Introduction
Atmospheric particulate matter with dynamic diameter less than 2.5 µm (PM2.5) is a crucial air pollutant due to its adverse effects on human health, atmospheric air, and climate change (Lemieux et al., 2004; Liu et al., 2017). Various sources such as vehicle exhaust, road dust, secondary inorganic aerosols, crustal dust, mineral dust, soil dust, open burning of wastes, residual oil combustion and a host of others have been attributed to particulate matter (Liu et al., 2017). Air pollution could be greatly associated with densely populated Nigerian cities such as Lagos, Abuja and Port-Harcourt (Sonibare and Jimoda, 2009). Also, the environmental laws and regulations are relatively nonexistent for discharging waste indiscriminately into the atmosphere (Thomson, 2001). However, previous studies have shown that transport related pollution in major cities is significant and remains the dominant source of urban pollution with severe health effects (Adebayo et al., 2017).
Several epidemiological studies have revealed that there exist strong relationships between atmospheric particulate matters and serious adverse human health impacts (Baumgartner et al., 2014; Jimoda et al., 2017), such as cardiovascular (Khuzestani et al., 2017), respiratory complications (Beuck et al., 2011; Jimoda et al., 2017; Shang et al., 2018), and reproductive diseases (Lewtas, 2007). Since the pollution caused by PM2.5 poses a greater threat to the people and government of Nigeria, therefore the government needs research-oriented information to control and reduce the sources of the fine particulate matter (PM2.5) (Sulaymon et al., 2018). In order to abate the adverse effects of air pollution caused by the ambient particulate matter, there is need to reduce the mass concentrations of the fine particulate matter. This can be best achieved by identifying the air pollutant’s sources and measure their contribution from the identified sources. Source apportionment of atmospheric particulate matter has been widely performed using several receptor models, such as positive matrix factorization (PMF) (Huang et al., 2015; Khuzestani et al., 2017; Wang et al., 2016), chemical mass balance (CMB) (Cai et al., 2016; Deng et al., 2017; Liu et al., 2017), Unmix (Deng et al., 2017; Jain et al., 2017; Ogundele et al., 2016) and principal component factorial analysis (PCFA) (Jain et al., 2017; Ogundele et al., 2016; Shang et al., 2018).
PMF has been adjudged as an advanced factor analysis and the most widely used source apportionment technique (Khuzestani et al., 2017; Owoade et al., 2016). One of the advantages of PMF is that the knowledge of the source profiles is not required (Huang et al., 2015; Owoade et al., 2016) but rather uses the concentrations and the uncertainties of the chemical species (Hopke, 2016). Several studies have utilized PMF model to identify the sources and their contributions towards the chemical components of the particulate matter (PM) (Khuzestani et al., 2017).
In the present research, two sampling sites in Lugbe area of Abuja were studied. Interestingly, this is the first study examining the chemical components of PM2.5 as well as their source identification and contributions in the area. The sampling was carried out during winter, spring, summer and autumn seasons in the year 2016. The data of PM2.5 chemical species was inputed into PMF model to identify the sources and quantify their contributions during each season. The results of this study will in no doubt provide adequate and useful information to the policy makers in the environmental sector in formulating sustainable control policies and strategies that could help to abate air pollution in Nigeria and other developing nations, thereby improving the health conditions of the populace.
Experimental methods
Study area
Lugbe (9.083º N, 7.533º E) is located in Abuja, the capital city of Nigeria and covers about 50 km2 area of land. It is a residential satellite town with dense population. It is cited along the usually-busy Abuja airport road. As a result of its proximity to the airport and city centre, the area has indeed attracted huge infrastructural developments. This has made it a good representative of a residential, educational, commercial, and traffic area. It is divided into northern, eastern, western, southern and central districts. The Lugbe-section of the airport road is characterized with heavy traffic during the rush hours (morning and evening) every day. The study area is characterized with four different seasons: winter (December – February), spring (March – May), summer (June – August) and autumn (September – November). The winter season has cold air with no precipitation and higher wind speed (Fig of frequency counts of wind speed); spring is known for moderate to high wind speed; summer is dominated with high temperatures and regular rainfall; and autumn has high temperatures and steady heavy rainfall. The details of the air mass flow and the seasonal meteorological conditions of the study area can be found in Fig. S8 and Table 1, respectively.
Ambient sampling
A pair each of quartz filters and Teflon filters were simultaneously loaded into the sampler. Quartz fibre filters (90 mm, 2500QAT-UP, Pall Life Sciences) were pre-baked in a furnace at 500 oC for at least 5 h. This was done to remove organic impurities before sample collection (Baoshuang et al., 2017). Immediately after baking, the filters were kept in silica gel desiccators for more than 72 hrs. Sampling was carried out at a flow rate of 16.7 L/min (Zhang et al., 2008). In total, 150/246 PM2.5 samples were obtained for four seasons (winter, spring, summer and autumn) in 2016. The samples were sealed in aluminum foil covered Petri dishes and stored under dry conditions at -20 oC before both gravimetric and chemical analysis were carried out to avoid evaporation of volatile components (Khuzestani et al., 2017). The collected filter samples and the blanks were gravimetrically analyzed. The details of the analysis have been described in many previous studies (Jimoda et al., 2017; Sulaymon et al., 2018; Wang et al., 2016). Briefly, filters were weighed using an electronic microbalance (Mettler Toledo AX025) with ±0.01 mg sensitivity in a humidity and temperature controlled room (45 % and 22 oC). The gravimetric analysis was carried out before and after sampling. Each filter was weighed in triplicate and the mean value of the readings was obtained for the analysis.
Chemical analysis of samples
Each sample was analyzed for 21 elements (Na, K, Cu, Zn, Pb, Cr, Cd, Ni, Ca, Mg, Al, Ba, Fe, Li, Mn, Sr, Se, V, Sb, Sn and Si), eight water-soluble inorganic ions (Na+, Ca2+, Mg2+, K2+, NH4+, NO3-, Cl-, and SO42-), elemental carbon (EC) and organic carbon (OC). For the elements, a quarter section of each filter was extracted in Teflon bombs using 6 mL nitric acid (Optima grade, Fisher Scientific) and 2 mL hydrogen peroxide (Optima grade, Fisher Scientific) for the microwave-aided digestion procedure. The concentration of the elements was quantified by high resolution inductively coupled plasma-mass spectrometry (ICP–MS; ELAN DRC-e, PerkinElmer). Blank filters were also analyzed using the same procedure above (Shang et al., 2018; Zhang et al., 2008).
Concentrations of water-soluble ions were quantified using ion chromatographic system (ICS-3000, Dionex Inc., USA). A quarter section of each filter was extracted with 5 mL of ultrapure water (18.2 M?·cm, Milli-Q Advantage, Millipore,USA) inside an ultrasonic bath for 25 min at 40 oC (Deng et al., 2018; Shen et al., 2007). The concentrations of the ionic species were corrected using the values of field blanks. Prior to the analysis of the ions, standard solutions were first analysed thrice and low relative standard deviations were obtained. The method detection limits (MDLs) ranged between 0.001-0.03 ?gm-3 and 0.003-0.012 ?gm-3 for anions and cations, respectively.
Organic carbon (OC) and elemental carbon (EC) in PM2.5 were quantified using the Sunset thermal optical carbon analyzer (Sunset Model-4, USA) adopting the National Institute for Occupational Safety and Health (NIOSH) method. The method in details can be found elsewhere (Deng et al., 2018; Schauer et al., 2003). Briefly, a 1.7 cm2 quartz filter punch was put in an oven and heated up to 850 °C in a pure helium atmosphere. As a result of this, all organic compounds were converted into CO2 gas using manganese dioxide (MnO2) as catalyst. The CO2 was quantified using a non-dispersive infrared (NDIR) system. Along the line, some organic compounds got converted to elemental carbon (EC) via pyrolysis process. The oxidized elemental carbon from the filter was quantified following the same procedure used for the organic carbon as the temperature stepped up to 850 oC. The measured concentrations of OC and EC were corrected by subtracting the field blanks from the sample concentrations. MDLs of 0.1 ?gm-3 and 0.04 ?gm-3 were respectively obtained for OC and EC (Khuzestani et al., 2017).
Receptor model
Positive matrix factorization (PMF) model was used for the source apportionment in this study. PMF5.0 is a factor analysis-based model which has been utilized widely for source apportionment studies (Huang et al., 2015; Zhang et al., 2008). The data of PM2.5 mass concentrations, eight water-soluble inorganic ions (Na+, Ca2+, Mg2+, K2+, NH4+, NO3-, Cl-, and SO42-), 21 elements (Na, K, Cu, Zn, Pb, Cr, Cd, Ni, Ca, Mg, Al, Ba, Fe, Li, Mn, Sr, Se, V, Sb, Sn and Si), EC and OC were included in the PMF model. The sources of these chemical species were identified using the results of the PMF model. The species below the MDLs were substituted with half of the MDLs while their uncertainties were calculated by multiplying their corresponding MDLs by a factor of 5/6. The analytical uncertainty of the species was obtained by adding the MDLs to 5 % (for PM2.5 mass, OC and EC) and 10 % (ions and elements) of the measured values (Ito et al., 2004; Khuzestani et al., 2017). Detailed explanation of basic procedures of PMF model can be sourced elsewhere (Hopke, 2016; Xie et al., 2012). In total, 150/246 samples obtained from the two sampling sites during the four seasons were utilized in this PMF model analysis. To obtain an optimum number of factors for the PMF model, a range of 5-8 factors were used iteratively and repetitively to evaluate the effect of the number of factors on the model outputs.
The backward trajectory analysis
The motive of carrying out backward trajectories was to classify those trajectories whose geographic origins were similar (Liu et al., 2014; Liu et al., 2017; Squizzato and Masiol, 2015). The hybrid single-particle Lagrangian integrated trajectory model software (HYSPLIT 4.9 version) was used to calculate the air mass backward trajectories using the Global Data Assimilation System (GDAS) one degree archive. The 120 hr backward trajectories with 1 h interval whose arrival height was chosen as 500 m above ground level (AGL) at the sampling sites were computed using a vertical velocity model and 6 hr interval between each starting time every day (Heo et al., 2013).
Bivariate concentration polar plot analysis
In this study, the plotting of bivariate concentration polar plot for each source factor obtained from PMF model was carried out and used to identify the locations as well as the features of the potential sources impacting the study area. Concentrations of each source are illustrated to vary by wind speed and wind direction. Further details of the analysis can be found in literatures (Carslaw and Beevers, 2013; Grange et al., 2016; Khuzestani et al., 2017). Briefly, the trio of wind speed, wind direction and concentration data are partitioned into wind speed-direction bins. This is followed by the calculation of the mean concentration for each bin using modeling and smoothing techniques. In order to aid interpretation, adjacent bins with mean concentration values were enhanced by using a smoothing technique. As a result of the smoothing technique used, the color illustrations should not be taken as the real values (Carslaw and Beevers, 2013; Grange et al., 2016; Khuzestani et al., 2017).
Results and discussion
The chemical components in PM2.5
The average mass concentrations of PM2.5 and its chemical components for the four seasons are shown in Table 2. The highest concentration (142.1405 ?gm-3) of the ambient PM2.5 was recorded in winter while the lowest (83.7023 ?gm-3) was observed in summer. Also in winter, the PM2.5 mean concentration (145.9255 ?gm-3) at site TK was slightly higher than that at LTH (138.3555 ?gm-3) and this was experienced during the remaining seasons of spring, summer and autumn (Table 2). This can be due to the fact that site TK has many sources around than that of LTH as TK is known to be a commercial centre coupled with very high traffic recorded incessantly and as well serves as a location to many consumption factories. The annual average PM2.5 mass concentrations recorded at the two sampling sites (105.7018 and 102.2325 ?gm-3 for TK and LTH, respectively) were very much higher and more than four times that of United States Environmental Protection Agency (USEPA) whose standard (15 ?gm-3) has been adopted by Nigeria for PM2.5. This is a serious pollution and poses great threat to both lives and environment.
The average concentrations of OC and EC for the two sampling sites during different seasons are enumerated in Table 2. The seasonal average concentrations of OC ranged between 15.6676-25.6741 ?gm-3 and 13.4148-15.6212 ?gm-3 for TK and LTH, respectively while that of EC ranged between 4.5084-9.5626 ?gm-3 and 3.1775-4.1222 for TK and LTH, respectively. The highest mean concentrations of OC (25.6741 mg/m3) and EC (9.45626 ?gm-3) were recorded in winter and lowest (15.6676 and 4.5084 ?gm-3 for OC and EC, respectively) in summer in TK site (Table S1). In LTH site, the highest mean concentrations of OC (15.6212 ?gm-3) and EC (4.1222 ?gm-3) were recorded in winter while the lowest (13.4148 and 3.1775 ?gm-3 for OC and EC, respectively) were observed in spring and autumn, respectively. The average concentration of EC in winter is twice that of summer in TK site. The reasons for this might be due to meteorological factors and regional transport during winter season. Interestingly, during winter, electric energy was basically used for heating as coal-burning for residences was never a practice. Hence, the highest concentrations of OC and EC could be likely attributed to a long-range regional transport. Furthermore, the scatter plots of OC/EC for the winter, spring, summer, and autumn seasons are illustrated in Fig. 5. The observed significant positive correlations between OC and EC during winter (R2 = 0.923), spring (R2 = 0.844), summer (R2 = 0.890), and autumn (R2 = 0.852) seasons show that both OC and EC share a common source (vehicle exhaust). Additionally, the highest concentrations of nitrate (8.2072 ?gm-3) and sulfate (12.2900 ?gm-3) occurred in TK site during summer and winter, respectively while in LTH site, NO3- (3.3601 ?gm-3) and SO42- (9.9455 ?gm-3) were the highest in winter and summer, respectively (Fig. 2 or Table S2). The highest concentration of NH4+ (5.13 ?gm-3) was recorded during spring in TK site while the lowest (2.10 ?gm-3) occurred in LTH site during winter. The analysis of the seasonal variations of the mean concentrations and percentages of mass and chemical components of PM2.5 was also carried out and the results are illustrated in Fig. 3.
The correlation analysis of the average concentrations of the chemical components in PM2.5 between the two sampling sites (TK and LTH) in Ogbomoso was carried out. Statistical parameters such as R2, slope, standard error of the slope, intercept, standard error of the intercept, and F value were obtained using line of best fit in the scatter plots. The analysis of the slopes and intercepts was carried out to examine their statistical significance and to assess the local/regional influence of the chemical components between the two sampling sites (Table S3). If the absolute (1 minus slope) value is found to be greater than two times the standard error of the slope, or the absolute value of intercept is found to be greater than two times the standard error of the intercept, then the influence is classified to be local, otherwise it is regional (McGinnis et al., 2014). The statistical analysis of the slopes and intercepts of the chemical species are illustrated in Table S3. As shown in Table S3, there are indications that differences among the chemical species between the two sites exist as the slopes and intercepts of the majority of the chemical species were found to be statistically significant. This indicates that both local and regional influences are established among all the major chemical components. In all the chemical species, the slopes and intercepts of Na+, Mn, Ba and Se were found to be statistically non-significant, thereby having regional influence over the sampling area.
Evaluation of PMF results
Considering the PMF results with factors 5–8 (Fig. S3-S5), a 7-factor model was found to be physically reasonable for the whole campaign. The scatter plot of the linear correlations between the predicted and measured PM2.5 mass concentrations in the two sampling sites are shown in Fig. 3. The G-space plots between the factors were found to have weak relationship with negligible regression coefficients (R2). This shows that all factors were accurately resolved by PMF and independent of one another (Fig. S6). The seven factors identified by PMF (Fig. S5) include industrial source, secondary aerosol, vehicle exhaust, mineral dust, crustal dust, residual oil combustion source and road dust.
The average mass concentration and percentage contributed by the sources to the total PM2.5 mass is provided in Table B, Fig. C (Supplementary, from PMF model). According to Fig. S7, industrial source was identified with high loading of cadmium (Wang et al., 2016; Wei et al., 2014; Yao et al., 2016; Yu et al., 2013), which contributes 86.91 % mass. Moderate loadings of Ba (19.40 %) and Se (16.71 %) were also noted for this factor profile. Industrial source contributed 2.56 ?gm-3 (2.90 %) to PM2.5. The secondary inorganic aerosols source was identified by the higher loadings of nitrate (NO3- -79.32 %), ammonium (NH4+-53.95 %) and sulfate (SO42- -37.77 % mass) (Sharma et al., 2014; Wei et al., 2014). This result is in line with that of Alleman et al. (2010). These chemical components are usually derived when NO3 and SO2 gasses are oxidized as well as during the neutralization process of NH3 gas (Wang et al., 2016). The secondary inorganic aerosol source contributed 12.85 ?gm-3 (14.7 %) to PM2.5.
The vehicle exhaust source is characterized by higher loadings of EC (70.31 %), OC (59.19 %), Zn (40.36 %) and Cu (31.71 %) (Wang et al., 2016; Yu et al., 2013). EC, which is a major emission from combustion sources has been widely identified as a tracer for diesel exhaust (Yin et al., 2010). Chemical species such as EC, OC, Zn, Cu, Ba, Pb and Mn have been identified as tracers for vehicle exhaust (Begum et al., 2011; Lough et al., 2005) and some of these (Cu, Zn, Ba, Mn and Fe) are tracers of brake-wear which are also used as markers of traffic re-suspension (Querol et al., 2008). Also, high loading of Zn in PM2.5 has been used as a tracer of vehicle exhaust (Brown et al., 2007). While vehicle exhaust has been identified as a major source of PM2.5 and about 10-80 % is contributed to PM2.5 by vehicle exhaust in Indian cities (Sharma et al., 2016), it is also identified as one of the main sources of OC (Zhang et al., 2013). Vehicle exhaust has contributed 21.87 ?gm-3 (25 %) to PM2.5 in this present study.
The mineral dust source was dominated by higher loading of water soluble re-suspended dust element (Mg) (Beuck et al., 2011; Khuzestani et al., 2017), which contributed 77.33 %. Gupta et al. (2007) also used OC as a tracer of mineral dust and OC has a moderate contribution of 32.36 % to PM2.5 in this study. In total, mineral dust contributed 23.65 ?gm-3 (27 %) to PM2.5 mass. The crustal dust source (Fig. 2) was dominated by higher shares of crustal elements including Ca (71.62 %), Br (71.51 %), Al (71.49 %), Si (62.01 %), Ca2+ (58.44 %), K (44.72 %) and Fe (44.31 %). In addition, some high to moderate loadings of Ti (30.59 %), Pb (29.46 %), Mg (24.77 %), Mn (16.65 %), Cr (12.36 %) and Ni (9.25 %) were also noted for this factor profile (Khuzestani et al., 2017; Wang et al., 2016). Elements such as Al, Si, Ca and Fe have been used as tracers for crustal dust (Alleman et al, 2010). Also, higher loadings of Br, Ti, Mn and Mg were used to identify crustal dust in a study by Lim et al. (2010). The crustal dust source accounted for 21.47 ?gm-3 (24.5 %) to PM2.5 mass.
The residual oil combustion source was identified by higher loadings of nickel (59.14 %) and vanadium (51.36 %). Several studies have used Ni and V as tracers for the residual oil combustion source (de Foy et al., 2012a; Wang et al., 2016). Additionally, high loadings of elements such as Mn (58.37 %), Fe (44.02 %) and Mg (42.07 %) were also noted for this source profile. The residual oil combustion source only accounted for 0.1 % of PM2.5 mass, therefore it is categorized to be a trace source. The road dust source was controlled by the higher amounts of Sn (73.36 %), Li (71.46 %) and Ba (62.80 %). A source with higher loadings of Ba can be identified as road dust with its origin being traced to mobile transportation on untarred roads which makes the precipitated soil particles to be resuspended (Lim et al., 2010; Sowlat et al., 2013). Due to the existence of several unpaved roads around the sampling sites and in Lugbe as a whole, the reason mentioned above can be justified. Road dust factor contributed 5.05 ?gm-3 and accounted for 5.8 % of PM2.5 mass.
Seasonal contributions of the sources
The source contributions for winter, spring, summer and autumn seasons are provided in Fig. 6. The average contributions of the industrial source are 1.124, 1.155, 1.642 and 0.120 ?gm-3 for winter, spring, summer and autumn, respectively. The secondary inorganic aerosol (SIA) source contributed largely during spring (1.780 ?gm-3) while the lowest (0.458 ?gm-3) was observed during winter. The contribution of SIA during summer and autumn are 1.036 and 0.922 ?gm-3, respectively. Vehicle exhaust contribution was observed to be very low during summer (0.277 ?gm-3) and highest (1.598 ?gm-3) during autumn (Fig. 6).
The mean mineral dust contribution was significantly low during autumn compared to winter (Fig. 6). The mean contributions of mineral dust are 1.297, 0.990, 1.084 and 0.631 ?gm-3 for winter, spring, summer and autumn, respectively. Crustal dust also follows the trend of mineral dust but its summer was lower compared to that of mineral dust. Across the four seasons, crustal dust contributed 1.761, 0.907, 0.684 and 0.630 ?gm-3, respectively. The boxplot showing the road dust follows the same trend with mineral dust in which the highest (1.650 ?gm-3) was recorded during winter and the lowest (0.636 ?gm-3) in autumn. The road dust contribution was fluctuating between spring and autumn (Fig. 6). The reason for the higher contributions of mineral dust, crustal dust and road dust sources during winter season may be due to the fact that massive road constructions are usually carried out between January and March in Nigeria while construction activities are passive in April and October due to heavy rainfall (Ref). In addition, Lugbe is an area that hosts several construction companies and many roads were under construction during the sampling periods. The mean contributions of the residual oil combustion as a source were 0.947, 0.840, 0.487 and 1.687 ?gm-3 for the winter, spring, summer and autumn, respectively.
When an OC/EC ratio is above the range of 2.0-2.2, it is an indication of the presence of secondary organic aerosols in the atmosphere of that area (Liu et al., 2017). In our study, the annual mean OC/EC ratio was 1.33, a value below the range mentioned above. This is in contrary to the result of a study conducted in Haikou, China where the annual mean OC/EC ratio was as high as 2.6 (Liu et al., 2017). Furthermore, the highest OC/EC ratio was recorded at TK site in spring (2.25), while the lowest was observed at LTH site in winter (0.99) (Fig. S9).
Backward trajectories analysis
The trajectories were plotted to trace the origin and transport pathways of air masses during the sampling periods in the area (Fig. S10). During the winter period, four clusters were obtained to represent different wind transport directions. Clusters #1 (23 %) and #2 (43 %) represent the flows coming from northeast (NE) and dominated the transport directions while clusters #3 (22 %) and #4 (12 %) were coming from the northern part of the sampling area. In spring, clusters #1 (36 %) and #2 (53 %) which dominate the transport directions (89 % in total) originated from southwestern (SW) part of Lugbe. The remaining 11 % was shared between clusters #3 (10 %) and #4 (1 %) which emanated from the western and northern parts, respectively. In addition, majority of the air mass (97 %) during the summer period was transported from the southwestern (SW) direction to the receptor site while the remaining 4 % was traced to the western part of the area. The contributions of clusters #1, #2 and #3 were 53, 35 and 9 %, respectively. During the period of autumn, four clusters were also found to represent different wind transport directions. Clusters #2 (55 %) and #3 (17 %) represent the flows transporting from southwest (SW) and dominated the transport directions while clusters #1 (20 %) and #4 (8 %) were approaching the sampling area from the eastern and northeastern (NE) part, respectively.
Local/regional transport analysis
The Lugbe area shows the regional long-range transportation of crustal dust mainly by south easterly wind directions and moderate wind speeds were controlling their higher concentrations. With northeast and southwest directions, low concentration (8 ms-1) while moderate to high concentration (2.5? PM2.5?3.5 ?gm-3) at low wind speed (ws8 ms-1) in both southwest and northeast directions. Also, there is moderate-high concentration recorded at high wind speed (ws;10 ms-1) in northeast and high concentration (1? PM2.5?1.2 ?gm-3) at low wind speed (ws10 ms-1), high concentration (;1.6 ?gm-3) of PM2.5 was recorded. The industrial factor displays potential locations mostly from north-eastern parts of Lugbe. In addition, a relatively local-range transport with high concentrations was observed with the source near the centre of the study area.
The bivariate concentration polar plot of mineral dust factor shows that the source is largely controlled by sources found in the central and south-eastern parts of the study area at moderate wind speed. Low-medium concentration in northeast direction was traced to high wind speed (ws;10 ms-1) while high concentration (PM2.5?1.3 ?gm-3) at low-medium wind speed in southeast and at high wind speed in southwest directions. Also, a relatively long-range transport was observed in areas closer to the western part of the study area and controlling high concentrations at higher wind speed. The bivariate concentration polar plot of road dust factor shows that the source was greatly influenced by regional medium-range transportation of external sources from north eastern direction at a moderate-high wind speed, thereby controlling the higher concentrations of road dust factor in the study area. There was low-medium concentration of PM2.5 at low and high wind speed across different directions while high concentration (PM2.5?2.0 ?gm-3) was related to low wind speed in northeast direction. The secondary inorganic factor, according to the bivariate concentration polar plot (Fig. S11), was found to exhibit local patterns with the sources majorly concentrated in the central parts of the study area with their higher concentrations being controlled by the variable wind speeds. In addition, in the northeast direction, low concentration (PM2.5?0.5 ?gm-3) was traced to high wind speed (ws;10 ms-1), high concentration (PM2.5?1.0 ?gm-3) at low wind speed (ws?6 ms-1) in northwest and high concentration (0.9? PM2.5?1.0 ?gm-3) at low-high wind speed in western and southwest directions.
