Considerations for evaluating green infrastructure impacts in microscale and macroscale air pollution dispersion models

Abstract

Green infrastructure (GI) in urban areas has been acknowledged to reduce the pollutant concentrations. However, there are limited modelling options to evaluate its impact on air quality. The aim of this review is to address the following question: how can the GI be considered in readily available dispersion models to allow evaluation of their impacts on pollutant concentrations and health-risk assessment? We examine the published literature on the parameterisation of air pollutants deposition velocities and dataset that are required for deposition scheme. The deposition schemes treating the GI in detail are complex, resource intensive and involve an abundant volume of input data. Further, we evaluate the limitations of different air pollution dispersion models at two spatial scales-microscale (i.e. 10–500 m) and macroscale (i.e. 5–100 km)-in considering the GI effect on air pollutants concentration and exposure alteration. We conclude that appropriate handling of GI characteristics such as aerodynamic effect, deposition of air pollutants and surface roughness in dispersion models is necessary for understanding the mechanism of concentration simulation by GI at different spatial scales. Several other processes such as air pollutant transformation due to GI emitted biogenic volatile organic compounds (bVOC), the fraction of stomatal blocking water film and change in uptake rate due to temperature increase also influence deposition rates and change the air pollutants concentration; most of them are not treated in dispersion models, mainly because these are not developed for such purposes. Furthermore, the impact of GI on air pollutants concentration and health risk assessment (e.g., mortality, morbidity) are partly explored. The United States Department of Agriculture (Forest Service) developed i-Tree, which with BenMap developed by the US EPA, have used to estimate the health outcomes of annual air pollutants removal by deposition over GI canopy at the macroscale. However, studies relating the air pollution health risk assessment due to change in short-term exposure by pollutants concentration redistribution at the microscale and enhanced atmospheric pollutants dilution by increased surface roughness at macroscale because of GI, along with deposition, are rare. Suitable treatment of all physical and chemical processes in coupled dispersion-deposition model and assessment against real scenarios are vital for health risk assessment.

Graphical abstract

1. Introduction

Traffic emissions are a major source of air pollution in urban areas that adversely impact human health and the environment (Kumar et al., 2016, 2015, 2013; Sharma and Khare, 2001). The adverse health effects have been studies for regulated air pollutants such as particulate matter including PM₁₀ (≤10μm; de Hoogh et al., 2014; Li et al., 2018; Talbi et al., 2018), PM_2.5 (≤2.5μm; Ham et al., 2017; Huang et al., 2012; Rivas et al., 2017), Ozone (O₃; Abramesko and Tartakovsky, 2017; Bao et al., 2018; Goodman et al., 2018), nitrogen dioxides (NO_x; Jeanjean et al., 2017; Knibbs et al., 2018; Muttoo et al., 2018) and unregulated pollutants such as ultrafine particles (Clifford et al., 2018; Kumar et al., 2017; Pacitto et al., 2018), black carbon (Chang et al., 2015; De Matteis et al., 2015; Rivas et al., 2017), volatile organic compounds (VOCs; McDonald et al., 2018) and heavy metals (Aksu, 2015; Onder and Dursun, 2006). Green infrastructure (GI) is considered as one of the potential solutions for improving near-road air quality and limiting personal exposure to local sources (Abhijith et al., 2017). Usually, GI reduces the pollutant concentration by offering more surface area to increase the dry deposition, pollutants redistribution and atmospheric turbulence. GI could have different meanings under different contexts and depends on in which context it is used. For instance, eco-friendly constructions that have a low carbon footprint and energy efficiency are also known as green buildings/infrastructure in structural engineering. At the same time in Geo-environment engineering, GI is the group of trees and plants to stop the soil erosion. Benedict and McMahon (2002) defined GI “as an interconnected network of green space that conserves natural ecosystem values and functions and provides associated benefits to human populations”. In a broad sense, GI cover from all type of grasslands, forests, wetlands to an individual level species of tree and plant. For urban air quality, the important GI is street trees, green walls and roofs, roadside hedges, parks and grasslands to the urban boundary where air exchange is significant. In this review, we focus only on urban green vegetation that includes a group of trees, plants and/or hedges having large leaf area index (LAI) with an aim to draw critical discussion on how these could be considered in various readily available dispersion models for their impact assessment.

Apart from influencing factors such as wind speed and direction, topography and meteorology, the air pollutants dispersion also varies by surface roughness or geometry and deposition characteristic of GI. For example, a potential reduction of up to 50% compared with no green wall and roofs in street canyon can be expected by GI at microscale (i.e. 10–500 m; Britter and Hanna, 2003) in street canyon and neighbourhood environments (Gromke, 2011; Pugh et al., 2012) and that could be as little as 5% through available urban trees at macroscale (i.e. 5–100 km) citywide (Bottalico et al., 2017; Currie and Bass, 2008; Nowak, 1994). These reductions come from various reasons such as enhanced dilution due to enhanced atmospheric turbulence owing to increased surface roughness (Pleijel et al., 2004), air pollutants concentration redistribution (Abhijith et al., 2017; Baldauf, 2017) and dry deposition that is strongly influenced by LAI (Jayasooriya et al., 2017; Selmi et al., 2016). To reveal the dominant process in consideration of GI effect on air pollution would require a specified designed experiment, e.g. by measuring the air pollutant reduction at different distances from source in combination with different GI. The coupled green infrastructure and air pollutant dispersion modelling to estimate the dry deposition is based on consideration of GI as a porous medium in CFD (Jeanjean et al., 2015) and wind tunnel (Gromke et al., 2012) and as whole in receptor models (Chen et al., 2016; Li et al., 2016; Maleki et al., 2016), time of interaction (Pugh et al., 2012) and air pollutant concentration over the surface area (Jeanjean et al., 2016; Tiwary et al., 2009). At macroscale, the dilution of air pollutants due to GI-induced enhanced atmospheric turbulence has not been studied much, but many researcher (Barnes et al., 2014; Venkatram et al., 2013) have shown increasing surface roughness could decrease the ground level air pollutant concentration. As summarised in Table 1, many researchers have assessed the effect of GI on urban air quality at different spatial scales. Past works have also investigated the impact of GI on the reduction of particulate matter (Janhäll, 2015), urban heat island mitigation (Akbari et al., 2001; Bowler et al., 2010; Luber and McGeehin, 2008), reduction in noise pollution (Chiesura, 2004; Pathak et al., 2011), reduction in pollutant exposure as passive roadside barriers (Abhijith et al., 2017; Gallagher et al., 2015, 2013), stormwater management (Czemiel Berndtsson, 2010; Shaneyfelt et al., 2017) and considering GI as a part of natural capital (Chenoweth et al., 2018).

Table 1.

Summary of relevant studies considering GI modelling at the microscale (street and neighbourhood) and macroscale (city).

City (modelled area)	Pollutant concentration measurement techniques	Approach for GI consideration	Author (year)
Toronto (1216 ha)	Field measurement	UFORE (i-tree)	Currie and Bass (2008)
Beijing (30,121 ha)	Field measurement	UFORE (i-tree)	Yang et al.(2004)
Chicago (60300 ha)	Field measurement	Nowak method (i-tree)	Nowak (1994)
Sacramento (23600 ha)	Field measurement	Mcpherson method	Mcpherson and Simpson (1999)
Baltimore (21000 ha)	CMAQ version 4.7.1	i-tree, CMAQ + WRF	Cabaraban et al.(2013)
Leicester (400 ha)	OpenFOAM software (CFD)	OpenFOAM software (CFD)	Jeanjean et al.(2015)
Marylebone (72 ha)	OpenFOAM software (CFD)	OpenFOAM software (CFD)	Jeanjean et al.(2017)
Antwerp (32 and 64 ha)	OpenFOAM software (CFD)	OpenFOAM software (CFD)	Vranckx et al.(2015)
Lisbon (45.50 ha)	URVE code (CFD)	URVE code (CFD)	Amorim et al.(2013)
Aveiro (64 ha)	URVE code (CFD)	URVE code (CFD)	Amorim et al.(2013)
Leicester (400 ha)	OpenFOAM software (CFD)	OpenFOAM software (CFD)	Jeanjean et al (2016)
Shanghai (0.18 ha)	FLUENT (CFD)	FLUENT (CFD)	Li et al.(2016)
Bari (0.645 ha)	FLUENT (CFD)	FLUENT (CFD)	Buccolieri et al.(2011)
Santiago (96720 ha)	Field measurement	UFORE (i-tree)	Escobedo and Nowak (2009)
Shanghai (47100 ha)	Field measurement	Statistical analysis	Yin et al.(2011)
Florence (10200 ha)	Field measurement	iTree software	Bardelli et al (2011)
Syracuse (6500 ha)	Field measurement	iTree software	Nowak et al.(2013)
Auburn (306 ha)	Field measurement	iTree software	Martin et al.(2013)
West Midland (960000 ha) and Glasgow (300000)	FRAME model	Statistical analysis	McDonald et al. (2007)
Berlin (200 ha)	Field measurement	Land use regression model	Ghassoun et al.(2017)
Strasbourg (7830 ha)	Field measurement	iTree software	Selmi et al.(2016)
Chicago (58830 ha)	Monitoring stations	Nowak method (iTree)	Yang et al.(2008)
Chapel Hill, NC (16ha)	Field measurement	CTAG (CFD)	Steffens et al.(2012)
Chapel Hill, NC and generic near-road environments (16 ha)	Field measurement	CTAG (CFD)	Tong et al.(2016)
Woodside, CA	Field measurement	CTAG (CFD)	Deshmukh et al. (2018)

However, the uncertainty in GI modelling can lead to overestimation of total dry deposition that can result in under-predication of exposure assessment. For instance, use of constant surface resistance in dispersion model would overestimate dry deposition by approximately ten-times higher (Cape et al., 2008) and low-resolution concentration data in deposition model can lead to overestimation by a factor of two (Schrader et al., 2018). Therefore, it is essential to consolidate and synthesise previous investigations on the considerations of GI in micro and macro scale air pollutant dispersion model for: (i) researchers to identify gaps in understanding coupled modelling of air pollutant deposition and dispersion into atmospheric model, and (ii) decision-makers to assess the potential of GI to evaluate air quality benefits and incorporate accordingly in future urban planning. We summarise the potential consideration of GI at the microscale and macroscale air pollution modelling. These summery can assist in the formulation of coupled models to obtain a more realistic estimate of atmospheric chemical budgets.

Furthermore, the previous articles on GI-air pollutants interaction have discussed the methods to estimate dry deposition of air pollutants with numerical models (Wesely, 2000), application of atmospheric models for particles dispersion (Holmes and Morawska, 2006), assessment of deposition velocities of air pollutant to different species (Hirabayashi et al., 2011), methods and controlling factors for particulate matter dry deposition (Mohan, 2016), assessment of deposition and thermal effect of urban trees (Buccolieri et al., 2018), bi-directional air pollutant exchange between green infrastructure and atmosphere (Massad et al., 2010), detrimental effects of particulate matters on green infrastructure (Litschke and Kuttler, 2008) and estimation of deposition velocities based on lane use (Schrader and Brümmer, 2014). However, there is still a need to understand the consideration of GI to estimate the dry deposition in atmospheric dispersion models for both microscale and macroscale. Janhäll (2015) provided a thorough literature review on urban green infrastructure effects on particulate matter concentration at different spatial scales. Our review extends its scope by focusing on the importance of different resistance in the estimation of deposition velocity for both gaseous pollutants and particulate matter and their consideration in coupled dispersion-deposition modelling. In particular, we (i) carry out a detailed review of parameterisation for GI modelling to estimate deposition velocity; (ii) provide a comprehensive summary of design inputs (e.g., meteorological, GI and topographical parameters) to evaluate the different resistance and deposition velocity for their estimation by deposition models; (iii) evaluate the effectiveness of microscale and macroscale air pollution dispersion models to estimate the pollutant concentration reduction through GI; (iv) evaluate the parametric uncertainties in coupled GI and dispersion modelling; and (v) discuss a numerical framework for linking GI, air pollution and public health outcomes.

2. Scope and outline

The scope of this review is limited to GI that includes roadside vegetation (street canyons/open roadside trees, hedges and bushes), grasslands including parks and gardens and urban forest within the micro-and macro-scale simulations. In case of green walls and inclined roofs, the direction of deposition velocity is not usually perpendicular to implanted surfaces. The discussion on green walls, inclined roofs or any other artificial system as well as considerations of wind direction impacts is not considered in detail in this work. The review starts with critically synthesising GI modelling and relevant technical input parameters that are required for deposition modelling (Section 3). A state-of-the-art review of consideration of GI in air pollution dispersion models at microscale and microscale simulations to predict spatiotemporal concentrations are discussed in Sections 4 and 5, respectively. Section 6 describes the challenges in the simulation of real scenarios such as GI deposition velocity, GI spatial distribution, the influence of local meteorological parameters and pollutant transformation. Section 7 focuses on mathematical approaches that link GI, air pollution and public health outcomes, followed by a summary of topic areas covered, their conclusions and future outlook (Section 8).

3. Description of the GI modelling systems in the deposition scheme

Deposition scheme, which is a part of air transport models, uses mathematical equations for describing atmospheric, absorption and settling processes within mixing layer, to estimate the accumulated quantity of air pollutants over any solid surface area. The following specific conditions make GI considerations distinct in typical numerical simulations: (i) aerodynamic effect in the form of air pollutants concentration redistribution (Hefny et al., 2015); (ii) deposition of air pollutants (Nowak et al., 2018); and (iii) surface roughness to enhance atmospheric turbulence (Barnes et al., 2014). At the microscale, roadside GI such as hedges may act as a filtering barrier between air pollution sources and receptors and can reduce personal exposure. On another hand at the macroscale, GI such as urban forests, parks, gardens and hedges increase atmospheric dilution as well as act as a sink (deposition) for atmospheric pollutants. The harmful gaseous pollutants and airborne particles are deposited while passing over the surface of GI. The deposited amount (F; g m⁻²) to GI is proportional to the deposition velocity (V_d; m s⁻¹), time of exposure (t; sec) and pollutant concentration (C; g m⁻³) (Bottalico et al., 2016; Jeanjean et al., 2016; Tiwary et al., 2009).

$$F=V_d\cdot C\cdot t$$ (1)

V_d for different gaseous pollutant (Eq. 2) can be calculated as the inverse sum of aerodynamic resistance (R_a; s m⁻¹), quasi-laminar boundary layer resistance (R_b; s m⁻¹) and canopy resistance (R_c; s m⁻¹) (Janhäll, 2015; Tallis et al., 2011; Tiwary et al., 2009). For particulate matter, Giardina et al.(2018) have proposed a new deposition model based on electrical analogy scheme to estimate V_d by using both total resistance R_z (= R_a + R_bp) and settling velocity (V_s) (Eq. 3). The schematic diagram for gaseous pollutant collection, through dry deposition, as indicated in Fig 1.

Figure 1.

Schematic diagram showing the resistance relationship with Ohm’s law in electrical circuits where R_a is aerodynamic Resistance; R_b is quasi-laminar boundary layer Resistance; R_c is canopy/surface resistance; R_c1 is atmospheric buoyancy, lower canopy and in-canopy resistance; R_c2 is stomatal, mesophyll and cuticular resistance; and R_c3 is canopy soil, ground, water, snow or any other surface resistance; adopted from (Fowler, 1981).

$$V_d=\frac{1}{R_a+R_b+R_c}forgaseouspollutants$$ (2)

$$V_d=\frac{V_s}{1-e^{-\left(\left[R_z\right]V_S\right)}}forparticulatematter$$ (3)

The R_a depends on the atmospheric turbulence over the surface and is independent of the species (Eq. 4); it dominantes typically from 10 to 100 m above the ground level (Cherin et al., 2015; Zhang et al., 2017).

$$R_a=\frac{1}{k{u}_{*}}\left(\ln \frac{z_h}{z_0}-{\Psi }_h\right)$$ (4)

Where k is the von Karman constant (0.4); Z_h is the reference height; z₀ is the surface roughness height above the displacement plane; u_* is the friction velocity depend upon the atmospheric turbulence and 𝛹𝛹_ℎ is a stability function of momentum depend upon Pasquill stability class 203 and calculated based on Monin-Obukhov length (L; m) using Eq. (5):

$$L=\frac{u_{*}^3{c}_p\rho \overline{T}}{kgH}$$ (5)

c_pis the specific heat at constant pressure; 𝑇𝑇 the average temperature; and H the sensible heat (Koloskov et al., 2007).

The R_b and R_bp affects the deposition of air adjacent to the surface and depends on molecular properties of the pollutant and roughness of the surface (Eqs. 6 and 7 for gaseous pollutant and 209 particulate matter, respectively) and it dominantes typically from 0 to 10 cm above the 210 deposition surface (Giardina and Buffa, 2018; Zhang et al., 2017).

$$R_b=\frac{2S{c}^{\frac{2}{3}}}{P{r}^{\frac{2}{3}}k{u}_{*}}\text{for}\text{gaseous}\text{pollutants}$$ (6)

Sc is the Schmidt number (υ/D), υ is the kinematic viscosity of air, D is the molecular diffusivity of the pollutant, determined from Stokes-Einstein equation $\left(D=\frac{kT{C}_c}{3\pi \mu D_p}\right)$ with K the

Boltzmann constant; T the absolute temperature; μ the air dynamic viscosity; and C_c the Cunningham factor and P_r is the Prandtl number $\left(=\frac{C_{p}v}{k}\right)$ depend on C_p is the heat capacity per

unit volume of the air, υ is kinematic viscosity, and k is thermal conductivity.

$$\frac{1}{R_{bp}}=\left(\frac{1}{R_{bd}}+\frac{1}{R_{bi}}+\frac{1}{R_{bi}+R_{bt}}\right)\text{ for particulate matter}$$ (7)

R_bd depends on the Schmidt number, R_bi on surface conditions and Stokes number, and R_bt on 219 dimensionless particle relaxation time.

The R_c is the most uncertain resistance and varies with the nature of surface and type of green infrastructure (Baldocchi et al., 1987; Escobedo and Nowak, 2009; Fowler, 1981; Hirabayashi et al., 2012; Janhäll, 2015; Jayasooriya et al., 2017; Lagzi et al., 2013; Morani et al., 2014; Nowak et al., 2006; Tiwary et al., 2009; Vranckx et al., 2015; Wesely, 1989; Zhang et al., 2003). Most of the deposition models used Eq. (8) to evaluate canopy resistance (Seinfeld and Pandis, 2006; Walmsley and Wesely, 1996; Wesely, 1989; Zhang et al., 2017)

$$\frac{1}{R_C}=\frac{1}{R_S+R_m}+\frac{1}{R_{lu}}+\frac{1}{R_{dc}+R_{Cl}}+\frac{1}{R_{ac}+R_{gS}}$$ (8)

When gaseous pollutants enter the substomatal cavity in the leaves’ canopy, the diffusion of 228 gaseous material is resisted through stomata of the leaves know as stomatal resistance (R_s) and 229 through aqueous media of the spongy mesophyll cells known as mesophyll resistance (R_m). 230 Zhang et al. (2017) reported the empirical Eq. (9) based on Wesely (1989) to assess the stomatal 231 resistance for green infrastructure.

$$R_S=R_i\left(1+\frac{1}{200\left({(G+0.1)}^2\right)}\right)\frac{400}{T_S\left(40-T_S\right)}\frac{D_{H2O}}{D_X}$$ (9)

R_i is the minimum canopy stomatal resistance depend on land cover, G is the solar radiation, T_s is the surface temperature (°C), D_H2O/D_x is the ratio of diffusivity between the water vapour and the gaseous pollutant concerned. The R_s is primarily a function of the size of the stomata and time of opening or closing of stomata based on plants’ photosynthesis requirement (Zhang et al., 2017). The R_m depends on the chemical properties of pollutant such as solubility and reactivity and type of vegetation species. While interacting within mesophyll cells, pollutants 239 transform from air to liquid and then diffuse into aqueous media of the spongy mesophyll cells. Xiao and Zhu (2017) recognised that the pollutant, after entering the substomatal cavity, feels resistance offered through physical barriers and biochemical components inside the cell wall, chloroplast envelope, cytosol and stroma between substomatal cavity and chloroplast of leaf and can be modelled as Eq. (10).

$$R_m=\frac{RT}{H_c}{R}_{air}+R_{liq}$$ (10)

R is the gas constant (bar m³ K⁻¹ mol⁻¹), T is the absolute temperature (K), H_c is the Henry law constant (bar m³ mol⁻¹) is a dimensionless number used to convert R_air to its liquid-phase equivalent resistance and R_liq is the summation of all series of liquid resistances (resistance of the cell wall and membrane, resistance of the cytosol between chloroplast and the cell wall, 249 resistance of the chloroplast envelope and resistance of the stroma.

The area over leaf other than the stomatal opening is the cutin polyester membrane impregnated and the waxy surface of the leaf that offers deposition surface for air pollutant is known as cuticle (Ruggiero et al., 2017) and the resistance to deposition is known as a cuticle or cuticular resistance (R_lu). Because of the permeation of air pollutants are negligible through cuticles 254 (Grünhage and Haenel, 1997), the resistance is an order less than stomatal resistances (Wesely and Hicks, 1977). The value of cuticle resistance is subjected to a degree of wetness of the surface and pH value (Massad et al., 2010) that divided into dry cuticle resistance and wet cuticle resistance (Walmsley and Wesely, 1996). The single model to evaluate the cuticle resistance for a wide range of green infrastructure, organs or pollution condition is not possible because deposition and gas exchange phenomena are the functions of wettability, the degree of polarity and apolarity of plant surfaces, retention and quality of surface-deposited liquids (Fernández et al., 2017). Zhang et al (2002a) have estimated the cuticle resistance for the dry and wet conditions by Eq. (11) and Eq. (12), respectively.

$$R_{lu}{}_{dry}=\frac{R_{lu}{}_{dry0}}{e^{0.03RH}LA{I}^1/4u_{*}}$$ (11)

$$R_{lu\text{ wet }}=\frac{R_{lu}{}_{wet0}}{LA{I}^{1/2u_{*}}}$$ (12)

Where R_{lu dry0} and R_{lu wet0} are reference values of dry and wet cuticle resistance varies with the 266 type of green infrastructure. Due to the porosity of GI, air pollutants may enter into the lower canopy of dense vegetation where atmospheric buoyancy resistance (R_dc) is dominated by buoyant convection force that depends on the amount of sunlight heats the surface or lower canopy, and the angle of terrain 270 (Eq. 13).

$$R_{dc}=100\left(1+\frac{1000}{G+10}\right)\times {(1+1000\theta )}^{-1}$$ (13)

θ is the is the slope of the local terrain in radians. The R_dc in the lower canopy is independent 273 of wind speed but in some cases when winds are able to penetrate into the lower canopy, 274 especially on the sides of hills, could change the mixing force (Wesely, 1989).

The resistance to the uptake of air pollutant by leaves, twig, bark and other exposed surface within the lower canopy is known as lower canopy resistance (R_cl). It depends on canopy structure such as bark area index, porosity and areal density. Zhang et al. (2017) reported Eq. (14) to estimate the R_cl as:

$$R_{cl}={\left(\frac{{10}^{-5}{H}_c}{R_{cl.So2}}+\frac{f_0}{R_{cl,03}}\right)}^{-1}$$ (14)

f₀ is the reactivity factor for gases, R_{cl, SO2} and R_cl,O3 is baseline R_cl for SO₂ and O₃, as given in 281 Wesely (1988).

Many deposition models assume that deposition over the ground surface under green infrastructure was negligible but some report that the amount of deposition at surface varies from 20–30% depending on the type of air pollutants (Meyers and Baldocchi, 1988). The resistance offered by the green infrastructure to the air pollutant while passing through the canopy to the ground surface known as in-canopy resistance (R_ac). It is a function of canopy 287 height and LAI of GI and can be modelled using Eq. (15) (Erisman et al., 1994).

$$R_{ac}=\frac{h_c{b}_{c}LAI}{u_{*}}$$ (15)

Canopy height h_c is the height above the ground to the top of green infrastructure and b_c is an empirical constant taken as 14 m⁻¹. The canopy soil resistance (R_gs) is a resistance to the uptake of air pollutant by soil surface.

The deposition of air pollutant depends on pH value of soil, relative humidity, surface temperature, soil moisture content, ambient concentration of air pollutant, pollutant type and solar radiation (Erisman et al., 1994). The soil resistance, R_soil is the similar canopy soil 295 resistance and its parameterisation is given in Eq. (16) (Jacobson, 2005).

$$R_{gs}/R_{soil}={\left(\frac{{10}^{-5}{H}_c}{R_{gs,So2}}+\frac{f_0}{R_{gs,03}}\right)}^{-1}$$ (16)

R_{gs, SO2} and R_gs,O3 is baseline soil resistance for SO₂ and O₃ are given in Wesely (1988). The Surface resistance (R_s) accounts the amount of air pollutant deposition to any surface including soil, snow, concrete or any other surface. Canopy resistance is also a subpart of the R_s. Erisman (1994) reported temperature-dependent (Eqs. 17–18) snow-covered R_s for SO₂ and NH₃ as:

$$R_{snow}=500s{m}^{-1}atT<-1^{°}C$$ (17)

$$R_{snow}=70(2-T)s{m}^{-1}at1<T<-1^{°}C$$

For particles, addition parameter V_s, requires to estimate V_d, for particles diameter up to 50 μm according to the Stokes law (Eq. 19)

$$V_s=\frac{d_p^{2}g\left({\rho }_p-{\rho }_a\right)c_c}{18{\mu }_a}$$ (19)

ρ_p is the density of the particles, ρ_a is the density of the ambient air, d_p is particle diameter and g is the gravitational acceleration.

Alternative methods to evaluate the resistances used for dry deposition estimation for gaseous pollutants are also reported in the literature (Alfieri et al., 2008; Bennett et al., 1973; Ganzeveld and Lelieveld, 1995; Gong et al., 2017; Irmak and Mutiibwa, 2010; Jiang et al., 2017; Kerstiens, 2006; Kumar et al., 2014; Lhomme and Montes, 2014; Magnani et al., 1998; O’Dell et al., 1977; Rodný et al., 2016; Wesely, 1989; Wong et al., 2018; Zhang and Shao, 2014; Zhou et al., 2017) and Khan et al. (2017) have discussed other models for dry particle deposition models. A summary of the typical value of different resistance are listed in the Supplementary Information, SI, Table S1, and parameters required for dry deposition estimation are summarised in Table 2.

Table 2.

Summary of parameters required to calculate the dry deposition over GI surfaces.

Resistance	Meteorological parameters	Topographical parameters	GI parameters	Pollutant parameters
Ra	Temperature; density of air; Specific heat; sensible heat; Friction velocity; Wind speed and direction	Terrain data; Building geometry	-	Source location and elevation; Source outlet velocity; Source geometry
Rb	Kinematic viscosity; Thermal conductivity; Air dynamic viscosity; Temperature; Cunningham factor, particle relaxation time, thermal conductivity; Heat capacity per unit volume	Land cover/use data; Surface roughness	-	Molecular diffusivity; particle diameter
Rs	Air temperature, solar radiation; Solar elevation angle; Diffusion and directbeam solar radiation; Conductance-reducing effects of air temperature	The angle between the leaf and the sun	Photosynthetically Active Radiation (PAR); Leaf water potential; vapour pressure deficit, LAI; Water vapour pressure deficit;	Molecular diffusion
Rm	-	-	Net photosynthesis rate per leaf area; Type of species,	Henry law constant; Absolute temperature, gas constant
Rlu	Atmospheric acids; Ambient air temperature; Relative humidity; Seasonal conditions,	-	The thickness and chemical composition of leafsurface water-layer; LAI; Pollutant concentration over leaf; Cuticle surface area; Formation, growth and fate of water films; Type of species; Age of leaf	Pollutant concentration; Rate of pollutant interaction, pollutant composition,
Rdc	Solar radiation; Solar elevation angle	Slope of the local terrain	-	-
Rcl	-	-	Bark area index; Porosity; Areal density; Stem area	Henry law constant; Absolute temperature; Gas constant, reactivity factor for gases; Baseline lower canopy resistance for SO2 and O3
Rac	Friction velocity; Wind speed and direction	-	Canopy height; Leaf area Index	-
Rgs and Rsoil	Relative humidity; Ambient concentration of air pollutant and solar radiation	pH value of soil; Soil moisture content	-	-
Rb (for particles)	Schmidt number; Air kinematic viscosity; Friction velocity; Air dynamic viscosity; Stokes number; Density of air; Ambient Temperature; Relative humidity	Surface rough; Land use cover	LAI; Height of canopy	Particle’s Brownian diffusivity; Cunningham factor; Particle’s settling velocity; particle relaxation time; Particle diameter; Particle density;
Vs	Density of air; Gravitational acceleration; Air kinematic viscosity	-	-	Particle diameter; Particle density; Cunningham factor

The most widely used deposition model is i-Tree Eco (www.itreetools.org) that map the measured air pollutants concentration by monitoring stations over each GI canopies then estimates the air pollutants sequestration rates based on air pollutants concentration, LAI of GI, meteorological data. For example, Selmi et al. (2016) estimated 88 t of air pollutants removal by urban trees during one year (July 2012 to June 2013) in Strasbourg city, France. However, dispersion aspects of GI mainly air pollutants redistribution and enhanced atmospheric dilution are not considered in i-Tree Eco.

4. Considerations of GI in microscale models

Microscale models are used to predict the air quality near the source where air pollutant dispersion is dominated by source characteristics such as source induced turbulence, pollutant chemistry, local meteorology, source geometry and surrounding building, terrains and GI, flow alternation and many more. In street canyons, aerodynamic effects have a greater impact on local air pollution levels than the deposition (Vos et al., 2013). However, in open road conditions, both aerodynamic and deposition effects are important (Tong et al., 2016). Here, we have reviewed several models that have been developed to estimate exposure

concentrations by capturing the temporal and spatial variation of air pollutant at the microscale and how these models consider GI in the simulations. Though these models are not developed to study the impact of GI on exposure reduction, these could still be partly used to assess the GI impact of air quality. Table 3 presents a summary of the differences in GI considerations for various microscale models that are described in subsequent text.

Table 3.

Summary of the consideration of differences processes in various microscale models.

Microscale models	Consideration of different process
	Absorption	Filtration	Aerodynamic effect	GI geometry	Pollutant tolerance limit
Box and wind tunnel models	No	Yes	Yes	Yes	No
Gaussian plume models	Yes	No	No	No	No
Receptor models	Yes	Yes	No	No	No
CFD models	Yes	Yes	Yes	Yes	No
Hybrid models	Yes	No	Yes	Yes	No

Box and wind tunnel models are based on the principle of conservation. The study domain is assumed as isolated, from the surrounding, into which physical and chemical process of air pollutant dispersion, dilution and deposition are simulated. The consideration of roadside GI on microscale models (especial urban street canyon) was intensively studied through laboratory experiments in a wind tunnel by Gromke and co-workers (Gromke, 2011; Gromke et al., 2012; Gromke and Blocken, 2015; Gromke and Ruck, 2007). In these models, air pollutants are normally forced to pass through the vegetation that can increase the deposition velocity by allowing more time and amount of pollutant to interact with GI (Janhäll, 2015). However, under ambient conditions, air pollutants only interact with the surface area of GI while passing above or around (Abhijith et al., 2017). Moreover, synthetic materials are used to represent the GI which produces the uncertainties in the simulation of R_c.

Gaussian plume models were most commonly used as dispersion model to estimate the concentration of air pollutant by solving the set of mathematical equations in three dimensions considering usually the point source. To incorporate stability conditions and plume rise, the model can use any one of the five stability classification schemes (Lapse Rate scheme, PasquillGifford scheme, Turner scheme, Sigma-Theta scheme and Richardson number) and any one (Briggs and Holland method) of the two plume rise formulations (Awasthi et al., 2006). Models also have two different vertical and horizontal dispersion coefficients to simulate the vertical and horizontal dilution depend upon stability class and distance from the source. In Gaussian models, air pollutant concentration is independent of GI characteristic. With computational advancement, Gaussian plume model limitations have been minimized in modified/advanced models that estimate pollutant concentrations with a combination of different sources’ contribution, which are spatially distributed sources (such as multiple points, line and area sources). These models use meteorological data and surface roughness height to compute surface friction velocity, Monin-Obukhov length, the wind speed and direction at a reference height etc. to define atmospheric boundary layer properties and used modified vertical and horizontal dispersion coefficient based on wind tunnel, field experiment or empirically. RLINE takes meteorological input from AERMET (Cimorelli et al., 2005), vertical and horizontal dispersion with empirical constants obtain based on field tracer studies and wind tunnel simulation (Venkatram et al., 2013) and have been tested against independent field studies (Snyder et al., 2013). RLINE does not incorporate the effect of GI (roadside vegetation) but it can predict pollutant concentration near to measured field data due to modified dispersion coefficient. HIWAY-2 uses the steady-state Gaussian model to estimate the concentration of nonreactive pollutants from highway traffic at receptor located in relatively uncomplicated terrain and unable to consider complex terrain with GI (Peterson, 1980). This model uses only three stability classes (unstable, neutral and stable) and more realistic concentration estimation adjacent to the highway with respect to original version because of updated dispersion coefficients (Peterson, 1980; Sharma and Khare, 2001). Other Gaussian models are also used modified dispersion coefficient to include the effect of non-porous barriers. ADMS-Road is a Gaussian plume air dispersion model that uses the boundary layer depth and Monin-Obukhov length to define the atmospheric boundary layer properties rather than the simplistic Pasquill-Gifford stability classes. This model calculates dry deposition removal by estimating R_a, R_b component and uses a constant value of R_c for the whole modelled domain that could lead to uncertainty in predicting air quality (Apsley, 2017). Usually, the advanced Gaussian models are not able to simulate the aerodynamic effects of GI that dominate the spatial distribution of air pollutant at the microscale.

Receptor models are based on a set of mathematical relationships to estimate the source contribution at the measured receptor point. Watson et al.(2002) have described the procedures for using a receptor model to estimate the contribution of sources at a receptor locations. Receptor models are simple, explanatory and highly precise to estimate air pollutant concentration at receptor point (Watson, 1984). Many studies (Belis et al., 2013; Gardner and Dorling, 1999; Han et al., 2004; Liu et al., 1996; Song et al., 2001; Wåhlin et al., 2006; Wahlina et al., 2001) have been reported the usage of receptor model applications apply to air quality measurement. The receptor models can be used to quantify the potential of GI as a whole (Chen et al., 2016, 2015; Heal et al., 2012; Li et al., 2016; Maleki et al., 2016; Yin et al., 2011) but while predicting air quality, they are unable to handle the non-linear behaviours of dry deposition, spatial distribution of GI and other parameters such as complex local meteorological parameters, air moisture and wind velocity.

Computational Fluid Dynamics (CFD) models are an effective and powerful tool to simulate wind flow and mass transfer numerically. Most of the CFD models solve the governing nonlinear Navier Stokes equations, which are conservation of mass, momentum and energy along with transport and/or any other user specific equation, with the help of any conventional methods such as Finite Volume Method, Finite Element Method, Finite Difference Method and Spectral Methods. Many CFD models (Amorim et al., 2013; Baik et al., 2009; Costabile et al., 2006; Jeanjean et al., 2017, 2015; Kwak et al., 2018; Marciotto et al., 2010; Sanchez et al., 2016) have also been developed by researchers to simulate the complex wind flows and pollutant transfer problems at different scales. Although CFD model has the capacity to deal with complex shaped geometries, wind-induced turbulence and air pollutant transformation to predict the air quality (Amorim et al., 2013; Costabile et al., 2006; Jeanjean et al., 2015; Kwak et al., 2018; Lateb et al., 2016; Sanchez et al., 2016) but still requires more validation to simulate the effect of modelled geometries on the wind velocity and air quality prediction in urban setting (Huang et al., 2009; Sini et al., 1996). GI is considered as porous media and its pollutant removal efficiency is modelled as function LAI or Leaf Area Density (LAD) by assuming constant deposition velocity (Jeanjean et al., 2017; Pugh et al., 2012). Well-configured CFD models are capable of resolving the wind flow field to estimate the air pollutant removal via atmospheric turbulence and Brownian motion near the surface. However, the effect of different species, leaf size, soil moisture content and other parameters to define canopy resistance are neglected in most of such studies. Therefore, CFD models could usually estimate less deposition than actual and predict higher pollutant concentration. Steffens et al. (2012) reported the first modeling study on the effect of vegetation barriers on plume dispersion near roadways through coupling turbulence and aerosol dynamics to capture both the aerodynamic and deposition effects. The modeling results show that the vegetation barriers can reduce UFP concentrations, but the level of reduction depends on meteorological conditions. For example, increased wind speed leads to more reduction in particles smaller than ~50 nm, but slight increase in particles larger than ~50 nm as a result of interactions among aerodynamic resistance, impaction and residence time. The predicted effect of wind speed by Steffens et al. (2012) agreed well with the observed patterns from a later field study (Lee et al., 2018), even in terms of the critical size (predicted 50 nm vs observed 60 nm).

Hybrid models are combination of two or more models which are either used in series to generate desirable output (Karamchandani et al., 2009; Korek et al., 2016; Sharma et al., 2013; Sun et al., 2017) or complementary to target specific problem such as development of modified flow field due to complex terrain, estimation of dry deposition because of GI, generate spatial meteorological data etc. within domain (Fallahshorshani et al., 2012). Eulerian grid models, Lagrangian puff dispersion or trajectory models along with Gaussian plume models are the classical examples of hybrid models. Worldwide, researchers are developing new hybrid models to simulate the complex air pollutant dispersion processes under different environmental conditions. Currently, the major problem in hybrid models is the linkage between models, which results in researchers having to develop unique methodologies for linking the different models and dealing with specific case requirements linked for example to data availability. For instance, LAI can be estimated by different models such as model that relates LAI to digital cameras values (Casadesús and Villegas, 2014), mathematical equations for LAI, absorbed photosynthetically active radiation and net primary production from remote sensing (Gower et al., 1999), by mathematical regression model (Blanco and Folegatti, 2003), models that used satellite data to estimate LAI (Aboelghar et al., 2010; Chen et al., 2005; Xavier and Vettorazzi, 2004) and many direct or indirect methods (Bréda, 2003; Gower et al., 1999). Hybrid models are capable to consider the GI in air quality prediction but because of the data required for consideration of GI (Table 2), uncertainty in data generation, spatial and temporal variation of data, there is no accepted hybrid approach to consider the GI in prediction of the air pollutant concentration.

The impacts of GI in urban areas on pollutant dispersion are not considered in many microscale models. A comprehensive understanding of individual aspects such as dry deposition (Nowak et al., 2006); filtration (Chen et al., 2017) and air pollutant concentration redistribution (Gromke and Ruck, 2008; Miao et al., 2016) are needed to consider in urban air quality simulation. For dry deposition, an additional sink term may be used in computational models to simulate the gaseous pollutants absorption into leave matrix through stomata. The mathematical model is used to estimate deposition shown in Eq. (1) is applicable when total LAD and concentration point belong to two different cells mostly in the macroscale model. Therefore, the Eqs. (2) and (3) describe a mathematical model to estimate V_d consider R_a that is resistance between the height of concentration modelled and the GI canopy, are applicable for forest canopy only. On the other hand, the microscale models have much finer resolution and the concentration is resolved around GI that has the impact of R_a so V_d should not be same as macroscale. For instance in CFD simulation, sink term is proportional to a concentration within a cell, V_d and LAD shown in Eq. 20 (Buccolieri et al., 2018; Jeanjean et al., 2016; Vos et al., 2013) where V_d is generally assumed same as macroscale.

$$S_d=-LAD\cdot V_d\cdot C$$ (20)

Apart from trees and hedges, surface like grass, soil and water also offer deposition but this deposition is not considered in microscale modelling. The filtration capacity of GI is the amount of gas and particle retained into its volume without absorption. It is most important for particles deposition inside the GI canopy. Furthermore, modelling particles deposition is more complex than gases because of re-suspension under high wind speed. The amount of re-suspension depends on wind speed and amount of epicuticular wax available on leave in different seasons (Zhang et al., 2017). Hefny et al.(2015) have discussed the GI-induced aerodynamic effects with different approaches; (i) implicit approach (represented by surface roughness); and (ii) explicit approach (represented by porous media) for inclusion of GI in numerical modelling, and concluded that explicit approach is more physically realistic over implicit approach for simulation of wind flow field and dispersion modelling. The built-up street canyon configuration and vortexes are shown in Fig 2. These vortexes in street canyon help to reduce the concentration through dilution. The flow around GI are dependent on the location of the latter in a built-up street, interaction with street canyon vortexes, surrounding environment and meteorological conditions. The porous behaviour makes the aerodynamic effect of GI different from solid barriers. Two zones – a low turbulence zone due to flow abatement and a high turbulence zone due to a sudden change in flow deflection – are generated due to vortexes interacting with the street GI (Vos et al., 2013). Generally, low turbulence zone shows the high concentration owing to relatively lesser mixing and the part of air pollutants that pass through GI loss momentum and get stagnant (Santiago et al., 2017). Tong et al. (2016) employed the CFD-based CTAG model to evaluate six different vegetation barrier configurations, and identified vegetation–solid barrier combination (i.e., solid barrier followed by vegetation barrier) as one of most effective design options, via promoting vertical mixing (through solid barriers) and enhancing deposition (through vegetation barriers). The CTAG modeling results also revealed that a highly porous roadside vegetation barrier containing large gaps within the barrier structure could increase downwind pollutant concentrations, consistent with the findings in a field study (Deshmukh et al., 2018). In conclusion, the choice of microscale model to study the impact of GI on local air quality simulation depend on available input parameters, simulation time, the importance of GI-pollutant interaction processes, meteorological condition and surrounding urban geometry.

Figure 2.

A schematic of the changes in the formation of vortex (shown by blue lines) in urban street canyon during the interaction of the flow with the leeward GI, showing vortex formation (a) in street canyon under perpendicular wind flow (Gromke, 2011) as well as cross-section and top view of (b, c) GI free street canyon, (c, d) street canyon with hedges, (e, f) street canyon with tree.

5. Consideration of green infrastructure on Macroscale models

Macroscale models are used to predict the air quality around the source where air pollutant dispersion is dominated by the meteorological and topographical conditions such as wind velocity and direction, ambient temperature, terrains slope, surface roughness and deposition. While considering GI in macroscale models, the important aspects are deposition velocity (Verbeke et al., 2015) and change in friction velocity and surface roughness (Barnes et al., 2014; Britter and Hanna, 2003). Here, we have reviewed methodologies for considering GI in macroscale air dispersion models. Table 4 summaries the differences in GI considerations in various macroscale models that are discussed in the subsequent text.

Table 4.

Summary of the consideration of different processes in various macroscale models.

Processes	Macroscale models
	Gaussian plume models	Modified Gaussian models	Statistical models	Receptor models	CFD models	Hybrid models
Absorption	No	Yes	Yes	Yes	Yes	Yes
LAI	No	No	Yes	No	Yes	Yes
Land cover	No	Yes	Yes	No	Yes	Yes
Surface roughness	No	Yes	No	No	Yes	Yes
Terrain data	No	Yes	No	No	Yes	Yes
VOC emissions	No	No	No	No	No	No
Coupled dispersion- deposition	No	Yes	Yes	Yes	Yes	No
Background Concentration variation	No	No	No	No	No	Yes

At macroscale, Gaussian plume models have been widely used for air quality prediction from a point source and have some limited applicability such as flat terrain, no local flow and circulation and single source. Apart from inbuilt assumptions of Gaussian plume model such as continuous steady source, chemically inert pollutant, bell-shaped distribution of pollutants in the horizontal and vertical direction and constant meteorological conditions. Gaussian models can only represent the dispersion under the described above conditions and are unable to handle complex environmental conditions, topography, other atmospheric chemistry and the air pollutant removal capacity of GI through dry deposition. Modified Gaussian models such as ADMS-Urban, AERMOD, CALPUFF and SCREEN3 are the most commonly used amongst the scientific community. ADMS-Urban is a Gaussian plume air dispersion model that computes local flow along with turbulence due to surface roughness based on Nguyen et al.(1997) and dry deposition velocity with constant surface resistance (Apsley, 2017). With constant surface resistance, ADMS-Urban is unable to incorporate the effect of different land cover on deposition. AERMOD is a steady-state Gaussian plume model, developed by American Meteorological Society, that takes meteorological inputs from AERMET (Cimorelli et al., 2005) and terrain data from AERMAP (Langner and Klemm, 2011). AERMOD has relative leaf area index that is used, rather than actual LAI that depends on nine land use categories and five seasonal categories, to compute deposition velocity based on Wesely (1988). Lin et al. (2018) have shown in their large-eddy simulation study that total particle deposition is sensitive to LAI. CALPUFF is a non-steady-state Lagrangian Gaussian puff model; it uses CALMAT generated wind field or wind velocity and other inputs such as surface roughness and terrain data to compute dispersion coefficient (Bai et al., 2018). CALPUFF estimates deposition velocity without considering the R_dc and R_cl due to GI porosity and other surfaces (e.g., grass, soil) in the lower canopy that could lead to uncertainty in dry deposition amount. These Gaussian models are unable to simulate the (i) effect of flow deflection and flow abatement near to receptor; (ii) pollutant tolerance limit of different GI species; (iii) characteristic (shape and dimensions) of GI types; and (iv) effect of GI inside the urban area.

Statistical models are mathematical models that predict the air quality with the help of unknown constants estimated based on measured data such as land use regression models (Rao et al., 2017), machine learning models (Kleine Deters et al., 2017), and Monte Carlo (Mallet and Sportisse, 2008). The statistical models need to be trained for past data sets to assess the impact of variables in the conditions under which they were initially trained. Land use regression models have been used widely to estimate air quality by finding mathematical relations between parameters such as land cover (LAI, types and spatial distribution of GI, land use pattern), meteorological data, emission data and ambient concentration (Cattani et al., 2017; de Hoogh et al., 2014; Habermann et al., 2015; Hoek et al., 2008; Yli-Pelkonen et al., 2017). Because these models do not consider the physical relationship between emissions and air quality under given sets of meteorological and topographical conditions (Shahraiyni and Sodoudi, 2016; Wolf et al., 2017; Zhang and Ding, 2017), effects of GI on air quality remains unanswered in statistical models.

CFD models at macroscale take high computational time and resources even for a smaller size of domain but are able to capture the effects of GI at a fine spatial resolution. The modelling methodology and limitations for considering GI in CFD models, box or wind tunnel models, receptor models and hybrid models are discussed (Section 4).

GI reduce air pollutants concentration through pollutants removal by dry deposition and enhancing atmospheric dilution by surface roughness at the urban scale. Most macroscale simulations use the average deposition value for every cell within a simulated domain because grid size is more than the dimension of GI that introduces uncertainty in air quality prediction. The seasonal variation leading to change in LAI, VOC emission, non-linear deposition behaviour and particles resuspension are important input, apart from meteorological data, terrain data and surface roughness, that needs to consider in dispersion modelling and could change air quality in macroscale. Generally, deposition models estimate air pollutants removal over GI and neglect increased atmospheric dilution by their surface roughness during an assessment of GI impact. For instance, Bodnaruk et al. (2017) have used i-Tree Eco to estimate air pollutants removal equal to 173 t. yr⁻¹ in 95 km² area by increasing tree cover from 24% to 44.4% between 2010 to 2040 in Baltimore, US.

6. Challenges in considering GI into dispersion modelling at microscale and macroscale

The atmospheric dispersion models predict air quality under the influence of different inputs such as meteorological data, topographical data, GI data and source emissions. Apart from the complexity of environmental processes and interface interactions (Holnicki and Nahorski, 2015; Irwin, 2014), the variance in measured and estimated concentrations with presence of GI is because of (i) pollutants measurement error, (ii) model input uncertainties (table 2), (iii) simplification of deposition processes (Section 3), and (iv) treating GI in numerical solution (Sections 4 and 5). Past studies have reported effects of input data uncertainties on pollutants concentration at different length scales. For instance, meteorology data uncertainty at mesoscale (Gilliam et al., 2015; Godowitch et al., 2015), emission data uncertainty at macroscale (Diez et al., 2014; Holnicki and Nahorski, 2015) and at mesoscale (Saikawa et al., 2017), topography and land use data uncertainly at macroscale (Zou et al., 2016), surface roughness uncertainly at macroscale (Barnes et al., 2014). The challenges in model inputs and processes especially during consideration of GI in air quality simulation are mainly related to (i) spatial-temporal variation of GI characteristics such as shape and size, porosity, pollutant tolerance limit and pollutant sink; (ii) pollutant transformation due to GI; and (iii) influence of meteorological and topographical data such as temperature, humidity, terrain slope, wind speed and direction on deposition process. The uncertainties in model input data that cause predictions of air pollutants could vary up to a factor of four for identical solutions (Lohmeyer et al., 2002). The most important challenge is consideration of the spatio-temporal distribution of GI and its characteristics because each GI type is different from others with respect to its location, LAI, porosity, species and geometries. Past studies (Aubrun et al., 2005; Buccolieri et al., 2011; Gromke et al., 2012; Gromke and Ruck, 2008; Miao et al., 2016) have discussed the effects of GI on local wind velocity and dispersion at different length scale. An additional challenge is the seasonal LAI variation that affects the change in porosity and geometry of GI (wind flow alternation) and change in pollutant absorption rate (deposition velocity). GI species are also having different tolerances to different air pollutants (Appleton et al., 2009; Yang et al., 2015) that may cause short-term damage (immediately visible sign on leaves) and long-term damage (premature leaf drop, reduced growth and species death). Furthermore, the meteorological and topographical conditions have an influence on deposition processes and are not typically included in model parameterizations. For example, high ozone is more likely to co-occur with high temperature (including during heat waves) and low humidity condition has been reported in many studies (Filleul et al., 2006; Hou and Wu, 2016; Zhang et al., 2017; Zhang and Wang, 2016). Similarly, high PM_2.5 are also reported during high temperature and low wind speed (Tai et al., 2010; Zhang et al., 2017). For example, Kavassalis and Murphy (2017) have reported the coincidence of low ground-level O₃ concentration and high relative humidity in US between 1987 and 2015. This is because O₃ uptake is unintentionally high by plants’ stomata during the high relative humidity. Furthermore, Zhang et al. (2003) have highlighted the combined effect of strong solar radiation 596 (> 200 Wm⁻²) and wet condition (rainfall or morning dew) on R_s and introduce the term W_st 597 (the fraction of stomatal blocking water film) in R_c calculation, Eq. (20).

$$\frac{1}{R_c}=\frac{1-W_{st}}{R_s+R_m}+\frac{1}{R_{lu}}+\frac{1}{R_{dc}+R_{cl}}+\frac{1}{R_{ac}+R_{gs}}$$ (20)

The pollutant transformation due to the presence of GI is another challenge in air dispersion modelling. VOCs such as isoprene, monoterpenes and sesquiterpenes are released by GI and can have a considerable effect on air pollutants concentration during heat waves (Churkina et al., 2017). These VOCs can undergo chemical transformation and produce ozone and PM in the presence of high nitrogen oxide concentration (Seinfeld and Pandis, 2006). This study observed that the GI-air pollutants interaction (deposition rate) is not constant which is assumed in many dispersion modelling but actually has a high spatial-temporal variation that depends on GI types and spatial distribution as well as meteorological and climatic factors such as relative humidity, ambient temperature and wind velocity.

7. Framework for linking GI, air pollution and health outcomes

The knowledge of linkage between public health outcomes and GI-induced air pollution reduction is important to assess public health benefits (Fig. 3). We focus here on a basic framework that links the GI’s reduction of atmospheric pollutant concentration and the potential exposure of individuals to assess the resultant health risk related to air pollution (Table 5). An air pollution health risk assessment (APHRA) is a mathematical approach that estimates the expected health impact of short-or long-term exposure to air pollutants in different variables such as age group, environmental conditions and socioeconomic status (WHO, 2016). Here, we present an APHRA framework to link health outcomes with short-or long-term exposure alternation due to presence of GI. Several epidemiological studies have reported a wide range of health impacts such as respiratory symptoms, hospital admission and/or premature deaths, associated with excess exposure to air pollution. These health outcomes could be quantified through a number of premature deaths, years of life lost (YLLs), disability-adjusted life years (DALYs), or change in life expectancy (Health Organization Regional Office for Europe, 2015). For instance, Pope et al. (2009) have reported life expectancy increase by 0.64 years for per 10 μg m⁻³ decrease in PM_2.5 for 51 US cities. The inputs data required to estimate the health outcomes, shown in Eq. (21), are population data (P), baseline rates of death or hospital admission (M), change in air pollutant concentration (ΔQ) and relative risk (RR) (Andreão et al., 2018; Künzli et al., 2000; Sacks et al., 2018).

$$\Delta \text{M}=\left(1-e^{-RR*\Delta Q}\right)*P*M$$ (21)

Where ΔM is a change in health effects due to air pollution and RR is a change in health effect for an unit change in an air pollutant concentration. The main steps for APHRA is performed via (i) direct measurements of air pollution concentrations of individual’s exposure (Steinle et al., 2015); (ii) indirect measurements by estimating the pollutants concentration by modelling or fixed site monitors (Beelen et al., 2014; Brauer et al., 2016). For APHRA, the use of the indirect method is more common over the direct method that is mostly used in industrial or occupational health risk assessment. At a national scale study for 2425 urban and 3094 rural areas in year 2010, Hirabayashi and Nowak (2016) used fixed site monitoring data and applied pollutants concentration change due to deposition over the tree to identify the air pollutant removal and health benefits with increase in LAI or percentage of tree cover. The uncertainty in APHRA is brought by factors such as (i) ambient air, which is a complex pollutant mixture and making estimation of exposure assessment is challenging because monitoring stations are not mapping full domain, (ii) baseline disease burden is not properly recorded especially under developed countries, and (iii) assumptions made during the derivation of concentration-response function such as smoking condition, indoor air pollutant exposure and medical condition (WHO, 2016). The presence of GI could further increase uncertainty in APHRA by (i) adding allergen pollens that may trigger other diseases, (ii) bVOC emissions that may transform air pollutants and increase particular pollutant concentration locally, and (iii) altering the exposure by air pollutant concentration redistribution. Furthermore, USDA Forest Service developed i-Tree with BenMap (US-EPA) have used to estimate the health outcomes of annual air pollutants removal by the deposition over GI canopy at the macroscale. However, APHRA due to change in short-term exposure by pollutants concentration redistribution at the microscale and enhanced atmospheric pollutants dilution by increased surface roughness at the macroscale, because of GI, have not been assessed along with deposition.

Figure 3.

Framework for linking between GI, air pollution and public health.

Table 5.

Summary of relevant studies that have quantified the linkage between GI, air pollutant reduction and health benefits.

Author (year)	City; Model used	Summary
Tiwary et al.(2009)	London (UK); (ADMS-Urban + statistical model)	• PM10 removal has been estimated through different GI combinations with two species (sycamore maple (Acer pseudoplatanus L.), Douglas fir (Pseudotsuga menziesii Franco) and grassland using computational model. • PM10 reduction varied between 0.03 to 2.33 t ha yr−1 with different GI combination. • The health effects noted were a reduction in 2 premature deaths per year and 2 respiratory hospital admissions per year.
Powe and Willis (2004)	Britain(UK); (Monitoring stations+ statistical model)	• The absorption of SO2 and PM10 via forests (more than 2 ha in area) were estimated based on National Air Quality Information and Forest Commission spatial distribution data of woodland. • The forest can absorb large quality of air pollutants, for example, 385,695–596,916 metric tonnes of PM10 and 7715–11,215 metric tonnes of SO2 per year • The above air pollutants reduction would be equal to 5–7 deaths per year and 4–6 hospital admission per year.
David J. Nowak et al.(2013)	10 U.S. cities; (Monitoring stations + i-Tree)	• PM2.5 removal was estimated for 10 U.S. cities with existing trees cover using i-Tree model and U.S. EPA monitors concentration. • Total amount of PM2.5 removal varies from 4.7 to 64.5 tonnes per year. • The equivalent mortality reduction from 0.1 to 7.6 death per year.
Nowak et al.(2018)	86 Canada cities; (Monitoring stations + i-Tree)	• The change in air pollutants (NO2, O3, PM2.5, SO2, and CO) concentration have been estimated through the iTree model and health impacts were studied. • The total air pollutant removal was 7500 t to 21,100 t and average removal rate was 3.72 g/m2/year. • The overall health impacts of urban trees are included avoidance of 30 human mortality in all cities.
Hirabayashi and Nowak (2016)	U.S.; (Monitoring stations + i-Tree)	• Resultant changes in concentration for the four air pollutants (NO2, O3, PM2.5 and SO2) were modelled by the iTree model using deciduous and evergreen trees with varying LAI and 2010 census data. US EPA’s BenMAP has been used to link air quality improvement to human health improvement was estimated. • In urban areas, annual mean air pollutant concentration was 15.5 (μgm−3) for NO2, 61.7(μgm−3) for O3, 10.0 (μgm−3) for PM2.5 and 4.9 (μgm−3) for SO2. Changes in concentration increased as the LAI increased but the relation is non-linear. • A comprehensive national database of deciduous and evergreen trees with varying LAI and its effects on air quality and human health in the United States was developed
Nowak et al.(2014)	U.S.; (Monitoring stations + i-Tree)	• Avoidable health impacts and associated economic benefits of four air pollutants (NO2, O3, PM2.5 and SO2) removal by trees and forest in the US were estimated for 2010. • The estimated quantity of air pollutants removal was 17.4 million tonnes based on hourly pollution data and daily total tree cover and LAI though i-Tree model. • The existing trees can help in avoiding more than 850 incidences of human mortality and 670,000 incidences of acute respiratory symptoms.
Rao et al.(2014)	Portland; (Monitoring stations + LUR)	• LUR have been used to estimate a decrease in NO2 concentration by an urban tree in Portland. • The estimated removal of NO2 was 0.57 ppb for every 10 ha trees. • The annual health benefits are approximately 21,000 fewer incidences and 7000 fewer days of missed school due to asthma exacerbation for 4–12 year-olds; 54 fewer emergency visits across people of all ages; and 46 fewer cases of hospitalization due to respiratory problems triggered by NO2 in the elderly. • The potential of an urban forest to reduce the air pollutant (NO2) and hence provide health benefits are approximately 7 million USD due to reduced incidence of respiratory problems.
Bodnaruk et al.(2017)	Baltimore (US); (Monitoring station + i-Tree)	• The monetary benefits of increasing tree cover from 24% to 40% have been estimated under different spatial GI distribution. • An additional annual 173 ton air pollutants removal was predicted at maximum potential tree cover of 44.4%. • The monetary benefits of these air pollutant removal on human health were estimated equal to 6.3 million USD.
City of Woodland (2018)	Woodland (California); (Remote sensing + i-Tree)	• The total air pollutants removal and monetary benefits of existing 14.5% urban tree canopy have been assessed using high-resolution aerial imagery and remote sensing software for 2010. The U.S. EPA’s BenMAP Model was used to estimate monetary values resulting from changes in air pollutants concentration. • The analysis estimated that Woodland’s tree canopy annually removing 40 tons of air pollutants (includes CO, NO2, O3, SO2 and PM10) and save 15.3 million gallons of stormwater. • The monetary values of health benefits resulting from air pollutants removal have been estimated as equal to 1.8 million USD.
City of Sacramento (2018)	Sacramento (California); (Remote sensing + i-Tree)	• The U.S. EPA’s BenMAP Model was used to estimate monetary values resulting from air pollutants removal. These removals have been estimated using highresolution aerial imagery and remote sensing software for existing 19.1 % urban tree canopy in 2010. • The analysis estimated that Sacramento’s tree canopy annually removing 392.4 tons of air pollutants (includes CO, NO2, O3, SO2 and PM10) and save 58 million gallons of stormwater. • The monetary values of health benefits resulting from air pollutants removal have been estimated as equal to 18.8 million USD.

Some studies highlighted that the overall urban GI is associated with a decrease in mortality and morbidity. For instance, De Keijzer et al (2017) linked air pollution concentration and urban vegetation to standardized mortality rates using Poisson regression and to life expectancy using linear regression. This study was based on mortality data from 2148 small areas with an average population of 20750 inhabitants between 2009 and 2013 in Spain. It found an increase in life year of 0.17 (95% CI: 0.07, 0.27) with an increase in average greenness in urban areas by interquartile range. The same study also found reduction in life years of 0.90 (95% CI: 0.83, 0.98), 0.20 (95% CI: 0.16, 0.24), 0.13 (95% CI; 0.09, 0.17) and 0.64 (95% CI; 0.59, 0.70) due to an increase of 5 μg m⁻³ for each of PM₁₀, O₃, NO_x and 2 μg m^–3 in PM_2.5 respectively. In another study, Lovasi et al (2008) have examined the association between asthma prevalence in 4–5 years old children and the number of street trees in New York and found that an increased trees density was associated with 29% lower asthma prevalence (RR = 0.71 per 343 trees/km², 95% CI; 0.64, 0.79). Furthermore, this association could be a due reduction in air pollutants concentration by urban GI. On another hand, several studies (King et al., 2014; Selmi et al., 2016; Yang et al., 2008) have found air pollutants removal between 58.9 and 99.6 kg ha⁻¹ year⁻¹ through GI but health benefits were not estimated presumably owing to lack of availability of health data.

8. Summary, conclusions and future outlook

This review paper discussed various aspects related to consideration of GI in microscale and macroscale atmospheric dispersion models. The mathematical description of different processes and the relevant inputs required to simulate GI effect for estimating air pollutants removal under deposition scheme are discussed. Microscale and macroscale air pollutants dispersion models are surveyed with respect to their physical and chemical representation of GI into the modelling system and limitations to assess the effect of GI on air pollutants concentration estimation. The non-linear behaviour of GI deposition response with meteorological parameters and other challenges such as spatial-temporal variation of GI characteristic and its effects on air pollutant transformation are briefly studied. Moreover, the importance of GI in health risk assessment through numerical framework linkage between GI, air pollution and health is examined.

The following key conclusions are drawn:

• A detailed review of parameterisation for GI modelling to estimate deposition velocity allows concluding that for gaseous pollutants, V_d is dominated by R_c that is larger than R_a and R_b. The favourable conditions for gaseous pollutants to be absorbed in plant leaves depends on a series of parameters like LAI, size of stomata opening, photosynthesis rate per leaf area, PAR. On another hand, for particulate matter, V_d is highly depended on particle’s diameter as is governed by a U-shaped curve in which it is minimum for particles size between 0.4 and 0.9 μm in aerodynamic diameter.

• The synthesis of data inputs (as listed in Table 2) to evaluate the different resistances and deposition velocity for their estimation shows the simplification of deposition scheme may lead to large uncertainty in air pollutant removal estimation.

• The deposition schemes show the V_d should be taken differently for microscale and macroscale simulations because the pollutants concentration is resolved around GI that has the effect of R_a in microscale models. Generally, V_d value used in dispersion models is measured with assuming downward flux of pollutant over GI (forest area) in the field. However, traffic emissions with a roadside GI are a different configuration where sources are located under GI canopy.

• Usually, the dispersion models (Sections 4 and 5) are not originally developed to assess the effect of GI on air pollutants concentration estimation. However, they can still be able to capture a number of processes listed in Tables 3 and 4 for GI considerations at the different length scales. There is a yet a need of coupled GI-dispersion models that can incorporate the air pollutants-VOCs chemistry, GI pollutants tolerance limits and non-linear pollutants deposition under different meteorological conditions.

• The numerical framework that links GI, air pollution and health outcomes help to estimate the benefits of GI based on a reduction in short-and long-term air pollutants exposure reduction in the term of mortality, morbidity and monetary values. These benefits are not only dependent on the air pollutants deposited amount over the surface offered by GI.

Many numerical methods have been developed to simulate additional physical and chemical processes due to GI in atmospheric dispersion model at different spatial scales to estimate air pollutant removal (Section 3). These processes depend on many factors such as meteorological data, GI characteristic and air pollutant concentration. Moreover, the deposition scheme in details are complex, resource intensive and require a large number of input data (such as meteorological parameters, topographical parameters, GI parameters and pollutant parameters, as listed in Table 2), which is not handy for dispersion models. Different air pollutants dispersion methodologies used for air pollutant concentration simulation at various spatial scale have been discussed (Sections 4 and 5) along with the summary of their consideration of different GI related processes in table 3and 4. The individual spatial case has specific flow, mixing characteristic and sensitive to input parameters while simulating the GI effects on air pollutants concentration. It is, therefore, necessary to identify the individual spatial conditions to understand the dominating physical and chemical processes that need to incorporate in detail during simulation of real-world problems and simplification of various others processes to reduce the computational time and resources. For instance, the predominat processes in air pollution dispersion models for microscale is the aerodynamic effects that are strongly influenced by local meteorology, source characteristics and surrounding geometry and much stronger than the deposition. Therefore, the implicit approach for inclusion of GI at microscale would lead to the unrealistic prediction of air quality. Several other processes also influence deposition velocity and alter the air pollutants concentration have been reported in the literature (section 6) but most of them are not available in any air pollutants concentration simulation methodologies. Churkina et al. (2017) have stated that vegetation emits more VOCs with rising temperatures and contribute to the ozone which had reached ~60% during the heat wave in July 2006 and ~40% during hot summer days in 2014 in Berlin. It is found in health risk assessment studies that GI is the important linkage between air pollution and public health and has a significant impact not only in air pollutant reduction but also decrease the exposure of air pollution in the day to day life.

This review identified the mechanism of air pollutant removal through the deposition process and the relevant key processes (aerodynamic effects, flow alteration and atmospheric dilution) at different spatial scales while consideration of GI in dispersion modelling. The review of the topic areas also highlighted a number of research unanswered questions that require focus of future research studies: (i) what is the proportion of deposition and dilution in reduced pollutant concentrations for large-spaced GI such as urban parks, grassland and forest; (ii) what is the effect of filtering capacity, pollutant tolerance limit and temperature sensitivity of different GI types on dispersion models; (iii) how does the V_d change if vehicular emissions are forced towards the roadside GI due to local wind flow; and (iv) what modification should be required for combined GIs such as hedges with trees, grass with hedges or trees in dispersion model. Future studies should also develop the methodologies to consider the wind-dependent GI porosity in dispersion models. More studies are needed to develop approaches for consideration of green roof and green walls in dispersion models at different spatial scales.

Research highlights.

• GI could reduce the exposure through deposition and pollutants redistribution at microscale.

• Enhanced atmospheric turbulence and deposition are key processes at macroscale.

• Deposition process is different at microscale and macroscale simulations.

• Presence of GI can alter air pollutants concentration and bring positive health benefits.

• Lack of input data makes consideration of GI in dispersion models challenging.