S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

ava i l ab l e a t www.sc i enced i rec t . com

www.e l sev i e r. com/ loca te / sc i to tenv

Impact and uncertainty of a traffic management intervention:Population exposure to polycyclic aromatic hydrocarbons

Sotiris Vardoulakisa,⁎, Zaid Chalabia, Tony Fletchera, Chris Grundya, Giovanni S. Leonardib

aPublic & Environmental Health Research Unit, London School of Hygiene & Tropical Medicine, Keppel Street, London WC1E 7HT, UKbCentre for Radiation, Chemical, and Environmental Hazards, Health Protection Agency, Chilton, Didcot, Oxfordshire OX11 0RQ, UK

A R T I C L E I N F O

⁎ Corresponding author. Tel.: +44 2079272442;E-mail address: sotiris.vardoulakis@lshtm.a

0048-9697/$ – see front matter © 2008 Elsevidoi:10.1016/j.scitotenv.2008.01.037

A B S T R A C T

Article history:Received 5 November 2007Received in revised form11 January 2008Accepted 16 January 2008Available online 4 March 2008

In urban areas, road traffic is a major source of carcinogenic polycyclic aromatichydrocarbons (PAH), thus any changes in traffic patterns are expected to affect PAHconcentrations in ambient air. Exposure to PAH and other traffic-related air pollutants hasoften been quantified in a deterministic manner that disregards the various sources ofuncertainty in themodelling systems used. In this study, we developed a generic method forhandling uncertainty in population exposure models. The method was applied to quantifythe uncertainty in population exposure to benzo[a]pyrene (BaP) before and after theimplementation of a traffic management intervention. This intervention would affect themovement of vehicles in the studied area and consequently alter traffic emissions, pollutantconcentrations and population exposure.Several models, including an emission calculator, a dispersion model and a GeographicInformation Systemwere used to quantify the impact of the traffic management intervention.We established four exposure zones defined by distance of residence postcode centroids frommajor road or intersection. A stochastic method was used to quantify the uncertainty in thepopulation exposuremodel. Themethod characterises uncertainty using probabilitymeasuresand propagates it applying Monte Carlo analysis. The overall model predicted that the trafficmanagement scheme would lead to a minor reduction in mean population exposure to BaP inthe studied area. However, the uncertainty associated with the exposure estimates was muchlarger than this reduction. The proposedmethod is generic and provides realistic estimates ofpopulation exposure to traffic-related pollutants, as well as characterises the uncertainty inthese estimates. Thismethod can beusedwithin a decision support tool to evaluate the impactof alternative traffic management policies.

© 2008 Elsevier B.V. All rights reserved.

Keywords:Environmental exposurePAHUrban air pollutionSpatial variabilityParametric uncertaintyDispersion model

1. Introduction

Increasing attention has been drawn to population exposure totraffic-related air pollutants, due to the decline of industrialemissions in developed countries and the relatively large num-bers of residents exposed to exhaust gases and particles in urbanareas. Among these pollutants, several polycyclic aromatichydrocarbons (PAH) with relatively high carcinogenic potencyare causing concern (Armstrong et al., 2004). A target value of

fax: +44 2075804524.c.uk (S. Vardoulakis).

er B.V. All rights reserved

0.25 ng/m3 for benzo[a]pyrene (BaP) in ambient air (annualaverage to be achieved by the end of 2010) has been set in theUK National Air Quality Strategy, as a marker for the group ofPAH. Furthermore, the EuropeanCommunity's fourthAir QualityDaughter Directive (2005/107/EC) has set a target value of 1 ng/m3

for BaP (annual average).PAH are a mixture of chemical species arising from a variety

of combustion processes, with road traffic being a major sourcein urban areas. Since road traffic has been recognised as the

.

245S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

main source of air pollution in cities, traffic planning and man-agement schemes have been increasingly implemented in orderto reduce population exposure to vehicle exhausts as well ascongestion.

Traffic density and/or distance from main roads have beencommonly used as indicators of exposure to primary traffic-related air pollution in epidemiological studies. The simplestapproach is touse circular buffer zones to define exposure levelsaround residence locations of cases and controls. For example,English et al. (1999) constructed a circular buffer zone of 169 mradius around geocoded residences as an approximation ofthe area contributing to air pollution exposure of individuals.Wilkinson et al. (1999) used thedistance to nearestmain road asameasure of exposure of childrenwith asthma to traffic-relatedpollution. In this study, a radius of 150 m around the residencepostcode centroid was used as the spatial limit of traffic expo-sure. In a similar study, Gordian et al. (2006) used road segmentlengths multiplied by traffic volume to create a weighted mea-sure of traffic-related exposure within a circular buffer zone of100 m around residences. However, the circular buffer zoneapproach does not take into account atmospheric processesthat may affect pollutant dispersion, transformation andremoval. For that reason, severalauthorshaveusedair pollutionmonitoring andmodelling techniques to estimate the impact ofdifferent road traffic interventions on air quality (Torp andLarssen, 1996; Beevers and Carslaw, 2005; Owen, 2005).

Several markers of personal exposure to traffic-related airpollution have been used in the past to assess health risks forchildren and/or adults. NO2 has been commonly used as anexposure indicator due to its well established health effects andthe relatively easy access to availablemonitoring data and emis-sion factors (Raaschou-Nielsen et al., 1996). PAH are in generalpoorly characterised in terms of emissions strength and spatialdistribution, due to the limited amount of experimental dataavailable in most countries (Chen and Liao, 2006). In this study,our objective is to develop a probabilistic modelling method forquantifying the uncertainty in population exposure to traffic-related pollutants, going beyond common deterministic simula-tions.We illustrate thismethod by estimating exposure levels toBaP before and after the implementation of stopping/parkingrestrictions in Stratford Road in Birmingham (UK).

2. Methods

2.1. Location and data used

Stratford Road (A34) is a major road axis linking central Birming-hamwith theM42 ring roadof the city in the southeast. It is a verybusy street passing through residential and shopping areas, withthree to four-storey buildings almost continuously lining bothsides of the street. The total width of the street is approximately22 m and the average building height 12–14 m. The annual ave-ragedaily traffic (AADT) flow is around29,000vehicles/day.Due tocongestion, the average vehicle speed ranges from 10 to 25 km/h,with many stop-starts near junctions.

An air quality monitoring station was operated in StratfordRoad measuring continuously CO and PM10 with a standardinfrared analyser and a TEOM monitor, respectively. AlthoughPAH concentrations were not measured in Stratford Road, we

obtained background BaP data from a permanent suburbanmonitoring station (Birmingham East, National Air QualityArchive, www.airquality.co.uk) at approximately 6 km distancefrom the monitoring location in Stratford Road. A modifiedAnderson GP1-S sampler, capturing both gas and particle phasePAHonglass fibre andpolyurethane filterswasusedat this site tocollect samples over 15-day exposure periods. Hourly CO, PM10

andNOXdatawerealsoobtained fromthesamesiteandaveragedover 15-day periods. Air quality monitoring, sampling andanalysis in Birminghamfollowed the standardQA/QCproceduresof the UK national air quality monitoring network.

We obtainedmeteorological data includingwind speed anddirection, temperature and atmospheric pressure from theMet Office weather station in Coleshill and global radiationdata from the University of Birmingham (approximately 11 kmand 4 km from Stratford Road, respectively). Road traffic datawere obtained from Birmingham City Council and populationcounts in each postcode from the 2001 Census Standard AreaStatistics (for England and Wales).

2.2. Dispersion modelling

The above information was used to develop and apply a geogra-phic modelling method, based on a combination of an emissioncalculator, an atmospheric dispersion model, an empirical dis-tancemodel, and a stochastic approach for handling uncertainty.

BaP was used as a marker for the group of PAH. CompositeBaP emission factors for the baseline (pre-intervention) scenariowere estimated using available information on vehicle type-specific BaP emission factors for cars runningonunleadedpetrol(with and without catalytic converter), diesel cars, light andheavy duty vehicles, and buses (Wenborn et al., 1999). Due to thelimited information on specific BaP emission factors for 2003,weused the well-characterised PM10 emission factor profiles forspecific vehicle types and speeds embedded in the EmissionFactors Toolkit (EFT version 2e, Casella Stanger 2003) to derivecomposite BaP emission factors for that year. Total BaPemissions (i.e. particle-bound and gas-phase) have been takeninto account, although it is thought that carcinogenic PAHroughly correspond to particle-bound PAH (Levy et al., 2001).Over 90%of totalBaP is expected tobe found in theparticlephasein roadside environments, which makes it an appropriateindicator for carcinogenic PAH (Lee et al., 1995).

In order to predict post-intervention BaP concentrations,we assumed that the vehicle speed in Stratford Road willincrease by an average of 12%, as predicted by the WestMidlands traffic simulation model. By contrast, the AADT isnot expected to be affected by the Red Route scheme accordingto results from the same model. Nevertheless, we expectthe uncertainty associated with the AADT to be higher afterthe implementation of the Red Route scheme than before theintervention, due to unforeseen changes in drivers' behaviour.

TheOperationStreet PollutionModel (WinOSPM)was used tosimulate BaP levels in Stratford Road. This is a semi-empiricaldispersion model that calculates total pollutant concentrationsonboth sidesof a street canyonbyadding the roadside (CR) to theurban background contribution (CB). Gaussian plume and boxmodel theory are applied for the calculation of the roadsidecontribution, using the street dimensions, meteorological data(wind speed and direction, air temperature and solar radiation),

Fig. 1 – Exposure buffer zone and postcode centroids aroundStratford Road.

246 S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

road traffic information (hourly traffic volume, average vehiclespeed and emission factors) as input data. A more detaileddescription of the model is given elsewhere (Berkowicz, 2000;Vardoulakis et al., 2003). A recent evaluation of WinOSPMagainst PM10 and CO monitoring data from Stratford Roadshoweda very good level of agreement,with Pearson correlationcoefficients and normalizedmean-square errors of 0.82 and 0.19for PM10, and 0.75 and 0.29 for CO (Vardoulakis et al., 2007).

2.3. Spatial variability

Several experimental studies have demonstrated the high spatialvariability of air pollution in urban areas, mainly in the vicinity ofbusy streets (Croxford et al., 1996; Vardoulakis et al., 2005). How-ever, only a few studies have focused on the spatial variability ofPAH in roadside environments (Lee et al., 1995;Nielsenet al., 1996;Levyetal., 2003). For that reason, studiesonthesmall-scale spatialvariability of particulate matter (PM) and NO2 have been alsoreviewed (Monnet al., 1997;Weijers et al., 2004). It shouldbenotedthat the chemical composition of PM can vary substantially withdistance from the road, as kerbside PM is expected to have largercontributions of traffic-related species, such as black carbon andPAH. Similarly, concentration gradients of NO2 can only provide arough indication of PAH variability, since their atmospheric life-times are different than those of PAH. The most relevant studiesin terms of PAH spatial variability are discussed in this section.

Levy et al. (2003) carried out PAH measurements at variedupwind and downwind distance (5–75m) from several roads inRoxbury, Massachusetts (USA). They identified strong hori-zontal gradients for particle-bound PAH (i.e. concentrationsgenerally decreased with distance from the kerb), especiallyon the downwind side of the road. Lee et al. (1995) measuredPAH concentrations at three distances (3, 30 and 100m) from atraffic source in southern Taiwan. The reported PAH composi-tion varied substantially with distance from the kerb, with BaPmass adsorbed on airborne particles being approximately 2.4times higher at 3 m than at 100 m.

Weijers et al. (2004) experimentally determined the spatialvariability of particle number concentrations (PNC), andparticles with aerodynamic diameter above and below 1 μm(PMN1 and PM1 respectively) in roadside, urban and coastalenvironments in the Netherlands using an instrumented van.According to their results, both particle number and massconcentrations decay exponentially with distance away fromthe road, with the decay being larger in the case of PMN1. Thisis expected as finer particles (PM1) are likely to travel longerdistances due to lower deposition velocities. Weijers et al.(2004) found that the distance from the road over which thePM1 roadside enhancement had decreased by 90%was around120 m. The estimated spatial decay parameters for PNC, PM1

and PMN1 were 0.01, 0.02 and 0.03 m−1, respectively.Since most BaP is bound on fine particles, we have as-

sumed that BaP concentration (C) decays exponentially withdistance x (20 mbxb150 m) from Stratford Road, according tothe relationship:

C xð Þ ¼ CR �exp�d� x� D0ð Þ½ � þ CB ð1Þ

whereC(x) is the total BaP (ng/m3) atdistancex (m) fromthekerb,CR is the BaP (ng/m3) roadside enhancement (i.e. total concen-trationminus the background concentration) inStratfordRoad, δ

is the decay parameter (δ=0.03 m−1), D0 is the initial distanceafter which the concentration decays exponentially (D0=20 m),and CB is the BaP (ng/m3) background concentration outside thebuffer zone (i.e. at a distance from the road greater than 150 m).According to this relationship, BaP concentrations have reachedbackground levels at approximately 150 m away from the road.Apart from the distance away from the kerb, we have alsoconsidered the distance from major intersections along Strat-ford Road as an additional exposure determinant (Beyea et al.,2006). This is because vehicles in acceleration/decelerationmode emit exhaust pollutants at higher rates. Based on theabove assumptions, the following four PAH exposure categorieshave been established:

(I) Very high exposure levels for residents living in front lineproperties in Stratford Road, within a distance less than100 m from the nearest major intersecting street (AADTN

10,000 vehicles/day):

CI ¼ e� CR þ CB ð2Þ

where ε=1.25 and CI is the BaP concentration in the firstexposure zone.

Table 1 – Dispersion model input parameters

Variable Baseline Red Route

Meanestimate

Variance Meanestimate

Variance

Emission factor(μg/veh km)

3.823 0.7132 3.566 0.4841

Background(ng/m3)

0.165 0.0025 0.165 0.0025

AADT(veh/day)

28,930 7.5728×106 28,930 13.468×106

247S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

(II) High exposure levels for residents living in front line pro-perties in Stratford Road at a distance greater than 100 mfrom the nearest major intersecting street (AADTN10,000vehicles/day):

CII ¼ CR þ CB ð3Þ

where CII is the BaP concentration in the second exposurezone.

(III) Moderate exposure levels for residents living beyondthe first line of houses in Stratford Road, but within150 m distance from the middle of this road in bothdirections, obtained by integrating the right hand side ofEq. (1) between x=D0 and x=D:

CIII ¼ 1D� D0

Z D

D0

CR �exp�d� x� D0ð Þ½ � þ CBð Þ � dx ð4Þ

where CIII is the BaP concentration in the third exposurezone.Assuming that D0=20 m and D=150 m, relationship (4)becomes:

CIII ¼ z� CR þ CB ð5Þ

where z=0.25.

(IV) Low exposure levels for residents living in properties atdistances greater than 150 m away from Stratford Road,which corresponded to the urban background PaB levelsobserved at Birmingham East monitoring site:

CIV ¼ CB ð6Þ

where CIV is the BaP concentration in the fourth expo-sure zone.

We assume that all residents in each zone are exposed to thesameBaP levels. The sizeof the exposedpopulation ineach zoneis determined by summing the number of residents within thezone. Fig. 1 displays a Geographical Information System (GIS,ArcView3) generatedplotof StratfordRoadand its surroundings.

Table 2 – Number of residents in exposure zones I–III

Zone Description Postcodes

I Junction 20II Street canyon 75III Surrounding 409

The 150 m buffer zone is shown as the shaded area around thecontinuous line which represents Stratford Road. The crossesgive the locations of major junctions and the dots give thelocations of the population-weighted postcode centroids (i.e. thegeographic centre of eachpostcode). It should benoted that eachpostcode refers to a block of flats or a number of houses and thepostcodes contain different population sizes. In the area of ourstudy, residentsnumbers ranged from0 (non-residentialproper-ties) to 323 per postcode.

2.4. Parametric uncertainty

In this study, we only deal with the uncertainty in the para-meters of the overall population exposure model (comprisingthe air dispersionmodel and the spatiallydistributedpopulationsize model) ignoring any uncertainty associated with thestructure of the system. Four major sources of parametricuncertainty have been identified: (a) the BaP emission factors,due to the limited information available on vehicle type andspeed-specific BaP emissions; (b) the traffic volume in StratfordRoad, due to the limited amount of traffic monitoring datacomplicated further by a large number of intersecting streets;(c) the background BaP concentrations, due to the low spatialresolutionof availablemeasurements; (d) the sizeof the residentpopulation ineachof theexposure zones, due to theallocationofresidents numbers to postcode centroids.

The uncertainty analysis comprises the following twosteps: (i) The characterisation of the uncertainty in each ofthe four parameters. (ii) Thepropagationof thoseuncertaintiesthrough the overall model. With regards to the first step, theuncertainty in each of the parameters is characterised using aprobability density function (PDF). Without loss of generality,log-normal distributions are used to characterise the uncer-tainty in emission factors, vehicle density and BaP backgroundlevels, while uniform distributions were used to characterisethe uncertainty in the numbers of residents within each expo-sure zone. Other statistical distributions could have been usedinstead, if supported by empirical evidence.

We obtained approximate estimates of the log-normal distri-butionsof theuncertainparameters (a), (b) and (c) for thepre- andpost-intervention scenarios from available emission factors,traffic volumes and QA/QC information (Table 1). The meanestimatesandvariancesof theemission factors inparticularwerecalculated from minimum, maximum and best estimate BaPemission factors reported by Wenborn et al. (1999) for differentfuel and vehicle types (with andwithout catalytic converter), andprojected to the year 2003 using the EFT. The centres of theuniform distributions of the uncertain parameter (d) were deter-mined from the 2001 Census Standard Area Statistics. Further-more, we performed a simple sensitivity analysis on thedimensions of the buffer zones around Stratford Road in orderto estimate the ranges of the uniform distributions (Table 2).

Houses Population Range

94 232 [201,295]696 1639 [1576,1670]

2775 9341 [6319,12127]

Fig. 2 – Pre-intervention probability distributions of BaP concentrations in exposure zones I–IV.

248 S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

The second step usesMonte Carlo simulations to propagatethe uncertainty in the model parameters to the model out-come. Assuming that the four parameters are uncorrelated,each parameter is sampled independently using its probabilitydensity function (PDF). A histogram of the model outcome isconstructed from the outputs of the simulation runs. Properlynormalized, the histogram provides an approximation of thePDF of the overall model output.

Fig. 3 – Post-intervention probability distributions

3. Results

Initially, we examined the correlation between BaP and othertraffic-related pollutants observed at the Birmingham Eastmonitoring site during a period of 5 years (2001–2006). Therewas evidence that BaP was mainly related to road transportsources, as it correlated strongly with CO (R=0.84) and NOX

of BaP concentrations in exposure zones I–IV.

Table 3 – Pre- and post-intervention BaP exposure levels

Zone Description Baseline Red route Change

Median STDV Median STDV

I Junction 0.634 0.124 0.585 0.100 0.048 (7.9%)II Street canyon 0.543 0.104 0.502 0.085 0.041 (7.8%)III Surrounding 0.261 0.061 0.245 0.053 0.017 (6.7%)IV Background 0.161 0.059 0.157 0.052 0.005 (2.8%)

249S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

(R=0.87),whichareboth indicators of vehicle emissions (Harrisonet al., 2003). A weaker correlation was observed between BaPand PM10 (R=0.54), which is consistent with the more diversesources of PM in theurban environment (e.g. vehicle exhausts,re-suspended dust, secondary particles, etc.). In an earlierstudy, Lim et al. (1999) demonstrated that road traffic con-tributed 80–82% of 4–7 ring PAH in Birmingham city centre.Finally, we found that pollutant concentrations in StratfordRoad correlated reasonablywell with concentrations obtainedfrom the Birmingham East monitoring station in 2003 (R=0.56for CO and R=0.70 for PM10).

The simulation results were obtained by running the pop-ulation exposure model 100 times for the pre-interventionscenario and 100 times for the post-intervention scenario. Alarger number of simulation runs could be used, following thesamemethodology, but in practice thenumber of simulations islimited by the computational time taken to execute eachmodelrun. For each simulation, independent combinations of para-meter values were used in the air dispersion and spatiallydistributed population size model. The parameters were ob-tained by randomly sampling their values from their respectivePDFs.

Figs. 2 and 3 show respectively the histograms of the pre-and post-intervention modelled annual mean BaP concen-trations in each of the four exposure zones. The histogramsare presented in 10 bins of equal size (spanning the rangefrom 0.0 to 1.0 ng/m3). The median concentrations indicatethat the implementation of the Red Route scheme in Strat-ford Road is predicted to bring about a reduction in annualmean BaP concentrations of 6.7–7.9% in the directly affectedareas (zones I–III) (Table 3). These reductions were foundto be statistically significant at 99% confidence interval for

Fig. 4 – Pre- and post-intervention probability distribut

zones I and II, and at 90% confidence interval for zone III, using atwo-tailed t-test. Although the estimated reductions were sta-tistically significant, the uncertainty in these estimates wastypically more than twice the change in concentrations, indi-cating that the estimated uncertainty was much greater thanthe reductions. As expected, the pre- and post-intervention BaPlevels in zone IV were not statistically different.

The roadside/urban background BaP ratio in Birminghamwas 3.37 and 3.20 for the baseline and Red Route scenarios,respectively. These values are in very good agreement with theratio of 3.14 observed during a winter sampling campaign inCopenhagen (Nielsen et al., 1996), despite the large differences inabsolute BaP levels between the two settings.

In order to compare total population exposure levels aroundStratford Road, we calculated the population-weighted averageBaP concentration (Iexp):

Iexp ¼

Xi

Ci � PiXi

Pið7Þ

where Ci is the total annual mean pre- or post-interventionpollutant concentration in exposure zone i and Pi is the pop-ulation size in the same zone. Zone IV data were excluded fromthe Iexp calculation, because the pre- and post-intervention Ci

values were not statistically different. Monte Carlo simulationswere also used to propagate the uncertainty in Ci and Pi to theuncertainty in the population-weighted average concentrationfor both the pre- and post-intervention scenarios. The pre- andpost-intervention histograms for Iexp are presented in Fig. 4,showing a 7.4% reduction of the median population-weightedexposure after the Red Route intervention.

ions of population-weighted exposure to BaP (Iexp).

250 S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

4. Discussion

The modelling methodology followed in this study providedexposureestimates that canbe readilycomparablewithexistingambient air quality objectives, without disregarding the sub-stantial uncertainty associated with the model input para-meters. The simulated annualmean BaP concentrations rangedfrom 0.068 to 0.985 ng/m3 for the baseline (pre-intervention)scenario and from 0.071 to 0.873 ng/m3 for the Red Route (post-intervention) scenario, showing a small overall reduction inconcentrations after the introduction of the Red Route scheme.

In the present study, we focused only on a relatively smallstretch of Stratford Road in order to demonstrate our metho-dology. However, the same approach can be used to estimatepopulation exposure to PAH in the larger area affected by theRed Route scheme or other traffic management intervention.We used the population-weighted average concentration tocompare exposure levels in cases where changes in popula-tion densitymay occur. It should be noted that this measure isnormalised with respect to the total size of the affected pop-ulation, which means that only changes in the distribution ofresidents within exposure zones I–III may have an impact onthe measure.

Deterministic modelling methods do not account for themany uncertainties associated with model parameters. Para-metric uncertainty is often disregarded in scenario simula-tions, due to the complexity and high operational cost of thedispersion models that are commonly used. In this study, wehave developed and illustrated a probabilistic methodology toestimate changes in population exposure to BaP, taking para-metric uncertainty into account. The uncertainty related tothe structure of the model has not been quantified. This me-thodology is generic and can be applied to estimate populationexposures to a wide range of pollutants using different dis-persion and population density models.

More complex regional or urban-scale dispersionmodels canbe used to estimate population exposure to other air pollutants,such as NOX, which are well-characterised in the urban envi-ronment (Physick et al., 2007). However, if the uncertainty in theemissions is large, as in our case study, and only a limitedamount of monitoring data are available for model verification,then a simpler modelling methodology that takes uncertaintyinto account is more appropriate (Beyea and Hatch, 1999). Theaccuracy of the population exposure modelling is in any casebounded by the accuracy of the overall model parameters suchas vehicle emission factors, local traffic data, background pollu-tion concentrations, meteorological data, information on localtopography and population distribution, which are rarely avail-able in detail (Lobscheid and McKone, 2004). Clearly, more BaPmonitoring data near roadways with variable traffic counts andvehicle speeds would help reduce the uncertainty of the trafficpollution model exposure estimates.

We assumed that BaP concentrations decay exponentiallywith distance from the road, which is a reasonable assumptionaccording to previous studies (Beyea and Hatch, 1999; Sioutaset al., 2005). Vertical pollution gradients may have a significanteffectonpopulationexposure inbuilt-upareas (Vardoulakisetal.,2002), but they were not taken into account in this study due tothe relatively small height of residential buildings surrounding

Stratford Road. For simplicity, we allocated population counts todiscrete exposure zones. Although there is some criticism onusing discrete exposure zones such as road buffers (Jerrett et al.,2005;Molitor et al., 2007), we believe that the estimated change inexposure levels due to the Red Route intervention would not besignificantly affected by the use of buffer zones.

Geographic referencing facilitates data visualisation andanalysis. However, the use of postcode centroids to estimatepopulation exposure is based on the assumption that allresidential properties with the same postcode lie within thesame exposure zones. This approximationmight not be realisticin less densely populated areas, where the same postcode maybesharedby residents livingat considerabledistances fromeachother. Another significant source of uncertainty may be thedifferences between day- and night-time numbers of residents,which are not reflected in the census data. It should be kept inmind that time–activity patterns of residents living within themodelling domain can vary widely. Other authors have demon-strated that address geocoding can introduce substantialuncertainty in epidemiological studies (Ward et al., 2005;Zandbergen, 2007).

Finally, we assumed that the indoor/outdoor ratio of BaPconcentrations was approximately equal to 1. Several experi-mental studies have demonstrated that indoor concentrationsof traffic-related pollutants, such as NO2, in non-smokinghouseholds are strongly correlatedwithoutdoor concentrations,which are dominated by road traffic emissions (Singer et al.,2004). Fromme et al. (2004) analysed for PAH indoor and outdoorair samples of respirable dust obtained inBerlin, Germany. Theirresults suggested that indoor PAH in non-smoking environ-ments could bemainly attributed to road traffic emissions, withindoor to outdoor BaP concentration ratios of 0.9. This ratio wasin good agreement with results reported from a PAH samplingsurvey carried out in Chicago, USA (Li et al., 2005).

5. Conclusions

In this paper, we have presented a probabilistic method forquantifying the uncertainty in population exposure to airbornePAH (using BaP as a marker) and plausible changes in exposurelevels due to the implementation of a traffic management inter-vention which is expected to reduce congestion in amain trafficaxis near Birmingham (UK). Themethodwas based on the deve-lopment of a modelling framework that combined atmosphericemission and dispersion models, GIS and population densitymaps. A stochastic approachwasused to characterise the uncer-tainty in the model parameters and to propagate it throughoutthemodel, so as to obtain realistic uncertainty ranges associatedwith the different exposure levels.

The simulations predict a marginal reduction in PAH expo-sure levels within uncertainty bounds in Stratford Road due tothe implementation of the Red Route scheme. We believe thatthe quantification of exposure levels in the form of probabilitydistributions is necessary towards providing a comprehensivedecision support tool for policy makers. The information onuncertainty in population exposure can be integrated with in-formation on uncertainty in other criteria of relevance to policymakers for the comparative evaluation of traffic managementoptions using multi-criteria decision analysis.

251S C I E N C E O F T H E T O T A L E N V I R O N M E N T 3 9 4 ( 2 0 0 8 ) 2 4 4 – 2 5 1

Acknowledgements

This studywas carried out as part of the Pollutants in the UrbanEnvironment (PUrE) project funded by EPSRC Sustainable UrbanEnvironment Programme. We are grateful to our partners inPUrE for their support; Dr. Ben Armstrong (LSHTM) for com-ments on the manuscript; Birmingham City Council, NationalAir Quality Information Archive, and British Atmospheric DataCentre for the datasets provided. The census data are Crowncopyright reproduced with the permission of HMSO. Any viewsor opinions presented in this paper are those of the authors anddo not represent the views of the United Kingdom HealthProtection Agency.

R E F E R E N C E S

Armstrong B, Hutchinson E, Unwin J, Fletcher T. Lung cancer riskafter exposure to polycyclic aromatic hydrocarbons: a reviewand meta-analysis. Environ Health Perspect 2004;112(9):970–8.

Beevers SD, Carslaw DC. The impact of congestion charging onvehicle emissions in London. Atmos Environ 2005;39:1–5.

Berkowicz R. OSPM — a parameterised street pollution model.Environ Monit Assess 2000;65:323–31.

Beyea J,HatchM.Geographicexposuremodeling: avaluableextensionof geographic information systems for use in environmentalepidemiology. Environ Health Perspect 1999;107:181–90.

Beyea J, Hatch M, Stellman SD, Santella RM, Teitelbaum SL,Prokopczyk B, et al. Validation and calibration of a model usedto reconstruct historical exposure to polycyclic aromatichydrocarbons for use in epidemiologic studies. Environ HealthPerspect 2006;114:1053–8.

Chen SC, Liao CM. Health risk assessment on human exposure toenvironmental polycyclic aromatic hydrocarbons pollutionsources. Sci Total Environ 2006;366:112–23.

Croxford B, PennA,Hillier B. Spatial distribution of urbanpollution:civilizing urban traffic. Sci Total Environ 1996;189/190:3–9.

English P, Neutra R, Scaif R, Sullivan M,Waller L, Zhu L. Examiningassociations between childhood asthma and traffic flow usinga geographic information system. Environ Health Perspect1999;107:761–7.

Fromme H, Lahrz T, Piloty M, Gebhardt H, Oddoy A, Ruden H.Polycyclic aromatic hydrocarbons inside and outside ofapartments in an urban area. Sci Total Environ 2004;326:143–9.

Gordian ME, Haneuse S, Wakefield J. An investigation of theassociation between traffic exposure and the diagnosis ofasthma in children. J ExpoAnal Environ Epidemiol 2006;16:49–55.

Harrison RM, Tilling R, Callen Romero MS, Harrad S, Jarvis K. Astudy of trace metals and polycyclic aromatic hydrocarbons inthe roadside environment. Atmos Environ 2003;37:2391–402.

JerrettM,ArainA,KanaroglouP,BeckermanB,PotoglouD,SahsuvarogluT, et al. A reviewandevaluationof intraurbanair pollutionexposuremodels. J Expo Anal Environ Epidemiol 2005;15:185–204.

Lee WJ, Wang YF, Lin TC, Chen YY, Lina WC, Ku CC, et al. PAHcharacteristics in the ambient air of traffic-source. Sci TotalEnviron 1995;159:185–200.

Levy JI, Houseman E, Spengler J, Loh P, Ryan L. Fine particulatematter and polycyclic aromatic hydrocarbon concentrationpatterns in Roxbury, Massachusetts: a community-based gisanalysis. Environ Health Perspect 2001;109:341–7.

Levy JI, Bennett DH, Melly SJ, Spengler JD. Influence of trafficpatterns on particulate matter and polycyclic aromatichydrocarbon concentrations in Roxbury, Massachusetts. J ExpoAnal Environ Epidemiol 2003;13:364–71.

Li A, Schoonover TM, Zou Q, Norlock F, Conroy LM, Scheff PA, et al.Polycyclic aromatic hydrocarbons in residential air of ten

Chicago area homes: concentrations and influencing factors.Atmos Environ 2005;39:3491–501.

Lim LH, Harrison RM, Harrad S. The contribution of traffic toatmospheric concentrationsofpolycyclic aromatichydrocarbons.Environ Sci Technol 1999;33:3538–42.

Lobscheid AB, McKone TE. Constraining uncertainties about thesources and magnitude of polycyclic aromatic hydrocarbon(PAH) levels in ambient air: the state of Minnesota as a casestudy. Atmos Environ 2004;38:5501–15.

Molitor J, Jerrett M, Chang CC, Molitor NT, Gauderman J, Berhane K,et al. Assessing uncertainty in spatial exposure models for airpollution health effects assessment. Environ Health Perspect2007;115(8):1147–53.

Monn C, Carabias V, Junker M, Waeber R, Karrer M, Wanner HU.Small-scale spatial variability ofparticulatematterb10μm(PM10)and nitrogen dioxide (NO2). Atmos Environ 1997;31:2243–7.

Nielsen, T, Jorgensen, HE, Larsen, JC, Poulsen, M. City air pollution ofpolycyclicaromatichydrocarbonsandothermutagens:occurrence,sources and health effects. Sci Total Environ 1996;189/190:41–9.

Owen B. Air quality impacts of speed-restriction zones for roadtraffic. Sci Total Environ 2005;340:13–22.

Physick WL, Cope ME, Lee S, Hurley PJ. An approach for estimatingexposure to ambient concentrations. J Expo Sci EnvironEpidemiol 2007;17:76–83.

Raaschou-Nielsen O, Olsen JH, Hertel O, Berkowicz R, Skov H,Hansen AM, et al. Exposure of Danish children to trafficexhaust fumes. Sci Total Environ 1996;189/190:51–5.

Singer BC, Hodgson AT, Hotchi T, Kim JJ. Passive measurement ofnitrogen oxides to assess traffic-related pollutant exposure forthe East Bay Children's Respiratory Health Study. AtmosEnviron 2004;38:393–403.

Sioutas C, Delfino RJ, Singh M. Exposure assessment foratmospheric ultrafine particles (UFPs) and implications inepidemiologic research.EnvironHealthPerspect 2005;113:947–55.

Torp, C, Larssen, S. Modeling population exposure to airpollution near the road network of Norway, and the effects ofmeasures to reduce the exposure. Sci Total Environ 1996;189/190:35–40.

Vardoulakis S, Gonzalez-Flesca N, Fisher BEA. Assessment oftraffic-related air pollution in two street canyons in Paris:implications for exposure studies. Atmos Environ2002;36:1025–39.

Vardoulakis S, Fisher BEA, Pericleous K, Gonzalez-Flesca N.Modelling air quality in street canyons: a review. AtmosEnviron 2003;37:155–82.

Vardoulakis S, Gonzalez-Flesca N, Fisher BEA, Pericleous K. Spatialvariability of air pollution in the vicinity of a permanentmonitoring station in central Paris. Atmos Environ2005;39:2725–36.

Vardoulakis S, Valiantis M, Milner J, ApSimon H. Operational airpollution modelling in the UK — street canyon applicationsand challenges. Atmos Environ 2007;41:4622–37.

Ward MH, Nuckols JR, Giglierano J, Bonner MR, Wolter C, Airola M,et al. Positional accuracy of two methods of geocoding.Epidemiology 2005;16:542–7.

Weijers EP, Khlystov AY, Kos GPA, Erisman JW. Variability ofparticulate matter concentrations along roads and motorwaysdetermined by a moving measurement unit. Atmos Environ2004;38:2293–3002.

Wenborn, MJ, Coleman, PJ, Passant, NR, Lymberidi, E, Sully, J, WeirRA. Speciated PAH inventory for the UK. AEAT-3512/REMC/20459131. Abingdon, Oxfordshire: AEATechnology; 1999. August.

Wilkinson P, Elliott P, Shaddick G, Thakrar B, Walls P, Falconer S.Case-control study of hospital admission with asthma inchildren aged 5–14 years: relation with road traffic in northwest London. Thorax 1999;54:1070–4.

Zandbergen PA. Influence of geocoding quality on environmentalexposure assessment of children living near high traffic road.BMC Public Health 2007;7:37.