a cluster analysis of long range air transport pathways and associated pollutant concentrations...

9
A cluster analysis of long range air transport pathways and associated pollutant concentrations within the UK Jacob Baker Division of Environmental Health and Risk Management, School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK article info Article history: Received 16 July 2008 Received in revised form 17 October 2009 Accepted 21 October 2009 Keywords: Cluster analysis Air pollution Trajectory sets Ozone Midlands abstract A cluster analysis of four-day back trajectories for January 1998 to December 2001 arriving mid-afternoon in Birmingham, UK at three different within boundary layer arrival heights has been performed in order for a better understanding of the pollution meteorology inuencing this region. The time period was purposely chosen to encompass the Pollution in the Urban Midlands Atmosphere eld campaign. Six natural synoptic scale transport patterns were identied with three, strong-westerly, westerly and slow-easterly, showing seasonal variation in frequency. Signicant differences in air pollutant concentrations and behaviour were found between air mass cluster types when they were analysed with measurements taken from an urban background site in Birmingham and a rural site in Harwell. Highest concentrations of primary pollutants were associated with a slow-easterly air mass from mainland Europe, while lowest concentrations were associated with south-westerly and strong-westerly air masses passing over the Atlantic Ocean. The polluted slow-easterly air mass was associated with highest ozone concentration for the warm season and lowest ozone concentration for the cool season. This could be explained by photochemical ozone production during the warm season and NO x titration of background ozone during the cool season, when photo- chemically inactiveconditions prevailed. A wealth of information was determinable, including inuences of short and long-range transport and photochemical productivity. Ó 2009 Elsevier Ltd. All rights reserved. 1. Introduction The rationale for the present study was to better understand the conditions for which air pollution, in particular ozone episodes, occurred within the UK Midlands. In addition a better understanding of pollution meteorology was sought for placing the PUMA (Pollu- tion in the Urban Midlands Atmosphere) eld campaign (Harrison et al., 2006) into its more general context and for gaining insights that could lead to improvements in phototrajectory air quality models (Derwent et al., 1996; Walker et al., 2009; Baker, submitted for publication). This eld campaign, which took place during the summer of 1999 and winter of 1999/2000 was part of the National Environmental Research Council's Urban Regeneration and the Environment (URGENT) thematic programme. Ozone in the troposphere is formed from complex reactions involving volatile organic compounds (VOC) and nitrogen oxides (NOx). Its formation may take place over several hours or days and may be associated with precursor emissions hundreds, or even thousands of kilometres away. Thus both chemistry and atmo- spheric transport need to be understood for developing knowledge of the causes of ozone episodes. With an understanding of this pollution meteorology, the inuence of policy on meeting air quality objectives may then be assessed. Back trajectories from trajectory models, which make use of archived meteorological data, are potentially useful for identifying the source regions for pollutants measured at a receptor. However, indi- vidual trajectories have uncertainties associated with the resolution and accuracy of the meteorological data and by any simplifying assumptions used in the trajectory model, which ultimately limits their usefulness (Stohl, 1998). To some extent, this may be overcome by calculating a large number of back trajectories and subjecting them to a cluster analysis. This is a multivariate statistical technique which groups the individual trajectories of an ensemble into a smaller number of clusters, where the errors in the individual trajectories tend to average out. In this way sourceereceptor relationships may be determined in a statistical sense, helpful in developing an under- standing of the pollution climatology of a location. Cluster analysis of back trajectories has previously been used to support the interpretation of precipitation composition (Dorling et al., 1992a, 1992b), particulate matter (Wang et al., 2004; Abdal- mogith and Harrison, 2005; Borge et al., 2007; Cabello et al., 2008) and ozone (Moody et al., 1995; Sirois and Bottenheim, 1995; Brankov et al., 1998; Cape et al., 2000). The majority of such studies use back trajectories with a timescale of 3e5 days. This is generally E-mail address: [email protected] Contents lists available at ScienceDirect Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv 1352-2310/$ e see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2009.10.030 Atmospheric Environment 44 (2010) 563e571

Upload: jacob-baker

Post on 04-Sep-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

lable at ScienceDirect

Atmospheric Environment 44 (2010) 563e571

Contents lists avai

Atmospheric Environment

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

A cluster analysis of long range air transport pathways and associatedpollutant concentrations within the UK

Jacob BakerDivision of Environmental Health and Risk Management, School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK

a r t i c l e i n f o

Article history:Received 16 July 2008Received in revised form17 October 2009Accepted 21 October 2009

Keywords:Cluster analysisAir pollutionTrajectory setsOzoneMidlands

E-mail address: [email protected]

1352-2310/$ e see front matter � 2009 Elsevier Ltd.doi:10.1016/j.atmosenv.2009.10.030

a b s t r a c t

A cluster analysis of four-day back trajectories for January 1998 toDecember 2001 arrivingmid-afternoon inBirmingham, UK at three different within boundary layer arrival heights has been performed in order fora better understanding of the pollutionmeteorology influencing this region. The time periodwas purposelychosen to encompass the Pollution in the Urban Midlands Atmosphere field campaign. Six natural synopticscale transport patterns were identified with three, strong-westerly, westerly and slow-easterly, showingseasonal variation in frequency. Significant differences in air pollutant concentrations and behaviour werefound between air mass cluster types when they were analysed with measurements taken from an urbanbackground site in Birmingham and a rural site in Harwell. Highest concentrations of primary pollutantswere associated with a slow-easterly air mass from mainland Europe, while lowest concentrations wereassociated with south-westerly and strong-westerly air masses passing over the Atlantic Ocean. Thepolluted slow-easterly air mass was associated with highest ozone concentration for the warm season andlowest ozone concentration for the cool season. This could be explained byphotochemical ozone productionduring the warm season and NOx titration of background ozone during the cool season, when photo-chemically “inactive” conditions prevailed. Awealth of informationwas determinable, including influencesof short and long-range transport and photochemical productivity.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

The rationale for the present study was to better understand theconditions for which air pollution, in particular ozone episodes,occurredwithin theUKMidlands. In addition a better understandingof pollution meteorology was sought for placing the PUMA (Pollu-tion in the Urban Midlands Atmosphere) field campaign (Harrisonet al., 2006) into its more general context and for gaining insightsthat could lead to improvements in phototrajectory air qualitymodels (Derwent et al., 1996; Walker et al., 2009; Baker, submittedfor publication). This field campaign, which took place during thesummer of 1999 and winter of 1999/2000 was part of the NationalEnvironmental Research Council's Urban Regeneration and theEnvironment (URGENT) thematic programme.

Ozone in the troposphere is formed from complex reactionsinvolving volatile organic compounds (VOC) and nitrogen oxides(NOx). Its formation may take place over several hours or daysand may be associated with precursor emissions hundreds, or eventhousands of kilometres away. Thus both chemistry and atmo-spheric transport need to be understood for developing knowledgeof the causes of ozone episodes. With an understanding of this

All rights reserved.

pollution meteorology, the influence of policy on meeting airquality objectives may then be assessed.

Back trajectories from trajectory models, which make use ofarchivedmeteorological data, are potentially useful for identifying thesource regions for pollutants measured at a receptor. However, indi-vidual trajectories have uncertainties associated with the resolutionand accuracy of the meteorological data and by any simplifyingassumptionsused in the trajectorymodel,whichultimately limits theirusefulness (Stohl, 1998). To some extent, this may be overcome bycalculating a large number of back trajectories and subjecting themto a cluster analysis. This is a multivariate statistical techniquewhich groups the individual trajectories of an ensemble into a smallernumber of clusters, where the errors in the individual trajectoriestend to average out. In this way sourceereceptor relationshipsmay bedetermined in a statistical sense, helpful in developing an under-standing of the pollution climatology of a location.

Cluster analysis of back trajectories has previously been used tosupport the interpretation of precipitation composition (Dorlinget al., 1992a, 1992b), particulate matter (Wang et al., 2004; Abdal-mogith and Harrison, 2005; Borge et al., 2007; Cabello et al., 2008)and ozone (Moody et al., 1995; Sirois and Bottenheim,1995; Brankovet al., 1998; Cape et al., 2000). The majority of such studies use backtrajectories with a timescale of 3e5 days. This is generally

J. Baker / Atmospheric Environment 44 (2010) 563e571564

a compromise between having sufficient time to describe the long-range transport and the decreasing accuracy of individual backtrajectories the further they are projected backward in time (Stohl,1998). For secondary pollutants, such as ozone, sufficient time shouldbe allowed for their formation, if relationships between them andtheir precursors wish to be investigated.

In the present study, a cluster analysis of four-day back trajec-tories has been undertaken to determine the significant transportpatterns influencing ground level ozone and its precursors aswell asother pollutants at two sites, BirminghamEast, an urban backgroundsite and Harwell, a rural site over a four year period (1998e2001).The analysis focussed on the mid-afternoon period when ozoneconcentrations are generally at their maximum, and hence mostrelevant to air quality control. In order to improve the accuracy of thestudy cluster analysis was undertaken on small sets of trajectoriesrather than individual trajectories. As far as this author is aware thisis the first time in which clustering of trajectory sets rather thanindividual trajectories has been carried out for this type of work.

2. Methodology

3-D four-day back trajectories were calculated using the NOAAHybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT-4)model, employing archived NCEP/NCAR global reanalysis meteo-rological data (Draxler and Hess, 1998). Air parcel trajectoriesarriving in Birmingham, UK during the mid-afternoon at 15:00 hGMT were computed for each day for a four year period from thebeginning of 1998 to the end of 2001. Mid-afternoon arrival timeswere chosen as this generally corresponds to the daily maximum inobserved ozone concentration. Back trajectories were determinedin sets of three for arrival heights at 10, 400 and 800 m in orderto take into account uncertainty within individual trajectorydeterminations. Previous phototrajectory modelling studies by usof ozone during the summer PUMA field campaign, resulted inreduced modelling bias during pollution episodes, when thistransport uncertainty was taken into account (Walker et al., 2009;Baker, submitted for publication). Typical mid-afternoon boundarylayer heights for this location are expected to vary in the400e2000 m range over the year (Smith and Hunt, 1978). Hence,the back trajectories are expected to be generally well inside or justabove the boundary layer upon arrival at the receptor location.Any significant change in the back-trajectory with change in arrivalheight provides a measure of the trajectory uncertainty. In reality,turbulence and convection would make the arrival height a ratherill-defined quantity. 3-D trajectories, which take into accountvertical movement of the air parcel, and considered to be the mostaccurate trajectory type (Draxler and Hess, 1998) were employed.

Borge et al. (2007) have recently critiqued the air parcel trajectorycluster methodology, emphasising the importance of specifyingadequate arrival heights in the computation of the back trajectories.As they mention, past studies have shown that differences in thetrajectory arrival height can have significant effects on the calculatedtransport pathway, due to possible large variation in the wind speedand direction with height above the ground. However it is debatableas to what constitutes an adequate arrival height. The approachdeveloped in this study essentially overcomes this issue by includingseveral arrival heights in forming trajectory sets to be clustered ratherthan clustering individual trajectories. Hence this approach consti-tutes a distinctmethodological improvement and takes account of thevertical shear.

A non-hierarchical k-means algorithm was developed to clusterthese daily trajectory sets based on those described by Hartigan(1975) and Dorling et al. (1992a). Here clusters were determinedthrough a variance minimisation process involving swapping oftrajectory sets betweenneighbouring clusters until convergence onto

final clusters. These final clusters reveal an “objective” structure tothe input trajectory set ensemble. “Seed” clusters are required to startthe process. In general, the final clusters represent a local rather thana global minimum and are somewhat dependent on the chosenseed clusters. In this study three different choices of seed clusterswere chosen to partly overcome this issue. Anothermodificationwasthe choice of great circle distance rather than Euclidean distanceas the sub-elements within the clustering routine. Although themajority of previous studies have used Euclidean distance, wherea planar as opposed to spherical geometry is assumed, it is moreappropriate and correct to use great circle distances for clustering airparcel trajectories which move over the Earth's surface.

Four years of daily trajectory sets were used as input to thealgorithm, corresponding to 1461 trajectory sets. Each trajectory setwas composed of three trajectories arriving in the mid-afternoonat an arrival height of 10, 400 and 800 m above ground level. Henceeach set was comprised of 3 � 96 hourly (longitude, latitude) coor-dinate pairs. In order to find the most appropriate i.e. most naturalnumber of clusters a procedure similar to that outlined by Dorlinget al. (1992a) was followed. In the present case a hundred initialclusters were specified and the total within-cluster variance deter-mined upon convergence of the k-means routine. The number ofclusters was then successively reduced by one until all trajectorysets had been lumped into a single cluster. The total within-clustervariance was then examined as a function of number of clusters.When two clusters are merged that are significantly different a largeincrease in thewithin-cluster variance is observed, indicating that anoptimum number of clusters had previously been reached.

The air quality data used in this study was obtained from the UKNational Automatic and Urban Rural monitoring Network (AURN).Two locations were chosen, Birmingham East, an urban backgroundmonitoring site within 8 km of the PUMA field campaign site (Har-rison et al., 2006), andHarwell, a ruralmonitoring site to the South, inOxfordshire. Hourly averaged data was obtained for 1998e2001. Formost of the analysis mean mid-afternoon pollutant concentrationswere taken, when ground level ozone is generally at its maximum.Although the focus of this study was on ozone and its precursors,other air pollutants were also analysed. Table S1 (Supplementarymaterial) gives the data capture of the various pollutants recordedat the two monitoring sites over the study period. Differences ofpollutant concentrations between the two sites are indicative of localsources and conditions, while similarities will tend to be associatedwith long-range pollutant transport. Since Harwell is a rural locationit would be expected to be less influenced by local sources than theBirmingham East site.

The air quality data was analysed according to air mass clustertype using SPSS (v.15). The significance of the inter-cluster variationwas checked using the KruskaleWallis test. Pair-wise comparisonof pollutant concentration was assessed using Tukey's studentisedmultiple comparison test.

3. Results and discussion

3.1. Clustering of four years of back trajectories

The results of the clustering are shown in Fig. 1. This figurepresents the change in total within-cluster variance with number ofclusters. A significant change is observedwhen reducing the numberof clusters from six to five, indicating six natural clusters for thisperiod. This result is consistent with a previous study, where oneyear (April 1992eMarch 1993) of daily averagedmeteorological datafrom an automatic weather station located on the University ofBirminghamcampuswas examined using principal components andhierarchical agglomerative cluster analysis (McGregor and Bamzelis,1995). It was reported that a plot of a cluster agglomeration

0

10

20

30

40

50

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15Number of clusters

% c

hang

e of

with

in c

lust

er v

aria

nce

Significant change in within cluster variance whenreducing clusters from six to five. Six clusters aretherefore retained.

Fig. 1. The percentage change of within-cluster variance with cluster number for the clustering of daily trajectory sets arriving at Birmingham between January 1998 and December2001. Three different methods of selecting seed trajectories were chosen (corresponding to closed circles, open circles and crosses).

J. Baker / Atmospheric Environment 44 (2010) 563e571 565

coefficient against cluster number had a distinct break at a clusternumber of six, suggesting six major air mass types for the studyperiod. They justified this number through reference to Lamb'sseven major weather type classifications for the UK.

These six natural clusters, which are representative of distinctsynoptic scale transport patterns or air mass types, are shown inFig. 2. They have been labelled according to their overall wind

Fig. 2. The six air mass clusters identified for the 1998e2001 period, each represented by avof 10 m, (black circles), 400 m (grey circles) and 800 m (light grey circles). Clusters correspon6) slow-southerly air masses.

speed and direction or provenance; Arctic, strong-westerly, slow-easterly, westerly, south-westerly and slow-southerly. Differencesin the trajectory pathways with arrival height give an indication ofaverage vertical wind shear and pathway uncertainty. Similaritiesexist between these clusters and those determined (five) by Capeet al. (2000) at Mace Head Ireland for 1995e1997. The seasonalfrequency variations of these transport patterns are presented in

eraged 4-day back trajectories, arriving mid-afternoon at Birmingham at arrival heightsd to 1) Arctic, 2) strong (fast) westerly, 3) slow-easterly, 4) westerly, 5) south-westerly,

J. Baker / Atmospheric Environment 44 (2010) 563e571566

Fig. 3. Three show a distinct seasonal variation; the strong-westerlies are more common during the late autumn-winter periodthan during the springesummer period, the slow-easterlies aremore prominent during the spring, and the westerlies are morecommon during the summereearly autumn period.

Incidentally, when clustering using individual trajectories (ratherthan trajectory sets) and Euclidean distance, as done in previousstudies, there were no clear breaks when plotting the equivalent ofFig. 1 and hence no identification of the natural number of clusters.Furthermore poor correspondences between clusters were obtainedwith mappings of 67%, 71% and 61% between the clusters deter-mined using the trajectory set method introduced in this study andthose determined using individual trajectories, with arrival heightsof 800 m, 400m and 10m, respectively. Poor correspondences werealso found between the clusters determined using the differentindividual trajectories. This is a weakness associated with the use ofindividual trajectories rather than trajectory sets as discussed in theprevious section.

3.2. Comparison with pollutant concentrations: annually averaged

When pollutant concentrations are analysed according to airmass cluster type, distinct differences become evident. Table 1presents the annually averaged mid-afternoon pollutant concen-trations at the Birmingham East and Harwell sites, according to air

0

5

10

15

20

25

30

35

40

45

Jan Feb Mar Apr May JunM

% F

requ

ency

cluster1: arcticcluster2: fast Wcluster3: slow E

0

5

10

15

20

25

30

35

40

45

Jan Feb Mar Apr May JunM

% F

requ

ency

cluster4: Wcluster5: SWcluster6: slow S

Fig. 3. Seasonal frequency variation of air mass type for the 1998e2001 period.

mass cluster type. For this and following tables, highest and lowestconcentrations are emboldened and underlined, respectively. Thelast column gives the variation factor defined simply as; vf¼ (largestvalue)/(smallest value). 1,3-Butadiene and ozone generally have thesmallest concentration variation across the air mass types, whilesulphur dioxide and nitric oxide have the largest variation.

These pollutants are associated with a range of atmospheric life-times and sources and hence yield different information. 1,3-Buta-dienehas a short atmospheric lifetimeof typically severalhoursand isassociatedwith local trafficemissions.Benzene incontrasthasamuchlonger lifetime of about 10 days or more (Seinfeld and Pandis, 1998)andhencemaybeassociatedwithboth longandshort range transportof traffic emissions, and is generally considered to be amarker speciesfor anthropogenic VOC emissions. Sulphur dioxide has a variableatmospheric lifetime of a few days. In Europe major sources are coalandoilfiredpower stationsandheavy industrywithemissions via tallstacks (100e300 m). PM10 has a complex range of both primary andsecondary sources, with lifetimes in the range of a few days to overa week, with both local and long-range sources potentially contrib-uting (APEG, 1999; Abdalmogith and Harrison, 2005; Borge et al.,2007). Nitric oxide concentrations tend to be high close to emissionsources but rapidly decrease as the air ages due to dispersion andreaction with ozone. As a result of the fast photochemistry thatinterconverts NO, NO2 and O3, it is also useful to consider NOx andtotal oxidant (Ox ¼ O3 þ NO2) (e.g. Clapp and Jenkin, 2001).

Jul Aug Sep Oct Nov Deconth

Jul Aug Sep Oct Nov Deconth

Lines drawn between points are indicative only. See text for further details.

Table 1Mean mid-afternoon pollutant concentrations by air mass cluster type for1998e2001 at Birmingham East and Harwell AURN sites. All units are in mg m�3,unless specified. NOx is given as total NO2 equivalence. Figures in bold (underlined)give maximum (minimum) values. vf corresponds to the variationfactor ¼ (maximum value/minimum value).

Parameter(mg m�3)

Air Mass Cluster Type vf

1(arctic)

2(fast W)

3(slow E)

4(W)

5(SW)

6(slow S)

Birmingham East1,3-Butadiene 0.26 0.26 0.25 0.24 0.27 0.29 1.2Benzene 1.84 1.82 2.03 1.62 1.84 2.03 1.3CO 0.34 0.31 0.37 0.29 0.28 0.30 1.3SO2 7.4 5.0 10.5 7.0 4.7 5.5 2.3PM10 19.0 19.0 29.1 18.8 17.8 21.8 1.6NO 11.0 7.8 14.3 10.0 7.2 8.4 2.0NO2 28.2 25.3 30.0 24.0 23.3 25.9 1.3NOX 44.8 37.1 51.6 39.0 34.2 38.4 1.5O3 43.5 44.3 55.0 48.5 42.6 47.4 1.3Ox (ppbv) 36.5 35.4 43.2 36.8 33.5 37.2 1.3

Harwell1,3-Butadiene 0.09 0.08 0.11 0.09 0.08 0.10 1.3Benzene 0.71 0.60 1.04 0.60 0.49 0.74 2.1SO2 4.5 1.6 11.4 2.5 1.2 2.3 9.7PM10 15.0 15.3 26.8 16.4 15.6 20.0 1.8PM2.5 8.9 7.5 15.6 8.6 7.7 11.2 2.1PMcoarse 6.2 7.8 11.2 7.8 7.9 8.8 1.8NO 3.0 2.0 6.2 3.5 2.3 3.4 3.0NO2 14.2 9.1 22.8 12.3 9.1 14.0 2.5NOX 18.7 12.2 32.2 17.6 12.6 19.2 2.6O3 59.3 64.4 68.9 64.9 61.0 64.7 1.2Ox (ppbv) 37.1 37.0 46.4 38.9 35.3 39.6 1.3

Table 2Mean cool season (OctobereMarch) mid-afternoon pollutant concentrations by airmass cluster type for 1998e2001 at Birmingham East and Harwell AURN sites. Allunits are in mg m�3, unless specified. NOx is given as total NO2 equivalence.

Parameter(mg m�3)

Air Mass Cluster Type vf

1(arctic)

2(fast W)

3(slow E)

4(W)

5(SW)

6(slow S)

Birmingham East1,3-Butadiene 0.34 0.29 0.37 0.36 0.32 0.32 1.3Benzene 2.36 2.05 2.91 2.43 2.26 2.37 1.4CO 0.43 0.34 0.52 0.41 0.33 0.36 1.6SO2 7.8 5.5 10.7 8.5 5.3 6.6 2.0PM10 19.6 20.1 26.2 21.8 18.8 21.7 1.4NO 15.2 8.8 26.9 17.1 8.6 11.1 3.1NO2 35.8 29.0 39.2 35.5 28.5 32.9 1.4NOX 58.9 42.4 80.2 61.5 41.5 49.7 1.9O3 30.4 39.0 25.7 32.0 35.7 29.6 1.5Ox (ppbv) 34.0 34.7 33.3 34.6 32.8 32.0 1.1

Harwell1,3-Butadiene 0.10 0.09 0.15 0.11 0.09 0.12 1.6Benzene 0.91 0.67 1.56 0.83 0.60 0.99 2.6SO2 3.95 1.80 7.81 3.54 1.34 2.79 5.8PM10 14.0 15.1 23.1 15.4 14.0 20.1 1.6PM2.5 8.5 7.6 14.3 9.3 7.9 12.1 1.9PMcoarse 5.5 7.5 8.8 6.1 6.2 8.0 1.6NO 3.4 2.0 10.2 5.3 2.4 4.3 5.2NO2 18.3 10.1 28.7 19.2 11.1 18.5 2.8NOX 23.5 13.1 44.2 27.3 14.8 25.1 3.4O3 49.5 62.7 38.3 50.0 59.1 47.5 1.6Ox (ppbv) 34.4 36.6 34.1 35.0 35.4 33.4 1.1

Table 3Mean warm season (AprileSeptember) mid-afternoon pollutant concentrations byair mass cluster type for 1998e2001 at Birmingham East and Harwell AURN sites. Allunits are in mg m�3, unless specified. NOx is given as total NO2 equivalence.

Parameter(mg m�3)

Air Mass Cluster Type vf

1(arctic)

2(fast W)

3(slow E)

4(W)

5(SW)

6(slow S)

Birmingham East1,3-Butadiene 0.18 0.16 0.20 0.18 0.19 0.27 1.7Benzene 1.25 1.10 1.59 1.16 1.21 1.76 1.6CO 0.23 0.21 0.27 0.21 0.21 0.25 1.3SO2 6.8 3.3 10.4 6.1 3.8 4.6 3.2PM10 18.4 15.8 30.9 16.8 16.4 21.9 2.0NO 5.9 4.9 6.6 5.3 5.3 6.2 1.4NO2 19.1 14.2 24.4 16.5 15.8 20.1 1.7NOX 27.9 21.4 34.3 24.4 23.5 29.2 1.6O3 58.6 60.0 72.7 59.2 52.8 61.8 1.4Ox (ppbv) 39.3 37.4 49.1 38.2 34.6 41.4 1.4

Harwell1,3-Butadiene 0.07 0.07 0.08 0.08 0.07 0.08 1.2Benzene 0.48 0.36 0.74 0.44 0.34 0.54 2.2SO2 5.21 1.06 13.56 1.84 0.93 1.89 14.6PM10 16.4 15.9 29.3 17.2 18.1 20.0 1.8PM2.5 9.3 7.3 16.3 8.3 7.6 10.6 2.2PMcoarse 7.1 8.5 13.0 8.9 10.6 9.4 1.8NO 2.5 2.3 3.2 2.1 2.2 2.8 1.5NO2 9.1 5.7 18.6 7.1 6.2 10.7 3.3NOX 12.9 9.2 23.5 10.3 9.5 14.9 2.5O3 71.5 70.5 87.8 74.7 64.0 78.9 1.4Ox (ppbv) 40.5 38.2 53.6 41.1 35.2 45.1 1.5

J. Baker / Atmospheric Environment 44 (2010) 563e571 567

The slow-easterly air mass is significantly the most pollutedcluster type at both sites e see Table 1. For rural Harwell this airmass is associated with highest concentrations for all pollutantsstudied. These concentrations would be expected to be associatedwith long-range transport of pollutants emitted from mainlandEurope. Birmingham East (BE) has higher traffic related pollutantconcentrations (1,3-butadiene, benzene, NO and NOx) than Har-well, and the difference can be attributed to sources within theurban background. Incidentally, the results indicate that the slow-southerly air mass is transporting additional local traffic emissionsto the BE site, resulting in the highest concentrations of 1,3-buta-diene and benzene for this cluster type. The reduced ozone at theBE site compared to Harwell, can be largely explained by titrationwith locally emitted NOx. The slow-easterly air mass has highestconcentrations of ozone and total oxidant indicative that it is themost photochemically active over the annual cycle, associated withreactions of VOC (as indicated by benzene) and NOx emitted alongits path. The least polluted cluster types are associated with south-westerly and strong-westerly air masses, which spendmost of theirtime over the unpolluted North Atlantic ocean (see Fig. 2).

3.3. Comparison with pollutant concentrations:seasonally averaged

To consider the seasonal dependence of the pollutants the annualperiod was divided into two seasons; a “dark” or “cool” season, cor-responding to OctobereMarch, when the Sun is predominantly overthe Southern Hemisphere, and a “bright” or “warm” season, corre-sponding to AprileSeptember, when the Sun is predominantly overtheNorthernHemisphere. In the followingweuse “cool” and “warm”

to label these two seasons. Tables 2 and 3 present the seasonallyaveraged mid-afternoon pollutant concentrations at the two sites.Figs. 4 and 5 graphically present this information, where the cool andwarmseason, aswell as annual concentrations are compared for eachpollutant and cluster type. ForHarwell (Fig. 5), where PM10 and PM2.5

were measured, it is more instructive to plot PMcoarse and PM2.5 withthe latter defined as PMcoarse ¼ PM10 � PM2.5.

Most pollutant concentrations are significantly higher duringthe cool season than during the warm season, except for ozone,oxidant and PMcoarse which are significantly higher during thewarm season, and PM2.5, PM10, and SO2, which are sometimes loweror higher depending on monitoring site and air mass cluster type.The pollutants with increased concentrations during the cool

0.0

0.1

0.2

0.3

0.4

1 2 3 4 5 6

1,3-Butadiene (µg/m3)

0.0

0.5

1.0

1.5

2.0

2.5

3.0

1 2 3 4 5 6

Benzene (µg/m3)

0

2

4

6

8

10

12

1 2 3 4 5 6

SO2 (µg/m3)

0.0

0.1

0.2

0.3

0.4

0.5

0.6

1 2 3 4 5 6

CO (µg/m3)

0

5

10

15

20

25

30

35

1 2 3 4 5 6

PM10 (µg/m3)

0

5

10

15

20

25

1 2 3 4 5 6

NO (µg/m3)

0

10

20

30

40

1 2 3 4 5 6

NO2 (µg/m3)

0

20

40

60

80

1 2 3 4 5 6

NOx (µg/m3)

0

20

40

60

80

1 2 3 4 5 6

O3 (µg/m3)

0

10

20

30

40

50

1 2 3 4 5 6

Ox (ppbv)

Fig. 4. Annual (black), cool season (grey) and warm season (white) mean mid-afternoon concentrations at Birmingham East by air mass cluster type. Clusters correspond to: 1)Arctic, 2) strong (fast) westerly, 3) slow-easterly, 4) westerly, 5) south-westerly, 6) slow-southerly air masses. Error bars correspond to the standard errors of the means.

0.00

0.05

0.10

0.15

1 2 3 4 5 6

1,3-Butadiene (µg/m3)

0.0

0.5

1.0

1.5

1 2 3 4 5 6

Benzene (µg/m3)

0

5

10

1 2 3 4 5 6

SO2 (µg/m3)

0

5

10

1 2 3 4 5 6

PMcoarse (µg/m3)

0

5

10

15

1 2 3 4 5 6

PM2.5 (µg/m3)

0

2

4

6

8

10

12

1 2 3 4 5 6

NO (µg/m3)

0

5

10

15

20

25

30

1 2 3 4 5 6

NO2 (µg/m3)

0

10

20

30

40

50

1 2 3 4 5 6

NOx (µg/m3)

0

20

40

60

80

1 2 3 4 5 6

O3 (µg/m3)

0

10

20

30

40

50

60

1 2 3 4 5 6

Ox (ppbv)

Fig. 5. Annual (black), cool season (grey) and warm season (white) mean mid-afternoon concentrations at Harwell by air mass cluster. Error bars correspond to the standard errorsof the means. See text for further details.

0

10

20

1 2 3 4 5 6

Secondary Ox Increment (ppbv)

Fig. 6. Warm season increments of secondary oxidant determined at Birmingham East(grey) and Harwell (white) for each air mass cluster type.

J. Baker / Atmospheric Environment 44 (2010) 563e571570

season are associated with surface based anthropogenic emissions(1,3-butadiene, benzene, CO, NOx). Interestingly, the warm tocool season concentration increases for these pollutants are rathersimilar. Considering benzene and NOx a factor of 1.9 � 0.3 isobtained for the increase, averaging over each site and over eachcluster. The most likely main factor for this increase in concentra-tion during the cool season is the decreased vertical dilution asso-ciated with the decreased cool season boundary layer height (e.g.Smith and Hunt, 1978). Other factors are likely to be less important,for example, emissions of these pollutants. From the work ofUtembe et al. (2005) and the references therein, factors of about0.93 and 1.07may be expected for benzene andNOx respectively forthewarm to cool season differences in emission, which when takentogether would cancel each other out. Another factor to consider isthe seasonal difference in horizontal wind speed as this will influ-ence the dilution of emissions. To check for possible seasonaldifferences in mid-afternoon wind speed, data from WMO meteo-rological stations near to the two air quality monitoring sites wereexamined over the same study period (Coleshill, Brize Norton andBenson). The overall difference was of the order of 5%, with greaterwind speeds during the cool season, and hence this only contributesa 0.95 factor to the expected seasonal difference in concentration.

Considering the pollutants with an increased concentration duringthe warm season (O3, Ox, PMcoarse), the ozone and total oxidant incre-ments can be interpreted as arising from the warm season photo-chemicalproduction. Inaddition forozoneduring thecool season thereis clear evidence of concentration loss associated with NOx titration.This latter processwill also contribute to the raisedNO2 concentrationsduring the cool season. PMcoarse is comprised of coarse particles withaerodynamic diameter between 2.5 mm and 10 mm. These tend to beformed by surface based mechanical processes and may have atmo-spheric lifetimes of up to a few days. The increased PMcoarse concen-tration in the warm season (see Fig. 5) may be interpreted as due toincreased mechanical production associated with the increasedatmospheric turbulence and drier land surfaces. There is also likely tobe an additional secondary source contribution associated withcondensation of photochemically produced low-volatility compounds(e.g. nitric acid, oxidised VOC species) onto the particles. The orderingof PMcoarse concentrationwith airmass cluster type does not generallyfollow that found for other pollutants; although the slow-easterly airmass still has highest concentrations, it is the Arctic air mass (ratherthan the south-westerlies and strong-westerlies) that has the lowestconcentrations. The reason for this difference is due to the dominanceof natural sources over anthropogenic sources for this pollutant.

For both seasons the slow-easterly air mass has generally thehighest pollutant concentrations.However, although it has thehighestozone concentration during the warm season, it has the lowestozone concentration during the cool season. This may be explained asfollows. The slow-easterly air mass is heavily polluted with primarypollutants, such as NOx and VOC. During thewarm season there is netproduction of O3 from NOx e VOC photochemistry. However, duringthe cool season the photochemistry is somewhat inactive and NOxtitration of the background ozone dominates. Similar effects havebeenobservedpreviously (e.g. see Sirois andBottenheim,1995). In factduring the cool season, the least polluted south-westerly and strong-westerly air masses have highest ozone concentrations, which isrepresentative of the background ozone concentration.

Oxidant levels for the cool season, which are independent ofthe NOx titration process, are generally uniform across all air masstypes, with mean values across all cluster types of 33.6 � 1.1 and34.8� 1.1 ppbv for the BE and Harwell sites, respectively. These valuescanbe interpretedas the lower tropospherebackgroundvalue,over theAtlantic and European region. Net secondary production of oxidantduring thewarm season, for each air mass type, can be determined byconsidering the warm season increment in total oxidant, after

accounting for primary nitrogen dioxide emissions. Fig. 6 presents theresults of this analysis, where it was assumed that NOxwas on averageemitted as 7% NO2, over the 1998e2001 period. The slow-easterly airmass (cluster 3) has the highest secondary oxidant increment with anacross site average of 19 ppbv, followed by the slow-southerly airmass (cluster 6)with an across site average of 11ppbv.Hence the slow-easterly air mass cluster type is about twice as photochemicallyproductive as the next most polluted air mass type. The least pollutedsouth-westerly and strong-westerly air mass cluster types yield thelowest secondary total oxidant increment with across site averages of1.3 and 2.6 ppbv, respectively.

PM10 is composed of bothfine and coarse particles. The results forthe coarse particles (PMcoarse) measured at Harwell have alreadybeen discussed. Now consideration will be given to the results forPM2.5 measured at Harwell and PM10 measured at the BE site. PM2.5,comprising fine particles, shows significant increments and decre-ments with cluster type when going from the cool to the warmseason. Fine particles have both primary and secondary sources (e.g.APEG, 1999). Primary anthropogenic combustion particles fromtraffic would be expected to have the same concentration decre-ments during the warm season as that shown for other similarsurface based anthropogenic emissions (e.g. NOx and benzene).Conversely, secondary particulate matter arising from gas to particleconversionwould be expected to have increased concentrations. Theoverall result will depend on which factor is dominant. PM2.5 asso-ciated with the slow-easterly air mass shows an increment for thewarm season, whereas most other air mass types show decrements.This is indicative of significantwarm seasonproduction of secondaryPM2.5 for the slow-easterly air mass. This secondary PM productionlikely explains the warm season increment of PM10 for this air massat the BE site, where other air mass types have decrements. Thisconclusion is consistent with the findings of Abdalmogith and Har-rison (2005).

Finally it is worth commenting on the results for sulphur dioxide.Highest concentrations are found for the slow-easterly air mass,while lowest concentrations are found for the south-westerly andstrong-westerly air masses, consistent with that found for otherprimary pollutants. For all cluster types except the slow-easterly, theBEsitehasgreater SO2 concentrations than thatmeasuredatHarwell,indicativeof sourceswithin itsurbanbackground.Concentrationsaregenerally greater during the cool season as might be expected fora within boundary layer emitted anthropogenic primary pollutant.However at Harwell, for the slow-easterly and Arctic air mass clus-ters, sulphurdioxide concentrations are significantly elevated duringthe warm season. This is especially marked for the slow-easterly airmass. In fact the variation factor across the cluster types for thispollutant at Harwell, of 10 and 15 for the annual and warm seasonperiods (see Tables 1 and 3), is significantly higher than that for any

Table 4Daily ozone exceedances and 90th and 95th percentile values by air mass clustertype for 1998e2001 at Birmingham East and Harwell AURN sites.

Air Mass Cluster Type

1(arctic)

2(fast W)

3(slow E)

4(W)

5(SW)

6(slow S)

N 217 212 246 343 205 238Birmingham East (mg m�3)[O3 (hourly max)]day � 100 1 1 36 6 0 20[O3 (hourly max)]day � 150 0 0 8 0 0 1[O3 (mean)]day � 100 0 0 1 0 0 090th ([O3 (hourly max)]day) 80 80 114 82 78 9895th ([O3 (hourly max)]day) 84 84 132 90 84 106

Harwell (mg m�3)[O3 (hourly max)]day � 100 4 4 60 25 5 45[O3 (hourly max)]day � 150 0 0 13 0 0 3[O3 (mean)]day � 100 0 0 5 0 0 290th ([O3 (hourly max)]day) 88 88 130 96 90 10695th ([O3 (hourly max)]day) 96 94 150 108 94 119O3 pollution order 1 3 2

J. Baker / Atmospheric Environment 44 (2010) 563e571 571

other pollutant � monitoring site combination. This indicates theexistence of a significant local source of SO2 at Harwell and in factthere is a coal-fired power station located about 10 km to the north-east (Lee et al., 1999). As shown in Fig. 2, the slow-easterly air masstype containsa significantnorth-easterlycomponent,whichexplainsthe high concentrations associated with this cluster type. Similarlythe northerly Arctic cluster is also affected by it. The seasonal varia-tion cannowbe interpreted as increased groundingof the SO2 plumefromthepower stationduring thewarmseason compared to the coolseason, associated with enhanced turbulence and convection in thelower atmosphere during the warm season.

3.4. Comparison with pollutant concentrations: ozone episodes

Table 4 provides information on ozone episodes associated withthe six air mass types. The concentrations can be compared to theUK air quality standard for ozone of 100 mg m�3 (50 ppbv). Thesevarious metrics all show that ozone episodes are most significantlyassociated with the slow-easterly air mass, followed next by theslow-southerly air mass. These results are consistent with theconclusions drawn from the analysis of the warm season incre-ments in secondary oxidant e see Fig. 6.

4. Conclusion

A novel k-means cluster algorithm was developed and applied tothe analysis of four-day back trajectories for January 1998eDecember2001 arriving mid-afternoon in Birmingham, UK at three differentarrival heights for the identification and description of the mainsynoptic scale air transport patterns associated with this region.Clustering was undertaken on trajectory sets rather than individualtrajectories to account for vertical shear that has affected previousstudies. The clustering analysis revealed a natural break of trajectoriesinto sixmainairmass clusters. Threeof these clusters showedseasonalvariation in frequency, with strong-westerlies more frequent duringwinter, westerlies more frequent during the summer, and slow-east-erlies more frequent during the spring. Highest primary pollutantconcentrations were associated with the slow-easterly air masscluster, which passed over continental Europe, while lowest pollutantconcentrations were associated with south-westerly and strong-westerly air mass cluster types, which spend most of their time overthe Atlantic. The slow-easterly air mass cluster was associatedwith highest ozone concentrations for the warm season but lowestconcentrations for the cool season. This could be explained by photo-chemical ozone production during thewarm season andNOx titration

of background ozone during the cool season, when photochemically“inactive” conditions prevailed. Consideration of secondary oxidantincrements during the warm season and daily ozone exceedancesenabled the air mass types to be ranked according to photochemicalproductivity with slow-easterlies followed by slow-southerlies beingthemost photo-reactive airmass types. Information on both short andlong-range transport was possible using this approach.

Appendix. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.atmosenv.2009.10.030.

References

Abdalmogith, S.S., Harrison, R.M., 2005. The use of trajectory cluster analysis toexamine the long-range transport of secondary inorganic aerosol in the UK.Atmospheric Environment 39, 6686e6695.

APEG, 1999. Source Apportionment of Airborne Particulate Matter in the UnitedKingdom. Department of Environment, Transport and the Regions, London, UK.

Baker, J. Photochemical trajectory modelling of ozone pollution episodes in the UKWest Midlands, Science of the Total Environment, submitted for publication.

Borge, R., Lumbreras, J., Vardoulakis, S., Kassomenos, P., Rodriguez, E., 2007. Analysisof long-range transport influences on urban PM10 using two-stage atmospherictrajectory clusters. Atmospheric Environment 41, 4434e4450.

Brankov, E., Rao, S.T., Porter, P.S., 1998. A trajectory-clustering-correlation meth-odology for examining the long-range transport of air pollutants. AtmosphericEnvironment 32, 1525e1534.

Cabello,M.,Orza, J.A.G.,Galiano,V.,2008.Airmassoriginandits influenceover theaerosolsize distribution: a study in SE Spain. Advances in Science and Research 2, 47e52.

Cape, J.N., Methven, J., Hudson, L.E., 2000. The use of trajectory cluster analysis tointerpret trace gas measurements at Mace Head, Ireland. Atmospheric Envi-ronment 34, 3651e3663.

Clapp, L.J., Jenkin, M.E., 2001. Analysis of the relationship between ambient levels OfO3, NO2 and NO as a function of NOx in the UK Source. Atmospheric Environ-ment 35, 6391e6405.

Derwent, R.G., Jenkin, M.E., Saunders, S.M., 1996. Photochemical ozone creationpotentials for a large number of reactive hydrocarbons under Europeanconditions. Atmospheric Environment 30, 181e199.

Dorling, S.R., Davies, T.D., Pierce, C.E., 1992a. Cluster analysis: a technique for esti-mating the synoptic meteorological controls on air and precipitation chemistry-method and applications. Atmospheric Environment 26A, 2575e2581.

Dorling, S.R., Davies, T.D., Pierce, C.E.,1992b. Cluster analysis: a technique for estimatingthe synoptic meteorological controls on air and precipitation chemistry-resultsfrom Eskdalemuir, South Scotland. Atmospheric Environment 26A, 2583e2602.

Draxler, R.R., Hess, G.D., 1998. An overview of the HYSPLIT-4 modelling system fortrajectories, dispersion and deposition. Australian Meteorological Magazine 47,295e308.

Harrison, R., Yin, J., Tilling, R., Cai, X., Seakins, P., Hopkins, J., Lansley, D., Lewis, A.,Hunter, M., Heard, D., Carpenter, L., Creasey, D., Lee, J., Pilling, M., Carslaw, N.,Emmerson, K., Redington, A., Derwent, R., Ryall, D., Mills, G., Penkett, S., 2006.Measurement and modelling of air pollution & atmospheric chemistry in theUKWest Midlands conurbation: overview of PUMA Consortium project. Scienceof the Total Environment 360, 5e25.

Hartigan, J.A., 1975. Clustering Algorithms. John Wiley, New York.Lee, D.S., Dollard, G.J., Derwent, R.G., Pepler, S., 1999. Observations on gaseous and

aerosols components of the atmosphere and their relationships. Water, Air andSoil Pollution 113, 175e202.

McGregor, G.R., Bamzelis, D., 1995. Synoptic typing and its Application to theInvestigation of weather air pollution relationships, Birmingham, UnitedKingdom. Theoretical and Applied Climatology 51, 223e236.

Moody, J.L.,Oltmans,S.J., Levy II,H.,Merril, J.T.,1995.Transportclimatologyof troposphericozone: Bermuda, 1988e1991. Journal of Geophysical Research 100, 7179e7194.

Seinfeld, J.H., Pandis, S.N., 1998. Atmospheric Chemistry and Physics. Wiley, New York.Sirois, A., Bottenheim, J.W., 1995. Use of backward trajectories to interpret the

5- year record of PAN and O3 ambient air concentrations at Kejimkujik NationalPark, Novia Scotia. Journal of Geophysical Research 100, 2867e2881.

Smith, F.B., Hunt, R.D., 1978. Meteorological aspects of the transport of pollutionover long distances. Atmospheric Environment 12, 461e477.

Stohl, A., 1998. Computation, accuracy and applications of trajectories e a reviewand bibliography. Atmospheric Environment 32, 947e966.

Utembe, S.R., Jenkin,M.E., Derwent, R.G., Lewis, A.C., Hopkins, J.R., Hamilton, J.F., 2005.Modelling the ambient distribution of organic compounds during the August2003 ozone episode in the Southern UK. Faraday Discussions 130, 311e326.

Wang, Y.Q., Zhang, X.Y., Arimoto, R., Cao, T.J., Shen, Z.X., 2004. The transportpathways and sources of PM10 pollution in Beijing during Spring 2001, 2002and 2003. Geophysical Research Letters 31, L14110.

Walker, H.L., Derwent, R.G., Donovan, R., Baker, J., 2009. Photochemical trajectorymodelling of ozone during the summer PUMA campaign in the UK WestMidlands. Science of the Total Environment 407, 2012e2023.