distribution characteristics inside the canyons with VDs compared to the case of regular canyon , depending to the wind direction.   (Figs. 8a and 9a). For canyons with both VDs, the ACH is the largest (106.5%), since the airflow components (ACH VD1 and ACH VD2 ) entering through both VDs are almost twice as large as when the VD is on one side. For oblique wind conditions, the variation of ACH is more pronounced with the change of VD location, where the canyons with the leeward VD have higher ACH values than that of the canyons with the windward VD. This is because the airflow component (ACH VD1 ) entering the canyon through the leeward VD is much higher than the airflow component (ACH VD2 ) flowing through the windward VD (see Table A2). As α increases, the ACH of the canyons first increases and then decrease slightly. The maximum ACH is 60°, 30°, and 60° wind direction when VDs are located at the upwind building, downwind building, and both buildings, respectively. For the canyons with the leeward VD, the airflow component (ACH Side1 ) decreases with increasing α, but the airflow components (ACH Top and ACH VD1 ) increase more (Table A2). Therefore, as α increases, the ACH of the canyon increases, and it has a maximum value (146.0%) when α = 60° and a minimum value when α = 0°. At α = 75° and 90°, the ACH value of the canyon decreases compared to the case of α = 60°. This is due to the significant decrease in the airflow component (ACH-Side1 ) compared to the changes in the airflow components (ACH Top and ACH VD1 ). For the canyons with the windward VD, the airflow component (ACH Side1 ) decreases with increasing α. The airflow component (ACH VD2 ) has the lowest value at α = 45° and the highest value at α = 90°. The airflow component (ACH Top ) increases with the increase of α, reaching a maximum value at α = 45° and then decreasing again. According to these changing characteristics of the airflow components, the ACH of the canyon with the windward VD reaches a maximum at α = 30° (118.8%) and a minimum at α = 75° (104.9%). For the canyons with both VDs, the airflow component (ACH Side1 ) decreases with increasing α, while the airflow component (ACH VD1 ) entering the canyon through the leeward VD increases significantly. The airflow component (ACH Top ) first increases with increasing α, reaches a maximum at α = 60°, and then decreases again. Therefore, the maximum ACH (155.8%) of the canyon occurs at α = 60° and the minimum ACH occurs at α = 0°.

ACH of the street canyons
In addition, the mean component ( ACH ) of the canyons with VDs decreases and the turbulence component (ACH′) gradually increases as α increases (see Table A1). This means that the effect of turbulence on the canyon ventilation becomes more pronounced with the increase of α. Figures 8, 9, and 10 show the airflow and pollutant distribution patterns on the vertical central plane of the canyon according to seven wind directions (α = 0°, 15°, 30°, 45°, 60°, 75°, 90°) when VDs are at the upwind building, downwind building, and both buildings. As can be seen in Figs. 8, 9, and 10, depending on the VD location and wind direction, completely different airflow and pollutant patterns develop inside the canyon.

Airflow and pollutant distribution on the vertical central plane
For the canyons with the leeward VD (Fig. 8), more pollutants accumulate near the windward side compared to the leeward side. When α = 0° (Fig. 8a), the weak airflow entering through the leeward VD flows upward, creating small vortices above the roof of two buildings and a clockwise vortex near the lower part of the windward wall. Since the wind direction is parallel to the canyon, pollutants are not easily removed upward, and it mostly accumulates near the ground. When α = 15° (Fig. 8b), the airflow entering through the leeward VD flows upward along the windward wall, forming a small counterclockwise vortex near the lower part of the leeward wall. However, since the airflow velocity is still slow, the pollutants cannot easily move upward and finally form a high-concentration region on the ground near the windward wall. From α = 30° (Figs. 8c-g), a large counterclockwise vortex and a small clockwise vortex are formed near the leeward side by the airflow entering through the leeward VD. Meanwhile, the core of the large vortex gradually moves toward the windward side as α increases. This is because the strength of airflow entering the canyon through the leeward VD increases as α increases (Table A2). Therefore, For the canyons with the windward VD (Fig. 9), the flow field inside the canyon varies considerably with the wind direction and the pollutant distribution is determined accordingly. For the case of α = 0° (Fig. 9a), the flow field and pollutant pattern inside the canyon are opposite to those of the canyon with the leeward VD (Fig. 8a). When α = 15° and 30°, a part of the free-stream flows down along the windward wall and then exits through the windward VD. Therefore, when α = 15°, some pollutants accumulate at the windward VD (Fig. 9b). When α = 30°, the airflow velocity flowing to the windward VD increases, so the vortex formed inside the canyon is more developed. Therefore, more pollutants accumulate on the leeward side compared to the case of α = 15° (Fig. 9c). When α = 45° (Fig. 9d), no airflow enters the canyon top plane and the airflow entering through the canyon entrance creates a large clockwise vortex within the canyon by Fig. A2d. The pollutants generated at the canyon floor are accumulated near the leeward wall by this vortex. In addition, this vortex is stronger than the α = 15° and 30° cases, so more pollutants accumulate on the leeward wall. When α = 60°, pollutants accumulate near the ground due to the weak airflow flowing from the ground to the canyon top plane (Fig. 9e). For the α = 75° and 90°, the reversed-weak airflow entering the canyon through the windward VD creates a large clockwise vortex is formed near the windward wall, and pollutants are removed upward by this vortex (Figs. 9f and g). When α = 90°, the vortex is more developed compared to the case of α = 75°, but more pollutants accumulate near the leeward side because of the weak airflow entering through the windward VD.
For the canyons with both VDs (Fig. 10), pollutants remove from the canyon by the airflow directly across the VDs at both sides. Since the velocity of this airflow increases with increasing α, the pollutant patterns on the windward VD decrease. As α increases, a part of this airflow entering through the leeward VD is reflected by the windward wall and then flows towards the leeward wall again, finally merges with the free-stream. That is, the airflow draws an "S" shape on the vertical central plane of the canyon, forming two vortices inside the canyon. In addition, most of this airflow entering through the leeward VD flows to the windward VD. By this strong airflow, the pollutants are removed directly from the canyon (Fig. 10b-g). As α increases, the amount of residual pollutants at the windward VD decreases due to the increased velocity of this airflow.

Airflow and pollutant distribution at the pedestrian respiration planes
In this section, we evaluated the effects of the VD location and wind direction on the dispersion of traffic pollutants at the pedestrian respiration level. Figure 11 illustrates the variation of the mean K values on both pedestrian respiration planes of the canyons under different wind directions.
For the canyons with the leeward VD (Fig. 11), except for the case of α = 0° (98.2% of the regular canyon), the K values on the leeward respiration planes are almost zero.
The K values on the windward respiration planes have a maximum value at α = 15° (166.3%) and decrease with increasing α. When α = 15°, the K pattern are compressed to the windward wall by the airflow entering obliquely to the downwind building (Fig. A1b). Therefore, compared to the case of α = 0°, the K on the leeward respiration plane sharply decreases to almost zero, while K on the windward respiration plane becomes higher. As α increases, the airflow component ( ACH VDI ) entering through the leeward VD increases and fewer pollutants are removed through the canyon outlet, mostly upward (Figs. A1b-g and 8b-g).
For the canyons with the windward VD, as α increases, pollutants accumulate near the leeward wall by the airflow flowing toward the upwind building (Fig. A2). Therefore, the K values on the leeward respiration plane increase significantly with increasing α, while those on the windward respiration plane decrease (Fig. 11). When α = 60° (Fig. A2e), a slow flow zone forms in the central part of the canyon because the smallest airflow component (ACH VD2 ) enters through the windward VD (Table A2). Therefore, some pollutants accumulate in the middle of the windward wall, and the K value of the windward respiration plane increases again compared to the case of α = 45° (Fig. 11b).
For the canyons with both VDs, when α = 0°, the K pattern has a symmetric "cheliped" shape inside the canyon, and the K values on both respiration planes are equal (81.9% of the regular canyon) (Fig. A3a). As α increases, the airflow flows to the windward wall ( Fig. A3b-g), so the K value on the leeward respiration plane is almost zero (Fig. 11a). When α = 15°, more pollutants accumulate near the windward wall than in the case of other wind directions because the airflow velocity inside the canyon is the slowest (Fig. A3b). Therefore, the highest K value (102.5% of the canyon) on the windward respiration plane are found at α = 15°. As α increases, the velocity of airflow flowing to the downwind building increases and the pollutants accumulated in the windward VD gradually decrease. Therefore, as α increases, the K value on the windward respiration plane decreases, and the K value at the pedestrian respiration height of the windward VD region also decreases (Fig. A4).

Pollutant distribution on the canyon walls
Finally, we analyzed the pollutant concentration distribution on the canyon walls to investigate the pollutant exposure risk to near-road residents. Figure 12 shows the variation of the mean K values with wind direction for both walls of the canyon.
For the canyons with the leeward VD (Fig. 12), the mean K value on the leeward wall is much lower than that on the windward wall. When α = 0°, pollutants move towards the canyon exit by the parallel wind ( Fig. A1a and Fig. A5a). In this case, the K on the leeward wall is the highest (71.1% of the regular canyon) because the airflow velocity entering through the leeward VD is the slowest compared to the case of other wind directions. For oblique wind directions, the pollutants move to the windward side by the airflow entering through the leeward VD, so the K values on the leeward wall decrease with increasing α. As α increases, the K value  Fig. 12 Mean K values at the canyon walls: a leeward wall; b windward wall on the windward wall increases and has a maximum value at α = 30° (678% of the regular canyon) (Fig. 12b). From α = 45°, the airflow velocity entering through the leeward VD increases considerably (Table A1), so more pollutants move upward the canyon along the windward wall, and the K value on the windward wall decreases again (see Fig. 8c-g, A1c-g). At α = 90°, the K patterns have symmetric distribution as the airflow enters the canyon perpendicularly through the leeward VD (see Fig. A1g).
In contrast, for the canyons with the windward VD, the mean K value on the leeward wall is much higher than that on the windward wall (Fig. 12). The case of α = 0° is the opposite of the case of the canyon with the leeward VD (Figs. A5a and A6a). For oblique wind conditions, pollutants move to the leeward side by the airflow entering the leeward wall obliquely (Fig. A2b-g). As α increases, the incidence angle of the airflow entering the leeward wall through the windward VD increases. Therefore, pollutants move toward the canyon entrance, and the K on the leeward wall increases significantly, reaching a maximum value at α = 90° (113%) (Fig. A6b-g). However, the K on the windward wall first decreases and then increases again, with a maximum value (167.7%) at α = 60°. This is because the airflow component (ACH VD2 ) entering through the windward VD is the lowest when α = 60°, so pollutants do not move easily to the leeward wall (Table A2).
Compared to the previous two cases, the mean K values on both walls of the canyons with both VDs are much lower and decreases with increasing α (Fig. 12). In addition, the K values on both walls are much lower than those of the regular canyon. As α increases, the airflow velocity entering through the leeward VD increases (Fig. 10b-g), so more pollutants are moved towards the windward VD, and finally the K values on both walls decrease. Therefore, at α = 0°, the K on the leeward wall has a maximum value (63.4% of the regular canyon). However, the windward wall has a maximum K value at α = 15° (102.1%). This is because when α = 15°, pollutants accumulate on the lower part of the windward wall by the weak airflow entering the windward wall through the leeward VD ( Fig. 10b and A3b). When α increases from 30 to 90°, more pollutants are removed from the canyon by the strong airflow passing through both VDs, so the K values on both walls decrease ( Fig. A7c-g).

Discussion
The purpose of this study is to predict the effect of wind direction on the traffic pollutant dispersion inside 3D street canyons with void decks, thereby providing urban designers with informed guidance for constructing street canyons with the highest ventilation capacity and lowest pollutant exposure risk levels.
This study found that the VD location and wind direction have significant effects on the ventilation capacity of the canyon. As can be seen in Fig. 7, the ACH values of the canyons with VDs were higher than that of the regular canyon. This result provides similar evidence to previous studies that introduction of VD design in the street canyons can improve the ventilation capacity of the canyons and greatly increase the dispersion rate of traffic pollutants (Huang et al. (2021); Chen and Mak (2021)). The ACH values of the canyons with leeward VD or both VDs first increase significantly with increasing α, then decrease slightly at α = 75°, but it has high ACH value at α = 90°. This result differs from the result of regular canyon with the lowest ventilation capacity at α = 90° (Soulhac and Salizzoni 2010;Huang et al. (2019);). The canyons with VDs have higher ventilation capacity than the regular canyon, mainly due to the strong free-stream enters through VD into the canyon. With increasing α, the mean airflow component entering through VD increases significantly, thus removing more pollutants from the canyon floor. This phenomenon is in good agreement with previous results that VD can increase the wind speed at the pedestrian level within street canyons (Weerasuriya et al. (2020); Huang et al. (2020)).
In addition, our findings can provide a scientific foundation for health and sustainable urban planning can construction in tropical cities by quantitatively assessing the exposure risk of pedestrians and near-road residents. Previous studies have shown that pollutant concentrations inside the street significantly dependent on the wind direction ; Jeanjean et al. (2015); Jon et al. (2022)). In general, wind directions can be clarified as northwest (NW), north (N), northeast (NE), east (E), southeast (SE), south (S), southwest (SW), and west (W), and vary with season, day, even hour. However, since each city has its own dominant wind direction, it is of great practical significance to study the effect of oblique wind direction. According to the results of our study, the canyon with VDs has the highest ACH when VDs are located at both buildings and the lowest ACH when VD is located at the downwind building for the same wind direction (Fig. 7). This result is supported by the results of Sin et al. (2022) under the perpendicular approaching wind condition. For the canyons with both VDs, the pollutant exposure risk for pedestrians and near-road residents was significantly reduced in all wind directions compared to the regular canyon (Figs. 11 and 12). Thus, the "both VDs" design can be positively supported in tropical cities where wind directions change frequently. For the canyon with the leeward VD, the pollutant exposure risk for pedestrians and residents near the leeward wall was completely eliminated (except for the case of α = 0°), but those near the windward wall will slightly were increased compared to the regular canyon (Figs. 11 and 12). In this case, the expected wind direction is 60°, representing the maximum ACH value (Fig. 7). For the canyon with the windward VD, the maximum ACH and the reduction in pollutant exposure level occurred at 30°. Therefore, the angle to the dominant wind direction can be set at 30° (Fig. 7). As mentioned above, the results of this study can guide urban planning to design street canyons with high ventilation capacity more scientifically according to the dominant wind direction of the city.

Limitations and future works
There are two main limitations in the study that can be improved with future work.
First, this research adopts the street canyons with ideal VD. In actual situation, buildings with VDs are lifted by pillars. However, previous studies have not considered the effects of pillars to simplify numerical simulations ; Chew and Norford (2019); Huang et al. (2021); Chen and Mak (2021)). Since the pillars of the VD increase the airflow resistance, it is expected that the airflow velocity entering through the VD in the actual situation will be lower than the case of this study. Therefore, the K may be slightly higher on the pedestrian respiration planes and canyon walls. The role of pillars has not been elucidated and may be discussed in future research.
Second, this paper investigates the effects of wind direction inside the symmetric street canyon with VDs. In addition to symmetric street canyons, there are many asymmetric street canyons in actual cities. Different building height ratios and building lengths can create completely different airflow patterns within the street canyons, which can affect the diffusion of pollutants Ming et al. 2018). Therefore, the effect of wind direction inside the VD street canyon with different building height ratio and building length can be considered in future works.

Conclusions
In this paper, the effects of VD location and wind direction on the airflow and pollutant dispersion inside the street canyon are discussed. For 3D canyons with VDs, the airflow entering through VDs the flow field significantly changes the flow fields and pollutant distribution inside the canyon. The analysis of CFD simulation results leads to the following key conclusions: (1) VDs allow more airflow into the canyon, greatly improving the ventilation capacity of the canyon. Therefore, the ACH values of the canyons with VDs are much higher than that of the regular canyon, especially at high α. Regardless of the VD location, the canyons with both VDs have the highest ACH, followed by canyons with the leeward VD.
In addition, regardless of the VD location, the ACH of the canyons with VDs first increases and then slightly decrease with the increase of α.
(2) For the canyons with VDs, pollutants are removed from the canyon by airflow entering or leaving the VD. Therefore, building wall and the pedestrian respiration plane on the side where VD is located have much lower K values than that of the regular canyon. Thus, for the canyons with both VDs, the K values on both pedestrian respiration planes and walls are lowest. With the increase of α, the K values decrease on both respiration planes, except for the leeward respiration plane of the canyon with the windward VD.
To sum up, VDs provide better ventilation capacity than regular canyon under the different wind conditions, and greatly reduce the pollutant exposure level of pedestrians and near-road residents. These findings are useful for considering the traffic pollution damage according to the wind direction when designing street canyons with VDs in tropical cities with frequently changing wind directions.