Emission factors of long-lived volatile organic compounds from the 2019-2020 Australian wildfires during the COALA campaign

In 2019/2020, Australia experienced its largest wildfire season on record. Smoke covered hundreds of square kilometers 10 across the southeastern coast and reached the site of the 2020 COALA (Characterizing Organics and Aerosol Loading over Australia) field campaign in New South Wales. Using a subset of nighttime observations made by a proton-transfer-reaction timeof-flight mass spectrometer (PTR-ToF-MS), we calculate emission ratios (ERs) and factors (EFs) for 21 volatile organic compounds (VOCs). We restrict our analysis to VOCs with sufficiently high lifetimes to be minimally impacted by oxidation over the ~8 h between when the smoke was emitted and when it arrived at the field site. We use oxidized VOC to VOC ratios to assess 15 the total amount of radical oxidation: maleic anhydride/furan to assess OH oxidation, and (cis-2-butenediol + furanone)/furan to assess NO3 oxidation. We compare ERs calculated from the freshest portion of the plume to ERs calculated using the entire nighttime period. Finding good agreement between the two, we are able to extend our analysis to VOCs measured in more chemically aged portions of the plume. Our analysis provides ERs and EFs for 9 compounds not previously reported for temperate forests in Australia: acrolein, pentanones/methylbutanal, methyl propanoate, methyl methacrylate, propene, maleic anhydride, 20 benzaldehyde, methyl guaiacol, and methylbenzoic acid. We compare our results with two studies in similar Australian biomes, and two studies focused on US temperate forests. We find mixed agreement for EFs presented from previous studies of Australian wildfires, and generally good agreement with studies focused on fires in the Western US. This suggests that comprehensive field measurements of biomass burning VOC emissions in other regions may be applicable to Australian temperate forests.

(EFs, in units of kg VOC emitted/kg fuel burnt) for temperate forests based on measurements in North America (Akagi et al., 2013;Burling et al., 2011), despite differences in fuel type, which is known to influence the speciation of VOCs emitted (Coggon et al., 2016;Hatch et al., 2017;Guérette et al., 2018). Recent studies address a limited number of VOCs emitted from wildfires in 40 Australian forests and savannahs (Paton-Walsh et al., 2014;Lawson et al., 2015;Wang et al., 2017;Guérette et al., 2018). These have shown that EFs of some VOCs (e.g. formic acid, ethane, monoterpenes, acetonitrile) can be 3 -5 times higher than those in measured in the US.
A complicating factor in deriving EFs from field observations is accounting for the influence of chemical processing. EFs are ideally based on observations close to the fire. When this is not possible, indicators of plume chemical age, such as oxidized VOC 45 (OVOC) to VOC ratios, can be used to diagnose the relative age of a plume. During the day, downwind VOC concentrations are primarily influenced by OH-initiated oxidation. At night, NO3-initated oxidation can significantly influence observed VOC concentrations (Decker et al., 2019;Kodros et al., 2020). There are several methods in existence for assessing daytime oxidation, but fewer are known for the night (De Gouw et al., 2006;Liu et al., 2016;Gregory et al., 2018;Decker et al., 2019). In this work, we use the maleic anhydride-to-furan ratio introduced in Gkatzelis et al. (2020) to assess OH oxidation. We examine the use of a 50 new OVOC/VOC ratio, cis-2-butenediol+furanone-to-furan, as an indicator of nighttime oxidation.
Here, we use observations from a proton-transfer-reaction time-of-flight mass spectrometer (PTR-ToF-MS) during the 2019-2020 Australian wildfire season to derive emission factors of 21 compounds, including 9 compounds for which there are no previous observations. We examine a subset of smoke-influenced nighttime observations made by a PTR-ToF-MS during the 2020 COALA (Characterizing Organics and Aerosol Loading over Australia) field campaign. NO3-initiated oxidation dominated the chemical 55 processing late in the night, as the plume travelled ~8 h to the field site from large, highly active fires to the south. Using co-located PTR-ToF-MS and FTIR measurements of CO2, CO and CH4, we derive EFs for 21 longer-lived VOCs (τBBVOC+NO3 ≥ average transport time), 9 of which have not been determined for Australian biomes. We compare these results with five related studies, two focused on Australian temperate forests, two focused on US temperature forests, and one reporting EFs used to represent temperate biomes across the globe. We find generally good agreement across several of these studies and discuss potential reasons 60 for discrepancies seen in EFs for select compounds.
2 Field Site and Instrument Description:

Field Site and Active Fires
The COALA field site was located in Cataract Scout Park (34.247 • S, 150.825 • E) at 400 m above sea level, 15 km inland, and 30 65 km to northwest of the nearest urban area (Wollongong, NSW). Fig. 1 shows the field site relative to the fires active between 1 Feb and 5 Feb 2020. We use the Suomi VIIRS thermal anomalies product filtering for points at high confidence levels to avoid counting any reflective false positives from plains or urban centers. Also plotted is the normalized difference vegetative index (NDVI) which is determined from measurements aboard the MODIS Terra satellite (Didan, 2021

PTR-ToF-MS and supporting observations
VOCs were measured using an Ionicon PTR-ToF-MS 4000 which operated with a mass resolution between 2000-3000 FWHM m/Δm and at a mass range spanning m/z = 18-256. The drift tube was held at a temperature of 70° C, pressure at 2.60 mbar, and 80 an E/N = 120 Td (electric field to molecular number density ratio). The instrument was housed in a climate-controlled unit, connected to a 15 m long, 1/4" OD PTFE insulated line attached to a 10 m tall mast, placing inlet height 0.5 m above canopy height. An assist pump was attached pulling an additional 3 SLPM for a residence time of 2.5 s. Peak separation of 1 min averaged spectra was conducted in Ionicon's PTR-Viewer 3 software.
Calibrations were performed using two VOC cylinders designed by Airgas on 31 Jan 2020, three days before measuring the smoke 85 event discussed here. The cylinders contained 17 compounds spanning a mass range of 33-154 Da and are shown in Table S1.
Many of these compounds are reported in the final EFs listmethanol, acetonitrile, acetaldehyde, acrolein, acetone, MVK+MACR, benzene, C8-aromatics, and C9-benzenes. All compounds used either do not fragment under these drift tube conditions or have known fragmentary peaks. Instrument zeros were determined using ultra-zero air. Limits of detection (3σ) for calibrated species are also given in Table S1 and range between 5-165 ppt. The raw counts per second (cps) were corrected for instrument 90 transmission, which was determined using a subset of the species in the calibration standards. Corrected cps are then normalized to the reagent ion signal (H3O + ccps x 10 6 , ncps) using the methodology described by Sekimoto et al. (2017). For compounds of interest not included in the calibration standards, we use the method described by Sekimoto et al. (2017), which yields uncertainties between 50-100%.
In addition to the PTR-ToF-MS measurements, we use observations of CO, CO2, and CH4 obtained from the collocated FTIR 95 system. Information of this instrument and its setup is provided in Griffith et al. (2012).   We use furan, a short-lived smoke tracer, and its oxidation products to determine which periods of the smoke event represent the 110 least oxidized plume. Furan is highly reactive with OH (kOH + furan = at 4.04 x 10 -11 cm 3 molec -1 s -1 298 K and 1 atm) and NO3 (kNO3 + furan = at 1.36 x 10 -12 cm 3 molec -1 s -1 at 298 K and 1 atm). OH-initiated oxidation produces maleic anhydride, which has low reactivity with both OH and NO3 (τOH = 3.99 days, τNO3 = 1.42 days with [OH]Avg = 2 x 10 6 molec cm -3 and [NO3]Avg = 8 x 10 7 molec cm -3 , with reaction rate constants from Grosjean (1992) and Bierbach (1994) and no reported direct emissions. The ratio of maleic anhydride-to-furan therefore provides a relative measure of the plume photochemical age. Using aircraft-based observations 115 of wildfire plumes in the Western US, Gkatzelis et al. (2020) found that maleic anhydride-to-furan ratios below 0.10 indicate the plume has undergone little OH processing.
Nighttime in-plume furan oxidation is dominated by NO3, with contributions from O3 (Decker et al., 2019). While many BBVOCs are highly reactive with NO3, there is substantially less research on indicators of NO3 oxidation. Decker et al. (2019) track NO3 chemistry using the ratio of total reactive nitrogen (NOy) to NOx, and Kodros et al. (2020) examine NO3-reacted products such as 120 nitrocatechol and nitrophenol of phenolic compounds (e.g. phenol, catechol, cresol). Measurements of NOy were not made during this field campaign, and NO3-products of phenols were subject to high uncertainty due to fragmentation in our PTR-ToF-MS measurement. We therefore examine a new indicator of NO3 processing using furan's dominant NO3 productscis-2-butenediol and furanone (Berndt et al., 1997). Both products are relatively long lived, with lifetimes estimated at τcis-2-butenediol=9 days and τfuranone= 8 h assuming an average concentration of [NO3] = 8 x 10 7 molec cm -3 (O'dell et al., 2020). Lab based studies and field 125 campaigns conducted in the US and Australia suggest that furan and furanone EFs are comparable, with study-averaged values for furan ranging from 0.132 -0.51 g kg -1 and 0.27 -0.57 for furanone (Andreae and Merlet, 2001;Akagi et al., 2011;Hatch et al., 2015;Stockwell et al., 2015;Liu et al., 2017;Koss et al., 2018;Selimovic et al., 2018). No furan EFs have been reported for Australian temperate forests and only one furanone EF is reported from Lawson et al. (2015) at a comparable value at 0.57 g kg -1 .
Additionally, emissions modeled in Decker et al. (2019) from wildfires suggest that furan and furanone are emitted in roughly 130 equal proportions. As such, we operate not on the assumption of negligible OVOC emissions, but that variability in OVOC/VOC ratios are driven by chemical aging. Fig. 2 shows furan enhancements, which begin later on Feb 3 than acetonitrile enhancements. Maleic anhydride concentrations are high during the initial period of the smoke event, suggesting significant OH-initiated processing throughout the day before the plume reached the site. After sunrise, furan decays faster than CO, and maleic anhydride concentrations begin to rise, again showing 135 the impact of OH-initiated oxidation. Cis-2-butenediol and furanone are both measured at m/z 85. Enhancements in m/z 85 are seen when the smoke arrives and vary throughout the night. Just prior to sunrise (04:00 -06:15, local time), both OVOC/VOC ratios rapidly decrease (Fig. 3), corresponding with a rise in furan, CO, and acetonitrile. Maleic anhydride/furan drops to 0.05, which is within the lower range of the chemically younger plumes reported by Gkatzelis et al. (2020). The ratio of m/z 85 to furan is around 2.5. While we cannot use this to quantify plume age since the two products are measured as a sum, we note that this 140 period constitutes the lowest ratio throughout the event, with surrounding periods having ratios 1.6 -2.8 times greater. We note that at a value of 2.5, this plume has likely undergone significant aging, despite this being the freshest smoke detected during the campaign.
The rapid decreases in OVOC/VOC ratios are unlikely to result from shifts in chemistry alone. Instead, this suggests a shift in meteorological conditions which brings in smoke from a closer source, in agreement with measured wind direction, which shift 145 from northeast to north at this time. We further investigate plume transport using a back-trajectory model. We use a HYSPLIT back-trajectory model (Stein, 2015) to determine the origin and transport time of the smoke arriving at the site throughout the smoke event. The meteorological input used is the Global Data Assimilation (GDAS) dataset. The model was set to assess trajectories at three different altitudes at 10 m, 500 m, and 1500 m above ground level (agl) to capture plume height. Our 150 period of interest spans from 17:00 Feb 3, just before CO enhancements are seen at the site, to 06:00 4 Feb when furan concentrations rapidly decrease. The model was set to calculate a new 12 h trajectory every hour during this time. Back trajectories are shown in Fig. 1. For every hour in the event (each represented by a color), one can track the origin of the sampled airmass 12 h in advance of its arrival.

Plume Origin and Transport Time
A shift in trajectories occurred between 17:00 and 18:00 3 Feb, corresponding with the arrival of the smoke plume as indicated by 155 observed CO enhancements. Subsequent trajectories originate near the fires located ~230-375 km from the field site on the southeast coast. The model shows that air masses initially kept at low altitude and were lofted to ~560 m agl when passing over the active fires ~25 km to the south, near Canberra (Fig. S1). The plume descended to 10 m agl as it reached the coast. The model suggests smoke sampled later in the evening (between 04:00 -06:00 4 Feb) spent more time over land compared to previous points in the event. This shift in trajectories and the increasing intensity of fires near Canberra during this time likely contributed to the 160 decrease in OVOC to VOC age marker ratios. Over the entire course of the event, HYSPLIT analysis suggests transport time from the fires to the field site is around 8 h (>200 km), but potentially shorter for the time frame immediately prior to sunrise (5-8 h, 25 km).

165
To identify compounds which would be suitable for EF derivation, we compare the list of measured ions with compounds identified in previous literature such as Brilli et al. (2014), Hatch et al. (2015), Gilman et al. (2015), Stockwell et al. (2015), Bruns et al.
(2017), Koss et al. (2018), and the PTR Library (Pagonis et al., 2019). To corroborate species assignment, we examine correlations of identifiable compounds with CO, acetonitrile, furans, and phenolic compounds which are well-established smoke tracers. We also examine tracer-tracer relationships, for instance the anti-trend between maleic anhydride and furan resulting from OH 170 oxidation. We exclude compounds with low proton affinities that are known to have humidity-dependent calibration factors (e.g., HCHO, HCN). This results in 150 identified VOCs species measured and identified during the smoke event.
We further filter our VOC list by two criteria. First, VOC + NO3 reaction rates must be included either in the NIST Chemical Kinetics Database (Manion, 2015) or Master Chemical Mechanism (v3.3.1) (Bloss et al., 2005;Jenkin, 1997;Jenkin et al., 2003;Saunders et al., 2003). Second, the VOC must have a significantly long lifetime against NO3 oxidation to be minimally impacted 175 over the 8 h transit time from the active fires to the field site (τBBVOC+NO3<8 h, again assuming [NO3] = 8 x 10 7 molec cm -3 ). This limits subsequent analysis to 21 long-lived VOCs.

Calculating Emission Ratios
An ER is defined here as the slope of a linear regression of a given VOC to CO (both in units of ppb). Following Guérette et al. (2018), ERs are reported if correlation between a given VOC and CO are well correlated, with R 2 ≥ 0.5. High correlation minimizes 180 the impact of the choice of regression method (e.g. orthogonal, York) on calculated slopes (Wu and Yu, 2018). We use a standard linear regression with no error weighting, noting that measurement uncertainty is substantially lower than natural variability.
We first derive ERs using all data from the "freshest" portion of the plume as determined from OVOC/VOC ratios (Marked "D" in Fig. 2). This produces 17 ERs that meet our criteria. We expect this period to provide the most accurate representation of original https://doi.org/10.5194/acp-2021-742 Preprint. Discussion started: 9 September 2021 c Author(s) 2021. CC BY 4.0 License. VOC emissions. We then calculate ERs for more aged portions of the smoke event (Periods A-C, Fig. 2), performing regression 185 analysis on the chemically distinct time periods. The start and end time of each period is determined by visual inspection of VOC/CO behaviors, which all exhibit similar distinct periods. Fig. 4 provides an example of the analysis using acrolein. We average the slopes from each of these lines to derive an average ER for the full smoke event and compare to just the freshest portion of the plume (Period D). We find the using only the freshest smoke compared to using all the data generates very similar results for 14 of the 17 compounds. Relative differences of the resultant ERs are within 1.5 -47 % with three outliers: butenes (302 %),

190
C8-aromatics (88%), and C3-benzenes (212 %). Five compounds have only 1 ER from all 4 periods (acetylene, pentene, maleic anhydride, benzaldehyde, methyl guaiacol) so there is no standard deviation, but the remaining 12 compounds from period D are captured within 1σ of ERs from periods A-D (shown in Fig. S2). Good agreement between methods allows us to extend our analysis beyond the freshest part of the plume, and therefore allows us to report ERs for a larger number of compounds. When focusing only on the freshest part of the plume, ethyne, pentene, maleic anhydride, and benzaldehyde must be excluded due to insufficient 195 R 2 with CO. All ERs reported here and used in EF calculation use the "average over evening" method and include these compounds.
Additionally, only one ER for CO2 and CH4 have been calculated using the dataset from periods A-D. Both these compounds are long-lived, and from visual inspection, they do not form distinct time periods like the VOC ERs (shown in Fig. 4).

Calculating Emission Factors
Emission factors are defined as the mass of some trace gas emitted per mass of dry biomass burnt. The most direct way of calculating this quantity is capturing total emissions released from a fire as well as knowing the quantity of fuel burnt. Unless experiments are conducted in a laboratory setting, these quantities are not known. As such, emission factors are calculated according to the carbon mass balance method (Akagi et al., 2011;Selimovic et al., 2018), using CO as the reference gas for the 21 where Fcarbon=0.5 and is the assumed carbon fractional content of the fuel as used in previous studies (Akagi et al., 2011;Paton-Walsh et al., 2014). MMX is the molar mass of compound X, MMC is the molar mass of carbon, ERX/CO is the CO ER of X, and ∑ERY/CO is the sum of ERCO2/CO, ERCH4/CO, and ERCO/CO. These ERs constitute the major volatilized carbon components of the 215 plume, but the resulting EFs may be overestimated by 1-2% (Andreae and Merlet, 2001) as this method assumes all volatilized carbon is detected including particulate carbon, VOCs, CO, and CO2.
EFs derived in this work are presented in Table 1 alongside results from 2 eastern Australia-based studies by Lawson et al. (2015) and Guérette et al. (2018), 2 western US-based studies sampling emissions from corresponding temperate fuel types by Liu et al. (2017) and Permar et al. (2021), and 1 study by (Akagi et al., 2011) that provides EFs for general temperate zones. First, in comparison with the Australia-based studies, Guérette et al. (2018) reports EFs notably larger than those presented in this work, with only benzene and C8-aromatics showing good agreement. Except for these two compounds and C3-benzenes, Guérette et al. (2018) reports larger EFs than Lawson et al. (2015) and none within agreement. Our results more closely agree with Lawson et al (2015) with methanol, acetone, and furanone EFs within 1σ, and acetonitrile and acetaldehyde falling within a factor of 2.

220
This agreement is likely due to both this work and Lawson et al. (2015) examining opportunistically intercepted smoke plumes that experienced some processing whereas Guérette et al. (2018) sampled near-source, controlled ground burns. Guérette et al. (2018) reports an acetonitrile EF ~4.5 times higher than this work and ~3 times greater than Lawson et al. (2015) constituting one of the largest disparities. This is attributed to the native and abundant Acacias which are N-fixing species located mainly in forest 235 understories. Their measurements likely had a higher proportion of this foliage constituting the total fuel load due to both proximity to the forest floor and resulting leaf litter. Another of the largest differences is MVK+MACR, which shows a disparity of ~6 times this work and 3 times that of Lawson et al. (2015). This is also most likely explained by differences in sampling approach in that proportional contributions of vegetation vary and plumes in Guérette et al. (2018) did not undergo any dilution or photochemical processing.

240
In comparison with US-based studies, ethyne, methanol, acetonitrile, butenes, acetone, and benzene agree across both studies within 1σ, with acrolein, pentanone, methyl propanoate, C3-benzenes, and methyl guaiacol agreeing very well with values reported by Permar et al. (2021). It should be noted that though within the estimated uncertainties, the value for methyl guaiacol reported by Permar et al. (2021) is ~3.5 times greater than the value in this work, which constitutes another of the largest disparities in this dataset. Additionally, ethyne and methanol agree well with values from Akagi et al. (2011). However, propene is on the range of 245 ~3 -4 times less than values from Akagi et al. (2011) and Permar et al. (2021), and while within a factor of 2 is outside of any variability in (Liu et al., 2017). Pentene is not within agreement of either US-based study as both have small corresponding variabilities (between 1 -5 %), and additionally is ~2.5 times less than the value in (Liu et al., 2017). This could result from differing fuel types. The EF for furanone in this work is also expectedly larger than both other values presented here at ~3 times greater than Permar et al. (2021). This is due to the plume sampled in this work undergoing the longest transport of any plumes 250 measured in other studies.
Perhaps an unexpected finding is that EFs derived in this work agree better with observations in the US than the Guerette et al.
(2018) study, which was in the same region as the COALA measurements. It should be noted that all studies except Guerette et al.
(2018) are from plumes sampled several km downwind. Differences previously characterized as arising from varying fuel types may actually result from measurement approaches to deriving EFs and proximity to emission source. Agreement across results 255 from this work and from the US-based studies lends credence to the use of newly presented EFs for modeling purposes in temperate Australian forests.

Conclusions
EFs were derived for a total of 21 trace gas species via measurements from a PTR-ToF-MS and an FTIR spectrometer. The COALA ground-based field campaign opportunistically sampled a sustained biomass burning plume from 3 -4 Feb 2020 during the 2019-260 2020 wildfire season in New South Wales, Australia. We determined via HYSPLIT trajectories that the most likely pathway traveled by the plume was from a distance ranging from ~230-375 km south from fires along the temperate forests of the east coast with contributions from more inland fires near Canberra, Australia. This plume lofted to an altitude of 500 m agl as it passed over active fires ~8 h out from the field site, before descending down to 10 m agl while traveling over the ocean, and reaching the site at 17:30 local time. All data used in the derivation of EFs was limited from sunset on 3 Feb to sunrise on 4 Feb as this period 265 showed the greatest enhancements of reactive BB tracers like furan. Through visual inspection, we partitioned this plume event into 4 portions, and calculated and averaged the individual ERs. We used two age marker ratios derived from furan radical oxidation to determine the freshest portion of the plume and found that ERs from this portion corresponded well with the averaged ERs  Lawson et al. (2015) and 2 others are well within a factor of 2, which indicates decent agreement. Furthermore, comparison with two recent US studies that report data on analogous temperate zones, as well as one report covering global temperate regions, show generally good agreement for 11 of the 21 compounds, with several others within a factor of 2. This closer agreement with these studies, as well as that of Lawson et al. (2015), is likely due to the measurement approach when deriving EFs as both US-based studies were aircraft campaigns, and the 280 Australia-based study intercepted a transported plume much like this work. Guérette et al. (2018) sampled controlled burns on a ground campaign virtually at the emission source. This indicates that variability previously ascribed to differing fuel types may be overshadowed by sampling approach and that comprehensive measurements from US-based studies may be useful for studying Competing interests. The authors declare that they have no conflict of interest.

295
Acknowledgements. This work was supported by NSF grant GR00003303. We thank Travis Naylor, Ian Galbally and all the UOW COALA team for their aid in conducting measurements during the field campaign and all input thereafter. We additionally gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for providing the HYSPLIT transport and dispersion model used for analysis in this publication. We acknowledge the use of data and/or imagery from NASA's Land, Atmosphere Near real-