Science of the Total Environment 461–462 (2013) 541–551

Contents lists available at SciVerse ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

CFD modelling of the aerodynamic effect of trees on urban airpollution dispersion

J.H. Amorim ⁎, V. Rodrigues, R. Tavares, J. Valente, C. BorregoCESAM & Department of Environment and Planning, University of Aveiro, 3810-193 Aveiro, Portugal

H I G H L I G H T S

• The effect of roadside trees on air quality is assessed for 2 European urban areas.• The analysis is targeted towards the air quality impacts at the pedestrian level.• A vegetative canopy model was coupled to a high detail CFD model.• Modelling accuracy is increased by considering the aerodynamic effects of trees.• Urban air quality can be optimised based on knowledge-based planning of green spaces.

⁎ Corresponding author. Tel.: +351 234 370 200; faxE-mail addresses: amorim@ua.pt (J.H. Amorim), vera

(V. Rodrigues), richard.tavares@ua.pt (R. Tavares), joancborrego@ua.pt (C. Borrego).

0048-9697/$ – see front matter © 2013 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.scitotenv.2013.05.031

a b s t r a c t

a r t i c l e i n f o

Article history:Received 4 December 2012Received in revised form 24 April 2013Accepted 13 May 2013Available online 8 June 2013

Editor: Xuexi Tie

Keywords:Street canyonUrban treesCO dispersionCFD modellingTraffic emissions

The current work evaluates the impact of urban trees over the dispersion of carbon monoxide (CO) emittedby road traffic, due to the induced modification of the wind flow characteristics. With this purpose, the stan-dard flow equations with a kε closure for turbulence were extended with the capability to account for theaerodynamic effect of trees over the wind field. Two CFD models were used for testing this numerical ap-proach. Air quality simulations were conducted for two periods of 31 h in selected areas of Lisbon and Aveiro,in Portugal, for distinct relative wind directions: approximately 45° and nearly parallel to the main avenue,respectively. The statistical evaluation of modelling performance and uncertainty revealed a significant im-provement of results with trees, as shown by the reduction of the NMSE from 0.14 to 0.10 in Lisbon, andfrom 0.14 to 0.04 in Aveiro, which is independent from the CFD model applied. The consideration of theplant canopy allowed to fulfil the data quality objectives for ambient air quality modelling established bythe Directive 2008/50/EC, with an important decrease of the maximum deviation between site measure-ments and CFD results. In the non-aligned wind situation an average 12% increase of the CO concentrationsin the domain was observed as a response to the aerodynamic action of trees over the vertical exchangerates of polluted air with the above roof-level atmosphere; while for the aligned configuration an average16% decrease was registered due to the enhanced ventilation of the street canyon. These results show thaturban air quality can be optimised based on knowledge-based planning of green spaces.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

Biosphere–atmosphere interactions, and in particular the behaviourof flows over and through plant canopies, have been subject of researchat different spatial scales and in numerous fields such as hydrology, ecol-ogy, climate and various engineering branches (Katul et al., 2004). Inwhat relates to the impact of trees on urban environments the analysishas been performed at different levels.While some studies have been fo-cused on the air pollutant removal capacity of trees by deposition and

: +351 234 370 309..moreira.rodrigues@gmail.comavalente@ua.pt (J. Valente),

rights reserved.

filtration mechanisms (Nowak et al., 2006; Tallis et al., 2011), othersanalysed the opposite effect in which trees are the reason for higher pol-lution levels due to different reasons (Gromke and Ruck, 2007; Ribeiro etal., 2009). The comprehensive review on this topic by Litschke andKuttler (2008) showed that accurate measurements of deposition veloc-ities of urban vegetation are still needed if the objective is to estimate theimprovement on air quality as a result of the filtration by plants of roadtraffic particle emissions. Numerical models (e.g., the UFORE modelfromNowak and Crane (1998)) have been applied to assess the potentialremoval of air pollutants induced by urban green spaces for current con-ditions and future scenarios (Nowak et al., 2006; Tallis et al., 2011). On theother hand, urban green canopies are also responsible for the emission ofairborne pollen antigenic particles (Pehkonen and Rantio-Lehtimaki,1994; Ribeiro et al., 2009) and biogenic volatile organic compounds

Table 1Closure constants for the k-ε model and URVE module.

cμ σk σε cd cε1 cε2 cε4 cε5 βp βd

0.09 1 1.3 0.2 1.44 1.92 1.5 1.5 1 4

542 J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

(Benjamin et al., 1996;Owenet al., 2003), the latter being relatedwith theozone-forming potential of vegetation (Benjamin and Winer, 1998). Theextent to which road traffic is capable of influencing the air qualityof neighbouring urban green areas has been addressed (Kuttler andStrassburger, 1999; Upmanis et al., 2001), as also the consequent bi-ological response of roadside trees (Li et al., 2010). For further readingsee the review by Tiwary and Colls (2010) on the role of vegetation inthe mitigation of air pollution.

In the basis of the current understanding of the turbulent flowwithin and around vegetation canopies, there is an extensive numberof outdoor measurements on natural canopies and wind tunnel experi-ments using artificial models. Wind tunnel measurements on artificialcanopies have played an important role for decades in the study of with-in canopy turbulence structure (see the reviews of Finnigan, 2000;Raupach and Thom, 1981). From the first model forests, applying woodpegs and plastic strips (Plate and Quraishi, 1965), to the recent extensivedatabase on idealised street canyon urban trees with varying permeabil-ity (Gromke and Ruck, 2007, 2009), much has been understood aboutthe aerodynamic effects of trees on ground level pollutant concentra-tions in urban-like geometries. Field campaigns have been carried out re-cently in urban streetswith the aimof investigating the effects of trees onflowpatterns, turbulent diffusion of airborne pollutants and temperaturedistribution (see Kikuchi et al., 2007).

On the computational field, several vegetation canopy models havebeen proposed and evaluated, which simulate the aerodynamic effectsof trees (Green, 1992; Liu et al., 1996; Hiraoka and Ohashi, 2008;Mochida et al., 2008). These numerical formulations have beenimplemented into Computational Fluid Dynamics (CFD)models and ap-plied to urban-like geometries to improve our understanding of the roleof vegetation on wind flow patterns and air pollutant concentrations(see Balczó et al., 2009; Bruse and Fleer, 1998; Buccolieri et al., 2009;Gromke et al., 2008; Liang et al., 2006; Mochida et al., 2008; Ries andEichhorn, 2001; Robitu et al., 2006). Most of these applications rely onkε turbulencemodels, except for Buccolieri et al. (2009) inwhich turbu-lence closure is obtained with a Reynolds stress model. These two tech-niqueswere evaluated byGromke et al. (2008) through the comparisonwith wind-tunnel data.

Although the effects of plant canopies on atmospheric turbulencehave been widely studied, a scientifically-based knowledge of the over-all impact of trees on urban environments is still required, particularlyin terms of the outcomes on air quality. Because themajority of numer-ical and experimental studies have been performed using idealised con-figurations and hypothetical scenarios, an open question still stands onthe overall understanding of the extent of the perturbations induced bytrees on air pollutants dispersion in real urban environments. In thisscope, the current research is focused on (i) the numerical modellingof the aerodynamic effects induced on local wind and turbulence fieldsby the mechanical drag of trees, and (ii) the related impacts on the dis-persion of traffic-emitted air pollutants for typical European street can-yon configurations. In the core of the work is the development of theURban VEgetation (URVE) module, which allows the simulation toaccount for the aerodynamic effects of trees over the 3D wind field.This module was coupled to two distinct CFD models and applied inthe simulation of the spatial distribution of carbon monoxide (CO)concentration in two Portuguese city centres.

In conclusion, this study aims to contribute to a better under-standing of the flow and dispersion of gaseous pollutants in urbanenvironments when accounting for the aerodynamic effects oftrees.

2. Urban vegetative canopy model development

The main concept behind the simulation of the aerodynamic effectof urban vegetation is the extension of the standard mean flow andturbulence equations with additional source terms for momentum,turbulent kinetic energy (k) and its dissipation (ε) that mathematically

represent the aerodynamics behind the interaction of leaves andbranches with the 3D wind flow. Consequently, the dispersion of theemitted air pollutants is conditioned by vegetation through this dis-turbed wind flow. The magnitude of this perturbation depends mostof all on the characteristics of the vegetation itself (e.g., size, porosity,location) and of the incoming air flow (e.g., velocity, direction,turbulence).

The extension of flow equations is made through the developednumerical module URVE, which is coupled to a specific solver thatprovides the numerical solution for the mean flow and turbulence, asdescribed in Section 2.2.

2.1. Numerical formulation

URVE code was designed to be coupled with a CFD model, basedon the application of the Reynolds-averaged Navier–Stokes (RANS)equations. Simplifications can be made in the general form of theequations that describe the 3D flow through the conservation of mass,momentum and energy, namely by the consideration of a steady-stateflow of a Newtonian, isotropic and incompressible fluid, in an inertial(non-accelerating) reference frame and neglecting the Coriolis force.Also, the Boussinesq approximation is applied removing the air densityfrom the calculation. The objective of this study was to capture the dy-namic turbulence induced by trees and, therefore, neutrally buoyantconditions were assumed.

Given the assumptions considered, the mass conservation equa-tion is expressed, using Einstein's summation notation (i and j are in-dices with values of 1, 2 and 3), as following:

∂ui

∂xi¼ 0 ð1Þ

where ui is the velocity and xi is the spatial coordinate.The loss of wind speed due to pressure and viscous drag forces

exerted by the leaves and branches can be expressed through a mo-mentum sink term (Fd) that is added to the momentum conservationequation, which is described by:

uj∂ui

∂xj¼ −1

ρ∂P∂xi

þ ∂∂xj

νe∂ui

∂xj

!þ Fd ð2Þ

where P is the pressure, ρ the air density, νe the effective viscosity, inwhich the turbulent (or eddy) viscosity, μt, is computed by combiningk and ε as follows:

νe ¼ ν þ νt ¼ ν þ cμk2

ε: ð3Þ

The turbulence closure is achieved through the calculation of k and εthat will be given in Eqs. (6) and (7). The value of cμ is also shown inTable 1. Fd is calculated according to the conventional parameterisationof the plant–airflow interaction expressed by Eq. (4), which neglectsviscous drag relative to form (or pressure) drag (Green, 1992; Katul etal., 2004; Li et al., 1990; Wilson, 1988).

Fd ¼ −cd⋅a⋅ Uj j⋅ui ð4Þ

where cd is themechanical drag coefficient, and a stands for the leaf areadensity (LAD), which is defined as the total one-sided leaf area of

Table 2Main characteristics of FLUENT and VADIS models.

FLUENT® VADIS

General characteristicsMulti-purpose commercial CFD model(v. 6.1.18 for UNIX platforms)

Microscale air quality CFD modeldeveloped at the University of Aveiro

Flow modellingUnstructured meshing scheme Structured meshing schemeEulerian approachSteady-state flow (1h averaged input/output values)RANS equationsTurbulence closure: k-ε model

Dispersion modellingEulerian approach Lagrangian approachNo chemical reactions

543J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

photosynthetic tissue per unit of canopy volume (Weiss et al., 2004).The magnitude of the wind speed vector, U, is calculated as follows:

Uj j ¼ ∑3i¼1u

2i

� �1=2 : ð5Þ

The turbulence closure is obtained through the application of thekε model (Launder and Spalding, 1974), which has been the mostwidely used turbulence model in wind engineering and near-field dis-persion problems (Hargreaves andWright, 2007) and has also been test-ed in other vegetation canopymodels (Mochida et al., 2008), as reviewedin Section 1. The kε turbulence model solves two additional transportequations, for k and ε. The values for the empirical constants of the kεmodel, presented in Table 1 are the ones originally defined by Launderand Spalding (1974).

The budget equations for k and ε are given by Eqs. (6) and (7), re-spectively. Similar to the budget equation formomentum, the turbulentinteraction between the airflow and the plant canopy is addressed byincluding the additional source terms Sk and Sε in the transport equa-tions of k and ε, respectively (Green, 1992; Wilson, 1988).

uj∂k∂xj

¼ ∂∂xi

νt

σk

∂k∂xi

� �þ νt

∂ui

∂xjþ ∂uj

∂xi

!∂ui

∂xj−ε þ Sk: ð6Þ

uj∂ε∂xj

¼ ∂∂xi

νt

σε

∂ε∂xi

� �þ cε1

εkνt

∂ui

∂xjþ ∂uj

∂xi

!∂ui

∂xj−cε2

ε2

kþ Sε : ð7Þ

The values for the closure constants cε1 and cε2, and the Prandtlnumbers for k and ε, σk and σε, are also shown in Table 1. The sourceterm Sk is given by (Green, 1992; Wilson, 1988):

Sk ¼ cda βp Uj j3−βd Uj jk� �

: ð8Þ

The values of βp and βd are shown in Table 1. Sk reflects the gener-ation of turbulence by the action of vegetation elements, and therebyβp represents the fraction of the mean flow kinetic energy convertedto wake-generated k by canopy drag. The second term in (8), on the con-trary, expresses the rapid dissipation of the generated wakes (Raupachand Shaw, 1982), and consequently βd is the fraction of k dissipated by“short-circuiting” of the Kolmogorov cascade (Kaimal and Finnigan,1994; Poggi et al., 2004). In this sense, Sk is a sum of the source and sinkof k due to the effect of the vegetative canopy.

The source term of ε, Sε, which is similar to the formulation of Sk, isgiven by (Green, 1992):

Sε ¼ cda cε4βp Uj j3 εk−cε5βd Uj jε

� �: ð9Þ

The values for the constants representing the vegetative canopyelements are case specific and thus will be presented and discussedin Section 3.2.2.3.

2.2. Module coupling

In order to obtain a solution for the 3D wind flow in the study do-main, the URVE module needs to be coupled to a CFD solver. Two dis-tinct CFD codes were used (both coupled and uncoupled with URVE)to evaluate the performance of this module under distinct conditions(e.g. weather) and urban configurations. Table 2 shows the main fea-tures of both models.

The first model is the multi-purpose commercial softwareFLUENT® (Inc, 2003), version 6.1.18 for UNIX platforms, which hasbeen applied in the simulation of flow and dispersion of fluids andparticles within confined and open complex geometries in differentscientific and technical domains. Its adequacy and feasibility to

urban air quality modelling was previously assessed (Borrego et al.,2003; Martins et al., 2009).

The other model is VADIS (Borrego et al., 2003), a fit-to-purposemodel specifically developed for the air quality simulation at microscale,which has been previously applied to European urban areas (Borrego etal., 2003; Borrego et al., 2004; Tchepel et al., 2010), as also in theintercomparison against experiments and other models (Sabatino et al.,2011).

In bothmodels, the 3Dwindflowwas simulated by applying a RANSprognostic model with a standard kε turbulence closure. For the CO dis-persion modelling, FLUENT applies an Eulerian approach while VADISuses a Lagrangian one. Due to the limited time and space scales of theanalysis CO was considered as a non-reactive chemical species in bothapproaches, as in previous modelling investigations (Crowther andHassan, 2002).

Fig. 1 schematically represents the methodological approach forthe numerical simulation of the aerodynamic effect of vegetation onwind flow patterns and CO dispersion.

Because these two CFD codes run in different programming lan-guages, URVE was written in C and in Fortran 90, respectively for thecoupling with FLUENT and VADIS.

3. Model application

The main goal of this work was to evaluate the interaction betweentrees and the dispersion of road traffic-emittedCO in a real 3Durban en-vironment with typical meteorological conditions. Two study domainswere defined in two Portuguese urban centres: the first in the country'scapital Lisbon, and the second in the medium-size town of Aveiro. Themain criteria for the selection of the computational domains were thefollowing: (i) the presence of a large number of densely foliaged talltrees flanking the main avenue, (ii) a significant road traffic flux, and(iii) the existence of an air quality station (AQS) inside the study area.

Air quality simulations were conducted for the two city-cases for aperiod of 31 h each, according to the averaging period for the calcula-tion of the daily eight hourmeanCOconcentration defined by theDirec-tive 2008/50/EC. Based on the prevailingmeteorological conditions andthe availability of traffic counting, Lisbon simulations comprise the pe-riod between 5 p.m. on March 5th and 12 p.m. on March 6th of 2002,while Aveiro simulations are from 5 p.m. on May 4th to 12 p.m. onMay 5th 2004.

3.1. Case study description

3.1.1. Study areasThe two defined computational domains consist of residential, com-

mercial and recreational areas. They both comprise important trafficthoroughfares and are characterised by the presence of a complexarray of buildings and irregular streets (see Fig. 2).

CFD model (FLUENT or VADIS)

+URVE moduleM

OD

ELS

INP

UT

OU

TP

UT

VA

LID

AT

ION

meteorology 3D mapping traffic emissions

3D wind field 3D concentration field

meteorological stationin the domain

air quality station in the domain

wind velocity and direction measured at a nearby meteo. station

3D coordinates of urbanelements from satelliteimagery and local obs.

estimated by the Transport Emission Model for Line

Sources TREM

Fig. 1. Schematic representation of the methodological approach.

544 J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

A brief description of the study area is given in Table 3. The differentcharacteristics of the canyon geometries in the two urban areas inducedistinct flow behaviours. In Lisbon the aspect ratio (representing therate between the height H of the canyon and its width W) indicates awide canyon where the buildings on both sides of the emission sourceare well spaced. In Aveiro, on the other hand, the buildings are moreclosely spaced. Both can be classified as symmetric canyons, i.e. thebuildings flanking the street have the same average height. As expected,in the narrow street canyon of the Aveiro study case the fraction of thecross-sectional area occupied by trees is higher than in the wider can-yon of Lisbon.

As can be seen in Fig. 2 and, in more detail, in Fig. 3, one of the mostrelevant characteristics of the study area is the presence of a large num-ber of densely foliaged tall trees that flank the emission sources alongtheir entire extension. Moreover, both AQS are located under the direct

Fig. 2. Satellite images of the Lisbon (a) and Aveiro (b) study areas (at the time of measuremand the location of the AQS is indicated by the white triangle.

influence of this dense canopy. Also, the probes are at approximately 2to 3 m from the leaves in the inferior boundary of the tree crowns.

The tree species in Lisbon are Platanus hybrida and Celtis australis,while in Aveiro they are Acer pseudoplatanus and Quercus robur. Theirdimensions will be listed in Table 5.

3.1.2. Meteorological conditionsThe wind velocity and direction values shown in Fig. 4 were ac-

quired at 10 m high meteorological masts located nearby the compu-tational domains. These data describe the mean wind flow conditionsthat will be used as inflow boundary conditions for the simulations(see Section 3.2.2.1). It can be observed from the Lisbon city casethat the wind direction was approximately constant from Northeast,indicating that there is a 45° deviation between the incoming flowand themain street canyon. Significant oscillations in thewind intensity

ents). The boundaries of the computational domains are shown by the white rectangle

Table 3General description of the study areas (Ld and Wd represent, respectively, the lengthand width of the built-up area considered in the computational domain, while H andW stand for the height and width of the buildings in the street canyon).

Studycase

Ld × Wd

(m × m)Average H/Wratio

Sectional area occupiedby trees (%)

Lisbon 550 × 530 0.33 5.4Aveiro 395 × 240 0.75 27.3

Time (h)16 20 0 4 8 12 16 20 0

Win

d d

irec

tio

n

S

W

N

E

S

Day5/3/2002 6/3/2002 6/3/2002 6-3-2002 6-3-2002 6-3-2002

Win

d V

elo

city

(m

.s-1

)

4

6

8

10

Wind directionWind velocity

Time (h)16 20 0 4 8 12 16 20 0

Day

4/5/2004 5/5/2004 5/5/2004 5/5/2004 5/5/2004 5/5/2004

Wind directionWind velocity

a

b

Fig. 4. Time evolution of mean hourly values of wind velocity (m·s−1) and direction atthe inlet boundaries in Lisbon (a) and Aveiro (b).

545J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

were observed during the study period, with a maximum variationbetween 4 and 7 m·s−1. In Aveiro, the wind velocity varied between5 and 10 m·s−1, while the wind direction was approximately fromNorthwest during the entire period.

The periods selected correspond to neutral stability conditions, inagreement with the numerical approach adopted in this study.

3.1.3. Road trafficThe main road in the selected Lisbon domain is the Liberdade Av.,

which is constituted by five central lanes and by two secondary lanesseparated from the main avenue with sidewalks. Ten other streetswere also considered, among which two also have significant trafficflux rates. In the Aveiro domain, the main emission source is the 25 deAbril Av., with a total of four traffic lanes. Another two avenues andseven streets were also considered. There is also a road tunnel underone of the avenues, with natural ventilation of traffic emissions by 15shafts (for the road segment length considered).

Vehicle fluxes data were acquired using automatic devices in Lisbonand manual counting in Aveiro. For the roads in which no data wasavailable, empirical rates expressing the relation with the traffic in thesurrounding roads were applied.

3.1.4. Air qualityHistorical data of CO concentrations were sourced from two ‘urban

traffic’ air quality monitoring stations of the national Portuguese net-work (APA, 2011). Additional information is given in Table 4.

For the Lisbon study case, themaximum8-h averageCO concentrationwas 1063 μg·m−3, corresponding to the period from 2 p.m. to 10 p.m.,which is considerably inferior to the CO limit-value of 10,000 μg·m−3.For the Aveiro study case, the maximum value for the 8-h periods was493.9 μg·m−3, from 7 a.m. to 3 p.m., which is also significantly lowerthan the limit-value. Hourly CO concentrations measured in the AQSwill be presented in Fig. 7 in comparison with the simulated values.

3.2. Modelling setup

The numerical simulations were conducted following the bestpractice guidelines proposed by COST Action 732 (Franke et al., 2007).

Fig. 3. Photographs of the Liberdade Avenue in Lisbon (a) and the 25 d

Simulations were performed in an hourly temporal frame assumingsteady-state conditions. Hourly averaged wind velocity and CO emis-sions were used as input for the calculations.

In order to evaluate the effects induced by the vegetativecanopy, simulations were performed for baseline conditions (URVEmodule activated), and for a scenario without trees (URVE moduledeactivated).

e Abril Avenue in Aveiro (b), both flanked by rows of dense trees.

Table 4Characterisation of the AQS located in the study domains.

AQSlocation

Geographicalcoordinates

Height abovesea level (m)

Distance fromnearest road(m)

Distance fromnearest building(m)

Liberdade av.(Lisbon)

38°43′16″N9°08′46″W

44 11.5 24.5

25 de Abril av.(Aveiro)

40°38′14″N8°38′53″W

20 6 18

546 J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

3.2.1. Computational domain and gridFollowing the recommendations fromCOST Action 732 (Franke et al.,

2007) for the simulation of urban flows with multiple buildings, thevertical, lateral and downwind extensions of the computational domainwere defined with a minimum of 5Hmax, where Hmax represents theheight of the tallest building.

The complexity of the city objects (buildings and trees) present inboth domains was reduced by assembling adjacent individual volumeswith similar characteristics. Specifically in the case of trees, the groupedelements were defined as parallelepipeds positioned at a given distanceabove ground, representing the average trunk height. Fig. 5 shows theresulting virtual buildings and trees created.

The generation of the city objects (porous and non-porous) and themeshing of the computational domain followed different procedures

Fig. 5. Computational domain generated by the models for the set of buildin

according to the CFDmodel. In VADIS an internal pre-processor createdthe obstacles based on the coordinates given by the user. The modelused a structured meshing scheme for the spatial discretisation of thecomputational domain. In FLUENT the CAD software Gambit® (version2.0.4) was used to build the objects and create an unstructured compu-tational grid (TGrid meshing tool), with variable cell sizes and shapes,that adjusts to more complex geometries. See Table 5 for more details.

3.2.2. Initial and boundary conditions

3.2.2.1. Inlet and outlet boundaries. Inflow boundary conditions for thehorizontal wind velocity components, u and v, were prescribed basedon the meteorological data presented in Section 3.1.2. Richards andHoxey's (1993) vertical profile equations were used to specify thevariation of velocity (U), k and ε with height at the inlet boundariesassuming neutral stability conditions:

U ¼ u�kv

lnzþ z0z0

� �; k ¼ u2

�ffiffiffiffifficμ

p ; ε ¼ u3�

kv zþ z0ð Þ ð10Þ

where u* is the friction velocity, extracted from the logarithmic profileof Eq. (10) with the wind velocity measured at the reference height of10 m, kv is von Karman's constant (0.41) and z0 is the surface roughness

gs (in dark grey) and trees (in light grey) in Lisbon (a) and Aveiro (b).

Table 5General description of the computational domains (Ld and Wd represent, respectively, the total length and width of the domain, including the open belt around the built-up area.Symbol # indicates number).

Study case Domain area Mesh Buildings Trees Roads

Ld × Wd (m × m) Type Resolution (m3) #Cells

#Sets

Height range (m) # Sets Min. crown height (m), Max. height (m) # Width range (m)

Lisbon 700 × 650 Unstruc. 0.08–1000 560,000 42 12–40 6 5 15 13 4–25Aveiro 800 × 800 Regular 1000 38,400 40 3–15 29 5 10 12 4–20

547J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

length, defined as 1.1 m, a typical average value for a urban area withmedium height and density (Grimmond and Oke, 1999).

3.2.2.2. Upper, lower and wall boundaries. In order to prevent a horizon-tal change from the inflow profiles along the domain, the values for thevelocities and the turbulence quantities given by the inflow profile forthe top of the computational domains were prescribed over the entiretop boundary, as suggested by Blocken et al. (2007) and Franke et al.(2007).

At ground and building surfaces a no-slip conditionwas imposed. Thestandard wall functions proposed by Launder and Spalding (1974) wereused. In the near-wall region the logarithmic law-of-the-wall for themean velocity was applied, while the production of k, and its dissipationrate, ε, at thewall-adjacentwere computed on the basis of the local equi-librium hypothesis. Wall roughness effects were modelled applying thelaw-of-the-wall modified for roughness. The roughness height parame-ter, Ks, was specified as 0.05 m, and the roughness constant, Cs, as 0.5 m.At walls, a zero-gradient boundary condition was applied for CO.

3.2.2.3. Tree boundary conditions. The assembled tree volumeswere de-fined in themodels asfluid zones, inwhich all transport equationsweresolved. The source of momentum and turbulence was defined in URVE.

The different parameters used in the definition of trees in Eqs. (4),(8) and (9) were defined in Table 1. The values for cε4 and cε5 werethe ones originally defined by Green (1992). For the mechanical dragcoefficient, cd, which typically ranges between 0.1 and 0.3 for most veg-etation, an average value of 0.2 was selected (Bruse and Fleer, 1998;Katul et al., 2004; Liang et al., 2006).

The spatial distribution of LAD (in this work given by a) is a keyparameter in describing the forest canopy characteristics (Lalic andMihailovic, 2004), but due to the difficulty and complexity inherentto the measurement of the vertical distribution of forest canopy ele-ments (see Law et al., 2001; Meir et al., 2000; Wang et al., 1992), espe-cially in a full scale analysis, few data are available, especially for trees inEuropean urban areas. Moreover, LAD values vary with respect to spe-cies, site fertility, time of day and year, weather conditions and evenwithin stands (Bréda, 2003). However, it can be extracted from theanalysis of Lalic and Mihailovic's (2004) work, that the average LAD in-side the crown of a large size tree is approximately 1 m2·m−3, with nosignificant variation with height. Consequently, the following simplevertical profilewas assumed for the type of vegetation existing in Lisbonand Aveiro:

a ¼ 0 0≤hbhc1 hc≤h≤ht

�ð11Þ

where h is the height above ground, hc is the lower average height of thecrown, and ht is the maximum average height of the tree.

3.2.2.4. CO source/sink. The roads defined in the computation were theones described in Section 3.1.3. All these emission sources were createdas fluid zones, in this case with amass source allocated. A numerical ap-proach to the traffic produced turbulence was not considered; instead,the referred fluid volumes were defined with a height of 3 m, and themass source was homogeneously distributed by the total number of

cells contained in these fluid zones, inducing a forced mixture of COwith the ambient air. Ventilation shafts were also considered for theAveiro simulation. In this case, the emission of each individual shaft is as-sumed to correspond to 1/15 the total emission in that segment length ofthe road tunnel.

Hourly averaged CO flow rates were prescribed for each road apply-ing the Transport EmissionModel for Line Sources (TREM), based on ve-hicles counting data (described in Section 3.1.3), and according to themodel cascade methodology validated by Borrego et al. (2003). TREMis based on the MEET/COST methodology. Emission factors are derivedfrom the average speed (approximately 50 km·h−1 in the consideredroads), which is an approach with good results when the influence ofdriving dynamics can be neglected (Sturm et al., 1998). The emission(ECO) is estimated for each road segment as follows:

ECO ¼ ∑i

eCO;i vð Þ � Ni

� �� L ð12Þ

where eCO,i(v) is the emission factor for CO and vehicle class i as a func-tion of average speed v; Ni is the number of vehicles of class i; and L isthe road segment length. For the selection of appropriate emission fac-tors, eCO,i(v) valueswere defined based on the aggregation of vehicles inthe following categories: passenger cars (gasoline and diesel), lightduty vehicles, heavy duty vehicles and urban busses. For each vehiclecategory, different classes were considered that distinguish engine age,type, and capacity, vehicle weight, fuel type (gasoline, diesel, LPG), andemission reduction technology (emission standards implementation as-sociated to vehicle age).

The CO removal by deposition processes occurring in the canopywas neglected, based on the fact that the capacity of urban trees to fil-trate CO is in the range from only 0.001 to 0.002% of its concentrationin ambient air (Nowak et al., 2006).

4. Modelling results analysis

The analysis of the CO emission values presented in Fig. 6 showsthe typical behaviour of downtown's traffic flow, with two distinctpeaks in the early morning and afternoon. In the Liberdade Av., Lisbon,traffic emissions varied between a minimum of 2.2 kg·km−1·h−1 at4 a.m., corresponding to a total of 500 vehicles·h−1, and a maximumof 24.5 kg·km−1·h−1 at 6 p.m., for a total of 5600 vehicles·h−1.Emissions start to increase at 6 a.m. and are kept high until 6 p.m.when they decrease, which is a typical behaviour of the traffic flowfor a large city downtown. In Aveiro the minimum emission was0.09 kg·km−1·h−1 for 14 vehicles·h−1 at 4 a.m., and the maximumwas 4.7 kg·km−1·h−1 for 1038 vehicles·h−1 at 7 p.m

In Fig. 7, the temporal variation of the mean hourly values ofmeasured/modelled CO concentrations is presented for both studycases. The data comparison reveals a good agreement between mea-sured and simulated concentrations over the entire period in both cities.Although in some periods there is an overestimating tendency whenURVE is used, in particular for Lisbon, there is a general increase ofmodelling performance with the coupling of the vegetation effect.

The existence of a meteorological station located inside the Aveirodomain allowed the evaluation of wind velocity simulations for this

Fig. 6. Time evolution of mean hourly values of CO emission (kg·km−1·h−1) estimatedby TREM for Lisbon (a) and Aveiro (b).

Time (h)0 4 8 12 16 20 0

Win

d v

elo

city

(m

.s-1

)

0

2

4

6

8

10

12

14

Meteo. station VADIS+URVE VADIS

Fig. 8. Time evolution of the mean hourly values of wind velocity (m·s−1) measured inthe meteorological station and simulated for the same location with and without theinclusion of URVE.

548 J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

study case. Results are shown in Fig. 8. As can be seen there is a verysignificant increase of the modelling performance with URVE.

The performance and accuracy of the model were evaluated by ap-plying the BOOT software and the model acceptance criteria definedby Chang and Hanna (2005) for air quality models assessment. Thefollowing parameters were calculated: average bias (d), normalisedmean squared error (NMSE), Pearson correlation coefficient (r), andfactor of two (FAC2). Results are shown in Table 6.

In all the cases, the statistical analysis indicated a better modellingperformance when the vegetative canopy was considered. The NMSEwas reduced from 0.14 to 0.10 in the Lisbon case study, and from 0.14

Time (h)16 20 0 4 8 12 16 20 0

CO

Co

nce

ntr

atio

n (

µg

.m-3

)

0

200

400

600

800

1000

1200

1400

1600

1800

Measured (AQS)

FLUENT+URVE

FLUENT

Time (h)16 20 0 4 8 12 16 20 0

0

100

200

300

400

500

600

700

Measured (AQS)

VADIS+URVE

VADIS

a

b

Fig. 7. Time evolution of the mean hourly values of CO concentration (μg·m−3) mea-sured in the AQS and simulated for the same locations with and without the inclusionof URVE, in Lisbon (a) and Aveiro (b).

to 0.04 in Aveiro. Nevertheless, the NMSE for all the simulations wassignificantly lower than the maximum of 1.5 defined by the acceptancecriteria of Chang and Hanna (2005). The r parameter also increaseswith the URVE module indicating a better correlation betweenmodelled and measured data. The negative value of d indicates someoverestimation when the effect of vegetation is considered. However,the underestimating tendency exhibited by themodel without the cou-pling with URVE is more pronounced.

Modelling uncertainty was assessed through the data quality objec-tives established by the Directive 2008/50/EC. According to the resultsobtained (also shown in Table 6), the maximum deviation of 50% wassurpassed when using the original versions of both FLUENT (59.70%)and VADIS (59.72%). This parameter decreased to 43.65% and 18.89%,respectively, with the consideration of the plant canopy, allowing to ful-fil the requirements established by the legislation.

Another conclusion that can be taken from Fig. 7 is that, in thegreat majority of the hourly simulations, the effects induced by theplant canopy on the wind flow lead to an increase of CO concentrationsat the location of both AQS. Themagnitude of this effect is mostly depen-dent on the orientation of the incoming wind in relation to the position-ing of the emission sources and buildings. According to the simulationresults there is an average increase on concentrations at the AQS of 35%in Lisbon, while in Aveiro the values are, in average, 175% higher withURVE. It can be observed from Figs. 9 and 10 that the effect of trees onair quality is extremely spatially dependent, mostly because the hetero-geneous positioning of trees induces complex wind flow patterns. It canbe inferred that, in fact, the Aveiro AQS is not representative of the overallimpact of trees on dispersion conditions in the studied area for the simu-lated meteorological conditions. It was concluded that in Lisbon, wherethewind direction is not alignedwith the street canyon, the average con-centration in the domain is 12%higher because of the aerodynamic actionof trees, while in Aveiro, where there is an approximate alignment, it is16% lower.

Fig. 9 shows the CO concentration field in Lisbon (for more detail,only a selected area is shown). As a result of the conjoint influence ofthe 3D configuration of trees and buildings over the wind flow an in-crease of concentrations is observed along the Liberdade Av., in partic-ular in the leeward side (buildings on the right side of the image) ofthe street canyon.

This behaviour is originated by the oblique roof-level incomingwinds that induce a counter-clockwise swirling flow along the canyon.As a result, this vortical airflow transports the pollutant emitted nearthe ground level by traffic towards the leeward side of the street canyon(through the open space under the trees crowns), where the CO istrapped by the decreased vertical exchange rate of air with the aboveroof-level atmosphere.

Table 6Statistical parameters for the assessment of modelling performance.

Statistical metrics Model accept-ance criteria CO concentration simulation Wind velocity simulation

Lisbon Aveiro Aveiro

FLUENT FLUENT + URVE VADIS VADIS + URVE VADIS VADIS + URVE

d – 137.13 μg·m−3 −61.32 μg·m−3 84.42 μg·m−3 −29.03 μg·m−3 −7.69 m·s−1 −0.59 m·s−1

NMSE b1.5 0.14 0.10 0.14 0.04 1.60 0.29r – 0.76 0.82 0.78 0.86 0.57 −0.04FAC2 >0.5 0.81 0.90 0.84 1.00 0.00 0.79Uncertainty (%) b50 59.70 43.65 59.72 18.89 – –

549J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

On the contrary, the alignment of trees with the incoming wind inthe Aveiro study case was shown to enhance ventilation, promotingCO dispersion inside the avenue and adjacent buildings, which is evidentin the comparative analysis of the concentration contours in Fig. 10. Itcan also be seen that despite the general improvement of air quality, inspecific areas as the AQS location, trees lead to the formation of addition-al hot-spots due to the rearrangement of vortical flow structures.

Therefore, the increase of concentrations at the AQS (Fig. 7) inAveiro is a consequence of (i) air recirculation induced by thebuilding's walls and trees, and (ii) the significant decrease of wind ve-locity at this spot (as already identified in Fig. 8), leading to the for-mation of a hot-spot in this particular zone.

Fig. 9. Close-up image of 3 m high horizontal streamlines and CO concentration field with10 a.m. Unfilled rectangles indicate trees blocks and the white triangle is the AQS location.

Fig. 10. Close-up image of 3 m high horizontal streamlines and CO concentration field with11 a.m. Unfilled rectangles indicate trees blocks and the white triangle is the AQS location.

5. Conclusions

In both cases, the coupling of the developed vegetative canopymodule URVEwith two CFD tools increased the accuracy of simulationscompared to the original codes. The Directive 2008/50/EC establishesfor ambient air quality assessment a maximum modelling uncertaintyof 50%. This threshold was surpassed by 10% in both cities when notconsidering the effect of trees. However, the uncertainty decreased to44% and 19%, respectively for Lisbon and Aveiro, with the considerationof the plant canopy implemented in URVE. A consistent increase ofmodelling performance was obtained also in the analysis of the NMSE,which was reduced by 29% (from 0.14 to 0.10) in Lisbon, and by 71%

and without the effect of trees in Lisbon. Contours refer to the period between 9 and

and without the effect of trees in Aveiro. Contours refer to the period between 10 and

550 J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

(from 0.14 to 0.04) in Aveiro. These values are substantially lower thanthe threshold of 1.5 defined by modelling acceptance criteria.

It was concluded that for an incoming wind direction of approxi-mately 45°with themain avenue (Lisbon study case) CO concentrationswere found to increase by an average of 12% due to the aerodynamic ac-tion of trees over the exchange rates of polluted air with the aboveroof-level atmosphere. On the contrary, for the nearly parallel wind(Aveiro) a general decrease of 16% on the CO concentrations was ob-served in the entire area due to an enhanced ventilation and dispersioncapacity of the street canyon. However, in some spots, as in the locationof the AQS, the concentration increased due to the lowerwind velocitiesand the formation of additional recirculation areas, which can be mis-leading given the overall benefit.

The results demonstrated that local air quality is strongly dependenton the synergies between themeteorological conditions, the 3D config-uration of the street canyon, and the presence of vegetation. The effectof the urban canopy (buildings and trees) on air pollutants dispersionwas shown to be complex and highly spatially dependent. These con-clusions point out to the importance of integrating the knowledge pro-vided by the application of CFDmodels in urban planning with the goalof optimising the role of green areas on human comfort and health.

Acknowledgements

The authors would like to acknowledge the financial support of the3rd European Framework Program and the Portuguese Ministry of Sci-ence, Technology andHigher Education, through the Foundation for Sci-ence and Technology (FCT), for the Post-Doc grants of J.H. Amorim(SFRH/BPD/48121/2008) and J. Valente (SFRH/BPD/78933/2011). Anacknowledgment is also given to J. Santos for his participation in thetraffic counting carried out in Aveiro.

References

APA. Air quality data for portuguese monitoring stations. URL:Lisbon, Portugal: APA;2011 [URL: http://www.qualar.org/ Accessed on November 2011].

Balczó M, Gromke C, Ruck B. Numerical modeling of flow and pollutant dispersion instreet canyons with tree planting. Meteorol Z 2009;18(2):197–206.

Benjamin MT, Winer AM. Estimating the ozone-forming potential of urban trees andshrubs. Atmos Environ 1998;32(1):53–68.

Benjamin MT, Sudol M, Bloch L, Winer AM. Low-emitting urban forests: a taxonomicmethodology for assigning isoprene andmonoterpene emission rates. Atmos Environ1996;30(9):1437–52.

Blocken B, Stathopoulos T, Carmeliet J. CFD simulation of the atmospheric boundarylayer: wall function problems. Atmos Environ 2007;41(2):238–52.

Borrego C, Tchepel O, Costa AM, Amorim JH, Miranda AI. Emission and dispersion model-ling of Lisbon air quality at local scale. Atmos Environ 2003;37(37):5197–205.

Borrego C, Tchepel O, Salmim L, Amorim JH, Costa AM, Janko J. Integrated modeling ofroad traffic emissions: application to lisbon air quality management. Cybern Syst2004;35(5–6):535–48.

Bréda NJJ. Ground‐based measurements of leaf area index: a review of methods, instru-ments and current controversies. J Exp Bot 2003;54(392):2403–17.

Bruse M, Fleer H. Simulating surface–plant–air interactions inside urban environmentswith a three dimensional numerical model. Environ Model Software 1998;13(3–4):373–84.

Buccolieri R, Gromke C, Di Sabatino S, Ruck B. Aerodynamic effects of trees on pollutantconcentration in street canyons. Sci Total Environ 2009;407(19):5247–56.

Chang JC, Hanna SR. Technical descriptions and user's guide for the BOOT statisticalmodel evaluation software package, version 2.0; 200562.

Crowther JM, Hassan AGAA. Three-dimensional numerical simulation of air pollutantdispersion in street canyons. Water Air Soil Pollut Focus 2002;2:279–95.

Finnigan J. Turbulence in plant canopies. Annu Rev Fluid Mech 2000;32(1):519–71.Franke J, Hellsten A, Schlünzen H, Carissimo B. Best practice guideline for the CFD sim-

ulation of flows in the urban environment. COST Action 732, quality assurance andimprovement of microscale meteorological models. COST Office; 200751.

Green S. Modelling turbulent air flow in a stand of widely spaced trees. Journal of Com-putational Fluid Dynamics and its Applications 1992;5(3):294–312.

Grimmond CSB, Oke TR. Aerodynamic properties of urban areas derived from analysisof surface form. J Appl Meteorol 1999;38(9):1262–92.

Gromke C, Ruck B. Influence of trees on the dispersion of pollutants in an urban streetcanyon — experimental investigation of the flow and concentration field. AtmosEnviron 2007;41(16):3287–302.

Gromke C, Ruck B. On the impact of trees on dispersion processes of traffic emissions instreet canyons. Bound-Lay Meteorol 2009;131(1):19–34.

Gromke C, Buccolieri R, Di Sabatino S, Ruck B. Dispersion study in a street canyon withtree planting by means of wind tunnel and numerical investigations — evaluationof CFD data with experimental data. Atmos Environ 2008;42(37):8640–50.

Hargreaves DM, Wright NG. On the use of the k-ε model in commercial CFD software tomodel the neutral atmospheric boundary layer. J Wind Eng Ind Aerodyn 2007;95(5):355–69.

Hiraoka H, Ohashi M. A (k-ε) turbulence closure model for plant canopy flows. J WindEng Ind Aerodyn 2008;96(10–11):2139–49.

Inc F. FLUENT 6.1 user's manual. In: Inc F, editor. Lebanon: Fluent Inc.; 2003. p. 1864.Kaimal JC, Finnigan JJ. Atmospheric boundary layer flows: their structure and measure-

ment. New York: Oxford University Press; 1994289.Katul GG, Mahrt L, Poggi D, Sanz C. One- and two-equation models for canopy turbu-

lence. Bound-Lay Meteorol 2004;113(1):81–109.Kikuchi A, Hataya N, Mochida A, Yoshino H, Tabata Y, Watanabe H, et al. Field study of

the influences of roadside trees and moving automobiles on turbulent diffusion ofair pollutants and thermal environment in urban street canyons. 6th InternationalConference on Indoor Air Quality, Ventilation & Energy Conservation in BuildingsIAQVEC Sendai, Japan; 2007.

Kuttler W, Strassburger A. Air quality measurements in urban green areas — a casestudy. Atmos Environ 1999;33(24–25):4101–8.

Lalic B, Mihailovic DT. An empirical relation describing leaf-area density inside the forestfor environmental modelling. J Appl Meteorol 2004;7:641–5.

Launder BE, Spalding DB. The numerical computation of turbulent flows. ComputMethods Appl Mech Eng 1974;3(2):269–89.

Law BE, Kelliher FM, Baldocchi DD, Anthoni PM, Irvine J, Moore D, et al. Spatial andtemporal variation in respiration in a young ponderosa pine forest during a sum-mer drought. Agr Forest Meteorol 2001;110(1):27–43.

Li Z, Lin JD, Miller DR. Air flow over and through a forest edge: a steady-state numericalsimulation. Bound-Lay Meteorol 1990;51(1):179–97.

Li HE, Li BT, Lan SF. Responses of the urban roadside trees to traffic environment.Int J Environ Technol Manag 2010;12(1):16–26.

Liang L, Xiaofeng L, Borong L, Yingxin Z. Improved k-ε two-equation turbulence modelfor canopy flow. Atmos Environ 2006;40(4):762–70.

Litschke T, Kuttler W. On the reduction of urban particle concentration by vegetation —

a review. Meteorol Z 2008;17(3):229–40.Liu J, Chen JM, Black TA, Novak MD. E-ε modelling of turbulent air flow downwind of a

model forest edge. Bound-Lay Meteorol 1996;77(1):21–44.Martins A, Cerqueira M, Ferreira F, Borrego C, Amorim JH. Lisbon air quality: evaluating

traffic hot-spots. Int J Environ Pollut 2009;39(3):306–20.Meir P, Grace J, Miranda AC. Photographic method to measure the vertical distribution

of leaf area density in forests. Agr Forest Meteorol 2000;102(2–3):105–11.Mochida A, Tabata Y, Iwata T, Yoshino H. Examining tree canopymodels for CFD prediction

of wind environment at pedestrian level. J Wind Eng Ind Aerodyn 2008;96(10–11):1667–77.

Nowak DJ, Crane DE. The Urban Forest Effects (UFORE) model: quantifying urban foreststructure and functions. Proceedings from integrated tools: Boise, Idaho, USA;1998. p. 714–20.

Nowak DJ, Crane DE, Stevens JC. Air pollution removal by urban trees and shrubs in theUnited States. Urban For Urban Green 2006;4(3–4):115–23.

Owen SM, MacKenzie AR, Stewart H, Donovan R, Hewitt CN. Biogenic volatile organic com-pound (VOC) emission estimates from an urban tree canopy. Ecol Appl 2003;13(4):927–38.

Pehkonen E, Rantio-Lehtimaki A. Variations in airborne pollen antigenic particlescaused by meteorologic factors. Allergy 1994;49(6):472–7.

Plate EJ, Quraishi AA. Modeling of velocity distributions inside and above tall crops. J ApplMeteorol 1965;4(3):400–8.

Poggi D, Porporato A, Ridolfi L, Albertson JD, Katul GG. The effect of vegetation densityon canopy sub-layer turbulence. Bound-Lay Meteorol 2004;111(3):565–87.

Raupach MR, Shaw RH. Averaging procedures for flow within vegetation canopies.Bound-Lay Meteorol 1982;22(1):79–90.

Raupach MR, Thom AS. Turbulence in and above plant canopies. Annu Rev Fluid Mech1981;13(1):97–129.

Ribeiro H, Oliveira M, Ribeiro N, Cruz A, Ferreira A, Machado H, et al. Pollen allergenicpotential nature of some trees species: a multidisciplinary approach using aerobi-ological, immunochemical and hospital admissions data. Environ Res 2009;109(3):328–33.

Richards PJ, Hoxey R. Appropriate boundary conditions for computational wind engineeringmodels using the k-ε turbulence model. J Wind Eng Ind Aerodyn 1993;46–47:145–53.

Ries K, Eichhorn J. Simulation of effects of vegetation on the dispersion of pollutants instreet canyons. Meteorol Z 2001;10(4):229–33.

Robitu M, Musy M, Inard C, Groleau D. Modeling the influence of vegetation and waterpond on urban microclimate. Sol Energy 2006;80(4):435–47.

Sabatino SD, Buccolieri R, Olesen HR, Ketzel M, Berkowicz R, Franke J, et al. COST 732 inpractice: the MUST model evaluation exercise. Int J Environ Pollut 2011;44(1):403–18.

Sturm PJ, Boulter P, De Haan P, Joumard R, Hausberger S, Hickmann J, et al. Instantaneousemission data and their use in estimating passenger car emissions, VKM-THD report.Austria: Verlag der Techn. Univ. Graz; 199842.

Tallis M, Taylor G, Sinnett D, Freer-Smith P. Estimating the removal of atmospheric par-ticulate pollution by the urban tree canopy of London, under current and future en-vironments. Landsc Urban Plan 2011;103(2):129–38.

Tchepel O, Costa AM, Martins H, Ferreira J, Monteiro A, Miranda AI, et al. Determinationof background concentrations for air quality models using spectral analysis and fil-tering of monitoring data. Atmos Environ 2010;44(1):106–14.

Tiwary A, Colls J. Air pollution measurement, modelling andmitigation. 3rd ed. Abingdon,UK: Routledge; 2010501.

551J.H. Amorim et al. / Science of the Total Environment 461–462 (2013) 541–551

Upmanis H, Eliasson I, Andersson-Skold Y. Case studies of the spatial variation of ben-zene and toluene concentrations in parks and adjacent built-up areas. Water AirSoil Pollut 2001;129(1/4):61–81.

Wang YS, Miller DR, Welles JM, Heisler GM. Spatial variability of canopy foliage in anoak forest estimated with fisheye sensors. For Sci 1992;38(4):854–65.

Weiss M, Baret F, Smith GJ, Jonckheere I, Coppin P. Review of methods for in situ leafarea index (LAI) determination: part II. Estimation of LAI, errors and sampling.Agr Forest Meteorol 2004;121(1–2):37–53.

Wilson JD. A second-order closure model for flow through vegetation. Bound-LayMeteorol 1988;42(4):371–92.