Dilution impacts on smoke aging: Evidence in BBOP data

Biomass burning emits vapors and aerosols into the atmosphere that can rapidly evolve as smoke plumes travel downwind and dilute, affecting climateand health-relevant properties of the smoke. To date, theory has been unable to explain variability in smoke evolution. Here, we use observational data from the BBOP field campaign and show that initial 30 smoke concentrations can help predict changes in smoke aerosol aging markers, number, and diameter. Because initial field measurements of plumes are generally >10 minutes downwind, smaller plumes will have already undergone substantial dilution relative to larger plumes. However, the extent to which dilution has occurred prior to the first observation is not a measurable quantity. Hence, initial observed concentrations can serve as an indicator of dilution, which impacts photochemistry and aerosol evaporation. Cores of plumes have higher concentrations than edges. By segregating the 35 observed plumes into cores and edges, we infer that particle aging, evaporation, and coagulation occurred before the first measurement, and we find that edges generally undergo higher increases in oxidation tracers, more decreases in semivolatile compounds, and less coagulation than the cores. https://doi.org/10.5194/acp-2020-300 Preprint. Discussion started: 6 April 2020 c © Author(s) 2020. CC BY 4.0 License.

The evolution of total particulate matter (PM) or organic aerosol (OA) mass from smoke has been the focus of 60 many studies, as PM influences both human health and climate. Secondary organic aerosol (SOA) production may come about through oxidation of gas-phase volatile organic compounds (VOCs) that can form lower-volatility products that partition to the condensed phase (Jimenez et al., 2009;Kroll and Seinfeld, 2008). SOA formation may also arise from heterogeneous and multi-phase reactions in both the organic and aqueous phases (Jimenez et al., 2009;Volkamer et al., 2009). In turn, oxidant concentrations depend on shortwave fluxes (Tang et al., 1998;Tie, 2003;Yang et al., 2009). Smoke particles contain semivolatile organic compounds (SVOCs) (Eatough et al., 2003); (May et al., 2013), which may evaporate off of particles as the plume becomes more dilute (Formenti et al., 2003;Huffman et al., 2009;May et al., 2013), leading to losses in total aerosol mass. Field observations of smoke PM and OA mass normalized for dilution (e.g. through an inert tracer such as CO) report that for near-field (<24 hours) physical aging, net PM or OA mass can increase (Cachier et al., 1995;Formenti et al., 2003;Liu et al., 2016;Nance et al., 1993;Reid et al., 1998;Vakkari et al., 2014Vakkari et al., , 2018Yokelson et al., constant (Brito et al., 2014;Capes et al., 2008;Collier et al., 2016;Cubison et al., 2011;Forrister et al., 2015;Garofalo et al., 2019;Hecobian et al., 2011;Liu et al., 2016;May et al., 2015;Morgan et al., 2019;Sakamoto et al., 2015;Sedlacek et al., 2018;Zhou et al., 2017). It is theorized that both losses and gains in OA mass are likely happening concurrently in most plumes through condensation and evaporation (Bian et al., 2017;Hodshire et al., 2019aHodshire et al., , 2019bMay et al., 2015), with the 75 balance between the two determining whether net increases or decreases or no change in mass occurs during near-field aging. However, there is currently no reliable predictor of how smoke aerosol mass (normalized for dilution) may change for a given fire.
Evolution of total aerosol number, size, and composition is critical in improving quantitative understanding of how biomass burn smoke plumes impact climate. These impacts include smoke aerosols' abilities to both act as cloud 80 condensation nuclei (CCN) and to scatter/absorb solar radiation, each of which is determined by particle size and composition (Albrecht, 1989;Petters and Kreidenweis, 2007;Seinfeld and Pandis, 2006;Twomey, 1974;Wang et al., 2008).
Particles can increase or decrease in size as well as undergo compositional changes through condensation or evaporation of vapors. In contrast, coagulation always decreases total number concentrations and increases average particle diameter; plumes with higher concentrations will undergo more coagulation than those with lower concentrations (Sakamoto et al., 85 2016).
Being able to predict smoke aerosol mass, number, size, and composition accurately is an essential component in constraining the influence of fires on climate, air quality, and health. Fires in the western United States region are predicted to increase in size, intensity, and frequency (Dennison et al., 2014;Ford et al., 2018;Spracklen et al., 2009;Yue et al., 2013). In response, several large field campaigns have taken place in the last 7 years examining wildfires in this region 90 Garofalo et al. 2019). Here, we present smoke plume observations from the Biomass Burning Observation Project (BBOP) campaign of aerosol properties from five research flights sampling wildfires downwind in seven pseudo-Lagrangian sets of transects to investigate aging of OA mass and oxidation, and aerosol number and mean diameter. A range of initial plume OA mass concentrations were captured within these flights and sufficiently fast (1 second) measurements of aerosols and key vapors were taken. We segregate each transect into edge, core, or intermediate regions of 95 the plume and examine aerosol properties within the context of both the location within the plume (edge, core, or intermediate) and the initial OA mass loading of the given location, with the differences in aerosol loading serving as a proxy for differences in dilution rates, as the extent to which dilution has occurred prior to the first observation is not a measurable quantity. We create mathematical fits for predicting OA oxidation markers and mean particle diameter given initial plume mass and physical age (time) of the smoke. These fits may be used to evaluate other smoke datasets and assist 100 in building parameterizations for regional and global climate models to better-predict smoke aerosol climate and health impacts.

Methods
The BBOP field campaign occurred in 2013 and included a deployment of the United States Department of Energy Gulfstream 1 (G-1) research aircraft in the Pacific Northwest region of the United States  105 Sedlacek et al., 2018) from June 15 to September 13. We analyze five cloud-free BBOP research flights that had seven total sets of across-plume transects that followed the smoke plume downwind in a pseudo-Lagrangian manner (see Figs. S1-S6 for examples; Table S1) from approximately 15 minutes after emission to 2-4 hours downwind . The G-1 sampling setup is described in Sedlacek et al., 2018;Kleinman et al., 2020).
Number size distributions were obtained with a Fast-integrating Mobility Spectrometer (FIMS), providing particle 110 size distributions nominally from ~20-350 nm (Kulkarni and Wang, 2006;Olfert and Wang, 2009); data was available between 20-262 nm for the flights used in this study. A Soot Photometer Aerosol Mass Spectrometer (SP-AMS) provided organic and inorganic (sulfate, chlorine, nitrate, ammonium) aerosol masses, select fractional components (the fraction of the AMS OA spectra at a given mass-to-charge ratio) (Onasch et al., 2012), and elemental analysis (O/C and H/C) (Aiken et al., 2008;Canagaratna et al., 2015). We use the f60 and f44 fractional components (the mass concentrations of m/z 60 and 44 115 normalized by the total OA mass concentration) and O/C and H/C elemental ratios of OA as tracers of smoke and oxidative aging. Elevated f60 values are indicative of "levoglucosan-like" species (levoglucosan and other molecules that similarly fragment in the AMS) (Aiken et al., 2009;Cubison et al., 2011;Lee et al., 2010) and are shown to be tracers of smoke primary organic aerosol (POA) (Cubison et al., 2011). The f44 fractional component (arising from primarily CO2+ as well as some acid groups; ) is indicative of SOA arising from oxidative aging (Alfarra et al., 2004;Cappa and Jimenez, 2010; 120 Jimenez et al., 2009;Volkamer et al., 2006). Fractional components f60 and f44 have been shown to decrease and increase with photochemical aging, respectively, likely due to both evaporation and/or oxidation of semivolatile f60-containing species and addition of oxidized f44-containing species (Alfarra et al., 2004;Huffman et al., 2009). O/C tends to increase with oxidative aging (Decarlo et al., 2008) whereas H/C ranges from increasing to decreasing with oxidative aging, depending on the types of reactions occurring (Heald et al., 2010). Thus, tracking H/C with aging may provide clues upon the types of 125 reactions that may be occuring. A Single-Particle Soot Photometer (SP2; Droplet Measurement Technologies) was used to measure refractory black carbon (rBC) through laser-induced incandescence (Moteki and Kondo, 2010;Schwarz et al., 2006). An Off-Axis Integrated-Cavity Output Spectroscopy instrument (Los Gatos, Model 907) provided CO measurements.
An SPN1 radiometer (Badosa et al., 2014;Long et al., 2010) provided total shortwave irradiance. The supporting information includes more details on the instruments used.

130
To determine the contribution of species X from smoke, the background concentration of X is subtracted off and normalized by background-corrected CO (ΔCO), which is inert on timescales of near-field aging (Yokelson et al., 2009), to correct for dilution. Increases or decreases of ΔX/ΔCO with time indicate whether the total amount of X in the plume has increased or decreased since time of emission. We background correct the number size distribution, OA, O, H, C, and rBC data in this manner by determining an average regional background for each species by using the lowest 10% of the CO data for a given flight with a similar altitude, latitude, and longitude as the smoke plume (excluding data from flying to and from the fire). Elemental O, H, and C are calculated using the O/C and H/C and OA data from the SP-AMS, allowing us to calculate ΔO/ΔC and ΔH/ΔC. We also background-correct f60 and f44 (using the mass concentrations of m/z 60, m/z 44, and OA inside and outside of the plume), we but do not normalize by CO due to these values already being normalized by OA.
We only consider data to be in-plume if the absolute CO >= 150 ppbv, as comparisons of CO and the number concentration 140 show that in-plume data has CO >150 ppbv and out-of-plume (background) data has CO < 150 ppbv. This threshold appears to be capturing clear plume features while excluding background air (Figs. S7-S11); we perform sensitivity analyses of our results to our assumptions about background and in-plume values in Section 3.
From the FIMS, we examine the background-corrected, normalized number concentrations of particles with diameters between 40-262 nm, ΔN40-262 nm/ΔCO. ΔN40-262 nm/ΔCO allows us to exclude potential influence of fresh nucleation 145 upon the total number concentrations, as the bulk of observed newly formed particles observed fell below 40 nm (Figs. S7-S11). Smoke plumes contain particles with diameters larger than 262 nm (Janhäll et al., 2009), and so although we cannot provide total number concentrations, we can infer how the evolution of ΔN40-262 nm/ΔCO will impact number concentrations overall. We also obtain an estimate of how the mean diameter between 40-262 nm, ! """" , changes with aging through: Where Ni and Dp, i are the number concentration and geometric mean diameter within each FIMS size bin, respectively.
All of the data are provided at 1 Hz and all but the SP-AMS fractional component data are available on the DOE ARM web archive (https://www.arm.gov /research/campaigns/aaf2013bbop). As the plane traveled at ~100 m s -1 on average, 155 data were collected every 100 m across the plume. The instruments used here had a variety of time lags (all <10 seconds) relative to a TSI 3563 nephelometer used as reference. The FIMS also showed an additional lag in flushing smoky air with cleaner air when exiting the plume with maximum observed flushing timescales around 30 seconds, but generally less (Fig.   S12). To test if these lags impact our results, we perform an additional analysis where we only consider the first half of each in-plume transect, when concentrations are generally rising with time ( Figure S12-S13), and our main conclusions are 160 unaffected. We do not test the impacts of other timelags.
We use MODIS Terra and Aqua fire and thermal anomalies detection data to determine fire locations (Giglio et al., 2006(Giglio et al., , 2008 and estimate the fire center to be the approximate center of all clustered MODIS detection points for a given sampled fire (Figs. S1-S6). Depending upon the speed of the fire front, the true fire location and center at the time of sampling is likely different than the MODIS estimates. To estimate the physical age of the plume, we use the estimated fire 165 center as well as the FIMS number distribution to determine an approximate centerline of the plume as the smoke travels downwind (Figs. S1-S6) and use mean wind speed and this estimated centerline to get an estimated physical age for each https://doi.org/10.5194/acp-2020-300 Preprint. Discussion started: 6 April 2020 c Author(s) 2020. CC BY 4.0 License.
transect. We did not propagate uncertainty in fire location, wind speed, or centerline through to the physical age, which is a limitation of this study.
3 Results and discussion (as Heading 1)

170
As a case example, we examine the aging profiles of smoke from the Colockum fire during the first set of pseudo-Lagrangian transects on flight 730b (Table S1). Fig. 1 provides ΔOA/ΔCO, ΔrBC/ΔCO Δf60 , Δf44, ΔH/ΔC, ΔO/ΔC, ΔN40-262 nm/ΔCO, and ! """" as a function of the estimated physical age; Figs. S14-S18 provide this information for the other pseudo- cases. However, the lowest two ΔCO bins tend more towards the physical edges of the plume and the highest two tend more towards the physical center of the plume (Figs. S2-S6). We do not use the data from the lowest 5% of ΔCO to reduce uncertainty at the plume-background boundary. We do not know where the plane is vertically in the plume, which is a limitation as vertical location will also impact the amount of solar flux able to penetrate through the plume.  the fire, ΔOAinitial. We show the 5-15 (edge) and 90-100 (core) ΔCO percentile bins here; Fig. S19 shows the same information for all four ΔCO percentiles. We note that although some of the physical ages appear to be at ~0 hours, this is 200 from a limitation of our physical age estimation method (Sect. 2), as no flights captured data before ~15 minutes after emission . Also included in Fig. 2 are the Spearman rank-order correlation tests (hereafter Spearman tests) that show correlation coefficients (R) with initial plume OA mass, ΔOAinitial (RΔOA,initial), and physical age (Rage) against each variable (for the correlations with ΔOAinitial, all transects in a set are given the same ΔOAinitial value). Figs. S13, S19-S21 show the same details as Fig. 2 but provide sensitivity tests to potential FIMS measurement artifacts (Fig. S13)

220
Two more important features of Δf60 and Δf44 can be seen within Fig. 2: (1) Δf60 and Δf44 depend on ΔOAinitial (RΔOA,initial = 0.38 and -0.5, respectively), with more concentrated plumes having consistently higher Δf60 and lower Δf44. (2) Differences in Δf60 and Δf44 for the nearest-to-source measurements indicate that evaporation and/or chemistry likely occurred before the time of these first measurements (assuming that emitted Δf60 and Δf44 do not correlate with ΔOAinitial).
The amounts of evaporation and/or chemistry depend on ΔOAinitial, with higher rates of evaporation and chemistry occurring 225 for lower values of ΔOAinitial. This result is consistent with the hypothesis that aircraft observations are missing evaporation and chemistry prior to the first aircraft observation (Hodshire et al., 2019b). The differences in ΔOAinitial between plumes may be due to different emissions fluxes (e.g., due to different fuels or combustion phases), or plume widths, where larger/thicker plumes dilute more slowly than smaller/thinner plumes; these larger plumes have been predicted to have less evaporation and may undergo relatively less photooxidation (Bian et al., 2017;Hodshire et al., 2019aHodshire et al., , 2019b. (Garofalo et al., 2019) segregated smoke data from the WE-CAN field campaign by distance from the center of a given plume and showed that the edges of one of the fires studied have less f60 and more f44 (not background-corrected) than https://doi.org/10.5194/acp-2020-300 Preprint. Discussion started: 6 April 2020 c Author(s) 2020. CC BY 4.0 License.

230
the core of the plume. Similarly, we find that the 730b flight shows a very similar pattern in f60 and f44 (Figs. S24-S25) to that shown in (Garofalo et al., 2019) (their Fig. 6). The 821b and 809a flights also hint at elevated f44 and decreased f60 at the edges but the remaining plumes do not show a clear trend from edge to core in f60 and f44. This could be as CO

235
concentrations (and thus presumably other species) do not evenly increase from the edge to the core for many of the plume transects studied (Figs. S2-S6). We do not have UV measurements that allow us to calculate photolysis rates but the inplume shortwave measurements in the visible show a dimming in the fresh cores that has a similar pattern to f44 and the inverse of f60 (Fig. S26; the rapid oscillations in this figure could be indicative of sporadic cloud cover above the plumes).
We also plot core and edge ΔH/ΔC and ΔO/ΔC as a function of physical age (Fig. 2d-e). Similar to Δf44, ΔO/ΔC 240 increases with physical age and is well correlated to both physical age and ΔOAinitial (Rage = 0.61 and RΔOA,initial = -0.42).
Conversely, ΔH/ΔC tends to be fairly constant or slightly decreasing with physical age and is poorly correlated to physical age and ΔOAinitial. A Van Krevelen diagram of ΔH/ΔC versus ΔO/ΔC (Fig. S27) indicates that oxygenation reactions or a combination of oxygenation and hydration reactions are likely dominant (Heald et al., 2010); however, without further information, we cannot conclude which reactions are occurring.
Both physical age and ΔOAinitial appear to influence Δf60, Δf44, and ΔO/ΔC: oxidation reactions and evaporation from dilution occur with aging, and the extent of photochemistry and dilution should depend on plume thickness. Being able to predict biomass burning aerosol aging parameters can provide a framework for interstudy-comparisons and can aid in modeling efforts. We construct mathematical fits for predicting Δf60, Δf44, and ΔO/ΔC: where X is Δf60, Δf44, or ΔO/ΔC and a, b, and c are fit coefficients. The measured vs. fit data and values of a, b, and c are shown in Fig. 3a-c. The Pearson and Spearman coefficients of determination (R 2 p and R 2 s, respectively) are also summarized in Fig. 3 and indicate moderate goodness of fits (R 2 between 0.21-0.25 for Δf60, between 0.53-0.58 for Δf44, and between 255 0.41-0.58 for ΔO/ΔC). Although no models that we are aware of currently predict aerosol fractional components (e.g. f60 or f44), O/H and H/C are predicted by some models (e.g., (Cappa and Wilson, 2012) and these fit parameters may assist in biomass burning modeling.
Other functional fits were explored (Figs. S28-S29), with 260 ( ) = ( ()(*(+, ) + ( ℎ ) + Eq. 3 ( Fig. S28) and ΔNinitial in the place of ΔOAinitial in Eq. 2 (Fig. S29) providing similar fits for Δf60 and Δf44. Aged Δf60 and Δf44 show scatter, limiting the predictive skill of measurements available from BBOP. The scatter is likely to variability in emissions due to source fuel or combustion conditions, instrument noise and response under large concentration ranges encountered in these smoke plumes, inhomogeneous mixing within the plume, variability in background concentrations not captured by our background correction method, inaccurate characterizations of physical age due to variable wind speed, and deviations from a true Lagrangian flight path. There may be another variable not available to us from the BBOP data that aids this prediction, such as photolysis rates. We encourage these fits to be tested out with further data sets and modeling.

Aerosol size distribution properties: ΔN40-262 nm/ΔCO and !!! 270
The observations of the normalized number concentration between 40-262 nm, ΔN40-262 nm/ΔCO (Fig. 2f), show that plume edges and cores generally show decreases in ΔN40-262 nm/ΔCO with physical age, with Rage = -0.25. The plume edges and cores with the highest initial ΔOA generally have lower normalized number concentrations by the time of the first measurement, and the edges generally have higher initial normalized number concentrations than the cores, potentially due to differences in coagulation rates. A few dense cores have normalized number concentrations comparable or higher than the 275 thinner edges, leading to no correlation with ΔOAinitial. We note that variability in number emissions (due to e.g. burn conditions) adds noise not captured by the R values.
The mean particle size between 40-262 nm, ! """" (Eq. 1), is shown to increase with aging ( Fig. 2g) for all plumes (Rage = 0.48). Coagulation and SOA condensation will increase ! """" and OA evaporation will decrease ! """" . The plumes do not show significant changes in ΔOA/ΔCO (Fig. 2a), indicating that coagulation is likely responsible for the majority of 280 increases in ! """" . The functional fits for Δf60 and Δf44 (Eq. 2; where X is ! """" in this case) can also moderately predict ! """" (R 2 p and R 2 s of 0.36 and 0.31; Fig. 3d) but do not well-predict ΔN40-300 nm/ΔCO (not shown). Sakamoto et al. (2016) provide fit equations for modeled ! """" as a function of age, but they include a known initial ! """" at the time of emission in their parameterization, which is not available here. ΔNinitial in the place of ΔOAinitial in Eq. 2 predicts ! """" similarly ( Fig. S29). As discussed in Section 3.1, scatter in number concentrations limits our prediction skill.

285
Nucleation-mode particles (inferred in this study from particles appearing between 20-40 nm in the FIMS measurements) are observed for some of the transects (S7-S11). Some days also show nucleation-mode particles downwind of fires in between transects (Figs. S7, S8, S9, and S11). Nucleation-mode particles appear to primarily occur in the outer portion of plumes, although one day did show nucleation-mode particles within the core of the plume (Fig. S11). Nucleation at edges could be due to increased photooxidation from higher total irradiance relative to the core (Fig. S26). However, given 290 the relatively small number of data points showing nucleation mode particles and limited photooxidation and gas-phase information, we do not have confidence in the underlying source of the nucleation-mode particles. The nucleation mode tends to be ~one order of magnitude less concentrated than the larger particles, and appears to be coagulating or evaporating away as the plumes travel downwind.

295
The BBOP field campaign provided high time resolution (1 s) measurements of gas-and particle-phase smoke measurements downwind of western U.S. wildfires along pseudo-Lagrangian transects. These flights have allowed us to examine near-field (<4 hours) aging of smoke particles to provide analyses on how these species vary across a range of initial aerosol mass loadings (a proxy for the relative rates at which the plume is anticipated to dilute as dilution before the first observation is not a measurable quantity) as well as how they vary between the edges and cores of each plume. We find 300 that although ΔOA/ΔCO shows little variability, Δf60 (a marker for evaporation) decreases with physical aging; Δf44 and ΔO/ΔC (markers for photochemical aging) increases with physical aging; and each correlate with both ΔOAinitial and physical age. ΔN40-262 nm/ΔCO decreases with physical aging through coagulation, with thicker plumes tending to show lower number concentrations, indicative of higher rates of coagulation. Mean aerosol diameter between 40-262 nm increases with age primarily due to coagulation, as organic aerosol mass does not change significantly. Nucleation is observed within a few of 305 the fires and appears to occur primarily on the edges of the plumes. Differences in initial values of Δf60 , Δf44, and ΔO/ΔC between higher-and lower-concentrated plumes indicate that evaporation and/or chemistry has likely occurred before the time of initial measurement and that plumes or plume regions (such as the outer parts of the plume) with lower initial aerosol loading can undergo these changes more rapidly than thicker plumes. We encourage future studies to attempt to quantify these chemical and physical changes before the initial measurement using combinations of modeling and laboratory 310 measurements, where sampling is possible at the initial stages of the fire and smoke. We also encourage future near-field (<24 hours) analyses of recent and future biomass burning field campaigns to include differences in initial plume mass concentrations and location within the plume as considerations for understanding chemical and physical processes in plumes.