Source apportionment and seasonal variation of PM 2 . 5 in a Sub-Saharan African city : Nairobi , Kenya

Sources of airborne particulate matter and their seasonal variation in urban areas in Sub-Saharan Africa are poorly understood due to lack of long-term measurement data. In view of this, filter samples of airborne particulate matter (particle diameter ≤2.5 μm, PM2.5) were collected between May 2008 and April 2010 at two sites (urban background site and suburban site) within the Nairobi metropolitan area. A total of 780 samples were collected and analyzed for particulate mass, black carbon (BC) and 13 trace elements. The average PM 2.5 concentration at the urban background site was 21 ± 9.5 μg m−3, whereas the concentration at the suburban site was 13 ± 7.3 μg m−3. The daily PM2.5 concentrations exceeded 25 μg m −3 (the World Health Organization 24 h guideline value) on 29 % of the days at the urban background site and 7 % of the days at the suburban site. At both sites, BC, Fe, S and Cl accounted for approximately 80 % of all detected elements. Positive matrix factorization analysis identified five source factors that contribute to PM 2.5 in Nairobi, namely traffic, mineral dust, industry, combustion and a mixed factor (composed of biomass burning, secondary aerosol and aged sea salt). Mineral dust and traffic factors were related to approximately 74 % of PM 2.5. The identified source factors exhibited seasonal variation, apart from the traffic factor, which was prominently consistent throughout the sampling period. Weekly variations were observed in all factors, with weekdays having higher concentrations than weekends. The results provide information that can be exploited for policy formulation and mitigation strategies to control air pollution in Sub-Saharan African cities.


Introduction
The air quality in major cities within Sub-Saharan Africa (SSA) is deteriorating due to several factors, which include an increasing urban population, unregulated traffic activities, poorly maintained vehicles, inadequate regulations and air pollution control policies (Zachariadis et al., 2001;van Vliet and Kinney, 2007;UN, 2010).In order to draft environmental regulations to control air pollution, high-quality and longterm measurement data are required.However, such data are not readily available in the SSA region due to a lack of longterm air pollution monitoring (Petkova et al., 2013;UNEP, 2013).Most studies on aerosols in cities within SSA are limited and usually based on short-term measurements (Gatari et al., 2009;Mkoma et al., 2010;Odhiambo et al., 2010;Sawyer, 2010).As a result, there is a poor understanding of long-term variations of airborne particulate matter (PM) and little is known about its sources, elemental constituents and seasonal variation.
Vehicle-related emissions (both exhaust and non-exhaust) account for a large portion of urban air pollution in developing countries (van Vliet and Kinney, 2007;Kinney et al., 2011).A study conducted in Nairobi by Kinney et al. (2011) reported that concentration levels of PM 2.5 at street level, next to a city road, frequently exceeded the World Health Organization (WHO, 2000) 24 h guideline of 25 µg m −3 .This poses serious questions concerning health risks for people who frequent the city on a daily basis.Vehicle-related emissions in the SSA region are exacerbated by the import of second-hand vehicles, which, for instance, in Kenya can be S. M. Gaita et al.: Source apportionment and seasonal variation of PM 2.5 in Nairobi up to 8 years old (KEBS, 2013b).In addition, the road network in Nairobi is progressively improving and this has translated into increased motorized traffic activities with a large part of the increased traffic load coming from old and poorly maintained vehicles.This improved road network has led to severe congestion problems and the exposure of many pedestrians to air pollution, especially during morning and evening rush hours (Graeff, 2010).Recently, there has been a rapid increase (from 50 000 units in 2008 to 140 000 units in 2011) in importation and usage of motorcycles in urban and rural areas (KEBS, 2013a).
Mineral dust accounts for a significant portion of PM and originates from unpaved road surfaces as well as wind-blown dust during dry seasons (Boman et al., 2009;Reeves et al., 2010;Lindén et al., 2012).The natural background concentration of mineral dust in Africa is bound to be influenced by increasing episodes of droughts (WMO, 2013), as well as increased anthropogenic activities which are related to economic growth.This will pose a challenge to formulation of strategies aimed at mitigating mineral dust effects on air quality.However, quantification of the contribution of mineral dust to urban PM will provide necessary information that can be used in drafting relevant air quality legislation.On a regional level, emissions from biomass burning are also significant and dependent on season (wet or dry).Savannah fires account for a large percentage of carbonaceous aerosols detected in regional PM, as has been reported from the extensive research campaigns carried out in southern Africa during the Southern African Regional Science Initiatives in 1992 and 2000 (SAFARI 1992 andSAFARI 2000) (Cahoon et al., 1992;Lindesay et al., 1996;Cachier et al., 1998;Andreae and Merlet, 2001;Formenti et al., 2003;Swap et al., 2003).
In Kenya, an air quality regulation has been drafted which is referred to as "The Environmental Management and Coordination (Air Quality) Regulations, 2008" (NEMA, 2013).This regulation does not set any air quality guidelines but instead proposes their formulation.It is interesting to note that the draft of air quality regulations in Kenya allows for burning of vegetation in the savannah grassland as part of cultural and traditional practices.
This paper presents results from a long-term measurement campaign initiated in 2008/2009 at two sites in the Nairobi metropolitan area.Samples were analyzed for their elemental content using X-ray fluorescence spectroscopy, and the results were evaluated using positive matrix factorization (PMF).The results are compared with available data from previous short-term studies in Nairobi and their implications for urban air pollution mitigation strategies are discussed.The overall aim is to identify sources of PM 2.5 and evaluate their seasonal variations based on measured concentrations of PM 2.5 , black carbon and 13 trace elements.

Description of the study area
The city of Nairobi lies some 200 km south of the Equator and occupies an area of about 684 km 2 .It has a population of about 3.2 million people, with a daytime population of about 4 million (KOD, 2012).The larger Nairobi metropolitan area had a population of 6.1 million as of 2007 and it is expected to grow to over 12 million by 2030 (MoNMD, 2008).This increase in population has resulted in an increase in commercial, traffic and industrial activities, especially within the central business district.The peri-urban areas of the city continue to experience both controlled and uncontrolled development in buildings and infrastructure, especially mushrooming of informal settlements (Karanja and Makau, 2012).In these settlements, solid waste is usually burned in open air and the majority of the households use kerosene and biomass-based fuels (charcoal, wood and vegetation) for domestic cooking (Egondi et al., 2013).The city's industrial area is located to the east and southeast and dominated by small-and medium-sized industries, which include food processing, power generation, chemical processing industries, battery manufacturing and scrap metal recycling.

Climate of the study area
Nairobi experiences a moderate climate even though it is located along the Equator.At a height of about 1795 m above sea level, the city has a subtropical high climate according to the Köppen climate classification (Peel et al., 2007), as opposed to the expected tropical climate.The regional climate of East Africa is influenced by the seasonal displacement of the Intertropical Convergence Zone (ITCZ) (Henne et al., 2008).The north-south-north annual shift of the ITCZ is driven by the apparent annual motion of the Sun and results in monsoon trade winds, which give rise to well-defined wet and dry seasons (Camberlin and Philippon, 2002;KMD, 2013).Severe changes in this cycle usually result in droughts or flooding in the East African region (WMO, 2013).The average annual rainfall is approximately 900 mm but varies from less than 500 mm to more than 1500 mm between years.The average daily temperature varies from about 17 • C in July and August to 20 • C in March (KMD, 2013).

Sampling sites and PM 2.5 sampling procedure
Sampling of PM 2.5 was carried out at two sites within the Nairobi metropolitan area.The first site was at the University of Nairobi main campus (1.279 • S, 36.817• E) (herein referred to as university and marked S1 in Fig. 1), which is within the city center.The second sampling site was established in a suburban area, within the compound of the United Nations Environmental Programme headquarter (1.234 • S, 36.818• E) (herein referred to as UNEP and marked as S2 in Fig. 1) and about 5 km to the north of the university site.
Table 1.Summary of the results from all filter samples for the collection period 22 May 2008 to 30 March 2010 at two sites in Nairobi, Kenya, which include detection limits, range, mean concentrations (ng m −3 ), standard deviations (SD) and percentage composition (% PM 2.5 ) for detected trace elements, BC and PM 2.5 .N is the number of valid samples.Cyclone samplers (Casella, Bedford, UK) were used to collect fine particles (PM 2.5 ) on pre-weighed polycarbonate filters with 0.4 µm pore size (Whatman International Ltd., Maidstone, UK).Samples were collected for 24 h per sample at a flow rate of 3 L min −1 , except on weekends, when samples were collected for 48 h per sample.At both the university and UNEP sites, sampling started at 10:00 local time (UTC+3).At the university site, PM 2.5 samples were collected from 22 May 2008 to 2 April 2010.The cyclone sampler was mounted on the rooftop of the engineering department building at about 17 m above ground level (a.g.l.).This site, being at an elevated location and close to the city center, represents an urban background setting.

Elements Detection limit
At the UNEP site, sampling was carried out from 16 April 2009 to 30 March 2010.The cyclone setup was placed on the rooftop of the administration block about 10 m a.g.l.The site is located in a suburban area compared to the university site, is surrounded by two local forests (Karura to the east and Gigiri to the south) and comprises office blocks and lowdensity residential houses.
A total of 780 valid samples were collected, of which 502 and 278 were from the university and UNEP sites, respectively.Loaded filters were analyzed for PM 2.5 mass concentration, black carbon (BC) and trace elements (from S to Pb).The PM 2.5 mass concentration was determined gravimetrically using a microbalance (Mettler Toledo model TM5).Empty and loaded filters were weighed after being conditioned at a relative humidity of 50 ± 10 % and a temperature of 20 ± 2 • C for 24 h.The BC concentration was analyzed using a BC reflectometer (ESM Emberline, model FH62 1-N).The reflectometer measures the absorption and reflection properties of the sample-loaded filter, whereby the reflected light intensity at 650 nm is observed to nonlinearly decrease as BC concentration increases on the filter (Gatari and Boman, 2003;Moosmüller et al., 2009).
Trace elements were determined using energy dispersive X-ray fluorescence (EDXRF) spectroscopy.The spectrometer uses a Philips diffraction X-ray tube and a Mo secondary target.The secondary target consists of a 1 mm thick Mo plate of 99.99 % purity.Fluoresced secondary X-rays propagate through two Ag collimators, giving a relatively focused beam of near-monochromatic X-rays for sample excitation.The spectrometer is laboratory built (University of Gothenburg, Sweden) in an optimized three-axial geometry that gives good signal-to-noise ratios for the measured elements (Boman, 1991).It was operated at a voltage of 50 kV and a current of 20 mA, and spectral information was acquired for a live time of 1000 s.The spectra were analyzed using AXIL (Analysis of X-ray spectra by Iterative Leastsquare fitting), which is a subprogram of the QXAS (Quantitative X-ray Analysis System) software from the International Atomic Energy Agency (Bernasconi et al., 2000).The subprogram assists in spectrum conversion and fitting.In this study, a calibration file was created using single element standards.The standards were run, fitted and evaluated for element's net peak area (P A ) and associated background area (B A ).The obtained P A and B A together with the known concentration (C A ) of the element were used to calculate the detection limit (DL A ; see Table 1) using the IUPAC equation below (van Grieken and Markowicz, 1993): B A P A In order to correct for the positive artifacts from organic carbon (OC), measured concentration values from the field blanks were subtracted from the measured samples' concentration values (Watson et al., 2009).

Data treatment and analysis
PM 2.5 , trace element and BC concentrations were further analyzed using PMF for purposes of source apportionment.PMF is a multivariate factor analysis method that decomposes a matrix of speciated sample data into factor contributions and profiles, which are then interpreted as to what source types they represent (Paatero, 1997).The EPA PMF 3.0 program from USEPA (Norris et al., 2008) was used for the analysis.Data below the detection limit (DL) were substituted with one-half of the respective DL, and uncertainty was set to ±5/6 times the DL (Polissar et al., 2001).During PMF analysis the number of factors was varied by between three and nine factors.From the analysis output, only five factors were seen to give meaningful results for possible PM sources and are thus presented herein.
Meteorology data for this study were collected from Jomo Kenyatta International Airport (JKIA), which is approximately 13 km to the southeast of the university site (data were purchased from AccuWeather Enterprise Solutions, New York, USA).In addition, back trajectories of air masses arriving at 100 m above the sampling sites were generated using the HYSPLIT_4 model (Draxler and Hess, 1998).Daily back trajectories were run for 72 h, clustered on a monthly basis and compared with monthly averages of measured hourly wind speeds and directions.

Meteorology during the sampling period
Sampling period was characterized by periods of long rains (mid-March to May), short rains (October to November) and dry periods (from January to mid-March and from June to September) in 2008 and 2010.However, in 2009 there was a pronounced drought (from March to October) and as a result, lower than normal amount of rainfall was recorded during the long rains season (Zwaagstra et al., 2010;Williams and Funk, 2011).
Measured wind direction mainly oscillated between north and east (Fig. 2) for the entire sampling period.Average hourly wind speed for entire sampling period was 4 m s −1 and varied from 0 to 33 m s −1 .On a monthly basis, the wind direction was predominantly northeasterly during the period from October to March, whereas the direction was mainly easterly and southeasterly during the months of April to September.The average relative humidity was 67 % and varied from 10 to 100 %, while the average temperature was 20 • C and ranged from a minimum of 8 • C to a maximum of 32 • C.
Figure 3 shows clustered back trajectories for the sampling period and illustrates that air masses arriving in Nairobi area mainly originated from the southeasterly (52 %) and easterly (48 %) directions.From April to October, the air masses originated from the southeasterly direction (back trajectories numbered 2 and 3), whereas from November to March, the direction of origin was easterly and northeasterly (back trajectories 1 and 4).There was a discrepancy between calculated back trajectories and recorded wind direction in the months of September and October.This discrepancy between local wind directions and regional air mass movement indicates that the Nairobi PM is influenced by both local wind dynamics and regional air mass movement.

PM 2.5 mass and elemental concentrations
Results for PM 2.5 BC and 13 trace elements concentrations from the university and UNEP sites are presented in Tables 1  and 2. The statistics in Table 1 are from combined data sets Table 2. Summary of the results from the university and the UNEP site for the time period from 16 April 2009 to 30 March 2010: range, mean concentrations and standard deviations (SD) for detected trace elements, BC (in ng m −3 ) and PM 2.5 (in µg m −3 ).N is the number of valid samples.from the two sites and depict a general picture of PM 2.5 and its constituents in the Nairobi metropolitan area.From all the filters collected, approximately 29 % of the elemental contents were detected, with the balance of 71 % attributed to the contribution from organic matter, nitrates, aluminosilicates and oxides of both detected and undetected elements.BC has the highest average concentration and accounted for 15 % of the PM 2.5 concentration, whereas trace elements accounted for 14 %.Average Fe concentration is the highest among the detected elements, followed by S and Cl, in that order.The average lead concentration is well below the WHO annual guideline concentration of 500 ng m −3 (WHO, 2000).This is probably due to the phasing-out of leaded gasoline in 2006 in Kenya.Due to the high altitude of Nairobi, a TEL-B mixture was being used (before 2006) as an anti-knock additive in gasoline, resulting in a Br / Pb ratio of 0.77 in particulate matter from fresh exhaust (Harrison and Sturges, 1983).
Table 2 shows the comparison of PM 2.5 and elemental concentrations at the university and UNEP sites for the common sampling period (16 April 2009 to 30 March 2010).At the university site, PM 2.5 concentration ranges from 3 to 53 µg m −3 , with an overall mean of 21 µg m −3 .The average PM 2.5 concentration exceeds the annual WHO guideline limit of 10 µg m −3 by a factor of 2. The 24 h WHO guideline was exceeded 29 % of the total sampling days.Elemental composition of PM 2.5 at the university site is dominated by BC, Fe, S and Cl.Average BC concentration accounts for 19 % of the of the total PM 2.5 concentration, whereas Fe, S and Cl accounted for a total of 9 %.
At the UNEP site PM 2.5 concentration ranges from 1.9 to 36 µg m −3 , with an overall mean of 13 µg m −3 .The average PM 2.5 concentration is about 30 % higher than the annual WHO guideline.At this site, the 24 h WHO guideline was exceeded 7 % of the total sampling days.BC has the highest average concentration of the identified components and individually accounts for 11 % of the total PM 2.5 concentration whereas all other elements account for 15 %.The average concentration of S is the second highest, followed by Cl, Fe and K, in that order.
The average PM 2.5 concentration at the university site is about 50 % higher than at the UNEP site.This observed difference points to inter-site variation due to factors such as the location of the measurement site, which in this case was that the UNEP site is located in a suburban area surrounded by forests (Fig. 1).The average BC concentration at the university site is more than twice the BC concentration at the UNEP site.Similarly, concentrations of Ca, Mn, Fe and Br show marked difference between the two sites.Despite the sites being in close proximity to each other, the university site, an urban background site, is seen to be under greater influence from PM sources compared to the UNEP site.In addition, observed ranges and standard deviations of PM 2.5 and elements concentrations at the two sites indicate variability of the PM sources at both sites during the sampling period.

Seasonal variation of the PM 2.5 , BC and some selected elements
Figure 4 shows the seasonal variation of PM 2.5 and some selected elements at the university site.The seasonal average concentration for PM 2.5 is highest during the cold and dry season in 2009 (25 µg m −3 ) and lowest during the short rains season (8.9 µg m −3 ) in the same year.The high seasonal averages in 2009, especially during the long rains season, are linked to the then ongoing drought.WHO air quality guideline for a 24 h period was frequently exceeded during this drought period.High BC concentrations were observed during the cold and dry season in 2009, as well as during the  Figure 5 shows the seasonal variation of PM 2.5 and selected elements at the UNEP site.The general trend for the PM 2.5 concentration is that of variation from high concentrations in 2009 to lower concentrations in 2010.A similar trend was observed for Zn, Pb, S, Fe and BC.However Cl was observed to be somewhat consistent throughout the sampling period.The highest seasonal average concentration (19 µg m −3 ) for PM 2.5 is measured during the cold and dry season in 2009, while the lowest seasonal average concentration is reported during the hot and dry season in 2010.This observation is supported by the high number of rainy days manually recorded at that site during the said period compared to the other part of the sampling period.
From comparison of the two sites, the observed seasonal variation is similar in 2009 but different in 2010.The observed trend for Pb concentrations at both sites shows similarity, and it is probably intermittent due to the decrease in airborne Pb as a result of banning the use of Pb in gasoline sold in Kenya.The observed trend of BC concentrations at both sites is equally similar, pointing to a common influence from possible local and regional sources.These observed differences in variations are attributed to differences in number and influences of possible PM sources, surrounding features such as forests at the UNEP site, and differences in observed weather patterns at each site.

Weekly variation of PM 2.5 , BC and some selected elements
Figure 6 shows the weekly variation of average concentrations of PM 2.5 , BC, S, Cl, K, Fe, Zn and Pb calculated from the combined data of samples collected at the university and UNEP sites.In general, there is a common trend in all the indicated concentrations, with weekdays (Monday to Friday) having higher concentration compared to the weekends (Saturdays and Sundays).This points to the influence of anthropogenic activities on airborne PM in Nairobi given that there is heightened human-related activities during the weekdays and minimal activities during the weekends (Kinney et al., 2011).The concentrations of PM 2.5 , BC, Cl, K and Fe do not display significant variations within the weekdays, and this observation indicates consistency of their possible sources.The concentrations of S, Zn and Pb display variations within the weekdays, indicating variability in their sources.

PMF analysis and source apportionment
PMF was applied to the combined data from the two sites to identify major PM 2.5 sources.The results from a five-factor solution are shown in Figs.7 to 11 and the individual factors are discussed hereinafter.
The first factor is attributed to both exhaust and nonexhaust traffic emissions (Fig. 7).The factor is characterized by a high contribution of BC, S, K, Fe and Zn and accounts for 39 % of the PM 2.5 .The factor shows a similar pattern as the elemental profile of PM 2.5 samples collected from a traffic-polluted site in Connecticut, USA (Gent et al., 2009).The factor is identified with exhaust traffic emissions due to the presence of S and BC (Zachariadis et al., 2001;Arku et al., 2008;Lestari and Mauliadi, 2009).The high S concentration comes from diesel sold in Kenya that contains 0.5 % S, and the visible black smoke from the exhaust pipes of many vehicles indicates incomplete combustion resulting in high emissions of BC.Furthermore, there are no regulations on vehicular emissions and the majority of the cars are imported as second-hand, although there is an age limit of 8 years.Non-exhaust traffic emissions such as resuspension of road dust (Fe, K, Ca and traces of Pb) and tire abrasion (Zn) also contribute to this factor (Chueinta et al., 2000;Salma and Maenhaut, 2006;Pant and Harrison, 2013).
The second factor is attributed to mineral dust (Fig. 8).It is dominated by crustal elements (Fe, Ca, K and Ti) and has a pattern similar to that of elements from soil samples collected across SSA, which were analyzed using the total Xray fluorescence technique (Towett et al., 2013).This factor highlights the contribution of local soil to PM 2.5 in Nairobi.The prolonged dry periods during the campaign contribute to the increased mineral dust loading to the sampled PM 2.5 .The absence of S, Zn and Pb in this factor marks the difference between the origin of the mineral dust in this factor and the resuspended road dust in the first factor, hence justifying both dust components as separate entities.
The third factor is attributed to a mixture of biomass burning, secondary aerosol particles and aged sea salt (Fig. 9).This mixed factor accounts for 13 % of the mass concentration.The presence of Cl and S points to marine influence, as shown by observed back trajectories (Fig. 3) that originate from the Indian Ocean, as discussed in the meteorology section (Sect.3.1).This implies that marine aerosols reaching Nairobi undergo, to some extent, partial replacement of Cl by S during the long-range travel from the Indian Ocean.Sea salt particles have been reported to lose Cl due to reaction with acidic compounds, sulfuric acid in this case, thus forming sulfate aerosol particles (Pio and Lopes, 1998;Nyanganyura et al., 2007).The S component originates from diesel and heavy fuel oil consumption within the Nairobi region.However, the inclusion of sea salt as part of the mixed factor is not strong enough, as this study did not measure Na, which is a better indicator of sea salt than Cl given that the latter can be lost from the particle phase during atmospheric transport (Andreae and Crutzen, 1997;Oum et al., 1998).The contribution from biomass burning is indicated by the presence of K, Zn and Rb.Biomass burning is a common phenomenon in the savannah plains that are located to the south of the Equator, as reported during the SAFARI campaigns (Lindesay et al., 1996;Formenti et al., 2003).
The fourth factor is attributed to combustion processes and accounts for 6 % of the mass concentration.It is characterized by BC, S, K, Zn and Pb (Fig. 10) and shows similarity with a combustion factor from a study in Skopje, Macedonia (Kovacevik et al., 2011), which was associated with biomass and fossil fuel combustion.The combustion activities that are likely to contribute to PM 2.5 in Nairobi include use of heavy fuel oil in industries and in thermoelectric power generation (BC and S), solid waste burning (BC, Zn and Pb), and domestic charcoal/wood fires/biomass burning (BC, K and Br) in the low-income households and slums, which account for over 80 % of settlements in Nairobi metropolitan area (Karanja and Makau, 2012;Egondi et al., 2013).
The fifth factor is attributed to industrial emissions and accounts for 7 % of the mass concentration.It is characterized by presence of Zn, Fe, Cl and K (Fig. 11).It is identified with emissions from metal recycling industries, which depend on scrap metals as raw materials (Querol et al., 2007).These metal recycling industries are located on the eastern and southeastern part of the city, which coincide with the general wind and air mass flow directions (Figs. 2 and 3).Contribution of Pb in this factor is below 1 %, but high concentrations of up to 427 ± 124 µg m −3 have been reported from a recycling plant in an industrial area (Were et al., 2012).
Figure 12 summarizes the percentage contribution of the identified PM 2.5 source factors.It is evident that traffic and mineral dust account for approximately 74 % of PM 2.5 in Nairobi.This observation points to the significance of anthropogenic influence on air quality given that traffic is human-related, whereas the mineral dust component originates from both natural and human-related activities.

University site
Traffic and mineral dust factors dominate at the university site (Fig. 13) and are generally inherent during the entire sampling period.It is important to note that the dip observed during the short rains season in 2009 was partly due to high uncertainty as a result of the majority of the filters collected at that time exhibiting concentrations below the detection limit.The mineral dust factor is prominent due to the then ongoing drought and its seasonal percentage contribution in 2009 ranges from 33 to 55 %.The drought effect is evident in that, during the long rains season in 2009, no significant change in mineral dust contribution (with seasonal percentage contribution of 33 %) is noted when compared with the cool and dry season in the same year (seasonal percentage contribution of 40 %).A low amount of rainfall ( 223 The traffic factor is almost consistent throughout the sampling period, except during the short rains season of 2009.During this period the seasonal percentage contribution was 11 % compared to other seasons, which had a range of between 35 and 69 %.The factor's contribution is seen to dominate in the beginning of 2010, with percentage contribution of 49 % during the hot and dry season.Further statistical analysis of data in relation to the traffic factor shows that the average ratio of Pb to PM 2.5 during the hot and dry season in 2009 deviates from the average ratio of Pb to PM 2.5 (0.0014) obtained for all seasons.Percentage-wise, it is 25 % higher than the average ratio.In addition, the average Br / Pb ratio in this season is 0.64, compared to a ratio of 0.77 reported from fresh vehicular exhaust by Harrison and Sturges (1983).This indicates that part of the Br from gasoline emissions was present in the vapor phase, as also reported from the city of Butare, Rwanda, by Maenhaut and Akilimali (1987).In addition, during this season the wind blows mainly from the northeast (the same direction as Thika Road and Globe Roundabout, which is the entry point for vehicles from central and northern Kenya).Furthermore, the contribution from tire abrasion was also a possibility (Salma and Maenhaut, 2006) given that many on-road vehicles rely on second-hand tires.
The mixed factor, though not prominent, is consistent throughout the sampling period but slightly pronounced during the short rains season in 2009, with a seasonal contribution of 26 %.During this season, wind direction is usually easterly to northeasterly, and therefore vehicular emission from the city center contributes S to this factor due to the S emissions from vehicles, as described in the sections above.In addition, there is a marine influence, as indicated by the presence of Cl.This is evident since back trajectories show that air masses arriving in Nairobi principally originate from the Indian Ocean (Fig. 8).
The combustion factor was not consistent through the sampling period, but was prominent from the hot and dry season in 2009, with a seasonal percentage contribution of 6 and 8 % during the long rains season of the same year.Thereafter the factor's percentage contribution drops to 4 % for the remaining seasons.This observed trend in combustion factor is an indication of the complexity of combustion sources in Nairobi such as unregulated burning of solid waste, extensive domestic use of kerosene, and use of biomass-based fuels (firewood and charcoal) in slum areas for energy generation (Salma and Maenhaut, 2006;Reinard et al., 2007).
The industrial factor showed a seasonal variation in that seasonal percentage contribution was highest (10 % in each season) during the cool and dry seasons in 2008 and 2009.The percentage contribution was also high during the long rains season in 2009 (8 %), but was significantly low (2 to 4 %) during the short rains, as well as all the hot and dry seasons.This observation highlights the contribution of metallurgical industries, since local winds and regional air masses were generally easterly to southeasterly during the seasons with high percentage contributions.Thus emissions from the said industries, which were located in the same (mainly southeasterly) direction relative to the sampling site, were transported to the site.

UNEP site
Figure 14 shows the trend in source factors at the UNEP site.Contribution of the factors is highest during the cool and dry season in 2009 and is dominated by the mineral dust, traffic and industrial factors.The contribution decreases towards the beginning of 2010.Wind direction during the cool and dry season is generally southerly to easterly, which is the same direction as the city center and industrial area in relation to this site.During the hot and dry season in 2010, the contribution from the mineral dust factor is minimal (percentage contribution of 22 %).This coincided with the observation that the UNEP site recorded 19 rainy days during that season.This observation highlights the erratic nature of rainfall pattern in Kenya during the sampling period (WMO, 2013) and its influence on the airborne particulate matter levels.
The traffic factor is consistently distributed throughout the sampling period but prominent during the long rains season in 2009 with a seasonal percentage contribution of 40 %.This is identified as resulting from resuspension of road dust from the neighboring highways.As was the case at the university site, the calculated Pb / PM 2.5 ratio shows seasonal variation.However the Pb / PM 2.5 ratio is slightly higher (about 28 %) than at the university site, indicating the significance of the road dust compared to other sources at the UNEP site.At the UNEP site, the Br / Pb ratio is 0.43.Even though use of leaded gasoline was banned in 2006 by the Kenya government, it is expected that the Pb concentration in the soil dust will last for long period and be somewhat constant, as reported from other world cities (Xu et al., 2012;Datko-Williams et al., 2014).In absolute concentration values, the university site had a higher concentration (23±16 ng m −3 ) compared to the UNEP site (17±14 ng m −3 ).Moreover, since the university site is close to the city center, it is influenced heavily by PM sources, as characterized by the reported 50 % higher PM 2.5 concentration than the concentration reported at the UNEP site.
The mixed factor is prevalent during the hot and dry season in 2010, with a seasonal percentage contribution of 45 %.During this season the wind direction is usually northeasterly, and thus S components from diesel-related emissions are transported from Thika highway towards UNEP site and its environs.In addition, back trajectories during this period are dominantly easterly and thus traveled over the city center and industrial and populated regions of the Nairobi metropolitan area.
The industrial factor shows a gradual decrease from the beginning of the sampling period towards the end the period.The seasonal percentage contributions show a gradual decline from 14 to 12 to 8 and finally to 0 %.Compared with a similar period at the university site, the observed trend is similar.Therefore, the explanation for this observation is the same as that given for the university site.
Lastly, the combustion factor is consistent throughout the sampling period but shows minimal contribution to PM 2.5 at this site.It does not show any seasonal variability and thus indicates that it is dependent on contribution from regional sources.This observation is attributed to the fact that the site is in a suburban area where combustion activities such as solid waste burning and domestic cooking using biomass, among other things, are not common practice since the area is in uptown and has essential services such as waste collection.

Weekly trend of PMF factors
Figure 15 shows classification of PMF factors by days of the week.Considering individual factors, the mineral dust factor shows the least difference between weekdays and weekends.This observation points to the fact that contribution of dust particles to airborne PM 2.5 is influenced by both anthropogenic activities and natural phenomena.During the sampling period there was heightened activity in road infrastructure, and given the working pattern in Kenya (working on weekdays and resting on weekends), roadconstruction-related activities such as quarrying of road materials were limited to weekdays (Kinney et al., 2011).
Weekly trend observed for the industrial factor is similar to the weekly working pattern in Kenya as explained previously.The majority of metallurgy industries in Kenya still use the open-hearth furnaces inherited from the colonial periods (early 20th century).The mixed factor was about 50 % lower on weekends compared to weekdays, and this implies that the factor was greatly influenced by anthropogenic activities compared to natural activities.This observation is supported by the fact that activities related to S emissions (consumption of diesel with high sulfur content) reduce to a minimum during weekends.
All the five factors have lower concentrations on weekends compared to weekdays.This trend is similar to observed human activities in the city center on a typical week, whereby weekdays are usually busy with both human and traffic activities compared to weekends (Kinney et al., 2011).Therefore it can be observed that all factors are impacted by anthropogenic activities.

Comparison of PM 2.5 with other studies
Since there were no other long-term studies of PM 2.5 and their elemental content in urban areas of Sub-Saharan Africa to compare with, PM 2.5 , elements and BC concentrations from this study were compared with concentrations from a selection of available short-term studies (Table 3).The African cities listed in Table 3 have similar characteristics to Nairobi and are reported to have their air quality influenced largely by mineral dust, combustion activities and vehicular emissions.PM 2.5 concentration from this study was half of that reported from measurements in 2006 and 2007 at a semirural area of the city of Accra, Ghana (Aboh et al., 2009).The main difference is probably the influence of the dust-filled Harmattan winds blowing from the Sahara to Accra.The same can be seen in a comparison between this study and the study in 2007 in Ouagadougou, Burkina Faso (Boman et al., 2009).In another study in Accra, the Harmattan wind conditions were not present during the sampling and the PM 2.5 concentration in this study are similar to the PM 2.5 concentration reported from a 3-week study conducted in 2006 at a low-income residential area in Accra, Ghana (Arku et al., 2008).The proximity to the desert is also reflected in the high PM 2.5 concentration in a 9-month study conducted between September 2010 and May 2011 at Cairo city center (Boman et al., 2013).
In comparison with a 2-week study (not in Table 3) conducted at street level in Nairobi by Kinney et al. (2011), the average PM 2.5 concentration from the university site was 50 % lower.In addition, average PM 2.5 concentration at the university site was a factor of 15 lower than average PM 2.5 concentration reported from a 1-week measurement campaign carried out along one of the major roads in Nairobi (Thika Road) by van Vliet and Kinney (2007).Observations from these two comparisons point to the discrepancy between background PM 2.5 concentrations and those from curb side concentrations in Nairobi.
The elemental composition of PM 2.5 from the semirural area of Accra had large concentrations of S, K and Fe as was the case with the composition of PM 2.5 in Nairobi in both this study and the study by Gatari et al. (2009) in 2001 pointing to a similarity in the sources of particles.A noticeable difference between this study and the ones by Gatari et al. (2009), Aboh et al. (2009) and Arku et al. (2008) is the higher concentration of Fe in the PM 2.5 particles at the university site in this study.This might partly be due to a difference between mineral dust compositions in the two cities but differences in measurement periods also play a role.The elemental concentrations at the low-income area in Accra (Arku et al., 2008) are comparable to the concentrations at the UNEP site in this study, illustrating the importance of the distance between the measurement site and major pollution sources, like the city center and busy roads.
In comparison to Cairo, Egypt, the elemental concentrations from this study were about 3 times lower than the concentrations reported from the 9-month study conducted between September 2010 and May 2011 at a city center site (Boman et al., 2013).PM 2.5 in Cairo had a large contribution from Ca, Cl and S. Since the city is within the Sahara region, its air quality is likely to be influenced by a mineral dust contribution from the desert.The higher concentrations of S and Cl in Cairo might be due to its location close to the Mediterranean Sea.The Pb concentration in Cairo is comparable to the Pb concentration in Nairobi in 2001, and thus higher than in this study.Egypt is one of the few countries in Africa where leaded gasoline is still allowed and the comparison points to the importance of banning leaded gasoline to decrease the health burden on the city inhabitants.

Conclusions
This study has shown that traffic-related emissions (both exhaust and non-exhaust) and mineral dust (both natural and anthropogenic) are both significant contributors to PM 2.5 in Nairobi.It can also be concluded that policies such as banning of Pb in gasoline have shown positive effects on reducing Pb concentration in PM 2.5 and hence reducing the risk of Pb toxicity to the urban inhabitants.Similarly, a reduction of sulfur from diesel fuel could lead to an improved air quality situation in Nairobi.Moreover, prevailing weather and climate patterns have been found to play a significant role by influencing the concentration and composition of airborne PM, as is the case in other cities in SSA.As shown from wind direction and air mass back trajectory analysis, industrial emissions need to be firmly controlled, as the industrial area in Nairobi is located in the upwind direction.Since this study points at factors contributing to the air quality situation in Nairobi, the outcome of this study can be used as a background for further policy actions.The contribution of PM 2.5 from the traffic factor can be controlled by inclusion of emission checks as part of annual vehicle inspection; this would significantly reduce air pollution in the city.On the other hand, natural background concentrations of mineral dust, as well as prevailing weather conditions, would give guidance when formulating any policy aimed at air pollution mitigation.In conclusion, the study has shown that enactment and adherence to air pollution regulations on identified source factors can significantly reduce air pollution levels in cities within SSA.
The Supplement related to this article is available online at doi:10.5194/acp-14-9977-2014-supplement.

Figure 1 .
Figure 1.GIS map of the Nairobi metropolitan area showing land use pattern (a) and transportation network (b).Original maps were modified to include sampling sites as shown on both maps -University of Nairobi (S1) and United Nations Environmental Programme (UNEP) headquarters (S2).Original maps were downloaded from the Center for Sustainable Urban Development's Nairobi GIS maps and database website (CSUD, 2010).

Figure 3 .
Figure 3. Clustered 72 h back trajectories ending in Nairobi at 100 m height above the ground for the period, 22 May 2008 to 30 March 2010.Calculated back trajectories originate principally from the southeast (trajectories 2 and 3) and east (trajectories 1 and 4).

Figure 4 .
Figure 4. Seasonal variation of PM 2.5 and some elements sampled at the university site.Included in the figure are the seasonal averages and standard deviations (enclosed in parentheses) as well as the World Health Organization air quality guideline for a 24 h period.

Figure 5 .
Figure 5. Seasonal variation of PM 2.5 and some elements sampled at the UNEP site.Included in the figure are the seasonal averages and standard deviations (enclosed in parentheses) as well as the World Health Organization air quality guideline for a 24 h period.

Figure 6 .
Figure 6.Weekly variation of PM 2.5 and some selected elements showing the trend based on calculated averages from respective days of the week.

Figure 7 .
Figure 7. Traffic factor from PMF analysis of PM 2.5 from this study compared with PM 2.5 from a traffic-polluted site in Connecticut, USA (Gent et al., 2009).

Figure 8 .
Figure 8. Mineral dust factor from the PMF analysis of PM 2.5 from this study compared with elements from soil samples from SSA countries (Towett et al., 2013).

Figure 9 .
Figure 9. Mixed factor (composed of biomass burning, secondary aerosol and aged sea salt components) from the analysis of sampled PM 2.5 marked by high concentration of S and Cl as well as substantial concentration of K.

Figure 10 .
Figure10.Combustion factor from the PMF analysis of PM 2.5 from this study compared with combustion factor from Skopje, Macedonia(Kovacevik et al., 2011).

Figure 11 .
Figure 11.Industrial factor from analysis of sampled PM 2.5 showing the contribution of metallurgy industries (Fe and Zn) to aerosol in Nairobi.

Figures
Figures 13 and 14 show the seasonal variation of the PMF factors at the university site and the UNEP site, respectively.The seasons are demarcated based on the description of the Nairobi climate by the Kenya Meteorological department (KMD, 2013).

Figure 12 .
Figure 12.Percentage contribution of source factors from PMF analysis to PM 2.5 collected in Nairobi, Kenya.Traffic and mineral dust factors are major contributors to airborne particulate pollutants.

Figure 13 .
Figure 13.Seasonal variation of the five factors obtained from the PMF analysis of the PM 2.5 composition at the university site.The traffic factor shows consistency throughout the sampling period, highlighting the significance of vehicular emissions to urban air pollution.Seasons have been demarcated for clarity.

Figure 14 .
Figure 14.Seasonal variation of the five factors obtained from the PMF analysis of the PM 2.5 composition at the UNEP site.The mixed factor constitutes contribution from biomass burning secondary aerosol and aged sea salt.The effect of the erratic weather pattern is evident from the dampened contribution of mineral dust factor in 2010.Seasons have been demarcated for clarity.

Figure 15 .
Figure 15.Weekly series of the five factors obtained from the PMF analysis of sampled PM 2.5 .The difference between weekdays and weekends highlights the anthropogenic contribution to air pollution in Nairobi.The mixed factor is attributed to contribution from aged sea salt, biomass burning and secondary aerosol.
PM 2.5 concentration at the university site in the month of July 2009 was comparable to the PM 2.5 concentration reported by Gatari et al. (2009) from a study conducted at an industrial background site in Nairobi during July 2001.However, in the month of July 2008, the PM 2.5 concentration was 30 % lower.The agreement between the PM 2.5 concentrations in July 2001 and July 2009, as well as the difference compared to July 2008, illustrates the influence of drought on PM concentration since there was a severe drought in July 2001, as was the case in 2009.The abundance of PM in the air in July 2001 can be seen in the higher concentrations of many elements in the study by Gatari et al. (2009) compared to the average concentrations at the university site in this study.A clear exception is the concentration of Pb, which was more than 3 times higher in 2001 compared to this study.This demonstrates the effect of banning Pb from the gasoline in 2006.

Table 3 .
PM 2.5 , BC and trace element concentrations from this study (Nairobi) compared with concentration from other cities in Africa.Concentrations are in ng m −3 .Values enclosed in brackets are the standard deviations.