Analytically unresolved hydrocarbon signature

As previously reported (Liggio et al., 2016), the total ion chromatogram of the GC-ITMS occasionally showed elevated and analytically unresolved hydrocarbons in the volatility range of C₁₁–C₁₇ with a saturation vapor concentration (C^*) of 10⁵ µg m${}^{-\mathrm{3}}<C^{*}<{\mathrm{10}}^{\mathrm{7}}$ µg m⁻³. An example is shown in Fig. 2.

Figure 2The top shows total ion chromatograms of air samples collected on 27 August 2013 from 18:04 to 18:14 UTC (red) and on 28 August 2013 from 13:43 to 13:53 UTC (blue). The total ion counts (TICs) of a headspace sample of ground-up bitumen collected after the campaign is superimposed (black). The gray area indicates the range over which IVOC signal was integrated. The bottom shows retention times of n-alkanes, determined after the measurement campaign by sampling a VOC mixture containing a C₁₀–C₁₆ n-alkane ladder.

Download

An offline analysis of the headspace above ground-up bitumen gave a similarly unresolved hydrocarbon signal (Fig. 2, black trace). In this particular case, the ambient air chromatogram also shows enhancements of lower-molecular-weight hydrocarbons (possibly from naphtha) that were not observed in the bitumen sample. The observed unresolved hydrocarbon feature is qualitatively similar to the “large chromatographic hump of unresolved complex mixtures” reported by Yang et al. (2011) during their analysis of bitumen extracts.

The major ions contributing to the unresolved signals in Fig. 2 are associated with alkanes (i.e., m∕z 55, 57, 67, 69; see Fig. S1 in the Supplement). In contrast, counts at masses associated with aromatics (i.e., m∕z 115, ${\mathrm{C}}_{\mathrm{9}}{\mathrm{H}}_{\mathrm{7}}^{+}$, and m∕z 91, ${\mathrm{C}}_{\mathrm{7}}{\mathrm{H}}_{\mathrm{7}}^{+}$) as reported by Cross et al. (2013) were negligible in both the bitumen headspace and polluted day samples. The resemblance of the unresolved hydrocarbon feature in ambient air with the bitumen headspace sample in terms of both volatility (i.e., elution time) and electron impact mass fragmentation is consistent with bitumen as the source of IVOCs at this site.

In the interpretation of the integrated IVOC signal, it is assumed that it is of primary origin, i.e., emitted directly from point sources in the vicinity of the measurement site. For the PCA, the unresolved signal was integrated from a retention time of 25 to 45 min (gray area in Fig. 2) in all ambient air chromatograms.

The IVOCs observed in this work likely encompass a portion of the total that is emitted. For example, IVOCs generated by combustion processes, such as diesel engine exhaust, are comprised of aliphatic alkanes, including cyclic and branched alkanes, and aromatics (Gentner et al., 2012; Zhao et al., 2015). The use of a chromatographic column in this work biases the IVOC signal towards hydrocarbon IVOCs since oxygenated compounds (i.e., alcohols and acids) will not elute from the analytical column. Furthermore, the recovery of VOCs from the pre-concentration unit, while reproducible and likely complete for n-alkanes, which bracket the bulk of IVOC emitted and whose calibration curves were linear, is not known for late-eluting compounds but is assumed to be sufficiently reproducible to yield a semiquantitative signal.

2.3.1 Selection of variables

A total of 22 variables whose ambient concentrations are dominated by primary emissions or which are formed very shortly after emission (such as the less-oxidized oxygenated organic aerosol (LO-OOA) factor observed by the SP-AMS; see below) were included in the PCA (Table 3). These variables included CO₂, CH₄, NO_y, CO, and SO₂, which are known to be emitted in the oil sands region from stacks, the mine fleet and faces, tailing ponds, and fugitive emissions (Percy, 2013). The median ratio of NO_x (= NO+NO₂) to NO_y was 0.85, consistent with the close proximity of the measurement site to emission sources and limited chemical processing. Because NO_x constituted a large fraction of NO_y, its temporal variation was captured by the latter, and it was not included as a separate variable in the PCA.

For this work, mixing ratios of all non-methane hydrocarbons (NMHCs) that were quantified (i.e., o-xylene, the n-alkanes decane and undecane, the aromatics 1,2,3- and 1,2,4-trimethylbenzene (TMB), and limonene and α- and β-pinene) were included as variables. In addition, the aforementioned unresolved signal associated with IVOCs was included as a variable by integrating total GC-ITMS ion counts (m∕z 50–425) over a retention time range of 25–45 min (retention index range of 1100 to 1700).

Gas-phase ammonia was included as a variable because elevated reduced nitrogen concentrations have been observed in the region and were linked to the use of ammonia on an industrial scale, for example as a floating agent and for hydrotreating (Bytnerowicz et al., 2010). Total sulfur and total reduced sulfur were added as tracers of upgrader stack SO₂ emissions and of “odors”, believed to be emitted from oil sands tailing ponds, which continue to be of concern in surrounding communities (Small et al., 2015; Percy, 2013; Holowenko et al., 2000).

Refractory black carbon was added as a variable since it is present in diesel truck exhaust and in biomass burning plumes and is hence a combustion tracer (Wang et al., 2016; Briggs and Long, 2016). Particle-surface-bound polycyclic aromatic hydrocarbons (pPAHs) were included because of their association with facility stack emissions and combustion particles in the area (Allen, 2008; Grimmer et al., 1987). Hydrocarbon-like organic aerosol (HOA) was included as a surrogate for fossil fuel combustion by vehicles (Jimenez et al., 2009). The LO-OOA factor was included as it appears to form rapidly after emission of precursors (Lee et al., 2018). Supermicron aerosol volume (PM₁₀₋₁, i.e., the volume of particles between PM₁₀ and PM₁) was also included as a tracer of coarse particles from primary sources, which are expected to be dominated by dust emissions.

Table 3Variables observed at the AMS 13 ground site during the 2013 JOSM campaign used for PCA.

^a Values were determined only from data points included in the PCA, not from the entire campaign. ^b Average and relative standard deviation were calculated before zeros were replaced with 0.5 × LOD. ^c RSD: relative standard deviation. ^d LOD: limit of detection. ^e ppt: parts per trillion by volume (10⁻¹²). ^f NA: data not available. ^g Calculated using 3 × standard deviation at ambient background levels.

Download Print Version | Download XLSX

To assess which components impact secondary product formation, a second PCA was performed that included variables mainly formed through atmospheric chemical processes and whose concentrations more strongly depend on air mass chemical age than those variables selected initially. In this PCA, odd oxygen (${\mathrm{O}}_x={\mathrm{O}}_{\mathrm{3}}+{\mathrm{NO}}_{\mathrm{2}}$), submicron aerosol ${\mathrm{SO}}_{\mathrm{4}\left(\mathrm{p}\right)}^{\mathrm{2}-}$, ${\mathrm{NO}}_{\mathrm{3}\left(\mathrm{p}\right)}^{-}$, ${\mathrm{NH}}_{\mathrm{4}\left(\mathrm{p}\right)}^{+}$, a second, more-oxidized OOA factor (MO-OOA), and PM₁ volume were included, increasing the total number of variables to 28 (Table 4). Furthermore, since oxidation of IVOCs leads to formation of SOA (Robinson et al., 2007; Lee et al., 2018), and the photochemical conversion of IVOC to SOA may adversely affect the PCA, a PCA without secondary and aerosol variables is presented in the Supplement (Table S10).

2.3.2 Treatment of input data

Data used in the PCA were averaged to match the time resolution of the GC-ITMS VOC and IVOC measurements, i.e., over 10 min long periods (spaced ∼1 h apart) set by the start and stop times of the GC-ITMS pre-concentration period. When concentrations were below their respective limit of detection (LOD; values are given in Table 3), half the reported LOD was used to minimize bias (Harrison et al., 1996; Polissar et al., 1998; Zhao et al., 2004; Guo et al., 2004). Prior to PCA, input variables were standardized to eliminate unit differences by subtracting the mean concentration ${\stackrel{\mathrm{‾}}{C}}_i$ of pollutant i from the concentration of sample k (C_i,k) and dividing by the standard deviation (s_i) of all samples included in the PCA.

Here, Z_i,k is the standardized pollutant concentration. In total, 218 data points from all identified species over the period of the campaign were used for the main PCA.

Table 4Variables added in the second PCA. Particle-phase concentrations, i.e., ${\mathrm{SO}}_{\mathrm{4}\left(\mathrm{p}\right)}^{\mathrm{2}-}$, ${\mathrm{NO}}_{\mathrm{3}\left(\mathrm{p}\right)}^{-}$, ${\mathrm{NH}}_{\mathrm{4}\left(\mathrm{p}\right)}^{+}$, and MO-OOA, were measured by aerosol mass spectrometry and account for PM₁ only.

Download Print Version | Download XLSX

2.3.3 PCA solutions

In this work, the varimax method (Kaiser, 1958) was used to rotate the loading matrix. This method is an orthogonal rotation (i.e., components are not expected to correlate), which minimizes the impact of high loadings, making the results easier to interpret (Kaiser, 1958). Several criteria (Table S10) were considered for component selection: the latent root criterion, i.e., on the basis that rotated eigenvalues must be greater than unity, the (cumulative) percentage of variance criterion, for which the extracted components account for >95 % of the variance, and the scree test (Fig. S2) (Thurston and Spengler, 1985; Guo et al., 2004; Hair et al., 1998; Cattell, 1966). For the optimal solution presented in the main paper, the 95 % variance criterion was chosen, providing a 10-component solution for the PCA with only primary variables and an 11-component solution for the PCA with both primary and secondary variables. Components 1 through 4 were consistent regardless of the number of components retained. Solutions with fewer and more components are presented in the Supplement.

Time series of each of the components were calculated by multiplying the original standardized matrix by the rotated loading matrix and were used to generate bivariate polar plots (Sect. 2.4).

3.2.1 PCA with primary variables

The loadings of the optimum solution are presented in Table 5. The 10-component solution accounts for a cumulative variance of 95.5 %. The communalities for the analysis, i.e., the fraction of total pollutant observations accounted for by the PCA, are all greater than 85 %, with the lowest communality obtained for the IVOCs (0.86).

In the following, an overview of the observed components is presented. Associations with r>0.7, r>0.3, and r>0.2 are referred to as “strong”, “weak”, and “poor”, respectively. Hypothesized identifications are given in Sect. 4 and are summarized in Table 6 and Fig. 4.

The component accounting for most of the variance of the data, component 1, is strongly associated with the anthropogenic VOCs (r>0.87), weakly associated with CH₄ (r=0.59), TRS (r=0.59), HOA (r=0.40), LO-OOA (r=0.45), CO (r=0.41), and the IVOCs (r=0.31), and poorly associated with NO_y (r=0.27) and rBC (r=0.30). Component 2 is strongly associated with the combustion tracers NO_y (r=0.82), rBC (r=0.77), HOA (r=0.74), and pPAH (r=0.94), weakly associated with CH₄ (r=0.39) and IVOCs (r=0.39), and poorly associated with ammonia (r=0.20), and undecane and decane (r=0.27 and 0.22, respectively). Component 3 is strongly associated (r>0.9) with the biogenic VOCs and weakly associated with CO₂ (r=0.48) and shows poor negative correlations with NO_y ($r=-\mathrm{0.26}$) and ammonia ($r=-\mathrm{0.24}$). Component 4 is strongly associated with SO₂ and TS (r=0.97 and 0.93, respectively) and poorly associated with NO_y (r=0.21) and LO-OOA (r=0.28).

Components 1 through 4 emerged regardless of the number of components used to represent the data, whereas the structure of components 5 through 10 only fully emerged in the 10-component solution (see Supplement). Hence, components 6 through 10 are somewhat tentative as many (i.e., 7–9) are single-variable components and have eigenvalues close to or below unity, i.e., account for less variance than any single variable. As a result, the interpretations of these components are subject to more uncertainty and are more speculative but are presented in the Supplement for the sake of completeness and transparency. For the purpose of this paper, this is inconsequential as components 6–10 are not associated with IVOCs.

Table 5Loadings for the 10-factor optimal solution (primary variables only). Coefficients with Pearson correlation coefficients r>0.3 are shown in bold font.

Download Print Version | Download XLSX

Table 6Hypothesized identifications of principal components.

Download Print Version | Download XLSX

3.2.2 Extended PCA with added secondary variables

The loadings of the optimum solution that includes primary and secondary variables are shown in Table 7. In this 11-component solution, the 10 components originally identified were preserved, though their relative order was changed, with the upgrader component moving from the fourth to second position. There was one new component (6), which encompassed only secondary species, including MO-OOA (r=0.92), O_x (r=0.33), ${\mathrm{NO}}_{\mathrm{3}\left(\mathrm{p}\right)}^{-}$ (r=0.36), PM₁ (r=0.31), and LO-OOA (r=0.31).

${\mathrm{NH}}_{\mathrm{4}\left(\mathrm{p}\right)}^{+}$, ${\mathrm{SO}}_{\mathrm{4}\left(\mathrm{p}\right)}^{\mathrm{2}-}$, and ${\mathrm{NO}}_{\mathrm{3}\left(\mathrm{p}\right)}^{-}$ are associated with the stack emissions component (2, with r=0.84, 0.84, and 0.44, respectively), which was also weakly correlated with PM₁ (r=0.44) and O_x (r=0.36). The association of secondary variables with the primary components suggests rapid formation of these secondary products on a timescale that is similar to the transit time of the pollutants to the measurement site. PM₁ correlated strongly with the major IVOC component (component 5, r=0.80), which was also weakly associated with LO-OOA (r=0.66) and ${\mathrm{NO}}_{\mathrm{3}\left(\mathrm{p}\right)}^{-}$ (r=0.59), as well as ${\mathrm{NH}}_{\mathrm{4}\left(\mathrm{p}\right)}^{+}$ and ${\mathrm{SO}}_{\mathrm{4}\left(\mathrm{p}\right)}^{\mathrm{2}-}$ (r=0.32 and 0.33, respectively).

Table 7Loadings for the 11-component solution with the inclusion of variables associated with secondary processes.

Download Print Version | Download XLSX

4.1.1 Component 1: tailing ponds (wet tailings)

Component 1 is strongly associated with anthropogenic VOCs (r>0.87) and weakly with TRS (r=0.59) and CH₄ (r=0.59). These pollutants originate from tailing ponds (Small et al., 2015), though it is unclear from this analysis how large a source tailing ponds are compared to fugitive emissions of these pollutants from the nearby processing (e.g., bitumen separation and mining) facilities.

Tailing ponds cover large areas of land and are used to slowly (on a timescale of years to decades) separate solid components, or tailings, from water used in bitumen extraction. Residual bitumen often floats to the top of the settling basins. Most tailing ponds are “wet” (as they contain residual naphtha that is used as a diluent during the transfer of tailings to the ponds) and emit VOCs, CH₄, and CO₂ (Small et al., 2015). The presence of o-xylene, TMB, and the n-alkanes in component 1 is consistent with the fugitive release of VOCs from residual naphtha, which contains these compounds (Siddique et al., 2008, 2011; Small et al., 2015). Furthermore, the observation of TRS and CH₄ from this source is consistent with the presence of anaerobic sulfur-reducing bacteria and methanogens within the ponds, which degrade not only the residual bitumen (Holowenko et al., 2000; Percy, 2013; Quagraine et al., 2005) but also the various components of naphtha (Shahimin and Siddique, 2017; Small et al., 2015). Overall, tailing ponds emissions explain much of the TRS and CH₄ concentration variability in this data set (Table 5) and in a recent aircraft study (Baray et al., 2018).

While component 1 correlates with CH₄ (r=0.59), it does not correlate with CO₂ (r=0.09). Emissions of CH₄ from tailing ponds due to methanogenic bacterial activity are well documented (Small et al., 2015; Yeh et al., 2010) and hence the correlation with CH₄ is not unexpected. Conversely, the lack of correlation with CO₂ seems inconsistent with emission inventories that generally present tailing ponds as large CO₂ sources (Small et al., 2015). One plausible explanation is that tailing ponds are a relatively small CO₂ source overall in the region and that other, larger CO₂ sources and sinks (such as photosynthesis and respiration by the vegetation surrounding the site) dominate the variance impacting the PCA results. It may also indicate that, at least on aggregate and for the particular ponds detected in this work, the emissions are in a regime in which the release of CH₄ dominates over CO₂, i.e., the ponds have, perhaps, become more anoxic than believed to be the case in previous studies and hence emit more CH₄ (Holowenko et al., 2000). For example, Small et al. (2015) showed that older tailing ponds (those without the addition of fresh froth or thickening treatments) tended to emit more CH₄, while newer ponds are associated with higher VOC emissions. It is likely that component 1 is dominated by the nearest pond (the Mildred Lake settling basin, 6–11 km SSE of AMS 13) and other tailings in the SE where the majority of air samples originated from. The Mildred Lake settling basin is one of the oldest in the region and is still actively being used; the correlation with CH₄ and VOC emissions is hence expected.

Component 1 is also associated with NO_y, rBC, CO, and HOA, though these correlations are relatively modest (r=0.27, 0.30, 0.41, and 0.40, respectively). These species typically originate from combustion sources, such as generators and motor vehicles, including diesel-powered engines powering generators or pumps; it is not obvious if and to what extent these are operated on or near tailing ponds, though. Satellite observations have shown elevated concentrations of NO₂ above on-site upgrader facilities, likely a result of emissions from extraction and transport sources (McLinden et al., 2012). In addition, one of the major highways of the region is located adjacent to the Mildred Lake settling basin and other major ponds in the region; highway traffic emissions (of CO, NO_y, rBC, and HOA) may hence also be partially included in component 1.

The bivariate polar plot shows that component 1 was observed when local wind speeds were from the SE and E of the measurement site (Fig. 5c), which is consistent with the notion that the Mildred Lake settling basin and emissions along Highway 63 and, potentially, more distant facilities are sources contributing to this component.

Component 1 is associated with the IVOC signature, though to a lesser degree than components 2 and 5. The association of the IVOC signal with component 1 is slightly poorer (r=0.31) than the association with component 2 (r=0.39), but significantly poorer than the association with component 5 (r=0.74). One possible explanation for the association of IVOCs with tailing ponds vapor is the presence of bitumen in the ponds that was not separated from the sand during the separation stage (Holowenko et al., 2000). This semi-processed bitumen would be expected to emit the same IVOC vapors as those that were observed in the lab (Fig. 2). Tailing ponds contain anywhere from 0.5 % to 5 % residual bitumen by weight (Chalaturnyk et al., 2002; Holowenko et al., 2000; Penner and Foght, 2010). As illustrated in Fig. 4a, some of this material floats on the ponds' surfaces, where IVOCs can partition to the air. Emission of IVOCs from bitumen floating on tailing ponds would be a function of many variables (e.g., diluent composition, extraction methodology, settling rate, temperature) and is thus not expected to be as persistent as CH₄ partitioning from the ponds to the above air or from exposed bitumen on the mine surface, leading to a lower overall correlation.

Component 1 is also weakly associated with LO-OOA (r=0.45). Liggio et al. (2016) found that the observed SOA is dominated by an OOA factor whose mass spectrum was similar to that of aerosols formed from oxidized bitumen vapors. The organic aerosol budget in this study was also dominated by an OOA factor, the LO-OOA (Lee et al., 2018). The association of LO-OOA with component 1 is thus consistent with its association with IVOCs.

4.1.2 Component 2: mine fleet and vehicle emissions

Component 2 strongly correlates with NO_y (r=0.82), rBC (r=0.77), pPAH (r=0.94), and HOA (r=0.74), which suggests a combustion source such as diesel engines. In the Athabasca oil sands, there is a sizeable off-road mining truck fleet consisting of heavy aggregate haulers. In addition, there are diesel engine sources associated with generators, pumps, and land-moving equipment, i.e., graders, dozers, hydraulic excavators, and electric rope shovels (Watson et al., 2013; Wang et al., 2016). Most of these non-road applications have been exempt from highway fuel taxes, on-road fuel formulation requirements, and after-engine exhaust treatment (Watson et al., 2013). Emissions from the hauler fleet and the stationary sources would fit the profile of component 2. Other diesel engines operated in the region include a commuter bus fleet, pickup and delivery trucks, tractor–trailers, and privately owned diesel-powered automobiles used to commute from the work sites to the major residential areas around Fort McMurray, whose emissions are likely captured by component 2 as well, though the magnitude of these relative to the mining truck fleet is not known. Consistent with component 2 being associated with an anthropogenic source is its poor correlation with undecane (r=0.27), likely arising from fugitive fuel emissions.

The bivariate polar plot (Fig. 5d) for component 2 and NO_y in particular (Fig. S4a) matches the location of Highway 63, which crosses the river to the SE of AMS 13 and bends to the E and is indicative of a line source. At the same time, some of the largest mining operations in the region, the Susan Lake Gravel Pit, Aurora North, Muskeg River, and Millennium mines are located to the NE and SE of AMS 13 as well. NO_y, rBC, and HOA (Fig. S4a, b and d) all appear to have dominating point sources to the S and E when wind speeds are 1–2 m s⁻¹. These directions are the same as the Fort McKay industrial park to the E and the Syncrude Mildred Lake facility parking lot to the S, which would have a higher concentration of vehicles emitting these pollutants in a smaller area, whose emissions would be in addition to those from industrial activities.

Component 2 is associated with the IVOC signature and CH₄ (both r=0.39). The mining activities bring bitumen to the surface; similar to what we had observed in lab experiments (Fig. 2, black trace), the surface exposure of bitumen during mining and on-site processing is expected to be associated with fugitive emissions of CH₄ (Johnson et al., 2016) and IVOCs.

Fine-fraction pPAHs are associated strongly with component 2, but no other components. Measurements of individual PAHs in snow and moss downwind from the oil sands facilities have identified multiple sources of PAHs in the Athabasca oil sands, which include windblown petroleum coke dust (also referred to as petcoke for short), a carbonaceous residual product from the upgrading of crude petroleum that is stockpiled on mine sites, and emissions from fine tailings, oil sands ore, and naturally exposed bitumen (Zhang et al., 2016; Jautzy et al., 2015; Parajulee and Wania, 2014). Given this diversity of known sources, the associations of PAHs with only a single component is surprising, though it indicates that emissions from the mining fleet (which would include diesel and, perhaps, windblown emissions from petcoke that is being transported) gave rise to most of the variability in surface-bound PAH concentrations in this data set. The petcoke emissions identified in the studies mentioned above are likely mainly associated with larger supermicron-sized particles, whose PAH content would not be detected by the pPAH measurement in this data set.

Component 2 is not associated with LO-OOA (r=0.11), even though IVOCs are associated with this component. This feature may indicate that the IVOCs emitted in component 2 are qualitatively different from those emitted by components 1 and 5, in that they are less likely to yield organic aerosol on the timescale of transport from emission to observation. One reason for the difference could be that the bitumen that is transported by the mining fleet is relatively freshly exposed, whereas the IVOCs released from tailing ponds or from mine faces (component 5) may have been oxidized to a greater extent and hence more prone to rapid aerosol formation.

There is no association of component 2 with CO₂ (r=0.08). This is somewhat unexpected as the trucks are expected to release CO₂ (Wang et al., 2016) but could be due to significantly larger CO₂ sources in the area dominating the observed CO₂ variability at AMS 13 (e.g., components 3 and 6). Furthermore, one would expect an association of non-road mining truck emissions with aromatics and alkanes. Component 2 exhibited only poor correlations with decane (r=0.22) and undecane (r=0.27) and no correlation with o-xylene (r=0.08), suggesting that other components (i.e., component 1) explained most of the variability in their concentrations at this site.

4.1.3 Component 5: surface-exposed bitumen and hot-water bitumen extraction

Component 5 correlates more strongly with the IVOCs (r=0.74) than with any other component and correlates strongly with LO-OOA (r=0.72), weakly with rBC (r=0.44), and poorly with HOA (r=0.25), NO_y (r=0.22), decane (r=0.23), undecane (r=0.20), and TRS (r=0.26). We interpret this profile as emissions from surface-exposed bitumen, which outgasses IVOCs.

One possibility is that these emissions occur on mine faces, where previously unexposed bitumen is brought to the surface as a result of mining. Only a relatively small portion of the mine faces are actively mined; those parts give rise to rBC and NO_y emissions from combustion engines in heavy haulers or generators powering equipment. The poor association of component 5 with TRS could be due to sulfur-reducing bacteria found on the surface of bitumen. However, most of the variability in TRS at AMS 13 is attributed to composite or “dry” tailing ponds given their more conducive environment to microbial activity.

Component 5 does not correlate with CO₂ ($r=-\mathrm{0.03}$) or with CH₄ (r=0.12), which is somewhat at odds with the notion of mine faces as the main source of IVOCs. The mine faces give rise to substantial fugitive emissions of CO₂ and CH₄ (Johnson et al., 2016) – these emissions are likely captured by component 6 in this analysis (see Supplement). It is unclear to what extent these greenhouse gases are released relatively quickly from “hot spots” (i.e., from a small number of locations) through surface cracks and fissures or by slow release from new material that is exposed and then releases greenhouse gases during material handling, transport, and processing (Johnson et al., 2016). IVOCs from surface-exposed bitumen are likely released by the latter mechanism and are temperature dependent. If the mine faces are indeed the main IVOC source, the analysis results presented here suggest that the IVOC emissions from surface-exposed bitumen on mine faces are decoupled from CH₄ emissions in time and appear as a distinct component and hence corroborate the hot spots or fast-release hypothesis, though clearly more work is needed to characterize greenhouse gas emissions from oil sands mine faces.

The association of IVOCs with component 5 may also be a result of fugitive emissions during the hot-water-based extraction of bitumen sand slurries during the separation phase of bitumen treatment. Generally, bitumen is extracted in a weak alkaline environment by aeration of the solution to optimize the separation of sand and bitumen (Masliyah et al., 2004). Unrecovered bitumen and naphtha then end up in tailings. The recovered bitumen and naphtha are moved to upgrader facilities where they undergo further treatment (such as coking or hydrotreatment). The magnitude of fugitive emissions during these downstream extraction processes could be large, considering the bitumen is heated and actively aerated. Future work should investigate IVOC fluxes near extraction plants and on mine faces.

Component 5 correlates strongly with LO-OOA (r=0.72), which is likely generated in part by photochemical aging of IVOCs. A back-of-the-envelope calculation using a k_OH of $\mathrm{1.8}\times {\mathrm{10}}^{-\mathrm{11}}$ cm³ molecule⁻¹ s⁻¹ based on the fact that used diesel exhaust IVOCs (Zhao et al., 2014) and an estimated midday OH concentration of 7×10⁶ molecules cm⁻³ (Liggio et al., 2016) gives a first-order lifetime of 130 min with respect to IVOC oxidation by OH during daytime. The photochemical age, estimated using relative concentrations of 124-TMB and n-decane and the method described by Borbon et al. (2013), during daytime was 1.0±0.4 h; assuming similar photochemical ages, we estimate that between 25 % and 50 % of the emitted IVOC is (potentially) oxidized during daytime (see Supplement). This oxidation will contribute SOA growth (Kroll et al., 2011). Hence, we expect some formation and growth of organic aerosol associated with component 5.

Finally, it is conceivable that a “natural” background of IVOCs exists in the region (since bitumen can be found at or near the surface in many parts of the region); such a natural background would also be included in component 5. However, this natural bitumen would have been exposed at the surface for geological timescales and, unlike unexposed, buried bitumen, likely would have lost most of its volatile content over that period. Furthermore, the mine faces occupy large swaths of land in the region (as evident from satellite imagery). Thus, the IVOC emissions are more likely due to anthropogenic activity than due to a natural phenomenon.

4.2.1 Component 3: biogenic emissions and respiration

Component 3 is strongly correlated with the monoterpenes α-pinene (r=0.98), β-pinene (r=0.98), and limonene (r=0.92) and is hence identified as a biogenic emissions source. This component is also weakly associated with CO₂ (r=0.48).

At AMS 13, CO₂ and the monoterpenes exhibit a very similar diurnal cycle: they are present in higher concentrations during the night than during the day (Fig. 3) due to a decrease in the boundary layer height (BLH) at night coupled with plant respiration of CO₂ and non-photochemical emission of monoterpenes (Fares et al., 2013; Guenther et al., 2012). During the day, mixing ratios of CO₂ are lower due to plant uptake and photosynthesis, and mixing ratios of terpenes are lower due to higher mixing heights and vertical entrainment and due to oxidation by O₃ and OH (Fuentes et al., 1996). Hence, the PCA gives a positive correlation of monoterpenes with CO₂ even though the physical processes, photosynthesis and respiration, work in the opposite direction.

The bivariate polar plots (Fig. S5a–c) show that the monoterpenes and CO₂ were observed in the highest concentrations when the wind speeds were low (<1 m s⁻¹), consistent with formation of a stable nocturnal boundary layer.

To corroborate this interpretation, the PCA was repeated with BLH estimated by a light detection and ranging (LIDAR) instrument (Strawbridge, 2013) added as a variable (Table S9 in the Supplement). Since BLH is not emitted by any source, it appears as a single variable component (r=0.90). The only other component that BLH (anti)correlates with is the biogenic component 3 ($r=-\mathrm{0.35}$).

The dominant monoterpene species observed was α-pinene, followed by β-pinene and limonene, though occasionally there was twice as much β-pinene than α-pinene in the sampled air. Some variability in this ratio is expected since emission factors vary considerably among tree species (Geron et al., 2000), which are not homogeneously distributed throughout the region (e.g., Fig. S1 of Rooney et al., 2012).

Simpson et al. (2010) observed enhancements of α-pinene and, to a greater extent, β-pinene over the oil sands (up to 217 and 610 pptv) compared to background levels of 20±7 and 84±24 pptv, respectively, during midday overflights (which occurred between 11:00 and 13:00 local time). Similar enhancements were also reported by Li et al. (2017), who observed emissions of biogenic hydrocarbons in the four facilities sampled, three of which showed a higher β- than α-pinene concentration. The PCA (Table 5) showed no significant correlation of α- and β-pinene with any of the anthropogenic components, which implies that the biogenic source strength is simply too large for any anthropogenic emissions of terpenes to be picked up in the analysis, especially considering that terpenes are relatively short lived.

The biogenic source shows poor anticorrelations with NO_y ($r=-\mathrm{0.26}$) and NH₃ ($r=-\mathrm{0.24}$). Many NO_y species (i.e., NO₂, HONO, peroxycarboxylic nitric anhydrides or PAN, and HNO₃) deposit to the forest canopy (Hsu et al., 2016; Min et al., 2014; Fenn et al., 2015); at night, when mixing heights are lower, their concentrations are expected to decrease faster than during the day and are thus out of phase with the CO₂ and terpene concentrations. The poor anticorrelation with NH₃ likely arises because the NH₃ emissions from plants are mainly stomatal and scale with temperature and are hence larger during the day than at night, anticorrelated with the terpene source (Whaley et al., 2018).

4.2.2 Component 4: Upgrader emissions

Component 4 is strongly correlated with SO₂ (r=0.97) and total sulfur (r=0.93). By far the largest source of SO₂ in the region are upgrader facilities, which emit as much as 6×10⁷ kg annually according to emission inventories (ECCC, 2013). Significant SO₂ emissions from upgrader facilities have recently been confirmed by aircraft studies (Simpson et al., 2010; Howell et al., 2014; Liggio et al., 2016). Component 4 is also poorly correlated with NO_y (r=0.21) but not with rBC (r=0.05), consistent with a non-sooty (i.e., lean) combustion source such as upgrader stacks. Strong enhancements in SO₂ were only observed intermittently as “spikes”, which is expected when sampling emissions from relatively few and discrete point sources.

Component 4 is not associated with CO₂ ($r=-\mathrm{0.12}$), even though inventories indicate that the upgrading facilities are the largest CO₂ source in the region (Furimsky, 2003; Englander et al., 2013; Yeh et al., 2010). In this data set, the lack of correlation of component 4 with CO₂ (and to some extent with PM₁₀₋₁ as well) likely arises mainly from a sampling bias as stack emissions were only observed during daytime, likely due to diurnal variability in the atmospheric boundary layer structure as explained below.

Most of the variability in CO₂ concentration at AMS 13 is due to surface-based sources that originate from large areas, especially biogenic processes (photosynthesis during the day and respiration at night, component 3) and anthropogenic surface sources such as those captured by component 6 (Sect. 4.2.3). Other anthropogenic pollutants, such as SO₂, NO_y, and CH₄, are not subject to large biogenically driven processes and are less affected than CO₂.

In contrast to surface sources, emissions from the >100 m tall stacks are comparatively undersampled and observed mainly during daytime, when vertical mixing brings elevated plumes to the surface, yet CO₂ concentrations are generally much lower than during the night due to uptake by vegetation. At night, pollutants emitted from stacks are injected above the likely very shallow nocturnal surface layer and were hence not observed at the surface. Vertical profile measurements of SO₂ stack plumes by a Pandora spectral sun photometer at Fort McKay during daytime have shown considerable vertical gradients and only occasional transport of SO₂ all the way to the surface (Fioletov et al., 2016).

The association of component 4 with CO₂ is negative because the stack emission source is observed only during the day when the large biogenic sink dominates and effectively masks the relatively small increase due to anthropogenic CO₂. In contrast, background concentrations of SO₂ are comparatively low, and the increase in SO₂ concentrations is readily picked up by the PCA.

It would be interesting to conduct a future study in winter when biogenic activities decrease; a wintertime PCA of surface measurements might be able to associate CO₂ enhancements with upgrading facilities, though boundary layer mixing heights would decrease as well, which would make a PCA using surface data even more challenging.

Component 4 does not correlate with PM₁₀₋₁ volume (r=0.09). It is clear that the emitted SO₂ will contribute to secondary aerosol formation downwind, such that a correlation of stack emissions with PM₁₀₋₁ volume might be expected. However, these secondary contributions will likely mostly be in the submicron aerosol fraction, which adds relatively little to PM₁₀₋₁ volume. Further, PM₁₀₋₁ volume is dominated by coarse particles from other primary sources, mostly windblown emission of sand from the mine surfaces, roadways, and perhaps bioaerosol (component 7, see the Supplement). These effects make PM₁₀₋₁ volume from stacks appear comparatively small, such that the variability in the larger, surface-based sources likely masks the contribution of stack emissions to PM₁₀₋₁ variability.

The bivariate polar plot of component 4 (Fig. S6d) shows that the largest magnitudes were observed when local winds were from the SE. The corresponding plot of SO₂ (Fig. S6a) reveals two more distinct sources: a larger one from the E and a smaller one from the SSE. However, only two facilities (Sunrise and Firebag) are located to the E at relatively large distances of 37 and 47 km, respectively. The largest known upgrading facility and SO₂ sources in the area (i.e., upgrading facilities located at the Mildred Lake and Suncor base plants) are located to the S and SE of AMS 13. Considering that the stack emissions are only observed intermittently, we speculate that there is a mesoscale transport pattern in the Athabasca River valley that channel emissions, such that the local wind direction and speed may be misleading as to the true location of these sources. For more extensive data sets, such phenomena may very well average out but perhaps did not in this case.