Sampling was conducted at an urban site in Montréal from 13 August to 11 November 2020. The sampling site, labeled as MTL, was located on the rooftop of campus MIL (12 m above ground level) at the University of Montréal (45∘31′21′′ N, 73∘37′14′′ W) in the neighborhood of Outremont. The site is characterized by a high density of residential and commercial premises.

A total of 80 PM2.5 filter samples (150 mm quartz-fiber filters, PALL) were collected with a 24 h resolution using a high-volume sampler (CAV-A/MSb, MCV S.A., Spain) operating at 30 m3 h-1. Filters were baked at 550 ∘C for 12 h before sampling to eliminate the organic impurities and kept at -20 ∘C until sampling. Collected filters were also stored at -20 ∘C until analysis. Organic species and elements were immediately quantified following the field campaign (i.e., within 3 months). Analyses of the water-soluble ions, sugars, OC, and EC were performed a year after the field campaign. Field blank samples were collected by loading filters into the sampler but without operating the sampler's pump. A total number of eight field blank samples were also analyzed with the same techniques as the sample filters. The concentrations of the species in the PM2.5 samples were corrected by subtracting the average field blank values. Additional QA/QC (quality assurance/quality control) procedures were applied on the different analysis techniques/protocols used (e.g., determination of detection limits and recovery, as well as validation using certified reference material) (Fakhri et al., 2023b). PM2.5 mass concentration was determined by weighing the filters before and after sampling using a VWR microbalance with a 1 µg readability (Abdallah et al., 2018; Fadel et al., 2022b).

The chemical analyses are detailed in Fakhri et al. (2023b) and will only be briefly presented here. Elemental carbon (EC) and organic carbon (OC) were analyzed using the thermo-optical transmission method on a Sunset Laboratory analyzer following the EUSAAR2 (European Supersites for Atmospheric Aerosol Research) protocol (Cavalli et al., 2010).

Major (Al, Fe, K, Mg, Na) and trace (Zn, Ni, Sr, Cu, Ti, Co, Cr, V, Mn, Cd, Mo, Sb, Pb) metals and metalloids were analyzed using an inductively coupled plasma mass spectrometer (ICP-MS) (PerkinElmer NexION 300x). In brief, a punch (18 mm2) was taken from the quartz filters and digested in a mixture of ultrapure HNO3 (4 mL of 67 %–70 % (w/w), Baker ARISTAR ULTRA supplied by VWR International), and HCl (1 mL of 35 %–38 % (w/w), TraceMetal Grade, Fisher Chemical) in a microwave reaction system (CEM MARSXpress MARS 230/60). The oven program was set to an initial temperature ramp reaching 180 ∘C in 5.5 min with a subsequent holding period of 9.5 min. Matrix-matched ICP-MS calibration standards were prepared using IV-ICPMS-71A (Inorganic Ventures), and quality control standards were prepared using QCS-27 (High Purity Standards). Two external calibration methods were used for sample analysis: one for trace metals and metalloids, where the calibration range was between 0.05 and 10 µg L-1, and the other for the analysis of major metals, where the calibration range was between 3 and 500 µg L-1. The calibration curves showed good linearity with R2 greater than 0.999. An internal standard solution comprised of Sc, Y, In, and Bi was analyzed along with the samples, with recovery ranging from 90 %–130 %. The analytical procedure was validated by considering a certified reference material NIST SRM 1648a (urban particulate matter, National Institute of Standards and Technology, United States of America). Data for a given metal were accepted as quantitative if NIST SRM 1648 recovery was 80 %–110 %.

Soluble anions and cations were analyzed by ionic chromatography after extracting a punch (20 mm) of each quartz filter by ultrasonic agitation in 6 mL of MilliQ water (ELGA LabWater Purelab Chorus 1, 18.2 MΩ at 25 ∘C) for 45 min, by means of a Dionex ICS-5000+ instrument. For anion analysis, a Dionex IonPac AS11-HC anion column (2×250 mm), a Dionex IonPac AG11-HC (2×50 mm) pre-column, and a Dionex AERS 500 suppressor were used. Thermo Scientific Dionex KOH with a gradient of 1–45 mM was used as the eluent in the anion analysis with a flow rate of 0.5 mL min-1. For cation analysis, a Dionex IonPac CS12 cation column (2×250 mm), a Dionex IonPac CG12 (2×50 mm) pre-column, and a Dionex CSRS 300 4 mm suppressor were used during the cation analysis. Thermo Scientific Dionex 12 mM methanesulfonic acid with a flow rate of 0.3 mL min-1 was used as eluent in the cation analysis.

Sugar alcohols were analyzed by ion chromatography with pulsed amperometric detection (IC-PAD) by means of a Dionex ICS-5000+ instrument using a Dionex CarboPac MA1 column (4×250 mm) and a Dionex CarboPac MA1 (4×50 mm) pre-column. Thermo Scientific Dionex 0.35 M NaOH with a flow rate of 0.6 mL min-1 was used as eluent in the sugar analysis.

The method used for the organic compound analysis was described elsewhere (Fadel et al., 2021; Fakhri et al., 2023b; El Haddad et al., 2011). Briefly, a punch of the sample filter (73 cm2) was spiked with two internal standards (D50-tetracosane and D6-cholesterol obtained from Sigma Aldrich) followed by an extraction using an accelerated pressurized solvent extraction device at 100 ∘C and 100 bar (Dionex ASE 350) and an acetone / dichloromethane mixture (1/1, v/v). After extraction, samples were concentrated to a volume of 200 µL using a constant gentle flow of nitrogen gas. A total of 50 µL of the extracts was then derivatized with the addition of 50 µL of N,O-bis(trimethylsilyl)trifluoroacetamide (BSTFA) (99 %, Sigma Aldrich) with 10 µL of pyridine (catalyst) (99 %, Sigma Aldrich) at 70 ∘C for 2 h. Aliquots of 2 µL of the derivatized extracts were immediately analyzed using gas chromatography–mass spectrometry (GC-MS) in split mode. In addition, aliquots of 2 µL of the non-derivatized fraction were also analyzed using the GC-MS in split mode. The gas chromatograph (Agilent 6890N) was equipped with a HP-5MS UI fused silica capillary column (5 %-phenyl, 95 %-methylpolysiloxane, 0.25 µm film thickness, and 30 m × 0.25 mm) and interfaced to an ion trap MS with an external electron ionization (EI) source (200 ∘C, 70 eV). Full scan mode was used in the mass range of 50–550 m/z.

The term “chemical mass closure” refers to the reconstruction of the measured weighed mass using just the chemical composition. It is done by comparing the combined masses of the chemical species to the gravimetric particulate matter mass (mgrav), wherein the reconstructed PM2.5 mass (mchem) is defined as the sum of organic matter (OM), EC, crustal matter, sea salt, secondary inorganic aerosol (SIA), and other elements that are not taken into account as minerals (Chow et al., 2015).

A chemical mass closure study was performed using the chemical composition measurements to estimate the contributions of the different components to the total PM2.5 mass concentration following the method reported by Fakhri et al. (2023b). Briefly, the contribution of sea salt is calculated by summing the six major ions (Sciare et al., 2005): 1Seasalt=Na++Cl-+ss-Mg2++ss-K++ss-Ca2++ss-SO42-. Ionic constituents such as K+, Ca2+, Mg2+, and SO42- are derived from both marine and non-marine sources. Therefore, it is necessary to discriminate sea salt (ss) from non-sea-salt (nss) contributions. Assuming that all sodium ions are of marine origin, the sea salt contribution can be calculated based on sea water composition as shown in Eqs. (2)–(5) (Genga et al., 2017; Sciare et al., 2005). Furthermore, non-sea-salt potassium, calcium, magnesium, and sulfate (nss-K+, nss-Ca2+, nss-Mg2+, and nss-SO42-) are calculated by subtracting the sea salt fraction (ss-K+, ss-Ca2+, ss-Mg2+, and ss-SO42-, respectively) from the total concentration of the ions (K+, Ca2+, Mg2+, and SO42-, respectively). 2ss-SO42-=0.252×Na+3ss-Ca2+=0.038×Na+4ss-K+=0.036×Na+5ss-Mg2+=0.119×Na+ In the chemical mass closure calculation, we assume that Na+ originates from sea salt, and the ratios between Na+ and other ions (SO42-, Ca2+, K+, and Mg2+) in sea salt aerosol are the same as those for seawater. However, it is possible that some Na+ originates from road salt, which is principally composed of NaCl, with a small amount of CaCl2 (Charbonneau, 2006). In this case, the contributions of SO42-, K+, and Mg2+ would be overestimated, and the true concentration of the road/sea salt component would be less than that calculated. In the extreme case of the component being derived entirely from road salt, the overestimation in the concentration of this component would be approximately 20 %, given the preceding equations. This error is relatively small because SO42-, K+, and Mg2+ have relatively small concentrations in sea salt.

In addition, secondary inorganic aerosol (SIA) is represented by the sum of nss-SO42-, NH4+, and NO3-. To take bound water into account a hydration multiplication factor of 1.29 was applied to convert the dry inorganic concentrations (SIA and sea salt) into hydrated species (Sciare et al., 2005; Genga et al., 2017).

The contribution of crustal matter (CM) (Eq.  6) was estimated by summing the concentrations of aluminum, silicon, calcium, iron, and titanium in their oxide forms (Huang et al., 2014). The coefficients in front of the elements correspond to the additional mass due to oxygen in the minerals. Silicon was not measured in this study and was indirectly determined by multiplying the measured aluminum concentration by a factor of 3.41 (Esmaeilirad et al., 2020). This factor is obtained from the ratio of Si and Al in the Earth's crust following Mason and Moore (1982). 6CM=2.2Al+2.49Si+1.63Ca+2.42Fe+1.94[Ti] To account for unmeasured O, N, S, and H atoms in OM, the conversion factor (CF) from OC to OM was derived using the equation OM = CF × OC. The method used to calculate the CF sums all the PM components while systematically varying the OM/OC conversion (Genga et al., 2017). To find the optimal CF to calculate OM from OC, the factor was varied from 1.2 to 2.1. The Pearson correlation (R) calculated between the reconstructed PM2.5 and the measured mass did not change significantly (0.978–0.979), but the highest correlation and the slope closest to 1 was obtained with CF = 1.6. The results of the chemical mass closure study are shown in Fig. S5.

The United States Environmental Protection Agency (USEPA) recommends using a health risk assessment model to determine the health risks from airborne elements. For each of the three exposure pathways, namely inhalation, ingestion, and dermal contact, the carcinogenic and non-carcinogenic health risks were evaluated for children and adults using the measurements taken at the MTL site. This study analyzes the non-carcinogenic risks of Al, Fe, Cu, Zn, Sb, Co, Cr(VI), Ni, V, Cd, Pb, and Mn as well as the carcinogenic risk of Cr(VI), Co, Ni, V, Cd, and Pb. The average daily dose (ADD in mg kg-1 d-1) for children (0–17 years) and adults (18–70 years) for the three exposure pathways and the exposure concentration through inhalation (ECinhalation in mg m-3) were calculated following Eqs. (7)–(10) (Fadel et al., 2022a, b; Roy et al., 2019). Each of the parameters used in the various formulas are listed in Table 1 and were retrieved from USEPA reports (USEPA, 2004, 2011). Moreover, since chromium toxicity is attributed to its hexavalent state, Cr(VI) concentration was determined as one-seventh of total Cr (Dahmardeh Behrooz et al., 2021; Hao et al., 2020; Fadel et al., 2022b). 7ADDingestion=CxIngRxEXPxEDxCFBWxAT8ADDdermal=CxSAxAFxABSxEXPxEDxCFBWxAT9ADDinhalation=CxInhRxEXPxEDxCFBWxAT10ECinhalation=CxEXPxEDxETxCFATx24

The hazard quotient (HQ), which is determined by dividing the ADD from each exposure pathway by a specified reference dose (RfD) (mg kg-1 d-1) for the same exposure route, can be used to measure the non-carcinogenic health risk effects from metals as presented in Eq. (11) (Fadel et al., 2022a, b; Hao et al., 2020). The non-cancer risk refers to the likelihood of developing health issues other than cancer as a result of exposure to chemical pollutants such as asthma, nervous system disorders, and cardiovascular and respiratory disorders (Fadel et al., 2022a).

To assess the overall potential for non-carcinogenic consequences produced by multi-element exposure for one exposure pathway, the hazard index (HIi) was calculated as the sum of the HQij (i is the exposure pathway which is either inhalation, ingestion, or dermal contact, and j is the targeted compound) (Eq. 12). HItotal represents the total hazard index (HItotal=HIinhalation+HIingestion+HIdermal). An HI value higher than 1 implies that the non-cancer risk merits attention and indicates that adverse health effects are likely to occur (Dahmardeh Behrooz et al., 2021). 11HQi=ADDiRfDi12HIi=∑HQij The carcinogenic risks (CRs) are estimated using Eqs. (13)–(15), which refers to the probability that a person might develop a cancer over a lifetime because of exposure to a carcinogenic chemical (Dahmardeh Behrooz et al., 2021; Fadel et al., 2022a, b). According to the USEPA, a cancer risk value between 10-6 (1 additional case per 1×106 people) and 10-4 (1 in 10 000) indicates that the carcinogenic risk is considered tolerable, while a value higher than 10-4 indicates that a serious risk of cancer exists (Bari and Kindzierski, 2016; Dahmardeh Behrooz et al., 2021). CRtotal represents the total carcinogenic risk (CRtotal=CRingestion+CRinhalation+CRdermal). 13CRingestion=ADDingestion×CSF14CRinhalation=ECinhalation×IUR15CRdermal=ADDdermal×CSF In the equations above, CSF is the cancer slope factor for a chemical in a specific exposure pathway (mg kg-1 d-1) and IUR is the inhalation unit risk (m3 mg-1) (Table S1 in the Supplement).

The USEPA PMF v5.0 software was used to identify and quantify the major emission sources in Montréal. PMF is a multivariate factor analysis tool that decomposes a data matrix X (n×m) into two matrices: source contributions G (n×p) and source profiles F (p×m), where n is the number of samples, m is the number of species, and p is the number of factors or sources. The PMF model requires the concentration dataset of the samples and associated uncertainty as inputs. The goal is to solve the chemical mass balance (Eq. 16) between the measured species concentrations and source profiles: 16xij=∑k=1pgikfkj+eij, where xij is the concentration of the species j in the ith sample, gik is the contribution of the kth source in the ith sample, fkj is the relative concentration of species j from the source k, and eij is the residual of species j in the ith sample. The values of gik and fkj are adjusted until a minimum value of Q (Eq. 17) for a given number of factors p is found: 17Q=∑i=1n∑j=1meijuij2, where e is the residual value and u is the uncertainty in a measurement. The residual value is the difference between the measured value and the PMF-modeled concentration of each compound. In the present work, samples below the detection limit (DL) were replaced by half of the DL and were given an uncertainty of 5/6 times the detection limit (Polissar et al., 1998). Missing samples were replaced by the median value of that species and were given an uncertainty of 4 times the median value (Polissar et al., 1998). When the concentration was greater than the DL, the uncertainty was calculated according to the USEPA guidelines (USEPA, 2014; Lee et al., 2022; Park et al., 2019): ((concentration×0.1)2+(0.5×DL)2). After screening the integrity of the input data, 27 species were included in the PMF model. The overall number of samples (80 samples) and the number of species complies with the ratio of at least 3:1, as proposed by Belis et al. (2019). The species included were OC and EC, major water-soluble ions (Na+, Cl-, NH4+, NO3-, and SO42-), and a selection of elements (Al, Fe, Ti, Cu, Sb, Cd, and Co). Levoglucosan was included as a tracer for biomass burning, 17α[H]-21β[H]-Hopane as a tracer for vehicular emissions, fatty acids (hexadecanoic acid and octadecanoic acid) for cooking activities, a set of n-alkanes for biogenic (C27, C29) and anthropogenic emissions (C20, C21, C24, C25), and finally a dicarboxylic acid (oxalic acid) and α-pinene oxidation products (pinic acid and cis-pinonic acid) as tracers for SOA.

All the included species were defined from weak to strong in the PMF model based on their signal-to-noise ratio (S/N). The PM species were classified as “bad” when the S/N ratio was less than 0.2, “weak” when the S/N ratio was between 0.2 and 2, and “strong” when the S/N ratio was greater than 2 (Esmaeilirad et al., 2020). The bad species are excluded from the analysis, while the uncertainty for the weak species is tripled. PM2.5 was designated as a “total variable” and was automatically classified as “weak”. All the included species were successfully modeled by PMF with their concentrations reconstructed accurately and were qualified as “strong” except for nitrate, which presented a S/N ratio of 0.9 and was defined as “weak”.

The final solution was selected based on several criteria such as (1) comparison of the resulting source profiles against the literature, (2) lack of correlation between the resolved factors, (3) correlation between the predicted vs. measured PM2.5 concentrations, (4) correlation between the modeled and measured species concentrations (R2 higher than 0.8), and (5) maximum individual mean (IM) and maximum individual standard deviation (IS) (Fig. S1). The R2 between the reconstructed and measured PM2.5 mass was 0.87 (slope = 0.90) (Fig. S2). The robustness of the PMF solution was tested by the two-error estimation method (bootstrap and displacement) as instructed in the PMF manual to ensure the solution was stable (Table S2) (USEPA, 2014).

Simulations were performed using the GEOS-Chem chemical transport model (version 14.0.1, doi:10.5281/zenodo.7271960) (Bey et al., 2001; Park et al., 2004). GEOS-Chem is driven by assimilated meteorology from the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2), at the NASA Global Modeling and Assimilation Office (GMAO). The atmosphere was resolved using 47 vertical layers from the surface to 0.01 hPa. The vertical resolution of the model is about 100 m near the surface, but it becomes coarser at higher altitudes. Boundary conditions were generated using a global simulation at 2∘ latitude × 2.5∘ longitude resolution. A nested grid is then used with 0.5∘ latitude × 0.625∘ longitude resolution, spanning 35 to 65∘ N, 90 to 50∘ W (Fig. S3) in order to include the full province of Quebec as well as the strong source regions of the Great Lakes region and the northeastern US. We use the recently developed treatment of wet scavenging described by Luo et al. (2020, 2019), which has been previously shown to yield better agreement for nitrate and ammonium concentrations over eastern North America.

Biomass burning emissions were simulated from the Copernicus Atmosphere Monitoring Service Global Fire Assimilation System (GFAS, Kaiser et al., 2012). We used global emissions from the Community Emissions Data System version 2 (CEDS v2, O'Rourke et al., 2021; Hoesly et al., 2018), except where overwritten by regional emissions inventories, as described in Keller et al. (2014). These regional emissions inventories included anthropogenic emissions from the US and most of Canada up to 54∘ N, provided by the National Emissions Inventory for 2016 (EPA, 2021a), with annual scaling factors to account for changes in emissions since 2016 (EPA, 2021b). Shipping emissions were provided by CEDS v2, and aircraft emissions were provided by the Aviation Emissions Inventory Code (AEIC, Stettler et al., 2011; Simone et al., 2013). Anthropogenic emissions of fine dust aerosols were from the Anthropogenic Fugitive, Combustion, and Industrial Dust (AFCID) inventory (Philip et al., 2017), while natural dust emissions were calculated according to the Mineral Dust Entrainment and Deposition (DEAD) parameterization (Zender, 2003). Formation of SOA was parameterized using the “simple” SOA scheme that treats all organic aerosol as non-volatile, as described in Pai et al. (2020). It has been shown to reproduce observed organic aerosol concentrations with similar skill to a more complex scheme. GEOS-Chem resolves mineral dust in four size bins spanning radii of 0.1–1.0, 1.0–1.8, 1.8–3.0, and 3.0–6.0 µm. In this study, the concentration of dust in particles smaller than 2.5 µm in diameter is estimated by adding 38 % of the concentration of dust in the 1.0–1.8 µm size bin to the concentration of dust in the 0.1–1.0 µm size bin (Fairlie et al., 2010; Zhang et al., 2013).

GEOS-Chem has previously been evaluated against NO2 concentrations from the USEPA Air Quality System sites (Silvern et al., 2019); satellite observations of aerosol optical depth, speciated aerosol concentrations from aircraft measurements over the United States (US), speciated surface observations from the USEPA Chemical Speciation Network (CSN), Interagency Monitoring of Protected Visual Environments (IMPROVE), and the Southeastern Aerosol Research and Characterization (SEARCH) network (Kim et al., 2015); and ozone concentrations from the Clean Air Status and Trends Network (CASTNET, Reidmiller et al., 2009). In all cases, GEOS-Chem has been shown to adequately resolve emissions, atmospheric processes, and large-scale transport. The model performance was evaluated against the observations presented in this study, and good correlation with the observed values was obtained (R=0.24–0.76) (Table S3). Some of the indicators are not within the acceptable limits (Abdallah et al., 2018), and this is likely because of the vertical and horizontal resolution of the model. Thus, the PMF results and the model outputs will be compared qualitatively. Furthermore, GEOS-Chem results have been previously used for source contribution analysis similar to the analysis presented in this study (Meng et al., 2019).

In order to examine the sensitivity of air pollutant concentrations to contributions from source regions, a base-case simulation is performed along with three sensitivity simulations, each with anthropogenic emissions from a geographic region turned off: noQC, noCA, and noUS. In the noQC simulation, anthropogenic emissions within the borders of the province of Quebec are not allowed. In the noCA simulation, anthropogenic emissions within the borders of Canada are not allowed, except for emissions from the province of Quebec (rest of Canada, RoC). In the noUS simulation, anthropogenic emissions within the borders of the contiguous US are not allowed. Emissions from shipping and aircraft are removed within the boundaries of the specified region, but emissions from biomass burning or natural sources are not changed. We note that for the purposes of masking emissions, each model grid cell is considered to be entirely within one province or country; the resolution of the provincial or national masks is the same as the model resolution. By calculating the differences in the concentrations of PM components in the sensitivity simulations compared to the base-case simulation, we qualitatively evaluate the proportions of air pollutants in Quebec due to sources within Quebec, sources in the RoC, and sources in the contiguous US.

The average PM2.5 concentration (and standard deviation) at the MTL site was 4±3 µg m-3. The concentration of PM2.5 in all samples collected at the MTL site was lower than the daily standard set by the WHO (15 µg m-3) (WHO, 2021) and the Canadian daily standard of 27 µg m-3 (EAR, 2019) (Fig. 1). The PM2.5 levels at the MTL site can be compared against nearby government monitoring stations to understand if there are large differences in concentrations and how concentrations have changed over recent years in this area of Montréal. The average concentrations at the MTL site were lower than that reported during the same sampling period at the Décarie station (6.6 µg m-3) located ∼ 5 km southwest of the MTL site near an intersection of two major highways (Fig. 2). Given that the Décarie station is strongly impacted by vehicle emissions, this difference is not surprising. In contrast, the average PM2.5 concentration at the MTL site is similar to that recorded at the Molson station (NAPS ID: S50134; ∼ 4 km northeast of MTL) (4.7 µg m-3), which is located in a mixed residential and industrial zone for the same sampling period. Furthermore, there is a general decreasing trend between 2017 and 2020 in PM2.5 levels at the government monitoring stations. Since the concentration of PM2.5 in 2020 was not too different in comparison with the previous years (2018 and 2019), the sources of PM2.5 identified in this study are likely to be similar to other years. It is important to mention that during our sampling period, Montréal was in partial lockdown, and public spaces (e.g., bars, gyms, cinemas, museums, libraries, and casinos) were closed due to the possibility of a second wave of the COVID-19 pandemic. Primary schools and some secondary schools were opened during that period. While these considerations suggest that the results presented here are also applicable to pre- and post-pandemic conditions, further studies are needed before generalizing the results of this study to other periods.

The average OC and EC concentrations (± standard deviation) were 1.31±0.76 µg m-3 and 0.17±0.15 µg m-3, respectively (Table 2). While EC originates from combustion sources and has a primary origin, OC can be directly emitted from combustion of fossil fuels, biomass burning, and cooking activities, and it can also be formed in the atmosphere through chemical reactions (Hallquist et al., 2009). The OC/EC ratio is a useful diagnostic ratio that provides information on the sources in PM2.5. An OC/EC ratio ranging between 0.3 and 1 was reported for diesel vehicles, a ratio between 1.4 and 5 was reported for gasoline operated vehicles, and a larger OC/EC ratio ranging between 4.1 and 14.5 was reported for biomass burning (Khan et al., 2021; Salameh et al., 2015). In this study, the ratio of OC/EC was 7.4±3.1, which is larger than those associated with vehicular emissions and consistent with the presence of biomass burning. Indeed, wood burning may be an important source of air pollutants in Quebec, and wood-burning stoves and fireplaces are used in both residential and businesses settings (e.g., pizzerias and bagel factories). The production of secondary organic carbon will also increase OC/EC ratios beyond values measured for primary sources. Indirect methods have been developed for identifying primary organic carbon (POC) or secondary organic carbon (SOC) (Shivani et al., 2019; Calvo et al., 2008; Joseph et al., 2012). 18SOC=OCtotal-EC×OCECmin One such method is summarized in Eq. (18), OCtotal is the measured OC and (OC/EC)min is the minimum ratio observed in the samples. Additional information on the calculation method is included in the Supplement. Using this method, the average contribution of SOC to the total OC and the standard deviation was 61±15 % at the MTL site, indicating a strong SOC contribution.

Some organic compounds are specific to a certain source and have been used in source apportionment studies as molecular markers. The average concentrations of selected organic compounds in PM2.5 measured at the MTL site are summarized in Table 3. Levoglucosan, generated by the pyrolysis of cellulose and often utilized as a particular marker of biomass burning (Simoneit, 2002), was the most abundant compound among the organic tracers examined. Due to its presence in lubrication oil used in both gasoline and diesel vehicles, 17α(H)-21β(H)-Hopane is a specific marker of traffic emissions, and it was the least prevalent measured organic species in the PM2.5 (El Haddad et al., 2009; Fadel et al., 2021). Arabitol and mannitol were well correlated (R=0.98; p<0.001) and are commonly described as markers for primary biogenic emissions, more specifically with fungal spores (Petit et al., 2019). Alkanes, specifically acyclic saturated hydrocarbons, can originate from both biogenic and anthropogenic sources. The n-alkane with the highest concentration was C29. In the literature, low-molecular-weight n-alkanes (< C27) are primarily associated with traffic emissions, whereas high-molecular-weight n-alkanes (C27, C29, C31) are associated with plant detritus because they are abundant in the epicuticular wax of plants (Rogge et al., 1993a, b).

To further evaluate the contribution of the anthropogenic and the biogenic sources to n-alkanes, the carbon preference index (CPI) was calculated (Fadel et al., 2021; Esmaeilirad et al., 2020). Two CPI parameters were adopted: the “overall CPI” for the whole range of n-alkanes (C15–C30) and the “high CPI” for the higher-molecular-weight n-alkanes (C25–C30). The detailed description of the calculation method is included in the Supplement. The average “overall CPI” value was 0.86±0.21, indicating the contribution of petrogenic sources. The “high CPI” values during the entire campaign were between 0.74 and 2.81, with a majority of the measurements in the anthropogenic range and the rest in the mixed anthropogenic–biogenic range (Fig. 3). The average “high CPI” was 1.56±0.48, indicating that larger n-alkanes at the MTL site are predominately anthropogenic with a lesser biogenic contribution.

The relationship between trace metals provides qualitative information on the sources of the measured elements. A significant correlation (R>0.95, p<0.05) was observed among Ti, Fe, Mg, V, K, Mn, and Na, indicating a common crustal source. Numerous studies have analyzed the chemical composition of traffic-related PM and have reported that Cu, Sb, and Cd are linked to non-exhaust emissions, more precisely from vehicular brake wear (Thorpe and Harrison, 2008; Lin et al., 2015; Pio et al., 2013; Mancilla and Mendoza; Pant and Harrison, 2013). In this study, a correlation of R=0.82 (p<0.05) was found between Sb and Cd, indicating that Sb and Cd originate from the same sources, most probably brake wear. Studies in other Canadian cities, namely Toronto and Vancouver, have also associated the elements Sb and Cd with non-exhaust emissions from road traffic (Celo et al., 2021). No correlation was found between Cu and the elements Cd and Sb (R<0.01, p<0.05), indicating that brake-wear debris was not an important source of Cu in Montréal.

Additionally, Co was correlated with Cr (R=0.79, p<0.01) and Cu (R=0.86, p<0.05), suggesting a common source from industrial emissions, similar to observations in Edmonton, Canada (Bari and Kindzierski, 2016). The Canadian Copper Refinery and Suncor Energy refinery can be identified as potential sources of Cr and Co in the Montréal region based on the National Pollutant Release Inventory (NPRI) data published by the Government of Canada (NPRID, 2022). It is important to mention that Cu, Cr, and Co could also originate from other sources such as traffic-related emissions and coal combustion (Riffault et al., 2015; Bari and Kindzierski, 2016; Thorpe and Harrison, 2008; Celo et al., 2021). However, these possibilities seem less important given the correlations between the elements as well as the lack of coal combustion in the province of Quebec. Lastly, no correlation was found between Zn, Pb, and Sb with Cl- (R<0.09, p<0.05), revealing that incinerators are not a potential source of these trace elements (Riffault et al., 2015; Rahn and Huang, 1999).

Among the water-soluble ions, SO42- presented the highest concentration followed by NH4+, NO3-, Ca2+, Na+, Cl-, and K+, and the species with the lowest concentration was Mg2+. The secondary inorganic aerosol components (sulfate, ammonium, and nitrate) accounted for 83 % of the total water-soluble ions and 32 % of the total PM2.5 mass (Fig. S5). The presence of secondary sulfate is supported by the strong correlation between sulfate and ammonium with a Pearson coefficient of 0.90 (p<0.005). This result, and the weak correlation between ammonium and nitrate (R=0.27, p<0.002), suggests that a significant fraction of ammonium in PM2.5 was associated with ammonium sulfate with a limited amount of ammonium nitrate. Nitrate concentrations were higher in the end of October and November in comparison with the warmer months (Fig. S6), consistent with a shift in equilibrium partitioning from gas-phase nitric acid to the particle-phase nitrate due to colder temperatures (Geng et al., 2013; Mantas et al., 2014). The total ammonium/total sulfate (TA/TS) molar ratio was used to evaluate if the site conditions were ammonia-rich (> 2) or ammonia-poor (< 2) (Joseph et al., 2012; Remoundaki et al., 2013). The TA/TS molar ratio was 1.24, highlighting that the concentrations of ammonium are not sufficient to neutralize the concentrations of sulfate and nitrate. Therefore, the latter anions could be associated with different cations such as Na+, Mg2+, K+, and Ca2+ and the formation of salts with these cations (e.g., NaNO3, MgSO4). Moreover, when considering the neutralization of all the cations (Na+, Mg2+, NH4+, K+, Ca2+) by all the anions (SO42-, NO3-, Cl-), the scatter plot (Fig. S7) presented a slope of 0.98, indicating a charge balance between the anions and cations.

The non-carcinogenic health risks from airborne elements through inhalation, ingestion and dermal contact were estimated. The HI exhibited the same trend for both children and adults and followed a decreasing order of Co > Mn > Pb > Cu > Cd > Ni > Cr(VI) > Sb > Zn > Al > Fe > V. Among the three exposure pathways, inhalation contributed the most to the total non-carcinogenic risk (HItotal). Overall, inhalation contributed 98 % for adults and 97 % for children to HItotal, ingestion contributed 1 % (adults) and 2 % (children), and dermal absorption was the remainder. Co was the largest contributor to the HIinhalation, with a contribution of 61 % for both adults and children. An HI value higher than 1 implies that adverse effects other than cancer such as cardiovascular and respiratory diseases are expected (Fadel et al., 2022a). In this study, HI and HItotal (HItotal=0.24 for adults and 0.36 for children) were below the level of 1, highlighting limited non-carcinogenic health hazards from PM2.5 metals.

The carcinogenic risk of each carcinogenic metal (i.e., Cr(VI), Co, Ni, V, Cd, and Pb) from the three exposures pathways was also calculated. Inhalation was the exposure pathway with the highest cancer risk to which 99 % and 98 % of the overall carcinogenic risk (CRtotal) for adults and children was ascribed, respectively. According to the USEPA, a cancer risk value between 10-6 (1 additional case per 1×106 people) and 10-4 (1 in 10 000) indicates that the carcinogenic risk is considered tolerable, while a value higher than 10-4 indicates that a serious risk of cancer exists (Bari and Kindzierski, 2016; Dahmardeh Behrooz et al., 2021).

In this study, the carcinogenic risk of V, Ni, Cd, and Pb was between 1.18×10-8 and 6.11×10-7, which is lower than 10-6. However, Co and Cr(VI) presented a cancer risk higher than 10-6 (Fig. 7), highlighting that more attention should be given to these trace metals. Based on PMF analysis, Co was associated with industrial emissions, and the correlation between Co and Cr (R=0.79, p<0.01) (Sect. 3.4) suggests a common source from industrial emissions for these two trace elements (Cr was not added in the PMF analysis because it was not well modeled). The sum of the risk levels posed by the six metals was 41.77×10-5 and 5.87×10-6 for adults and children, respectively, which is between the range of 10-6–10-4, indicating that the carcinogenic risk is considered tolerable.

These results show that even though industrial emissions presented a very small contribution in term of mass (0.26 µg m-3), the health risks associated with this source cannot be fully determined by the PM2.5 mass concentration alone, and the trace elements emitted from industrial sources source were found to have a potential health risk. Thus, this study highlights that mitigation strategies should also prioritize reduction in metals in addition to PM.