Formation and evolution of tar balls from northwestern US wildfires

Abstract. Biomass burning is a major source of light-absorbing black and brown
carbonaceous particles. Tar balls (TBs) are a type of brown carbonaceous
particle apparently unique to biomass burning. Here we describe the first
atmospheric observations of the formation and evolution of TBs from forest
fires. Aerosol particles were collected on transmission electron microscopy (TEM) grids during aircraft
transects at various downwind distances from the Colockum Tarps wildland fire.
TB mass fractions, derived from TEM and in situ measurements, increased from
<1 % near the fire to 31–45 % downwind, with little change in TB
diameter. Given the observed evolution of TBs, it is recommended that these
particles be labeled as processed primary particles, thereby distinguishing
TB formation–evolution from secondary organic aerosols. Single-scattering
albedo determined from scattering and absorption measurements increased
slightly with downwind distance. Similar TEM and single-scattering albedo results were observed
sampling multiple wildfires. Mie calculations are consistent with weak light
absorbance by TBs (i.e., m similar to the literature values 1.56−0.02i or
1.80−0.007i) but not consistent with absorption 1 order of magnitude stronger
observed in different settings. The field-derived TB mass
fractions reported here indicate that this particle type should be accounted
for in biomass burning emission inventories.



Aircraft Platform
The BBOP field campaign comprised two deployments -a wildfire deployment staged in the Pacific Northwest (Richland, WA) from July -mid-September 2013 and an agricultural (controlled) burn deployment staged in the lower Mississippi Valley (Memphis, TN) during October 2013 (Figure 1). Schmid et al. (2014) described the research platform used during the BBOP field campaign. The aircraft was equipped with an isokinetic aerosol inlet on the starboard side of the aircraft and 60 a dedicated trace gas inlet on the port side. Ambient aerosol samples are brought to individual instruments via conductive tubing while trace gas sampling utilized PFA tubing and/or T316 stainless steel. All instruments were time synced prior to every flight. The Gulfstream-1 (G-1) aircraft was outfitted for four measurement classes (i) microphysical properties, (ii) optical properties, (iii) trace gases, and (iv) radiation. Below are measurement details for those instruments in the present study. 65

Electron Microscopy.
Aerosol particles were collected onto TEM lacey-carbon substrate grids (Type 01881, Ted Pella, lacey Formvar/carbon, 200 mesh Cu grids) using a two-stage aerosol impactor sampler (AS-16W, Arios) that could collect 16 TEM grid samples per flight. Sampling times ranged from 1 to 30 minutes depending on particle loading. The lower and upper 50% cutoff 70 aerodynamic diameters for the samples used in this study are approximately 100 nm and 700 nm, respectively.
Particle analysis was carried out using a 120kV transmission electron microscope (JEM-1400, JEOL) equipped with an energy-dispersive X-ray spectrometer (EDS; Oxford instruments). A total of ~ 3300 particles were examined from representative images of the 16 Colockum Flats TEM grids in order to determine TB number fractions. A Gatan 628 singletilt holder was used for TB heating experiments (Adachi et al., 2017). Ns-soot volumes as determined from 2D TEM images 75 using fractal parameters from Adachi et al (2010) had the same qualitative trends derived from SP2 measurements of refractory black carbon (rBC). particles. A single-wavelength photothermal interferometer (PTI) was used to acquire in situ aerosol absorption at 532 nm 85 (Sedlacek and Lee, 2007;Cross et al., 2010).

Aerosol Mass Spectrometry
An Aerodyne SP-AMS (Onasch et al., 2012) was used to measure the non-refractory particulate matter (NR-PM), including organic and inorganic aerosol. Standard AMS measurement uncertainty is estimated to be 25% as discussed by Canagaratna 95 et al. (Canagaratna et al., 2007). The SP-AMS uses two methods for vaporizing particles. As in a conventional AMS, a resistively heated tungsten vaporizer flash volatilizes non-refractory aerosol in a vacuum followed by electron ionization and detection via mass spectrometry (Canagaratna et al., 2007). With dual vaporizers, light absorbing refractory aerosol are vaporized by a Nd:YAG laser, after which the particle beam encounters the tungsten vaporizer. The tungsten vaporizer was calibrated with ammonium nitrate and the laser vaporizer was calibrated with atomized Regal black multiple times during 100 the BBOP study following established protocols (Onasch et al., 2012;Canagaratna et al., 2007;Willis et al., 2014). Onasch et al. (2012 discuss the complications that may occur with dual vaporizers with respect to collection efficiencies (CEs). Lee et al. (2015) observed the CE of ambient NR-PM increased when both vaporization sources were used. During the BBOP campaign, the SP-AMS was operated using the tungsten vaporizer with or without laser vaporization; laser-on or laser-off modes, respectively, and also observed higher NR-PM for laser-on conditions. Changes between these two modes were 105 done from flight to flight, from one plume transect to the next, and sometimes on shorter time scales. The CE for the SP-AMS operating in laser-off mode was determined to be 0.5 from 2 intercomparison flights with a surface-based AMS at Mount Bachelor Observatory (Collier et al., 2016) and correlations with scattered light measured at 550 nm, where the massspecific scattering coefficients (MSC) are expected to be near 3.6 ± 1.2 m 2 g -1 (Hand and Malm 1990). All the data for the Colockum Tarp fire presented here were collected with the SP-AMS in laser-on mode. In order to determine the appropriate 110 CE for non-refractory aerosol under laser-on conditions, we compared laser-on with laser-off measurements for five similar wildfires measured during BBOP, in which 16 plume transect pairs were sequentially sampled under similar conditions for laser-on and laser-off. Averages of the NR-PM measured by the SP-AMS during these sequential plume transects were normalized by measured CO and/or light scattering and the laser-on to laser-off ratios calculated. Counting each transect Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-41 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 30 January 2018 c Author(s) 2018. CC BY 4.0 License.
pair as a data point and using the laser-off CE = 0.5 as the presumed ambient NR-PM loadings, we obtain a laser-on CE = 115 0.76 (1σ = 0.10) with CE values that range from 0.63 to 0.88 for a given transect pair.

Plume Sampling Strategy
For the wildfire flights, a pseudo-Lagrangian sampling protocol was employed in which flight transects orthogonal to the plume direction were conducted at selected distances downwind of the source. The plume age was calculated using prevailing wind speed and direction together with the assumption that the emission source was constant for all sampling 120 transects.

Photochemical Age and Plume Dilution
Photochemical age of the smoke plume was calculated using the ratio of NO x to NO y as described by Kleinman et al. (2008).
Effects of plume dilution were corrected by normalizing data stream (e.g., scattering, absorption, mass loading) by CO enhancement -a conserved tracer. 125

Colockum Tarps Fire
The Colockum Tarps fire sampled on the afternoon of 30-July 2013 is representative of at least eight wildfire plumes studied in BBOP with multiple transects covering ~2 to 4 hours of transport. The fire was first noted on 27 July and was declared 98% contained on August 15 3 . A total of ~80,000 acres, consisting of short grass, timber grass understory, and hardwood litter were burned. An analysis of the time dependence of gaseous and aerosol constituents shows that the fire is typical of 130 several other wildland fires sampled in the BBOP campaign, the description of which will be the subject of future publications. Figure 2 shows the aircraft ground track for the afternoon July 30 th flight. There are 13 plume crossings and one upwind leg.
Each is nearly orthogonal to the plume transport direction, as determined from aircraft wind observations. Locations of TEM samples are color-coded and labeled in Figure 3 in sequential order (1 to 16) and according to their downwind position 135 (T0 to T6). There are two sets of transects, referred to here as Set A and Set B, each of which cover a sampling distance from the main fire region to ~35 km downwind in 6 steps from A-T1 to A-T6 and B-T1 to B-T6 ( Figure 3). TEM sample 1 at location T0 was collected upwind of the Colockum Fire. Plume travel, or aging, times between consecutive transects is approximately 30 minutes, and the aging time probed is ~ 3 hours. Concentration of CO, a common combustion tracer, 3 http://inciweb.nwcg.gov/incident/3567/ Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-41 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 30 January 2018 c Author(s) 2018. CC BY 4.0 License. during a time series (Figure 3) ranges between approximately 0.2 ppm upwind to 5 ppm over fire hot spots near transects T1 140 and T2. Peak concentrations decrease with downwind distance due to dilution. NO x emitted by forest fires is oxidized to peroxyacetyl nitrate (PAN), organic nitrates, and HNO 3 , which collectively are detected as NO y (Fisher et al., 2010;Fisher et al., 2016). A qualitative measure of atmospheric oxidative processing is thereby provided by -log 10 (NO x /NO y ), as described by Kleinman et al., (2008), and its increase with downwind distance ( Figure 4). Little change occurs during the one hour separating the first set of transects (Set A) from the second set (Set B), 145 indicating fire conditions were approximately steady during that time period.

BBOP ARM Data Archive
All data used in the present analysis are publicly available on the Department of Energy ARM data archive (http://www.archive.arm.gov/armlogin/login.jsp). DOE ARM verifies data quality through quality assurance and data quality checks. 150

Tar Ball Production:
Four TEM images of aerosol particles with increasing photochemical age, collected near the active fire (0 hrs) and at three downwind distances, are shown in Figure  Over the fire, the TB number concentration is near zero and OA particles dominate. As the plume ages, the TEM images 160 reveal an increase in the number fraction of TBs, with the highest number fraction, 64%, observed in the most aged portion of the plume (A-T6 & B-T6 in Figure 3). In air that has the chemical signature of aged, diluted BB smoke (A-T0 in Figure   3, which was collected upwind of the fire), the number fraction of TBs is 36%. These number fractions are similar to those reported by others (Pósfai et al., 2003;Pósfai et al., 2004;Hand et al., 2005;China et al., 2013).
Since TEM number fractions reported here and elsewhere (Hand et al., 2005;China et al., 2013) are sensitive to the loss of 165 volatile particles, a more useful statistic is the number ratio of TBs to ns-soot, which is 3.8 and 7.6 for the downwind transects 15 and 16 ( Figure 3). The TB/ns-soot number ratio in aged, diluted BB air sampled upwind of the fire (i.e., background) is 10.8. These ratios are similar to the number ratio of 10 reported by China et al. (2013). If the Tóth (2014) mechanism -direct injection of precursor 'tar' droplets followed by rapid heating -were indeed responsible for TB formation, then the BBOP TEM images should reveal TBs at the source, since that is where the high-170 temperature zone, necessary for particle solidification, is located. Instead, there is a steady growth in TB/ns-soot mass ratio as the plume ages. This increase indicates that TBs are not directly emitted, but rather form (or transform) in the atmosphere within a few hours from emission. Measurements of TB abundance in smaller fires suggest that rapid changes can occur within 15-minute of atmospheric aging (Adachi and Buseck 2011). Adachi and Buseck (2011) observed 10-fold more TBs in the plumes of more-distant fires compared with nearby fires. In contrast to the present study, fresh and aged smoke 175 samples reported by Adachi and Buseck (2011) were in general from different fires. TEM samples with more TBs showed OA morphological changes consistent with an increase in viscosity. Apparently, atmospheric processing causes some OA particles that start out deformable to solidify (Adachi et al., 2010) and be recognized as TBs in the TEM images. The atmospheric processing leading to TBs may involve dehydration and oligomerization of low-volatility OA (Pósfai et al., 2004;Adachi and Buseck, 2011). 180 Sampling within a single fire as presented here provides further evidence that TBs are processed primary particles (see Discussion Section on nomenclature). The downwind increase in TB mass is due to an increase in number concentration rather than a change in particle size (green 25 th , 50 th , and 75 th percentile lines in the lower panel of Figure 6). To accomplish a number increase by condensation would require that particles smaller than the 100-nm sampling cutoff for accurate TEM imaging grow into a detectable size range, a feature not observed in the TEM measured size distributions from which 185 average TB diameter is calculated (upper trace in Figure 6). TEM samples often included background air with varying properties along with the plume being investigated and is likely responsible for the spread observed between the 25 th and 75 th percentile lines derived from individual data points (dashed and dot-dashed green lines in Figure 5).

Tar ball Mass Fractions
To date, the only way to definitively identify TBs has been through microscopy. However, issues associated with particle 190 volatility have severely limited evaluations of the radiative forcing contribution by this aerosol type. In response to this limitation, we combine TEM image analysis with SP2 and SP-AMS measurements to derive the first estimates of the mass loadings and mass fractions of TBs in wildland fire plumes. TB mass loadings and the TB mass fractions are derived by combining i) the number ratio of TBs to ns-soot determined from TEM images with ii) speciated aerosol concentration measurements from the SP2 and SP-AMS, averaged over the TEM collection time periods. 195 In our calculation of the TB mass fractions, we make three significant assumptions: First, TEM grids collect TBs and ns-soot with the same efficiency; Second, ns-soot, as determined by TEM, is the same as rBC measured with an SP2. A similar assumption was recently invoked by Adachi et al. (Adachi et al., 2016). For convenience, both quantities will be called "soot", and; Third, the SP-AMS collection efficiency (CE) for TBs is the same for other NR-PM. Adachi et al (2017) Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-41 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 30 January 2018 c Author(s) 2018. CC BY 4.0 License. studied the thermal vaporization of TBs by heating BBOP samples to 600 o C and found them to be less volatile than most 200 OA, suggesting that TBs may exhibit a different CE in the SP-AMS than other OA (see supplemental). However, given the current lack of chemical information on TBs in general, TB chemical signatures cannot be deconvolved from SP-AMS OA mass spectra. Therefore, calculations are done assuming TBs have the same CE as other OA.
Ambient particulate mass concentrations (e.g. µg m -3 at STP) of TBs, soot, non-refractory inorganics, and non-refractory organics are denoted as M TB , M soot , M IN , and M ORG , respectively. TB fractional contribution to total mass is denoted by f TB 205 and we define R as the ratio M TB /M soot derived from TEM. We have: The SP-AMS was used to measure M AMS , the non-refractory component of M total . 210 Except for a sensitivity calculation 4 , the SP-AMS collection efficiency, CE, is assumed to be the same for all non-refractory species; that is CE ORG = CE IN = CE TB . Equation 2 is written with separate contributions to total mass from TBs and organics.
However, in practice, neither M ORG nor M TB can be obtained from Eq. 2 because additional chemical information is needed to allow the SP-AMS to discriminate tar ball organic signals from other organic particulate matter. Combining Eqs. 1 and 2, 215 we obtain For an error analysis, it is convenient to restate Eq. 3b in terms of R, To a first approximation, sub-micrometer aerosol mass loadings in wildfire plumes consist of less than 10% ns-soot and inorganic constituents, with the remainder split between OA and TBs. TB/ns-soot number ratios derived from TEM (Adachi et al., 2010) are used to derive volume ratios by combining the observed TB diameters from TEM with the rBC volume equivalent diameters (VEDs) derived from SP2 measurements. As discussed above, this approach assumes that ns-soot, as determined by TEM, is the same as rBC measured by the SP2. Use of the SP2-derived ns-soot VED is considered superior 225 to a TEM-derived VED since the latter is very sensitive to fractal parameters used (Adachi et al., 2010). These volume ratios can be converted into mass ratios using assumed densities. For a TB density (Alexander et al., 2008) of 1.5 g cm -3 and an ns-soot spherule density (Park et al., 2004) of 1.8 g cm -3 , the TB/soot mass (volume) ratios are 16.3 (19.6) and 39. 3 (47.2) for the downwind transects 15 and 16, respectively. Figure 7a shows the averaged TB/soot mass as a function of plume age. 5 230 The calculation of f TB for a typical downwind sample is given in Table 1. A one standard deviation accuracy has been estimated by combining individual measurement errors (listed in Table 1) in quadrature, assuming that they are uncorrelated.
It is seen that the accuracy of f TB depends primarily on R, which we estimate has a 50% uncertainty. Figure 7b shows that TB mass fractions increase with plume age. Over the fire, this mass fraction is close to zero and increases in the aged plume to 31-45%, or approximately a third of the particle mass in the smoke plume. The error bars in 235

Nomenclature
The results presented here raise the question as to how best to describe TBs. Are they processed primary particles, as we have elected to describe them here, or are they secondary particles? The argument for the latter can be found in the atmospheric chemistry community where SO 2 is a primary pollutant but its oxidation product, sulfate, is called 250 secondary. Applying similar logic, if emitted organic particles are converted to tar balls via oxidation reactions TBs would be classified as secondary. However, within the aerosol community, secondary organic aerosol (SOA) has exclusively been used to describe particulate matter condensed from the gas phase. We did not observe TBs on the TEM grids exposed over the fire but they were found in increasing numbers downwind. Because we did not observe particle growth (Figure 7c) a gas-to-particle condensation mechanism for TB formation seems implausible. We have therefore elected to label TBs as 255 processed primary particles.

Inventory Implications:
Top-down, bottom-up comparisons of satellite-retrieved optical properties (e.g., aerosol optical depth or AOD) with inventory-based optical properties reveal a discrepancy that requires emissions to be scaled up by a factor ranging from 1.5 (Reddington et al., 2016) to over 10 (Bond et al., 2013;Lu et al., 2007;van der Werf et al., 2010;Kaiser et al., 2012;Kopacz 260 et al., 2010;Reid et al., 2009). Our results and those of Adachi et al. (2017) underscore the importance of studying TBs, including their physio-chemical and optical properties, such that new techniques can be developed to identify (e.g. via specific chemical signatures) and accurately account for TBs in bottom-up inventories. Several scenarios are possible, one of which is that TBs may have escaped accounting in wildfire inventories (Urbanski 2014;Reid et al., 2005), which would reduce the top-down/bottom-up discrepancy. Liu and co-workers (2017) recently highlighted the importance of quantifying 265 wildfire and prescribed burn emission factors.

Optical Properties
Inconsistencies in the values of the TB refractive index (Table 2) preclude a useful assessment of the radiative impact of these particles. Aerosol SSA was calculated on the basis of the mass fractions, mass absorption coefficient (MAC) and mass scattering efficiency (MSE) of each species. A mass-weighted average was used as if the aerosol was an external mixture. 270 More complicated schemes were not warranted in the absence of mixing state information. Using the mass fractions determined from Eqn. 3b (4) and the definition of SSA, we obtain Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-41 Manuscript under review for journal Atmos. Chem. Phys. where i labels the four aerosol components in our calculation.
Comparisons of the SSA derived from optical measurements with Mie calculations that assumed TBs were either weak 275 (1.56 -0.02i) or strong absorbers (1.67 -0.27i) are shown in Figure 8 as a function of plume age. When TBs are assumed to be stronger absorbers, as reported by Alexander et al. (2008) and Hoffer et al. (2016), the increase in TB mass fraction as a function of plume age should result in a large decrease in SSA. However, such a decrease is not observed. Instead, there is a slight increase in SSA with age. This comparison indicates that TBs in the Colockum Tarp wildfire are weaker absorbers than suggested in laboratory studies (Hoffer et al., 2016;Hoffer et al., 2017). Other wildfires sampled in BBOP had 280 comparable TB-to-soot ratios and age trends. SSA's were also similar in magnitude and time trend to the Colockum Tarp wildfire so we believe our observations at least pertain to fires in this region. Table 2, a log normal fit to the observed size distribution (count median diameter =242 nm, geometric standard deviation = 1.32), and a TB density of 1.5 g cm -3 . Results are given in Table 3 along with literature values used for the other aerosol components. 285

MAC and MSE for TBs were determined from Mie calculations based upon literature values of refractive index in
Plausible changes to optical parameters other than the MAC of TBs had little effect on the results in Figure 8.
The causes for disparate measurements of TB refractive index have not been identified. We note that high absorptivity TBs made in the laboratory via the Tóth et al. (2014) mechanism, such as those used by Hoffer et al. (Hoffer et al., 2016;Hoffer et al., 2017), are formed differently than the ambient TBs reported here. The other high absorption result in Table 3 reported by Alexander et al. (2008) is from an EELS (electron energy-loss spectroscopy) spectrum of a collected ambient aerosol, but 290 it is not clear that the TBs analyzed by Alexander have a BB origin. These comparisons suggest there may be variability in TB optical properties dependent upon fuel source, formation mechanism, and, potentially, aging.

Radiative Forcing Implications:
Incorporation of TBs into models of BB (Jacobson, 2014) have been hampered because the only technique for detecting these particles is single particle microscopy, which is an off-line, intensive technique with sampling issues (e.g., does not 295 quantify volatile aerosol lost in storage or upon electron beam irradiation). Our work shows that TBs can represent a significant fraction of the aerosol mass in some wildfire plumes and that regional effects of TBs over biomass burning dominated regions would be much more significant. Inclusion of TBs also will help to constrain the BC radiative forcing in biomass burning.

Acknowledgements 300
The authors gratefully acknowledge Ernie Lewis and Leah Williams for help with laboratory studies of TBs as well as discussions about the AMS collection efficiency in the derivation of the mass fraction expressions. This research was Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-41 Manuscript under review for journal Atmos. Chem. Phys. Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-41 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 30 January 2018 c Author(s) 2018. CC BY 4.0 License.

Upwind
Over the fire Downwind