Complementary aerosol mass spectrometry elucidates sources of wintertime submicron particle pollution in Fairbanks, Alaska, during ALPACA 2022

Fairbanks, Alaska, is a sub-Arctic city that frequently suffers from the non-attainment of national air quality standards in the wintertime due to the coincidence of weak atmospheric dispersion and increased local emissions. As part of the Alaskan Layered Pollution and Chemical Analysis (ALPACA) campaign, we deployed a Chemical Analysis of Aerosol Online (CHARON) inlet coupled with a proton transfer reaction time-of-flight mass spectrometer (PTR-ToF MS) and an Aerodyne high-resolution aerosol mass spectrometer (AMS) to measure organic aerosol (OA) and non-refractory submicron particulate matter (NR-PM₁), respectively. We deployed a positive matrix factorization (PMF) analysis for the source identification of NR-PM₁. The AMS analysis identified three primary factors: biomass burning, hydrocarbon-like, and cooking factors, which together accounted for 28 %, 38 %, and 11 % of the total OA, respectively. Additionally, a combined organic and inorganic PMF analysis revealed two further factors: one enriched in nitrates and another rich in sulfates of organic and inorganic origin. The PTR_CHARON factorization could identify four primary sources from residential heating: one from oil combustion and three from wood combustion, categorized as low temperature, softwood, and hardwood. Collectively, all residential heating factors accounted for 79 % of the total OA. Cooking and road transport were also recognized as primary contributors to the overall emission profile provided by PTR_CHARON. All PMF analyses could apportion a single oxygenated secondary organic factor. These results demonstrate the complementarity of the two instruments and their ability to describe the complex chemical composition of PM₁ and related sources. This work further demonstrates the capability of PTR_CHARON to provide both qualitative and quantitative information, offering a comprehensive understanding of the OA sources. Such insights into the sources of submicron aerosols can ultimately assist environmental regulators and citizens in improving the air quality in Fairbanks and in rapidly urbanizing regional sub-Arctic areas.

1 Introduction

Extremely cold urban regions of the Earth, such as in the Arctic, experience poor dispersion of atmospheric pollution, especially during the wintertime when unique meteorological characteristics, such as extremely low solar radiation and strong radiative cooling at the surface, are coupled with enhanced local anthropogenic emissions from heating, industry, and transport. A good example is the sub-Arctic city of Fairbanks, Alaska, where air quality standards are frequently violated during the winter, with concentrations of fine particulate matter (i.e. with aerodynamic diameters smaller than 2.5 µm; PM_2.5), exceeding the 24 h limit of 35 µg m⁻³ as defined by EPA's National Ambient Air Quality Standards (Dunleavy and Brune, 2020; Fairbanks Air Quality Plan, 2024). Not only is Fairbanks one of the cities with the most polluted wintertime air in the US, but it has also been declared a “moderate non-attainment area” since 2009. Due to the persistence of the problem, it was reclassified as a “severe non-attainment area” in 2017. Increased local anthropogenic emissions and poor atmospheric dispersion due to strong surface-based temperature inversions ($>\mathrm{0.5}\mathrm{°}\mathrm{C}{\mathrm{m}}^{-\mathrm{1}}$ in the lowest 10 m above the ground) are major causes of wintertime pollution in the region (Tran and Mölders, 2011; Mayfield and Fochesatto, 2013). Many research studies have recognized biomass combustion as the major source of aerosol in Fairbanks (Ward et al., 2012; Wang and Hopke, 2014; Kotchenruther, 2016; Ye and Wang, 2020; Haque et al., 2021) that drives overall PM_2.5 concentrations across the city during strong temperature inversion conditions (Robinson et al., 2023). A comprehensive study covering three winters from 2008 to 2011 apportioned 60 %–80 % of PM_2.5 mass at four locations in Fairbanks to emissions from residential wood stoves, the open burning of biomass, outdoor boilers, and other solid-fuel combustion (Ward et al., 2012). Source apportionment of year-round PM_2.5 in the past two decades (2008–2009 (Haque et al., 2021), 2005–2012 (Wang and Hopke, 2014), 2009–2014 (Kotchenruther, 2016), and 2013–2019 (Ye and Wang, 2020)) also revealed wood smoke as a major contributor to PM_2.5 loads: 47.5 % (Haque et al., 2021), 40.5 % (Wang and Hopke, 2014), ∼52 % (Kotchenruther, 2016), and ∼19 % (Ye and Wang, 2020). Wildfire activity and residential wood combustion are the major sources in summer and winter, respectively. The persistent role of wood-burning emissions in shaping the air quality of Fairbanks during winters triggered the implementation of a two-stage burn restriction in 2015 by the Alaska Department of Environmental Conservation (ADEC). The ADEC advisories restricted the operation of solid-fuel heating devices and required alternative heat sources to be used on days with weak atmospheric dispersion and PM_2.5 levels exceeding 25 µg m⁻³, as observed or forecasted (Fye et al., 2009; Czarnecki, 2017; Jentgen, 2022). Sulfate has been observed to be the second-largest component of PM_2.5 mass in Fairbanks (Ward et al., 2012; Wang and Hopke, 2014 ), forming ∼33 % of the annual average PM_2.5 mass (Ye and Wang, 2020). Isotope analyses have revealed 62 % of this PM_2.5 sulfate to be primary (e.g. from residential heating oil combustion) during the winters (Moon et al., 2023).

The aforementioned studies on air quality in Fairbanks have focused on PM_2.5 even though PM₁ has been recognized as the major cause of adverse health effects (Wang et al., 2015; Mainka and Zajusz-Zubek, 2019) due to its capability to spread deeper into the respiratory or cardiovascular system (Meng et al., 2013; Liu et al., 2013; Chen et al., 2017). Currently, efforts to monitor PM₁ are surprisingly scarce, even in “non-attaining” cities such as Fairbanks, underscoring the need for a better characterization of submicron aerosols to understand local sources and chemical composition and ultimately to inform public health and support policy decisions.

Mass spectrometric techniques have advanced over the years, featuring greater mass accuracy, resolving power, and sensitivity. For instance, the Aerodyne high-resolution time-of-flight aerosol mass spectrometer (HR-ToF AMS; called AMS from hereon) is a well-established method for quantifying non-refractory NR-PM₁. Aerosol vaporization at high temperatures and electron ionization result in substantive molecular decomposition, facilitating quantification with a high time-resolution (Decarlo et al., 2006) but at the cost of molecular-level information. This limitation has encouraged the rise of complementary techniques. For instance, extractive electrospray ionization (EESI)-ToF MS has been successfully deployed in Beijing (Tong et al., 2021) and in Zurich to resolve multiple OA sources (Stefenelli et al., 2019a; Qi et al., 2019). Although the instrument provides molecular-level information, its quantitative response is variable and selective for polar species, preventing its independent application for ambient measurements. Other measurement methods, such as a thermal desorption aerosol gas chromatograph (TAG) coupled to a flame ionization detector (FID) and a mass spectrometer (MS) (Williams et al., 2006) and filter inlet for gas and aerosol chemical ionization ToF mass spectrometer (FIGAERO-CIMS) (Lopez-Hilfiker et al., 2014), similarly offer better a chemical resolution than the AMS but at a lower temporal resolution. Semi-continuous measurements, such as those from TAG and FIGAERO-CIMS, may not capture the rapid variation in sources.

To improve the analysis of submicron OA in ambient air, a novel inlet system called the Chemical Analysis for Aerosol Online (CHARON) was developed to collect real-time measurements (Eichler et al., 2015). This inlet minimizes thermal and ionization-induced fragmentation of sampled OA by employing a low-temperature vaporization system (≤ 150 °C) coupled with a relatively softer ionization method, such as the proton-transfer reaction (PTR). The CHARON PTR-ToF MS (called PTR_CHARON from hereon) was successfully used for the characterization of OA from ship exhaust (Eichler et al., 2017); ambient OA in Lyon, France, and Valencia, Spain; and OA source apportionment in Innsbruck, Austria (Müller et al., 2017). Recently, the inlet was used to quantify individual compounds in laboratory-generated secondary organic aerosol (SOA; Lannuque et al., 2023) and complex mixtures, such as vehicular petrol emissions and atmospheric organic matter (Piel et al., 2019; Kostenidou et al., 2024). The system can also measure gas-phase species, creating the opportunity to explore volatile organic compound (VOC) precursor emissions or phase partitioning (Peng et al., 2023; Gkatzelis et al., 2018). Overall, PTR_CHARON and AMS are complementary techniques; the former features molecular-level information about the OA faction but has limited ability to detect particles below 150 nm (Eichler et al., 2015); the latter covers a smaller particle size range (i.e. >60 nm) and detects inorganic components too (Decarlo et al., 2006). Together, they provide an excellent combination of real-time and quantitative data on atmospheric ambient aerosol.

The detailed composition of submicron aerosol in Fairbanks – and other anthropogenically influenced sub-Arctic regions – is still not well understood. To address this issue, we deployed PTR_CHARON and an AMS in the urban centre of Fairbanks during the Alaskan Layered Pollution And Chemical Analysis (ALPACA) campaign as part of the French Climate-relevant Aerosol Sources and Processes in the Arctic (CASPA) project in January–February 2022 (Simpson et al., 2019, 2024). We aimed to determine the composition, concentrations, and sources of atmospheric NR-PM₁. In this paper, we present (i) an intercomparison of the performance of the two instruments focusing on OA quantitation, (ii) the identification of major OA sources in Fairbanks, and (iii) the source apportionment of organic and inorganic aerosol (e.g. ammonium, nitrate, and sulfate). These findings highlight the synergistic benefits of combining multiple analytical techniques and emphasize how soft ionization mass spectroscopic methods enhance molecular-level insights into particulate organic carbon. This integrated approach advances our understanding of the complex composition of particulate matter, offering valuable contributions to environmental characterization and source apportionment studies.

2 Methodology

2.1 Field campaign

The data presented in this study were collected during the ALPACA campaign in Fairbanks, Alaska, US, from 20 January to 26 February 2022. ALPACA is an international collaborative field experiment that aims to understand the sources of outdoor and indoor air pollution in the cold and dark conditions of Fairbanks' winter. The scientific objectives and broad preliminary findings of the experiment were recently reviewed (Simpson et al., 2024). All instruments used for this study were housed in a trailer parked at the Community and Technical College (CTC) of the University of Alaska, Fairbanks (64.84064° N, 147.72677° W; 136 m a.s.l.). The CTC is in the urban core of Fairbanks, close (within 40 m) to a central downtown road and parking area (Simpson et al., 2024); residential activities dominate the west of this locality, while the north and east have commercial activity.

The trailer was equipped with a suite of particle counters and mass spectrometers, featuring high temporal resolutions (ranging from 10 s to 2 min). A scanning mobility particle sizer (SMPS) and a multi-angle absorption photometer (MAAP) were utilized to measure the distribution of particles sized 15.1–661.2 nm and black carbon concentrations, respectively. A separate inlet was used for ${\mathrm{PM}}_{\mathrm{1}}/{\mathrm{PM}}_{\mathrm{2.5}}/{\mathrm{PM}}_{\mathrm{10}}$ measurements conducted with a commercial optical particle counter (OPC; model 1.109, Grimm Aerosol Technik, Germany) at a time resolution of 1 min. Two mass spectrometers, PTR_CHARON (150–1000 nm) and AMS (60–700 nm), were connected to the same inlet that sampled air at 3.5 m above ground level through a short (≈1 m) stainless steel tube with a $\mathrm{1}/{\mathrm{2}}^{\prime \prime }$ (1.27 cm) outer diameter extending through the trailer roof. A HEPA filter was placed upstream of the inlet for 1 h at regular intervals (twice a week) to measure the instrumental background. Additionally, meteorological data, including ambient temperatures at 3 and 23 m, wind speed and direction, and trace gases (namely CO, SO₂, O₃, NO and NO₂), were recorded as described in a previous study associated with the campaign (Cesler-Maloney et al., 2022).

2.2 Instrumentation

2.2.1 PTR-ToF MS: operation and data processing

The OA was quantified with a PTR-ToF MS (PTR-TOF 6000 X2, Ionicon Analytik GmbH, Austria) coupled with a CHARON inlet in near real time at a 20 s temporal resolution, i.e. PTR_CHARON. The CHARON inlet has been described in detail by Eichler et al. (Eichler et al., 2015), and its applications were further evaluated and improved in subsequent studies (Müller et al., 2017, 2019; Leglise et al., 2019; Piel et al., 2019; Peng et al., 2023). Here, the PTR-ToF MS was configured to alternate between the sampling of ambient air to measure VOCs for 15 min (not included in the current study) and the sampling of particulate matter through the CHARON inlet for 45 min. The instrument was operated at a low E$/$N of 65 Td (i.e. drift voltage or pressure; pressure, temperature, and voltage of the drift tube were set at 2.6 mbar, 120 °C, and 265 V, respectively) and in RF mode for optimal sensitivity. The thermodesorber was operated at 150 °C and 8 mbar; this combination of moderate temperature with low pressure expands the range of detection to include extremely low-volatility organic compounds (ELVOCs) as well (Piel et al., 2021). Raw data were obtained, as described in Sect. S1 in the Supplement, and pre-processed with an Ionicon Data Analyzer (IDA; version 1.0.0.2, Ionicon Analytik Ges.m.b.H., Innsbruck, Austria), followed by post-processing (i.e. background subtraction, conversion of raw signal to mixing ratios, temporal averaging, PMF input generation) with an in-house data-processing tool, PeTeR Toolkit (version 6.0; Igor 6.37). The error matrix was also calculated by PeTeR, taking into account uncertainties in ion counts and background signals. Among the resolved 1118 ions spanning the range of $m/z$ 50–425, only 336 were retained above the S$/$N, and 318 ions could be given a molecular formula based on the criteria described in Sect. S2. The PTR-ToF MS records raw signals in counts per second (cps) that were converted to mixing ratios according to the molecular identity determined for the detected ions and their protonation efficiencies (further details in Sect. S1). For comparison with the AMS, mixing ratios were converted to mass concentrations, i.e. µg m⁻³, using Eq. (S2) in the Supplement. Mass concentrations calculated for PTR_CHARON require a critical correction for the enrichment of sampled OA in the aerodynamic lens of the CHARON inlet (Eichler et al., 2015; Müller et al., 2017); further details are provided in Sect. S3. The total (or bulk) OA at a given point in time was the sum of mass concentrations of all ions, which was corrected for fragmentation using a previously reported method (Leglise et al., 2019), which increased the total OA mass concentrations by 17 %.

Species with $m/z>\mathrm{50}$ were retained for the PMF of OA, as molecules of $m/z$ 18–50 were presented in low concentrations, are expected to be too volatile to be present in OA, and were likely detected by PTR_CHARON as artefacts from the denuder. Time series were averaged to 2 min (from 20 s), and two matrices ($m/z\times$ time points) were extracted: (i) ion concentrations and (ii) their measurement uncertainties, using PeTeR. The final matrices had the following dimensions: 336×17 986. Where required, ion intensities (in either ppb or µg m⁻³) were normalized to the sum of all measured intensities.

2.2.2 HR-ToF AMS: operation and data processing

NR-PM₁ was monitored with a 1 min time resolution by a high-resolution time-of-flight aerosol mass spectrometer (AMS) (Aerodyne Research Inc., Billerica, US), extensively described by Decarlo et al. (2006) and Canagaratna et al. (2007). Briefly, ambient particles are sampled through a critical orifice, focused into a narrow beam by an aerodynamic lens, accelerated towards a standard vaporizer heated at 600 °C, and then ionized by electron impact (70 eV at 10⁻⁷ torr). Finally, the ions are analysed by a time-of-flight mass spectrometer. Standard calibrations were performed using 300 nm size-selected dried ammonium nitrate and ammonium sulfate particles at the beginning and the end of the campaign. Nitrate-equivalent values of sample mass concentrations were converted by applying relative ionization efficiencies (RIEs) for organics, nitrates, ammonium, sulfate, and chloride (1.4, 1.1, 3.15, 1.93, and 1.3, respectively). The collection efficiency (CE) has been calculated in PIKA using the composition-dependent CE (CDCE) method, following the approach of Middlebrook et al. (2012). The calculated CE values ranged from 1.00 to 0.35.

Data were averaged to 2 min and extracted as concentration and measurement uncertainty matrices ($m/z\times \text{time}$ points) using SQUIRREL version 1.65 and PIKA version 1.25 in Igor 8.04. Separate matrices (and subsequently PMF) were prepared for organic only (abbreviated as AMS_org) and by combining organic and inorganic species (abbreviated as AMS_{org + inorg}). The inorganic species included in the analyses were nitrate ($m/z$ 30, NO⁺ and 46, ${\mathrm{NO}}_{\mathrm{2}}^{+}$), sulfate ($m/z$ 48, SO⁺; 64, ${\mathrm{SO}}_{\mathrm{2}}^{+}$; 80, ${\mathrm{SO}}_{\mathrm{3}}^{+}$; 81, ${\mathrm{HSO}}_{\mathrm{3}}^{+}$; and 98, ${\mathrm{H}}_{\mathrm{2}}{\mathrm{SO}}_{\mathrm{4}}^{+}$), ammonium ($m/z$ 15, NH⁺; 16, ${\mathrm{NH}}_{\mathrm{2}}^{+}$; and 17, ${\mathrm{NH}}_{\mathrm{3}}^{+}$), and chloride ($m/z$ 35, Cl⁺ and 36, HCl⁺). Error matrices were calculated by PIKA based on uncertainty in ion counts, background signal, air beam correction, and electronic noise (Sueper, 2014). Atomic O$/$C and H$/$C ratios were calculated based on established methods (Aiken et al., 2007, 2008; Canagaratna et al., 2015). Where needed for comparison with PTR_CHARON, mass concentrations of polycyclic aromatic hydrocarbons (PAHs) were estimated from fragments as described previously (Herring et al., 2015) and levoglucosan was estimated as detailed in Sect. S4.

Species with $m/z$ 12–120 were retained for PMF in this study, excluding important PAHs detected up to $m/z$ 252; such PAHs were used as external tracers for factor identification. All PAHs were included in total OA quantification and associated comparisons. This exclusion is expected to cause underestimation below 2 % of the mass of some factors, particularly hydrocarbon-like organic aerosol (HOA) and biomass-burning organic aerosol (BBOA). Final matrices from the AMS_org and AMS_{org + inorg} analyses had the following dimensions: 193×24 762 and 205×24 762, respectively.

2.3 Source apportionment: positive matrix factorization

Source apportionment was performed using a PMF implemented in the multilinear engine (ME-2) (Paatero, 1997a, 1999). The PMF was configured and analysed using the SoFi (Source Finder) Pro interface (Canonaco et al., 2013) (version 8.4.1.9.1; Igor 8.04). PMF is a descriptive mathematical algorithm that describes the input data, i.e. measurements of several variables collected over time (here, $m/z\times \text{sampling}$ time points), as a linear combination of factors that have constant mass spectra associated with temporally varying concentrations of the spectral constituents (Paatero, 1997b; Paatero and Tapper, 1994). The mathematical expressions and functions of the PMF algorithm have been exhaustively detailed in previous studies (e.g. Tong et al., 2021; Stefenelli et al., 2019a; Chen et al., 2022; Chazeau et al., 2022). Below, we summarize the user-defined configurations applied in SoFi Pro to optimize the PMF of our datasets, PTR_CHARON, AMS_org, and AMS_{org + inorg}.

General methodology for PMF analysis

A preliminary PMF was performed without using a priori information to explore factor variability and source contributions to guide the selection of an optimal solution before applying constraints. We considered solutions ranging from 3 to 13 factors, applying a step-wise, cell-wise down-weighting approach: variables with $\text{S}/\text{N}<\mathrm{0.2}$ (“bad” variables) were down-weighted by a factor of 10, while those with $\mathrm{0.2}<\text{S}/\text{N}<\mathrm{2}$ (“weak” variables) were down-weighted by a factor of 2 (Paatero and Hopke, 2003; Ulbrich et al., 2009). Upon establishing some primary factors, such as cooking and biomass burning, which were successfully identified in unconstrained trials, we narrowed the range of possible solutions by applying the a-value approach, which allows for improved factorization by constraining the PMF with external data when available (Canonaco et al., 2013; Paatero, 1999). For instance, a factor profile from a PMF trial in the same experiment, a time series from an external tracer, or a well-established factor profile for a source from another experiment may be provided to the PMF as an “anchor” or “vector”, around which it can build a factor in its overall solution. The extent to which each PMF factor can diverge from the anchor is defined by the value of a (Tong et al., 2021), which varies from 0 to 1. This anchor can be provided for one or multiple factors and has been proven to improve the quality of PMF solutions compared to unconstrained trials (Tong et al., 2021; Stefenelli et al., 2019a; Chen et al., 2022).

Currently, there are no objective criteria for choosing the optimal number of factors; however, several criteria have been proposed in the literature to inform an appropriate choice (Chen et al., 2022; Zhang et al., 2011; Ulbrich et al., 2009; Crippa et al., 2014). The PMF solutions reported here were primarily selected based on their physical meaning, which was determined by the presence of known tracer compounds in the factors and temporal correlation with co-located measurements of external tracers (e.g. NO_x, SO₂). We selected eight-, four-, and six-factor solutions from PTR_CHARON, AMS_org, and AMS_{org + inorg}, respectively. The justification for these solutions is presented in Table S2 in the Supplement. Once the most suitable solution, i.e. the base-case, was established, bootstrap analyses were performed to assess its stability, evaluate uncertainties, and conduct a sensitivity analysis on the range of a values used. In an unblocked bootstrapping approach, the original matrices (both data and error) are perturbed by random resampling of the rows to create a new input of the same dimensions, resulting in some duplications and deletions throughout the input (Paatero et al., 2014). The need and application of this approach differed between PTR_CHARON and the two AMS datasets, as discussed in Sects. S5 and S6, respectively. Ancillary data on particle size distribution have been associated with mass spectrometry data in an additional PMF analysis (Sect. S7 in the Supplement). Finally, the quality of the solutions was assessed by the $Q/Q_{\text{exp}}$ values and from key diagnostic plots of residuals, as well as from the statistical stability across multiple runs (Figs. S5–S7 in the Supplement).

3 Results and discussion

3.1 Campaign overview

Figure 1 summarizes the meteorological conditions, chemical composition, and particle size distribution of NR-PM₁ observed from 20 January to 26 February 2022. High aerosol loads coincided with poor atmospheric dispersion due to low wind speeds ($<\mathrm{2}\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$) and low temperatures (below −10 °C), associated with strong surface temperature inversions. The temperature differences between 23 and 3 m a.s.l. ranged from 3 to 10 °C. The average values of BC and NR-PM₁, measured with MAAP and AMS, were $\mathrm{1.4}\pm \mathrm{1.4}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ and $\mathrm{8.3}\pm \mathrm{9.3}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$, respectively. During intense pollution events, the daily average concentrations of NR-PM₁ were 24–27 µg m⁻³. During the same sampling period at the NCore site (Fairbanks), PM_2.5 values of ∼25 and $\sim \mathrm{29}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ were reported (Robinson et al., 2023). Ancillary OPC measurements at the CTC site showed that the hourly PM₁ mass comprised up to 99 % of the PM_2.5. Organic matter was the predominant component of NR-PM₁ throughout the campaign, constituting $\sim \mathrm{66}\pm \mathrm{11}\mathit{\%}$ of its total mass, while chloride, ammonium, nitrate, and sulfate contributed 2±3 %, 3±3 %, 6±4 %, and 22±10 %, respectively. This finding aligns with previous studies in Fairbanks, where OA was the largest component of PM_2.5 mass (Ward et al., 2012; Ye and Wang, 2020; Robinson et al., 2024). Specifically, according to a recent study conducted from 2020 to 2021, ACSM analysis during wintertime demonstrated that inorganics formed less than 25 % of the PM_2.5 mass, with sulfate (∼10 %) and nitrate (∼ 8 %) being the predominant components (Robinson et al., 2024). Despite the different average concentrations, the fractional contributions of these non-refractory components remained almost invariable throughout the campaign (Fig. 1d). The detailed molecular-level composition of organics with PTR_CHARON reveals that a large majority of organics comprise only C, H, and/or O atoms, while only $\sim \mathrm{9}\pm \mathrm{4}\mathit{\%}$ of the OA_CHARON mass measured with this instrument was attributable to heteroatomic molecules, including organonitrates and organosulfates (Figs. 4 and S8). Generally, heteroatomic species cannot be distinguished at a resolving power of 5000 FWHM in complex environmental mixtures, such as atmospheric aerosol (Reemtsma, 2009). In this study, based on the low formula error and the lack of an appropriate alternate, we gave 53 low-concentration ions (<2 % of the total signal) CHOS or CHNO identities. However, due to the low confidence in their formula assignments, they were not considered for factor identification. Prominent peaks include $m/z$ 217.09 (C₁₂H₁₂N₂O₂), 219.09 (C₁₅H₁₀N₂), 123.05 (C₄H₁₀O₂S), and 151.08 (C₆H₁₄O₂S).

Figure 1 Overview of meteorological parameters and aerosol properties. The shaded areas indicate the periods when Stage 1 (red) and Stage 2 (black) advisories (“burn bans”) from the Alaska Department of Environmental Conservation were in place in Fairbanks. (a) Ambient temperature at 3 and 23 m and difference in temperature between the two heights, (b) wind speed and direction with the daily sunlight in terms of the NO₂ photolysis rate coefficient ($J_{{\mathrm{NO}}_{\mathrm{2}}}$) (Simpson et al., 2024), (c, d) absolute and fractional compositions of non-refractory fine particulate matter (NR-PM₁) from the AMS, and (e) size distribution of PM₁ from the SMPS.

On average, the OA mass loading recovered by PTR_CHARON (i.e. OA_CHARON) accounted for approximately 85 % of the OA mass measured by the AMS (i.e. OA_AMS). While the two instruments showed a good temporal agreement ($R^{\mathrm{2}}=\mathrm{0.60}),$ as depicted in Fig. 2a and b, measurements were biased either towards AMS_org or PTR_CHARON (i.e. distributed away from the 1:1 line in the scatter plot of Fig. 2c) during different periods of the campaign. These trends could be explained by the variation in the relative contributions of two major emission sources identified by both instruments in this study: on-road transport and biomass burning. OA_CHARON was comparable to OA_AMS, when the relative contribution of BBOA_{AMS, org} was more than 50 % of total OA_AMS and HOA_{AMS, org} (i.e. transport_CHARON) was less than 10 % (Fig. 2d and e). Similar trends were observed for some major constituents of BBOA, e.g. levoglucosan and a PAH (C₂₀H₁₂), as shown in Fig. S9. Part of such discrepancy can be traced back to the size transmission of particles, where sub-100 nm urban vehicular emissions are underestimated by PTR_CHARON (Guo et al., 2020; Pikridas et al., 2015; Louis et al., 2017; Kostenidou et al., 2020) and larger than 100 nm biomass burning emissions (Reid et al., 2005) are estimated well (Janhäll et al., 2010). Another part of the quantitative difference can be explained by the PTR limitation in ionization and the induced fragmentation of analyte ions. Tests conducted in our laboratory with five C₁₆–C₂₆ alkanes as markers of vehicular emissions revealed that they undergo extensive fragmentation, resulting in 2–4 times underestimation of their actual concentrations. In line with this, the ineffective ionization of saturated alkanes by PTR (Ellis and Mayhew, 2014) and their tendency to undergo dissociative ionization (Gueneron et al., 2015) have also been reported.

Figure 2 Comparison of total OA measured with PTR_CHARON and the AMS. (a) Absolute concentrations of OA measured with the AMS and OA_corr (fragmentation-corrected OA) from PTR_CHARON, (b) daily average concentrations of OA, and (c) scatter plot of total OA measured with the AMS and PTR_CHARON. Data points are coloured by the dates, and the legend is written as MM/DD/YY. Data points are sized by the geometric mean mass of the dM/dlogDp from SMPS (50–500 nm). The dashed line denotes the 1:1 relationship. Coefficients a and b denote the slope and the intercept for the linear regression (p≤ 0.05; solid line) and are written with ± 1 standard deviation.

3.2 Source apportionment

3.2.1 Overview of source apportionment

A four-factor solution was selected for the AMS_org measurements with three primary factors (i.e. hydrocarbon-like organic aerosol, HOA; cooking organic aerosol, COA; and biomass burning organic aerosol, BBOA) and an oxygenated or aged OA factor (i.e. OOA). The mass spectra and time series are presented in the Supplement (Fig. S10). The counterparts of these four factors were diagnosed in AMS_{org + inorg} based on a high temporal correlation (R²>0.9; Table S4 in the Supplement), along with two additional factors: a sulfur-rich factor (labelled sulf-OA) and a nitrate-rich factor (labelled AmNi) (Fig. 3). An eight-factor solution was selected for PTR_CHARON and is summarized in Figs. 4 and 5. To differentiate between corresponding factors retrieved from the different datasets, they have been assigned unique subscripts, e.g. COA_{AMS, org}, COA_{AMS, org + inorg}, or COA_AMS (i.e. referring to both AMS datasets) or COA_CHARON. Amongst the three datasets, COA, HOA (labelled “transport” in PTR_CHARON analyses), and OOA were common. A single BBOA factor was observed in AMS_org and AMS_{org + inorg}, while four chemically distinct but closely co-varying counterparts were detected by PTR_CHARON.

Figure 3 Overview of the positive matrix factorization output for NR-PM₁ measurements with AMS_{org + inorg}. Mass concentrations were normalized to the sum of the concentrations of all ions. Time series are overlaid with those of the corresponding factor (if available) in AMS_org and PTR_CHARON analysis or an external tracer. Correlation coefficients (R²; p≤0.05) are also provided, and slopes can be found in Table S5 in the Supplement or Table 1.

Figure 4 Normalized mass spectra of factors from the PMF of PTR_CHARON measurements. Mass concentrations are normalized to the sum of concentrations of all ions. Peaks are coloured by the molecular group (CHO, CHNO, CHOS, CH, CHN) of the formula assigned. Unassigned species are shown in black. Further information, such as tentative identities and formula errors, can be found in dataset 1 in the Supplement.

Figure 5 Diurnal profiles and complete time series of factors from the positive matrix factorization of PTR_CHARON measurements. In the second column, time series are overlaid on those of the corresponding factor in AMS_org and an external tracer or marker ion. Correlation coefficients (R²; p≤0.05) are also provided, and slopes can be found in Table S4 in the Supplement or Table 1.

Table 1 Linear regression (R²; p≤0.05) between the time series of factors derived from (a) PTR_CHARON, (b) AMS_org, and (c) AMS_{org + inorg} measurements, with external tracers and chemical species (S- and N-containing species and PAHs) measured with the AMS.

3.2.2 Organic aerosol from residential heating

Both AMS analyses indicate that biomass burning was a major source of PM₁ during the ALPACA campaign. On average, BBOA contributed $\mathrm{1.5}\pm \mathrm{1.9}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ (28±18 % of total OA_AMS) and $\mathrm{1.6}\pm \mathrm{2.2}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ NR-PM₁ (19±14 % of total NR-PM₁ mass). The mass spectra of BBOA_AMS featured a strong peak at $m/z$ 60 (${\mathrm{C}}_{\mathrm{2}}{\mathrm{H}}_{\mathrm{4}}{\mathrm{O}}_{\mathrm{2}}^{+}$) and 73 (${\mathrm{C}}_{\mathrm{3}}{\mathrm{H}}_{\mathrm{5}}{\mathrm{O}}_{\mathrm{2}}^{+}$) (Fig. S10A and B). These fragments are markers of anhydrosugars in wood-forming polymers, such as cellulose. Wood combustion has previously been estimated to be the largest emitter of aerosols in Fairbanks and surrounding areas, where it may produce as much as 80 % of the aerosol load. Wood-burning emissions are also the major driver of the spatial variability of PM_2.5 and BC in Fairbanks during strong atmospheric temperature inversions. Other typical residential heating sources of emissions in Fairbanks include coal, gas, and fuel oil.

The BBOA_AMS factor was strongly correlated with PAHs (R²≥0.7), while a moderate correlation was observed with SO₂ (R²=0.4; Table 1). While PAHs are a major component of biomass combustion emissions, the emission of SO₂ is largely associated with coal and oil combustion (Smith et al., 2011; Dunleavy and Brune, 2019). However, the AMS was unable to distinguish between multiple combustion-related sources. As shown in the diurnal plots in Fig. 3, the concentration of the BBOA_AMS factor enhanced at ∼1800 AKST stayed stable through the night and then decreased in the early morning. Its lowest mass concentrations occurred during the afternoon (13:00–15:00 AKST). Therefore, BBOA_AMS was associated with residential heating – the combustion of various fuels by residents within their homes (non-commercially), such as in wood-burning stoves, furnaces, and boilers, for heating their living spaces. We did not find evidence of OA or NR-PM₁ from commercial heat sources, such as power plants, likely due to their small contribution to surface-level aerosol, as smokestacks typically lie above the inversion layer.

PTR_CHARON apportioned $\mathrm{2.6}\pm \mathrm{3.4}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ of OA_CHARON, on average, to four distinct residential heating-related sources expressed as ResH1–4 (62±26 % of total OA_CHARON). These factors closely co-varied in time and were correlated reasonably well (R² = 0.5–0.7; Table S5) with the BBOA_AMS factor. In addition, combining all four residential heating-related factors in PTR_CHARON into a composite factor increased the correlation (R²) with AMS_org and AMS_{org + inorg} to 0.79 and 0.82, respectively, suggesting that PMF was unable to effectively separate these closely co-varying residential heating factors in the AMS dataset.

The four factors from PTR_CHARON were identified as different sources based on the distribution of key marker species and their correlation with external (e.g. trace gases) and internal (e.g. PAH) influences. Levoglucosan is used here as an internal tracer of biomass burning, being relatively stable under atmospheric conditions (Fraser and Lakshmanan, 2000). Protonated levoglucosan ($m/z$ 163) and its fragments (at $m/z$ 85, 127, and 145) were found in ResH1, ResH4, and ResH2 with 30 %, 26 %, and 14 % of the total signal, respectively (Fig. S11), suggesting that they originate from biomass wood-burning (Figs. 4 and S11). These three factors collectively accounted for $\mathrm{2.1}\pm \mathrm{2.5}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ (47±20 %) of total factorized OA_CHARON.

ResH1 includes low-temperature combustion markers. This factor is small as it contributes only an average of $\mathrm{0.5}\pm \mathrm{0.5}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ (14 %) of the total OA_CHARON, but it contains the highest fraction of levoglucosan (∼30 %). Approximately 65 % of the total signal of ResH1 is due to compounds with six or fewer carbon atoms, compared to heavier species present in the other factors (Fig. S13). The most abundant species are at $m/z$ 69.03 (C₄H₄O; furan) (Palm et al., 2020; Jiang et al., 2019), $m/z$ 87.04 (C₄H₆O₂; oxobutanal) (Brégonzio-Rozier et al., 2015), $m/z$ 97.03 (C₅H₄O₂; furfural), $m/z$ 109.0286 (C₆H₄O₂; benzoquinone) (Stefenelli et al., 2019b), and $m/z$ 115.04 (C₅H₆O₃; methyl-dihydrofuran) (Koss et al., 2018). Consistent with these molecular formulae, the concentration-weighted average O$/$C of ResH1 was relatively higher (i.e. 0.42) compared to other residential heating factors (O/C = 0.2–0.3). The most abundant species observed in ResH1 can be attributed to depolymerization reactions occurring during low temperatures and the early stages of the combustion process (Collard and Blin, 2014; Sekimoto et al., 2017).

ResH2 and ResH4 include OA from hardwood and pinewood combustion, respectively. Two more factors associated with wood burning were ResH2 and ResH4. Their average concentrations were 1.1±1.9 and $\mathrm{0.8}\pm \mathrm{0.9}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$, respectively, corresponding to 28 % and 20 % of OA_CHARON (Fig. 6). ResH2 was the dominant factor in the PMF and could reach $\sim \mathrm{37}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ during the most severe pollution episodes. Both factors were associated with particles greater than 300 nm (Fig. S12), typical of wood smoke (Glasius et al., 2006), and presented unique molecular signatures of different wood types (Fig. S11). Generally, the specific nature of wood cannot be inferred unambiguously because the emissions of known marker species, such as levoglucosan or methoxy phenols, vary not just with fuel used and its quality but also with the type of heating appliance, operational conditions, appliance efficiency, and stage in the combustion cycle (Fine et al., 2002; Alves et al., 2017). Regardless, several studies (Fine et al., 2002; Schauer and Cass, 2000; Kawamoto, 2017) have distinguished between softwood and hardwood by investigating the presence of marker compounds, which were also observed in our study, such as substituted phenols and resin acids (Fig. S11).

Figure 6 Campaign averages of mass concentrations apportioned to each factor in (a) PTR_CHARON, (b) AMS_org, and (c) AMS_{org + inorg} analyses. Slices of pies are equivalent to the average absolute concentrations. A complete time series of fractional contributions can be found in Fig. S14.

ResH2 featured an abundance of methoxy phenols, including C₇H₈O₂ (guaiacol), C₈H₁₀O₃ (syringol), C₁₀H₁₀O₃ (conifer aldehyde), C₆H₆O₂ (benzenediol or methyl-furfural), and C₈H₁₀O₂ (creosol), which collectively accounted for ∼9 % of the total signal compared to 1,%, 2,%, and 2 % in ResH1, ResH3, and ResH4, respectively. These compounds are important products of lignin pyrolysis in birch, aspen, and spruce and are usually detected in the gas phase at mild ambient temperatures (Kong et al., 2021). Guaiacol and syringol are depolymerization products of guaiacyl and syringyl units of lignin at 200–400 °C, and they rapidly transition to catechols, cresols, and phenols during secondary pyrolysis reactions at 400–450° C (Kawamoto, 2017). While guaiacols are emitted by both hardwood and softwood, semi- or low-volatility substituted syringols are emitted in higher amounts by hardwood combustion (Kawamoto, 2017; Fine et al., 2002, 2001; Schauer and Cass, 2000). In this study, derivatives of guaiacols, including C₁₀H₁₂O₂ (eugenol), C₁₀H₁₄O₂ (4-propyl guaiacol), and C₁₀H₁₀O₃ (conifer aldehyde), presented higher “relative concentration” (Eq. S3 in the Supplement) of 0.56–1.41 for ResH2 and ResH4 compared to ResH1 (<0). Other compounds, such as C₈H₈O₃ (vanillin), C₉H₁₀O₃ (acetovanillone), C₁₀H₁₂O₃ (propiovanillone), and C₁₀H₁₂O₄ (methyl-homovanillate), were predominantly found in ResH2. Similarly, substituted syringols, i.e. C₁₁H₁₄O₃ (methoxy eugenol), C₁₀H₁₂O₄ (acetosyringone), and C₁₁H₁₄O₄ (syringyl acetone, propionyl syringol, or sinapyl alcohol), were almost entirely associated with ResH2 as well. These compounds have been reported as markers of hardwood burning (Fine et al., 2001), implying a greater contribution of hardwood emissions to the ResH2 factor. In Alaska, relevant hardwood species include deciduous leafy trees such as paper birch, balsam poplar, and quaking aspen. (Alaska Department of Environmental Conservation, 2023).

The ResH4 factor presented a unique fingerprint characterized by oxygenated molecules bearing more than 13 carbon atoms (Fig. S13), such as C₁₆H₃₀O₆ ($m/z$ 319.21), C₂₀H₂₈O₂ ($m/z$ 301.21), C₂₀H₁₈O₄ ($m/z$ 323.12), C₂₀H₃₀O₂ ($m/z$ 303.24), and C₂₂H₁₈O ($m/z$ 299.14), in addition to the levoglucosan marker ions (26 % of the total signal). The intense signals from $m/z$ 301 (C₂₀H₂₈O₂) and $m/z$ 303 (C₂₀H₃₀O₂) (Fig. S11) are likely related to resin acids, dehydroabietic acid, and abietic acid, respectively, which are almost exclusively emitted from the thermal alteration of resins in coniferous species, and thus are indicative of softwood burning (Simoneit, 2002, 1999). Due to the presence of these compounds, ResH4 was interpreted as an OA factor influenced by softwood combustion. Softwood species in Alaska include trees with needles and cones, such as hemlock, cedar, and spruce (Adec, 2023).

ResH3 includes OA from heating oil combustion. This factor contributed to 16±9 % of the total OA_CHARON ($\mathrm{0.6}\pm \mathrm{0.6}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$). It showed the characteristic diurnal pattern of residential heating as it correlated quite well (R²=0.56) with BBOA_{AMS, org}. However, its chemical composition differed from that of the other residential heating factors. Notably, levoglucosan contributed to a smaller fraction of the total signal of ResH3 (i.e. 9 %) compared to other residential heating factors (14 %–30 %; Fig. S11), while PAHs represented a much larger fraction of its total signal (for instance, 30 %, 31 %, and 29 % of C₁₆H₁₀ ($m/z$ 203.09), C₁₈H₁₂ ($m/z$ 229.10), and C₂₀H₁₂ ($m/z$ 253.10); Fig. S13). These PAHs could be fluoranthene (or pyrene), naphthacene (or benzo(x)anthracene, chrysene), and benzo(x)pyrene (or benzo(x)fluoranthene), which have been reported in emissions of light oil combustion (Bari et al., 2009). Additionally, ResH3 was strongly correlated with SO₂ (R²=0.61), compared to a lower correlation (R²≤0.47) with the other residential heating factors. The residential combustion of heating oil is an important source of SO₂ in Fairbanks, compared to wood and coal. This is because ∼ two-thirds of households use oil-fired space heaters with high sulfur content of >1600 ppm (e.g. no. 1 and no. 2 fuel oil and waste motor oil) (Dunleavy and Brune, 2019). Consistent with the possibility of the ResH3 factor being associated with fuel oil emissions, the factor is characterized by particles smaller than 100 nm (Fig. S12), and due to the CHARON inlet's cut-off, its mass concentration was possibly underestimated.

3.2.3 Hydrocarbon-like organic aerosol

The HOA_AMS factors were characterized by notable peaks at $m/z$ 43 (${\mathrm{C}}_{\mathrm{3}}{\mathrm{H}}_{\mathrm{7}}^{+}$), 57 (${\mathrm{C}}_{\mathrm{4}}{\mathrm{H}}_{\mathrm{9}}^{+}$), 71 (${\mathrm{C}}_{\mathrm{5}}{\mathrm{H}}_{\mathrm{11}}^{+}$), 85 (${\mathrm{C}}_{\mathrm{6}}{\mathrm{H}}_{\mathrm{13}}^{+}$), and 99 (${\mathrm{C}}_{\mathrm{7}}{\mathrm{H}}_{\mathrm{15}}^{+}$), belonging to the $[{\mathrm{C}}_{\mathit{\text{n}}}{\mathrm{H}}_{\mathrm{2}\mathit{\text{n}}+\mathrm{1}}{]}^{+}$ series, typical of n- and branched alkanes. Additional peaks at $m/z$ 55 (${\mathrm{C}}_{\mathrm{4}}{\mathrm{H}}_{\mathrm{7}}^{+}$), 69 (${\mathrm{C}}_{\mathrm{5}}{\mathrm{H}}_{\mathrm{9}}^{+}$), 81 (${\mathrm{C}}_{\mathrm{6}}{\mathrm{H}}_{\mathrm{9}}^{+}$), 83 (${\mathrm{C}}_{\mathrm{6}}{\mathrm{H}}_{\mathrm{11}}^{+}$), 95 (${\mathrm{C}}_{\mathrm{7}}{\mathrm{H}}_{\mathrm{11}}^{+}$), 97 (${\mathrm{C}}_{\mathrm{7}}{\mathrm{H}}_{\mathrm{13}}^{+}$), 107 (${\mathrm{C}}_{\mathrm{8}}{\mathrm{H}}_{\mathrm{11}}^{+}$), 109 (${\mathrm{C}}_{\mathrm{8}}{\mathrm{H}}_{\mathrm{13}}^{+}$), and 111 (${\mathrm{C}}_{\mathrm{8}}{\mathrm{H}}_{\mathrm{15}}^{+}$) represented the $[{\mathrm{C}}_{\mathit{\text{n}}}{\mathrm{H}}_{\mathrm{2}\mathit{\text{n}}-\mathrm{1}}{]}^{+}$ and $[{\mathrm{C}}_{\mathit{\text{n}}}{\mathrm{H}}_{\mathrm{2}\mathit{\text{n}}-\mathrm{3}}{]}^{+}$ series, related to cycloalkanes (McLafferty and Tureč̌ek, 1993). These ions are associated with engine-lubricating oils, vehicular exhaust, and diesel fuel (Canagaratna et al., 2004). The HOA_AMS factors contributed 38±20 % and 21±14 % of the OA_AMS and NR-PM₁ mass, respectively (Figs. 6 and S14).

The unconstrained PTR_CHARON analysis was unable to apportion a road-transport-related factor; however, by constraining the factorization with the time series of a mobile petrol factor, identified in the gas-phase PTR-ToF MS analyses of the ALPACA campaign (Temime-Roussel et al., 2022), a small road-transport-associated factor was identified. For instance, the latter was strongly correlated with black carbon and NO_x (R² of 0.58 and 0.66; Table 1) and featured high contributions of C₈H₁₀ (xylene, ethylbenzene), C₇H₈ (toluene), and C₆H₆ (benzene) (Figs. 4 and S11). Despite exhibiting some reasonable diurnal trends peaking during the morning (09:00 AKST) and evening (17:00–16:00 AKST) rush hours (Fig. 5), the factor accounted for negligible concentrations ($<\mathrm{1}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$) and presented some unlikely species, such as $m/z$ 315.22 (C₂₁H₃₀O₂; possibly cannabidiol) absent in the unconstrained PMF. For instance, on average, $\mathrm{2.1}\pm \mathrm{3.0}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ of OA was associated with HOA_{AMS, org}, compared to only $\mathrm{0.1}\pm \mathrm{0.1}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ for the road transport_CHARON factor (Fig. 6). These discrepancies are largely instrumental, partly due to the poor transmission of the small particles (<100 nm) by the CHARON inlet and the limited sensitivity towards hydrocarbons by PTR, but other possible biases can be due to the heating oil OA signal interfering with HOA_AMS, as discussed in Sect. S8.

3.2.4 Cooking organic aerosol

Another primary factor identified in Fairbanks was cooking, likely arising from residential or commercial activities around the CTC. The COA_AMS factor featured a high abundance of ${\mathrm{C}}_{\mathit{\text{x}}}{\mathrm{H}}_{\mathit{\text{y}}}^{+}$ ions, along with prominent O₁ fragments at $m/z$ 55 (C₃H₃O⁺), 84 (C₅H₈O⁺), and 98 (C₆H₁₀O⁺), originating from organic acids (Mohr et al., 2009) and used as diagnostic markers of COA in urban settings (Sun et al., 2011). The $f\mathrm{55}/f\mathrm{57}$ value (i.e. the ratio of fractions of ${\mathrm{C}}_{\mathrm{4}}{\mathrm{H}}_{\mathrm{7}}^{+}$ to ${\mathrm{C}}_{\mathrm{4}}{\mathrm{H}}_{\mathrm{9}}^{+}$) was ∼3.00 for COA_AMS, compared to ∼1.04 for HOA_AMS (Fig. S10D). As a reliable tracer for COA remains unidentified in the AMS spectrum, a $f\mathrm{55}/f\mathrm{57}\text{ratio}>\mathrm{1}$ is considered a characteristic feature (Katz et al., 2021; Sun et al., 2011). The PMF analysis of PTR_CHARON also revealed a distinct COA factor dominated by the long-chain fatty acids C₁₈H₃₂O₂, C₁₈H₃₄O₂, and C₁₈H₃₆O₂, identified as linoleic, oleic, and stearic acids, respectively, contributing 11 %, 16 %, and 4 % to the total COA_CHARON mass (Figs. 4 and S11). These fatty acids are commonly found in OA from cooking oil and meat (Katz et al., 2021; Mohr et al., 2009). COA_CHARON contributed a maximum of ∼9 % of the total OA_CHARON mass and exhibited a unique diurnal pattern visualized in Fig. 5, with a minor maximum in the afternoon (lunchtime) and a second maximum in the evening (dinnertime). The average absolute concentrations of COA were 0.6±0.8 for the AMS and $\mathrm{0.1}\pm \mathrm{0.2}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ for PTR_CHARON. Such a discrepancy can be explained by the same reason discussed above and detailed in Sect. S9.

3.2.5 Oxygenated organic aerosol

Past source apportionment studies have reported multiple OOA factors differing in volatilities or oxygenation levels (e.g. Stefenelli et al., 2019a; Kumar et al., 2022; Cash et al., 2020). Here, we diagnosed only a single OOA factor in either AMS or PTR_CHARON measurements. Specifically, the OOA_AMS factors were identified based on a prominent peak at $m/z$ 43 (C₂H₃O⁺) and $m/z$ 29 (CHO⁺; Fig. S10A) and showed a strong correlation (R²=0.74) with OOA_CHARON. The average absolute concentrations of OOA_CHARON and OOA_{AMS, org} were 0.4±0.6 and $\mathrm{1.0}\pm \mathrm{2.1}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$, respectively. Notably, the most intense ions in the mass spectrum of OOA_CHARON have been tentatively assigned to $m/z$ 73.03 (C₃H₅O₂, e.g. methyl-glyoxal), $m/z$ 99.04 (C₅H₆O₂, e.g. oxo-pentanal), $m/z$ 113.06 (C₆H₈O₂, e.g. methyl-oxo-pentanal), $m/z$ 127.08 (C₇H₁₀O₂; e.g. heptadienoic acid), $m/z$ 137.06 (C₈H₈O₂; e.g. methoxy-benzaldehyde), $m/z$ 167.10 (C₁₀H₁₄O₂), $m/z$ 171.07 (C₈H₁₀O₄), and $m/z$ 185.10 (C₁₃H₁₂O; e.g. benzyl phenol). Among these compounds, some have previously been associated with the atmospheric oxidation or photolysis of BBOA (Montoya-Aguilera et al., 2017; Lignell et al., 2013; Smith et al., 2020), while others could be due to the oxidation of aromatic VOCs originating from road transport (Temime-Roussel et al., 2022). Few other species overlapped with the residential heating tracers, notably $m/z$ 163.06 (C₆H₁₀O₅; e.g. levoglucosan), $m/z$ 179.08 (C₁₀H₁₀O₃; e.g. conifer aldehyde), and $m/z$ 301.21 (C₂₀H₂₈O₂; e.g. dehydroabietic acid). However, given the prominence of wood burning as a major primary emission, the OOA is likely linked to BBOA. A recent study in Fairbanks identified wintertime OOA as a mixture of BBOA and SOA formed from non-photochemical processing using ACSM (Robinson et al., 2024). The examination of f44 versus f60 in the AMS_org dataset plot (Fig. S10C) is consistent with aged OOA derived from biomass burning, as previously demonstrated by Xu et al. (2023). Another recent source apportionment study using the HR-ToF AMS at a site close to the CTC did not identify an OOA factor, while BBOA, HOA, and a mixed primary factor (HOA, COA, etc.) comprised 45 %, 25 %, and 31 % of total OA, respectively (Yang et al., 2024). A limited OOA formation is plausible due to reduced solar light exposure during this period of the year (Cesler-Maloney et al., 2024). However, the absence of OOA is more likely a result of an unresolved organic fraction.

Sulfate and OOA. An intriguing insight about the OOA factor emerged from the AMS_{org + inorg} measurements, indicating the significant content of sulfur-containing compounds (Fig. S15). The AMS does not distinguish between the different sulfur-containing species, but following guidelines from previous works (Chen et al., 2019; Schueneman et al., 2021), we could explore the ratio of sulfur fragments to investigate the presence of different species such as hydroxymethanesulfonate (HMS; CH₂(OH)SO⁻³), ${\mathrm{HSO}}_{\mathrm{4}}^{-}$ (bisulfite), ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ (sulfate), and H₂SO₄ (sulfuric acid). An organosulfate content of $\sim \mathrm{0.8}\pm \mathrm{1.3}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ ($\sim \mathrm{20}\pm \mathrm{16}\mathit{\%}$) was then derived using the ratios of SO⁺ and ${\mathrm{SO}}_{\mathrm{2}}^{+}$ ions against ${\mathrm{SO}}_{\mathrm{3}}^{+}$, ${\mathrm{HSO}}_{\mathrm{3}}^{+}$, and ${\mathrm{H}}_{\mathrm{2}}{\mathrm{SO}}_{\mathrm{4}}^{+}$ ions, as detailed by Song et al. (2019). This value is in good agreement with previous reports from the same field campaign (Campbell et al., 2022; Robinson et al., 2024). Additionally, to mimic the potential matrix effects of wood-burning OA on sulfate fragmentation patterns, AMS spectra from a solution of (NH₄)₂SO₄ mixed with various amounts of levoglucosan (i.e. 0 %–80 % in mass) were compared to ambient data and PMF factors, as shown in Fig. S16a. Among the sulfate-rich factors, OOA_{AMS, org + inorg} exhibited lower ${\mathrm{HSO}}_{\mathrm{3}}^{+}$ to ${\mathrm{H}}_{\mathrm{2}}{\mathrm{SO}}_{\mathrm{4}}^{+}$ intensities, suggesting a higher fraction of organosulfate compounds, as is also evidenced in Fig. S16D and E by the strong correlation between the derived organosulfur fraction and sulfate ions in the OOA_{AMS, org + inorg} factor (R²=0.85, slope=0.57).

Further information on chemical composition was gathered by comparing the AMS results to the IC measurements from the PM_0.7 filters sampled as part of another ALPACA study (Dingilian et al., 2024). Despite the good correlations between the two datasets, as shown in Fig. 7a, the AMS underestimated sulfate, ammonium, and nitrate by 30 %, 26 %, and 35 %, respectively (see Sect. 2.2.2). Both the total estimated organosulfur fraction and the OOA_{AMS, org + inorg} factor presented robust correlations (R²>0.90) with the S_(IV) and HMS ions from the filter analysis and a somewhat weaker correlation ($R^{\mathrm{2}}>\mathrm{0.61}-\mathrm{0.68}$) with the ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ ion (Figs. 7 and S16F–I). The OOA_{AMS, org + inorg} factor was also strongly correlated with ammonium (AMS data R²=0.86, Table 1; filter IC R²=0.77, Fig. 7b), potentially promoting the formation of S_(IV) species (Campbell et al., 2024). This author also reported that S_(IV) species, including HMS, represented the major secondary organosulfur component of PM_2.5 in Fairbanks during wintertime, contributing 26 %–41 % of total sulfate (Campbell et al., 2022). Overall, the molecular-level composition of OOA from PTR_CHARON and the inorganic chemical information from AMS_{org + inorg}, as well as diurnal patterns with enhanced concentrations in the afternoon (Fig. 3), are indicative of chemical daytime processing, underscoring the need for further exploration of the atmospheric processing pathways involved.

Figure 7 Scatter plots showing the correlation (R²; p≤0.05) between inorganic species measured with the AMS and offline ion chromatography of chemical species in PM_0.7 collected on filters. Comparison of (a) total mass concentrations of sulfur- and nitrogen-containing species, (b) OOA_{AMS, org + inorg} factor with different species from IC analysis, and (c) sulf-OA factor with different species from IC analysis.

3.2.6 Additional insights from combined analysis of organic and inorganic AMS data

Two additional factors, sulf-OA (i.e. sulfur-rich OA) and AmNi (i.e. ammonium nitrate), were observed from the PMF of AMS_{org + inorg} (Fig. 3). Approximately 40 %–60 % of these factors' masses comprised sulfur and nitrogen species (Fig. S15).

Sulfur-rich organic aerosol. Sulf-OA is composed of sulfate (60 %), organics (30 %), ammonium (6 %), and nitrate (4 %). The chemical composition was further explored via the $f{\mathrm{HSO}}_{\mathrm{3}}/f{\mathrm{H}}_{\mathrm{2}}{\mathrm{SO}}_{\mathrm{4}}$ analysis detailed in Sect. 3.2.4, as shown in Fig. S16A, where the factor is positioned between H₂SO₄ and (NH₄)₂SO₄. The measured [NH₄]$/$[SO₄] ratio was 0.07, considerably lower than the theoretical mass ratio of 0.38 and 0.18 of (NH₄)₂SO₄ and NH₄HSO₄, respectively, indicating the acidic nature of sulf-OA (Chen et al., 2019). The factor was well correlated with SO₂ (R²=0.6) and moderately correlated with the ResH3 factor (R² of 0.33). The factor was also associated with ultrafine particles in the 50–80 nm range (Fig. S12D). Regardless of the low correlation, we speculate that ResH3 and sulf-OA originated from the same source, i.e. the residential combustion of heating oil, and their temporal disagreement may result from instrumental biases of the CHARON inlet in quantifying particles smaller than 100 nm (Fig. S12B and D). For instance, as shown in Fig. S12E and F, the organic ResH3 supersedes sulf-OA concentrations when larger particles are abundant, and it has lower concentrations of smaller particles.

This factor contained $\mathrm{0.6}\pm \mathrm{0.5}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$ ($\sim \mathrm{58}\pm \mathrm{26}\mathit{\%}$) of total sulfate measured with the AMS, and it dominated during the low-pollution periods, which were more frequent and lasted longer than the high-pollution events (Fig. 1). Other primary factors, HOA_{AMS, org + inorg}, COA_{AMS, org + inorg}, and BBOA_{AMS, org + inorg}, contained an additional 11±9 % of sulfate ($\mathrm{0.2}\pm \mathrm{0.2}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$), so collectively, primary factors made up 69±24 % ($\mathrm{0.7}\pm \mathrm{0.6}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$) of total sulfate. This value is in close agreement with a previous ALPACA study that reported ∼62 ±12 % of total ${\mathrm{SO}}_{\mathrm{4}}^{\mathrm{2}-}$ mass to be of primary origin (Moon et al., 2023).

AmNi factor. The second inorganic factor was composed of 35 % nitrate, 14 % ammonium, and 43 % organics, accounting for 71±23 % of the total nitrate measured by the AMS (R²=0.98). The average concentration of this factor and the nitrate species in it were 1.1±1.6 and $\mathrm{0.4}\pm \mathrm{0.5}\mathrm{\mu }\mathrm{g}{\mathrm{m}}^{-\mathrm{3}}$. The factor was more abundant when NO_x concentrations were high (above 130 ppbv; Fig. S17b). Its diurnal trend peaked at around 14:00 AKST (Fig. 3), roughly 3–4 h after the morning peak of HOA_AMS, and was associated with relatively small particles of 110 nm (Fig. S12D). A high contribution of aliphatic moieties characterized the organic fraction, and according to the difference in mass concentrations of HOA_{AMS, org} and HOA_{AMS, org + inorg} of 13 % (Fig. S17A), we speculate that some organic components of HOA_{AMS, org} were transferred to the AmNi factor (Fig. 6). All these elements suggest a probable contribution from the vehicular emissions to this factor. The presence of inorganic compounds provided additional variables to the PMF, thereby improving the resolution of factors into distinct AmNi and HOA_{AMS, org + inorg} factors.

4 Local environmental implications

During the period of the campaign, 12–48 h long ADEC advisories for wood-burning restrictions were implemented seven times. The variation in the relative contributions of ResH1–4 during these advisories is depicted in Figs. 8 and S18–S21. For all advisory events, ResH2 and ResH4, i.e. wood smoke, were the predominant contributors before and after the advisories were in place. A notable increase was observed in the ResH3 contribution, i.e. heating oil, during the second (stage 1), fifth (stage 1), sixth (stage 1), and seventh advisory events. At the same time, ResH2 (i.e. hardwood-related fuels) remained a prominent contributor to OA_CHARON during the third (stage 2), fourth (stage 1), and fifth (stage 1) advisories. Most households in Fairbanks use heating oil (∼72 % of residents), followed by wood (∼22 % of residents) (Dunleavy and Brune, 2019), which was not reflected in the relative contributions apportioned to ResH3. This can be linked to a higher PM₁ release from wood combustion per given volume of fuel compared to heating oil and/or an underestimation of ResH3 by PTR_CHARON being associated with a size smaller than 100 nm (Fig. S12).

Figure 8 Variation in the relative contributions of residential heating factors to total biomass-burning OA concentrations. For simplicity, only the second ADEC advisory implemented during the campaign is shown. Contributions are also shown for approximately 2 d before and after the advisory for comparison, along with their 6 h averages as box plots (white panels), when suitable data were available (e.g. periods with noisy data were omitted and the adjacent period is shown instead). For better visualization of variation in contributions, when the advisory was in place, 3 h averages are shown (grey panels). To account for a lag in the appearance of variations in emission sources, 1 h averages are shown for the beginning and end of the advisory event.

As expected, the absolute average concentrations of all factors were inversely related to ambient temperature; however, the per cent change differed considerably across factors. Specifically, as temperatures decreased from −10 °C to below −25 °C, the average absolute concentrations for transport_CHARON, COA_CHARON, OOA_CHARON, and ResH1–4 increased 0.25 times, 0.75 times, 9.0 times, 1.4 times, 25.1 times, 3.0 times, and 2.9 times, respectively (Fig. S22). The steep increase in the relative contribution of ResH2, associated with hardwood tracers, contrasts with previous reports based on surveys (Dunleavy and Brune, 2019) and organic tracers (Haque et al., 2021), indicating birch and spruce as the most popular firewood in Fairbanks. Laboratory studies have shown that the burning of softwood pellets of Douglas fir or eastern white pine emits less PM than hardwood pellets of the same volume, and this response varies based on the moisture content of the wood and the heating appliance used (Morin et al., 2022). Also, ResH2 comprises a broader spectrum of volatile and semi-volatile substituted phenolic species. Thus, it is likely to undergo efficient gas-to-particle partitioning at low temperatures, leading to increased OA loads (Ijaz et al., 2025).

5 Conclusion

A CHARON inlet coupled with PTR-ToF MS and HR-ToF AMS were deployed during the Alaskan Layered Pollution And Chemical Analysis (ALPACA) campaign. The PMF analysis of the AMS data revealed three primary factors: biomass burning, hydrocarbon-like, and cooking factors, accounting for 28 %, 38 %, and 11 % of the total OA, respectively. A combined organic and inorganic PMF analysis provided additional insights. It revealed the presence of organosulfur compounds, mostly associated with the OOA factor and with another sulfate-rich factor of an acidic nature. A nitrate factor associated with hydrocarbon-like OA and high NO_x levels was interpreted as aged road transport emissions. The PTR_CHARON PMF analysis could differentiate between four residential heating sources – one oil combustion and three wood combustion types associated with a low temperature and softwood or hardwood combustion. Such factorization was achieved with the support of specific tracers that CHARON could successfully identify as furans, aromatic alcohols (resorcinol, guaiacol, eugenol, syringol), aldehydes (furfural, conifer aldehyde), acids (benzoic, dehydroabietic, abietic, linoleic, oleic, and stearic), and various PAHs. Collectively, all residential heating factors accounted for 79 % of the total OA_CHARON. Cooking and road transport were also recognized as primary sources by PTR_CHARON. All PMF analyses could apportion a single secondary organic fraction accounting for 11 %–19 % of the total OA. This work demonstrates the complementarity of the two instruments and their ability to describe the complex chemical composition of PM₁ and its related sources. The enhanced deconvolution of closely co-varying sources of ambient pollution epitomizes the novelty of our study. It demonstrates the capability of PTR_CHARON to deliver detailed qualitative and quantitative insights, thus enabling a comprehensive understanding of organic aerosol sources. These advances can support environmental regulators and citizens' efforts in improving air quality in Fairbanks and in fast-urbanizing regional sub-Arctic areas.