The Effect of Vegetation Enhancement on Particulate Pollution Reduction: CFD Simulations in an Urban Park

Abstract

Vegetation in parks is regarded as a valuable way to reduce particulate pollution in urban environments but there is little quantitative information on its effectiveness. The aim of this study was to conduct on-site measurements and computational fluid dynamic (CFD) simulations to determine the aerodynamic and deposition effects of vegetation enhancement on particulate matter (PM) dispersions in an urban park in Xi’an, China. Initially, the airflow characteristics and deposition effects of vegetation were predicted and compared with measured air velocities and particulate pollution data to validate the numerical modeling. Then, associated coverage ratios and supplementary green areas (tree coverage ratio, crown volume coverage (CVC), and roof greening) were added to numerical simulations. After a series of numerical simulations and comparisons, results indicated that: (1) Numerical models with simplified vegetation method could reproduce the distribution of particulate matter concentrations in the real park environment; (2) with a tree coverage ratio >37.8% (or CVC > 1.8 m³/m²), the pedestrian-level PM_2.5 could meet the World Health Organization’s air quality guidelines (IT-1) standards in the park; (3) roof greening on leeward buildings produced greater PM removal effects compared with windward buildings; and (4) the most economical and reasonable tree coverage ratio and CVC to reduce atmospheric PM in urban parks should be 30% and 1.8 m³/m², respectively. These results are useful guidelines for urban planners towards a sustainable design of vegetation in urban parks.

1. Introduction

Atmospheric aerosol particulate matter (PM) refers to a variety of suspended solid and liquid pollutants. Inhalable particles, with an aerodynamics diameter smaller than 10 μm and 2.5 μm (PM₁₀ and PM_2.5), not only reduce human working efficiency and happiness [1,2,3], but inhalable particles can cause a range of respiratory and cardiovascular diseases and increase the incidence of malignant tumors [4,5,6,7]. Studies in Europe, the United States, and many Asian cities have shown that increased, and longer exposure to, particulate concentrations increase the risk of morbidity and mortality [8,9,10]. As a result, air pollution caused by PM has attracted much public and scientific attention.

Urban green infrastructure can relieve heat island effects and play an important role in improving the urban air quality [11,12,13,14]. Trees in green infrastructure can effectively reduce PM in urban environments [15,16,17,18,19]. Tree species, canopy size, canopy porosity, leaf area density (LAD), and tree arrangements around buildings can all affect the particulate diffusion [20,21]. Recently, numerous studies have discussed the deposition velocity or particulate capture efficiency of various tree species through wind tunnel experiments or field studies [22,23,24,25,26,27]. Methods include mass subtraction [28,29], membrane filters [30,31,32,33,34], and elution weighing coupled with particle size analysis by electron microscopy [35,36,37,38,39,40], along with other direct measuring methods with aerosol instruments [41,42,43]. These studies largely analyze the deposition rate of PM on leaves [44,45], as well as changes in retention duration of different PM sizes on leaf surfaces [46]. Numerical simulations generally quantitatively analyze aerodynamic and deposition effects of vegetation on PM dispersion in built environments [47,48,49,50,51,52,53,54,55,56], or investigate the influences of outdoor PM diffusion into indoor spaces by constructing three dimensional models [57,58]. Additionally, related research has been conducted on the deposition effects of green roofs on particulate matters [59]. Vegetation on roofs could reduce more pollution rather than crops grown near roads [60].

As a component of urban green infrastructure, parks play an important role in reducing PM in urban air [61,62]. Research demonstrates that the average pollutant concentration in parks is lower closer to the center of treed areas [63]. Yin et al. (2011) analyzed pollutant concentration changes in a Shanghai park and found that the crown volume coverage (CVC) increased from 0 to 2 m³/m², the TSP, SO₂, and NO₂ removal rates increased by 30%, 15%, and 10%, respectively [61]. Research elsewhere indicates that tree coverage ratio in urban parks is a major factor affecting PM retention [64].

Abundant field tests and simulation studies demonstrate that trees can effectively reduce atmospheric PM. However, studies have principally focused on effects of vegetation structure or tree species [24,27,29]. Due to instrumental constraint, results from field experimental studies can record parameters in only a limited time and space without fully reflecting the dynamic meteorological and environmental conditions, and using more stations would have been costly. Therefore, they do not provide generalize guidance on urban park design. With the aid of simulation based on computational fluid dynamics (CFD), the simplified models simulating the particulate matter dispersion in a block or urban scale has been significantly advanced [65]. Numerical simulations can solve the problem of environmental control, save time and labor, and are not limited by time and space. Using controlled variables in a street or urban scale simulation, the effect of trees on atmospheric particles dispersion can be directly calculated using the Reynolds-averaged Navier–Stokes (RANS) model combined with the generalized drift flux model [20,58,63].

In this study, we used CFD based models, including a standard k-ε model based on the RANS approach and a revised generalized drift flux model, to quantitatively investigate: (1) the accuracy of numerical models using simplified vegetation method; and (2) effects of changing supplementary green areas and associated coverage ratios (tree coverage ratio, CVC, and roof greening) on PM_2.5 and PM₁₀ reduction. Our results could provide optimal methods and quantitative indices for tree design in urban parks for the goal of environmental improvement.

2. Methods

2.1. Study Site

The study park is located in Xi’an, China (34°15′56.2″ N, 108°4′21.5″ E). The park’s footprint is 13.3 ha, and the building area is approximately 16,000 m². The park is a natural garden style with four main buildings (Bldg.1—China agricultural history museum, Bldg.2—Botanical museum, Bldg.3—Animal museum, and Bldg.4—Insect museum). The landscape elements in the park are as follows: buildings (accounting for 10 percent of the total area), plants (accounting for 75 percent of the total area), paving and square (accounting for 10 percent of the total area), water-filled pool (accounting for 5 percent of the total area). The north and east sides of the park are adjacent to a main thoroughfare. The west and south sides are adjacent to a university campus (Figure 1).

2.2. On-Site Measurement

Twelve monitored points were fixed in the park to record wind speed and direction (Kestrel 5500 wind speed meter, Nielsen-Kellerman Inc., Boothwyn, PA, USA) as well as PM concentrations (Aerocet 531S, Met One Inc., Grants Pass, OR, USA) at 1.5 m height. Monitored point A is fixed on the roof of the Bldg.3 to record incoming wind speed, wind direction and PM (PM₁₀ and PM_2.5) concentrations (Figure 2). The experiment was carried out from 30 October to 1 November 2018, from 9:00 h to 17:00 h every day. All measured points covered various environmental spaces in the park. Wind speed and direction were recorded once per minute and particulate concentration was recorded every 20 minutes. It was sunny without significant pollution source changes during on-site measurements and there were no other human factors close to monitored points that would cause concentration changes.

During the experiment, prevailing wind direction at monitored point A was easterly, and maximum wind velocity reached 2.8 m/s (Figure 3a). Particle concentrations for the same period in the three-day experiment were averaged hourly. The mean hourly PM₁₀ ranged between 297.3 and 427.2 μg/m³ and the PM_2.5 ranged between 95.6 and 157.6 μg/m³ (Figure 3b).

2.3. Computational Approach

2.3.1. Model Setup

The simulation model was constructed by using the vegetation, buildings, water-filled pool and type of underlying surfaces parameters based on the present condition of the park (Figure 4). According to the established guidelines [66], the distance between the inlet (inflow boundary) and the target area was set to W (W is the width of the target site) and the distance from the target area to the outlet (outlet boundary) was set to 3W. The distance between the left/right symmetric boundary and the target area was W. The distance between top and the ground boundary was set to 11H (H is the height of buildings in the target area, H = 15 m). The computational domain, with a dimension of 1290 m × 675 m × 165 m, was divided into three types of structural hexahedral meshes (coarse meshes: X_min = Y_min = Z_min = 0.27H; fine meshes: X_min = Y_min = Z_min = 0.13H; and finest meshes: X_min = Y_min = Z_min = 0.06H).

The outlet boundary condition was established with fixed pressure and zero gradients. Rough wall functions were fixed for the ground boundary. Corresponding constant horizontal velocity and turbulent kinetic energy of the inflow profile were fixed at the top boundary, while the left and right symmetric boundary was modeled as a slippage wall without a gradient.

The oncoming wind at the inlet is a gradient and is calculated using the equation:

$$u\left(z\right)=u_0{\left(z/z_0\right)}^{\mathsf{\alpha }}.$$

(1)

where u(z) is horizontal velocity at height z, and u₀ is the horizontal velocity at height z₀. In this model, u₀ = 2.8 m/s, z₀ = 16.5 m, and α = 0.25 [67].

The turbulent kinetic energy, k (m²/s²), and its dissipation rate, ε (m²/s³), are set as:

$$k=\frac{u_{*}{}^2}{\sqrt{C_{\mu }}}\left(1-\frac{z}{\delta }\right)$$

(2)

$$\epsilon =\frac{u_{*}{}^3}{Kz}\left(1-\frac{z}{\delta }\right)$$

(3)

where u_* is the friction velocity, δ is the depth of the boundary layer, and K is the von Karman’s constant. In this model, u_* = 0.52 m/s, K = 0.4, and C_μ = 0.09 [68].

Statistical data show that the predominant PM pollution sources in Xi’an in 2017 were mainly derived from atmospheric transportation [69]. To better replicate conditions during the on-site experiment, pollution sources were added to the inlet boundary in simulations. The average concentrations at monitored point A, which were added to the inlet boundary, were assumed to be constant. Particles in this domain were regarded as a continuum, and particle dispersion was assumed to have no effect on turbulence.

Trees in the model were parameterized as a one-dimensional column with a normalized LAD scaled to tree height [70]. LAD varied as a function of height modified by crown shape, height, and canopy edges. Different types of vegetation can be easily distinguished using non-uniform vertical distribution of LAD [71]. In this park, there are about 30 types of trees. To simplify computational simulations, six dominant species that account for more than 80% of vegetation were selected. Statistics on plant quantity, canopy diameter and height, and crown base height (m) of these trees were measured (

Table 1. Typical vegetation parameters in the park.

Typical Plants	Crown Volume Expressions	X (m)	Y (m)	CBH (m)	LAD (m2/m3)
Pinus tabuliformis Carr.	πx2y/12	3.5	4	1.8	0.40
Platycladusorientalis (L.) Franco	πx2y/12	3.0	6.5	1.0	2.20
Ginkgo biloba L.	πx2y/12	2.6	9	2.0	1.26
Acer truncatum Bunge	πx2y/6	4.2	3.8	1.5	1.54
Populus tomentosa Carr.	πx2y/4	4.0	9.8	2.0	1.38
Sophora japonica L.	πx2y/6	3.5	6.0	2.0	2.06

Due to the age similarity among different tree categories in the park, mean canopy height, width and crown base height of plants is used. X—crown diameter, Y—crown height, CBH—crown base height.

). Leaf area index (LAI) of the major trees was measured using a LAI-2200C Plant Canopy Analyzer (LI-COR Inc., Lincoln, Nebraska) (m²/m²) with the corresponding LAD was calculated using Equation (7). All typical trees were parameterized with their calculated LAD profiles based on tree height, crown diameter, and crown base height. These data were then written into user-defined functions to characterize the three-dimensional canopy of trees in CFD simulations.

The deposition velocities of PM_2.5 and PM₁₀ on the foliage of typical trees, grass and a water-filled pool in the park are listed in

Table 2. The deposition velocities of PM_2.5 and PM₁₀ on the foliage of typical trees, grass, and water-filled pools.

Particle Size Fraction	Deposition Velocity (Vd)(m/s)
	P. tabuliformis	P. orientalis	G. biloba	A. truncatum	P. tomentosa	S. japonica	Grass	Pool
PM10	0.0279	0.0343	0.0371	0.0364	0.0057	0.0325	0.0064	0.0047
PM2.5	0.0175	0.0458	0.0245	0.0922	0.0081	0.0454	0.0028	0.0001

[24,38,49,52,72]. Previous research has found that particle deposition on building envelopes was less than 0.03%, which we considered to be negligible for this study [52].

2.3.2. Simulation Description

We reproduced airflow and particle diffusion using three-dimensional steady-state isothermal flow field models. CFD simulations were aligned with COST Action 732 parameters [73]. The RANS approach model, consisting of the k-ε Murakami-Mochida-Kondo (MMK) closure scheme, was used. This model was modified based on the standard k-ε model, which accurately reproduces airflow fields around buildings. The Semi-Implicit Method for Pressure-Linked Equations (SIMPLE) algorithm with the Quadratic Upstream Interpolation for Convective Kinematics (QUICK) discretization scheme was applied to all governing equations. The scaled iterative convergence criteria for all parameters in simulations were set to 10⁻⁶. Simulations were run on an i7 2.67 GHz processor (Intel, Santa Clara, CA, USA). The Parabolic Hyperbolic or Elliptic Numerical Integration Code Series (PHOENICS; CHAM, London, UK, Edition: 2009) program was used to compute solutions.

In turbulence models, tree crowns are considered as a porous medium with individual tree branches are similar to the crown [74]. The drag and pressure produced by the tree crown decrease kinematic airflow energy. Therefore, resistance based on the momentum equation is considered to simulate vegetation influences on turbulence flow fields. The sink term is introduced to the momentum equation to express turbulence resistance by canopy layer:

$$S_{d,i}=-C_d\times LAD\times \left|U\right|\times u_i$$

(4)

where $C_d$ is the drag coefficient, |U| is the vector speed on foliage surface (m/s), and $u_i$. is the Cartesian velocity in i direction (m/s).

The LAD of vegetation is expressed by LAI, and defined as:

$$LAI=\underset{0}{\overset{h}{\int }}LAD·dz$$

(5)

$$LAD={\alpha }_m{\left(\frac{h-z_m}{h-z}\right)}^{n}exp\left[n\left(1-\frac{h-z_m}{h-z}\right)\right]$$

(6)

where h is the average canopy height. When 0 ≤ Z ≤ Z_m, n = 6 and Z_m ≤ Z ≤ h, n = 0.5. α_m is the maximum value of α at the perpendicular position Z_m. For computational efficiency, LAD is considered constant in the perpendicular direction and it can be calculated with canopy height and LAI:

$$LAD=LAI/h$$

(7)

The turbulence interaction between airflow and tree canopy can be expressed by additional source terms in momentum equation:

$$S_k=C_d\times LAD\times \left({\beta }_p{\left|U\right|}^3-{\beta }_d\left|U\right|k\right)$$

(8)

$$S_{\epsilon }=C_d\times LAD\times \left(C_{4\epsilon }{\beta }_p{\left|U\right|}^3\frac{\epsilon }{k}-C_{5\epsilon }{\beta }_d\left|U\right|\epsilon \right)$$

(9)

where β_p, β_d, C_4ε, and C_5ε are empirical constants, β_p denotes the mean fluid kinetic energy of wake flow K which is produced by drag force of canopy, and β_d represents the K kinetic energy that is dissipated by the short circuits of Kolmogorov energy gradients. In the present study, β_p and β_d as well as the closure constants C_4ε and C_5ε are 1.0, 3.0, 1.5, and 1.5, respectively [75,76,77].

The revised generalized drift flux model considers the slippage between particulates and fluid (air) phase. It is a corrected Eulerian model that regards particles as a continuum when solving the conservation equation of PM mass/quantity concentration, and is widely applied due to its simulation accuracy and efficiency. In the revised generalized drift flux model, three dimensional tree models enhance PM deposition through turbulence diffusion. Tree branches and leaves also absorb PM, and some PM might be suspended on leaves or washed away [22]. The aerodynamic and deposition effects of plants on PM are expressed by additional source terms (S_sink and S_resuspension). In this way, the revised generalized drift flux model can comprehensively and accurately describe and simulate plants’ influence on PM dispersion in real environments [20]. This model can be expressed as:

$$\frac{\partial \left[\left(V_j+V_{slip,j}\right)C\right]}{\partial x_j}=\frac{\partial }{\partial x_j}\left[{\epsilon }_p\frac{\partial C}{\partial x_j}\right]+S_c-S_{sink}+S_{resuspension}$$

(10)

Particle slippage velocity (V_slip) is defined by gravity, thermal force by the thermophoresis effect, particle fluctuation defined by turbulence and particle acceleration [78], and is calculated as:

$$V_{slip,j}={\tau }_p{g}_j+{\tau }_p{\displaystyle \sum }{F}_j+\frac{{\tau }_p}{C}{S}_{mj}-\frac{{\tau }_p}{C}\frac{\partial \left(V_{Pj}{V}_{pi}C\right)}{\partial x_i}$$

(11)

$$S_{mj}=\frac{\partial }{\partial x_i}\left[{\epsilon }_{p}C\left(\frac{\partial V_{pj}}{\partial x_i}+\frac{\partial V_{pi}}{\partial x_j}\right)\right]+\left[\frac{\partial }{\partial x_i}{\epsilon }_p\left(V_{pi}\frac{\partial C}{\partial x_j}+V_{pj}\frac{\partial C}{\partial x_i}\right)\right]$$

(12)

$${\tau }_p=\frac{C_c{\rho }_p{d}_p{}^2}{18\mu }$$

(13)

where V_j and V_slip,j are mean fluid (air) velocity and gravitational settling velocity of particles in direction j (m/s). C is particle concentration at the inlet (μg/m³). ε_p is turbulent diffusivity (m²/s), and that can be simplified to 1.0 [78]. S_c is the formation rate of particle sources (kg/m³s). S_sink is the mass of particle absorbed by vegetation per cubic meter within a unit of time (μg/m³). S_resuspension is the secondary pollutant generated by foliage per cubic meter within a unit of time [79]. V_pj and V_pi are particle velocities in j and i directions (m/s), respectively. τ_p is the particle relaxation time. g_j is the gravitational acceleration in j direction (m/s²). ΣF_j is the resultant force exerted upon the particle (m/s²). S_mj is the momentum source of particle in j direction kg/(m² s²). μ is molecular kinematic viscosity of air(Ns/m²). ρ_p is density of atmospheric particles (kg/m³). d_p is particle diameter (m). C_c is the Cunningham factor induced by slippage.

The effect of tree canopies on absorbing atmospheric particles is dependent on their LADs, deposition velocities and particulate concentrations in atmosphere, expressed as:

S_sink = LAD × V_d × C

(14)

The resuspension of particulate matter, as a term of volume source, is described as:

S_resuspension = S_sink × P_resuspension

(15)

P_resuspension = −0.00041v² + 0.017v − 0.0016

(16)

where V_d is particle deposition velocity on foliage (m/s); P_resuspension is the percentage of resuspended particles [79]; v is the magnitude of air velocity (m/s).

2.4. Case Description

To understand the influence of vegetation coverage on particulate concentration distributions in the park, the total vegetation coverage ratio was set to 70% (including tree coverage ratio and tree clearing ratio). This is the standard for vegetation coverage ratio in parks [80]. Tree coverage ratio, CVC, and roof greening were chosen as three variables, and their influences on particulate concentration distributions in the park were analyzed. The CVC is expressed as [61]:

$$CVC\left({\mathrm{m}}^3/{\mathrm{m}}^2\right)=\frac{\mathrm{Total} \mathrm{crown} \mathrm{volume} \mathrm{in} \mathrm{measurement} \mathrm{site}\left({\mathrm{m}}^3\right)}{\mathrm{Area} \mathrm{of} \mathrm{measurement} \mathrm{site} \left({\mathrm{m}}^2\right)}$$

(17)

Twelve scenarios were examined to meet simulation targets. Scenario 0 was the present condition of the park. Scenario 1 was the control group. Scenarios 3, 5, and 7 had different tree coverage ratios. Scenarios 2, 4, 6, and 8 had various CVCs by changing different types of tree canopy heights. Scenarios 9 and 10 had different pool coverage. Scenarios 10 and 11 were set with and without greening roof (

Table 3. Case settings in numerical simulations.

Scenario	Area Coverage Proportion (%)						CVC (m3/m2)
	Tree Coverage	Tree Clearing	Pool Coverage	Hard Paving	Building Coverage	Total
0 (Status quo)	25	50	5	10	10	100	1.15
1	-	-	-	90	10	100	-
2	5	65	-	20	10	100	0.1
3	60	10	-	20	10	100	2.4
4	60	10	-	20	10	100	3.6
5	30	40	-	20	10	100	1.2
6	30	40	-	20	10	100	1.8
7	15	55	-	20	10	100	0.6
8	15	55	-	20	10	100	0.9
9	30	40	10	10	10	100	1.2
10	30	40	5	15	10	100	1.2
11	30	40	5	15	10 (Green roofs)	100	1.2

).

3. Results and Discussion

3.1. Results of Model Evaluation

The accuracy of the airflow field and particle dispersion from the CFD simulation must be validated before the correct estimate the deposition effect of trees on particulate concentrations could be calculated. Simulations were based on the present conditions of the park. This enables the actual environment of the park (scenario 0) to be reproduced overlain by three grids (coarse meshes: X_min = Y_min = Z_min = 0.27H, fine meshes: X_min = Y_min = Z_min = 0.13H, and finest meshes: X_min = Y_min = Z_min = 0.06H). Specific-hour concentrations (PM_2.5 = 132 and PM₁₀ = 358 μg/m³), wind velocity (2.1 m/s) and wind direction (easterly) recorded at 14:30 h on 1 November 2018 were added in the inlet boundary. Three grid densities perform similar changing tendency of particle concentration and wind speed along the middle line (Figure 5). Grid independence was evaluated using the grid convergence index [81]. The grid convergence index between coarse and fine meshes was 4.61%, and was 3.87% between fine and the finest meshes that all meet the calculation requirements (<5%) [82]. Since an increase in the total grids could take more computational time, the above results indicate that the satisfactory grid independence may be archived by using fine meshes. Computations require only 49 hours and the accuracy is higher than with coarse meshes. Therefore, fine meshes were chosen in subsequent simulations of other scenarios (1–11).

Figure 6 depicts a comparison of the CFD predictions with measured PM₁₀, PM_2.5, and wind velocity at different monitoring locations on 1 November. Each data point denotes the specific-time value recorded at 14:30 h. The standard deviations of PM₁₀, PM_2.5, and wind velocity were 70.9 μg/m³, 30.9 μg/m³, and 0.49 m/s. Overall, the linear regression of PM₁₀ and PM_2.5 between CFD predictions and measurements all showed similar tendencies with a slope close to 1 and an R² greater than 0.94. The PM₁₀ concentration difference between simulation and measurement was less than 20%. Differences of PM_2.5 concentration were less than 10% except that at point D. Point D was close to edge of the park and adjacent to roads where PM_2.5 might be higher due to vehicle emission particulates, although we scheduled field experiments for times when traffic volume were less. It is noteworthy that the correlation between simulated and measured PM_2.5 is stronger than that of PM₁₀. In the simulation process, we hypothesized that particulates diffuse with airflow, and the mutual frictional force and drag among particulates are negligible. The friction and resistance between the larger size particles are greater, which makes the difference of PM₁₀ slightly larger than PM_2.5. The predicted wind velocities in windward regions (points G and E) agreed well with the measured data with discrepancies less than 10%. While velocities in leeward regions (points H and J) generally differed up to approximately 20% between the simulated and measured values due to blockage from buildings and trees. It also can be noticed that the simulated wind velocity at the monitored point of C deviated from that of measured data by 30.7%, owing to the CFD model used in our study was a RANS model including the k-ε MMK turbulence closure scheme, which was modified form the standard k-ε model and had better applicability to simulate airflow around buildings. Although previous studies indicated that Reynolds Stress Model (RSM) and Large Eddy Simulation (LES) reproduced better aerodynamic effect of plants compared with k-ε model [83], regarding validation of deposition model, we only considered the work in which k-ε model was employed [84]. Thus, the monitored points surrounded by dense trees or far away from buildings exerted relatively large variations between simulated and measured data.

3.2. Tree Coverage Ratio (%)

The concentration in the windward areas of the park was significantly higher than that in the leeward. With the increase of tree coverage ratio, PM₁₀ (PM_2.5) decreased significantly, and the concentration in the downwind areas was more variable than that in upwind areas. When the tree coverage ratios were rising from 15% (Scenario 7), 30% (Scenario 5), to 60% (Scenario 3), the concentration in the leeward areas of buildings decreased significantly, while the difference of concentration on windward areas of buildings was small (Figure 7). Due to building obstruction, PM₁₀ and PM_2.5 on leeward areas of buildings at the pedestrian level were small, but concentrations on the windward areas of buildings were relatively high, agreeing well with the results of Ji and Zhao (2014) [20]. In areas where trees were planted, the pedestrian-level concentration was clearly lower. Minimum concentrations in the park all occurred in the large-scaled tree planting areas. Concentrations close to the tree canopy were relatively low and decreased the concentration significantly at the pedestrian level. This pattern matches previous results [85,86], however, some studies demonstrated that concentrations at the tree canopy level were higher in the street canyons [50,53,87]. This appears to be caused by different pollution sources, for instance internal (e.g. traffic-emitted), rather than external sources transmitted from atmosphere as in our study. The reduction in ventilation was also responsible for the buildup of pollutants in street canyons. Additionally, these studies largely considered the aerodynamic effect of trees on pollutants, and the absorption effect was not fully taken into account. A previous study demonstrated that when the average wind speed was 5 m/s, trees could decrease PM_2.5 concentration by 4.6% through deposition and only 0.7% through aerodynamic effects [88].

To compare our results against the World Health Organization air quality guidelines (WHO AQGs), our study chose interim target-1 (IT-1) for PM_2.5 and PM₁₀ with 24 hour average concentrations of 75 and 150 μg/m³ as a criterion [89]. An increase in tree coverage ratio from 15% to 30% resulted in area proportion with pedestrian-level PM_2.5 less than 75 μg/m³ increasing from 26% to 50%, and area proportion with pedestrian-level PM₁₀ less than 150 μg/m³ increasing from 31% to 50%. However, this trend was not obvious when the tree coverage ratio increased from 30% to 60%. The area proportion with pedestrian-level PM_2.5 less than 75 μg/m³ only increased from 50% to 53%, and the area with pedestrian-level PM₁₀ less than 150 μg/m³ only increased by 2% (

Table 4. The area proportion with pedestrian-level PM₁₀ and PM_2.5 met the WHO criterion when the tree coverage ratios are 15%, 30%, and 60%.

Criterion	Tree Coverage Ratio (%)
	15 (Scenario 7)	30 (Scenario 5)	60 (Scenario 3)
PM10 ≤ 150 μg/m3
PM2.5 ≤ 75 μg/m3

).

To quantitatively investigate the effect of tree coverage ratio on particulate concentrations in the pedestrian space, the average concentration (1.5 m height) as a function of tree coverage ratio is displayed in Figure 8. To generalize the results, the concentrations at different tree coverage ratio were normalized by the concentrations at the inlet. The fitting curve of PM₁₀ was lower than that of PM_2.5. This is mostly due to turbulent diffusion and surface deposition reduction intensified by increasing tree coverage ratio, and larger size particles were easier to be deposited on leaf surfaces. At tree coverage ratio = 10%, the PM_2.5 was approximately 80% of the inlet concentrations, while PM₁₀ was 72% of the inlet value. At tree coverage ratio = 30%, the concentrations declined to approximately 62% (PM_2.5) and 50% (PM₁₀) of the inlet values. When the tree coverage reached 60%, the concentrations declined to approximately 50% (PM_2.5) and 40% (PM₁₀) of the inlet values. By substituting y = 75 μg/m³ in equation y = [1.177 − 0.17ln(x)] × C_in. It can be seen that when tree coverage ratio >38.1%, the pedestrian-level PM_2.5 could conform to the WHO AQGs (IT-1) standards.

3.3. Crown Volume Coverage (CVC)

PM₁₀ and PM_2.5 under CVCs = 1.2 (Scenario 5) and 1.8 m³/m² (Scenario 6) were compared when the tree coverage ratio was 30% (Figure 9). The concentration in the space between two windward buildings (Bldgs.3 and 4) was relatively high. The concentrations decreased sharply when they flow with air and diffuse to areas where trees were planted. Under the two CVC conditions, the concentration on windward surface of buildings was relatively high, but the concentration on tree planted regions and leeward surface of buildings was relatively low. When CVC increased from 1.2 to 1.8 m³/m², the PM₁₀ and PM_2.5 decreased significantly. The WHO interim target-1 (IT-1) for PM_2.5 and PM₁₀ with 24 h average concentrations was also selected as a criterion. When the CVC increased from 1.2 to 1.8 m³/m², the area proportion with pedestrian-level PM_2.5 less than 75 μg/m³ increased by 2% (from 48% to 50%), while the area with pedestrian-level PM₁₀ less than 150 μg/m³ increased from 47% to 51% (

Table 5. The area proportion with pedestrian-level PM₁₀ and PM_2.5 met the WHO criterion when the CVCs are 1.2 and 1.8 m³/m².

Criterion	CVC (m3/m2)
	1.2 (Scenario 5)	1.8 (Scenario 6)
PM10 ≤ 150 μg/m3
PM2.5 ≤ 75 μg/m3

).

The results and curve fitting between the average concentration (1.5 m height) and CVC are shown in Figure 10. The correlation between PM_2.5 and CVC was stronger than that of PM₁₀. The average concentration was found to decline exponentially with increasing CVC. This is consistent with the findings of Wu et al. (2018) [90]. The difference in concentration between PM_2.5 and PM₁₀ was initially small at CVC = 0.2 m³/m² and gradually increased towards CVC = 3.5 m³/m². This is a result of increased surface deposition mostly due to turbulent diffusion with increasing CVC. At CVC = 0.5 m³/m², the PM_2.5 was approximately 75% of the inlet concentration, while the concentration of PM₁₀ was 70% of the inlet value. Concentrations are significantly reduced with the increase of CVC from 0 to 2.0 m³/m². A break-even point was observed at CVC = 2.6 m³/m² where the particle concentration declined to the same level (50% of the inlet concentration). By substituting y = 75 μg/m³ in equation y = [0.66 − 0.17ln(x)] × C_in. The WHO AQGs (IT-1) specify that the CVC should be greater than or equal to 1.83 m³/m².

3.4. Greening Roof

Research demonstrates that indoor PM is significantly correlated with outdoors, and outdoor PMs mainly enter indoor spaces through transmission and penetration [91,92,93]. Therefore, the windward building (Bldg.4) and leeward building (Bldg.2) were chosen to analyze influences of roof greening on concentration close to building walls (0.5 m away from the building facade). Concentrations with and without roof greening were compared (Figure 11).

The concentration on windward facade of a building was significantly higher than that on the leeward facade. Concentrations gradually increased with height above the tree canopy, consistent with the results of Ji and Zhao (2014) [20]. It is clear in Figure 11a,b that the deposition effect of roof greening around the leeward façade (the difference between blue solid and dotted lines) was stronger than that of windward facade. The concentration on south facade of Bldg.4 (facade D) was higher than that on north one (facade A). Since the south wall was close to the ventilation corridor, particulate concentration in this region was relatively high. Concentration differences between the south and north facades of Bldg.2 were small. In Bldg.4, the concentration in the lower building space (height < 12 m) was higher with greening roof, but the concentration in the upper building space (height > 12 m) decreased significantly. PM₁₀ and PM_2.5 were respectively reduced by maximum values of 9.7 and 4.5 μg/m³. In Bldg.2, roof greening could decrease the concentration around each façade at different heights. Roof greening unsurprisingly appears to reduce roof level particulate concentrations more effectively. The maximum reductions of PM₁₀ and PM_2.5 were 40.4 and 13.8 μg/m³, respectively. Greening roof thus significantly reduced particulate concentration around facades of leeward buildings.

The temporal mean concentration at 1.5 m height above the tops of buildings with roof greening was lower than that without roof greening (Figure 12). The concentration with roof greening varied over a larger range, but was smaller than without greening. The median concentration with roof greening was lower than that without roof greening. We found that the difference between the median and the upper (lower) quartiles was smaller, indicating that particle concentration above roof was discrete with roof greening. Roof greening can further reduce particulate concentration at the top of leeward buildings (Bldg.2), and the average reductions of PM₁₀ was 7% and of PM_2.5 was 5%.

Particulate matter disperses within the airflow field. Roof greening above the windward building hinders airflow, causing PMs to increase in the lower layers of the windward building. Planting vegetation on the roof of the windward building, although it reduced the concentration near the upper level, is still not desirable. Thus, greening roof on leeward buildings, due to their relatively low wind speed, particles were easily deposited on leaf surfaces. Therefore, greening roofs should be used in an appropriate space, and greening roofs on leeward buildings are more effective and a better choice to reduce particles.

3.5. Removal Rate

Tree coverage and CVC are significant predictors of particulate concentration distributions. In this section, the influence of tree coverage and CVC on the average particulate concentration at the pedestrian level was investigated. The reduction rate of particulate concentration was calculated as:

$$\omega =(C_{ave,no greenery}-C_{ave})/C_{ave,no greenery}$$

(18)

where ω is the removal rate capability of greenery on PM₁₀ and PM_2.5. C_{ave,non-greenery} and C_ave are the average particulate concentrations at the pedestrian level with and without greenery.

Removal rates of pedestrian-level PM₁₀ and PM_2.5 are positively related to both tree coverage ratio and CVC (R² > 0.90) (Figure 13). An increase in tree coverage ratio from 0 to 30%, reduced concentration sharply. An increase in CVC from 0 to 1.8 m³/m², greatly decreased particulate concentration. This pattern agrees well with an experiment conducted in a Shanghai park [61]. Further increasing of tree coverage and CVC do not appear to have significant mitigating effect on particles. When trees are very abundant, there is airflow resistance and this obstruction of atmospheric circulation increases the local particles. So that removal rate stabilizes when CVC > 1.8 and tree coverage ratio >30%. From these results, we can concluded that number of trees planted in a park is not simply the more the better. The most economical and reasonable tree coverage ratio and CVC should be 30% and 1.8 m³/m².

In this study, the influence of vegetation enhancement on PM dispersion in an urban park was analyzed. Results demonstrate the utility of PM removal rate corresponding to tree coverage ratio and CVC as supplementary indices in urban park design. However, our investigations uncovered some limitations: (1) We only simulated airflows and PM dispersion with a single wind direction. However, real-world meteorological changes are complex and, thus, future studies should focus on multi-perspective simulation analysis with more complex meteorological parameters, i.e., different pollution levels, wind speed and direction, park spatial scale. (2) Pollution sources were assumed to dilute through the inlet boundary in this study, while internal pollution sources and regional traffic pollution were not taken into account. Our simulation hypothesized that particulate diffusion was continuous, without the influence of unsteady-state airflow changes. In practice however, airflow changes continuously and even steady-state airflow influences particulate diffusion. More experiments could verify these effects. (3) Actual green spaces in parks are generally arbor-shrub-grass structures. This study only considered the impact of trees. Future studies should include the comprehensive effects of more complex arbor-shrub-grass structure to improve the prediction accuracy for actual urban environments. (4) Vegetation growth is dynamic and the effects of vegetation on dust retention during different growth periods differed significantly in our simulations. Hence, our conclusions should be verified with field data. (5) Finally, effects of solar radiation, heat transmission among buildings, vegetation, water and underlying surfaces on particulate diffusion were not accounted for but should be considered to make simulations more realistic.

4. Conclusions

The objective of this study was to investigate the effects of supplementary green areas and associated coverage ratios on PM_2.5 and PM₁₀ concentration distributions in an urban park. Numerical simulations and field measurements were performed to examine the aerodynamic and deposition effects of trees on PM dispersion, which would provide architects and planners with a better understanding of the effect of park design on air quality by probing the impacts of changing CVC, tree coverage ratio, and supplementary green areas on atmospheric PM reduction. The primary results were summarized as follows:

(1)

Field data and simulations were strongly correlated in a simplification of the vegetation model (R² > 0.88), indicating that the model could reflect the real influence of trees on particulate concentration distributions in the park.

(2)

Assuming that pollution sources were diluted through the inlet boundary, the tree coverage ratio and CVC were the primary factors affecting PM dispersion. To ensure pedestrian-level PM_2.5 meets the standards of WHO AQGs (IT-1), tree coverage ratio should be greater than 37.8%, and CVC should be greater than 1.8 m³/m².

(3)

Roof greening can reduce particulate concentration at the top of the building. Particulate concentrations on vertical spaces and roofs of leeward buildings were lower than that of windward buildings. The average reduction ratio above the roof of a leeward building for PM₁₀ was 7% and for PM_2.5 was 5% comparing greening and non-greening roof.

(4)

When the tree coverage ratio increased from 0 to 30% and CVC increased from 0 to 1.8 m³/m², the PM reduction ratio increased significantly. These patterns remain stable as the tree coverage ratio and CVC continued to increase.

Our study shows that vegetation can absorb atmospheric PM directly in a tree canopy, and indirectly through ventilation. Therefore, urban park solutions are not simply the more the better, because trees can dampen ventilation. To improve particle reduction and save resources, the most economical and reasonable tree coverage ratio should be 30% and CVC should be 1.8 m³/m². When the roof is greened, vegetation should be planted on leeward buildings instead of windward buildings to remove more PM around building. These findings provide a useful strategy with the implementation of greening modifications in urban parks to mitigate atmospheric pollution.