Two methods were used to determine BrC in this study, an off-line filter-based approach and in situ measurements from a photoacoustic spectrometer (PAS). Following a number of past studies (J. Liu et al., 2014, 2015; Zeng et al., 2020), the light absorption of soluble BrC species (an operationally defined parameter) was measured by a liquid-based spectrophotometric method on solvent extracts of particle-laden filters. The spectrometer (USB-4000, Ocean Optics, Dunedin, FL) was coupled with a long-path waveguide capillary cell (2.5 m optical path; LWCC-3250; World Precision Instruments, Sarasota, FL) and a broadband UV–Vis–NIR light source (DH-mini; Ocean Optics, Dunedin, FL). A similar approach was used on a newly developed online mist chamber water-soluble particle collection system deployed for the first time in this study (Zeng et al., 2021). Detailed operating procedures and data processing are described elsewhere (Zeng et al., 2021). Here we only focus on the filter data since we are interested in more than just water-soluble BrC species. Briefly, aerosol particles with an aerodynamic diameter less than 4.1 µm were collected onto Teflon filters at intervals typically less than 5 min for samples within smoke plumes and a maximum of ∼ 20 min for background air sampling. Filters were extracted at a later date sequentially, first with water, then after air drying, with methanol. In each case, the resulting liquid extract was filtered (0.22 µm pore size syringe filter) and then injected into the LWCC via a syringe pump. The absorption spectra over wavelengths from 300 to 700 nm at ∼ 1 nm resolution were recorded relative to those of the pure solvent, resulting in the light-absorption spectra of water-soluble (WS) and methanol-soluble (MS) chromophores of species in the ambient aerosol particles. Light absorption measured in the waveguide (Aλ) is converted to a light-absorption coefficient (AbsλLWCC) using 1AbsλLWCC=AλVlVa⋅lln⁡10, where Va is the volume of air that passed through the filter, Vl is the volume of solvent used in the extraction and l is the LWCC optical path length (nominally 2.5 m). The total soluble (TS) light-absorption measurement is defined here as the sum of the light-absorption coefficients at each wavelength measured from the WS and MS extracts (AbsTS,λLWCC=AbsWS,λLWCC+AbsMS,λLWCC), as sequential extraction has been shown to be comparable to methanol extraction alone (J. Liu et al., 2015). (Note that water then methanol measurements of BrC were done to provide both WS and total BrC data, since some instruments are only capable of WS BrC measurements, e.g., the online systems in Zeng at al., 2021.) As will be discussed, this extraction does not necessarily account for all of the BrC, since some chromophores may not be soluble in these solvents. As in past studies, for simplicity, overall BrC levels are often characterized by light absorption at one wavelength, such as for the solvent method at 365 nm for both WS BrC and TS BrC (AbsWS,365nmLWCC and AbsTS,365nmLWCC). Light absorption data over the whole spectrum are available from the NASA data archive (FIREX-AQ 2019; https://www-air.larc.nasa.gov/missions/firex-aq/). For this method, the limits of detection (LODs) were determined by 3 standard deviations of the blank measurements and are 0.10 and 0.26 Mm-1 for WS and MS BrC, respectively, at 365 nm. At other wavelengths from 300 to 700 nm, the LODs range between 0.08–0.52 Mm-1 for both AbsWS,λLWCC and AbsMS,λLWCC. The uncertainties associated with the absorption measurements were calculated by propagating the uncertainties from sampling (air flow rates and sampling time), filter extraction and the absorption measurement and are estimated to be 16 % for WS BrC, 19 % for MS BrC and 25 % for TS BrC at 365 nm. The uncertainties are larger towards higher wavelength because the measured absorption is closer to the blank measurement; the uncertainties are ∼ 60 % for AbsWS,700nmLWCC and AbsMS,700nmLWCC, and therefore the uncertainty for AbsTS,700nmLWCC can exceed 85 %. There are several advantages of the BrC measurement based on aerosol extracts: (1) the majority of insoluble absorbers, for example, BC or mineral dust particles, were filtered out (Zeng et al., 2021), making it a direct measurement of BrC. (2) The light absorption can be measured over a broad wavelength spanning from UV to visible range at high spectral resolution (300–700 nm, at ∼ 1 nm resolution). However, aerosol particle size and morphology information is lost, so light absorption (AbsλLWCC) measured in the LWCC is not directly comparable to results from aerosol optical instruments since it does not consider particle size and other related effects. In the following, we denote the absorption coefficients for just chromophores in liquids by AbsλLWCC and the estimated coefficients for the chromophores in aerosol particles by bap,BrC,λ. A further limitation is that a fraction of non-polar chromophores may not be extracted efficiently in water or organic solvents (Corbin et al., 2019; Shetty et al., 2019). Table 2 defines many of the variables used.

A photoacoustic spectrometer (PAS) was deployed on the DC-8, providing real-time measurements of dry aerosol absorption of fine particles (diameters < 2.5 µm) at three wavelengths: 405, 532 and 664 nm (Lack et al., 2012; Langridge et al., 2013). In this instrument, the light at a specific wavelength absorbed by an aerosol particle is converted to an acoustic pressure wave that is detected with a microphone. Uncertainties for these data are estimated to be 20 %, mainly from calibration, pressure variation and optical alignment issues (Langridge et al., 2011). Note that the PAS does not directly measure BrC absorption; it must be calculated as the difference between total aerosol light absorption and light absorption by BC (here BC is referred to as the overall light-absorption properties of BC, which include the absorption by BC and any absorption enhancement due to coatings). Since BC, in most cases, dominates the light absorption at all wavelengths, BrC absorption inferred by optical instruments like the PAS can have a large uncertainty due to BrC being calculated by the subtraction of two similar magnitude numbers. In smoldering smoke plumes, this is less of an issue due to the high levels of BrC relative to BC. As the PAS only provided data at a limited number of wavelengths, mostly in the visible wavelength range, extrapolation from visible to UV wavelengths is required, which also leads to further uncertainties when using these data to infer optical properties over a broad spectral range (S. Liu et al., 2015).

Refractory black carbon (rBC) mass concentration was measured by a single-particle soot photometer (SP2), which quantified rBC mass in individual particles in the 0.090 to 0.550 µm size range (volumetric-equivalent diameter assuming 1.8 g cm-3 void-free density) based on the incandescence signal they generated when passing through a laser beam (Stephens et al., 2003; Schwarz et al., 2006). The amplitude of the BC incandescence signal is related to the amount of refractory material contained in the illuminated particle. Accumulation-mode rBC concentrations used for calculation of BC-specific absorption were extrapolated from the SP2 observations to account for rBC mass outside the detection range of that instrument (Schwarz et al., 2008). Log-normal fits to the size distributions of rBC observed by the SP2 were generated on a per-flight basis to allow for estimation of the undetected mass in that mode. Corrections were applied evenly to all the rBC concentrations of that flight. Note that the size distributions were strongly dominated by the fire-generated smoke and thus do not reflect background or urban rBC detected in transit, etc. Figure S1 in the Supplement shows a typical rBC size distribution from a flight sampling the Williams Flats fire (3 August 2019). This generates a corrective scaling factor of 1.12. The average correction factor for all western wild fires in FIREX-AQ was 1.08, with a standard deviation of 0.04 (13 flights). Use of a dilution system to reduce particle loads to single-particle instruments on the DC-8 increased total SP2 uncertainty to larger values than typical. Total uncertainty was estimated to be 40 % in integrated rBC mass mixing ratios in its size range with the dilution system.

OA mass concentration was measured for particles nominally up to 1 µm diameter by a high-resolution time-of-flight Aerodyne aerosol mass spectrometer (HR-ToF-AMS) (Decarlo et al., 2006, 2008). The uncertainty of the OA mass concentration was estimated to be 38 % (2σ), dominated by uncertainty in particle collection efficiency due to particle bounce and absolute and relative ionization efficiency (Bahreini et al., 2009). A extractive electrospray ionization time-of-flight mass spectrometer (EESI-ToF-MS) (Lopez-Hilfiker et al., 2019; Pagonis et al., 2021) measured the mass concentration of 4-nitrocatechol (4-NC). The instrument was limited to pressure altitudes below 7 km above sea level, and the measurement uncertainty was estimated to be 47 % (2σ). HR-ToF-AMS and EESI-ToF-MS shared a National Center for Atmospheric Research (NCAR) High-Performance Instrumented Airborne Platform for Environmental Research Modular Inlet (HIMIL) (Stith et al., 2009), together with a high-efficiency particulate air (HEPA) filter for background measurements and a calibration system (Pagonis et al., 2021).

Aerosol number size distributions were measured by a laser aerosol spectrometer (LAS, model 3340, TSI Incorporated, Shoreview, MN), operated behind a monotube Nafion dryer (Moore et al., 2021). The reported size range is from ∼ 100 nm to 4 µm (Brock et al., 2019). LAS was size-calibrated with mobility-classified ammonium sulfate particles. The LAS is known to undersize particles with real refractive indices less than ammonium sulfate (m=1.52+0i) while oversizing particles with larger real refractive indices or absorbing particles (Moore et al., 2021). The LAS uncertainty is estimated to be 20 % to account for variability in smoke aerosol refractive index and the LAS particle sizing calibrations.

Carbon monoxide (CO) mixing ratios were measured by a diode laser spectrometer system, referred to by its historical name Differential Absorption Carbon Monoxide Measurements (DACOM), with a measurement uncertainty of 2 ppbv (Warner et al., 2010; Sachse et al., 1991). O3 and NOx were measured with the NOAA Nitrogen Oxides and Ozone (NOyO3) four-channel chemiluminescence instrument, with a measurement uncertainty of 5–10 pptv ±3 % and 9 %, respectively (Pollack et al., 2010; Bourgeois et al., 2022). All the data presented here are at standard temperature and pressure (273 K and 1013 mbar). High-resolution 1 Hz data (e.g., measurements by PAS, BC, CO, O3, NOx, OA and 4-NC) were merged to low-time-resolution data (i.e., 10 s data or filter sampling interval) depending on the analysis performed.

For all smoke samples collected during FIREX-AQ, WS BrC (AbsWS,365nmLWCC) accounted for 45 % ± 16 % (mean ± standard deviation) of TS BrC at 365 nm (AbsTS,365nmLWCC) (Fig. S2, TS vs WS BrC at 365 slope = 2.23, intercept = 0, R2= 0.91), which is similar to levels (∼ 45 %) observed in fresh biomass burning plumes during the DC3 campaign that also investigated summertime western United States wildfires (J. Liu et al., 2015). This fraction is slightly lower than the observation in highly aged biomass burning in the ATom study (smoke transported from the continents to remote marine regions), which was 53 % ± 17 %, and could be explained by the ATom BrC being more oxidized (more aged) with a higher hygroscopicity (Duplissy et al., 2011).

The spectral characteristics of BrC are often characterized by the absorption Ångström exponent (AAE). To cover as many of the short wavelengths as possible, we calculated the AAE from the light absorption measured between 300 to 500 nm. (Note that calculating the AAE by fitting data with a power law is dependent on the wavelength range utilized; Moosmüller et al., 2011.) Including all measurements in the identified smoke plumes of this study, the AAE for TS BrC (sum of the water and methanol extracts) was on average 4.2 ± 1.6 (mean ± SD). This is lower than the AAE for just WS BrC (AAE = 5.1 ± 1.3) and likely a result from less-polar chromophores that are extracted in methanol but not water, absorbing more light at the higher wavelength range (J. Liu et al., 2015; Zhang et al., 2013). The AAE for soluble BrC (either WS BrC or TS BrC) measured in FIREX-AQ is comparable to the result from the ATom study (Zeng et al., 2020) but lower than for the DC3 (J. Liu et al., 2015). The AAE of the overall absorbing aerosol (that is, rBC plus lensing plus BrC) determined from fitting the PAS-measured absorption at the three wavelengths with a power law for these same smoke plumes was 1.49 ± 0.52 (mean ± SD), while the BrC determined from the PAS had an AAE of 2.07 ± 1.01 (mean ± SD).

The emission ratio of BrC (ERBrC) can be estimated from the various fires as the ratio of ΔBrC to ΔCO (i.e., NEMRBrC), assuming there was little atmospheric processing between the emission and time of measurement. To estimate emissions, we only use BrC data from wildfires with transport time less than 2 h. The results are shown in Fig. 2. For PAS-predicted BrC at a wavelength of 405 nm (the lowest PAS measurement wavelength), the ERBrC (slope) was 0.131 ± 0.001 Mm-1 ppbv-1 (the ± is the fit uncertainty in the slope at a 1σ confidence interval). For just water-soluble BrC, the ERBrC was 0.071 ± 0.003 Mm-1 ppbv-1 and for TS BrC 0.163 ± 0.006 Mm-1 ppbv-1. We calculated the soluble BrC ERs at 365 nm since this is the wavelength most often used to characterize BrC using a single wavelength. These data are BrC in the solution and not corrected for conversion to ambient particle (factor Kλ, discussed below, is not applied here) since many studies report BrC measured in solution. (Note that the TS BrC is higher than PAS BrC because it is given at a lower wavelength; see Fig. S3.)

The wildfire BrC optical properties can be mapped onto the classification proposed by Saleh (2020) to provide a rough characterization and test the approach as a parameterization. For the PAS and filter data, we determined the BrC AAE and the mass absorption cross-section (MAC). For the PAS, the MACPASBrC,532nm was determined from the ratio of bap,PASBrC,532nm to OA mass measured by the AMS, and AAEPASBrC was calculated from the power law fit to bap,PASBrC,405nm, bap,PASBrC,532nm and bap,PASBrC,664nm. PAS data are shown with the Saleh BrC characteristics identified by regions in the boxes. In Fig. 3a, these wildfires data are best characterized as M-BrC and show a weak trend with the rBC / OA ratio as a higher rBC / OA ratio (more flaming) tends to appear at the bottom left. Our smoke data also show little correlation between BrC AAEPASBrC and log10(MACPASBrC,405nm). Figure 3b shows a similar plot for the TS BrC (without applying the conversion factor Kλ to convert to aerosol absorption). Although many of these data are outside of Saleh's categorization, they are shifted to the upper left relative to the PAS data, consistent with the idea that PAS BrC contains relatively more weakly absorbing species (Atwi et al., 2022). Like the PAS BrC, TS BrC also does not show a correlation between AAE and log10(MAC) but possibly a weak trend with rBC / OA.

Starting from a wide perspective, we compare the downwind evolution of TS BrC (AbsTSBrC,365nmLWCC) of the FIREX-AQ data to the larger-scale evolution of smoke (RIM fire) reported by Forrister et al. (2015) from a previous study. The same filter-based measurement and analytical methods were used in both cases to determine TS BrC. Analysis of the uniquely large RIM fire that was studied on 2 separate days as it advected from California into Manitoba, Canada, showed a consistent decrease in the NEMRTSBrC (Fig. 8), and an observed half-life of TS BrC of 9 to 15 h was estimated. We have added the various measurements from this study to the RIM fire data.

First we note the more aged (10 to about 25 h) FIREX smoke data in Fig. 8, which were identified as smoke from the Tucker and two Williams Flats fires. The Tucker fire tends to follow the steady decay of the RIM fire, but most of the aged smoke measured in the 08/07 and 08/08 Williams Flats fires has significantly higher NEMRTSBrC at ages between 15 and 25 h (see Fig. S6). For these data, a trend is less clear, although for the Williams Flats 08/07 data, there was a decrease in NEMRTSBrC for the aged relative to the fresh smoke. But for the 08/08 data there seems to be no change with age, although the NEMRTSBrC are highly scattered for both the fresh and aged. Since these fires were not tracked continuously (there is a significant measurement gap of over 10 h in the plume evolution), consistency in the evolution of the these smoke plumes cannot be assessed, so causes for the NEMRTSBrC variability are uncertain. For example, fresher smoke from other fires could have contributed to the Williams Flats 08/08 plume when measured far downwind and would explain the higher NEMRTSBrC to what is expected for a plume of that age. However, in general (with 08/08 the exception), NEMRTSBrC for aged plumes was lower than fresh plumes, suggesting an overall decay of BrC on timescales greater than about 8 h.

Focusing on the higher density of measurements made closer to the fires where the transport ages were less than approximately 8 h, Fig. 8 shows that the RIM BrC data are within the range of the FIREX data, but the FIREX NEMRTSBrC data are highly variable with no clear trend with increasing plume age. Looking at each of the fresh plumes investigated in this study, Fig. 9 shows that a range of behaviors is seen. No consistent pattern of production nor depletion of BrC is observed in the FIREX-AQ data. In some plumes the data are highly scattered; in others there appears to be a consistent downward or upward trend, or no change in the NEMRTSBrC with increasing time. These trends are similar when using PAS BrC data averaged to the filter sampling times. A similar lack of consistent trends in WS BrC was observed from the Twin Otter measurements as part of FIREX-AQ (Washenfelder et al., 2022) and the WE-CAN study (Amy Sullivan, personal communication, 2020). Waxing and waning of the fire emissions, which were confirmed with the geostationary satellite fire radiative power (FRP) measurements during the period of the DC-8 sampling, and changes in source emission strength (and perhaps aerosol composition) during this time may also impact the downwind variability of the smoke plume (in addition to the dilution, photochemistry and semi-volatile partitioning processes) (Wiggins et al., 2020; Hodshire et al., 2019). The limitations of semi-Lagrangian sampling adds complexity, but the results suggest highly complex and variable BrC evolution. To investigate the changes in NEMRTSBrC, we look at the evolution of a specific BrC chromophore and study the variability of species along cross-plume transects.

One approach to evaluate the evolution of bulk BrC is to compare NEMRBrC as a function of plume age to a specific BrC species with known properties. 4-Nitrocatechol (4-NC) has been observed to be abundant in a variety of BrC sources, including primary emission from biomass burning (Lin et al., 2016) and in the secondary aerosols generated from aromatic precursors (Lin et al., 2015b; Vidović et al., 2020), and its evolution has been studied in detail (Zhao et al., 2015). For FIREX, 4-NC mass concentration was measured on 17 of the 23 flights. As it is one of the components of WS BrC, we compare 4-NC to WS BrC measured in the filter extract. We also compare it to PAS BrC in the aerosol particle phase; both are shown in Fig. 10. For all available data within smoke plumes, the ratio of the 4-NC absorption coefficient to the WS BrC absorption, both at the wavelength of 365 nm, was 18 % ± 16 % (mean ± SD), with a lower, middle and lower quartile range of 10 %, 19 % and 33 %. Figure 10a shows the statistics of this ratio for data grouped by estimated plume age. In the first 3 h, the following emission of 4-NC contributed about 23 % (median) to the WS BrC light absorption at 365 nm, although there was significant variability. For smoke plumes in the 3 to 6 h age range, the fraction of 4-NC to WS BrC decreased, with a median of about 18 %, and for plumes with transport ages greater than 18 h, 4-NC was essentially all lost; it contributed only about 0.2 % to the light absorption of WS BrC. The absorption ratio of 4-NC and all BrC (PAS BrC) at a wavelength of 405 nm shows a similar decay in the contribution of 4-NC with smoke age. This indicates that the bulk BrC in these smoke plumes had a lifetime that was significantly longer than the 4-NC. As one of the smaller (in terms of molecular weight) BrC chromophores, 4-NC has been found to have a short lifetime of ∼ 4 min from aqueous OH (3.2 × 10-14 M) oxidation after the photo-enhancement stage (Zhao et al., 2015).

Unlike BC, which is refractory, OA has a wide range of volatility (Huffman et al., 2009). Some chromophores that contribute to the overall BrC may also be semi-volatile and evaporate when the temperature increases or when the plume dilutes; however, this behavior cannot be inferred from the OA evolution since BrC is only a small mass fraction of OA. For the Sheridan fire on 15 August 2019, a sampling transect was made along the direction in which the plume was advecting away from the fire. In this particular case, the plume descended as it moved away, resulting in a ∼ 15 K temperature increase from the higher to lower altitude, providing an opportunity to investigate the evolution of BrC in terms of temperature-driven evaporation.

Variation in temperature and the NEMRs (to account for dilution) of various species along the plume as it descended is shown in Fig. 11. The NEMRrBC (black line) was fairly constant along the plume transect, implying that the combustion conditions (flaming vs smoldering) were relatively stable over time (i.e., since different downwind distances are related to different times of fire emissions). Thus, we assume that any observed changes with age were mainly from temperature-driven processes as the contribution of chemical aging should be negligible during this time period. Using the higher-frequency PAS BrC data, NEMRPASBrC (405 nm) essentially did not change along the plume; the coefficient of variation of NEMRPASBrC was less than 1 %. As noted above, 4-NC is less stable than bulk WS BrC, and in this plume it also displayed possible evidence of volatility-driven loss as it advected downwind, here, at an average rate of 5 % K-1. For just these data, the 4-NC contributes ∼ 10 % of the total particle BrC absorption closest to the fire and the fraction decreases to ∼ 3 % when temperature increases from 287 to 303 K. To maintain a constant NEMRBrC despite a decrease of 4-NC, BrC production may have compensated for evaporation loss, such as gas-phase 4-NC and other volatile chromophores to less-volatile species that partitioned back to the aerosol phase (Roman et al., 2022). Alternatively, a temperature effect on the particle chemistry resulted in an unchanged BrC absorption. Firm conclusions are not possible due to the high variability in 4-NC relative to loss trend for this small time period.

In most cases, the DC-8 flew into plumes approximately perpendicular to the direction of smoke transport, generating transverse plume transects, as shown in Fig. 1. We use these transects with the PAS BrC data to investigate variables that contribute to BrC variability. The filter BrC data could not be used in this analysis since one filter was collected for each transverse transect. Multiple processes, including evaporation and chemical reactions, that can be occurring simultaneously may be easier to resolve in a transverse transect analysis. In an idealized transect of a smoke plume, the aircraft would enter the plume from background air, then experience a positive concentration gradient from edge to plume center then a negative gradient from center to the other edge and finally exit the plume into background air. Burning is typically not an ideal point source that produces a plume which fans out as it advected away but often occurs in a region or along a line. Smoke generated at different rates along the whole burning area would then contribute to the concentrations of smoke species measured along the transect. If burning conditions or material burned varied in the region, this would complicate the analysis. To minimize this effect on aerosol properties, we focus on the analysis of three contrasting plume transects where in all cases the NEMRrBC was relatively constant (coefficient of variation of NEMRrBC < 10 %), suggesting minimal variation in overall particle emissions along the transect. The three plumes investigated are shown in Fig. 12.

Figure 12a shows one transect of the Williams Flats plume on 7 August 2019. The CO data suggest smoke from three major burning regions had merged into a single plume; three peaks in CO were observed, and these plumes had merged since background CO concentrations were not reached in the regions between the plumes. Figure 12d is one transect of the Castle plume on 12 August 2019, and Fig. 12g is one transect of the Williams Flats plume on 3 August 2019. In both of these cases, smoke from two intense burning regions had merged to some extent, based on the CO data. Note the differences in CO concentrations indicating the contrasting levels of emissions from these fires. In most of these cases, the BrC (PAS BrC absorption at 405 nm) profile along the transect had the same shape as CO, suggesting that BrC and CO had the same source and experienced a similar dilution process, but there were differences. Figure 12b, e, and h show NEMRrBC and NEMRPASBrC, which removes the effect of plume dilution. NEMRrBC is relatively constant, suggesting that rBC emissions for these fires did not significantly change (e.g., flaming vs smoldering). For BrC, if the NEMRPASBrC behaved as NEMRrBC, it would suggest little net effect of any atmospheric processes, other than a simple dilution effect on BrC concentration, or that during the dilution process, production balanced loss, but the NEMRPASBrC did vary to different extents in these three cases, and the variation was correlated with O3.

In the 7 August Williams Flats (Fig. 12a) and Castle fire transects (Fig. 12d), NEMRPASBrC and O3 concentration had a good positive correlation (Fig. 12c and f), suggesting O3 oxidation or related process (e.g., secondary processes) could possibly be linked to the observed BrC enhancement, indicated by the increasing NEMRPASBrC. In the Williams Flats 7 August transect, the O3 concentration was lowest in the center of the plumes (45, 12 and 29 ppbv, respectively, for the three CO peaks) and higher in the regions where the plumes mixed and CO was lower (O3 > 60 ppbv). Lower O3 in the plume centers was likely due to O3 titration by NOx with NO2 photolysis too slow to regenerate O3, consistent with the anticorrelation between O3 and NOx clearly seen in this transect (Fig. 12a). Higher O3 production in the diluted edges of the plume is discussed in detail by Xu et al. (2021), Wang et al. (2021) and Decker et al. (2021). NEMRPASBrC also tended to be higher in the edge regions between the plumes where O3 was higher, leading to a positive correlation between NEMRPASBrC and O3 (Fig. 12c). Concentrations of various species in this plume (08/07 Williams Flats; Fig. 12a, b, c) were much higher than the other two fires shown in Fig. 12.

For the Castle fire transect (Fig. 12d, e, f), smoke levels were much lower (much lower CO), and O3 may not have been significantly titrated by NOx (note low NOx levels). O3 was about ∼ 60 ppbv across the plume, but in this case, there was a slight enhancement in the center of the plume (Fig. 12d), along with NEMRPASBrC (Fig. 12e), again leading to a positive correlation with O3 (see Fig. 12f). For the Williams Flats fire on 3 August 2019 (Fig. 12g, h, i), which was more intense than the Castle fire but less than 08/07 Williams Flats (compare CO), O3 was higher in the center of the plumes along with NOx, but NEMRPASBrC was lower, being higher at the edges, leading to a negative relationship with NEMRPASBrC. A positive relationship between O3 and NEMRPASBrC may be linked to BrC photo-enhancement (O3↑→ NEMRBrC↑, or O3↓→ NEMRBrC↓), while an inverse relationship is indicative of photobleaching of BrC (O3↑→ NEMRBrC↓, or O3↓→ NEMRBrC ↑). These two possible divergent behaviors when BrC is oxidized by O3 have been observed in other studies. Sareen et al. (2013) observed this behavior for secondary BrC (formed with methylglyoxal and ammonium) and Fan et al. (2020) for aerosols from biomass burning. O3 could also just be acting as a tracer for other oxidation processes.

To look for evidence of these trends in all the data, for each plume transect the relationship between O3 and NEMRPASBrC was determined and then grouped as either a positive or negative relationship between NEMRPASBrC and O3. The transect-average NOx concentration was then compared for these two groups; results are shown in the box plot in Fig. 13. When NEMRPASBrC had a positive relationship with O3, consistent with O3 enhancing BrC absorption by generating additional BrC chromophores or transforming BrC to more strongly absorbing compounds, high NOx was more likely to be present. When NEMRPASBrC had a negative relationship with O3, suggesting O3 contributed to bleaching of BrC, NOx concentrations were generally lower. These observations are consistent with some previous studies. P. F. Liu et al. (2015) found that the presence of NOx was associated with the production of organonitrogen compounds via O3 oxidation, such as nitro-aromatics and organonitrates, which enhanced light absorption. However, other studies show fragmentation of chromophores on exposure to O3 in a NOx-free environment led to a decrease in BrC absorption (Pillar-Little and Guzman, 2017; Sun et al., 2019). Additionally, the reaction of NO2 with O3 produces the NO3 radical, which has been identified to be an important factor in BrC formation at night (Cheng et al., 2020; Mayorga et al., 2021; Selimovic et al., 2020). In dark optically thick smoke plumes where the nitrate radical loss by photolysis may be suppressed, high O3 and NOx could also be linked to increases in BrC (Cheng et al., 2020; Mayorga et al., 2021; Selimovic et al., 2020). The correlation is not perfect, since as seen in Fig. 13 (left box-and-whisker plot), there were many periods when BrC increased with increasing O3, and yet NOx levels were very low. This may be a limitation with our analysis, but there is some evidence that BrC can be formed without NOx through heterogeneous reactions of ozone with combustion particles (i.e., soot) (Kuang and Shang, 2020). Overall, the range of possible results demonstrate the complexity of processes that may affect BrC in fairly fresh wildfire smoke.

Dilution-driven evaporation resulting in BrC loss has been reported to be an important process in the WE-CAN airborne study, which investigated similar western United States wildfires in the summer before FIREX-AQ (Palm et al., 2020). Our analysis comparing the evolution of WS BrC to 4-NC and the change in BrC with changing plume temperature, however, suggested that the dominant fraction of BrC was not volatile. Also, if dilution had a large effect in the three plumes above, it is likely that it would have been difficult to discern any trends between NEMRPASBrC, O3 and NOx, which implies that dilution played a minor role compared to the effects of O3 on BrC. For example, along these transverse transects, air masses experienced different degrees of dilution; air masses at the edge of the plume, or where two plumes had merged, are more diluted with background air than those at the center of the plumes. In the transect from the Williams Flats fire on 7 August 2019 (Fig. 12a), the highest CO mixing ratio was ∼ 5600 ppbv, and the lowest was ∼ 1500 ppbv near where two plumes had intersected but still sampling in smoke (i.e., CO still significantly above background levels). This corresponds to a dilution ratio (the ratio of highest CO enhancement to the CO enhancement at a location of interest) of about 4. If only considering the effect of dilution-driven evaporation, the NEMRPASBrC profile would be similar to the CO profile (the CO change indicates degree of dilution between two regions). But the profile of NEMRPASBrC was opposite of this, which means other processes, possibly O3 oxidation in this case, drove the change in BrC absorption. This opposite pattern between CO and NEMRPASBrC also occurred in the transects of the Williams Flats fire on 3 August 2019 (Fig. 12g, h, i), but in this case a possible reason for the observed NEMRBrC shape was due to bleaching or oxidation of BrC by O3. In the Castle fire transect (Fig. 12d, e, f), the dilution effect was superimposed on the enhancement by O3 oxidation. From the analysis above, we conclude that the effect of O3 oxidation, or a process linked to O3 production, was stronger than dilution.

As noted in the Introduction, there are a host of factors that can affect BrC levels in an evolving smoke plume. We examined the FIREX-AQ dataset for other potential factors that might alter the optical properties of BrC, including relative humidity (RH) and aerosol liquid water content (LWC), for evidence of heterogeneous reactions, OH exposure (product of OH concentration and time), NH4 associated with ammonium- or amine-containing BrC production, optical thickness of the plumes (jNO2 values) and type of material burning, but no evidence was found for a consistent relationship with BrC evolution. Direct photolysis may also change the optical properties of BrC; however, the wildfire flight transects were made in the late afternoon or in the evening. Additionally, jNO2 in the center of the plumes was typically less than 5 % of the jNO2 level outside of the plume, so direct photolysis may not be a significant factor causing BrC bleaching within these plumes. Late afternoon measurements and dense optically thick smoke plumes could also depress OH oxidation, except in the upper levels and sides of the plume where photochemical OH production would be more likely (Wang et al., 2021). Wildfire smoke generated at different times of day may evolve differently due to the type of oxidants involved and extent of photochemical bleaching in the first few hours (i.e., emissions late in the day or at night versus emissions in the morning or early afternoon) The DC-8 rarely continuously flew at the top or edges of plumes, limiting investigation of the effect of OH on BrC aerosol in these more dilute regions. It is also possible that multiple simultaneous processes limited our ability to resolve individual ones. A positive matrix factorization (PMF) analysis did not show any consistent factors, which could be either due to lack of clear processes or that many were highly non-linear and not captured by the PMF analysis.