PM 10 Source Identification: A Case of a Coastal City in Colombia

This paper assesses the spatial variation of EC, OC, major, and trace elements in an industrialized coastal city, allowing identification and tracers of PM 10 emission sources. 83 samples (24 h average) were collected on quartz filters during the dry season using high-volume samplers. Major and trace elements were analyzed using ICP-AES and ICP-MS, whereas a thermal/optical carbon analyzer was used to determine OC and EC. Chemical characterization of major elements, SiO 2 , SO 42– , MgO


INTRODUCTION
Understanding atmospheric particles and chemical characterization is a continuing concern within public health, due to their possible adverse effects (Englert, 2004;Pey et al., 2010a). Studies of particulate matter (PM) with aerodynamic diameter < 10 µm (PM10) have become a major concern in developing countries related due to their association with morbidity and mortality, reduction in lung function, and exacerbation of airway illnesses (Agudelo-Castañeda et al., 2019;Hsu et al., 2016;Rodríguez-Villamizar et al., 2019). For this purpose, positive matrix factorization (PMF) has become the most extensively used pollution source resolution method in research around the world Jain et al., 2020;Pant and Harrison, 2012;Sharma et al., 2016Sharma et al., , 2014Shivani et al., 2019). While EPA PMF (V5.0) incorporates considerably enhanced error estimation methods, which try to estimate the errors happening to the inherent uncertainties in the data and the rotational ambiguity in the result (Hopke, 2016;Paatero et al., 2014). Moreover, the chemical composition of PM10, especially trace metal content, has been widely used to infer PM10 sources (Li et al., 2021;Sharma et al., 2014).
Although the importance of analyzing the global, regional, and local influences in this region, studies in Caribbean urban coastal zones are scarce (Agudelo-Castañeda et al., 2020;Aldabe et al., 2011;Arruti et al., 2011;Kalaiarasan et al., 2016;Li et al., 2021;Moreno et al., 2010a;Police et al., 2016) Because of their location and prevailing wind direction, the study area may receive a high influence from regional air masses, industrial sources, and traffic, causing high loading of trace elements. Even if these elements frequently account for a minor proportion of PM, they have substantial environmental and human health impacts (González-Castanedo et al., 2014;Hao et al., 2018;Minguillón et al., 2014;Ripoll et al., 2015). For instance, cadmium, cobalt, arsenic, chromium, nickel, and lead are considered human carcinogens (Liu et al., 2018). Regardless of the importance of public health and risk, there is no baseline information concerning atmospheric pollutant emissions, dynamics, and principal sources of trace elements in the study area. Therefore, this study makes a major contribution to research on chemical element sources in PM10 in a highly industrialized area. The first section (i) of this paper assesses the spatial variation of OC, EC, EC, major, and trace elements of PM10 in an industrialized Caribbean city. The next section (ii) focuses on the evaluation of PM10 emission sources focusing on the Enrichment Factor and chemical characterization. The last section (iii) discusses PMF profiles and air masses trajectories to deduce the causes of PM peak pollution. Therefore, this work will provide one of the first investigations through spatially resolved data of trace elements associated with PM10 that may be used for public health studies and fill the nonexistence of information in the region.

Study Area
The study area is located on the western side of the river delta of the Magdalena River, and nearby the Caribbean Sea. Barranquilla is the central economic and industrial hub of the Colombian Caribbean Region with 1,193,952 inhabitants (Barranquilla, 2018). The average annual rainfall is 750.1 mm in 55 days with dry and wet seasons, where the principal dry season is December-April. Bimodal precipitation consists of two periods in May-June and August-November. In July, a reduction of precipitations is registered, although less pronounced than in January. The annual average relative humidity is 83% with 1652 mm of evapotranspiration. The dry season (December-March) is characterized by its lower relative humidity of 78%. In these dry months, the sun on average shines for 8-9 hours. The month with the lowest solar brightness is October with 6 hours day -1 . During the sampling period (March 17-April 4, 2016), weather conditions were like climatological normals, with small deviations. Ambient temperature, relative humidity, wind speed, and direction data were taken from two Mark Davis Vantage PRO 2 stations located in two sampling points: Pies Descalzos and Cotediba. The prevailing wind was NE and NW for Pies Descalzos and Cotediba, respectively. Following the Köppen climate classification, the climate in the study area is tropical wet and dry savanna Aw, characterized by an average annual temperature of 28°C, with an average annual minimum of 25.1°C and a maximum of 31.4°C (IDEAM, 2016). The prevailing wind was from the northeast (31.4%) all year, with the second predominance of winds from the north (29.9%). The annual occurrence of calm winds (< 0.5 m s -1 ) was 9.7% with an annual average wind speed of 10-11 m s -1 (IDEAM, 2016). Wind speed and relative humidity ranged between 5.7-8.8 m s -1 and 77-87%, respectively, during the sampling period. Moreover, relative humidity decreased in the last week of March. The average ambient temperature was 28.6°C and 27°C for Pies Descalzos and Cotediba, respectively. Some industries use diesel fuel, while most use natural gas, supplementary information is included in SI1. The location of the sampling sites in the study may be observed in Table 1 and Fig. 1.

Sampling and Chemical Analysis
Simultaneous measurements of PM10 (airborne particle matter < 10 µm) concentrations at seven sampling sites representing regional, urban, and industrial environments were carried out. Sampling was conducted by the Environmental Authority of Barranquilla and filters were kindly provided as part of a joint project.
Airborne particle matter < 10 µm (PM10) was measured using high volume samplers (1.1-1.7 m 3 min -1 ) Model TE 6070-MFC PM10, TISCH environmental brand (EPA approved Manual Filters were subjected to different treatments in acidic digestion (HNO3:HF: HClO4) of ½ of each filter; and an analysis of organic carbon (OC) and elemental carbon (EC) in 1.5 cm 2 sections. The extracted solution from the acidic digestion was analyzed by Inductively Coupled Atomic Emission Spectrometry, ICP-AES, (IRIS Advantage TJA Solutions, THERMO) for the determination of the major elements, and Inductively Coupled Plasma Mass Spectrometry, ICP-MS, (X Series II, THERMO) for the trace elements. OC and EC were determined by a thermal/optical carbon analyzer (SUNSET), using protocol EUSAAR_2. Details of the procedure and chemical analysis are provided in the supplementary materials (SI1, SI2, SI4, SI5). Other analytical details may be found in Querol et al. (2009).

Positive Matrix Factorization (PMF)
Positive matrix factorization (PMF) is a multivariate factor analysis tool that partition the data into factor contributions, profiles, and a residual matrix Paatero, 1997;Paatero et al., 2014). The fundamental principle of the model is mass conservation may be assumed and thus, a mass balance analysis can be used to detect and apportion sources of atmospheric particles (Hopke, 2016). We used the EPA PMF v5.0 (U.S. EPA, 2018). PMF methodology is explained, briefly in supplementary materials (SI1) and another published study (Agudelo-Castañeda and Teixeira, 2014;Teixeira et al., 2015). The determination of factor profiles in PMF is an important process and depends on the goodness of fit to the original data (Agudelo-Castañeda and Teixeira, 2014). Consequently, six factors were used because they can be explained by known source patterns and physically realistic results. Major and trace elements were chosen based on the signal-tonoise ratio, the percentage of values above the MDL, and the database size requirements. Just 16 species were selected as strong, 29 were classified as bad and 11 were classified as weak. Weak species had their uncertainty magnified by a factor of 3, and bad species were eliminated. The model was run numerous times with a differing factors number (3-8) using random seed mode. Q theoretical value was very similar to the obtained Q in the model, thus representing an applicable uncertainty in the input. Scaled residuals values were in the recommended range (-3 to +3) (Comero et al., 2009). Details of PMF settings, regression diagnostics, and Q values results are provided as supplementary materials (SI1 , Table S1, S2, and S3).

Enrichment Factor (EF) Analysis
The enrichment factor (EF) is widely applied to identify the anthropogenic source of chemical elements (Hans Wedepohl, 1995). Calculation of the enrichment factor of each element detected in the study area was done using Ti as the reference element and composition of the upper continental (Fomba et al., 2013;McLennan, 2001). Eq. (1) was applied for the calculation of the enrichment factor of each sampling site in the study area: where EF is the enrichment factor of element x that represents the chemical element of interest.
Cx is the concentration of each element in the PM10 sample, where CR is the concentration of the reference element in the upper continental crustal. (Cx/CR)PM is the concentration ratio of element x to R element in the PM10 sample and (Cx/CR)Crust is the concentration ratio of X to R element in the crustal material. Table 2 summarizes the average concentration of major elements, PM10, OC, and EC (µg m -3 ). Results evidenced high PM10 levels (mean 51.7 ± 10.5 µg m -3 ), especially in P3 up to P7, the latter nearly 3-fold the WHO annual PM10 guideline. Spatial variation (see Table 1 and Fig. 1 for the location of sampling sites) results indicated that P1 (Portales de Sevilla, a residential area) had the lowest levels for almost all major elements, except for P2O5. Probably, because of their location, upstream of primary emission sources (the predominant wind direction was NE for the meteorological station in P2). Additionally, the highest PM10 concentrations were observed at the southernmost sampling sites: P6 (residential area) and P7 (residential area). Sampling site P3 is characterized by near quarries and exposed land, thus, the slightly higher PM10 concentrations compared with the nearby sites. This pattern was observed for crustal elements, too that presented contributions of Al and Mg elements ( Table 2). The sampling period corresponded to the dry season, where resuspension increases due to their dry climate. The soil of the study area is characterized by layers of sandy, marly, and loam limestones (limestone -clay -caliche), and coralline rocks, composed of Ca, Al, and Si (IGAC, 2017). Results showed, too, a high concentration for SiO2, MgO, and CaO, though with a slight spatial variation between sites. MgO was constant between sites, indicating the presence of limestone (dolomite). Moreover, SiO2 and CaO levels were lower in P1 and P2. Probably, as explained above, because P1 and P2 location is upstream of main emission sources and may be considered "background" sites. MgO, Al2O3, and SiO2 showed the highest concentrations in P3 due to the near quarries.

PM10 and Major Chemical Species
Cement industries and construction originated CaO high levels, i.e., P6 presented the highest concentrations for CaO, probably of some road construction works done nearby for a stream canalization.
The abundance of all these major elements confirms the effect of the resuspension of exposed land and road dust. The city, too, uses concrete pavement in mostly all the streets. This issue and the problem of inadequate rainwater sewerage system cause the deterioration of roads and possible production of dust that may be resuspended.
Furthermore, P4 (common industries and cemeteries) presented the highest concentrations for SO4, OC, and EC. This site is downstream of diverse industries (the prevailing wind direction for P5 was NW), especially several brickkilns and hazardous waste incinerators. Brickkilns may use combustible sulfur which is a precursor of sulfated aerosols, thus increasing SO4 concentrations and organic compounds.
A comparison of the findings with those of other studies confirms that the relatively high levels of K might indicate that biomass burning impact on PM10 levels is high (Rolong et al., 2021). Though, the high correlation K/Al (R 2 = 0.72) suggests that K is present in mineral matter. Also, results showed significant correlations between OC and K (R 2 = 0.64). The sum of mineral matter-related oxides accounts for the largest proportion of the PM10 mass (30-32% P1-P2 and 34-41% P3-P7). Fig. 2 presents the correlation between OC and EC in all sampling sites (a) and by site (b). Results showed significant correlations between OC and EC (R 2 = 0.77), although mostly all correlations were > 0.98 (Fig. 2(b)), except for the sampling site P3. These correlations and proportional relative rates of OC and EC emission to each other may indicate the presence of common dominant sources (Kong et al., 2010). Even if the sampling sites were highly impacted by vehicles and industrial sources, levels of EC were low (1 order of magnitude lower than most European and Indian cities, Mahapatra et al., 2021;Reche et al., 2011), and as expected much lower than OC, with OC/EC ratios close to 20 for P1 and P2 ( Fig. 2(b)) and 11-15 for P3-P7, possibly pointing to high production of secondary organic aerosol (SOA). P1 and P2 located the nearest to Isla Salamanca Natural Park on the northeastern side of the Magdalena River ( Fig. 1) showed similar OC/EC ratios with high correlations values (Fig. 2(b)). This natural Park suffers along in the dry season from several fires caused by the high temperatures or by burning the mangrove wood for coal production, cattle raising, or farming. The Caribbean region is particularly vulnerable to fire occurrence due to large seasonal water deficits (Hoyos et al., 2017). Thus, a high impact of biomass combustion and/or the influence of the dry climate, with high temperatures and relative humidity, the presence of hydric resources, and mangroves that favor SOA production, may account for these high OC/EC ratios. In the study area, roads are constructed with concrete, which requires higher energy use compared with an asphalt surface, which marks in higher fuel consumption hydrocarbon concentrations, resulting probably in high OC/EC values (Adamiec et al., 2016). On the other hand, P3 showed the lowest correlation value (R 2 = 0.44). This site is an industrial site impacted by hazardous waste incinerators and brick kilns.   Fig. 3 reveals highly enriched EF values for Pb in P1, P2, and P7 sites. The two latter sites are characterized by irregular Pb smelters for artisanal recovery of this metal in old batteries, which were in operation at the time of sampling. Moreover, P1 and P7 high EF values are probably due to their location near the harbors, thus receiving a high influence on shipping emissions. Consequently, high EF values in the city probably are due to Pb in emissions of irregular smelters and harbors. Moderately enriched EF values of P, principally in P1, are probably due to a fertilizer industry located near the sampling site. Moreover, fossil fuel combustion is estimated to be a major source of atmospheric-P in industrialized regions (Weinberger et al., 2016) besides the particle phase phosphorus emitted from biogenic processes (Rizzo et al., 2010). Table 3 shows the average concentration of trace elements for all sampling sites. As explained above, high EF values of Cu, Pb, and Ni were obtained. Moreover, high concentrations of Cu, Pb, and V were measured. Vanadium is a tracer of the influence of oil-burning (all sites). The maximum Cu value obtained for all samples was 496.2 ng m -3 . Cu mean values reached near 117.5 and 323.1 ng m -3 at P2 and P5, respectively, pointing to the influence of industrial emissions at these sites, such as smelter processes, besides road traffic (González-Castanedo et al., 2014). These results may confirm the high EF values for these two sites (P2 and P5).

Tracer Species for Source Identification
Pb reached sporadically near 340 ng m -3 values, without major correlation with other trace elements out of As and Sn. Probably pointing to infrequent emissions of a high-temperature industrial process, principally in P1-P2 sites due to their proximity to the industries area near the harbors. As explained above, these two sites presented high EFs values, whereas P1 showed   (Table 2 and Table 3), and P2 for the same elements plus Cu. These sporadic Pb and Cu concentrations associated with PM10 may indicate the influence of industrial processes and shipping emissions (Nunes et al., 2017;Russo et al., 2018;Viana et al., 2014b). Ni had similar concentrations in all sites, confirming the influence of vehicular traffic, except for P3. These results can be attributed to industry emissions such as cement and brick production (Taghvaee et al., 2018). Table 4 shows our results compared to previous studies. Fe concentration is less than that reported for the other cities, as well as the average concentration of Cr, which is lower than that reported for megacities in Latin America such as Mexico City and Bogotá. While Cr, Co, and Cd are close to those reported for other coastal cities in the Caribbean such as Belén, Costa Rica, and Cienfuegos, Cuba. Ni concentration is higher than that reported in other coastal cities, such as Navi, India, and Patras, Greece, and close to the reported concentration for Barcelona, Spain. Pb concentration is higher than that reported for all the cities in Table 4, except for Mexico and Barcelona.
The direct identification of anthropogenic emissions employing just these tracers is difficult because they are markers for many types of combustion processes occurring at the same time, such as industrial processes, traffic, petroleum refinery, and energy generation. Consequently, to better understand these values, some tracer species ratios were calculated for each sampling site (These results are like related studies in other cities -Tables 4 and 5-). Values confirmed the presence of different anthropogenic sources (Table 5). Cu/Pb ratio indicated traffic as a local source in P6 and P1 (Pey et al., 2010a). Moreover, Cr/Pb ratio confirmed traffic as a source in P6 and P7 (Font et al., 2015), while the Sb/Cu ratio fell in the range of traffic for P2, P5, and P6 (Pey et al., 2010a). Sn/Cu ratio, confirmed too, the traffic influence in P2 and P5 sites, where values fall in the range for wear-abrasive sources (Lin et al., 2015). The influence of shipping emissions was confirmed in all sampling sites as combustion processes that use crude oil as the main fuel, where vanadium (V) and nickel (Ni) are commonly identified as markers of shipping emissions (Viana et al., 2014a). The study area is downwind of the prevailing wind direction, thus receiving shipping emissions, confirming the V/Ni and La/Ce ratios values for this type of local source originated in the North and Northeastern side of the city. These two latter ratios correlated very well with shipping emissions, as much of the combustion from ships derives V and Ni associated principally with fine atmospheric particles, and therefore capable of traveling long distances and staying resuspended (Viana et al., 2014a). Moreover, the trace element ratio V/Co can be assigned to oil combustion according to literature (Ledoux et al., 2017) for P1, P2, and P3 sites. Furthermore, the Mn/V ratio could be useful for distinguishing between particles originating from oil burning or coal-burning; while from oil burning this ratio is << 1, from coal-burning, is >> 1 (Kong et al., 2011). Ratios ranged from 1.36 to 5.36 which meant that PM10 was more relative to heavy fuel oil combustion than coal burning in all sampling sites, probably originating from shipping emissions   as stated above. Also, these sites may be impacted by possible recoveries of industrial oil. Ratio Cu/Pb and Cr/Pb confirmed the presence of municipal waste incinerators in P3 (Font et al., 2015), where these ratios may be used as tracers. Slightly high Cu/Cd ratios in the study area confirmed the presence of additional emission sources of Cu from some industries, although, at some sampling points, these ratios were lower indicating the presence of close Cd sources, whereas Cu had also been recognized as markers for non-ferrous metal smelters (Kong et al., 2011).

PMF Profiles
Profiles obtained from the PMF model identified six sources for PM10. Results may be observed in Fig. 4. The identified sources were enriched road dust resuspension, biomass burning, trafficrelated emissions, heavy fuel oil combustion, industrial/smelter processes, and brick.
Factor 1 was defined as enriched road dust resuspension with soil mineral dust because of their high loadings of crustal elements enriched in OC, EC, V, and Cd, among others. Possible sources may include emissions from near quarries located within the study area, besides resuspension of road dust (Amato et al., 2009).
Factor 2 presented P2O5, MgO, and K2O with dominant concentrations. K is used as a key element for biomass burning or wood combustion (Sharma et al., 2014). As explained before, a significant correlation between OC and K might give evidence of biomass burning as a source of PM10 during the sampling period, which probably originated in the Isla Salamanca Natural Park on the northeastern side of the Magdalena River (Rolong et al., 2021).
Factor 3 accounts for crustal elements and Zr with dominant concentrations. Zr was identified as a tracer of the ceramic industry, clay or clinker loading/transport (Minguillón et al., 2013;Pey et al., 2009;Sánchez de la Campa et al., 2010). Several brickfields/ceramic industries that use coal or wood as main fuels have grown on the southwest side of the city near the P3 site. Consequently, this profile was defined as bricks/ceramics (Begum et al., 2011).
Factor 4 had high loadings of V, Cd, OC, EC, and SO4 2-. An increasing number of studies are showing that these elements are typical of heavy fuel oil combustion, probably originating from shipping emissions (Moreno et al., 2010b;Pey et al., 2010b;Viana et al., 2014b), as explained above.
Factor 5 was identified as industrial and smelter processes because presented high loadings of Cu. As explained above, Cu concentrations were high in some industrial sites, thus confirming this result. Several studies state Cu is a tracer of non-ferrous smelters (Pan et al., 2015) or even Cu-smelter processes (González-Castanedo et al., 2014).
Factor 6 presented high values for Ni, Cr, Cd, EC, and SO4 2-. These metals have been observed in other studies in the coarse fraction of atmospheric particle matter in high-traffic urban areas (Fomba et al., 2018). Road dust resuspension generated by abrasion of vehicle parts and pavements may originate from these elements (Salvador et al., 2012). Consequently, this profile was designed as traffic-related. As explained previously, the economic growth of the study area increased vehicle traffic and number.

Back Trajectory Analysis
HYSPLIT trajectories may be used to determine possible source regions contributing to air pollution in the study area or to determine air masses that may be affecting (Draxler et al., 2020;Rolph et al., 2017). All masses trajectories are in the supplementary information (SI3). PM10 may travel long distances and stay suspended in air (Sharma et al., 2014), thus 7 days were selected to calculate backward trajectories using HYSPLIT to understand the flow of pollutants from distant source regions. AGL heights of 500 m, 1000 m, and 1500 m were used to reduce the effects of surface friction (500 m) and to understand the behavior of the boundary layer. May be observed three different air masses trajectories during the sampling period (Fig. 5): one coming from the Northeast Atlantic Ocean and entering the northern region of Colombia, another one coming from east South America and crossing Venezuela, and the last one coming from the Caribbean Ocean and going through Venezuela. Recent studies demonstrate the possible influence of biomass burning in the Orinoquia region, thus increasing PM10 (Ballesteros-González et al., 2020).
This study has been of great importance for the environmental authority, since it has allowed us to recognize which are the main emitting sources of these trace metals, strengthening the management of air quality through the development of surveillance and control strategies, such as preparing the air quality management plan that contains programs and projects to reduce atmospheric pollution and trace metal level in Barranquilla, as well as the control of industries that impact air quality in the city. Our study suggests that to further reduce trace metal levels, we must encourage sustainable transport.

CONCLUSIONS
This paper assesses the spatial variation of chemical characterization of PM10 in an industrialized Caribbean city and evaluation of emission sources using PMF, EF, and diagnostic ratios. As it is the first research in the study area, obtained results explained the origin of diverse trace elements associated with PM10.
The results of this research support the idea of the contribution of resuspension of exposed land and road dust contribution to the study area.
Sites near brickkilns and hazardous waste incinerators contributed as precursors of sulfated aerosols.
Good correlations were found between OC and EC. Even if the sampling sites are highly impacted by vehicles and industrial sources, levels of EC were low, and as expected much lower than OC, possibly pointing to high production of secondary organic aerosol (SOA).
The impact of biomass burning was demonstrated, too.
High EF values for S were obtained for all sampling sites, indicating possible vehicular traffic emissions. Closer inspection of both Ni and Cu EF, and high EF values for S show similar values for all sites (except for P2 and P5 for Cu EF values), thus, confirming the vehicular traffic influence. Diagnostic ratios confirmed these results, which is the influence of traffic in the study area.
The influence of shipping emissions was confirmed in all sampling sites as combustion processes that use crude oil as the main fuel, where vanadium (V) and nickel (Ni) are commonly identified as markers; and high EF Pb values in P1 and P7 probably due to their location near the harbors. Also, the markers V and Ni pointed out the influence of heavy fuel oil combustion by the shipping emissions, too.
Cu mean values at P2 and P5 showed the influence of industrial emissions, such as smelter processes. PMF analysis confirmed some identified sources by ratios and EF values: enriched road dust resuspension, biomass burning, traffic-related emissions, heavy fuel oil combustion, industrial/ smelter processes, and brick.
These results could improve emission inventories and air quality management, overcoming the existing lack of information. The presented results show important implications regarding the scientific and environmental panorama.