Measurements were conducted in a trailer in downtown Fairbanks next to the University of Alaska Fairbanks Community and Technical College (64.84064° N, 147.72677° W, elevation 136 m above sea level) from 17 January to 25 February 2022 as part of the 2022 ALPACA campaign, unless stated otherwise. This location is referred to as the downtown site throughout the paper, which was co-located with Ijaz et al. (2025b). Ketcherside et al. (2025) took place in a residential neighborhood (referred to as the residential site) approximately 3 km away from the downtown site. Detailed information about the ALPACA campaign can be found in Simpson et al. (2024).

A proton-transfer-reaction time-of-flight mass spectrometry (PTR-TOF 6000 X2, Ionicon Analytik GmbH, Austria) was used to monitor VOCs. As described in Ijaz et al. (2025a), the PTR-ToF-MS was equipped with an inlet to switch between gas-phase and aerosol-phase species. The gas-phase was measured for 15 min every hour, and the aerosol-phase species for 45 min every hour. Both were measured at a 20 s time resolution. Only the gas-phase data was used for the PMF analysis in this work. The instrument was operated at a low E/N of 65 Td (i.e., drift voltage/pressure; pressure, temperature, and voltage of the drift tube were set at 2.6 mbar, 120 °C, and 230 V) and in RF mode for optimal sensitivity. Raw PTR-ToF-MS spectra were processed with the commercial Ionicon Data Analyzer (IDA) software to fit high-resolution mass peaks and extract the corresponding time series. Times series were post-processed (i.e., background subtraction, conversion of raw signal to mixing ratios, 2 min temporal averaging, PMF input generation) with an in-house data processing tool, PeTeR Toolkit (version 6.0; Igor 6.37).

Mixing ratios were calculated using Eq. (1): 1R=RH+H3O+×TrH3O+TrRH+×Udrift×Tdrift2k×τ×pdrift2 Where RH⁺ and H₃O⁺ are the signal intensities in counts per second (cps) of the individual protonated ion and the reagent ion corrected by their relative transmission (Tr) respectively, Udrift, Tdrift, pdrift and τ are the voltage (V), temperature (K), and pressure (mbar) residence time (s) of the ions in the drift tube respectively, k the reaction rate (cm³ s⁻¹) between the reagent ion and VOC. The average relative uncertainty regarding the concentration derived from the propagation of the relative uncertainties on the transmission and the k is estimated to be within ±30 %. This value aggregates the systematic uncertainty of the reaction rate constants, typically 25 %–30 % (Ellis and Mayhew, 2014; Zhao and Zhang, 2004), and the experimental ion transmission curve (15 %). The combined root sum-square calculation yields an estimated uncertainty of approximately 30 % for uncalibrated VOCs. While this uncertainty can be in the range of 20 %–50 % for specific compounds (Sekimoto et al., 2017), 30 % represents a reasonable overall estimate across the range of species considered in this study.

A 12-component mix at ppmV level (Apel Riemer Environmental Inc, Miami, USA) in the range of 32–134 amu diluted in pure nitrogen was used to experimentally determine the relative ion transmission at the beginning and at the end of the measurement campaign.

In this study, no general fragmentation correction was applied. As demonstrated by Pagonis et al. (2019), the absence of such corrections may lead to an underestimation of oxygenated and nitrogen-containing compound families. The magnitude of this bias is difficult to assess for the present dataset, as it depends on the fragmentation behavior of individual compounds. However, the PTR-ToF-MS was operated at a low E/N (60 Td), which minimizes fragmentation, and the resulting bias is therefore expected to be limited. It is worth noting that fragmentation corrections are only feasible when the protonated molecular ion is detected and can be uniquely assigned to the compound of interest. This is not the case for some O-containing compounds (e.g., > C₃ alcohols, > C₃ linear aldehydes) or for N-containing compounds, for which the protonated molecular ion is generally not observed in ambient air, and whose main fragments ions are not specific and overlap with fragment ions from abundant atmospheric hydrocarbons, such as alkanes and alkenes (Brown et al., 2010; Buhr et al., 2002; Duncianu et al., 2017).

Compound-specific proton-transfer reaction rate constants (k) were applied for VOCs where available in the literature (Cappellin et al., 2012). For all other species, a rate constant of 2 × 10⁻⁹ cm³ s⁻¹ was assumed; this is a common value used when a measured rate constant is unavailable (Cappellin et al., 2012; Wang et al., 2020, 2021). An additional correction was applied to ethanol using calibration coefficients obtained from post-campaign laboratory measurements, as significant fragmentation resulted in an underestimation of ethanol mixing ratios by a factor of 9 when default reaction rate constants were assumed. Post-campaign ethanol calibration was conducted using a certified 20-component gas reference mixture, for which the certified concentration uncertainty was 10.1 %. Instrument drift between campaign and post-campaign calibration was corrected using acetaldehyde, a compound common to both the campaign and post-campaign calibration cylinders with certified relative uncertainties of 5 % and 5.4 %, respectively. Therefore, the relative uncertainty obtained by uncertainty propagation by transferring the ethanol calibration to the post-campaign period is estimated as 12.5 %.

Alongside the PTR-ToF-MS, two additional formaldehyde measurements were made using a MIRA Ultra Gas Analyzer (Aeris Technologies) and COmpact Formaldehyde FluorescencE Experiment (COFFEE) (St. Clair et al., 2017). Correlations between all three instruments were high (r2 ranging from 0.69 and 0.81, Fig. S1 in the Supplement) but the PTR-ToF-MS measurements were about four times lower than both the MIRA Ultra Gas Analyzer and COFFEE. This is because formaldehyde has a proton affinity close to water, the back-reaction of protonated formaldehyde with water vapor becomes relevant and reduces the sensitivity of formaldehyde in the PTR-ToF-MS (Vlasenko et al., 2010). Although the PTR-ToF-MS measurements of formaldehyde were underestimated due to back-reactions with water vapor, we find no relationship between relative humidity or dew point and the magnitude of difference of PTR-ToF-MS compared to other formaldehyde instruments (Fig. S2). Therefore, the formaldehyde measured by the PTR-ToF-MS was scaled up by a constant factor of four to match these instruments, and the scaled data was used in the PMF input. Mixing ratios were corrected for background contribution using a cylinder of high purity nitrogen (99.999 purity).

Based on post-campaign calibration measurements using D₃–D₅ siloxane standards and comparison of their main fragmentation patterns with the closest matching ions observed in the campaign data, only ions at m/z 297.098 and 299.066, which are consistent, within the instrument mass accuracy, with ions previously reported for pure D₄ siloxane were detected (m/z 297.083 and m/z 299.062 corresponding to C₈H₂₄O₄Si₄H⁺ and C₇H₂₂O₅Si₄H⁺). However, the fact that both ions are mainly associated with the biomass burning factor (50 % and 42 % versus 3 % and 8 % for the VCP factor) indicates that their assignment to D₄ siloxane remains uncertain. Other plausible assignments, including C₂₀H₁₀O₃H⁺ (297.091) and C₂₁H₁₂O₂H⁺ (299.071), corresponding to compounds that are more likely to be associated with biomass burning emissions, cannot be excluded.

The PTR-ToF-MS is unable to detect short-chain alkanes with less than C₈ such as ethane, propane, and butane due to their low proton affinities. Very few works contain a comprehensive overview of both oxygenated VOCs and short-chain alkane VOCs in the US. The CalNex campaign, which used multiple instruments to measure both categories as described in Heald et al. (2020), found that short-chain alkanes made up on average 36.5 % of the total measured µg C sm⁻³, excluding organic aerosols. Heald et al. (2008) contains VOC measurements from other North American campaigns, although not all VOCs were measured for all campaigns. Ethane, propane, and butane range from 7.5 %–37.2 % of the measured µg C sm⁻³ for measurements done in the US and Canada, excluding organic aerosols. A direct comparison between these works and Fairbanks are difficult due to the unique climate of Fairbanks, but we expect short-chain alkanes and alkenes to be in a similar range.

Other non-VOC measurements made at the downtown site were used for comparison with the PMF results. PM₁ OA, sulfate, nitrate, and ammonium were measured by an Aerosol Mass Spectrometer (AMS). Black Carbon (BC) was measured by a Multi-Angle Absorption Photometer (MAAPS). Other downtown measurements include carbon monoxide (CO), SO₂, NO_x, ozone (O₃), temperature, wind speed and wind direction. Total ammonium was measured using a mist chamber coupled with ion chromatography (MC-IC) (Cofer et al., 1985), and total S(IV) was measured using a particle into liquid sampler coupled with ion chromatography (PILS-IC) (Orsini et al., 2003). Both MC-IC and PILS-IC methods have been discussed in other ALPACA papers (Campbell et al., 2024; Dingilian et al., 2024a; Simpson et al., 2024).

PMF is a mathematical model commonly used to estimate source contributions of VOCs datasets. Briefly, the PMF model uses a time series dataset matrix and corresponding error matrix as inputs. The user specifies a number of factors – in our case, the number of factors equates to the number of VOC sources. The PMF model attempts to split the main dataset into two matrices: a time series of factors, and the contribution of each VOC species to each factor. The user then interprets the resulting matrices to determine what source each factor corresponds to. More detailed information about the model can be found in literature (Paatero, 1999). We used SoFi Pro version 9 which utilizes the ME-2 engine (Canonaco et al., 2013; Paatero, 1999). Over 300 ions were included for PMF analysis. An error matrix was also calculated with the PeTeR tool by propagating uncertainties in ion counts and background signals, assuming Poisson counting statistics (de Gouw and Warneke, 2007). The data was down-weighted based on the recommendations of Paatero and Hopke (2003).

We first examine the campaign average for VOC measurements. The mean VOC mixing ratio during the campaign was 35.6 ppb. In this section, for simplicity, we report VOC mixing ratio ± one standard deviation for species which had a campaign average of 0.05 ppb or higher. This included 46 species which account for 33.6 ppb (94.3 % of the total VOC mixing ratio). It should be noted that there were two periods of enhanced pollution during the ALPACA campaign, as described in Simpson et al. (2024). Figure S3 shows PM₁ (measured by aerosol mass spectrometer, described below) and temperature during the campaign, with the pollution events highlighted.

Figure 1 shows the percent contributions and campaign-mean VOC mixing ratios by functional group (see Table S1 for details). Alcohols, aromatics, carbonyls, acids, non-aromatic hydrocarbons, and nitrogen-containing species are the dominant VOC classes. Alcohols were by far the most prominent category due to the high mixing ratio of methanol (6.8 ± 8.0 ppb) and ethanol (5 ± 7.4 ppb) during the campaign. The species with the third largest mean mixing ratio was formaldehyde at 2.3 ± 1.3 ppb. Common species from other functional group categories are (in decreasing order): toluene, C₈ aromatics, benzene, and C₉ aromatics for aromatics; formaldehyde, acetaldehyde and acetone for carbonyls; C₂H₄O₂ (acetic acid) and formic acid for acids. The speciation is similar with median values (Fig. S4).

Methanol and ethanol are highest from 21 to 25 January before the first pollution event, with methanol spikes up to 166 ppb and ethanol spikes up to 180 ppb. Interestingly, as shown in Table S2 and Fig. S5a, there was not much enhancement of methanol and ethanol during the cold polluted event compared to periods outside the pollution events, but they were significantly increased during the warm pollution event, with methanol spikes up to 133 ppb. Both are similar mixing ratios to measurements at the residential site in Ketcherside et al. (2025).

The methanol in wintertime Fairbanks is on average 1.2–2 times higher than other urban wintertime measurements in the US and China (Li et al., 2019; Millet et al., 2005), and 2–6 times higher than measurements in Europe (Languille et al., 2020; Legreid et al., 2007; Patokoski et al., 2014). The large discrepancy of methanol between Fairbanks and Europe may be because the US and Canada use methanol-based windshield wiper fluid and Europe uses ethanol-based. Legreid et al. (2007) attributes the high ethanol mixing ratios in Zürich, Switzerland to windshield wiper fluid. Slowik et al. (2010) attributes high levels of methanol due to windshield wiper fluid in Toronto, Canada, but did not report a mixing ratio. We further discuss the influence of windshield wiper fluid in Fairbanks in Sect. 3.2. Ethanol in Fairbanks is between 1.2–5 times higher than measurements in the US and Europe (Gkatzelis et al., 2021; Legreid et al., 2007; Millet et al., 2005).

The aromatics were mainly made up of toluene (2.5 ± 2.7 ppb), C₈ aromatics (xylenes and ethylbenzene, 2.1 ± 2.2 ppb), benzene (1.00 ± 0.92 ppb) and C₉ aromatics (0.62 ± 0.64 ppb). All species show enhancements during the polluted episodes, with similar mixing ratios during the cold and warm pollution events (Fig. S5b, Table S2). The mean aromatic mixing ratio in Ketcherside et al. (2025) is 11.9 ppb, almost double what is reported in this work. All aromatics are higher at the residential site, with C₈ aromatics having the largest discrepancy – nearly 3 times higher at the residential site than downtown. This likely reflects a greater influence of heating oil combustion in residential areas. The mixing ratios of aromatics in this work are 2–12 times higher than those reported during other North American wintertime campaigns in urban locations like NYC, NY (Gkatzelis et al., 2021); Boulder, CO (Gkatzelis et al., 2021); Pittsburgh, PA (Millet et al., 2005); Houston, TX (Sadeghi et al., 2022); and Toronto, Canada (Rivellini et al., 2024). Benzene, toluene and C₈ aromatic levels are comparable to urban and suburban wintertime measurements in China (Li et al., 2015, 2019).

The high aromatic mixing ratio in wintertime Fairbanks compared to other US cities is also likely due, in part, to differences in gasoline composition. Yano et al. (2016) measured the aromatic content of Anchorage, AK gasoline in 2013, after new gasoline regulations took place nationwide. They found benzene content was 1.3 % v/v, and the total aromatic content (benzene, toluene, ethylbenzene, and xylene) was 25 % v/v. This is more than double the average benzene content in gasoline in 2013 (0.6 % v/v), and about 1.3 times higher than the total aromatic content (19 % v/v) provided by the United States Environmental Protection Agency (EPA, 2017). The relatively high aromatic content in Alaskan gasoline, combined with increased vehicle emissions at cold temperatures, leads to high gas-phase aromatic mixing ratios.

The greatest carbonyl species was formaldehyde (2.3 ± 1.3 ppb), followed by acetaldehyde (1.8 ± 1.2 ppb) and acetone (1.2 ± 0.7 ppb). Nine other carbonyls had mean values between 0.05–0.17 ppb (Table S1). Formaldehyde, acetaldehyde and acetone are all enhanced during the pollution periods (Fig. S5c, Table S2). Formaldehyde is noticeably much higher during the cold event, whereas acetaldehyde is highest during the warm event. In contrast, measurements of formaldehyde (1.4 ppb) and acetaldehyde (1.3 ppb) at the residential site are on average 40 %–60 % lower compared to downtown measurements. Acetone has a similar mixing ratio range to other wintertime measurements in the US, Europe, and China, although acetaldehyde is 2–3 times higher (Gkatzelis et al., 2021; Legreid et al., 2007; Li et al., 2019; Millet et al., 2005). In another study, formaldehyde, acetaldehyde, and acetone are in a similar range to wintertime non-haze periods in Beijing, and about half as much when compared to haze periods (Rao et al., 2016).

Acetic acid (1.7 ± 1.4 ppb) and formic acid (0.88 ± 0.32 ppb) are the two most prominent acid species. Although these compounds often spike at the same time, in many periods formic acid increases while acetic acid decreases (Fig. S5d, see 28–29 January, 3–4 February, 21–23 February), and they are poorly correlated (Fig. S6), suggesting that they are likely arising from different sources, instead of a shared source like wood heating. Acetic acid at the residential site is lower (1.3 ppb) compared to the downtown site, but formic acid is similar. Formic acid is unique in that it is slightly lower during both pollution events compared to the clean periods (Table S2). This indicates a lack of primary sources of formic acid in downtown Fairbanks, and may be evidence of dark aging, which is discussed more in Sect. 3.2. Both formic acid and acetic acid are moderately lower than urban and suburban wintertime measurements in China (Li et al., 2019).

Mixing ratios of nonaromatic hydrocarbons and nitrogen species are generally low. The two most prominent nonaromatic hydrocarbon species are C₃H₆ (1.7 ± 1.4 ppb) and C₅H₈ (0.39 ± 0.31 ppb) (Fig. S7a), but these ions may also include fragmentation of larger VOCs (Coggon et al., 2024; Gueneron et al., 2015). Similarly, the most commonly measured nitrogen-containing ion was NO2+ fragments (Fig. S7b), a common fragment for many organic nitrate species (Aoki et al., 2007; Hansel and Wisthaler, 2000).

Nine factors were resolved from the PMF analysis. These factors are: gasoline, diesel, windshield wiper fluid, acetic acid, wood heating, heating oil, volatile chemical products (VCPs), aged-transported, and aged-local. Section S1 describes how the nine-factor solution was chosen, while Sect. S2 describes how VOC profiles, diurnal cycles, and correlation with external measurements were used to assign the PMF factors. A summary of the factors is shown in Fig. 2 and Table S3. A log-scale version of Fig. 2 is shown in Fig. S8, all species are shown in Figs. S9–S11, and time series of all factors is shown in Fig. S12. Diurnal cycles of all factors are shown in Fig. 3. In the following, we define a VOC as negligible to a factor's composition when a VOC contributes less than 3 % of its average campaign mixing ratio to a factor. The likely assignment of the VOC “C₂H₄O₂” is acetic acid, however we refer to it by its chemical formula to avoid confusion with the acetic acid factor.

Ketcherside et al. (2025) also used PTR-ToF-MS data in Fairbanks to conduct PMF of VOC measurements in Fairbanks, but there are a few main differences. First, their instrumentation was located in a residential area 2.6 km away from the downtown site. Therefore, their analysis is likely more influenced by residential factors, like space heating. Second, in their PMF analysis they included species that were measured at the residential site by instruments besides PTR-ToF-MS, like the gas-phase species CO and NO_x, as well as aerosol-phase PM₁ species like BC, sulfate, nitrate, and ammonium. They also include SO₂ measured at the downtown site. In our analysis, we only use PTR-ToF-MS data as inputs for PMF analysis, but we do compare the PMF outputs to the corresponding downtown measurements of these species. Third, their analysis is based on hourly measurements, while ours is based on 2-minute data. They also use the EPA PMF 5.0 software while we use SoFi Pro V9; however, both are based on the ME-2 engine (Canonaco et al., 2013; Norris et al., 2014) so the PMF procedures should be largely similar. More information on the differences between these two sites and the instrumentation at the residential site can be found in Simpson et al. (2024) and Ketcherside et al. (2025).

Despite the different methodology, there are many similarities between our work and the results in Ketcherside et al. (2025). Both studies found factors related to aged-transported air masses (which they call ozone), local aged air, heating oil, wood heating (i.e., biomass burning), and traffic. Their study identifies two biomass burning factors (one for short-lived VOCs and one for long-lived), while we only have one. We find two traffic factors, while in their six-factor solution they only find one. Their traffic diurnal cycle is most similar to our gasoline factor. We find the wiper fluid, VCP, and acetic acid factor, which they did not. They found that methanol was largely associated with traffic emissions, which is similar to our assignment of methanol to windshield wiper fluid. They also did not observe the large spikes of C₂H₄O₂ that we find, which helps to confirm that our analysis of C₂H₄O₂ are local to the downtown site. They mainly attribute C₂H₄O₂ and formic acid to biomass burning, while we find that much of the C₂H₄O₂ is in its own factor (the acetic acid factor), and formic acid is mainly attributed to the aged-transported factor. Figure 5 presents pie charts of the PMF factor contributions from our work and Ketcherside et al. (2025), It is notable that wood heating and heating oil are a significantly smaller fraction in our work compared to theirs. Figure S18 can be used for further comparison between the two studies.

Both our study and Ketcherside et al. (2025) find traffic to be the largest source and heating oil to be the second largest source of aromatic VOCs. However, in our results the gasoline factor dominates, making up 75 % of the total benzene, toluene, C₈ aromatics, and C₉ aromatics, while heating oil only makes up 13 %. In Ketcherside et al. (2025), gasoline makes up 48 % and heating oil makes up 39 %. This large difference likely due to the increased influence of space heating at the residential site compared to the downtown site. The large amount of aromatic VOCs attributed to VOCs in both works is in contrast to the Alaska Department of Environmental Conservation's emission inventory for ALPACA, which contributes little to no aromatic species to heating oil, as described in Zhang et al. (2026). The results of these works show a need to update the emission inventories to include heating oil emissions of aromatic species. Based on these findings, tracer modelling of aromatics during ALPACA show improved agreement with observations when these emissions are included (Zhang et al., 2026).

Ketcherside et al. (2025) attributes 18 % of formaldehyde to heating oil, whereas we do not see any formaldehyde associated with heating oil. A possible reason for this discrepancy could be due to the different time resolution used in our works. The measurements in Ketcherside et al. (2025) were averaged to one hour, while our measurements are every 2 min. Heating oil and diesel fuel are compositionally similar but the combustion process is different, so the emissions are comparable but not identical. Our two-minute data may be able to resolve the differences between heating oil and diesel emissions more effectively. Another reason is that since their work was conducted at a residential site with less traffic than our downtown site, diesel emissions are expected to be much smaller and thus may be more difficult to distinguish from heating oil. As the downtown site is much more dominated by traffic emissions, the variation of formaldehyde related to heating oil may be difficult to distinguish from other factors. Qualitatively, their heating oil factor contains characteristics of both our diesel and heating oil factors – for example, a 5 am increase similar to our diesel factor, and an afternoon increase present in our heating oil factor.

Our wood heating factor appears to be most comparable to the short-lived biomass burning factor in Ketcherside et al. (2025), including similarities between the furan, methylfuran, furfural, and methylfurfural assignments, and their diurnal cycles (Fig. 3, Table S4). Meanwhile, our aged-transported factor has similarities to the chemical composition of their long-lived biomass burning factor (notably formic acid and formaldehyde, as well as furans and other aldehydes, Table S4) but the diurnal cycle of the aged-transported factor is closer to their ozone factor diurnal cycles. The discrepancy between chemical composition and diurnal cycles may be because 98 % of ozone gas was applied to their ozone factor. Although our aged-transported factor is correlated with observed ozone, there are also periods where observed ozone and the aged-transported factor are not in agreement (Fig. S15). Their ozone factor describes the transport of background levels of ozone from all wind directions, but our aged-transported factor is highest when wind direction is from the east.

We cannot compare the source apportionment of species like CO, NO_x, BC, and PM₁ sulfate because we used these species as external validation to compare with the PMF results. However, we do find that CO, NO_x, and BC are well-correlated with traffic (Fig. 4), which is consistent with these species being associated with their traffic factors.

Ijaz et al. (2025b) conducted PMF in Fairbanks using PM₁ measurements with both a PTR-ToF-MS with a CHARON inlet and an AMS. We focus on their PTR-ToF-MS results here since that is the same instrument used in our analysis. They found four residential heating-related factors, including three biomass burning factors and one heating oil factor, which contribute on average 2.6 µg m⁻³ (62 %) of total measured OA. Although this is a much larger fraction of OA mass compared the 14 % of total VOCs we attribute to residential heating, the absolute mass is much lower than our wood heating (8.6 µg m⁻³) and heating oil (12.1 µg m⁻³) factors. Ijaz et al. (2025b) attribute about 1.6 % (0.06 µg m⁻³) of measured OA to traffic, significantly less than the 29 % (36.6 µg m⁻³) of VOCs that we attribute to the gasoline and diesel factor. They also identify a cooking factor, while neither our work nor Ketcherside et al. (2025) resolved a cooking factor. This comparison shows major differences between the sources of PM₁ OA and gaseous VOCs. Comparison of VOC and aerosol partitioning is also explored in Ijaz et al. (2025a).

High concentrations of HMS and related aldehyde-S(IV) species are a unique feature of Fairbanks wintertime PM_2.5 (Campbell et al., 2022, 2024; Dingilian et al., 2024a; Holen et al., 2025; Moon et al., 2023) and so are investigated more here. The MC-IC measures gas-phase and PM­_2.5 aerosol-phase species. This is important for total ammonium, which could have a significant portion in the gas-phase. In extremely cold environments like Fairbanks, high concentrations of total ammonium can increase aerosol pH to 4-5, greatly facilitating secondary aqueous chemistry of species like HMS and sulfate (Campbell et al., 2024). We compare MC-IC total ammonium (TA) measurements to the PMF factors in order to help determine the sources of TA.

Gasoline vehicles may be a large source of total ammonium in wintertime Fairbanks. The diurnal cycle of TA has many similarities to the gasoline factor, including an early morning increase and 05:00 pm LT peak on weekdays (Fig. S19). We find that TA correlates best with the gasoline factor (r = 0.659), followed by the VCP factor (r = 0.620) as shown in Fig. 4. TA is also correlated to external factors CO (R2 = 0.47), NO_x, (R2 = 0.42), and somewhat with BC (R2 = 0.36), which are commonly associated with vehicle emissions (Fig. S20). Vehicle emissions have also been shown to be a large source of ammonia (Ji et al., 2025). Therefore, a large fraction of TA may be coming from gasoline vehicle emissions. VCPs may also be a large source of TA, but since VCPs include a broad category of industrial and personal care products, it is more difficult to pinpoint exact sources.

HMS formation may be highly dependent on ammonia emissions from gasoline vehicles because of ammonia's effects on aerosol acidity, and formaldehyde emissions from diesel vehicles because formaldehyde is an HMS precursor. PILS-IC total S(IV) correlates best with diesel (r = 0.636), followed by gasoline (r = 0.516) and wood heating (r = 0.515), as shown in Fig. 4. The total S(IV) during the ALPACA campaign was found to be mostly HMS in Dingilian et al. (2024a) and Holen et al. (2025). Total S(IV) has a diurnal cycle most similar to diesel during weekdays and most similar to gasoline on the weekends (Fig. S21). Furthermore, total S(IV) has a similar diurnal cycle to formaldehyde on weekdays and NH_x on the weekends (Fig. S22). Although SO₂ gas is a precursor to HMS, there are few similarities to the S(IV) and SO₂ diurnal cycles (Fig. S22), which may be because SO₂ gas (campaign mean 9.6 ppb) is often much higher than formaldehyde gas (campaign mean 2.3 ppb), making formaldehyde mixing ratios more often the limiting factor in HMS formation (Dingilian et al., 2024a). The diesel factor is not the largest source of formaldehyde, but formation of HMS related to the gasoline and diesel factors is likely helped by both factors being emitted in the same place, and the mixing of air that occurs from moving vehicles.

As the site in Ketcherside et al. (2025) is residential, we expect VOC species related to vehicle emissions to be lower compared to our downtown site. This is true for formaldehyde, which is on average 60 % lower at the residential site compared to downtown. Assuming the same is true for total ammonia, secondary aerosol formation of S(IV) and other species may be elevated at the downtown site compared to the residential site.

Since the formation of HMS and other S(IV) species are secondary reactions, many other factors need to be considered, and this is not a complete picture of what drives aqueous S(IV) chemistry in Fairbanks. However, this first investigation seems to indicate that vehicle emissions are involved in S(IV) formation.