Light Absorption by Brown Carbon over the South-East Atlantic Ocean

Biomass burning emissions often contain brown carbon (BrC), which represents a large family of lightabsorbing organics that are chemically complex and therefore difficult to estimate their absorption of incoming 20 solar radiation, resulting in large uncertainties in the estimation of the global direct radiative effect of aerosols. Here we investigate the contribution of BrC to the total light absorption of biomass burning aerosols over the South-East Atlantic Ocean with different optical models utilizing a suite of airborne measurements from the ORACLES 2018 campaign by introducing an effective refractive index of black carbon (BC), meBC=neBC+ikeBC, that accounts for all possible absorbing components at 660 nm wavelength to facilitate the attribution of absorption 25 at shorter wavelengths. Most values of the imaginary part of the refractive index, keBC, were larger than those commonly used for BC from biomass burning emissions, suggesting contributions from absorbers beyond BC at 660 nm. The TEM-EDX single particle analysis further suggests that these long-wavelength absorbers might include iron oxides, as iron is found to be present only when large values of keBC are derived. Using this effective BC refractive index, we 30 find that the contribution of BrC to the total absorption at 470 nm (RBrC,470) ranges from ~5-15 %, with the organic aerosol mass absorption coefficient (MACOA,470) at this wavelength ranging from 0.25±0.34 m2 g-1 to 0.43±0.12 m2 g-1. The core-shell model yielded much higher estimates of MACOA,470 and RBrC,470 than https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c © Author(s) 2022. CC BY 4.0 License.

wavelengths. Another methodology involves measuring the absorption of organics that have been extracted with either an organic solvent or water and represents, thus far, the only way to directly measure BrC absorption (Wong et al., 2019). The drawback to this approach is that not all organics can be extracted with one or two solvents, as highlighted by the work of Chen and Bond (2010), who reported extraction efficiencies of ~70 % in water and ~90 % in methanol. This inability to extract all organics means that insoluble organic substances remain unknown 5 since they are not measured. This, in turn, could lead to cases where the absorption properties of the extracted organics might be different from those derived from in situ measurements. In addition, this method is carried out offline and requires elaborate laboratory analysis. Still another approach to estimate the absorption of BrC is through optical closure, determine the BrC absorption as the difference of the total measured absorption and that of BC calculated using Mie theory (Saleh et al., 2014;Liu et al., 2015). In this approach, the accuracy of BrC 10 absorption relies heavily on the accuracy of BC absorption calculation. Values commonly used for the refractive index of BC (mBC) from BB emissions in these calculations have ranged from 1.5+0.3i to 1.95+0.79i (Liu et al., 2015;Chylek et al., 2019;Taylor et al., 2020), which will lead to large differences in BC absorption simulation results (Taylor et al., 2020). To date there is no consensus on the best value of mBC. Another factor influencing the BC absorption calculation, and hence the estimated BrC absorption, is the mixing state of BC and non-BC 15 components within particles. Liu et al. (2015) used the core-shell (CS) Mie model and Rayleigh-Debye-Gans approximation to investigate the effect of BC microphysics on the estimation of BrC absorption and found it to be highly sensitive to the model treatment. Saleh et al. (2014)

compared internal and external mixtures of BC and
BrC and found internally mixed cases yielded smaller BrC absorption than externally ones. Similar to the AAE attribution method, the presence of absorbing materials other than BC and BrC can lead to errors in the attribution 20 of BrC absorption. For example, light absorbing FeOx was found to be common in field studies, especially for BB emissions (Ito et al., 2018), yet few studies perform measurements of size distributions and chemical composition by these particle types, making it difficult to separate their contribution to the total absorption from that of BrC.
The savannah regions in Africa experience widespread annual BB events from July to October, which 25 are estimated to account to approximately 1/3 of global BB emissions (van der Werf et al., 2010). These aerosols are transported westward over the South-East Atlantic, making this region an ideal natural laboratory for investigating the absorption of BrC from BB emissions. In this study, we estimate the absorption of BrC using the optical closure method utilizing in-situ aircraft measurements and offline single particle analysis from the ORACLES (ObseRvations of Aerosols above CLouds and their intEractionS) 2018 campaign (Redemann et al., 30 2020). An effective refractive index of BC (meBC), which attempts to capture the absorption of all possible absorbing components at 660 nm, is introduced to facilitate the absorption attribution at shorter wavelengths. The core-shell model and homogeneous models are applied and compared in this study. The range of values of the https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License. organic aerosol mass absorption coefficient (MACOA,470) and contribution of BrC to the total absorption at 470 nm (RBrC,470) using the optical closure method are obtained. ORACLES was a three-year NASA-funded field campaign to investigate the influence of BB emissions from southern Africa on regional and global climate (Redemann et al., 2020). We investigated aerosol optical 10 properties during seven research flights (RF) from the ORACLES 2018 campaign (Fig. 1): RF05_1, RF05_2, RF05_3, RF6_1, RF6_2, RF10, and RF11. The specifications for each flight can be found in Table 1 in Dang et al. (2021). These flights were chosen as they provided investigations of aerosol properties from the marine boundary layer (MBL) and in the free troposphere, from relatively aged aerosols to highly aged aerosols, and are comprehensive in online and offline measurements with few missing data. RF05_1 occurred in mixed cloud and 15 aerosol layer at an altitude of ~1 km, RF05_2 was within the MBL at an altitude of ~380 m, the others were in the aerosol layer above the MBL at an altitude equal to or greater than 2 km. The aging time is taken as the first interception of back trajectories with fires; note this age is strictly minimum age as air can get repeated injections of smoke. The age was estimated to be ~155 h for RF05_1 and ~40 h for RF05_3, RF6_1, RF6_2, RF10, and RF11. The age of RF05_2 was uncertain as its 7-day back trajectories were within the MBL and did not reach the 20 https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License.

Site and Instrumentation
fires, but based on the result from the Weather Research and Aerosol Aware Microphysics (WRF-AAM) model, it was ~9 h older than RF05_1, and thus the most aged among all samples.
Particle properties measured were non-refractory submicron aerosol composition from an Aerosol Mass Spectrometer (AMS), mass concentrations of refractory BC and mixing state of BC-containing particles from a Single Particle Soot Photometer (SP2), particle number size distribution (PNSD) from an Ultra-High-Sensitivity  for the Neph, whose humidity is not controlled. Data for BC mass concentration less than 0.1 μg/m 3 or absorption coefficient at 660 nm less than 1.5 Mm -1 were not included. Measurements were averaged to 10 s and adjusted to STP values at 273.15 K and 1013 hPa. More details of the instrumentation are provided in Section S1 of the supplemental information.

Optical Models
From the TEM images, most of the particles were found to be nearly spherical, with >70% of the particles having aspect ratios <1.5; therefore, we applied Mie theory in our optical simulations. We investigated the sensitivity of our closure simulations to four different models -the ideal core-shell (CS) model and three homogeneous mixing models: 1) the volume mixing (VM) model, 2) the Maxwell-Garnett (MG) model, and 3) 20 the Bruggemann (BG) model. We did not investigate the externally mixing model, as our TEM analysis showed that most of the particles were internally mixed (Dang et al., 2021).
In our optical closure, particles were separated into BC-containing particles and BC-free particles, whose size distributions were calculated with the PNSD of all particles from the UHSAS and APS, and BC 2-D size and mixing state (i.e. coating thickness) distribution of BC-containing particles from the SP2. We applied the 25 aforementioned four models to BC-containing particles. Detailed descriptions and inputs of all four models can be found in Section S2 in the supplement. We assumed that the non-BC components in both populations were homogeneously well mixed and calculated the refractive index of the mixture, mBC-free, with the VM rule, which assumes that the index of refraction of the mixture is the sum of the volume-weighted indices of refraction of the individual components. The mass concentrations of SO ! "# , NO $ # , NH ! % , and Clmeasured by AMS were converted where n mol represents the number of moles. As potassium salts are the most frequently detected salts from TEM- where mi and Vi represent the refractive indices and volumes of aforementioned salts; mOA and VOA are those for OA. The refractive indices of salts are taken from Table 1 in Kuang et al. (2020). The quantity mBC-free is assumed to be the same value for both BC-free particles and the non-BC component in BC-containing particles. The effective refractive indices for the VM, MG, and BG models can then be obtained by the corresponding mixing rules (Section S2 in the supplement). The density of BC is assumed to be 1.8 g cm-3 (Bond and Bergstrom, 2006; 10 Liu et al., 2017). The refractive indices of BC and OA will be presented in the following section.

Optical calculation procedure
https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License.   We introduce an effective refractive index of BC, meBC, to represent the refractive index of BC and any other absorbing components at 660 nm in this study. The meBC is defined as meBC = neBC+ikeBC, where neBC and keBC are the real and imaginary parts of meBC, respectively. The meBC is proposed based on three considerations.

10
Firstly, the numerical value selected for mBC has a large impact on the absorption attributed to BC ( Fig. 3 and Fig.   S3), and hence on the estimated absorption by BrC, but as note above, there is no consensus on the best value of mBC yet. Secondly, the lack of measurements of possible absorbers other than BC and BrC such as magnetite makes it difficult to correctly estimate the BrC absorption, so approaches that take these components into consideration are required. Thirdly, aged BrC is usually considered to be non-absorbing at long visible 15 wavelengths (Chen and Bond, 2010;Lack et al., 2012), thus we assume meBC does not include the absorptivity of BrC and therefore the BrC absorption can be differentiated from absorption from other substances. This assumption has not considered tar balls, which are ubiquitous in fresh plumes and absorbing at infrared https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License. wavelengths; considering the aging time of our samples, the absorption of tar balls is expected to be small or negligible (Sedlacek III et al., 2018;Adler et al., 2019;Li et al., 2019).
The meBC is named the effective refractive index of BC as it is supposed to be contributed mainly or entirely by BC. The meBC would be the same as mBC if the absorption at 660 nm is solely contributed by BC; however, if other absorbing components are present, the imaginary part (keBC) of meBC would be greater than that 5 (kBC) of mBC, the amount of which depends on the absorptivity and relative amount of these absorbers. The retrieval of meBC is illustrated in Fig. 2; the real refractive index of OA (nOA) is retrieved as well because it has an impact, albeit small, on the BC absorption calculation ( Fig. 3 and Fig. S3). Note that OA is assumed to be non-absorbing with the imaginary refractive index kOA=0 in the whole calculation. We determined the values of meBC and mOA by minimizing the sums of the squares of the relative differences between the measured and calculated scattering 10 and absorption coefficients at 660 nm for each of the four models: is determined by interpolated using the scattering Ångström exponent from scattering coefficients measured by the nephelometer (Section S3 in the supplement). Only absorption is used in the calculation of c " for RF05_1 and RF05_2, as measured and simulated scattering coefficients are not comparable (discussed in detail in Section S1.2 in the supplement): ) for each flight using the different models. The values of mBC and mOA are usually considered to be very weakly dependent on, or independent of, wavelength in the visible spectrum (e.g., Chang and Charalampopoulos, 1990;Moteki et al., 2010;Saleh et al., 2014). As will be discussed later in Section 3.1, absorbers that contribute to meBC besides BC are 25 mainly regarded as magnetite, whose refractive index is generally invariant between wavelengths 470 and 660 nm. Thus, the assumption of constant meBC and mOA between 470 and 660 nm is reasonable in our study, although it may lead to underestimations of BrC absorption for highly aged particles, as discussed further in Section 3.2.
In addition, the absorption by BrC may be slightly underestimated without incorporating tar balls into the calculation. As illustrated in Fig. 2, the absorption at shorter wavelengths is assumed to be determined by that The quantity RBrC,λ, defined as 13(,@ = 8?7,13(,@ / 8?7,@ *48 is the fractional contribution of BrC to the total absorption at wavelength λ, which is assumed to be zero at 660 Similarly, the mass absorption coefficient of BC (MACBC) is defined as:   of coated BC particles and mass-equivalent diameters of BC core, respectively. The MR100 is the MR for particles with 100 nm BC core only, i.e. Dc equals to 100 nm. As stated in Section 2.3, values of meBC for RF05_1 and RF05_2 are constrained only by absorption coefficients with Eq. 9. To justify its feasibility, we derived meBC for other flights using Eq. 9 and found the differences are less than 5% for both the real and imaginary parts. No result for RF05_3 was obtained with CS model because no meBC value has been achieved in the retrieval.

20
The variation of keBC shows a similar pattern among different models (Fig. 4). Those values derived from the CS and VM models are the highest and lowest, respectively, consistent with modelling results for absorption in previous studies (e.g. Taylor et al. 2020), and those from the MG and BG models are between the other two and are very close to each other. The values of keBC for RF05_1, RF05_2, RF06_1, and RF06_2 from the CS model have greatly exceeded the largest kBC of commonly used values of mBC (grey shaded region in Fig. 4  internally mixed with salts/organics (e.g., example of RF05_3 in Fig. 5) or aggregates attached to non-BC components (e.g., example of RF05_2 in Fig. 5). Figure 5 shows representative TEM images of different types of BC-containing particles for each flight. Although a fair number of BC particles homogeneously mixed with salts and OA were detected from TEM for RF10 (Fig. 5), which seems to resemble homogeneous models, this sample 5 is thickly coated, with MR equal to 7.4 and MR100 equal to 22.5, largely exceeded the threshold value of MR for the CS model specified in Liu et al. (2017). In addition, results of BrC contribution to the total absorption also suggest that the CS model for RF10 is more reasonable, which will be discussed in detail in Section 3.3.  of keBC for RF05_1, RF05_2, RF05_3, RF06_1, and RF06_2, for which iron (Fe) was detected from the TEM-EDX (as shown in the bar chart in Fig. 5), are greater than those for RF10 and RF11, for which no iron was detected. Therefore, we suspect higher keBC values are related to iron oxides (FeOx) that absorbs at 660 nm, which we suspect to be magnetite, based on the following three points. Firstly, magnetite absorbs strongly and uniformly over the visible spectrum with the imaginary refractive index ranging from 0.58 to 1.0 according to literature (Amaury et al., unpublished data, http://www.astro.uni-jena.de/Laboratory/OCDB/mgfeoxides.html;Ackerman and Toon, 1981;Zhang et al., 2015), and can thereby contribute to the high values of keBC at 660 nm. Secondly, magnetite can be transformed at high temperatures from Fe(III), such as goethite and hematite (Till et al., 2015;Ito et al., 2018), which happen to be the two most abundant forms of FeOx in African dust (Formenti et al., 2014).

15
Therefore, it is speculated that a part of the magnetite is converted from Fe(III) during biomass burning. Thirdly, magnetite can be emitted by anthropogenic combustion, such as steel manufacturing, oil combustion, and vehicle emissions (Liati et al., 2015;Ito et al., 2018;Kurisu et al., 2019). Its significance in radiative forcing and carbon cycle has been investigated and highlighted in recent studies (Moteki et al., 2017;Ito et al., 2018;Lamb et al., 2021). Although there is no study on pyrogenic iron in sub-Saharan Africa yet, we speculate that anthropogenic 20 combustion emissions in Africa, especially in coastal areas, may also contribute magnetite. https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License. Among all flights, RF05_3, collected at the uppermost aerosol layer (Fig. 4), is an exception, since the values of the absorption coefficient at 660 nm calculated with all four models are considerably smaller than the measured ones (Fig. S3). The MG, BG, and VM models yield keBC values greater than 1.5, and no value was attained with the CS model. The MACBC of RF05_3 is the highest among all those investigated with values of 20.0±0.8, 17.8±0.8, and 14.3±0.7 m 2 g -1 at 470, 530, and 660 nm, respectively. The absorption enhancement, EAbs, 5 defined as the ratio of MACBC to the value for uncoated BC reported by Bond and Bergstrom (2006), is 2.3±0.1 for all three wavelengths. To the best of our knowledge, except for modelling or laboratory studies of thickly coated particles (Bond et al., 2006;Jacobson, 2012;Peng et al., 2016), such high values of EAbs, particularly at long visible wavelengths, are rarely reported in field measurements (Cui et al., 2016). Taylor et al. (2020) presented relatively high MACBC values of 20±4, 15±3, and 12±2 m 2 g -1 at 405, 514, and 655 nm, respectively 10 and an EAbs of 1.85±0.45 for the CLARIFY 2017; however, these values are still smaller than those for RF05_3.
Particles in CLARIFY 2017 are universally thickly coated, with median MR values of 8-12, and are therefore believed to have a largely enhanced lensing effect, which is not the case for RF05_3, which for which MR (Fig.   4) was equal to 1.6 and MR100 was equal to 5.3. The calculated absorption coefficient for RF05_3 at 660 nm, using mBC of 1.95+0.79i and mOA of 1.65+0i, was only 59% of the measured absorption coefficient (Fig. S3),

15
implying that the remaining 41% of absorption was contributed by other absorbers. This huge absorption from other absorbers in this flight may indicate a much larger amount of magnetite compared to other flights. In addition, we noticed titanium (Ti) on the particles analyzed from RF05_3. Formenti et al. (2014) found the mineral dust in western Africa consists mainly of clays, quartz, iron, and titanium oxides, which together represent at least 92% of the dust mass. Although titanium oxides are not absorbing, several forms of Ti have been reported to strongly 20 absorb at visible wavelengths (Pflüger and Fink, 1997;Palm et al., 2018). Therefore, we suspect this large absorption may also be related to absorbing titanium compounds.

Mass absorption coefficient of OA (MACOA)
https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License.  (Taylor et al., 2020), suggesting BrC 10 bleaching during transport considering our result is at a much longer wavelength (Che et al., 2021). RF05_1 and RF05_2, whose particles are equally or even more aged than those in CLARIFY, showed unexpectedly high MACOA,470 values of 1.18±0.54 m 2 /g and 1.07±1.60 m 2 /g, respectively. Although MOA for these samples are low (Fig. 6), measurements from all instruments are within the detection limit, which would argue that the values are valid. Wang et al. (2018) used airborne measurements to constrain their global model and found 15 the best MACOA to represent the measurements is 1.33 m 2 /g for freshly emitted BB OA at 365 nm. Lin et al. (2017) investigated relatively fresh BB aerosols subject to atmospheric processes during a night-long BB event in an urban environment and got a MAC of 0.9 m 2 /g for water extractable BrC at 470 nm under the peak BB episode.
Our values of MACOA,470 for RF05_1 and RF05_2 are higher than those for much fresher BB aerosols, which seems unrealistic, as BrC bleaching is expected to occur during transport. One possible explanation is that 20 secondary BrC formation occurred, perhaps through aqueous-phase chemistry during transport (Hems et al., 2021), for RF05_1 and RF05_2, which were sampled in or close to the MBL with higher RHs. Saleh et al. (2013) reported that secondary BrC can be more absorbing than primary BrC at short visible wavelengths; however, to the best of our knowledge, such high MAC for secondary BrC have not been documented (Kasthuriarachchi et al., 2020).
As we assume a constant meBC over the investigated spectrum, i.e. 470-660 nm, if there are some 25 components with strong absorption at short visible but not at long visible wavelengths, the meBC would be underestimated at shorter wavelengths and therefore MACOA would be overestimated. Hematite, whose imaginary refractive index ranges from 0 to 1.0 at 470 nm and 0 to 0.4 at 670 nm (Zhang et al., 2015;Go et al., 2021), is the second most abundant FeOx in western African dust (Formenti et al., 2014). Particles in RF05_1 and RF05_2 experienced approximately 6 days of transport and thus may have had more opportunities to mix with hematite 30 and therefore lead to an overestimation of MACOA. While lacking measurements on these absorbers, it is difficult to verify our speculation. A modified SP2 has been reported by (Yoshida et al., 2016) that can discriminate black-https://doi.org/10.5194/acp-2021-1000 Preprint. Discussion started: 28 January 2022 c Author(s) 2022. CC BY 4.0 License. coloured magnetite and red-coloured hematite; thus, we recommend adding such measurements in BB investigations.    Fig.   7, with averages of all flights below 5%, roughly a factor of three smaller than RBrC,470.

10
The extreme underestimation from the AAE attribution method is mainly due to the fact that the AAE and absorption coefficients used in this method are not derived from BC alone, but include contributions from other absorbing substances. Take RF06_1 as an example, the contribution of BrC derived from the AAE attribution method is 2%, approximately a factor of five smaller than that from our optical closure method with the BG model, 11% ( Fig. 7 and 8). As illustrated in Fig. 8, the absorption coefficient at 530 nm used in Eq. 14 15 and 15 includes the contribution of BrC, which accounts for 4% of the total absorption, and that of absorbers beyond BC and BrC, which accounts for 21%. Note this attribution is based on the BC absorption coefficient calculated from the BG model with mBC=1.95+0.79i, which will vary with the value of mBC. If we remove the 4% BrC from 8?7,B$> *48 in Eq. 14, the contribution would increase to 6%, two times larger than 2%, indicating that even though BrC may exist as a small portion at the long wavelength, its impact on the AAE attribution method can be 20 substantial. Thus, we recommend that application of any optical properties-based attribution method to use absorption coefficients at the longest possible wavelength to minimize the influence of BrC, and in the meanwhile, to account for potential contributions from other absorbing materials.

Conclusions
We investigated the contribution of BrC to the total absorption with different models utilizing and RBrC,470 from the three homogeneous mixing models were fairly close to each other, while those from the CS