In situ measurements of NO, NO2, NOy, O3, and SO2 were conducted using commercial instruments (Thermo Scientific Inc.) modified to measure at 1 s time resolution. Ammonia (NH3) measurements were made at 1 s time resolution using a Los Gatos Research (LGR) NH3/H2S Analyzer, model 911-0039. Gas phase elemental Hg (GEM) was measured with a Tekran 237X instrument (Tekran Instruments Corporation) modified to allow a reduced sampling time of 2 min (McLagan et al., 2021; Cole et al., 2014). CO, CO2, and CH4 were measured with a cavity ring down spectroscopy instrument (Picarro G2401 m). A second Picarro G2401 m instrument was used to measure total carbon (TC, in units of ppm C) by passing the sample air through a catalyst to convert all carbon species to CO2. Total non-methane organic gases (NMOGT), in mixing ratios units of ppm C, was quantified by subtracting the ambient CH4, CO, and CO2 measurements (instrument without the catalyst) from the TC measurements (see Sect. S1.0 in the Supplement for more details).

Individually speciated NMOGs (and some inorganic species) were measured with a chemical ionization mass spectrometer (CIMS), with a proton transfer time-of-flight mass spectrometer (PTRMS), and through an advanced whole air sampler (AWAS) using canisters. In addition, integrated cartridge-based samples were taken. The CIMS (a modified Tofwerk/Aerodyne Api-ToF) was operated using iodide as the reagent ion providing 1 s time resolved measurements for 30 compounds (Table S2). The PTRMS (Ionicon Analytik GmbH, Austria) used chemical ionization with H3O+ as the primary reagent ion providing 1 s measurements for a suite of organic compounds. For those compounds with no available gas standard, a relative response factor was calculated with reaction rate constants using the method described in Sekimoto et al. (2017) and guided by the work of Koss et al. (2018) (“calculated” compounds). Integrated “grab” samples (20–30 s) were collected from the aircraft using the AWAS with offline analysis. The AWAS provided speciated measurements of hydrocarbons (≤C10), but no oxygenates. Overlapping compounds/isomers that were measured by both the PTRMS and AWAS, and between the PTRMS and CIMS, were handled as described in Sect. S1.1.4 in the Supplement. Integrated gas-phase samples were collected using an automated adsorbent tube (i.e. cartridge) sampling assembly with offline analysis (Ditto et al., 2021; Sheu et al., 2018; Khare et al., 2019). These samples provided targeted measurements of gas-phase compounds ranging in volatility from C10 volatile organic compounds (VOCs) to C25 semivolatile organic compounds (SVOCs) including hydrocarbons (CH), and functionalized compounds containing 1 oxygen atom (CHO1) and 1 sulfur atom (CHS1).

Particle chemistry was obtained with a high-resolution aerosol mass spectrometer (AMS) (Aerodyne) providing mass concentrations of particle species including total organics (OA), NO3, SO4, and NH4 for particles less than ∼1 µm. Particle size distributions were measured between 60 and 1000 nm at 1 s time resolution using the ultra high sensitivity aerosol spectrometer (UHSAS; Droplet Measurement Technologies). Refractory black carbon (rBC) was measured using a single-particle soot photometer (SP2; Droplet Measurement Technologies).

In-plume mixing ratios and the relative contribution of individually measured NMOG species to the sum of those species (ΣNMOG) are shown for 13 chemical classes in Fig. 5. (See Fig. S7 in the Supplement for separate SP and NP chemical classes). The largest chemical classes include carbonyls (acids, aldehydes, and ketones), alcohols, hydrocarbons (alkanes, alkenes, alkynes), aromatics (including furans, phenol, benzene, and toluene), and nitriles. Hydrocarbons (i.e. CxHy, including some aromatics) were responsible for just over half of the ΣNMOG (53 %) (Fig. S8 in the Supplement), with 29 % identified as alkenes such as ethene, propadiene, and propene, 19 % alkanes, predominantly ethane, and 3 % alkynes, almost entirely acetylene. Non-aromatic oxygenates accounted for an additional 36 % of the ΣNMOG with roughly equal contributions (9 % to 12 %) from acids, aldehydes and alcohols, and a smaller fraction from ketones (5 %). Including other oxygenated compounds such as furanoids and phenol/phenol derivatives, all oxygenates (CxHyOz) comprised 42 % (Fig. S8) of the ΣNMOG.

A similar range of compound classes has been observed in previous field and laboratory studies, noting that the measured compound suite between studies varies to some extent. For example, Simpson et al. (2011) found a similar distribution of compound classes with 57 measured NMOG species, based on discrete canister samples, in boreal forest wildfires. In that study, oxygenates (non-aromatic) comprised a smaller portion of NMOG (29 %) as major emitted species like acetaldehyde and acetic acid (Fig. 8) were not included. Other studies have also found oxygenates to be a large portion of NMOG emissions across multiple fuel types, including those similar to the current study, ranging from 51 %–68 % (Permar et al., 2021; Koss et al., 2018; Gilman et al., 2015; Akagi et al., 2011) with a range of 25 %–55 % reported in Hatch et al. (2017). The fraction of NMOG oxygenates in the present study (42 %) was closer to those reported in Hatch et al. (2017) when only the most relevant fuel types of pine and spruce were considered (55 % and 43 %, respectively). Similar to previous work (Koss et al., 2018; Stockwell et al., 2015; Hatch et al., 2015), emissions of substituted oxygenates like furanoids (furans + derivatives) and phenolic compounds were observed. Furanoids contributed 4 % of the ΣNMOG mostly due to furfural, furan, and methyl furan, while phenolic compounds e.g. guaiacol, methyl guaiacol, contributed 0.5 % of the ΣNMOG (Fig. S9 in the Supplement). Although these emissions were less abundant in the present study, they represent important OH reactants (Coggon et al., 2019; Koss et al., 2018; Gilman et al., 2015) with phenols being implicated as precursors to brown carbon formation in secondary organic aerosol (SOA) (Palm et al., 2020).

Biogenic emissions of terpenoids including isoprene, monoterpenes, carvone, sesquiterpenes, camphor/isomers, and terpine-4-ol/cineole/isomers were elevated in the plumes collectively reaching ∼2.4 ppbv, and contributing ∼2 % to the ΣNMOGs (Fig. S9). Isoprene was ∼66 % of these compounds with an additional 32 % from monoterpenes. Emissions of isoprene from biomass burning has been observed from a wide range of fuel types (Hatch et al., 2019). As isoprene is not stored by plants and the measurements were taken ∼14 h post flaming, it was likely emitted as a combustion product.

In this study, furfural was the most abundant oxygenated aromatic compound and a factor of 5 times higher than that of phenol. Although Koss et al. (2018) found that phenol and furfural emissions were similar for most fuels tested in the laboratory, furfural emissions derived from multiple wildfires sampled in Permar et al. (2021) were similar to those in the present study, and a factor of 1.6 higher for phenol. As phenol emissions are associated with lignin pyrolysis (Stockwell et al., 2015; Simoneit et al., 1999), the lower emissions in the current study could be because the lignin content in the fuel mixture was lower than fuels used in previous laboratory studies or that most of the phenolic compounds were emitted during the earlier phases of the fire. Several modelling studies have indicated that aromatics and terpenes are insufficient to explain SOA formation in biomass burning plumes (e.g. Hodshire et al., 2019) suggesting the importance of inclusion of other aromatic species such as phenolics and furanoid compounds. However, models typically do not include reactions involving phenolic and furanoids species, especially substituted compounds like furfural, guaiacol, and methyl guaiacol. Box model simulations have also shown that incorporation of OH oxidation of furan, 2-methyfuran, 2,5-dimethylfuran, furfural, 5-methylfurfural, and guaiacol leads to 10 % more O3 formed (Coggon et al., 2019).

Offline analysis of cartridge samples showed a wider range of hydrocarbons and functionalized gas-phase organic compounds not observed in the PTRMS, CIMS, and AWAS measurements, including I/SVOC compounds in the wildfire plume. ERs (Table S7 in the Supplement) for species containing carbon and hydrogen, and with either sulfur or oxygen (i.e. CH (hydrocarbon), CHS1, and CHO1 type molecules) accounted for a sizeable fraction of carbon in the C10 to C25 range. Additional contributions are expected from more highly functionalized organics in the gas (and particle) phase not reflected in the CH, CHO1, and CHS1 compound classes (e.g. gas-phase species with multiple oxygen atoms like vanillic acid or acetovanillone, and gas-phase species containing combinations of oxygen and nitrogen atoms (CHONs), Ditto et al., 2021, 2022). ERs in the plume varied across the carbon number range; in general, the highest ratios were observed for the complex mixture of hydrocarbons (i.e. CH compounds) broadly peaking at C20–C23 in the SVOC range, with a larger contribution from C10 compounds including monoterpenes. By comparison, the complex mixture of CHO1 compounds was slightly lower in abundance than CH with contributions from C10 monoterpenoid emissions or monoterpene oxidation products. CHS1 IVOC–SVOCs were the lowest abundance species quantified. CHN1 compounds represent another observed contributor of IVOCs–SVOCs; the sum of all CHN1 compound ion abundances was two orders of magnitude smaller than the sum of all CHO1 species. We note that for CHN1, this qualitative comparison is in terms of ion abundances only, given a lack of appropriate standards to calibrate for the mass spectrometer's response to the complex mixture of reduced nitrogen-containing I/SVOCs.

EFs were estimated to be 1.4 ± 0.037–2.4 ± 0.063 gkg-1 for CH, 0.81 ± 0.078–0.81 ± 0.079 gkg-1 for CHO1, and 0.21 ± 0.0033–0.22 ± 0.0060 gkg-1 for CHS1 species, for a total EF of 2.4 ± 0.12–3.5 ± 0.15 gkg-1 (Table A1). Here, the uncertainty represents measurement uncertainty associated with the conversion from signal to mass, and the reported ranges show lower and upper limit EF values that account for a contaminated background and that assume no background concentrations, respectively (as described above). These estimates accounted for C11–C25 species and focused on I/SVOCs to avoid double counting the monoterpenes and C10 monoterpenoid species, as they were already accounted for in the PTRMS data. It is noted that the concentrations estimated for the cartridge samples may be sensitive to variations in sampling efficiency within the under-wing sampling pod across C10–C25 though these effects are expected to be minimal for the adsorbent tubes used in this study (Ditto et al., 2021; Sheu et al., 2018). These emission estimates expanded the characterized spectrum of organic species to include IVOC/SVOCs in boreal forest fire emissions, which until now, had only been available from laboratory measurements (Hatch et al., 2018). However, the observed emissions of the complex mixture of hydrocarbons and functionalized species may include contributions from the re-volatilization of compounds previously emitted from upwind oil sands operations and deposited in the forest ecosystem, as noted in Ditto et al. (2021).

Measurements of TC, along with the speciated measurements from the PTRMS, CIMS, AWAS, and cartridges, provided a unique opportunity to reconcile the TC budget in a wildfire. Figure 6 shows the TC partitioning based on derived EFs (Sect. 3.4); overlapping compounds from the individual measurement methods were handled as described in Sect. S1.1.4. The total EF for all carbon-containing compounds was 1652 gkg-1 and, as expected, CO2 was the dominant contributor comprising >90 % of TC. CO contributed 7.0 % followed by a contribution from NMOGT of 1.9 %, with even smaller contributions observed from CH4 (0.5 %) followed by OC and BC (not shown) at <0.5 %. The two magnified pie charts (right side), representing the low and high I/SVOC EF estimates, show the percent breakdown of the measured NMOGs, and the remaining unidentified portion of NMOGT. The EF values (gCkg-1) are identified in the box below. The ΣNMOG EFs (for PTRMS + CIMS + AWAS measurements), totalling 13.6 ± 0.9 gCkg-1, accounted for 43 % ± 3 % of the NMOGT EF of 31.2 ± 4.7 gCkg-1 (refer to Fig. S10 in the Supplement for the individual SP and NP breakdowns). The ΣNMOG uncertainties were estimated by summing in quadrature the individual compound EF uncertainties for the SP and NP separately, with these uncertainties subsequently summed in quadrature to derive the average ΣNMOG uncertainty (Fig. 6). The cartridge data showed the presence of a range of larger molecular weight I/SVOC compounds between C10 and C25 with an additional 2.1 to 3.0 gCkg-1 representing 7 % to 10 % of NMOGT. Together, all of the speciated NMOG measurements in this study accounted for 50 % ± 3 % to 53 % ± 3 % of NMOGT. The remaining carbon mass was unidentified comprising 47 % ± 15 % to 50 % ± 15 % of NMOGT. Despite using four state-of-the-art measurement techniques resulting in an extensive measurement suite, almost half of NMOGT remained unidentified. This is consistent with previous work estimating ∼50 % of NMOGT by mass as unidentified (Akagi et al., 2011). It is noted, however, that the magnitude of the unidentified portion is partly affected by uncertainties in the speciated measurements. For example, many of the “calculated” PTRMS compounds are uncertain by an estimated factor of ∼2 (Sect. S1.1.1 and Table S1 in the Supplement). Nevertheless, a portion of the unidentified species likely consisted of challenging-to-measure-VOCs and larger I/SVOCs that were highly functionalized or contained molecular features like reduced nitrogen groups (e.g. amines) that have been observed in the gas and particle phase at various sites (Ditto et al., 2020, 2022). The presence of I/SVOCs in biomass burning emissions has been previously observed in laboratory experiments (e.g. Koss et al., 2018; Hatch et al., 2018, 2017; Bruns et al., 2016) with smoldering more likely to emit a higher fraction of compounds with low volatility than higher temperature processes (Koss et al., 2018). The unidentified portion may also have been comprised of nitrogen-containing organics (Sect. 3.1). Studies that included measurements of a larger range of nitrogen-containing organics in biomass burning emissions estimated that they comprised <5 %–6 % of the total nitrogen budget (Lindaas et al., 2020; Gilman et al., 2015), and thus an even smaller fraction of NMOGT. Advancing analytical techniques to expand the suite of NMOG speciation will enable further reconciliation of the TC budget which is important for assessing secondary formation processes in the atmosphere.

Volatility distributions can help track the full range of organic species to assess their partitioning between the condensed and gas phases (Donahue et al., 2011). Figure 7 shows the fractional sum of all NMOG EFs within each volatility bin in terms of saturation concentration ranges (log⁡10Co, µgm-3) for the low I/SVOC EF estimate. Co values were estimated using the parameterization developed by Li et al. (2016). NMOG emissions from this fire spanned a large range of volatilities from log⁡10Co of -2 to 10 µgm-3 across SVOC to VOC categories. The bin-averaged O/C ratio based on the measurements increased with reduced volatility reflecting the presence of compounds with additional oxygen-containing functional groups. The highest fraction of emissions was present as VOCs with 81 % having log⁡10Co>6 µgm-3, 9 % as IVOCs having 4µgm-3<log⁡10Co<6µgm-3, and 10 % as SVOCs having log⁡10Co<3 µgm-3. These results align with laboratory studies showing that oxygenates comprised more than >75 % of IVOCs across a range of biomass types with IVOCs accounting for ∼11 % of the ΣNMOG (Hatch et al., 2018). Figure 7 encompasses the range of volatilities based on all the identified NMOGs in this study that is expected to represent initial emission conditions for modelling downwind chemistry. However, improved speciation, particularly of lower volatility compounds, is needed to further expand the range of volatilities and advance knowledge in gas-to-particle partitioning processes.

The PM1 EF (6.8 ± 0.8 gkg-1) represents the total of all particle component species as measured by the AMS. The PM1 EF of 6.8 ± 1.1 gkg-1 (Fig. 8a) (accounting for estimated mass differences due to particle diameters (Sect. S1.1.2 in the Supplement)) falls in the lower end of the large range previously observed for boreal forest wildfires (18.7 ± 15.9 gkg-1; Fig. 9a). The few PM EFs for BFF19 (n=5) over a limited range of MCEs (i.e. 0.89 to 0.93) shows significant variability consistent with previous work (Jolleys et al., 2015; Akagi et al., 2011; Cubison et al., 2011; Hosseini et al., 2013). OA, accounting for 90 % of PM1, has the largest EF, with comparatively lower EFs for NO3, BC, NH4, and SO4 (Figs. 8a and S6). This reflects the dominant particle-phase organic carbon content of the burned fuel and correspondingly lower fractions of nitrogen and sulfur-containing compounds. Similar high organic fractions have been previously observed in biomass burning emissions (Liu et al., 2017; May et al., 2014; Hecobian et al., 2011). ERs similarly highlight the dominant OA emissions. Although the magnitude of EFs between the SP and NP are within their derived uncertainties (Fig. S11), the ERs showed differences by up to 70 % for NH4 (Fig. S12) suggesting some differences in photochemistry between the two plumes. EFs and ERs for chemically-speciated particle species derived in this study represent the first such measurements under boreal forest wildfire conditions. EFs for chemically-speciated compounds are not found in BFF19 (except BC), but when compared with available values for US temperate forest wildfires (Liu et al., 2017), are found to be lower for OA (Fig. 9a), SO4, NO3, and NH4 by factors of 3.7, 5.0, 5.3, and 3.0, respectively. Although differences in fuel type burned between the present study (mature Jack pine, boreal spruce, boreal mixed-wood) and Liu et al. (2017) (mixed conifer, grass, brush, and chaparral) may influence the chemical composition of emissions, these large differences suggest the importance of other factors in controlling OA emissions. The lower OA emissions under smoldering conditions in the current study compared to Liu et al. (2017) with higher combustion efficiencies (0.877 to 0.935) conflicts with some findings showing increased OA emissions with lower fire intensities (Liu et al., 2017; Burling et al., 2011). However, the relationship between EFOA and combustion efficiency can be impacted by multiple factors such as OA loading, gas–particle partitioning related to dilution, and fuel moisture content (May et al., 2014). The EFOA in the current study (6.6 ± 2.6 gkg-1) lies in the range of EFOA reported for prescribed burns across three temperate ecosystems (2.8 ± 1.6 to 11.2 ± 2.7 gkg-1) (May et al., 2014). This may imply that the low intensity, surface, and smoldering wildfire conditions in the present study (Sect. 3.1) may be similar to prescribed burn conditions which are typically low intensity fires that are restricted to the forest floor and understory, and conducted under controlled and consistent meteorological and fuel moisture conditions (Yokelson et al., 2013; Carter and Foster, 2004). Inorganic PM emissions, however, are likely more dependent on fuel elemental composition than combustion efficiency (Liu et al., 2017). Differences in fuel composition between boreal and temperate forest ecosystems are inferred through comparisons of NOx and SO2 emissions. For example, the average NOx and SO2 EFs for boreal forests are lower than the average EFs for temperate forests by factors of 2.5 and 3.0, respectively. The lower NOx and SO2 emissions from boreal vs. temperate forest wildfires are likely reflective of the reduced S and N content in boreal biomass (Bond-Lamberty et al., 2006) relative to conifer (Miesel, 2012) fuels in the western US, and the possible influence of lower anthropogenic sources of nitrogen and sulfur atmospheric deposition in boreal forests (Jia et al., 2016).

The largest average EFs for inorganic gases (Fig. 8b; separate NP and SP Fig. S11) were from reduced nitrogen compounds dominated by NH3 (0.63 ± 0.14 gkg-1) and followed by HCN (0.31 ± 0.07 gkg-1), with lower EFs for oxidized nitrogen compounds such as NO2 (0.15 ± 0.04 gkg-1) and HONO (0.02 ± 0.012 gkg-1). This is consistent with previous work identifying elevated emissions of NH3 and HCN during smoldering conditions, whereas emissions of HONO and NOx are primarily associated with flaming combustion (e.g. Roberts et al., 2020; Akagi et al., 2013; Yokelson et al., 1997; Griffith et al., 1991). The EFs for CO2 and CO from the present study are comparable within uncertainties of that previously reported for BFF19 (Table A1). However, EFs for most other gaseous inorganic species were lower than the BFF19 EF average including NH3, HONO, and NOx by factors of 4.0, 20, and 7.1, respectively (Fig. 9a). There are only a limited number of studies reporting EFs for these compounds in the BFF19 category. For example, there are only four previously reported BFF19 EFs for NH3 (2.5 ± 1.8 gkg-1) showing a large range of values. Although these comparisons are limited by the few reported values in the literature, the differences indicate a strong sensitivity towards factors like fire intensity, chemical reactivity, fuel type and moisture, and meteorology. In contrast, EFs for HCN derived in the current study (0.31 ± 0.07 gkg-1) lie within the range of BFF19, LAB18, and TFF21 values (0.28 ± 0.06 to 0.53 ± 0.30 gkg-1), (Fig. 9a–c, respectively), and does not vary widely suggesting that HCN may be less sensitive to burning characteristics. HCN is of concern due to its impacts on human health particularly since biomass burning emissions are responsible for the majority of the global HCN (Moussa et al., 2016, and references therein).

In Fig. 8c, the top 25 average EFs for gas-phase organic species are shown in decreasing order of magnitude. The most abundant emissions were from the lower molecular weight compounds; such trends are generally in agreement with previous field-based measurements for a range of fuel types (e.g. Permar et al., 2021; Andreae, 2019; Liu et al., 2017; Simpson et al., 2011; Urbanski et al., 2009). Excluding CH4, the largest EFs were associated with methanol, followed by ethene, ethane, acetic acid, acetaldehyde, formaldehyde, and acetone ranging from 1.9 ± 0.4 to 0.82 ± 0.22 gkg-1 for these compounds. Noting some variations related to differences in measurement methods, other studies have identified many of these same species as dominating biomass burning emissions (e.g. Permar et al., 2021; Simpson et al., 2011; Akagi et al., 2011). For example, Simpson et al. (2011) found that five of the same compounds in the present study including formaldehyde, methanol, ethene, ethane, and acetone were in the top 10 NMOG EFs from aircraft-based measurements made of boreal forest wildfires in northern Saskatchewan, Canada, and within ∼300 km of the current study. In the present study, the top 24 NMOG compounds accounted for 81 % of the ΣNMOG by total molecular mass with lower emissions from the remaining 149 measured compounds. In western US wildfires, small emissions from 151 species were found to account for almost half of ΣNMOG (Permar et al., 2021). EFs for the NP and SP generally agreed within their uncertainties with larger differences for some of the more reactive species like isoprene, monoterpenes, and furan. For example, the SP EF for isoprene was a factor of 3.4 lower than that for the NP (0.64 ± 0.34 gkg-1) (Fig. S13). Although the reasons for these differences are not yet known, observations of higher O3 in the SP (52.4 ± 3.0 ppbv) compared to the NP (44.7 ± 3.6 ppbv) suggest the influence of higher oxidant chemistry in the SP emissions compared to the NP.

To compare the total NMOG derived in the present study with those from previous studies that typically sum up their speciated measurements, i.e. ΣNMOG, estimates were made using two methods: (1) increasing the ΣNMOG to account for the unidentified portion of NMOGT; and (2) adjusting the NMOGT to reflect the total molecular mass (not just the carbon portion). For method 1, the ΣNMOG EF (including the I/SVOCs) in this study (24.5 ± 1.6 to 25.6 ± 1.6 gkg-1) was increased by 50 % and 47 % (Fig. 6), respectively, equalling 36.8 ± 11.3 to 37.6 ± 12.2 gkg-1. This estimate assumes that the carbon distribution is the same as the identified, speciated measurements. For method 2, based on the speciated measurements, the average molecular mass was 100 gmol-1 and the average carbon number was 6 resulting in ∼28 % ± 24 % of the molecular fraction represented by atoms other than carbon. Adjusting the NMOGT of 31.2 ± 3.8 gCkg-1 upwards by 28 % ± 24 % to reflect the additional molecular mass results in a NMOG of 39.9 ± 5.8 gkg-1. The resulting estimated NMOG in this study of 36.8 ± 11.3 to 39.9 ± 5.8 gkg-1 lies between the estimated average of 58.7 gkg-1 for the BFF19 (Fig. 9a) and those estimated from the ΣNMOG EFs of 25.0 gkg-1 (LAB18) (Fig. 9b), and 26.1 ± 6.9 gkg-1 (TFF21) (Fig. 9c) derived from laboratory- and field-based studies (Table A1). In contrast to the current work, previous estimates of NMOGT are likely to underestimate total NMOG emissions as they typically represent the sum of measured species only. Some studies have attempted to account for NMOGT by including the sum of measured plus estimates of “unknown” portions of NMOGs (ΣNMOGs) (Permar et al., 2021; Koss et al., 2018; Stockwell et al., 2015; Gilman et al., 2015). The BFF19 EF was recently doubled from 29.3 ± 10.1 to 58.7 gkg-1 to account for unidentified NMOGs where the ΣNMOGs were measured by FTIR, GC, and PTRMS (Andreae, 2019; Akagi et al., 2011). These results support that doubling the ΣNMOG provides a reasonable estimate of the NMOGT. It is noted that the average BFF19 NMOG EF is ∼1.5 times higher than that derived in the present study, however this may reflect variability in NMOG emissions even within the same boreal biome.

Although it is known that acidic compounds are emitted from biomass burning, few studies have quantified their emissions, particularly under field conditions (Andreae, 2019; Veres et al., 2010; Yokelson et al., 2009; Goode et al., 2000). In this study, EFs for 22 organic acidic compounds were derived (Table A1) representing the most detailed set of organic acid EFs from biomass burning for any ecosystem (Andreae, 2019). The largest EFs for these compounds include acetic acid, C4 oxo-carboxylic acids, and pyruvic acid, all of which are found among the top 24 NMOGs (Fig. 8c). For those measurements that are available for comparison, EFs in the present study were lower for formic acid and acetic acid compared to BFF19, and were also lower than in LAB18 and TFF21, ranging from factors of 1.7 to 8.8 (Fig. 9a–c). Organic acids that were in common with TFF21 and LAB18 (Table A1) had lower EFs, with the exception of pyruvic acid, which was substantially higher (> factor of 37) in the present study. Differences in fuel type may be an important factor in the variability of these comparisons. Based on laboratory experiments, Veres et al. (2010) found a large range (factor of 5 to 13) of organic acid emissions with different fuel types suggesting that the lignin content of the fuel could be a source of biomass burning organic acid emissions. Emissions for 10 organic acids that have not previously been reported, and several inorganic acids including nitrous acid, isocyanic acid, and peroxynitric acid, are included in Table A1. These acids, representing 9 % of the ΣNMOGs (Fig. 5), are an important class of oxygenates as they can form additional PM (Reid et al., 2005) and influence the hygroscopicity of smoke particles (Rogers et al., 1991; Kotchenruther and Hobbs, 1998).

Isoprene and monoterpenes, with similar EFs ∼0.41 ± 0.19 gkg-1, represented 16th and 18th, respectively, of the top 24 NMOG EFs in this study. Terpenes are known to be emitted from a range of biomass burning fuels (Andreae, 2019, and references therein), but there have been few measurements in boreal forest wildfire plumes (Simpson et al., 2011; Andreae, 2019). It is noted that PTRMS measurements of IVOCs like sesquiterpenes likely represent lower limits as they tend to be easily lost to sample inlet lines due to their low volatility. The isoprene average EF of 0.42 ± 0.26 gkg-1 was more than a factor of 5 higher, while the monoterpenes EF, 0.41 ± 0.19 gkg-1, was substantially lower than the only reported EF for boreal forest wildfires (Simpson et al., 2011). The difference in EFs for isoprene would be even greater if only the NP EF (0.64 ± 0.34 gkg-1) is compared (if it is assumed that isoprene emissions were influenced by photochemical losses in the SP). As the present study and the Simpson et al. (2011) study were conducted in similar locations (i.e. boreal forest region within ∼300 km of each other), and comparable background levels, these differences are likely driven by fire combustion state, despite having similar study-averaged MCEs. The majority of monoterpenes are stored in plant tissues (resin stores) for long periods of time, but isoprene is synthesized and immediately released by plants, and can also be emitted as a combustion product (Ciccioli et al., 2014; Akagi et al., 2013). Hatch et al. (2019) found that a wide range of terpenoids are released across a variety of biomass types with variable emissions that were dependent on plant species, and specifically related to their fuel resin stores. In the present study, monoterpenes may have “boiled-off” through distillation processes in the early stages of the fire resulting in lower monoterpenes emissions at the aircraft sampling time, ∼14 h post-flaming. In contrast, the Simpson et al. (2011) study sampled comparatively earlier and more intense fire stages where higher monoterpene emissions were likely released from live or recently fallen trees that still contained significant resin stores. The monoterpenes EF reported by Simpson et al. (2011) was likely even higher given only two monoterpenes were speciated and emissions of other terpenes were likely (Hatch et al., 2019). Higher isoprene emissions in the present study compared to Simpson et al. (2011) could be related to the comparatively larger smoldering component. Although limited data exist on the release of isoprene as a function of fire intensity, negative relationships between isoprene and MCE were observed in Australian temperate forest fires (Guérette et al., 2018) and wheat fields (Kumar et al., 2018).

Several furanoid compounds also exhibited significant emissions (Fig. 8c) including furfural, furan, and methyl furan ranking 12th, 19th, and 22nd of the top 24 organic compounds, respectively. Emissions of furanoids have been observed for a wide range of fuel types (Hatch et al., 2017; Simpson et al., 2011). Agreement within uncertainties was found with BFF19 for furfural and furan (Fig. 9a). The EFs for furan (0.39 ± 0.19 gkg-1) and furfural (0.65 ± 0.31 gkg-1) were also similar to that in LAB18 (Fig. 9b) and TFF21 (Fig. 9c), and other ecosystems (Andreae, 2019), suggesting their emissions were relatively insensitive to fire intensity and fuel mixture. The comparisons in Fig. 9 indicate that for the higher emitting species, the current results are fairly similar, but for the lower emitting species, these results are lower than previous reported values. These comparisons provide context for the emissions reported in the present study and moves towards improved statistics to better constrain wildfire emissions. Additional factors are considered to explain variability in emissions between this study and other reported values, and within this study (NP vs. SP). Differences and variability in burn conditions (e.g. fire intensity, winds, fuel density, flame dynamics, fuel moisture) likely influence these comparisons; the screen 1 measurements in the present study were taken from 09:00–10:00 LT when the fire was in a low intensity, smoldering state, while those in Permar et al. (2021) and Simpson et al. (2011) took place during mid-day under active wildfire conditions. Aircraft measurements in general have a higher probability of sampling variable burn conditions compared to laboratory studies (Hodshire et al., 2019), and as such, aircraft-derived EFs are likely to reflect variability for reactive species as speculated earlier with isoprene. Particularly for reactive species that can exhibit complex variation across plumes, EFs (and ERs) derived by integrating across plumes can be biased low, (Sect. 2.5; Peng et al., 2020; Decker et al., 2021). Also, EFs derived using TC in this study may result in lower, albeit small, EFs compared to reported values that do not account for all the carbon (estimated to be 1 %–2 %, Akagi et al., 2011).

EFs derived in the present study are compared with those that are currently incorporated into the emissions component of the FireWork modelling system using the Forest Fire Emissions Prediction System (CFFEPS). CFFEPS uses EFs allocated for three combustion states (flaming, smoldering, and residual) and for eight species including lumped non-methane hydrocarbons (NMHC) based on United States vegetation data compiled in Urbanski (2014) (Table 3 in Chen et al., 2019). Figure 9d (bolded compounds) shows that the smoldering EFs in the present study were comparable for CO, CH4, and NMOG, but lower for PM1 (PM2.5), NH3, SO2, and NOx by factors of 3.4, 2.4, 6.6, and 17, respectively. The lower EFs for these species implies that the CFFEPS EFs would not adequately capture their total emissions under smoldering conditions for the boreal fuel in the current measurement study.

For incorporation into numerical air quality models, total organic gas (TOG=NMOG+CH4) emissions are typically split into detailed chemical components using chemical mass speciation profiles, and converted to lumped chemical mechanism species. In the FireWork modelling system, the smoldering combustion TOG is split into components based on EPA's SPECIATEv4.5 profile (#95428) (US EPA, 2022; Urbanski, 2014 – Supplement Table A.2, boreal forest duff/organic soil). This profile is ultimately compiled using laboratory data from Yokelson et al. (2013), Bertschi et al. (2003), and Yokelson et al. (1997) based entirely on US fuel types. EFs in the present study were found to be generally lower than the laboratory-based EFs for 74 species in common ranging from factors of 1.7 to 8.5 including for monoterpenes, formic acid, phenol, and furan (Fig. 9d). A few species including furfural, propane nitrile, and ethyl styrene are comparable, while isoprene and pyruvic acid are notably higher by factors 2 to 5.3.

For a research version of the FireWork system, the component speciation is mapped to the SAPRC-11 chemical mechanism species (Carter and Heo, 2013) with detailed oxygenated compounds and aromatic species, largely to better represent SOA formation processes. For comparison with the measurement-derived speciation profile in this study, EFs were first mapped to SAPRC-11 species and normalized by the total identified mass species fraction without unknowns to obtain mass fractions of relevant model mechanism species (Table S9 in the Supplement). Comparing the normalized mass fractions for similar mechanism species (Fig. S14 in the Supplement) showed a substantially lower fraction of reactive alkanes (ALK5) with an estimated 5 % in this study compared to 28 % in the SPECIATEv4.5 wildfire smoldering profile. Mass fractions in this study are notably higher for the ACYL, ETHE, and ISOP lumped model species by factors of 13, 7, and 51. The mass fraction of CH4 is also different with 24 % of TOG in this study compared to 4 % from the SPECIATE4.5 profile. The measurement-derived chemical speciation profile is expected to be different from the average speciation profile from EPA's SPECIATEv4.5 due to differences in chemical species identification, fuel type, fire and measurement conditions, and uncertainties on how measured compounds are mapped to lumped mechanism species. The emissions profile developed in the present study can be used to improve predictions of wildfire smoldering emissions specific to the Canadian boreal forest.

Wildfires generally exhibit a diurnal cycle with fire intensities maximizing late afternoon and diminishing at night having important implications for fire emissions (Chen et al., 2019). Evaluating modelled emissions throughout the diurnal cycle with observations is a critical step in verifying smoke predictions. Emissions models mostly parameterize diurnal fire emissions with prescribed profiles that distribute daily total emissions to hourly. In CFFEPS, a diurnal profile is applied to allocate daily burn area to hourly intervals, with highest activity in the late afternoon. The actual fuel consumed, and thus, hourly emissions, is then calculated with depth of burn estimates driven by hourly meteorology (Chen et al., 2019). In Fig. 10, for the wildfire in the present study, the hourly CFFEPS-predicted emissions (orange dots) for selected compounds are shown between 24 June 2018 17:00 UTC and 25 June 2018 21:00 UTC, spanning the aircraft sample time (red arrow at 15:00 UTC). After 21:00 UTC, the discrepancy between the CFFEPS-predicted emissions and FRP increased as a result of rain that passed through the area that is not considered in the model bottom-up emission estimates (not shown in the figure). The burning phases are outlined in the figure where flaming (light pink background) is assumed to occur when the atmospheric conditions alongside fire behaviour and emissions model outputs infer a fireline intensity >4000 kWm-1 (Stocks et al., 1989), and a smoldering fire (blue background) for intensity <4000 kWm-1. The fire intensity distinction between flaming and smoldering roughly aligns with the observed minimum for this particular fire with the fire radiative power (FRP, grey dots) retrieval from the GOES-16 satellite sensor of 500 MW where smoldering occurs <500 MW and flaming for >500 MW. The 500 MW threshold over the approximately 1700 ha of actively smoldering area observed by overnight VIIRS thermal detections gives an estimated energy density of 0.29 MWha-1. This energy density threshold for smoldering <0.29 MWha-1 found in this study is in agreement with O'Brien et al. (2015) who found flaming combustion at >0.4 MWha-1 for lower intensity flaming fires and smoldering combustion at lower energy densities. The FRP represents the sum over all hotspots of this fire for each 15 min observation period. Emission rates in metric tonnes per hour (th-1) were derived from selected aircraft measurements using a mass balance method that was designed to estimate pollutant transfer rates through virtual screens using aircraft flight data (Gordon et al., 2015) (see Sect. S1.2). Emission rates were estimated to be 29 ± 2.1 th-1 for PM1, 433 ± 26.7 th-1 for CO, 0.65 ± 0.03 th-1 for NOx (as NO), and 2.7 ± 0.16 th-1 for NH3 (red arrows). Emission rates were also derived from satellite observations (black arrows) for CO, NOx, and NH3. Emissions of CO were estimated using a flux method as described in Stockwell et al. (2022) using TROPOMI satellite observations yielding 1670 ± 670 th-1 at 19:06 UTC and 4050 ± 1620 th-1 at 20:48 UTC. NOx emissions (9.1 ± 3.4; scaled to tNOh-1 at 19:06 UTC) were derived from the TROPOMI NO2 dataset using an exponentially modified Gaussian approach (Griffin et al., 2021) (not enough high-quality observations for the 20:48 UTC overpass). NH3 emission rates (5.6 ± 3.9 th-1) were derived from CRIS satellite observations at the satellite overpass time of 19:00 UTC by applying a flux method (Adams et al., 2019).

The aircraft measurements were taken when the FRP was low reflecting a smoldering surface fire. However, the satellite overpass occurred ∼4 h later than the aircraft measurements close to the FRP daily maximum, after which rain passed through the area. The CFFEPS model, exhibiting a prescribed diurnal pattern, captures the increase in NOx and NH3 emissions between that derived from the aircraft and satellites transitioning from a smoldering to predominantly flaming fire; NOx emissions increased by a factor >10, whereas the NH3 emissions increased by a factor of approximately 2. This is in agreement with recent laboratory measurements that found that the release of NOx is favoured during the flaming stage and the release of reduced forms of nitrogen, such as NH3, is favoured during the smoldering phase (Roberts et al., 2020) (also see Fig. 4). However, the CFFEPS CO emission rates do not track the increase in CO emissions between the aircraft-derived value and the two TROPOMI values, indicating that the CO EF for flaming is low in the model.  This highlights the need to validate model emission rates with measurements to adjust and update the EFs accordingly.

The aircraft- and satellite-derived emission rates for CO, NOx, and NH3 were each ratioed to FRP (in units of th-1MW-1, referred to as Rspecies/FRP) to represent the two end burning states, ie. smoldering and flaming conditions. Rspecies/FRP values were estimated for the flaming and smoldering phases of the fire and it was assumed that flaming occurred for FRP>500 MW and smoldering for FRP<500 MW. The CO Rspecies/FRP values were roughly twice as large during smoldering compared to flaming. For the two satellite overpasses during the flaming phase of the fire, the RCO/FRP values were within the uncertainties (19:06 UTC RCO/FRP=0.47 ± 0.25 th-1MW-1; 20:48 UTC RCO/FRP=0.43 ± 0.23 th-1MW-1). The RNOx/FRP value for NOx is also twice as large for flaming compared to smoldering, and for NH3, the RNH3/FRP value is ∼5 times larger for smoldering than flaming. Total emissions were then estimated by integrating the GOES FRP over the period 24 June 2018 17:00 UTC to 25 June 2018 23:00 UTC (after which no more hot spots were detected by GOES and the fire presumably extinguished), and applying the derived smoldering and flaming Rspecies/FRP values. Assuming that the fire went out when GOES did not observe any hot spots, total emissions for this fire of CO, NOx, and NH3 are estimated at 22 000 ± 8700, 104 ± 42, and 84 ± 33 t, respectively. If the fire is assumed to have continued burning when GOES did not detect any fire hot spots (between 22:00–04:00 UTC and 07:00–15:00 UTC, with an FRP of 150 MW (∼GOES detection limit; Roberts et al., 2015), the emissions increase to 24000±9600, 106 ± 43, and 98 ± 39 t, respectively, providing an upper limit of emissions. The combination of aircraft- and satellite-derived emission estimates for multiple species helps to obtain the diurnal variability of emissions and to obtain more complete details on the emission information across different burning stages.