Source Apportionment of Inorganic and Organic PM in the Ambient Air around a Cement Plant : Assessment of Complementary Tools

In this study, we analyzed the sources of ambient PM inorganic and organic components near a cement plant using fossil fuels as well as alternative fuels, such as sewage sludge and biomass. Source apportionment methodologies, i.e., principal component analysis (PCA) and multivariate curve resolution by alternating least squares (MCR-ALS), and carbon isotope analysis (δC) were used to determine the potential sources and their contributions. Four sources of PM10 main tracer compounds constituents were identified: marine and secondary inorganic aerosol, cement plant/industrial, traffic and crustal. The contributions of those sources varied significantly depending on the period of the year. However, the inorganic tracer PM species in the area were mainly released by combustion sources, namely traffic and the activity of the cement plant, especially in winter months. The analyses of tracer organic compounds also indicated combustion sources, i.e., biomass burning and fossil fuel combustion, as the predominant contributors to ambient air PM (62, 59 and 69%, in PM10, PM2.5 and PM1, respectively). Organic dust was a significant source of PM10 (33%) while its contribution was found to be minor in the finest fractions (9 and 2% in PM2.5 and PM1, respectively). Results of δC corroborated a significant contribution of combustion sources, traffic or biomass fuel as well as a higher influence of mineral (calcite) powder in larger particles.


INTRODUCTION
Cement production is associated with the emission of a wide range of pollutants including gases (i.e., NO x , SO x , HF, HCl) and particle-bound pollutants such as metals (i.e., As, Cd, Pb, etc.) or organic compounds, (i.e., polycyclic aromatic hydrocarbons (PAHs), etc.).The emissions of particulate matter (PM) generates most of the noticeable impact and concern in the surroundings of cement plants and takes place at different points along the cement manufacturing process (quarrying of raw materials, packing of the product, combustion process, vehicle movements, etc.) (Marcon et al., 2014).PM source apportionment of a cement plant becomes complex when this installation is located in the vicinity of other industrial and urban areas that have additional sources, such as traffic and power generation (Karagulian et al., 2015).PM is associated with many health effects that include asthma, pulmonary injury, cardiorespiratory diseases and recently it was nominated as Group I carcinogen by the International Agency for Research on Cancer (IARC) (Hamra et al., 2014).It is known that the size of particles plays a key role on determining their potential of being harmful.Due to the evidence of those adverse effects, in Europe the ambient air quality Directive (2008/50/EC) sets limits for particulate matter with aerodynamic diameter smaller than 10 and 2.5 µm (known as PM 10 and PM 2.5 respectively).Levels for smaller particles are not legislated yet, although it is known that the smaller particles, especially ultrafine particles (those with a diameter smaller than 100 nm), are the most hazardous ones since they may penetrate deeper into the lungs and eventually into the blood stream (Kelly and Fussell, 2012).Notwithstanding, PM-induced adverse health effects are linked to their chemical composition, particularly to trace metals and toxic organic compounds.However, data on PM composition is still limited for the smallest PM fractions (Tsai et al., 2015).
Composition and levels of PM can vary dramatically among locations depending on the source emission strength as well as meteorological conditions and local topography.Nevertheless, in order to control levels to safeguard health and minimize environmental damage, it is necessary to determine the contribution of the different sources.Receptor models (RM) have been extensively used to apportion PM sources based on concentrations determined at monitoring sites (Karagulian et al., 2015).The fundamental principle of receptor modelling is that mass and species conservation can be assumed and therefore a mass balance analysis can be used to identify and apportion sources of airborne PM in the atmosphere (Viana et al., 2008b).Diverse RMs have been widely described (Belis et al., 2013).Briefly, they can be classified as: 1) Chemical Mass Balance (CMB) models that require an input information of the emission profiles of the sources to identity their contributions to each sample, and 2) Multivariate Models (that include Principal Component Analysis (PCA), Multivariate Curve Resolution (MCR) and Positive Matrix Factorization (PMF)) that are able to identify the number of sources and the associated profiles of the sources based on marker species (Shi et al., 2014).Other tools such as enrichment factors (EF) and diagnostic ratios (DR) may help to identify pollutant sources, however they are not able to reconstruct the sources contribution of the PM collected in a sampling site (Galarneau, 2008;Hernández-Mena et al., 2011;Banerjee et al., 2015).The DR consists on comparing the ratio of marker species in source profiles with those from ambient samples in order to identify the most important sources in a region (Robinson et al., 2006).On the other hand, the EF compares the ratios of the atmospheric concentrations of various elements with the corresponding ratios in geological material to determine if a certain element has natural or anthropogenic origin (Enamorado-Báez et al., 2015).Different authors have recently reviewed the use of receptor models to analyze the sources of PM and its organic fractions in Europe (Viana et al., 2008a;Belis et al., 2013;Karagulian et al., 2015).They encourage the use of advanced factor analysis techniques that are able to deal with heterogeneous and complex data.In turn, they also note the common limitations in such type of analyses including the lack of markers for some sources, components or factors that may represent mixtures of emission sources, or source signatures that change with time.Another critical drawback is that a large number of samples are required for such kind of studies because although levels of PM 10 and PM 2.5 are currently widely monitored, their characterization is still limited due to the high cost that entails their chemical characterization.Thus, the combination of different types of receptor models is proposed to solve the limitations of each of the current models, by constructing a more robust solution based on their strengths (Viana et al., 2008a;Zeng et al., 2010).In turn, there are new tools such as stable isotopic analysis, that have been rarely used in receptor model studies and can also be used as complementary source identification technique (Pieri et al., 2013).
In previous studies we evaluated the levels and composition, regarding inorganic and organic species of PM 10 , PM 2.5 and PM 1 around a cement plant located in Barcelona (Sánchez-Soberón et al., 2015;2016).The objectives of the present study were: 1) to establish the most significant sources of PM in the area of study by applying PCA, MCR and isotopic analyses; and, at the same time, 2) to evaluate how these three methodologies can complement each other especially when dealing with relatively small data sets.

Receptor Models Principal Component Analysis (PCA)
Principal component analysis (PCA) is a widely known technique of multivariate analysis that has been largely used for the identification of the potential sources of pollutants in the environment (Viana et al., 2008a;Contini et al., 2010;Mari et al., 2010;Contini et al., 2012;Engelbrecht and Jayanty, 2013).The objective of a PCA is to derive principle components (PCs) as a linear combination of the original variables, which provides a description of the data structure with a minimum loss of information.These PCs are a set of orthogonal variables that best span the data variances in a way that they reflect the patterns of the correlations among the pooled initial variables (Zou et al., 2015).In order to equally weight the data variance it is important to use an appropriate data pretreatment.
In this study, PCA was performed using XLStat software.The Pearson correlation was used in order to homogenize the variances of the data.Varimax rotation was used to maximize the sum of the variances of the squared loadings.

Multivariate Curve Resolution by Alternating Least Squares (MCR-ALS)
Multivariate Curve Resolution-Alternating Least Squares method is similar to PCA (Jaumot et al., 2005), basically, the main difference between PCA and MCR is that the latter method is based on the application of more natural constraints like non-negativity instead of the orthogonality constraints that are fixed in PCA.The variance explained by the different components is not orthogonal (non-overlapped) as in PCA (Jolliffe, 2002).Natural sources in the environment are rarely orthogonal; thus the MCR methods may have a physically sounder interpretation than orthogonal database decomposition methods.MCR-ALS was introduced for the first time as a multivariate analysis tool for PM source apportionment by Tauler et al. (2009) and has successfully been applied for environmental source apportionment (Terrado et al., 2009) including organics in urban ambient air PM (Alier et al., 2013;Elorduy et al., 2016).
In our study, the MATLAB MCR-ALS toolbox interface was used for the analysis in order to obtain the loadings and scores of the components (Jaumot et al., 2005).Individual concentrations were divided by the standard deviation of the variable in order to homogenize the variance.The component scores (potential sources) were used to regression analyses with respect to the PM mass concentrations to quantify the daily contributions of each of the identified sources.

Isotopic Analysis
Stable isotope analysis is a tool capable of tracing the sources of atmospheric particles in urban air by comparing the carbon isotopic composition (δ 13 C) of ambient samples with those characteristic of potential sources (Widory et al., 2004;Gorka and Jedrysek, 2008;Puig et al., 2008;López-Veneroni, 2009).Carbon isotopic (δ 13 C) signature of total carbon in particulate matter can provide information about its origin as long as two premises are fulfilled: 1) the different potential sources have different carbon isotopic composition; 2) the carbon isotopic signature of the potential sources remain almost constant or vary in a known proportion throughout the processes they experience in the air (Boutton, 1991).The stable isotope composition of an element is expressed in delta (δ) notation, which is given in per mil (‰) units.For carbon, δ 13 C is calculated as following: where the carbon standard is a belemnite fossil (Belemnitella americana) from the Peedee Cretacic formation (North Carolina, United States), abbreviated as VPDB.Many authors have analyzed the carbon isotopic composition of the total carbon of ambient samples to discriminate the source of atmospheric particles by comparing the carbon isotopic composition of ambient samples with those of potential sources.The initial studies distinguished anthropogenic from natural particulate carbon emissions (Cheselet et al., 1981;Cachier et al., 1985).Later, Widory et al. (2004) differentiated between road traffic and industrial particle sources, and between diesel, fuel oil and several sources using both carbon and lead isotopes.Recently, Mkoma et al. (2014) studied the contribution of the burning of C3 and C4 plant species to the aerosols.Unlike PCA and MCR, isotopic analysis is a non-statistical fingerprinting technique and therefore only a few samples are required for the interpretation of the data.

Data Sets and Chemical Compounds Analyzed
Chemical tracer compounds were previously determined in ambient air PM 10 , PM 2.5 and PM 1 samples collected at a cement plant located in the surroundings of Barcelona (Sánchez-Soberón et al., 2015;2016).Details on the study area as well as on the sampling and analysis methodologies of major, inorganic and organic elements were previously described (Sánchez-Soberón et al., 2015;2016).In brief, 24-h samples were collected in quartz fibre filters (QFF) with high-volume active samplers.The studied facility is located within an urban area affected by a number of different industries.Moreover, two highways with heavy traffic cross the settlement.Fig. S1 of the study area is shown in the Supplementary Material.The plant has an annual production close to 1 million of cement tons and alternative fuels (they are authorized to use sewage sludge, meat and bone meal, refuse derived fuels and biomass) supply 20% of the total energy consumed.The first data set was composed by 23 variables, including the concentrations of 19 metals and metaloids (Al, As, Ca, Cd, Ce, Co, Cr, Cu, Fe, K, Li, Mn, Ni, Pb, Sb, Sn, Sr, Ti, V), OC + EC (Organic and Elemental Carbon) and 3 anions (Cl  (2 hopanes (17(H)α-21(H)β-29-norhopane and  17(H)α-21(H)β-hopane), 4 organic biomass combustion products (galactosan, mannosan, levoglucosan, dehydrabietic acid), 3 saccharides and derivates (α-glucose, β-glucose, manitol) and 2 dicarboxylic acids (succinic acid, glutaric acid)), determined in PM 10 , PM 2.5 and PM 1 samples (8 samples per mass fraction) collected in December 2014.Simultaneous data of inorganic and organic species was only available for 4 days in December 2014.
Apart from the chemical compounds described above, carbon isotopic composition was also measured in the PM 10 , PM 2.5 and PM 1 samples collected in November-December 2014.For that purpose a piece of approximately 1 cm 2 of all the quartz filters containing the particulate matter of each different size was cut off.The filter piece was folded using a pair of flat tipped tweezers, placed into a tin cube together with vanadium pentoxide as a catalyser to achieve complete combustion.The tin cubes containing the samples were then analysed for their carbon isotopic composition using an Isotope Ratio Mass Spectrometer (Delta C Finnigan MAT) with an Elemental Analyser Carlo Erba Flash 1112 and an interface Conflo III (Finnigan MAT).Isotope data was reported as δ 13 C values.A database with δ 13 C values of the potential sources in the area was constructed and is shown in Fig. 1, comprising: 1) Mineral fraction sources, including cement and kiln dust, and street dust; 2) Combusted Biomass fuels from C3 plants; 3) Combusted fossil fuels; 4) Combusted sewage sludge; 5) Traffic sources; 6) Deposited particles on a Petri dish in the school sampling point.For street dust as well as from 2 to 4 sources, carbon isotopic fingerprints were obtained from bibliography (more details are given in section 3.3).Isotopic fingerprints for cement, kiln dust, traffic and total particles were characterized in this study.Cement and kiln dust were obtained from the studied plant.Traffic samples were collected by placing a quartz fiber filter in the tailpipe of both, diesel and gasoline cars.Deposited particles were collected in a Petri dish left during five days in the sampling location.Values of δ 13 C on those samples were obtained with the same method than the one described for the ambient samples.

RESULT AND DISCUSSION
Particles of different sizes are released during different steps of the cement manufacturing process.In this study, the sources of PM 10 , PM 2.5 and PM 1 in the neighborhood of a cement plant were inferred by considering its content of inorganic and organic species by means of two receptor model methodologies, PCA and MCR.The results are discussed and compared below.In turn, the δ 13 C of the total carbon present in ambient PM samples was compared to the carbon isotopic signature of potential sources.

Sources of Major and Inorganic Elements of PM
Although major and inorganic elements were determined in three PM fractions (PM 10 , PM 2.5 and PM 1 ) (Sánchez-Soberón et al., 2015), in this case the PCA was only performed to the PM 10 fraction since a significant number of tracers were found below the detection limit in PM 2.5 and PM 1 (40 and 50%, respectively).For calculations, when a concentration was below the LOD, the value was assumed to be half of that limit (ND = 1/2 LOD).Four components were resolved in such a way that they could explain the major variance while the loadings could be interpreted as profiles of the potential sources in the area.Components were sorted according to the amount of explained variances.Hence, the 4-dimensional model explained 86% of the variance.The Kaiser-Meyer-Olkin coefficient was 0.718 (higher than 0.5) indicating a good performance of the PCA analysis.Subsequently, MCR analysis was performed and again four components were selected.In this case, the model explained 90% of the variance.Figs. 2 and 3(a)-3(d) present the factor loadings obtained by means of PCA and MCR, respectively.It can be observed that in spite of the different constraints considered by the two models, orthogonality and non-negativity, respectively, similar loading profiles were found in both cases.The interpretation of the four components for PCA and MCR was as follows: C1: Cement plant/Industry.This component was associated with two different emissions: mineral matter and combustion.PC1 showed high correlations with Al, Ca, Li and Ce.These elements, especially Al and Ca, are associated with crustal/mineral materials (Yatkin and Bayram, 2008;Belis et al., 2013).They are characteristic not only of resuspension of soil dust but also of mineral industries (ceramic or cement) as well as of quarries (Yatkin and Bayram, 2008).Therefore, the raw materials used in the cement plant located in the area may be responsible for the emissions of those elements.Correlations were observed between Ni and V which are typical tracers of fuel-oil/petcoke combustion (Viana et al., 2008a).Those elements may be related to the cement plant activity together with other industrial activities in the same area (brick, plastics and packaging factories among others).The V/Ni ratio is normally within the range 0.5-2.0 at regional background air monitoring sites across Europe (Pey et al., 2009;Moreno et al., 2010).On the other hand, the ratio V/Ni from the combustion of refineryproduced materials may be highly variable, depending on the origin of the crude, and sometimes overlap with that from mineral dust (Moreno et al., 2010).In our samples, the V/Ni value was higher than 2, being reasonable to associate the origin of those compounds in the area with petcoke and fuel oil combustion.In addition, K is an indicator of biomass combustion that could be correlated with the use of sewage sludge as alternative fuel in the cement plant as well.
C2: Traffic.This component was related to combustion processes and tire/break wear.PC2 is characterized by a high contribution of OC + EC (indicator of combustion processes) with a notable contribution of NO 3 -that may derive from the oxidation of NO x emissions from traffic emissions.Regarding heavy metals, Cd is associated with the engine emissions of vehicles (Pey et al., 2013), while As, Cu, Pb, Sb and Sn are associated with break wear (Belis et al., 2013;Pey et al., 2013).
C3: Crustal.This factor showed a high correlation with Mn and Fe (and Ti in the case of PCA).Those elements generally have a mineral origin and are generally correlated with other elements such as Al and Ca (Pey et al., 2013).However, in our case, Al and Ca appeared more associated with the activity of the cement plant (component 1), that also entails the use of mineral materials.In any case, the EF values for Mn and Fe were < 5, indicating a geological origin (Hernández-Mena et al., 2011).
C4: Marine and Regional Secondary Inorganic Aerosol (SIA).This component was characterized by Cl -and SO 4 2-, which suggests the contribution of marine aerosols.Regarding SO 4 2-, its origin is secondary aerosol (Viana et al., 2008a).Since SO 4 2-is related to Cl -, and the plant is located at a distance of 8 km from the sea, we considered that the origin of this element may be linked to marine as well as aerosols of regional origin.In turn, Fig. 4 shows the average contribution of the potential sources during the different sampling periods.It can be observed that the contribution of the sources varied significantly depending on the sampling period, showing that PM levels and composition depend not only on meteorological conditions but also on the intensity of irregular emissions such as traffic and some industrial activities.In October 2013, the contributions of traffic, marine and regional secondary aerosol and the cement plant/industry were similar (33, 29 and 26%, respectively), being those of crustal (12%) lesser.In December 2013, PM levels were clearly dominated by combustion sources namely, cement plant/industrial and traffic (66 and 22%, respectively), while crustal and marine and regional secondary aerosol were minor (9 and 3%, respectively).However, in July 2014 the highest contribution to total PM was associated with marine and regional secondary aerosol (50%), which was related to the active sea-breeze system in summer in this area.In the same period, a notable contribution was also observed for crustal, (35%), being low the contributions of the cement plant/industry (11%) and traffic (3%).Finally, in November/December 2014 the contribution of the cement plant/industry was the smallest (8%) of the four evaluated periods.The abated contribution of the cement plant to PM levels in this period is in agreement with the reduction of the production activity of the plant during the sampling period, since the plant was lessening its activity (from 25/11/14 to 08/12/14) until it was completely ceased (from 09/12/14 to 17/12/14).Therefore, crustal had a significant role as PM emitters (45%) followed by traffic (32%) and marine and regional secondary aerosol (15%).From these results, we can see that inorganic PM species in the area are mainly released by combustion sources, namely traffic and the activity of the cement plant, especially in winter months while in summer marine and regional secondary aerosol are more important.Nevertheless, during the last sampling period, in November/December 2014, the reduction of activity of the plant was noticeable in ambient inorganic PM components.

Sources of Organic PM
PCA was applied on a matrix of scaled concentrations of 24 organic compounds (13 PAHs and 11 organic tracer compounds) measured in three PM mass fractions (PM 10 , PM 2.5 , PM 1 ) collected in November/December 2014 to explore the amount of variance that could be explained by a reduced number of components and to identify the most Fig. 3. Loadings (a-d) and daily contributions (e-h) of PM 10 inorganic species derived from MCR analysis.determinant associations among the organic tracer compounds analyzed in this study.A four-component solution was selected from the analysis involving a 93.5% of explained variance.Addition of a fifth component did not showed any further relevant environmental information.Following the optimal number of components obtained by PCA, MCR-ALS was applied to the scaled column-wise data matrix (i.e., without mean-centering) and non-negativity constraints, resulting in four components which accounted for 93.2% of the total data variance.Despite the limited size of our dataset as well as the different constraints considered by the two models; again, a good correlation between PCA and MCR-ALS models (Fig. S2 in the supplementary material).A reliable interpretation of sources according correlated molecular tracer compounds was found.Figs.5(a)-5(d) shows the MCR loadings of the individual organic compounds in the components.Based on the abundance of molecular tracer compounds the four components were identified as: C1: Biomass burning.This component was related to biomass burning due to the high correlation with well known tracers of biomass burning such as levoglucosan, and its isomers, dihydroabietic acid and retene (Kourtchev et al., 2011;van Drooge et al., 2014), as well as other PAHs.
C2: Fossil fuel combustion.This factor showed a high correlation with compounds characteristic of fossil fuel combustion such as hopanes, which are found in the lubricating oils used by both gasoline-and diesel-powered motor vehicles (Křůmal et al., 2013) and some PAHs.
C3: Soil dust.This component was associated with soil dust resuspension due to the high contribution of glucose and mannitol as well as dicarboxylic acids and hopanes.Glucose is present in soil dust.Among the sugar alcohols, mannitol is one of the most abundant (Bauer et al., 2008;Schmidl et al., 2008).Mannitol is also associated with airborne fungal spores (Schmidl et al., 2008).C4: Regional Secondary Organic Aerosols (SOA).This component was characterized by a high contribution of secondary aerosol species (succinic and glutaric acid) (Hsieh et al., 2008) together with traces of PAHs.In the atmosphere, non-methane volatile organic compounds from traffic emissions are transformed through photochemical reactions to dicarboxylic acids such as succinic and glutaric acids.Therefore, the origin of this component is probably from long-and middle-range atmospheric transport.
The loadings of the individual compounds in the components (Figs.5(a)-5(d)) were used to estimate the daily contribution of the four identified sources of organic compounds in the studied area (Figs.(5e)-5(h)).Different sampling periods can be differentiated in the case of organic compounds: 1) when the plant was totally operative (from 18/11/14 to 24/11/2014), and 2) when the plant was reducing its activity (from 01/12/14 to 7/12/2014).Fig. 6 shows the average estimated contributions of identified sources and processes for each of the three PM mass fractions.Organic dust was the major contributor to the PM 10 fraction (33%), followed by biomass burning (29%), fossil fuel combustion (23%) and SOA + regional (15%).According to organic tracers, it was not possible to differentiate if biomass burning was related to the cement plant, which uses sewage sludge as alternative fuel, or to burning of pruning wastes in the area, neither if fossil fuel was related to the cement plant or to traffic.In PM 2.5, biomass burning was dominant (34%) followed by regional Secondary Organic Aerosols (32%), fossil fuel combustion (25%) and organic dust (2%).Finally, PM 1 was dominated by combustion products from biomass combustion (37%) and fossil fuel (32%) origin followed by regional Secondary Organic Aerosols (29%) and organic dust (2%).As it has been found in other studies our results indicate that main sources of the finest PM components are linked with combustion processes.In fact, major sources of fine particles in urban areas are vehicular emissions especially of diesel vehicles (Shirmohammadi et al., 2016).In our case, however biomass combustion also seems to have a considerable contribution in PM 2.5 and PM 1 levels what could be related to the plant activity.

Carbon Isotope Analysis
Bulk carbon isotope analysis was applied in this study as a complementary source identification technique in order to explore its capacity to further refine the insight on the origin of the PM sampled in the vicinity of the cement plant.Despite the low number of samples, this method allows the fingerprinting of organic PM sources based on the carbon isotopic composition of the samples, which on its term can be compared to potential sources.For this purpose, the δ 13 C was determined in the carbon content of the same PM 10 , PM 2.5 and PM 1 samples, collected in November/December 2014, where organic compounds have also been determined, and compared with those of different types of potential contaminant sources in the area (Fig. 1).The potential sources included and indicated in Fig. 1 were: 1) Mineral fraction sources (cement and kiln dust measured in this study corresponding to points 1 and 2, respectively; and point 3 obtained from López-Veneroni ( 2009)), 2) Combusted biomass fuels of C3 plants origin (points 4 to 9, obtained from Garbaras et al. (2015)), 3) Combusted fossil fuels (points 10 to 23, being points 10, 11 and from 13 to 18 obtained from Widory et al. (2004) and 12, 19 and 20 from Andres et al. (1994)), 4) Combusted sewage sludge (point 21, obtained from Garbaras et al. (2015)), 5) Traffic sources (point 22, unleaded gasoline from an exhaust pipe; and point 23, diesel from an exhaust pipe, both analytically determined Fig. 5. Loadings (a-d) and daily contributions (e-h) of the analysed PM 10 , PM 2.5 and PM 1 organic compounds in the identified components of the MCR-ALS. in this study), 6) Deposited particles on a Petri dish at a school located near the cement plant (point 24, analytically determined in this study).More details on the sources of Fig. 1 are indicated in the supplementary material.In Fig. 1 it can be observed that the δ 13 C of the PM samples collected around the cement plant (points 25 to 30) were in the range of combustion sources.More specifically, the analyzed samples were within the range of two main groups of sources, namely group 1) biomass fuels, which include agricultural waste, wood pellets, as well as sewage sludge (points 5, 6 and 21, respectively) and group 2) fossil fuels, corresponding to domestic fuel oil, diesel particles, petroleum and unleaded gasoline (points 17, 18, 19, 22 and 23, respectively); while cement and kiln dust from the plant (mineral fraction in Fig. 1) are far from the samples collected around the cement plant.Hence, fuel combustion and biomass combustion sources are clearly dominating in the ambient PM samples of study, which is consistent with the results of the source apportionment analysis of the organic compounds, which were determined in the same samples.The former analysis, however, was able to differentiate between biomass and fuel combustion sources, while the isotopic results are overlapping.
Although values of samples around the cement plant are far from the δ 13 C of mineral fraction, if we take into account that the cement plant uses sewage sludge as alternative biomass fuel, the values of δ 13 C in ambient PM were feasible with a contamination coming from the cement plant but also from traffic (diesel and gasoline combustion particles), domestic heating (domestic fuel oil) or fossil fuel used in the cement plant (petroleum).In order to discriminate the contribution of the cement plant, δ 13 C values of samples collected when the cement plant was not operating were compared to those of operational days.It can be seen in Fig. 1, that point 26, which represents the average δ 13 C value of samples collected when the cement plant was not operating, is within the range of δ 13 C values obtained during the operational period.This result indicates that the cement plant emissions would not be significant in ambient PM during the overall sampling period or it could mean that the cement plant emissions cannot be detected since the δ 13 C signature is very similar to the value of the average emissions from other sources.
Regarding differences in the δ 13 C of PM according to size, a small but significant increase in the δ 13 C value was noted when increasing PM size (Fig. 7).PM 1 , PM 2.5 and PM 10 presented average δ 13 C values of -26.8 ± 0.1‰, -26.4 ± 0.1‰ and -25.8 ± 0.4‰, respectively.Results indicate that PM 2.5 is mainly composed of PM 1 while there is an enrichment of 13 C in the coarse fraction between 2.5 and 10 µm.This increase in δ 13 C matches with the presence of calcite in PM 2.5 and PM 10 , (revealed by DRX analyses, see Fig. S3 in the Supplementary Material) but not in PM 1 .Hence, the differences in δ 13 C observed in PM samples depending on the size (Fig. 7) indicate the effect of the mineral contribution in the coarser particle samples.These results are in agreement with the differences observed in the organic tracer composition of the different PM size fractions (Fig. 6), where higher contributions of organic dust were noted in PM 10 in comparison with the smaller PM 2.5 and PM 1 fractions.Geogenic calcite usually presents δ 13 C values close to 0%.Moreover, sample point 24, that represented solid deposited material with no size selection and therefore containing larger particles, corroborates the influence of the mineral part on the δ 13 C since it showed a δ 13 C value of -24.5 ± 0.3‰ which is even higher than that found in PM 10 (-25.8 ± 0.4‰), in agreement with the presence of geogenic calcite.Therefore the mineral contribution of the cement plant starts to be visible when the particle size increases, however, its contribution to the total carbon in the ambient PM was always notably smaller than the contributions of the combustion sources that according to δ 13 C analysis dominate in the vicinity of the cement plant.

CONCLUSIONS
Particle matter is one of the pollutants that generate greatest concern around cement plants.Ambient PM is composed by inorganic and organic compounds and can be released from other sources that include combustion processes (traffic, power generation, etc.), mineral origins or secondary aerosols formation.These sources may be active simultaneously, resulting in a complex mixture of compounds difficult to appoint to one source.Moreover, external factors, such as meteorological conditions, may have equal influences on the compounds in the air shed, resulting in correlations among compounds that may not have the same emission source.It is therefore complex to apply source apportionment tools on a dataset.In addition, receptor models require large databases that, such as in our case, are not always available because of the high costs that entail the collection of the large number of samples as well as the analysis of the numerous tracers required.In our case, we used available PM ambient data regarding inorganic and organic compounds near a cement plant to establish the most significant sources of those pollutants by applying source apportionment methodologies PCA and MCR.In addition, we analyzed the δ 13 C in the carbon content of PM ambient samples to identify sources by comparing with the δ 13 C of potential contaminant sources.
In spite of the different constraints considered by PCA and MCR-ALS and the the small datasets available, similar sources could be identified by both methodologies according to the correlations found between tracer compounds: marine and secondary inorganic aerosol, cement plant/industrial, traffic and crustal.MCR-ALS allowed quantifying those sources.The contributions of those sources varied significantly depending on the period of the year and even the year of collection.However, it could be noted that tracer PM species in the area were mainly released by combustion sources, namely traffic and the activity of the cement plant/industry, especially in winter months while in summer marine and secondary inorganic aerosol were more important.The reduction of activity of the plant was noticeable in ambient inorganic PM components.PCA, MCR-ALS and δ 13 C on the organic dataset indicated that PM around the cement plant was mainly influenced by combustion sources, biomass burning and fossil fuel combustion.However, according to MCR-ALS and δ 13 C it was not clear if all was emitted by the cement plant (burning fossil and sewage sludge as alternative fuel), since other activities, such as traffic, domestic heating and biomass removal from fields could also be involved.In spite of the limitations of our datasets regarding their size in comparison with traditional source apportionment studies, the overall results indicate that the three studied techniques, PCA, MCR and δ 13 C agree between each other, even though they have been applied by using different type of data.PCA and MCR-ALS allowed differentiating cement plant emissions from traffic by means of correlations found between tracers of combustion and mineral matter, in the case of the cement activity, and between tracers of combustion and engine emissions and tire/break, in the case of traffic.On the other hand, PCA, MCR and δ 13 C on the organic dataset empathized the significance of combustion sources in the area from both biomass burning and fossil fuel combustion.In the case of carbon isotopes, the lack of differences between operational and non-operational days of the plant showed the significant impact of traffic during the sampling period.These results confirm that the different techniques can be used together to corroborate each other and to give more reliability to the results; or, they can also be used separately in studies where only few data is available or there is some design limitation.

Fig. 1 .
Fig. 1. δ 13 C values of different sources, types of fuels and materials susceptible of being used as fuels together with δ 13 C of the PM ambient samples.Error bars represent deviation between analytical duplicates.Description of each sample is listed in the supplementary material.

Fig. 2 .
Fig. 2. Loading profiles for each component derived from PCA analysis of PM 10 inorganic species.

Fig. 4 .
Fig. 4. Percentages of source contributions to main PM components of the four factors derived from MCR classified by 4 sampling periods.

Fig. 6 .
Fig. 6.Estimated contributions of identified sources and processes on the observed PM mass fractions in the vicinity of the cement plant.

Fig. 7 .
Fig. 7. Average δ 13 C values in PM 1 (blue diamonds), PM 2.5 (red squares), and PM 10 (green triangles) samples collected around the cement plant while it was operating, from November 17 th to December 7 th , 2014.Error bars represent deviation between analytical duplicates.