Air Pollution in AccraNeighborhoods: Spatial,Socioeconomic, and TemporalPatternsK A T H I E L . D I O N I S I O , †

R A P H A E L E . A R K U , § , |

A L L I S O N F . H U G H E S , ⊥

J O S E V A L L A R I N O , †

H E A T H E R C A R M I C H A E L , ‡

J O H N D . S P E N G L E R , †

S A M U E L A G Y E I - M E N S A H , § , # A N DM A J I D E Z Z A T I * , †

Harvard School of Public Health, Boston, Massachusetts,Harvard College, Cambridge, Massachusetts, Department ofGeography and Resource Development, University of Ghana,Legon, Ghana, Cyprus International Institute for theEnvironment and Public Health, Nicosia, Cyprus, Departmentof Physics, University of Ghana, Legon, Ghana, andEnvironmental Science Program, University of Ghana,Legon, Ghana

Received October 27, 2009. Revised manuscript receivedFebruary 9, 2010. Accepted February 16, 2010.

This study examined the spatial, socioeconomic status (SES),and temporal patterns of ambient air pollution in Accra, Ghana.Over 22 months, integrated and continuous rooftop particulatematter (PM) monitors were placed at a total of 11 residentialor roadside monitoring sites in four neighborhoods of varying SESand biomass fuel use. PM concentrations were highest inlate December and January, due to dust blown from the Sahara.Excluding this period, annual PM2.5 ranged from 39 to 53 µg/m3

at roadside sites and 30 to 70 µg/m3 at residential sites; meanannual PM10 ranged from 80 to 108 µg/m3 at roadside sitesand 57 to 106 µg/m3 at residential sites. The low-income anddensely populated neighborhood of Jamestown/Ushertown hadthe single highest residential PM concentration. There wasless difference across traffic sites. Daily PM increased at allsites at daybreak, followed by a mid-day peak at some sites, anda more spread-out evening peak at all sites. Average carbonmonoxide concentrations at different sites and seasons rangedfrom 7 to 55 ppm, and were generally lower at residentialsites than at traffic sites. The results show that PM in thesefour neighborhoods is substantially higher than the WHO AirQuality Guidelines and in some cases even higher than the WHOInterim Target 1, with the highest pollution in the poorestneighborhood.

Introduction

Although more than 60% of sub-Saharan Africa’s (SSA)population is currently rural, Africa’s urban population isgrowing faster than that in any other world region (1). Despitethis trend, there is limited data on air pollution in SSA cities,especially for particulate matter (PM) which is consideredthe best indicator of the health effects of pollutant mixtures.For example, a comprehensive review found that in 2000annual PM data were available for only 3 of 212 cities withpopulation g100,000 in SSA (2).

In high-income countries, distances to major roads orlarge stationary sources are important predictors of PMpollution, and air pollution tends to be higher in lowersocioeconomic status (SES) communities (3-9). Sources ofurban air pollution in SSA and other developing regionsinclude industrial emissions, transportation, household andcommercial biomass use, and resuspended dust fromunpaved roads. These cities often have large informal “slum”communities with poor environmental conditions (10). Whilethe spatial and SES patterns of air pollution may be differentfrom those in high-income countries, few studies haveempirically examined such patterns (11-14). Even fewer ornone have measured both fine (PM2.5) and coarse particles,or have considered PM pollution in relation to communitySES.

Our study aimed to address this important data gapthrough systematic collection and analysis of primary airpollution data in Accra, Ghana. The results advance ourunderstanding of the levels and the spatial, temporal, andSES patterns of air pollution in a growing city in a low-incomedeveloping country.

Study Location. Our study took place in four neighbor-hoods in Accra, the capital city of Ghana. Accra is located onthe Gulf of Guinea and has an area of more than 250 km2.Land elevation ranges from 0 to 30 m above sea level. Accrahas grown substantially in recent decades with the populationof Accra Metropolitan Area (AMA) increasing from 600,000in 1970 to 1.7 million in 2000.

The four study neighborhoods lie on a line from the coastto the northern boundaries of the AMA: Jamestown/Ush-ertown (JT), Asylum Down (AD), Nima (NM), and East Legon(EL) (Figure S1 and Table S1). JT is one of the oldestneighborhoods in Accra and lies between the coast and AccraBusiness Center; AD and NM are located approximately 3km inland, separated by Ring Road Central; EL is 10 km inlandand lies north of Accra International Airport. JT and NM arepoor, densely populated communities where biomass is thepredominant household fuel. Biomass is also used for small-scale commercial purposes, such as cooking street food. Alongthe JT coast, fish are smoked over wood fires and goats areroasted over burning tires doused with kerosene (15). A largebusy road with a central bus station runs through NM. BothJT and NM experience much activity throughout the day,including markets and small vendors. AD is a middle class,mostly residential neighborhood, bordered by the Ring RoadCentral, one of the largest and busiest roads in Accra. Fewerpeople use biomass fuels and street food vendors are lesscommon in AD than in JT and NM. EL is an upper-class,sparsely populated, and residential neighborhood with mostfamilies living on large plots of land. Most homes havemodern indoor kitchens and use liquefied petroleum gas.The streets are quiet during the day. The main road in EL hasheavier traffic primarily during the morning and eveningcommute periods.

* Corresponding author e-mail: majid_ezzati@harvard.edu; tele-phone: +1-617-432-5722; fax: +1-617-432-6733.

† Harvard School of Public Health.‡ Harvard College.§ Department of Geography and Resource Development, Uni-

versity of Ghana.| Cyprus International Institute for the Environment and Public

Health.⊥ Department of Physics, University of Ghana.# Environmental Science Program, University of Ghana.

Environ. Sci. Technol. 2010, 44, 2270–2276

2270 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 7, 2010 10.1021/es903276s 2010 American Chemical SocietyPublished on Web 03/05/2010

Study Design. We evaluated the spatiotemporal variabilityof ambient air pollution in these four neighborhoods usinga combination of integrated and continuous rooftop moni-tors. The study took place between November 2006 andAugust 2008, following a pilot study in July 2006 (16). Wedesigned our study to examine the seasonal and daily patternsof air pollution, as well as its variation between neighborhoodsand to a lesser extent within each neighborhood; withinneighborhood analyses compared measurements near majorroads and those in residential parts of the neighborhood.

Monitors were placed on the rooftops of a total of 11homes and businesses in the four study neighborhoods(Figure S1). To achieve the above aims, we selected moni-toring sites such that one site in each neighborhood was ona road with medium-heavy traffic for the whole day or partsof the day. One or two other sites in each neighborhood wereselected in areas judged to be typical of residential parts ofthe neighborhood. These residential (-R1 and -R2) sites werelocated on secondary roads or alleys with no or significantlyless traffic than the neighborhood traffic site (-T), thoughthe residential sites still may have pollution sources nearby,especially biomass fuels in JT and NM. The other criteria forselection of sites were equipment safety, access to electricity,and uninterrupted access to our equipment for operationand maintenance. Characteristics of measurement sites areprovided in Table S2.

Because we had a limited number of integrated monitorsand because operating and maintaining monitors in distantneighborhoods was time-intensive, we restricted simulta-neous operation to five sites in two neighborhoods. To achievethe above study aims, we used the following design:Throughout the first year of the study (November 2006 toAugust 2007), we operated monitors at a residential and atraffic site in NM. We also operated three sites for 6-7 weeksin each of the other three neighborhoods with the detailedschedule shown in Figure S2. Therefore, in study year 1 NMwas a reference neighborhood, relative to which all otherneighborhoods were measured. A reference neighborhoodwas necessary because measurements in other neighbor-hoods were in different months. To examine whether thedifference between other neighborhoods and NM varied byseason, in the second year of the study (September 2007 toAugust 2008), we continued to operate NM-T and NM-R whileplacing the remaining three monitors at JT-R1, AD-R1, andEL-R1 for the entire study year (Figure S2). Year 1 measure-ments were conducted for contiguous 48-h periods, whileyear 2 measurements were conducted for one 48-h periodevery six days.

From October to April, predominant winds in West Africablow southwest from the Sahara, creating dry and dustyconditions, a phenomenon known as the Harmattan (17-19).Accra residents have observed that the winds are strongestand the dust highest during the last few days of Decemberand through the month of January, also confirmed by ourdata. We operated monitors at JT-R1, JT-R2, JT-T, NM-R1,and NM-T during these 3-5 weeks of highest dust concen-trations in measurement year 1, and at JT-R1, AD-R1, EL-R1,NM-R1, and NM-T in measurement year 2. In this manuscript,we use “Harmattan” to refer to this shorter, more dustyperiod. Specific dates that denote the Harmattan period inour measurements and analysis are provided in Figure S2.

At each monitoring site, we measured integrated PM2.5

and PM10 using a gravimetric method. A total of 568 PM2.5

and 571 PM10 samples were collected during year 1 of thestudy and 311 PM2.5 and 314 PM10 samples during year 2 ofthe study (see Table S3 for details by site). Of these, 58 werePM2.5 duplicates (i.e., side-by-side measurements) and 64were PM10 duplicates. A total of 56 PM2.5 and 56 PM10 fieldblanks were collected over these two years. We also measuredPM2.5 and PM10 continuously at as many sites as we had

equipment available. The numbers of site-days of continuousmeasurements were 960 for PM2.5 and 554 for PM10 in year1, and 656 (PM2.5) and 404 (PM10) in year 2 (see Table S3 fordetails by site). We measured integrated carbon monoxide(CO) using passive monitors. All pollutant measurementswere colocated. We did not measure SO2 and NO2 becausean earlier pilot study showed that their concentrations wererelatively low and there was little spatial variation (16).Electricity outages were particularly common and lengthy instudy year 1 because droughts had reduced electricitygeneration. We used 12-V 100-Ah batteries connected tochargers and inverters as a power backup system at eachmonitoring site so that the regular unannounced or plannedelectricity outages in Accra had limited effect on ourmeasurements.

Materials and MethodsPollutant Measurement and Analytical Methods. IntegratedPM. See Text S1 for description of measurement methods.

Measured concentrations were used only if the pumpsoperated for g90% of the 48-h measurement period and ifthe average flow rate was within 10% of the intended rate.This excluded 169 PM2.5 and 123 PM10 measurements in year1 and 33 PM2.5 and 37 PM10 measurements in year 2. The21-30% year 1 exclusion was greater than the 12-13%exclusion in year 2 because power outages were morecommon and substantially longer in year 1 and on somedays affected measurements despite our power backupsystem. Eight additional measurements in year 1 and 10 inyear 2 were excluded for miscellaneous quality controlreasons, including incorrect labeling, damage to the filter,and broken connections in the air flow system.

All PM concentrations were blank corrected. Whereduplicate measurements were taken, the two measurementswere averaged to use all available data. See Text S1 forquantitative information on blanks and duplicates.

Continuous PM. We measured continuous PM2.5 and PM10

using DustTrak Model 8520 and SidePak Model AM510monitors (TSI Inc.). See Text S1 for details.

PM concentrations measured using light scattering aresubject to error, because factory calibrations use specificaerosols whose characteristics (e.g., shape, size, density, andrefractive index) may differ from those in field studies, andbecause factors such as relative humidity (RH) affectmeasurements (20-22). For the same reason, measurementerror can vary across days. We adjusted measured continuousPM in a two-step process: In the first step, we standardizedthe minute-by-minute records for the effects of RH, usingrelationships from previous studies (20). See Text S1 for RHdata sources and methods. In the second step, we correctedall minute-by-minute PM records in each 48-h measurementperiod using a correction factor (CF) so that the average ofcontinuous PM measurements was equal to the integratedgravimetric PM level over the same period and at the samelocation. In the above approach, the first step removes theeffect of RH variation on measured PM within a single dayand the second step ensures that the measurements arecorrected against the gravimetric measurement which hassubstantially less error than nephelometers.

We calculated unique CFs for PM2.5 and PM10, for each48-h period, and at each site. The median (interquartile range)of CFs were 0.71 (0.56-1.13) for PM2.5 and 1.07 (0.91-1.34)for PM10. When the integrated sample was excluded for asite-period (see above), or when the duration of continuousdata was <90% of the 48-h measurement period, we imputedthe CF as described in Text S1.

Integrated CO. See Text S1 for details of measurementmethods.

Statistical Analysis. We calculated mean annual PMconcentrations for year 2 directly using the measured data,

VOL. 44, NO. 7, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2271

because measurements were done throughout the entire year.In year 1, measurements at sites other than NM-R and NM-Twere done only for 6-7 weeks. We used measured data tocalculate the mean in seasons when monitoring was con-ducted and a regression model to predict the mean in seasonswithout measurement. The regression model utilized therelationship of each of the other neighborhoods to thereference neighborhood (NM) to predict PM concentrations,represented by the equation below, run separately for PM10

and PM2.5:

where neighborhood is JT, NM, AD, or EL; site-type isresidential or traffic; season is November-December,February-April, April-July, or July-September; and * de-notes statistical interaction.

In this specification, all data contribute toward determin-ing the effect of being in a specific neighborhood, and ofbeing a residential or a roadside measurement site, whiledata from NM inform the seasonal patterns. The premise ofthe model is that differences between neighborhoods,controlled for being a residential vs roadside site, can beextrapolated across seasons. We tested this assumption usingyear 2 data, and found that the neighborhood differencesdid not seasonally differ in 8 out of 9 pairwise comparisonsfor PM2.5 (6 out of 9 if data were log-transformed) and in 9out of 9 pairwise comparisons for PM10 (8 out of 9 if datawere log-transformed). See Text S1 for further informationon model performance. In sensitivity analysis, we used ln(PM)as the dependent variable. The differences with concentra-

tions in Figure 2 were 8.1% for PM2.5 (range 1-17%) and7.7% for PM10 (range 1-15%).

Confidence intervals (CIs) for annual mean PM for year2 were calculated directly using the measured data based onsample sizes. Because annual means in year 1 are based oncombinations of measurements and predictions from aregression equation, they have larger uncertainty. The CIsfor year 1 were calculated using the measured data in seasonswhen a site had measurements and using the joint uncertaintyof the regression coefficients plus that caused by regressionresidual in seasons when PM for a site was predicted. As aresult, annual means for year 1 have larger uncertainty thanthose for year 2 (with the exception of NM where estimatesfor both years were based on measurement), reflecting theuse of a prediction equation. The season-specific means andstandard errors were combined into annual ones usingstandard techniques for variance of averages, weighted byduration of each season.

All analyses were done using Matlab 7.9.0, Stata 10, andthe open-source package R version 2.8. See Text S2 for studystrengths and limitations.

Results and DiscussionIntegrated PM. Excluding the peak Harmattan period, overthe two measurement years, mean annual PM2.5 ranged from39 to 53 µg/m3 at roadside sites and 30 to 70 µg/m3 atresidential sites; mean annual PM10 ranged from 80 to 108µg/m3 at roadside sites and 57 to 106 µg/m3 at residentialsites. Overall, PM concentration was lower in year 2 than inyear 1 at all sites, except for PM10 at the AD and JT residentialsites; due to the relatively large uncertainty associated withthe prediction model, none of the differences were statistically

FIGURE 1. Integrated PM concentrations (a) PM2.5 and (b) PM10.See Figure S3 for logarithmic scale on the vertical axis.

PM ) �0 + �1neighborhood + �2site-type + �3season +�4neighborhood*site-type + ε

FIGURE 2. Annual mean PM (excluding Harmattan period) byneighborhood and site type for (a) PM2.5 and (b) PM10. Thehorizontal lines show the most recent WHO Air QualityGuidelines (AQG) (10 µg/m3 for PM2.5 and 20 µg/m3 for PM10)and Interim Target 1 (IT-1) (35 µg/m3 for PM2.5 and 75 µg/m3 forPM10).

2272 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 7, 2010

significant (Figure 2). If Harmattan were included, annual PM2.5

and PM10 means would increase by 21-23 and 33-36 µg/m3,respectively, in year 1 and by 11-14 and 20-26 µg/m3,respectively, in year 2. These levels are substantially higherthan the World Health Organization (WHO) Air QualityGuidelines (AQG) of 10 µg/m3 (PM2.5) and 20 µg/m3 (PM10)and in some cases even higher than the WHO Interim Target1 (IT-1) of 35 µg/m3 (PM2.5) and 75 µg/m3 (PM10) (23) (Figure2). These annual means are similar to or slightly lower thanthose in large cities in South and East Asia and in the EasternMediterranean, but substantially higher than those in LatinAmerica and high-income countries (2). Non-HarmattanPM2.5 concentration at a single peri-urban site northeast ofAccra during a period overlapping our first year of study waslower than those at our monitoring sites by 7-47 µg/m3;PM10 at the peri-urban site was lower than that at our trafficsites and JT residential sites, but was higher than that ofresidential sites in AD, NM, and EL by 6-20 µg/m3 (24).

Over the full measurement period, PM2.5:PM10 ratiosranged from 0.4 to 0.7, with JT-R sites having the highestratios after accounting for measurement season (detailedresults in Table S4). Traffic sites had slightly lower ratiosthan residential ones in each neighborhood and season,except during Harmattan. There was evidence of someseasonal variation in the fine and coarse particle contribu-tions. The (Spearman) correlation between PM2.5 and thecoarse fraction (calculated as PM10 minus PM2.5 on eachmeasurement day) ranged from 0.76 to 0.98 in study year 2,when measurements took place over the whole year. Thisrelatively high correlation implies that in this uniqueenvironment, the same or at least co-occurring sourcescontribute to both fine and coarse size fractions. Identifyingthe specific sources requires further analysis, e.g., of PMcomposition; correlated sources may include biomass burn-ing for cooking contributing to fine PM, and oil dropletsfrom frying food contributing to the coarse fraction. Therewas more heterogeneity in fine-coarse correlation in themeasurements from the first year of the study, with somesites having significantly lower correlations than thoseobserved in year 2, either due to seasonal factors or due toreal interannual differences.

Between-Day Patterns of PM Pollution. The single mostnotable seasonal pattern of PM was the large peak in lateDecember and January in both measurement years, reachingas high as 300-800 µg/m3 for PM2.5 and 600-1200 µg/m3 forPM10 in January 2007 (Figures 1 and S3). The peaks in 2008were smaller than the previous year, which may reflect year-to-year differences in Harmattan winds. Previous multiyearstudies had also recorded very high PM concentrations duringHarmattan with year-to-year variation (17, 18, 25).

While we had expected wind-blown desert dust to havehigher coarse fraction, PM2.5:PM10 ratio during Harmattanwas similar to those observed during the rest of the year(Table S4). Our finding is consistent with those of previousstudies, which found the proportion of fine particles inHarmattan dust was inversely proportional to the distancefrom the Sahara, and relatively high in Southern Ghana(17, 18).

Beyond the large peak in December-January, PM con-centrations were also high in November-December in year1, and in February-April in both years (Figure 1, Figure S3,and Table 1); both of these correspond to the longerHarmattan phenomenon (17-19). Whether there are otherreasons for interannual differences in PM seasonality requiresfurther investigation but may include meteorological factors,e.g., variations in rainfall, and/or different source patterns.PM concentrations were consistently lowest in April-August,possibly because these months are during and immediatelyafter the main rainy season.

There was no significant association between PM pollutionand day of week in our data, thus day of week was not usedin further analysis.

Residential (-R) vs Roadside (-T) PM Pollution. With theexception of JT, PM was higher at traffic sites than residentialsites in all neighborhoods, by 8-14 µg/m3 for PM2.5 and 19-46µg/m3 for PM10; with the exception of year 1 PM10 in NM, theresults were not statistically significant but their consistentpattern across sites is unlikely to be due to chance. Further,multivariate regression analysis of all data found that trafficsites on average had higher PM, but the effect varied byneighborhood (see interaction terms in Table 1). Studies inhigh-income countries generally find a strong influence frommajor traffic routes on PM concentrations (7, 8). Thetraffic-residential difference in these Accra neighborhoodswas possibly smaller than those in high-income countriesand was reversed in JT. The lower roadside concentration atJT may be due to two factors: first, traffic was diverted fromthe main road next to this site to a nearby road soon afterwe began measurements, dampening the local traffic effect.Second, JT has the highest density of residential and small-commercial biomass use, even relative to NM, potentially

TABLE 1. Association of 48-h PM Data with Neighborhood,Site Type, and Season in Multivariate Analysis

coefficient (95% CI) p-value

dependent variable: PM2.5; n ) 342; R2 ) 0.81constant 71.9 (67.1, 76.7) <0.001

site-typeresidential 0.0 NAtraffic 8.3 (4.1, 12.5) <0.001

neighborhoodNima (NM) 0.0 NAJamestown (JT) 32.8 (26.6, 39.0) <0.001Asylum Down (AD) -3.8 (-9.7, 2.2) 0.211East Legon (EL) -4.0 (-10.1, 2.2) 0.207

seasona

November-December 0.0 NAFebruary-April -25.1 (-31.1, -19.1) <0.001April-July -48.6 (-54.2, -43.0) <0.001July-August -48.7 (-54.6, -42.7) <0.001

site-type * neighborhoodb

traffic * Jamestown -25.5 (-34.3, -16.7) <0.001traffic * Asylum Down 5.3 (-2.3, 13.0) 0.171traffic * East Legon 0.9 (-8.0, 9.8) 0.848

dependent variable: PM10; n ) 352; R2 ) 0.69constant 122.1 (112.1, 132.2) <0.001

site-typeresidential 0.0 NAtraffic 19.1 (10.1, 28.1) <0.001

neighborhoodNima (NM) 0.0 NAJamestown (JT) 35.0 (22.1, 47.9) <0.001Asylum Down (AD) -9.1 (-21.8, 3.5) 0.157East Legon (EL) -5.8 (-18.9, 7.3) 0.385

seasona

November-December 0.0 NAFebruary-April -21.1 (-33.9, -8.4) 0.001April-July -78.0 (-89.7, -66.3) <0.001July-August -80.0 (-92.3, -67.1) <0.001

site-type*neighborhoodb

traffic * JT -33.6 (-52.0, -15.1) <0.001traffic * AD 27.2 (10.9, 43.5) 0.001traffic * EL 4.8 (-13.9, 23.5) 0.614

a Harmattan was not included as a season in theregression analysis because measurements during thosefew weeks would be outliers. b The interaction terms canbe interpreted as the additional effect of being a traffic sitein specific neighborhoods, above and beyond the averagetraffic site effect and neighborhood effect.

VOL. 44, NO. 7, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2273

leading to disproportionately higher contributions from suchsources. Within different neighborhoods, PM2.5 differed by1-13% between the two residential sites; PM10 differed by4-27%.

Despite the differences in levels, roadside and residentialsite PM levels were correlated across days in JT, NM, and AD,with correlation coefficients ranging from 0.69 to 0.96 forPM2.5 and 0.91 to 0.99 for PM10, i.e., their day-to-day variationstracked one another. Roadside-residential correlation wasmuch lower in EL, 0.04-0.13 for PM2.5 and 0.30-0.44 forPM10. Lower correlation in EL may be because the residentialand traffic sites were farther, and because in at least JT andNM, biomass use was more common around both residentialand traffic sites.

Between-Neighborhood Patterns. In measurement year 1,roadside PM2.5 was relatively similar across neighborhoodswith between neighborhood differences below 11 µg/m3

(Figure 2). Similarly, roadside PM10 in JT, NM, and EL variedby less than 3 µg/m3 but was 17 µg/m3 higher at AD-Tcompared to the second highest neighborhood; none of thesedifferences were statistically significant, partly due to themodel prediction uncertainty. PM10 may have been higherat AD-T because it is located on the busy Ring Road Central,with contributions from mobile sources as well as resus-pended dust which contains higher coarse-fraction PM.

JT had the single highest residential PM, 21-33 µg/m3

higher than the next highest neighborhood for PM2.5 and 35µg/m3 higher for PM10 over the two measurement years; thecomparisons with each other neighborhood were significantor borderline significant at p ) 0.05 for different years andsize fractions, also confirmed in multivariate analysis (Table1). This is likely due to the extensive use of biomass fuels,especially wood, for home and small-commercial cookingpurposes in a very densely populated area. As describedearlier, JT is also bordered by a stretch of the Accra coastwhere fish smoking and goat roasting are done by burningwood or old rubber tires. There was substantially lessdifference in residential site concentrations across otherneighborhoods, with slightly higher concentrations in NMin year 1 and little systematic pattern in year 2. Residential

PM concentrations in neighborhoods other than JT werewithin 5 µg/m3 of one another for PM2.5 and 14 µg/m3 forPM10 over the two years of the study; none of these differenceswere statistically significant (Figure 2). Developed countrystudies have generally found higher air pollution in low SEScommunities (4), which is consistent with our finding aboutJT residential sites. There was no apparent relationshipbetween PM and SES gradient at our traffic sites.

Beyond differences in absolute levels, 48-h concentrationsof PM were correlated across neighborhoods. Pair-wise cross-neighborhood correlation coefficients between residentialsites when there were simultaneous measurements in allneighborhoods ranged from 0.86 to 0.98 for PM2.5 and from0.93 to 0.998 for PM10, excluding the Harmattan period.

Continuous PM. After standardizing for RH, daily PM atresidential and traffic sites in all neighborhoods increasedfrom around 03:00, peaking or stabilizing between 06:00 and07:00 (Figure 3). This morning rise is likely to be caused bythe schedule of major sources, including morning residentialand small commercial cooking, bakery operation, andmorning traffic. In simultaneous measurements, this morningrise occurred earlier in JT than in NM, and earlier in NM thanin EL. This pattern is consistent with two possible explana-tions: first, prevailing winds from the southwest may carryPM over the city, leading to the observed staggering of PMrise based on the location of downwind neighborhoods. Asecond explanation is that cooking and traffic patterns differacross neighborhoods, contributing to the differences intiming of the morning rise in PM. Distinguishing the role ofthese factors requires further investigation.

Traffic sites in all neighborhoods and residential sites inJT and NM also had a midday peak which may correspondto midday commercial traffic and cooking of street food.After a slight decline in the afternoon, all sites also had abroad and attenuated PM peak between 17:00 and 20:00; therelative magnitude of this evening peak varied across sites.This evening peak may correspond to the evening rush hour,biomass use for household cooking, and the use of kerosenelamps for lighting as sundown occurs around 18:00 in Accra.

FIGURE 3. Continuous PM concentration (a) PM2.5 and (b) PM10. The measurements were standardized for variation in RH throughoutthe day and corrected against gravimetric measurements as described in Materials and Methods. In each panel, measurements fromall days over the measurement period are averaged. The peaks in individual days are narrower and vary in their timing. See FigureS4 for results without standardization for RH.

2274 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 7, 2010

Most of these activities occur over a longer period at nightthan in the morning.

Figure S5 shows continuous PM corrected against thegravimetric measurements but not standardized for RH.Comparison with Figure 3 shows that standardization forRH variability throughout the day qualitatively changes theresults and should be considered in analysis of within-daypatterns of PM.

Integrated CO. The available CO measurements werelower at residential sites than at the corresponding trafficsites in the same neighborhood, with differences betweenresidential and traffic sites within a neighborhood rangingfrom 9 to 31 ppm (Figure 4). The exception was AD-R1, whichhad CO levels very similar to AD-T. AD-R1 was located on asecondary road with medium-heavy traffic during morningand evening commute times. This pattern shows the influ-ence of mobile sources on CO, even in relation to biomassburning as the NM residential site was located in an areawhere residents commonly used biomass.

With the exception of one residential site in EL, availableCO measurements at our measurement sites were near orwell above the United States Environmental ProtectionAgency (U.S. EPA) National Ambient Air Quality Standards(NAAQS) of 9 ppm for an 8-h average (Figure 4) (26).

AcknowledgmentsFunding for this research was provided by the NationalScience Foundation (Grant 0527536). Laboratory support wasprovided by the Harvard NIEHS Center for EnvironmentalHealth. We thank the residents of Nima, Jamestown/Ushertown, Asylum Down, and East Legon for their hospi-tality, Nana Prempeh and Adam Abdul Fatah for fieldassistance, and the Legal Resources Center and the Depart-ment of Geography and Resource Development at theUniversity of Ghana for valuable help with logistical ar-rangements. Julian Marshall and Adam Both providedvaluable advice regarding the standardization of continuousPM for RH.

Supporting Information AvailableDetailed measurement and statistical methods (Text S1), andstudy strengths and limitations (Text S2); population densityand fuel use in the study neighborhoods (Table S1), char-acteristics of measurement sites (Table S2), details of datacollection by site (Table S3), and PM2.5:PM10 ratios bymeasurement site and season (Table S4); map of studyneighborhoods and measurement sites (Figure S1), mea-surement schedules in study years 1 and 2 (Figure S2),integrated PM concentrations on a log-scale (Figure S3),average daily pattern of RH during the study period (FigureS4), and continuous PM concentration without standardiza-tion for variation in RH throughout the day (Figure S5). This

information is available free of charge via the Internet athttp://pubs.acs.org/.

Literature Cited(1) United Nations Department of Economic and Social Affairs

(Population Division). World Urbanization Prospects: The 2006Revision; Population Division of the Department of Economicand Social Affairs of the United Nations Secretariat: New York,2007.

(2) Cohen, A.; Anderson, R.; Ostro, B.; Pandey, K. D.; Krzyzanowski,M.; Kunzli, N.; Gutschmidt, K.; Pope, A.; Romieu, I.; Samet, J.,et al. Urban ambient air pollution. In Comparative Quantifica-tion of Health Risks: Global and Regional Burden of DiseaseAttributable to Selected Major Risk Factors; Ezzati, M., Lopez,A. D., Rodgers, A., Murray, C. J. L., Eds.; World HealthOrganization: Geneva, 2004; pp 1353-1433.

(3) Buzzelli, M.; Jerrett, M. Racial gradients of ambient air pollutionexposure in Hamilton, Canada. Environ. Plan. A 2004, 36 (10),1855–1876.

(4) O’Neill, M. S.; Jerrett, M.; Kawachi, I.; Levy, J. I.; Cohen, A. J.;Gouveia, N.; Wilkinson, P.; Fletcher, P.; Cifuentes, L.; Schwartz,J. Health, wealth and air pollution: Advancing theory andmethods. Environ. Health Persp. 2003, 111, 1861–1870.

(5) Levy, J. I.; Houseman, E. A.; Spengler, J. D.; Loh, P.; Ryan, L. Fineparticulate matter and polycyclic aromatic hydrocarbon con-centration patterns in Roxbury, Massachusetts: a community-based GIS analysis. Environ. Health Persp. 2001, 109 (4), 341–7.

(6) Weijers, E. P.; Khlystov, A. Y.; Kos, G. P. A.; Erisman, J. W.Variability of particulate matter concentrations along roads andmotorways determined by a moving measurement unit. Atmos.Environ. 2004, 38 (19), 2993–3002.

(7) Holmes, N. S.; Morawska, L.; Mengersen, K.; Jayaratne, E. R.Spatial distribution of submicrometre particles and CO in anurban microscale environment. Atmos. Environ. 2005, 39 (22),3977–3988.

(8) Hoek, G.; Meliefste, K.; Cyrys, J.; Lewne, C.; Bellander, T.; Brauer,M.; Fischer, P.; Gehring, U.; Heinrich, J.; van Vliet, P.; , et al.Spatial variability of fine particle concentrations in threeEuropean areas. Atmos. Environ. 2002, 36, 4077–4088.

(9) Puustinen, A.; Hameri, K.; Pekkanen, J.; Kulmala, M.; Hartog,J. d.; Meliefste, K.; Brink, H. t.; Kos, G.; Katsouyanni, K.;Karakatsani, A.; , et al. Spatial variation of particle number andmass over four European cities. Atmos. Environ. 2007, 41, 6622–6636.

(10) Songsore, J.; McGranahan, G. The political economy of house-hold environmental management: Gender, environment andepidemiology. World Dev. 1998, 26 (3), 395–412.

(11) Etyemezian, V.; Tesfaye, M.; Yimer, A.; Chow, J. C.; Mesfin, D.;Nega, T.; Nikolich, G.; Watson, J. G.; Wondmagegn, M. Resultsfrom a pilot-scale air quality study in Addis Ababa, Ethiopia.Atmos. Environ. 2005, 39 (40), 7849–7860.

(12) Padhi, B. K.; Padhy, P. K. Assessment of intra-urban variabilityin outdoor air quality and its health risks. Inhalat. Toxicol. 2008,20 (11), 973–979.

(13) Jackson, M. M. Roadside concentration of gaseous and par-ticulate matter pollutants and risk assessment in Dar-es-Salaam,Tanzania. Environ. Monit. Assess. 2005, 104 (1-3), 385–407.

(14) Weinstein, J. P.; Hedges, S. R.; Kimbrough, S. Characterizationand aerosol mass balance of PM2.5 and PM10 collected in Conakry,Guinea during the 2004 Harmattan period. Chemosphere 2010,78 (8), 980–988.

(15) Obiri-Danso, K.; Hogarh, J. N.; Antwi-Agyei, P. Assessment ofcontamination of singed hides from cattle and goats by heavymetals in Ghana. African J. Environ. Sci. Technol. 2008, 2 (8),217–221.

(16) Arku, R. E.; Vallarino, J.; Dionisio, K. L.; Willis, R.; Choi, H.;Wilson, J. G.; Hemphill, C.; Agyei-Mensah, S.; Spengler, J. D.;Ezzati, M. Characterizing air pollution in two low-incomeneighborhoods in Accra, Ghana. Sci. Total Environ. 2008, 402(2-3), 217–31.

(17) Breuning-Madsen, H.; Awadzi, T. W. Harmattan dust depositionand particle size in Ghana. CATENA 2005, 63 (1), 23–38.

(18) Afeti, G. M.; Resch, F. J. Physical characteristics of Saharan dustnear the Gulf of Guinea. Atmos. Environ. 2000, 34, 1273–1279.

(19) McTainsh, G. Harmattan dust deposition in northern Nigeria.Nature 1980, 286, 587–588.

(20) Chakrabarti, B.; Fine, P. M.; Delfino, R.; Sioutas, C. Performanceevaluation of the active-flow personal DataRAM PM2.5 massmonitor (Thermo Anderson pDR-1200) designed for continuouspersonal exposure measurements. Atmos. Environ. 2004, 38,3329–3340.

FIGURE 4. Integrated CO by neighborhood and site.

VOL. 44, NO. 7, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2275

(21) Day, D. E.; Malm, W. C.; Kreidenweis, S. M. Aerosol lightscattering measurements as a function of relative humidity. J.Air Waste Manage. 2000, 50, 710–716.

(22) Laulainen, N. S. Summary of conclusions and recommenda-tions from a visibility science workshop: technical basis andissues for a national assessment of visibility impairment; PNL-8606; U.S. Department of Energy, Pacific Northwest Labora-tory, 1993.

(23) World Health Organization (WHO). Air Quality Guidelines:Global Update 2005; WHO Regional Office for Europe: Copen-hagen, 2006.

(24) Aboh, I. J. K.; Henriksson, D.; Laursen, J.; Lundin, M.; Ofosu,F. G.; Pind, N.; Lindgren, E. S.; Wahnstrom, T. Identification

of aerosol particle sources in semi-rural area of Kwabenya,near Accra, Ghana, by EDXRF techniques. In EuropeanConference on X-Ray Spectrometry, Cavtat, Dubrovnik, Croatia,2009.

(25) Sunnu, A.; Afeti, G.; Resch, F. A long-term experimental studyof the Saharan dust presence in West Africa. Atmos. Res. 2008,87 (1), 13–26.

(26) U.S. Environmental Protection Agency. National Ambient AirQuality Standards (NAAQS); Office of Air Quality Planning andStandards, September 30, 2009.

ES903276S

2276 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 7, 2010