lable at ScienceDirect

Environmental Pollution 159 (2011) 954e962

Contents lists avai

Environmental Pollution

journal homepage: www.elsevier .com/locate/envpol

Monitoring of heavy metal concentrations in home outdoor air using moss bags

Marcela Rivera a,b,c,d,*, Harald Zechmeister e, Mercedes Medina-Ramón a,b,d, Xavier Basagaña a,b,d,Maria Foraster a,b,c,d, Laura Bouso a,b,d, Teresa Moreno f, Pascual Solanas g,h, Rafael Ramos g,h,Gunda Köllensperger i, Alexandre Deltell j, David Vizcaya a,b,c,d, Nino Künzli a,k, l

aCentre for Research in Environmental Epidemiology CREAL, Barcelona, SpainbMunicipal Institute of Medical Research (IMIM-Hospital del Mar), Barcelona, SpaincUniversitat Pompeu Fabra, Barcelona, SpaindCIBER Epidemiología y Salud Pública (CIBERESP), SpaineUniversity of Vienna, Faculty of Life Sciences, Vienna, Austriaf Institute of Environmental Assessment and Water Research (IDÆA-CSIC), Barcelona, SpaingResearch Unit, Family Medicine, Girona, Jordi Gol Institute for Primary Care Research (IDIAP Jordi Gol), Catalan Institute of Health, Catalunya, SpainhDepartment of Medical Sciences, School of Medicine, University of Girona, SpainiUniversity of Natural Resources and Applied Life Sciences, Vienna, Austriaj Polytechnic School, GREFEMA, University of Girona, Spaink Swiss Tropical and Public Health Institute, Basel, SwitzerlandlUniversity of Basel, Basel, Switzerland

The long-term spatial distribution of heavy metals, measured with mo

ss bags, is mainly determined by proximity to bus lines.

a r t i c l e i n f o

Article history:Received 6 August 2010Received in revised form1 December 2010Accepted 9 December 2010

Keywords:Traffic-related air pollutionOutdoor exposureSpatial distributionDeterminantsBusesParticulate matterNitrogen dioxide

Abbreviations: ADT, average daily traffic; Al, alummony; Cd, cadmium; Cr, chromium; Cu, copper;molybdenum; Sn, tin; Zn, zinc; NO2, nitrogen dioxidparticulate matter; PM2.5, PM10, the fraction of aerodynamic diameter less than 2.5 mm and 10 mm respectRegister (Registre Gironí del Cor) Study.* Corresponding author.

E-mail address: arivera@creal.cat (M. Rivera).

0269-7491/$ e see front matter � 2010 Elsevier Ltd.doi:10.1016/j.envpol.2010.12.004

a b s t r a c t

One monitoring station is insufficient to characterize the high spatial variation of traffic-related heavymetals within cities. We tested moss bags (Hylocomium splendens), deployed in a dense network, for themonitoring of metals in outdoor air and characterized metals’ long-term spatial distribution and itsdeterminants in Girona, Spain. Mosses were exposed outside 23 homes for two months; NO2 wasmonitored for comparison. Metals were not highly correlated with NO2 and showed higher spatialvariation than NO2. Regression models explained 61e85% of Cu, Cr, Mo, Pb, Sb, Sn, and Zn and 72% of NO2

variability. Metals were strongly associated with the number of bus lines in the nearest street. Heavymetals are an alternative traffic-marker to NO2 given their toxicological relevance, stronger associationwith local traffic and higher spatial variability. Monitoring heavy metals with mosses is appealing,particularly for long-term exposure assessment, as mosses can remain on site many months withoutmaintenance.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

It is well known that air pollution affects human health(Dockery, 2009; Kunzli and Tager, 2005). Atmospheric particulatematter (PM) induces adverse effects on the respiratory andcardiovascular systems through various mechanisms (Nel et al.,

inum; As, arsenic; Sb, anti-Pb, lead; Pt, platinum; Mo,e; NOx, nitrogen oxides; PM,sol particles with the aero-ively; REGICOR, Girona Heart

All rights reserved.

2001; Simkhovich et al., 2008; WHO, 2006). There is emerginginterest in the toxic features of ultrafine particles, as a source ofhighly oxidative constituents bound to their surface (de Kok et al.,2006). Several recent studies identified particle bound heavymetals as particularly relevant in causing the health effects ofparticles (Bell et al., 2009; Chen and Lippmann, 2009) such asinduction of oxidative stress (Nel et al., 2001) and mitochondrialdamage (Li et al., 2003), which ultimately lead to an increase incardiovascular morbidity and mortality (Araujo and Nel, 2009).

Most existing studies relied on environmental concentrations ofPM-constituents measured at a few locations. A primary challengein the assessment of long-term exposure to local traffic-relatedpollutantse such as ultrafine particles and related constituencieseis its high spatial variation on the local scale along busy roads andstreet canyons, which are a typical feature of many compact

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962 955

European cities (Hoek et al., 2008). Therefore, to characterize long-term exposure to these local pollutants it is not sufficient to rely onone or a fewmonitoring stations. The spatial contrasts of pollutantsemitted by traffic can be very large within the first hundred metersof roads (Monn, 2001), reaching background levels after some100e200 m only (Hoek et al., 2008). A dense network of long-termmeasurements is thus needed to capture the spatial contrasts ofthese contaminants (Harrison and Deacon, 1998). This has beendone extensively with passive samplers using mostly NOx andoccasionally other traffic-related gases, used as markers of traffic-related pollutants. Due to the saturation of samplers, repeateddeployment of passive samplers for a few (often two) weeks isneeded to describe long-term average conditions. Conventionalmeasurement of PM mass concentration or particle bound toxi-cants, including metals, is even more demanding and costly, andthus long-term monitoring at a high spatial resolution isprohibitive.

The purpose of our study was to characterize the long-termsmall-scale spatial distribution of heavy metals, using mosses asbiomonitors of particle bound metals, and to identify the maindeterminants of the spatial distribution of these metals in Girona(Spain). We also compared the spatial distribution of heavy metalsto that of nitrogen dioxide (NO2) as commonly used markers oftraffic-related local pollution.

Biomonitoring techniques, such as moss sampling, allow moni-toring pollutants for long periods of time. With the active bio-monitoring technique a small amount ofmoss is exposed in the sitesof interest and analyzed subsequently (also known as “moss bag”method (Goodman and Roberts, 1971)). Accumulation of depositedparticulate matter takes place on the moss for as long as it isexposed, which can be several months. These features are veryappealing for the use ofmoss bags as long-termmarkers of airbornemetals in epidemiologic studies investigating health effects of long-term exposure to air pollution. Elemental deposition in moss bagshas been shown to be highly correlated both spatially and tempo-rally with PM10 measurements, with Pearson correlations for, e.g.,Cu reaching 0.94 in mosses versus PM10 (Cao et al., 2009). Recentstudies have used moss bags to measure urban pollution fromseveral sources (Fernandez et al., 2004; Makholm and Mladenoff,2005; Tretiach et al., 2007), and a few have focused on traffic-related pollution (Anicic et al., 2009a; Cao et al., 2009; Naszradiet al., 2007; Viskari et al., 1997). Our study is the first to use thetechnique to characterize home outdoor exposure at people’s resi-dences which is of particular relevance in epidemiological research.In fact, our study is part of the exposure assessment strategy of theREGICOR-Air project (www.regicor.org), an ongoing cohort studyconducted in Girona, Spain, which aims to determine the long-termeffects of local traffic-related air pollution on cardiovascular health,including atherosclerosis. (Kunzli et al., 2005).

2. Materials and methods

Wemonitored ambient air concentrations of heavy metals by placing moss bagson balconies of REGICOR participants’ homes for 2 months. The moss bags weredistributed over 23 sites across Girona and Salt (Spain). The heavy metal concen-trations in mosses were then measured in order to define their distribution. Traffic-related variables were collected to establish their relationship to the metalconcentrations. Palmes NO2 passive samplers were deployed in parallel with themoss bags.

2.1. Study area

The geographic area of study is located in Girona and Salt, neighboring cities ofthe Girona province, in northeastern Spain, with 96.188 and 29.985 inhabitants (INE,2009) and an area of 39.1 and 6.59 km2 respectively. These cities were chosenbecause of their broad contrasts in NO2 levels. NO2 is considered a marker of traffic-related emissions, and thus, substantial contrast of other traffic-emitted pollutantssuch as heavy metals was expected.

The monitoring was performed at 23 sites: 20 homes of the REGICOR studyparticipants and research partners, two street locations and one park (see map inFig. 1), selected to cover all residential sectors of the study area and to be homo-geneously spread across space and the previously characterized NO2 distribution.Only non-smoking households with an accessible balcony were selected for thedeployment of the mosses and NO2 samplers. In line with our objective of small-scale pollutant characterization monitoring density was rather high (one moss bagper 0.5 km2).

2.2. Collection, pre-treatment and exposure of mosses

Hylocomium splendenswas chosen as the biomonitor because it has beenwidelyused and its uptake capacity has been investigated thoroughly (Zechmeister et al.,2003). Collection and pre-treatment were made following internationallyapproved techniques (Harmens et al., 2010; Rühling and Steinnes, 1998). On March2008, naturally grownmosses were collected in a 50-m2 parcel in the Großarl Valley,a rural area close to Salzburg (Austria). The moss was cleaned from any litter andadhering macroscopic particles, brown leaves were cut off and discarded and onlythe green parts of the mosses representing the growth of the last two years werekept. The remaining moss material was rinsed three times with double distilledwater, dried at 40 �C for 72 hours and homogenized by mixing all materials toprevent differences in the moss bags composition. Four subsamples were taken forpre-exposure analysis to determine pollutants background levels. Mosses (5 g) werepacked in polypropylene bags, airtight sealed and sent to the study center (Barce-lona) where the preparation of the moss bag for exposure took place.

The bags were prepared by putting moss loosely on a wooden frame of10 cm� 10 cm, equipped with a fine polypropylene net of 1 cm� 1 cm mesh sizethat maintained the leaves within the frame using the system (see Fig. 1) developedby Zechmeister et al. (2006). The monitors were exposed for 8 weeks in May andJune 2008. The moss bag was set on the balcony at a sheltered location to avoid anywet deposition as well as a potential washout of already deposited particles. Whensuch a location did not exist, an acrylic roof was fixed on top of the moss frame. Thebalconies were at heights between 3 and 21 m above the street level.

Most studies using moss bags expose them for 1e4 months. We selected anexposure time of eight weeks knowing that accumulation of air pollution tracersincreases linearly with time over a 4-month period (Anicic et al., 2009b).

While exposed, the mosses were not irrigated as few differences between wetand dry moss bags have been found (Anicic et al., 2009b) and the simpler logistics ofthe dry moss technique makes it more suitable for monitoring at participants’homes.

To measure NO2, a diffusion tubewas installed next to the moss, exposed for onemonth and then replaced by a new tube deployed during the second visit. Duringthis visit (June 2e6) we also verified that the measurement instruments wereworking properly. At the third and final visit (June 30eJuly 4) the samplers werecollected and packed in tightly sealed PVC bags. Questionnaires were administeredduring the visits to identify unusual situations related to the measurement instru-ments and environmental conditions.

2.3. Chemical analysis

After monitoring, the chemical analysis was performed in the University ofNatural Resources and Applied Life Sciences, and the Umweltbundesamt GmbHlaboratories (Vienna). A detailed description of the analytical method is given in theSupplementary data. Briefly, the moss was dried at 30 �C and grinded prior tomicrowave digestion in nitric acid and hydrogen peroxide. Concentrations of Al, As,Cd, Cr, Cu, Mo, Pb, Sb, Sn and Zn were measured by inductively coupled plasmasector field mass spectrometry (ICP-SFMS) carried out on an Element 2 ICP-SFMS(Thermo Fisher, Bremen, Germany). Quality control was performed by analyzingmicrowave digestion blanks prepared with ultra-pure water; adding Indium (1 mg/L)to all samples before ICP-SFMS as an internal standard; and using the certifiedreference material TM 27.2 (low level fortified sample for trace elements) for cali-bration quality.

The chemical analysis of the NO2 samplers was performed at the AEA Energy &Environment laboratories (London). NO2 was analyzed colorimetrically (Bran andLuebbe autoanalyzer) as described by Targa et al. (2008).

2.4. Determinants of spatial distribution of pollutants

To identify local determinants of the distribution of heavy metals in ambient air,geostatistical techniques were used to derive information on potential variables foreach location, namely traffic intensity (average daily traffic ADT), percentage ofheavy vehicles, number of bus lines and stops in the nearest street, distance fromsampler to traffic, height above the street, degree of urbanization and character-ization of the site’s surroundings such as crossroads and open areas. The proximityfrom the site to a river was also considered as three rivers cross Girona. The OnyarRiver, flowing through downtown, has major streets on both sides. Traffic count datawere provided by several sources including the urban mobility plan (PMU-Girona,2007) and a city noise mapping project (Deltell, 2005). Site characterization datawere collected by manual measurements using high-resolution maps from the

Fig. 1. Enrichment factors of antimony in mosses (mg/g of moss) and picture of the moss monitor.

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962956

Girona City Council website (UMAT, 2008). A list of all variables examined aspotential predictors is given in the Table 1.

2.5. Statistical analysis

The heavy metal concentrations were measured by the net enrichment, i.e. theconcentration after exposure minus the initial concentration.

The spatial distribution of metals and NO2 was described by the ratio of theinterquartile range to the median, which is a robust measure analogous to thecoefficient of variation. Correlation betweenmetals wasmeasured using Spearman’srank correlation coefficients.

To assess the effect of all the potential determinants mentioned above, univar-iate and multiple variable linear regression models were fitted for each heavy metaland for NO2. Backward stepwise linear regression was used to determine the mostrelevant variables to be included in each model in order to get the highest adjusted

coefficient of determination (R2). Cross-validation R2 was computed by the leave-one-out method to assess the predictive ability of the model to a new dataset (Efronet al., 1993).

Multivariate analysis was conducted to identify metal distribution patterns atthe sites. Factor analysis reduced the number of observed variables by groupingtogether metals with similar correlations (Pérez, 2001). The aim was to substitutethe individualmetals by the first factors carrying a large proportion of the variability.We standardizedmetals to havemean zero and variance one and used the principal-factor technique with varimax rotation. Factor loadings, which measure the corre-lation between the new factors and the original variables, are reported. The resultingfactors were predicted for each site and the association between these and thepotential determinants was assessed by linear regression.

Complementary to factor analysis we performed a cluster analysis using parti-tioning around medoids (PAM) on standardized metals and NO2 to derive clusters ofelements that showed a similar distribution on all sites (Zechmeister et al., 2006).

Table 1Predictor variables. N, minimum, maximum, mean and standard deviation for continuous variables, median and interquartile range (iqr) for ordinal variables and frequency (%)for categorical variables.

Predictor variable N Min Max Mean SD Median IQR %

Traffic flow in the nearest street (veh/100 per day) 21 0 260 50.6 72.6Heavy vehicles in the nearest street (veh/100 per day) 21 0 26 3.1 5.7Street width (m) 23 3 22 9.3 4.5Distance to the nearest street (m) 21 1 87 15.7 18.9Distance to the nearest crossroad (m) 23 4 230 62.8 59.3Moss height above street (m) 20 3 21 8.7 4.8Number of bus lines in the nearest street (passing in front of the house) 23 0 4 0.0 1.0Number of bus lines passing within 150 m along the street 23 0 5 1.0 2.0Number of bus lines (150 m buffer) 23 0 5 2.0 3.0Number of bus stops (50 m buffer) 23 0 2 0.0 0.0Number of bus stops (150 m buffer) 23 0 5 2.0 3.0Home floor number 23 0 7 1.0 2.0Number of crossroads (100 m along street) 23 0 3 1.0 2.0Number of crossroads (300 m along street) 23 0 6 3.0 2.0River (100 m buffer) 23 0 1 0.0 0.0River (300 m buffer) 23 0 1 0.0 0.0Number of open areas (100 m along street) 23 0 6 2.0 5.0Number of open areas (300 m along street) 23 0 11 5.0 9.0Number of gaps (100 m along street) 23 0 5 0.0 5.0Number of gaps (300 m along street) 23 0 8 1.0 8.0

Building densityIsolated house 5 21.7Buildings in 300 m along street (one side) 4 17.4Buildings in 300 m along street (two sides) 14 60.9

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962 957

Statistical analyses were performed with Stata 10.0, R 2.9.2 and ArcGis 9.1packages.

3. Results

The moss bags were deployed and collected successfully at allparticipants’ homes. There was an unexpected heavy rainingseason during the first month of measurements but it did not affectthe samplers and the protection roofs remained in place.

The initial concentrations and detection limits of heavy metalsare shown in Table 2. For all metals, except platinum (Pt), thepercentage of detection in the moss was 100%. Pt was detected onlyin two mosses (8.7%), thus data for Pt are not shown.

Descriptive statistics for net enrichment of pollutants in mossesare shown in Table 2. The spatial variation, assessed by the inter-quartile range over the median, was higher for most metals(0.81e8.16) as compared to NO2 (0.54), with the exception of Zn

Table 2Detection limits (mg/g), background levels (mg/g) and net enrichment of heavy metals in

Metals LODa LOQb Initial conc.c mean Initial conc.c sd

Aluminum (Al) 1.95 6.49 151.88 15.08Antimony (Sb) 0.0032 0.0106 <LOD <LODArsenic (As) 0.0069 0.0230 0.11 0.0144Cadmium (Cd) 0.0182 0.0607 0.04 0.0015Chromium (Cr) 0.0075 0.0251 <LOD <LODCopper (Cu) 0.0108 0.0361 4.18 0.2535Lead (Pb) 0.0039 0.0131 2.42 0.1191Molybdenum (Mo) 0.0012 0.0040 0.11 0.0119Platinum (Pt) 0.0002 0.0007 <LOD <LODTin (Sn) 0.0032 0.0107 0.09 0.0041Zinc (Zn) 1.08 3.59 13.66 1.00NO2

e e e e e

a LOD: limit of detection.b LOQ: limit of quantification.c Concentration before exposure.d Interquartile range.e Measured with passive samplers.

(0.47) and Cr (0.54). The highest spatial variability was observed forAs (8.16) and Sn (1.26). Influential data points in the linearregressions of individual metals were found for two of the threenon-residential sites (street; park). The high concentrations ofmetals in these sites were expected given the proximity to streetswith high traffic intensity. Consequently, these outliers were notincluded in the results of regression models and factor analysispresented here.

Most metals showed low correlations with NO2 (Supplementarydata). The Spearman rank coefficient ranged from0.15 for Cd to 0.61for Sb andwas significant only for Sb and Zn. The correlation amongmetals ranged from 0.03 (between Al and Cd) to 0.92 (between Aland Cr). Several metals were highly correlated. That was the casefor Cr, Cu and Sb; Mo, Sb and Sn; and Cu and Al, respectively.

Two factors were derived through factor analysis that explained90.8% of the total variance of metals (NO2 was not included inderiving these factors) (Fig. 2). The first factor explained 73% of the

mosses (mg/g of moss) and ambient NO2 concentrations (mg/m3).

Net enrichment (N¼ 23)

Mean SD Min Median Max IQRd IQR/median

501.06 460.85 190.24 356.89 2442.45 290.69 0.810.31 0.32 0.13 0.18 1.32 0.20 1.100.06 0.15 �0.07 0.02 0.44 0.14 8.160.04 0.02 0.01 0.04 0.07 0.03 0.871.11 0.71 0.62 0.82 3.44 0.45 0.544.09 3.63 1.43 2.81 14.87 2.89 1.030.92 1.06 �0.14 0.62 3.89 0.64 1.020.22 0.18 0.07 0.15 0.77 0.16 1.07e e e e e e e

0.40 0.53 0.10 0.18 2.00 0.22 1.2623.48 6.68 14.91 23.26 35.80 10.89 0.4723.89 9.32 8.57 22.20 41.06 11.98 0.54

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962958

variability and is a compound of all metals except Cd. The secondfactor explained the remaining 17.8% and is positively correlatedwith Cd, Pb and Zn and negatively correlated with As.

The cluster analysis conducted with outliers revealed fourclusters (see results in Supplementary data), three with the samegrouping of metals indicated by the factor analysis and one inde-pendent cluster formed by NO2. Regarding the metals clusters onewas formed by As only, Zn and Cd contributed to the second onewhile the third grouped the remaining metals (Mo, Cr, Sn, Cu, Sb, Aland Pb). When removing the outliers, NO2 and Zn became part ofthe large cluster grouping most metals. The hypothesized structureof the components and allocation of groups to particular sourcesare considered in the discussion.

Regression models (Table 3) explained between 61 and 85% ofthe spatial variation of Pb, Zn, Mo, Cr, Sb, Sn and Cu and 40% of Al.Heavy metal concentrations were strongly associated with thenumber of bus lines passing in the nearest street. This variable wasthemost significant andwas a strong predictor for Al, Sb, Cr, Cu, Mo,Sn and Zn. It explained between 75 and 85% of Mo, Cr, Sb, Sn and Cuvariability. For illustration, a scatter of Cu versus the number of buslines passing in the nearest street is shown in Fig. 3. Pb was asso-ciated with the number of bus stops and with the density ofbuildings, and Zn with the presence of a river within 150 m, andwith the number of bus lines and within a 150 m buffer. None of thestudied covariates were associated with Cd or As. Heavy vehiclestraffic and the characterization of the site’s surroundings explained72% of the variability of NO2. None of the other variables (moss

Factor 1

Fact

or 2

0.2 0.4 0.6 0.8 1.0

−0.4

−0.2

0.2

0.4

0.6

0.8

As

Sb

Cd

Cr

Cu

Pb

Mo

Sn

Zn

Al

Fig. 2. Factor loadings for the factor analysis of all metals using the principal-factormethod. The two principal factors explain 90.8% of the metals variance (N¼ 21). A longarrow indicates that the metal is well represented by the two factors, while shortarrows indicate that the metal is not well characterized in this new two-dimensionalspace. Two arrows with a similar direction (angle) imply a high correlation betweentwo metals, and the closer the arrow to one of the axis the higher the correlation of themetal with the factor represented in that axis.

height above street, distance to nearest street, street width, etc.)were significant predictors of the pollutants’ variability.

Metals were more strongly associated with bus and total trafficthan NO2. The first factor combining the metals was highly corre-lated with average daily traffic (Pearson correlation r: 0.82,p< 0.0001) and the bus lines circulating in the nearest street(r: 0.91, p< 0.0001). In comparison, the correlation of NO2 withthese variables was lower (ADT r: 0.59, p: 0.008; bus lines r: 0.45,p: 0.043). Both NO2 and the factor combining metals were associ-ated with heavy vehicles.

When the first factor resulting from factor analysis was used asthe response variable in a regressionmodel the number of bus linespassing in the nearest street explained 82% of its variation (Table 3).The second factor was not significantly associated to any of thestudied variables.

4. Discussion

This is the first study using mosses at home outdoors to monitorthe small-scale inner-city spatial distribution of ambient heavymetals. Analysis of moss samples provides information on the long-term spatial distribution of particle bound metals demonstratinga high potential for epidemiological studies. A significantpercentage of the ambient metal concentration was explained bythe number of bus lines passing in the nearest street. To put thefindings in context, we discuss the main predictors of the metaldistribution, and the use of metals as markers of traffic-relatedpollutants. We also discuss on the practical advantages of themethod and the characteristics of the measurements.

In the present studymetals showed higher spatial variation thanNO2. The distribution of most of the metals was explained by thebus traffic intensity in the nearest street. The mean distance fromthe house facade to this street was 15.7 m. The distribution of NO2,on the other hand, was determined by the traffic intensity in thenearest street and by the number of open areas within 300 m andthe number of crossroads within 100 m. The two latter measure-ments are indirect indicators of traffic at a medium-scale distancefrom the site. NO2 e a marker of both primary and secondarypollutants e is thus not only influenced by the local emissions butalso by the dispersion from neighboring areas and sources. Onceemitted, the pollutants disperse and the ambient concentrationchanges with distance creating a gradient. The distance untilreaching background concentrations is generally lower for PMmass(and consequently, for the metals bound to it) than for NO2 (Zhouand Levy, 2007). Therefore, the spatial variation of metals,compared to that of nitrogen dioxide, is higher andmore influencedby the distance to the emission sources i.e. traffic. Metals are majorcontributors to the oxidative properties of PM (Chen and Lippmann,2009). Oxidative activity is considered a highly relevant toxico-logical characteristic but no monitors are yet available to directlymeasure redox activity on a mass scale needed in epidemiologicalstudies (Borm, 2008). Thus, monitoring metals by mosses appearsto be an attractive option.

The variable that explained the highest fraction of the metalsvariability was the number of bus lines in the nearest street. InFig. 3, the point representing a house with four bus lines in thenearest street did not have unduly influence in the results, as evi-denced by the usual regression diagnostics. The R2 of the Curegression with and without this data point is 85 and 65% respec-tively. In Girona, the number of bus lines is a direct indicator of bustraffic flow given that all lines have a frequency of 15 minthroughout the day. As it is common in Spain, most buses operateon diesel. In 2007, 98% of the total of registered buses in Gironaprovince ran on diesel, but they only accounted for 0.16% of the totalvehicle fleet (DGT, 2009). Other heavy vehicles were not significant

Table 3Linear regression models for outdoor heavy metals (mg/g of moss), factor 1 of factor analysis (dimensionless) and NO2 concentrations (mg/m3).a,b

N Coef. Std. Err. p [95% CI] R2 c R02 d R2(cv)e

Aluminum (Al) 21 0.4 0.14 0.29Number of bus lines in nearest street 91.06 25.37 0.002 37.96 144.16Constant 339.79 28.77 <0.001 279.58 400.00

Antimony (Sb) 21 0.8 0.53 0.51Number of bus lines in nearest street 0.23 0.03 <0.001 0.18 0.29Constant 0.15 0.03 <0.001 0.08 0.21

Chromium (Cr) 21 0.77 0.48 0.66Number of bus lines in nearest street 0.39 0.05 <0.001 0.29 0.49Constant 0.74 0.06 <0.001 0.63 0.86

Copper (Cu) 21 0.85 0.65 0.68Number of bus lines in nearest street 2.66 0.25 <0.001 2.13 3.19Constant 2.07 0.29 <0.001 1.47 2.67

Lead (Pb) 21 0.61 0.61 0.24Number of bus stops (50 m buffer) 0.81 0.24 0.004 0.30 1.32Buildings in 300 m along street (one side) 1.11 0.45 0.024 0.17 2.06Buildings in 300 m along street (two sides) �0.19 0.36 0.611 �0.95 0.58Constant 0.47 0.33 0.169 �0.22 1.15

Molybdenum (Mo) 21 0.75 0.38 0.52Number of bus lines in nearest street 0.12 0.02 <0.001 0.08 0.15Constant 0.12 0.02 <0.001 0.08 0.16

Tin (Sn) 21 0.82 0.60 0.52Number of bus lines in nearest street 0.35 0.04 <0.001 0.27 0.43Constant 0.12 0.04 0.009 0.04 0.21

Zinc (Zn) 21 0.67 0.62 0.37Number of bus lines in nearest street 2.31 0.82 0.011 0.59 4.03Number of bus lines (150 m buffer) 1.67 0.48 0.003 0.67 2.68River (300 m buffer) 10.67 2.75 0.001 4.86 16.48Constant 16.26 1.46 <0.001 13.18 19.34

Factor 1 of factor analysis 21 0.82 0.55Number of bus lines in nearest street 0.88 0.09 <0.001 0.68 1.08Constant �0.46 0.11 <0.001 �0.68 �0.24

Nitrogen dioxide (NO2) 19 0.72 0.70 0.61Traffic flow on nearest street (veh/100 per day) 0.06 0.02 0.006 0.02 0.10Number of crossroads (100 m along street) 3.56 1.36 0.019 0.66 6.45Number of open areas (300 m along street) �1.20 0.33 0.003 �1.90 �0.49Constant 23.21 3.15 0 16.50 29.91

a Two outdoor samplers located in streets are not included in regressions of individual metals and NO2.b Cadmium and Arsenic were not associated to any of the studied predictors.c R2 Adjusted for multiple regression.d R02 Adjusted coefficient of determination without site with 4 bus lines in the nearest street.e Cross-validation coefficient of determination.

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962 959

determinants of the spatial variation of metals, suggestinga particular role of bus emissions. Our results are in line withprevious studies concluding that a small fraction of the vehicle fleetmay be responsible for large parts of the total traffic emissions(Zhang et al., 1995). In Athens (Greece) the main determinant ofdaily PM10 and NOx emissions was the percentage of diesel vehiclesin the fleet (Kassomenos et al., 2006). In Shanghai, diesel buses andtrucks (with few emission controls), accounting for the 5.5% of thevehicle fleet, were responsible for 65% and 56% of NOx and PMemissions from vehicles respectively (Wang et al., 2008). Theexposure assessment done in cities such as Gironamay thus need todifferentiate between buses, other heavy duty vehicles, andbetween diesel and gasoline powered vehicles.

The first factor of the factor analysis explained about 73% of thevariation inmetals. All metals, except Cd, showed a high correlationwith this factor. Therefore, this new composite variable can beregarded as an index of overall exposure to heavy metals. Asexpected, the model of the first factor captured the important effectof the number of bus lines, with R2 of 82%. This factor is thus anindicator of the particles emitted or resuspended by buses. In linewith results of factor analysis, the cluster analysis detected that NO2and As behaved with two independent patterns, and Cd and Zn also

showed a common pattern that could be differentiated from themain metal group (Al, Sb, Cr, Cu, Pb, Mo, Sn). This grouping can bedue to similar emission sources. For example, Cu, Sb, and Mo, theelements showing the highest correlations, are present in the cars’brake pads and derive from abrasion processes while braking. Theconcentration of these metals is seriously influenced by dustdeposition (Reimann et al., 2001), and thus, Al and Cr (dust-relatedelements) belong to the same main metal group. We consider thatthemain emission source of heavymetals in or study is traffic giventhat the influence of other sources was minimal. Coal and othercombustion processes did not affect our results as the study wasconducted during the summer season, in non-smoking households,and on balconies not adjacent or connected to the kitchen. More-over, there are no major emissions from industrial sources in thecity, and the few industrial areas of Girona are in the outskirts of thecity, thus affecting, if any, background levels rather than localvariation.

The sample size of 23 data points limits a more in-depth stratifiedanalysis. However, for the small town of Girona/Salt, the welldistributed measurement sites provide a sufficient data base toexplain keydeterminants of the localmetal distribution, confirmedbyrelativelyhigh cross-validation coefficientsof determination (Table 3).

05

1015

Cu

net e

nric

hmen

t (µg

/g o

f mos

s)

0 1 2 3 4Number of bus lines in front of the site

Fig. 3. Scatter plot for outdoor Cu (mg/g of moss) versus number of bus lines in thenearest street (N¼ 21).

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962960

4.1. Characteristics of the measurements and advantages andlimitations of the method

We successfully implemented a protocol for a two-monthmonitoring period but this could easily be extended to long-termmonitoring. Mosses receive water and nutrients by atmosphericdeposition, and during this process PM is deposited and accumu-lated on its surface. The uptake depends on the nature of theelements only, and is independent of the physiological condition ofthe plant (Zechmeister et al., 2003). For the two months period ofour study, the uptake rate could be considered constant over time(Goodman and Roberts, 1971). Previous studies have successfullyused mosses as active biomonitoring device from at least one week(Ratcliffe, 1975), to up to five months (Anicic et al., 2009b). Theadvantages of a longer monitoring period are that the character-ization of long-term exposure is more accurate (as the metalcontent increases with time) and more relevant in the assessmentof chronic health effects. Thus, metals occurring in very lowconcentrations, such as platinum, could be monitored as well.Alternatively, more sensitive analytical techniques would berequired for such elements.

The moss monitors come with very relevant differences andadvantages to the conventional PM sampling approaches. Thepurpose of our study was not to validate the element concentra-tions in mosses or to formally compare themwithmetals in PM (forthis purpose see (Cao et al., 2009)), but we will discuss on thepractical characteristics of both methods. With approximately 5 gper sampler, themoss quantity needed for monitoring is very small.Thus, moss sampling can be done in sustainable ways by collectingsmall portions of mosses from a dense moss colony in clean areaslike the Austrian Alps.While the logistics to organize deployment issimilar to more traditional sampling approaches, the mosssamplers do not require maintenance or access to any utility(electricity, water, pumps, etc.). The installation of the mosssamplers was feasible at home outdoors as it was in previousstudies done in different settings such as tunnels (Zechmeisteret al., 2006), region-wide or industrial areas (Adamo et al., 2003;Fernández et al., 2007; Zechmeister and Hohenwallner, 2006).

The sampler is small, it makes no noise, and participants werepleased with the placement. Site visits done after the first andsecondmonths revealed no problemswith the biomonitors. Qualitycontrol visits may indeed not be necessary and could be replaced byoccasional phone calls to reassure the mosses are still in place.Similar to NO2 samplers, mosses can be deployed at residences

where people spendmost of their time, thus, results provide a goodmarker for personal exposure. PM monitoring is far moredemanding and expensive (logistics, space, installation, mainte-nance, pumps, electricity, filter handling, personnel for repeatedmaintenance visits, etc) and therefore they are usually used only ata few locations and monitoring time is restricted to days or a fewweeks. This, however, is not appropriate when the goal is todescribe the long-term within-city contrasts in PM-constituents.While laboratory costs to analyze a moss or a PM filter are similar,the advantage of using mosses becomes most evident if theobjective is long-term exposure monitoring (e.g. 4e12 months).The moss bags technique accurately measures the element depo-sition trends and is cost-effective compared with PM monitoring.

It is worth noting that the particles deposited in the moss coverthe entire spectrum of particle sizes, i.e. dry atmospheric deposi-tion, and that the measurement resulting from their analysisaccounts for the total particles accumulated over the monitoringperiod. Metals distribute in all PM fractions with a tendency ofhigher contents of Sb in the coarse fraction and of As, Cd, Cr, Cu, Mo,Pb, Sn and Zn in the fine fraction (Moreno et al., 2006). Given therecent evidence of association of both coarse (Brunekreef andForsberg, 2005; Lipsett et al., 2006; Yeatts et al., 2007) and fine(Bell et al., 2009) PM fractions with adverse health outcomes,determining metal concentrations in the whole range of depositedparticle size is thus relevant.

The net enrichment for As and Pb was negative (initialconcentration higher than final) in nine and one sites respectively.For As, this may indicate a true loss of the element during exposure,given that it is not well retained (Zechmeister et al., 2003).However, negative values for Pb may reflect low enrichment anda relatively high measurement error, since mosses have goodbinding capacity for this metal (Ruhling and Tyler, 1970).

One drawback of the biomonitoring technique is that a directinference of the environmental concentrations of pollutants cannotbe made. Studies have reported high correlations between Cucontent in moss and PM10 (Cao et al., 2009) and between metalconcentrations in moss with the unsedimentable dry depositedparticles (<0.8 mm) (Vasconcelos and Tavares, 1998) and theatmospheric bulk deposition (Berg et al., 1995). Deriving environ-mental concentrations is not needed when using mosses inepidemiological studies. These do not necessarily require absoluteconcentrations but tools to properly rank the spatial distribution ofair quality within and across communities. Monitoring air qualitywith mosses does achieve this goal effectively.

As part of the REGICOR-Air project, PM2.5 was measured dailythrough high-volume air sampling in the three non-residentialmoss locations, for the same time period. Elemental concentrationsof heavy metals on the filters (Al was not measured) were averagedfor the two months period and compared with moss results. Whilethree data points are not valid for a formal assessment of correla-tions it was reassuring to see that metals such as As, Sb, Cd, Cu, Snand Zn followed the same pattern across the three sites. Moreover,we compared the results of the two methods with the Spearmanrank correlation across the ranking of elements in each sites; the r’sof 0.58, 0.73, and 0.63 respectively suggest that metals in PM2.5 andmosses are determined by similar characteristics of the localenvironment. A perfect correlation is not expected given thatmetals also bind to the PM fraction above 2.5 mm, which alsodeposits in mosses.

5. Conclusions

In conclusion, mosses can be used tomeasure the distribution ofmetals at home outdoors within cities. For the purpose of long-term exposure studies, mosses are an attractive alternative to PM

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962 961

monitoring and complementary or possibly an alternative to theuse of NOx as a marker of local traffic-related pollution. Heavymetals showed higher spatial variation and associations with trafficand traffic markers than NO2. These features and the knowntoxicity of heavy metals indicate that moss-based biomonitoring isan appealing approach for exposure assessment in epidemiologicalstudies.

Acknowledgements

Project funded by L’Agence Française de Sécurité Sanitaire del’Environnement et du Travail AFSSET, and Spanish Grant FISPI060258. Ms. Rivera received funding from DIUE GC/FSE(2009FI_EX 00041), Dr. Medina-Ramón from the Spanish Ministryof Science and Innovation (JCI-2007-47-767), Dr. Künzli fromICREA, Catalonia, Spain.

Appendix. Supplementary data

Detailed information on the chemical analysis, tables of theSpearman correlation coefficients across pollutants and figures ofthe clusters derived from principal component analysis.

Supplementary data associated with this article can be found inthe online version, at doi:10.1016/j.envpol.2010.12.004.

References

Adamo, P., Giordano, S., Vingiani, S., Castaldo Cobianchi, R., Violante, P., 2003. Traceelement accumulation by moss and lichen exposed in bags in the city of Naples(Italy). Environmental Pollution 122, 91e103.

Anicic, M., Tasic, M., Frontasyeva, M.V., Tomasevic, M., Rajsic, S., Mijic, Z., Popovic, A.,2009a. Active moss biomonitoring of trace elements with Sphagnum girgen-sohnii moss bags in relation to atmospheric bulk deposition in Belgrade, Serbia.Environmental Pollution 157, 673e679.

Anicic, M., Tomasevic, M., Tasic, M., Rajsic, S., Popovic, A., Frontasyeva, M.V.,Lierhagen, S., Steinnes, E., 2009b. Monitoring of trace element atmosphericdeposition using dry and wet moss bags: accumulation capacity versus expo-sure time. Journal of Hazardous Materials 171, 182e188.

Araujo, J.A., Nel, A.E., 2009. Particulate matter and atherosclerosis: role of particlesize, composition and oxidative stress. Particle and Fibre Toxicology 6, 24.

Bell, M.L., Ebisu, K., Peng, R.D., Samet, J.M., Dominici, F., 2009. Hospital admissionsand chemical composition of fine particle air pollution. American Journal ofRespiratory and Critical Care Medicine 179, 1115e1120.

Berg, T., Røyset, O., Steinnes, E., 1995. Moss (Hylocomium splendens) used as bio-monitor of atmospheric trace element deposition: estimation of uptake effi-ciencies. Atmospheric Environment 29, 353e360.

Borm, P.J.A., 2008. Future interactions in particle toxicology: the role of PFT. Particleand Fibre Toxicology 5, 5.

Brunekreef, B., Forsberg, B., 2005. Epidemiological evidence of effects of coarseairborne particles on health. European Respiratory Journal 26, 309e318.

Cao, T., Wang, M., An, L., Yu, Y., Lou, Y., Guo, S., Zuo, B., Liu, Y., Wu, J., Cao, Y., Zhu, Z.,2009. Air quality for metals and sulfur in Shanghai, China, determined withmoss bags. Environmental Pollution 157, 1270e1278.

Chen, L.C., Lippmann, M., 2009. Effects of metals within ambient air particulatematter (PM) on human health. Inhalation Toxicology 21, 1e31.

de Kok, T., Driece, H.A.L., Hogervorst, J.G.F., Briede, J.J., 2006. Toxicological assess-ment of ambient and traffic-related particulate matter: a review of recentstudies. Mutation Research e Reviews in Mutation Research 613, 103e122.

Deltell, A., 2005. Elaboració del mapa acústic de trànsit de la ciutat de Girona.University of Girona, Spain (in Catalan).

DGT, Direccion General de Trafico. Available from: http://www.dgt.es/portal/es/seguridad_vial/estadistica (accessed September 2009).

Dockery, D.W., 2009. Health effects of particulate air pollution. Annals of Epide-miology 19, 257e263.

Efron, B., Tibshirani, R., Tibshirani, R.J., 1993. An Introduction to the Bootstrap.Chapman & Hall, New York.

Fernandez, J.A., Aboal, J.R., Carballeira, A., 2004. Identification of pollution sourcesby means of moss bags. Ecotoxicology and Environmental Safety 59, 76e83.

Fernández, J.Á., Aboal, J.R., Real, C., Carballeira, A., 2007. A new moss biomonitoringmethod for detecting sources of small scale pollution. Atmospheric Environ-ment 41, 2098e2110.

Goodman, G.T., Roberts, T.M., 1971. Plants and soils as indicators of metals in the air.Nature 231, 287e292.

Harmens, H., Norris, D.A., Steinnes, E., Kubin, E., Piispanen, J., Alber, R.,Aleksiayenak, Y., Blum, O., Coskun, M., Dam, M., De Temmerman, L.,Fernandez, J.A., Frolova, M., Frontasyeva, M., Gonzalez-Miqueo, L.,

Grodzinska, K., Jeran, Z., Korzekwa, S., Krmar, M., Kvietkus, K., Leblond, S.,Liiv, S., Magnusson, S.H., Mankovska, B., Pesch, R., Ruhling, A., Santamaria, J.M.,Schroder, W., Spiric, Z., Suchara, I., Thoni, L., Urumov, V., Yurukova, L.,Zechmeister, H.G., 2010. Mosses as biomonitors of atmospheric heavy metaldeposition: spatial patterns and temporal trends in Europe. EnvironmentalPollution 158, 3144e3156.

Harrison, R.M., Deacon, A.R., 1998. Spatial correlation of automatic air qualitymonitoring at urban background sites: implications for network design. Envi-ronmental Technology 19, 121e132.

Hoek, G., Beelen, R., de Hoogh, K., Vienneau, D., Gulliver, J., Fischer, P., Briggs, D.,2008. A review of land-use regression models to assess spatial variation ofoutdoor air pollution. Atmospheric Environment 42, 7561e7578.

INE, Instituto Nacional de Estadística. Available from: http://www.ine.es (accessedNovember, 2009).

Kassomenos, P., Karakitsios, S., Papaloukas, C., 2006. Estimation of daily trafficemissions in a South-European urban agglomeration during a workday. Eval-uation of several what if scenarios. Science of the Total Environment 370,480e490.

Kunzli, N., Jerrett, M., Mack, W.J., Beckerman, B., LaBree, L., Gilliland, F., Thomas, D.,Peters, J., Hodis, H.N., 2005. Ambient air pollution and atherosclerosis in LosAngeles. Environmental Health Perspectives 113, 201e206.

Kunzli, N., Tager, I.B., 2005. Air pollution: from lung to heart. Swiss Medical Weekly135, 697e702.

Li, N., Sioutas, C., Cho, A., Schmitz, D., Misra, C., Sempf, J., Wang, M., Oberley, T.,Froines, J., Nel, A., 2003. Ultrafine particulate pollutants induce oxidative stressand mitochondrial damage. Environmental Health Perspectives 111, 455e460.

Lipsett, M.J., Tsai, F.C., Roger, L., Woo, M., Ostro, B.D., 2006. Coarse particles andheart rate variability among older adults with coronary artery disease in theCoachella Valley, California. Environmental Health Perspectives 114, 1215e1220.

Makholm, M.M., Mladenoff, D.J., 2005. Efficacy of a biomonitoring (moss bag)technique for determining element deposition trends on a mid-range (375 km)scale. Environmental Monitoring and Assessment 104, 1e18.

Monn, C., 2001. Exposure assessment of air pollutants: a review on spatialheterogeneity and indoor/outdoor/personal exposure to suspended particulatematter, nitrogen dioxide and ozone. Atmospheric Environment 35, 1e32.

Moreno, T., Querol, X., Alastuey, A., Viana, M., Salvador, P., de la Campa, A.S.,Artinano, B., de la Rosa, J., Gibbons, W., 2006. Variations in atmospheric PMtrace metal content in Spanish towns: illustrating the chemical complexity ofthe inorganic urban aerosol cocktail. Atmospheric Environment 40, 6791e6803.

Naszradi, T., Badacsonyi, A., Keresztenyi, I., Podar, D., Csintalan, Z., Tuba, Z., 2007.Comparison of two metal surveys by moss Tortula ruralis in Budapest, Hungary.Environmental Monitoring and Assessment 134, 279e285.

Nel, A.E., Diaz-Sanchez, D., Li, N., 2001. The role of particulate pollutants inpulmonary inflammation and asthma: evidence for the involvement of organicchemicals and oxidative stress. Current Opinion in Pulmonary Medicine 7,20e26.

Pérez, C., 2001. El sistema estadístico SAS. Prentice Hall, Madrid.PMU-Girona, Pla de mobilitat urbana PMU del municipi de Girona e Urban mobility

plan for Girona, 2007. Available from: http://www.girona.cat/mobilitat/plamobilitat.html (accessed 2008e2009).

Ratcliffe, J.M., 1975. An evaluation of the use of biological indicators in an atmo-spheric lead survey. Atmospheric Environment 9, 623e629.

Reimann, C., Niskavaara, H., Kashulina, G., Filzmoser, P., Boyd, R., Volden, T.,Tomilina, O., Bogatyrev, I., 2001. Critical remarks on the use of terrestrial moss(Hylocomium splendens and Pleurozium schreberi) for monitoring of airbornepollution. Environmental Pollution 113, 41e57.

Rühling, A., Steinnes, E., 1998. Atmospheric Heavy Metal Deposition in Europe1995e1996. Nordic Council of Ministers, Copenhagen.

Ruhling, A., Tyler, G., 1970. Sorption and retention of heavy metals in woodlandmoss Hylocomium splendens (Hedw.) Br et Sch. Oikos 21, 92e97.

Simkhovich, B.Z., Kleinman, M.T., Kloner, R.A., 2008. Air pollution and cardiovas-cular injury: epidemiology, toxicology, and mechanisms. Journal of the Amer-ican College of Cardiology 52, 719e726.

Targa, J., Loader, A., The Defra working group on harmonisation of diffusion tubes,2008. Diffusion tubes for ambient NO2 monitoring: practical guidance forlaboratories and users. AEA Energy and Environment.

Tretiach, M., Adamo, P., Bargagli, R., Baruffo, L., Carletti, L., Crisafulli, P., Giordano, S.,Modenesi, P., Orlando, S., Pittao, E., 2007. Lichen and moss bags as monitoringdevices in urban areas. Part I: influence of exposure on sample vitality. Envi-ronmental Pollution 146, 380e391.

UMAT. Servidor de Cartografia, 2008. Available from: http://www.ajuntament.gi/umat/cat/cartografia.php. (Accessed 2008e2010).

Vasconcelos, M.T.S.D., Tavares, H.M.F., 1998. Atmospheric metal pollution (Cr, Cu, Fe,Mn, Ni, Pb and Zn) in Oporto city derived from results for low-volume aerosolsamplers and for the moss Sphagnum auriculatum bioindicator. The Science ofthe Total Environment 212, 11e20.

Viskari, E.L., Rekilä, R., Roy, S., Lehto, O., Ruuskanen, J., Kärenlampi, L., 1997. Airbornepollutants along a roadside: assessment using snow analyses and moss bags.Environmental Pollution 97, 153e160.

Wang, H.K., Chen, C.H., Huang, C., Fu, L.X., 2008. On-road vehicle emission inventoryand its uncertainty analysis for Shanghai, China. Science of the Total Environ-ment 398, 60e67.

WHO, 2006. Health risks of particulate matter from long-range transboundary airpollution. European Centre for Environment and Health, Bonn Office. WorldHealth Organization, Europe.

M. Rivera et al. / Environmental Pollution 159 (2011) 954e962962

Yeatts,K., Svendsen, E., Creason, J., Alexis,N.,Herbst,M., Scott, J., Kupper, L.,Williams,R.,Neas, L., Cascio, W., Devlin, R.B., Peden, D.B., 2007. Coarse particulate matter(PM2.5e10) affects heart rate variability, blood lipids, and circulating eosinophilsin adults with asthma. Environmental Health Perspectives 115, 709e714.

Zechmeister, H.G., Dullinger, S., Hohenwallner, D., Riss, A., Hanus-Illnar, A., Scharf, S.,2006. Pilot study on road traffic emissions (PAHs, heavy metals) measured byusing mosses in a tunnel experiment in Vienna, Austria. Environmental Scienceand Pollution Research 13, 398e405.

Zechmeister, H.G., Grodzinska, K., Szarek-Lukaszewska, G., 2003. Bryophytes. In:Markert, B.A., Breure, A.M., Zechmeister, H.G. (Eds.), Bioindicators and

Biomonitors: Principles, Concepts, and Applications. Elsevier, Amsterdam, pp.329e375 (Chapter 10).

Zechmeister, H.G., Hohenwallner, D., 2006. A comparison of biomonitoring methodsfor the estimation of atmospheric pollutants in an industrial town in Austria.Environmental Monitoring and Assessment 117, 245e259.

Zhang, Y., Stedman, D.H., Bishop, G.A., Guenther, P.L., Beaton, S.P., 1995. Worldwideon-road vehicle exhaust emissions study by remote-sensing. EnvironmentalScience and Technology 29, 2286e2294.

Zhou, Y., Levy, J., 2007. Factors influencing the spatial extent of mobile source airpollution impacts: a meta-analysis. BMC Public Health 7, 89.