Water-Soluble Anions in PM 10 Samples Collected in the Metropolitan Area of Costa Rica: Temporal and Spatial Variations

: The concentrations of water-soluble anions present in PM 10 collected from ambient air in four sites of Costa Rica’s greater metropolitan area (GAM) were analyzed. Samples were collected over a 24-h period for three days during the 2011–2018 period, using high-volume air samplers with mass ﬂow controller. The PM 10 samples were collected using quartz ﬁber ﬁlters, which were analyzed using ion chromatography to identify organic and inorganic ions. The annual averages for PM 10 , F − , Cl − , and SO 42 − showed signiﬁcant differences between the sampling sites, with Uruca (UR) consistently registering annual averages above the maximum concentration established by Costa Rica’s air quality regulations (30 µ gm − 3 ). The ions analyzed contributed 33%, 34%, 35%, and 37% of the PM 10 mass for sampling sites UR, HA, AL, and CA, respectively. Using Spearman correlations and principal component analysis (PCA), the following contributions were identiﬁed: biomass burning, secondary particle formation processes from mobile and stationary emissions, and biogenic emissions. For each variable, seasonal patterns and trends were analyzed using time series with additive decomposition.


Introduction
Costa Rica's greater metropolitan area (GMA) is a multinuclear region, comprising the cities of Alajuela, Cartago, Heredia, and San Jose, which function as subcenters, favoring poor-quality public transportation and vial infrastructure exclusively featuring a radial road system that hinders mobility in the region. This geographical area includes 3.83% (2.044 km 2 ) of the total national territory, but houses approximately 60% of the population (2.6 million), representing the most urbanized, populated, and economically active region in the country [1]. Demographic and economic growth patterns and a lack of adequate territorial planning have generated a sustained deterioration of air quality in the GAM, where particles constitute one of the main air pollutants [2].
The World Health Organization (WHO) has indicated that particulate matter (PM) is one of the pollutants that most affects people, with issues such as cardiovascular and respiratory disease resulting from chronic exposure [3][4][5][6][7]. Likewise, recent studies have associated the concentration of PM in urban areas with lung damage and increased mortality [8]. PM 10 is considered one of the pollutants with the greatest effect on human health, causing immune system reactions, lung irritation, cell damage, asthma, chronic bronchitis, cancer, and even death [9]. Particulate matter also impacts the environment, generating stress on the ecosystem and promoting its degradation and the loss of sensitive species [10]. The continuous exposure of plants to air pollutants makes them susceptible to their effects. PM tends to deposit on plant leaves, causing acute damage over short periods when the concentrations are high and generating necrotic lesions. If the exposure is over a broader period under low concentrations of the pollutant, then chronic effects, such as leaf injury, stomatal damage, premature senescence, decreased photosynthetic activity, disturbed membrane permeability, and reduced growth and yield, appear. Likewise, morphological impacts are observed as a decrease in the area and number of leaves due to the obstruction of the stomata caused by the particles, which disrupt the physiological activity of the plant [11]. Particulate matter also reduces photosynthetically active radiation (PAR) by altering the surface optical properties, affecting the photosynthetic activity of the plant and increasing the temperature of the leaves [11,12]. Similarly, the ecosystem's soils are affected by the particulate matter deposition, which alters their chemical composition and pH and results in diverse effects depending on the origin and nature of the particles, among these the reduction of available nutrients in the soil or the introduction of heavy metals into the ecosystem, inhibiting plant growth [13]. The composition of the PM can vary widely depending on the source and the environmental conditions. Among the substances that comprise these particles, studies have detected fatty acids, n-alkanes and iso-alkanes, phthalate esters, siloxanes, sterols, sugars and sugar alcohols, lignin tracer compounds and resin acids, dicarboxylic acids, polycyclic organic compounds, organic nitrogen compounds, and products from secondary oxidation of monoterpenes [14].
The fraction of water-soluble components (WSC) in the particles contain both inorganic ions and organic compounds, which exert an influence on the activity of condensation nuclei of the cloud, the size of the particulate material, and the toxicity [15]. The concentration of ionic inorganic species accounts for approximately one-third of the fine fraction and is mainly influenced by prevailing weather factors, geographical conditions, and the variability of the emission sources [16]. Several studies have shown that secondary inorganic ions (SO 4 2− , NO 3 − and NH 4 + ) dominate water-soluble ionic species and define the acidic or alkaline nature of the particles [17]. However, there have been few studies aimed at quantifying the presence of water-soluble organic compounds. Thus, monocarboxylic and dicarboxylic acids, for example, are key groups of organic compounds that have been identified in atmospheric particles [18]. Both formic and acetic acid, as well as oxalic acid, have also been detected in samples of particles collected in urban and rural environments [19]. These acids have a high water solubility and the potential to modify the hygroscopic properties of the particles, including their size and the activity of the condensation nuclei [20]. In the particulate phase, most of these acids are neutralized as formate, acetate, and oxalate salts [21], derived mainly as a product of biomass burning or from secondary reactions in the atmosphere [22,23]. Although particulate oxalate is highly hygroscopic, its concentration is much lower when compared to sulfate and nitrate [24]. However, a study on Brazil's urban areas showed that oxalate is the water-soluble organic ion with the highest concentration on PM 10 particles [25]. The chemical composition of the particles has a significant effect on global weather patterns [26], human health [27], and air quality.
In this study, the temporal and spatial variations of water-soluble anions present in samples of PM 10 particles collected in the metropolitan area of Costa Rica were analyzed to determine the possible sources that originate them and, thus, guide the actions that must be included in the respective air quality management plans.

Sampling
For the PM 10 sampling, 4 monitoring sites were selected Table 1. The sites were representative of commercial, industrial, and residential areas, all located in Costa Rica's metropolitan area Figure 1.
Sampling campaigns were conducted from January 2011 to December 2018. Twenty-fourhour PM 10 samples were collected once a day for three days using high-volume air samplers (Hi-Vol) with MFC (mass flow controller), operating at a flow rate of (1.13 ± 10%) m 3 min −1 Atmosphere 2021, 12, 1264 3 of 14 with 10 µm size-selective inlets (Thermo Andersen, Massachusetts, USA). Table 2 summarizes the number of samples collected per year.   Sampling campaigns were conducted from January 2011 to December 2018. Twentyfour-hour PM10 samples were collected once a day for three days using high-volume air samplers (Hi-Vol) with MFC (mass flow controller), operating at a flow rate of (1.13 ± 10%) m 3 min −1 with 10 µm size-selective inlets (Thermo Andersen, Massachusetts, USA). Table  2 summarizes the number of samples collected per year.
Temperature and pressure during the sampling were determined to calculate the actual air volume for each sample under standard conditions (25 °C and 1 atm). Quartz fiber filters (Whatman CAT No. 1851-865, Massachusetts, USA) were used for particle sampling, which were desiccated for 24 h after sampling in a desiccator under the following conditions: temperature of 15-30 °C and humidity of less than 40%. PM10 mass concentrations were measured gravimetrically by weighing the quartz fiber filters (twice) before (blank filter) and after (total filter) sampling using an analytical balance.  Temperature and pressure during the sampling were determined to calculate the actual air volume for each sample under standard conditions (25 • C and 1 atm). Quartz fiber filters (Whatman CAT No. 1851-865, Massachusetts, USA) were used for particle sampling, which were desiccated for 24 h after sampling in a desiccator under the following conditions: temperature of 15-30 • C and humidity of less than 40%. PM 10 mass concentrations were measured gravimetrically by weighing the quartz fiber filters (twice) before (blank filter) and after (total filter) sampling using an analytical balance.

Filter Extraction
Water soluble ions were extracted from a 2.54 cm by 23 cm strip of each quartz filter used for the gravimetric analysis. The strip was placed in a 150 mL beaker containing 50 mL distilled and deionized water (DDW) (18.2 MΩ resistivity). The beakers were placed in an ultrasonic water bath for 30 min at room temperature. The solution was then poured into a volumetric flask and filled to the mark with DDW. The extracts were filtered through 0.45 µm pore-size microporous membranes before instrumental analysis.

Chemical Analysis
A water-soluble anions analysis was performed by dual-microbore-suppressed ion chromatography using a DIONEX ICS-3000 machine with a quaternary pump. For the anion analyses, the instrument was equipped with an IonPac AS14A column (8 mM Na 2 CO 3 /1 mM NaHCO 3 as the eluent). Ionic species were identified and quantified by interpolation on a calibration curve of 7 standard solutions and were prepared from certified commercial solutions (MERCK Suprapur solutions, Hong Kong, China). A fresh calibration curve was prepared for every 20 samples, together with a dissolution of quality control of 5 mgl -1 prepared from a certified DIONEX synthetic sample. Detection limits for each ion are shown in Table 3.

Statistical Analysis
The statistical analysis was conducted using Minitab software. First, an Anderson−Darling normality test was used to compare the empirical cumulative distribution function of the sample data with the expected distribution function as if the data were normal, wherein a large enough difference (a p-value less than the significance value) rejects the null hypothesis [28]. An analysis of variance (ANOVA) was then conducted. As this study contained several datasets, each with a significant number of factors, the ANOVA was used even when these showed abnormal behavior, as this parametric test provides a higher resolution than non-parametric methods and is effective with a large enough sample size [29]. The ANOVA was used to compare the mean value of the ions per sampling site and year. While the test can evidence statistically significant differences between two or more levels of a given factor, it does not specify between which factors these differences were observed [30]. Thus, Tukey's method of multiple comparisons was used in conjunction with the ANOVA to create confidence intervals between each pair of differences analyzed, enabling us to specify between which levels the differences shown in the ANOVA were observed [29]. Additionally, a Spearman correlation test was used to verify the correlation between two variables when the relation between them was nonlinear, measuring the monotonic relationship between two continuous or ordinal variables. In this case, the p-value indicated the statistical significance of the correlation, while the Spearman coefficient indicated its strength [31]. Lastly, a principal component analysis (PCA) was conducted. This technique helps decrease the dimensionality of the variables by using the correlations or levels of association between them to classify a large number of variables based on their correlational proximity. This method consists of obtaining the main components, scaled for one of the sample sights, and taking from these only those components with a variance of greater than 1, allowing the groupings to be observed graphically. All data were processed using the R Core Team (2020).

Time Series
To analyze the temporal variation of the parameters measured (trends and seasonal patterns), respective time series were constructed, for which the information needed to be divided in a uniform manner. For this, an average monthly observed concentration was obtained for PM 10 , F − , Cl − , NO 3 − , and SO 4 2− from January 2011 through December 2018, leading to a series of 96 datasets for each variable and sampling site. The remaining results, which pertained to a statistically insignificant portion of the total, were replaced through linear interpolation, considering the anterior and posterior values. This process did not affect the trend analysis for each time series. Once the database was refined, obtaining a single (average) monthly value, we proceeded to generate a time series for each variable per sampling site. All data were processed using the R Core Team (2020). As such, charts corresponding to the series were generated, as well as those of their respective decompositions into their various components.
The Mann-Kendall trend statistical test was applied to account for enough evidence in the significance determination of the trend (consistently increasing or decreasing, p < 0.05). A p-value higher than 0.05 does not necessarily imply a nonexistent trend in the longitudinal analysis; instead, it means a not significant result in statistical terms. Table 4 shows the averages and the standard deviations of the data of concentrations of PM 10 and water-soluble anions obtained for the four sampling sites during the 2011−2018 period.

Concentrations of Water-Soluble Ions and PM 10
An Anderson−Darling normality test was performed for each of the parameters evaluated on each collecting site during the sampling period, showing that most of the parameters followed a non-normal distribution.
Upon conducting the variance analysis for the sampling sites, significant differences were found in at least one mean value for PM 10 , F − , Cl − , and SO 4 2− , with a significance level of 5%. The average values for the remaining parameters showed no significant differences between the sites. Figure 2 shows the results of the Tukey test, displaying significant differences between the following sites: UR−AL (PM 10  . This zone features a high concentration of industrial and commercial activities combined with an increased circulation of light vehicles and heavy cargo and combustion processes from point sources. Differences between AL and all other sites were also found. This site is located downwind from the metropolitan area and is dominated by commercial activities, though with a lower concentration of manufacturing industries. Sampling site HA is also different, as it is the only site located in a residential zone, and it is near an important high-traffic road.  (19) 27 (7) 28 (9) 26 (7) 25 (6) 26 (9) 26 (8)   . This zone features a high concentration of industrial and commercial activities combined with an increased circulation of light vehicles and heavy cargo and combustion processes from point sources. Differences between AL and all other sites were also found. This site is located downwind from the metropolitan area and is dominated by commercial activities, though with a lower concentration of manufacturing industries. Sampling site HA is also different, as it is the only site located in a residential zone, and it is near an important high-traffic road. The ions analyzed contribute to 33%, 34%, 35%, and 37% of the PM10 collected for sampling sites UR, HA, AL, and CA, respectively, per the following sequence of individual contribution: SO4 2− > Cl − > NO3 − > PO4 3− > NO2 − > C2O4 2− > HCOO − > F − > CH3COO − . As shown in Figure 3, the PM10 concentrations recorded for the sampling sites during the dry season (December-April) were slightly higher than those recorded during the rainy season (May-November), with some key exceptions. In the dry period, the metropolitan area is influenced by trade winds from the Caribbean Sea, which can have speeds of up to 30 kmh −1 and cause a considerable drop in the The ions analyzed contribute to 33%, 34%, 35%, and 37% of the PM 10 collected for sampling sites UR, HA, AL, and CA, respectively, per the following sequence of individual contribution: SO 4 2− > Cl − > NO 3 − > PO 4 3− > NO 2 − > C 2 O 4 2− > HCOO − > F − > CH 3 COO − . As shown in Figure 3, the PM 10 concentrations recorded for the sampling sites during the dry season (December-April) were slightly higher than those recorded during the rainy season (May-November), with some key exceptions. In the dry period, the metropolitan area is influenced by trade winds from the Caribbean Sea, which can have speeds of up to 30 kmh −1 and cause a considerable drop in the region's precipitation levels. For the same reason, recorded concentrations of Cl − , mainly resulting from the effects of the marine spray, were higher in the dry season. In the case of ions such as nitrate, the formation of which is linked to a transformation from the gaseous to the particle phase, concentrations were higher during the rainy season than the dry season in the majority of cases, possibly because temperatures are lower during this period, while the trade winds that help clear pollutants from the metropolitan area are reduced.
Several authors have used the ion NO 3 − to trace mobile emissions sources through combustion gases created by vehicle exhaust, while SO 4 2− indicates emissions from stationary sources. Given this, the relation between the concentration of these ions was used to evidence the influence of mobile emissions as opposed to point sources [32,33]. Figure 4 shows the monthly averages of this relationship throughout the sampling period.
As shown above, the relation between NO 3 − and SO 4 2− showed an upward trend for all sampling sites in the 2011-2016 period, while the ratio was more stable in 2016-2018. This could be due to two fundamental variables: (2) A policy recently implemented by the Costa Rican Oil Refinery to make two types of fuel oil available to the industrial market, one of which has a much lower sulfur content than the regular one, which contains around 2% [35].
Atmosphere 2021, 12, x FOR PEER REVIEW 9 of 16 region's precipitation levels. For the same reason, recorded concentrations of Cl − , mainly resulting from the effects of the marine spray, were higher in the dry season. In the case of ions such as nitrate, the formation of which is linked to a transformation from the gaseous to the particle phase, concentrations were higher during the rainy season than the dry season in the majority of cases, possibly because temperatures are lower during this period, while the trade winds that help clear pollutants from the metropolitan area are reduced.
Atmosphere 2021, 12, x FOR PEER REVIEW 10 of 16 Several authors have used the ion NO3 − to trace mobile emissions sources through combustion gases created by vehicle exhaust, while SO4 2− indicates emissions from stationary sources. Given this, the relation between the concentration of these ions was used to evidence the influence of mobile emissions as opposed to point sources [32,33]. Figure 4 shows the monthly averages of this relationship throughout the sampling period.  Several authors have used the ion NO3 − to trace mobile emissions sources through c bustion gases created by vehicle exhaust, while SO4 2− indicates emissions from statio sources. Given this, the relation between the concentration of these ions was used to evid the influence of mobile emissions as opposed to point sources [32,33]. Figure 4 shows monthly averages of this relationship throughout the sampling period.  According to the Mann−Kendall test results, only AL (p = 0.00002) and UR (p = 0.03789) sites presented a significant statistical trend. But UR and HA exhibited the highest peak ratios, explained by the high vehicular traffic around those areas. It is also relevant to point out that the metropolitan area was affected by emissions from the Turrialba Volcano during the sampling period, presenting significant activity between 2014-2016. However, not the whole time the plume was pointing to the metropolitan area. Table 5 shows the Spearman coefficients for the gravimetric concentrations of PM 10 , as well as the ions evaluated in the sampling sites. All of these showed a moderate to strong correlation (HA = 0.520, UR = 0.577) between SO 4 2− and NO 3 − , which are secondary aerosols that form via the oxidation of SO 2 and NO x emitted during the burning of fossil fuels. Likewise, a significant correlation between PM 10 and these ions of secondary origin was identified. For HA, AL, and UR, a moderate correlation was found between HCOO − −CH 3  2− in particulate material are primary emissions and the photochemical processes that occur at the atmospheric level. C 2 O 4 2− could originate from oxalic acid in the gaseous phase, which reacts with pre-existing particles via the coagulation of particles or originates from heterogeneous reactions within large droplets [36]. SO 4 2− and C 2 O 4 2− did not present significant correlations in all the sampling sites, which could indicate the presence of other primary sulfate sources (e.g., sea salt, combustion, and crustal sulfate), which are as important as secondary formation through in-cloud processing, masking any correlation. Further research is needed for this particular finding.  Figure 5 shows the results of the PCA obtained for each sampling site. For AL and UR, three factors explained the majority of the data variance. Otherwise, HA and CA required four factors. The first factor presented similar parameters in common for AL, HA, and UR, related to PM 10 , HCOO − , CH 3 COO − , Cl − , NO 2 − , NO 3 − , SO 4 2− , and C 2 O 4 2− . This suggests that these three sites share the same primary and secondary sources of particulate matter, where combustion emissions are strongly present around the sampling sites. However, for CA, the first factor mainly accounted for F − , Cl − , NO 2 − , NO 3 − , SO 4 2− , and C 2 O 4 2− , pointing to a significant influence of primary emissions related to fuel burning (mainly biomass). The second factor included PM 10 , CH 3 COO − , and PO 4 3− for most sites, with eigenvalues higher than one. This indicates the influence of particles from biomass burning, except for CA, where PM 10 , HCOO − , CH 3 COO − , and Cl − were more important, showing the presence of secondary aerosols from biogenic origin. In all sites, the third factor was more important for F − , HCOO − , PO 4 3− , and C 2 O 4 2− and related to multiple sources of contribution (mineral dust, biomass burning, fuel emissions).  Figure 5 shows the results of the PCA obtained for each sampling site. For AL and UR, three factors explained the majority of the data variance. Otherwise, HA and CA required four factors. The first factor presented similar parameters in common for AL, HA, and UR, related to PM10, HCOO − , CH3COO − , Cl − , NO2 − , NO3 − , SO4 2− , and C2O4 2− . This suggests that these three sites share the same primary and secondary sources of particulate matter, where combustion emissions are strongly present around the sampling sites. However, for CA, the first factor mainly accounted for F − , Cl − , NO2 − , NO3 − , SO4 2− , and C2O4 2− , pointing to a significant influence of primary emissions related to fuel burning (mainly biomass). The second factor included PM10, CH3COO − , and PO4 3− for most sites, with eigenvalues higher than one. This indicates the influence of particles from biomass burning, except for CA, where PM10, HCOO − , CH3COO − , and Cl − were more important, showing the presence of secondary aerosols from biogenic origin. In all sites, the third factor was more important for F − , HCOO − , PO4 3− , and C2O4 2− and related to multiple sources of contribution (mineral dust, biomass burning, fuel emissions).

Analysis of the Time Series
The behavior of the PM10 values and the concentration of the main water-soluble anions through the formulation of time series for each sampling site were analyzed. As shown in Figure 6, all of the parameters evaluated for UR demonstrated well-defined seasonal patterns. For instance, the concentrations of PM10 and Cl − showed higher values during January-April (the dry season) and July, when the rainy season begins to wind down. As explained above, this behavior is partly due to the influence of prevailing winds in the metropolitan area that influence rainfall patterns and are seen as a critical factor in removing atmospheric particulate matter. Additionally, the ions generated through secondary transformation processes, such as NO3 − , presented a differentiated pattern with higher values during the rainy season due to a drop in temperatures and winds and lessening the chance of contaminant removal.

Analysis of the Time Series
The behavior of the PM 10 values and the concentration of the main water-soluble anions through the formulation of time series for each sampling site were analyzed. As shown in Figure 6, all of the parameters evaluated for UR demonstrated well-defined seasonal patterns. For instance, the concentrations of PM 10 and Cl − showed higher values during January-April (the dry season) and July, when the rainy season begins to wind down. As explained above, this behavior is partly due to the influence of prevailing winds in the metropolitan area that influence rainfall patterns and are seen as a critical factor in removing atmospheric particulate matter. Additionally, the ions generated through secondary transformation processes, such as NO 3 − , presented a differentiated pattern with higher values during the rainy season due to a drop in temperatures and winds and lessening the chance of contaminant removal. tained increase in the country's vehicular fleet and a reduction in average circulation speeds throughout its vial infrastructure [2]. F − showed an inverse behavior, with a sustained downward trend possibly attributable to reduced contributions from biomass burning in this site. The gravimetric concentration of PM10 and ions showed similar trends across the four sampling sites studied. According to the results of the Mann−Kendall test, only the trends obtained for F − (p = 0.00003), Cl − (p = 0.00145), and SO4 2− (p = 0.047) turned out to be statistically significant.

Conclusions
The results obtained from this study suggest that the nature of the processes responsible for PM10 generation in the selected sampling sites can vary depending on the predominant land-use patterns in each zone related to the local emissions sources and atmospheric photochemical reactions. UR and AL, which had a significant influence from commercial and industrial activities, showed an important contribution of secondary sources to the particulate matter composition, given the correlations between SO4 2− -NO3 − , HCOO − −CH3COO − , HCOO − −NO3 − , CH3COO − −NO3 − , and C2O4 2− −NO2 − , while CA exhibited a stronger correlation for F − and C2O4 2− , directly related to biomass burning. As for HA, despite being a residential area, it showed a major influence of primary particulate source contributions. The PCA results suggest that AL, HA, and UR sites have the same primary and secondary sources of particulate matter. In contrast, in CA, the primary sources, like biomass burning, seem to be more critical.
The concentrations of anions present in the PM10 samples represented between 33 and 37 percent of the total mass collected, with SO4 2− , Cl − , and NO3 − being the most abundant elements. As for the organic water-soluble anions, the HCOO − and CH3COO − concentrations were higher in AL, while C2O4 2− was dominant in UR. In the case of PM 10 , as well as for NO 3 − , the data series showed a slight upward trend, possibly attributable to an increase in primary emissions from mobile sources due to a sustained increase in the country's vehicular fleet and a reduction in average circulation speeds throughout its vial infrastructure [2]. F − showed an inverse behavior, with a sustained downward trend possibly attributable to reduced contributions from biomass burning in this site. The gravimetric concentration of PM 10 and ions showed similar trends across the four sampling sites studied. According to the results of the Mann−Kendall test, only the trends obtained for F − (p = 0.00003), Cl − (p = 0.00145), and SO 4 2− (p = 0.047) turned out to be statistically significant.

Conclusions
The results obtained from this study suggest that the nature of the processes responsible for PM 10 generation in the selected sampling sites can vary depending on the predominant land-use patterns in each zone related to the local emissions sources and atmospheric photochemical reactions. UR and AL, which had a significant influence from commercial and industrial activities, showed an important contribution of secondary sources to the particulate matter composition, given the correlations between SO 4 2− -NO 3 − , HCOO − −CH 3 COO − , HCOO − −NO 3 − , CH 3 COO − −NO 3 − , and C 2 O 4 2− −NO 2 − , while CA exhibited a stronger correlation for F − and C 2 O 4 2− , directly related to biomass burning. As for HA, despite being a residential area, it showed a major influence of primary particulate source contributions. The PCA results suggest that AL, HA, and UR sites have the same primary and secondary sources of particulate matter. In contrast, in CA, the primary sources, like biomass burning, seem to be more critical.
The concentrations of anions present in the PM 10 samples represented between 33 and 37 percent of the total mass collected, with SO 4 2− , Cl − , and NO 3 − being the most abundant elements. As for the organic water-soluble anions, the HCOO − and CH 3 COO − concentrations were higher in AL, while C 2 O 4 2− was dominant in UR.
Both the concentration of PM 10 and its respective anions presented well-defined seasonal patterns. The PM 10 and Cl − values, for example, were higher during the dry season than the rainy season due to the influence of trade winds from the Caribbean Sea, which considerably decreased the effect of precipitation as a means of particle removal. NO 3 − , in contrast, showed an inverse behavior, possibly due to higher temperatures during the rainy season combined with a decrease in the removal of pollutants by way of winds.
The relation between NO 3 − and SO 4 2− concentrations showed an upward trend for the 2011-2016 period for all sampling sites, while the ratio between these anions remained stable for 2016-2018. This evidences the influence of a nearly 8% annual increase in the national vehicular fleet in the first case and new policies designed to improve the sulfur content of industrial fuels in the second. UR and HA exhibited the highest peak ratios during the sampling period since both sites are very near high-traffic roads. It can also not be ignored that this ratio was influenced by emissions from natural sources such as volcanic activity.
The low SO 4 2− and C 2 O 4 2− correlations in all the sites suggest that primary sulfate sources could provide a significant contribution to the particulate matter composition, in addition to the well-known secondary formation due to in-cloud processing.

Data Availability Statement:
The database is available from the corresponding author upon reasonable request.