Supporting information for : Production of particulate brown carbon during atmospheric aging of wood-burning emissions

Wall loss correction Solving the equations in Section 3.1 requires the determination of the time-dependent concentrations of the different absorbing species, which may be governed by their photochemical production or decay as well as by diffusion, electrostatic and gravitational losses to the walls. Assuming all particles are equally lost to the walls, an inert, non-volatile species, X, follows a first order decay:


INTRODUCTION
Atmospheric aerosols contribute to radiative forcing either directly by absorbing and scattering light or indirectly by acting as cloud-condensation and ice nuclei.While black carbon (BC) from combustion processes is the most efficient light-absorbing aerosol component, organic aerosols (OA) may also absorb solar radiation (Alexander et al., 2008;Chen and Bond, 2009;Kirchstetter et al., 2004).This light-absorbing OA, denoted as brown carbon (BrC), absorbs most strongly at shorter UV-visible wavelengths (Andreae and Gelencsér, 2006;Hoffer et al., 2005).
Global chemical-transport model estimates indicate that the BrC contribution to the positive radiative forcing of climate by anthropogenic aerosols may not be negligible (Feng et al., 2013;Jo et al., 2016;Lin et al., 2014;Wang et al., 2014).
Unlike BC, whose light absorption properties are relatively constant across sources (Bond et al., 2013), BrC is composed of a wide range of largely unknown compounds, which exhibit highly variable spectral dependence and absorption efficiencies.For example, reported imaginary indices of refraction for different organic species, which describe the absorption of these compounds, span two orders of magnitude (Lu et al., 2015).Because it is impractical to experimentally separate BrC from non-absorbing OA, optical properties are typically determined for the bulk OA of a given source.The large variability of BrC fraction in combustion aerosol may contribute to the wide variation in reported properties of BrC containing OA.
Biomass burning OA, which contributes two-thirds of the global budget of directly-emitted primary OA (POA), is expected to be a considerable source of BrC (Chakrabarty et al., 2010;Hecobian et al., 2010;Lack and Langridge, 2013;Liu et al., 2014).The variability in reported light absorption properties of biomass burning OA with fuel type and burn conditions remains a major obstacle complicating its treatment in climate models (Lu et al., 2015;Saleh et Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License. al., 2013).Residential biomass burning is typically characterized by a more efficient combustion, than open burning.
Residential wood burning represents a substantial contribution to anthropogenic combustion emissions (Bond et al., 2013), especially in urban atmospheres, and is considered the largest source of OA in Europe during winter (Denier Van Der Gon et al., 2015).
Upon photo-oxidation, biomass-burning emissions produce secondary organic aerosol (SOA) at concentrations similar to or exceeding the primary organic aerosol (POA) (Bertrand et al., 2017;Bruns et al., 2015Bruns et al., , 2016;;Corbin et al., 2015a;Grieshop et al., 2009).There is a growing body of evidence that light absorption by OA change with OH exposure (aging) owing to the production of secondary BrC or to the transformation of primary BrC (Heringa et al., 2011;Lee et al., 2014;Zhao et al., 2015).However, these effects have not yet been systematically investigated and must be quantified to assess the climate effects of primary and aged biomass burning OA.
Here, we show that both POA and SOA from residential biomass burning emissions aged in controlled smog chamber experiments contain BrC.Wavelength dependent, mass-normalized absorption cross-sections (MACs) of POA and SOA are presented from online aerosol measurements as a function of aging for the first time.
Complementary measurements of filter-extract absorbance (conducted in different solvents) are used to obtain the imaginary refractive index and to investigate the solubility of BrC in fresh and aged OA.While results presented here are related to flaming residential wood combustion emissions and cannot therefore be generalized, the approach used can be extrapolated for the characterization and quantification of the contribution of BrC in other primary and aged emissions.

Smog chamber experiments
Laboratory measurements were conducted in an 8 m 3 Teflon smog chamber (Bruns et al., 2015;Platt et al., 2013) installed within a temperature-controlled housing.Conditions in the chamber were maintained to represent winter time in Europe, i.e. relative humidity ranging between 50 -90%, at 263 K (Bruns et al., 2015(Bruns et al., , 2016)).Beech wood was combusted in a residential wood stove.Primary emissions were sampled through heated lines at 413 K, diluted by a factor of ~14 using an ejector diluter (DI-1000, Dekati Ltd.), then sampled into the chamber, which provided an additional ten-fold dilution.The overall dilution was a factor of 100 to 200.As we aimed to sample only flamingphase emissions into the chamber, samples were taken when the modified combustion efficiency (ratio of CO 2 to the sum of CO and CO 2 ) was > 0.90.Despite maintaining the same combustion conditions, the resulting organic fraction in the different samples was highly variable, indicating that these samples are representative of a mixture of preignition and flaming emissions (with varying contributions of each combustion stage).
After injection of the primary emissions and stabilization of the concentrations, nitrous acid (HONO) was continuously added, which dissociates upon irradiation (λ<400 nm) and forms the hydroxyl radical (OH).Then, 9times deuterated butanol sample (butanol-D9, 98%, Cambridge Isotope Laboratories) was subsequently injected into the chamber.The decay of butanol-D9 was used to infer the time-resolved OH exposure of the sampled aerosol (Barmet et al., 2012).The chamber was exposed to UV lights for ~3.5 hours.
Particles were collected onto filters (47 mm Tissue-quartz, Pall Corporation, 26 L min -1 for 30-32 min) for offline optical measurements and the determination of elemental carbon (EC) mass.Three filters were collected during each experiment, namely i) a primary aerosol filter sample ("primary"), ii) a slightly aged aerosol ("Aged1", OH exposure ~ 1x10 7 molecules cm -3 h), collected 30 minutes after the UV lights were switched on, and iii) an aged aerosol ("Aged2", OH exposure ~ 4x10 7 molecules cm -3 h), collected at the end of the experiment (see Figure S1 for the sampling periods).A charcoal denuder was installed upstream of the filter sampler to remove organic gases.
Filters were stored at 253K until analysis.
In addition to the characterization of the particle optical properties detailed in the next section, a set of online and offline techniques were used for the characterization of the gaseous and particulate emissions before and after aging.The non-refractory particle size-segregated chemical composition was measured with a high resolution (HR) timeof-flight aerosol mass spectrometer (AMS) (DeCarlo et al., 2006).Details related to the AMS data analysis and calibration can be found elsewhere (Bruns et al., 2015(Bruns et al., , 2016)).A scanning mobility particle sizer was used to measure the size distribution of the evolving aerosol.Organic gases were monitored by a proton transfer reaction time-of-flight mass spectrometer (PTR-MS, [H 3 O + ] reagent ion, Ionicon Analytik GmbH) (Bruns et al., 2017), following the same procedure as in Klein et al. (2016).Additionally, elemental carbon (EC) mass concentration was measured offline using a sunset thermo-optical analyzer, following the EUSAAR2 protocol (Cavalli et al., 2010).

Optical measurements
Aethalometer.A dual-spot aethalometer (Magee Scientific aethalometer AE33, Aerosol d.o.o.) was used for realtime aerosol light attenuation measurements at seven wavelengths (λ = 370, 470, 520, 590, 660, 880 and 950 nm) (Drinovec et al., 2015).The instrument measures the attenuation coefficient (b ATN ) of a light beam transmitted through a filter tape loaded with aerosol samples.The use of the sampling flow (here, 2 L min -1 ), integration time for the measurement (here, 1 minute), and automated dual-spot loading compensation to obtain b ATN has been described by Drinovec et al. (2015).
The loading compensated b ATN was used to infer the aerosol absorption coefficient, b abs , using a constant wavelength independent correction factor C, which accounts for multiple scattering within the filter matrix (Weingartner et al., 2003): The loading compensated b ATN at 880 nm from the AE33 is further used to infer the equivalent-BC mass concentration, M eBC : where  ATN is the mass attenuation cross-section of BC deposited on the filter of the AE33.M eBC inferred from The manufacturer default values are 1.57for C and 12.2 m 2 g -1 for  ATN at 880 nm, which corresponds to a MAC BC (880 nm) of 7.77 m 2 g -1 at (Gundel et al., 1984, Drinovec et al., 2015).However, these three parameters depend on aerosol properties.Here, we have determined the C value by applying Equation 1 to b ATN measured by the aethalometer and the absorption coefficient,  abs MWAA , measured by a multi-wavelength absorbance analyser, MWAA (Massabò et al., 2015;Massabò et al., 2013).The MAC BC (880 nm) was determined using Equation 4to compare  abs MWAA from the MWAA measurements with EC mass from the Sunset thermo-optical analyzer (see Figure 1A&B and Section 4.1 for detailed discussion).Following this procedure, the MWAA and Sunset analyser will be defined as reference methods for absorption coefficient and EC mass concentration, respectively.Note that data from these reference methods were only available with low time resolution and for a subset of all samples.
Thus, the aethalometer anchored against these reference methods, was used to obtain the wavelength dependent absorption coefficients and the eBC mass concentrations with high time resolution using Equations 1 and 2, respectively.Processing the loading compensated AE33 attenuation coefficients with C value and MAC BC , determined with independent MWAA and Sunset analyser measurements, ensures that the inferred  abs () (Equation 1) and M eBC (Equation 2) have minimal bias compared to respective true values.
MWAA measurements.The MWAA (Massabò et al., 2015;Massabò et al., 2013) was used as reference method for the aerosol absorption coefficient.It measures the absorption coefficient  abs MWAA (λ) of particles deposited on on standard filter samples.It is composed of five laser diodes, with λ = 375, 407, 532, 635 and 850 nm, acting as light sources and placed above the filter, an automated sample-changer, and three low-noise UV-enhanced photodiodes.The first photodiode is placed behind the filter for the analysis of transmittance measurements, while the other two photodiodes are positioned at specific angles between the sources and the loaded filter to perform reflectance measurements.These transmittance and reflectance measurements are used together with a radiative transfer model (Hänel et al., 1987) , which takes into account multiple scattering within the particle/filter layer, to retrieve both the total optical thickness and the particle-filter-layer single scattering albedo, providing the absorption coefficient  abs MWAA (λ) values.These calculations largely follow the approach implemented in the multi-angle absorption photometer (Petzold and Schönlinner, 2004).

UV-visible
where l is the optical path length.
The absorbance measurements are aimed at inferring the imaginary part of the refractive index.For this,  abs,OA−sol () is transformed to the absorption coefficient of the bulk OA in the pure form,  abs,OA−bulk (Sun et al., 2007): where  OA is the bulk density of OA (assumed to be 1.5 g cm −3 , typical of wood-burning OA; (Corbin et al., 2015a;Moosmüller et al., 2009;Sun et al., 2007)),  OA is the extracted OA mass, and  solvent is the solvent volume.The bulk absorption coefficient directly leads to the imaginary part of the OA refractive index,  OA , in pure form (Moosmüller et al., 2009): Inserting Equation 6into Equation 7 eventually provides (Liu et al., 2015a): The mass of organics dissolved in the solution could not be quantified.Therefore, we use an upper limit value for compared to those calculated based on Equation 8 may be related to low OA extraction efficiency or to nonextractable highly absorbing material and results shall be discussed accordingly.

Determination of absorption Ångström exponents and mass absorption cross-sections
In this section we describe the methodology adapted for the determination of the mass absorption cross-sections (MACs) for the different aerosol material from the Sunset, MWAA and aethalometer measurements.The assumptions and limitations underlying these calculations are clearly stated.We also explain the relationship between the MACs and the wavelength dependence of the overall absorption.
Definition of the absorption Ångström exponent .The wavelength dependence of the overall absorption due to both BC and BrC has often been described assuming a power law: where α is the Ångström absorption exponent, often determined by fitting the absorption coefficient measurements across the entire wavelength range.Equation 9 is an empirical simplification, which breaks down when different components having different spectral dependence contribute to the absorption, e.g. a mix of BrC and black carbon (e.g., Moosmüller et al., 2011).In practice, different values of α would be obtained for different choices of λ ranges, and therefore we alternatively calculated two-wavelength absorption exponents according to where  is a wavelength of interest (in nm) and ref  is the reference wavelength, here 880 nm.This reference wavelength was chosen, because BC is expected to fully dominate light absorption in this range (Laskin et al., 2015).Black carbon is known to have an α between 0.9 and 1.1 (Bond et al., 2013;Kirchstetter et al., 2004;Liu et al., 2015b), whereas BrC, which preferentially absorbs at shorter wavelength, has a higher α (Laskin et al., 2015;Saleh et al., 2013).Thus, we interpret an increase of ) , ( can potentially change due to other effects such as a wavelength dependent lensing effect on absorption by BC (e.g., Lack and Langridge, 2013) or the restructuring of BC aggregates during aging.The former effect was negligible under our conditions, as elaborated on below.The latter, if it occurs during aging, would be attributed to SOA absorption in our approach.However, this is not an issue if our values are accordingly applied in e.g.model simulations, following the same assumption as in our approach.This means that the potential restructuring effects must implicitly be considered within the MAC() of SOA, while the MAC() of BC must be kept fixed.

Determination of MAC BC and MAC POA using the absorption Ångström exponent
In a mixture of n absorbing species, the total absorption at any wavelength may be written as the sum of the absorbance of each of the species.Accordingly, Equation 10 can be expressed for a multi-component system In equation 12, the summation now only goes over the n-1 organic species, which contribute to light absorption.
The fresh combustion aerosol exclusively contains BC and POA as absorbing species.In Equation 13, M OA (t 0 ) is the mass concentration of primary organic aerosol measured by the AMS at  0 .
MAC BC (t 0 ,880nm) was inferred from the MWAA and Sunset thermo-optical analysis and shown to be independent of the experimental conditions (Section 4.1; Figure 1A).Absorption coefficients  abs ( 0 , ) are obtained from the high time resolution attenuation measurements by the aethalometer referenced to the MWAA absorption measurements as described above.( 0 , , 880 nm) is derived from  abs ( 0 , ) and  abs ( 0 , 880 nm) using Equation 10.This leaves only 2 free parameters in Equation 13, MAC BC (t 0 , λ) and MAC POA (t 0 , λ).These were determined by fitting Equation 13to ( 0 , , 880 nm), M OA (t 0 ), MAC BC (t 0 ,880nm) and  abs ( 0 , 880) data measured in all experiments for fresh emissions at  0 .This approach contains the implicit assumption that the two MAC values are also independent of experimental conditions, and therefore these MACs should be considered as average values.The accuracy of these MAC values obviously depends on the accuracy of the absorption and mass measurements.First, a systematic bias in the C value potentially caused by a systematic bias in the MWAA measurements propagates to an identical bias in both MAC BC (t 0 , λ) and MAC POA (t 0 , λ).Second, a systematic bias in the Sunset EC mass measurements yields a corresponding inverse bias in MAC BC (t 0 , λ), while MAC POA (t 0 , λ) remains unaffected.Third, a systematic bias in the AMS POA mass yields a corresponding inverse bias in MAC POA (t 0 , λ), while MAC BC (t 0 , λ) remains unaffected.Equation 13shows that  of the primary aerosol at a certain wavelength is largely driven by MAC POA ( 0 , ), i.e. the optical properties of POA, and by the ratio abs ( 0 ,880) , which reflects the relative contributions of POA and BC to total primary aerosol mass.

Determination of MAC SOA
The MAC of SOA, MAC SOA , can be generally defined as: where  abs,SOA and  SOA are the absorption coefficient and mass concentration of SOA, respectively.In the aged Note that inferring  abs,POA+BC (, ) from  abs (, 880 ) implicitly accounts for the decrease in the BC and POA absorption due to wall losses.
SOA was obtained as total organic minus POA mass concentration: The POA mass concentration in the aged aerosol can be inferred from the initial OA mass concentration in the fresh emissions by accounting for the wall losses using Equation S1 and the wall loss time constant  (see Section Wall loss corrections in the SI): MAC SOA can be calculated for every data point in time and for all aethalometer wavelengths from 370 to 660 nm (MAC SOA defined to be zero at λ  880 nm), as all quantities on the right hand side of Here,   and   are the number of particles and particle diameter, respectively, in the i th size bin, and  OA is the real part of the refractive index of the OA (which is assumed to be  OA = 1.5 typical for organic material; Lu et al., 2015).The MAC of particles with diameter   , MAC  Mie , was calculated using the Mie Code by Peña and Pal (2009) (incorporated into Igor Pro 6.3, WaveMetrics, OR, USA by Taylor et al., 2015).MAC  Mie also depends on the density of OA, for which we assume a value of  OA = 1.5 g cm −3 (see Section 2.2), as the volume specific absorption crosssection obtained from Mie theory needs to be converted to a mass specific absorption cross-section.We note that as we have used the same value of  OA in the calculation of both MAC  Mie and  OA (), MAC OA,bulk becomes independent of the assumed  OA value.
Assuming spherical particles and neglecting the presence of BC in these particles may seem inappropriate.However, calculations considering BC and assuming core-shell morphology revealed (1) limited sensitivity of the resulting MAC OA to this assumption and (2) a higher than measured lensing effect.Therefore, a substantial fraction of the OA seems to be externally mixed and to dominate the measured size distribution (see also Section in the MAC OA inferred from  OA of the UV-visible absorbance measurements was estimated by combining an estimated 20 % precision with a detection limit of 0.3 m 2 g -1 in quadrature.

Verification of MAC BC and C value
As mentioned above, the determination of MAC BC (880nm) requires the determination of the absorption coefficients at λ and the BC mass.We used the aethalometer to obtain the absorption coefficients with high time resolution, while absolute values were scaled to match MWAA data, which we defined as our reference method.The aethalometer was also used to obtain eBC mass concentrations with high time resolution, while absolute values were scaled to match EC mass measured by the Sunset thermo-optical measurement using the EUSAAR-2 protocol, which we defined as our reference method.Here we start by proving the concept of our scaling approaches and provide average values for MAC BC (880nm) and aethalometer-C which are required subsequently.
Figure 1A shows the correlation between the MWAA measured absorption coefficient at 880 nm and the Sunset thermo-optical EC mass measurements.MWAA absorption measurements at 880 nm is determined by extrapolating the absorption coefficients at 850 nm using an α determined from the ratio between the absorption coefficients at 850 nm and 635nm.The corresponding MAC BC (880nm), determined as the slope of the linear fit through all data, is 4.6 ± 0.7 m 2 g -1 .This value matches the data at all three levels of aging, i.e. for the primary, Aged1 and Aged2 filter samples, within experimental uncertainty (see Figure S2 in the Supplement for more information).This average MAC BC (880nm) is also very similar to values reported for "pure" BC (4.7 ± 0.7 m 2 g -1 at 880 nm) (Bond et al., 2006), indicating no significant lensing effect on absorption by BC from primary or secondary OA.This can also be observed from the time resolved attenuation measurements by the aethalometer at 880 nm (Figure S3), suggesting that little (<10%) to no increase in the attenuation coefficients upon SOA formation.
If the OA and the BC were internally mixed, the observed variability in the mass fraction of OA (  ) from 0.1 to 0.9 for the fresh and aged samples would result in a high variability in the MAC BC (880nm), with values higher than those reported in the literature, according to Mie calculations assuming core-shell internal mixtures.However, this is not the case.Based on this observation, we conclude that the particles studied are likely not core-shell internal mixtures, although we have measured a mono-modal aerosol population growing during SOA production (Figure S4).An explanation for the occurrence of an external mixture could be that the primary OA and BC particles may have been externally mixed after these species were emitted separately during combustion, preferentially during the pre-ignition and flaming phases, respectively (Corbin et al., 2015a(Corbin et al., , 2015b;;Heringa et al., 2011).These phases may occur consecutively during a burn or simultaneously in different parts of the stove.MAC BC (880nm) found to be constant supports our approach described in Section 2.2 using scaled aethalometer data for BC mass and treating MAC BC at all other wavelengths as a constant across all experiments during the data retrieval process described in Section 3.1.
Figure 1B shows the correlation between  ATN,AE33 and  abs,MWAA measured by the aethalometer and MWAA, respectively.The two variables correlated very well, indicating a constant aethalometer C-value, which is the ratio between  ATN,AE33 and  abs,MWAA (Equation 1), of 3.0±0.3,independent of the type of the aerosol sampled.This is also reflected in the probability density function of individual C-values shown in Figure S2 where the standard deviation is found to be as small as   ~ ± 10% .Such constant ratio justifies our approach of applying this single C value for all conditions in order to scale the time resolved attenuation measurements by the aethalometer to the MWAA reference method.
Note, the manufacturer's default values, which were not applied in our case, are 1.57for C and 12.2 m 2 g -1 for  ATN at 880 nm, which implies an underlying MAC BC (880 nm) of 7.77 m 2 g -1 (Gundel et al., 1984, Drinovec et al., 2015).Therefore, factory default  abs () would have a substantial systematic high bias for the wood combustion aerosols of this study.Meanwhile, the  ATN calculated at 880 nm, which is the product of the C value and MAC BC (Equation 3), is consistent with the manufacturer value of  ATN ( ATN values determined here are 15% higher, 13.8 m 2 g -1 in this study compared to the value of 12.2 m 2 g -1 provided by the manufacturer), and the factory default M eBC would agree well with the true  BC , determined here.

Optical properties of BC, POA, and SOA
In this section we derive the wavelength dependent mass absorption cross-sections for BC, POA and SOA.In Figure 2, we display the evolution of (370, 880) as a function of OH exposure.Figure 3 shows the relationship between (, 880) and  OA for primary and aged aerosols.α of primary emissions.The (370, 880) values computed for the primary aerosol (OH exposure = 0 molecules cm -3 h) ranged between 1.3 and 1.7 (Figure S5), which is within the range reported previously for biomass-burning emissions (Kirchstetter et al., 2004;Lewis et al., 2008;Zotter et al., 2016).The (, 880) is close to that of pure BC (~0.9-1.1;Bond et al., 2013;Zotter et al., 2017) for small f POA , while increasing f POA corresponded to a distinct increase of (, 880).This increase provides clear evidence for the contribution of primary BrC to the absorption at lower wavelengths (shown explicitly in Equation 13).The f POA ranges from 0.23 to 0.59, which is lower than f POA reported for open burning emissions (e.g., f POA ~0.75, Ulevicius et al ( 2016)), because our wood-stove emissions feature a more efficient combustion.The systematic decrease in (, 880) with increasing  reflects the more-efficient light absorption by BrC at shorter wavelengths (Moosmüller et al., 2011), and shows that the power law wavelength dependence is an inaccurate oversimplification for this mixed aerosol.
Evolution of α with aging.Figure 3B shows that upon aging, the OA fraction rapidly increased (a typical time series of raw data is shown in Figure S1), reaching an average value of 0.81 (full range for aged OA: 0.74 < f OA < 0.89) at high OH exposures (> 2×10 7 molecules cm -3 h), and resulting in a corresponding increase of  BC+POA+SOA (370nm, 880nm).The increase of  BC+POA+SOA (370nm, 880nm) and  OA were always correlated and plateaued at OH exposures beyond ~2×10 7 molecules cm -3 h, as seen in Figure 2. Also, note in Figure 2 that the highest  BC+POA+SOA (370nm, 880nm) were reached, on average 1.8, during experiments where the f OA was highest.
Such strong correlation between SOA formation and  BC+POA+SOA (370nm, 880nm) suggests the production of substantial amounts of brown SOA.A similar relationship is observed between  BC+POA+SOA (λ, 880nm) and f OA for higher wavelengths as shown in Figure S6.Similar to the case of POA, a systematic decrease in (, 880) with increasing  is observed, reflecting the preferential absorption of BrC SOA at shorter wavelengths.We note that  BC+POA+SOA (370, 880) as a function of OA f for all experiments lies below the overall trend for the primary aerosol (dashed line in Figure 3B), implying that MAC SOA (370nm) was smaller than MAC POA (370nm).
Determination of MAC BC and MAC POA .We determined best-fit values for MAC BC (λ) and MAC POA (λ) from the data shown in Figure 3A. Figure 3A includes least-squares fits of Equation 13 to the data, with MAC BC (λ) and MAC POA (λ) as fit parameters.The fit results are shown in Table 1.The obtained fit value of MAC BC (370nm) was 13.7 m 2 g -1 (GSD 1.1), higher but not statistically significantly different from the value suggested by Bond et al. (2013) of 11.1 m 2 g -1 with a 95% confidence interval of 3.5 m 2 g -1 , considering α BC =1.Meanwhile, the mean MAC POA (370nm) value, equal to 5.5 m 2 g -1 , obtained under our conditions for domestic wood burning is ~2.4 times higher than that obtained by Saleh et al. (2014) for open biomass burning primary emissions, suggesting the presence of more-strongly absorbing organic material under our conditions (this comparison is continued in Section 4.3).
Determination of MAC SOA .The MAC SOA (λ) values, determined using Equation 19, are shown in Figure 4 and Table 1.MAC SOA (370nm) was 2.2 m 2 g -1 (GSD 1.39), a factor of 2.5 smaller than MAC POA (370nm), but approximately an order of magnitude higher than values reported for ambient oxygenated aerosols or laboratory SOA from biogenic and traditional anthropogenic precursors such as terpenes and methyl-benzenes (Clarke et al., 2007;Lambe et al., 2013;Liu et al., 2016;Romonosky et al., 2015).The predominant SOA precursors identified in wood smoke comprise (methyl)naphthalene(s) and phenol derivatives from lignin pyrolysis (Bruns et al., 2016;Ciarelli et al., 2016), the oxidation products of which are expected to be highly light absorbing due to the presence of aromatic moieties in the SOA (Bruns et al., 2016;Laskin et al., 2015).In this regard, it is not surprising that the MAC SOA (370nm) values obtained here are similarly high as those obtained from methanol-extracted SOA from guaiacol and naphthalene oxidation (0.5-3.0 m 2 g -1 , Romonosky et al., 2015).
Uncertainties and variability in MAC BC , MAC POA and MAC SOA .Table 1 shows that the uncertainties in the fitted MAC BC (λ) are relatively low (< 10%), increasing with decreasing λ.By contrast, the uncertainties in the fitted MAC POA are much higher (GSD = 1.2-1.5)and increase with increasing λ.The relative residuals between the measured and fitted (, 880nm) for primary emissions showed small biases of only 0.07 (Figure S7).The corresponding RMSE (root mean square error) was 0.13, showing that the obtained average values may represent the data well.MAC SOA values determined were highly variable between experiments with a GSD = 1.39 and 2.42 for =370 nm and 660 nm, respectively.We expect the variabilities in MAC SOA and of MAC POA to be related to changes in the organic aerosol chemical composition between different burns, since the variability of MAC BC (λ) was relatively small.In Section 4.3, we discuss this variability further using the results of an additional and independent analysis.
MAC BC , MAC POA and MAC SOA wavelength dependence.The relationships between the MAC SOA (), MAC POA () and MAC BC () and wavelength appear to fall on three unique lines in the range 660 nm to 370 nm when plotted in log-log space, as shown in Figure 4 (Figure S8 shows the same data plotted on a linear scale).This indicates that a power-law approximation provides a good description of the behavior of individual components within this wavelength range from 370 nm to 660 nm.Accordingly we fitted the power law coefficients to the data shown in  .This yielded α BC = 1.2, α POA = 4.6, and α SOA = 5.6.Note that α BC in the range 660 nm to 370 nm obtained from this fit is very similar to α BC values that can be inferred by extrapolating the data shown in Figure 3A to f OA =0.The high α values obtained for the organic fractions are consistent with previous measurements for BrC containing POA (e.g.Chakrabarty et al., 2010Chakrabarty et al., , 2013) ) although, to our knowledge, this is the first study to report α SOA without performing a solvent extraction.
Evolution of MAC OA with aging.In Figure 5, we examine whether the absorption profile of SOA evolved with aging.A change in MAC SOA (370nm) or  SOA with increasing OH exposure may indicate either a change in the mass-specific absorption of the condensing SOA species with time, or a change (e.g."bleaching") in the MAC of pre-existing POA. Figure 5 indicates that neither of these scenarios was the case.Both MAC SOA (370nm) and  SOA were statistically independent of the OH exposure, for exposures up to 40 molec.OH cm -3 h.This signifies that under our conditions and within our measurement uncertainties the optical properties of the additional organic mass formed was constant with aging, under the assumption that the light-absorption properties of POA were negligibly influenced by aging.Most of the variability in MAC SOA () discussed above is therefore related to experiment-toexperiment differences rather than to the extent of OH exposure, as it is also shown below.

Solubility of BrC in methanol and water
Figure 6 shows the MAC OA (370nm) determined from the water and methanol extracts against the MAC OA (370nm) determined from the online measurements.The MAC OA (370nm) from online measurements was estimated by subtracting the contribution of BC assuming a constant MAC BC (370nm) = 13.7 m 2 .g - as obtained in this work (Table 1).We performed all the calculations and comparisons at λ = 370 nm, as the signal to noise ratio of the absorption coefficients measured by UV-visible spectroscopy and the contribution of BrC to the total carbonaceous absorption are highest at this wavelength.The MAC of the extracts was computed from the k OA through Mie calculations.Repetition of both water and methanol extracts yielded results that were consistent within 10% (Figure S11).Average raw absorption spectra are shown in Figure S12.
Figure 6B shows excellent correlation between the MAC values obtained from the the k OA of the methanol extracted OA with the in-situ method described above.This correlation suggests that none of the assumptions employed in either method led to substantial errors in precision, providing direct support for our results.A similar relationship was observed between k OA and the MAC OA (370nm) determined from the online measurements (Figure S13), showing that this relationship is not sensitive to assumptions underlying the Mie calculations.It further suggests that the wide variability observed in the MAC OA values of different burns, as seen Figure 6, most likely reflects real variability in the optical properties of POA and SOA rather than random noise or experimental errors in the retrieved quantities.MAC OA retrieved based on the k OA of the water soluble OA show substantially more scatter than observed in Figure 6B (for both primary and aged data), suggesting a variable extraction efficiency in the case of water, which we also attribute to variability in the OA composition.
The orthogonal, uncertainty-weighted linear regression in Figure 6B shows that the methanol extracts explain 46 ± 10% of the online MAC.(Note that, in this analysis, aged OA refers to the sum of POA and SOA for aged samples.)Considering the simplifying assumptions that were necessary for our Mie analysis and those related to online MAC OA calculations, we consider this an adequate agreement.In particular, the assumption of a perfect extraction efficiency of OA in methanol may have been violated (see Section 3.4).Conversely, the fit in Figure 6A indicates that the apparent MAC of water-soluble species was a fourth of the respective methanol MAC, according to the slope of only 12 ± 3%. .This strong disagreement shows that the BrC in our samples was hardly water soluble, even for the most aged samples.As we expect that the majority of OA in our samples formed by wood pyrolysis (Di Blasi, 2008;Corbin et al., 2015b;Shafizadeh, 1984), we can compare our results directly to those of Chen and Bond (2010), who also found that primary wood-pyrolysis BrC was water insoluble.Moreover, the waterinsoluble nature of the light absorbing components of SOA is in line with the results by Bruns et al. (2016) who showed that the precursors of SOA in these experiments were predominantly aromatic compounds.

Comparison of k OA with literature
The results above highlight the variability in the OA absorption properties.In this section, we discuss potential reasons for this variability and compare our results to literature. Figure 7 shows the imaginary refractive index of methanol-extracted OA at 370 nm, k OA,methanol (370nm) (Equation 8), as a function of M BC /M OA and aging.The data are plotted against M BC /M OA instead of  OA to allow for a direct comparison with literature (see Figure S14 for a plot against  OA ).An approximately linear trend of k OA,methanol (370nm) with M BC /M OA is seen in log space.This agingindependent relationship may be useful in, for example, atmospheric scenarios where wood-burning OA is a dominant aerosol component but its exact degree of aging is unknown.⁄ can be estimated.For these calculations, we have estimated the wall losses using two approaches as described in the SI.
The  SOAP,WLC  POA,WLC ⁄ was on average equal to 7.8 (GSD = 1.4) and  OH was estimated as 2.7×10 -11 molecule -1 cm 3 (GSD = 1.4), consistent with the chemically speciated data obtained by a proton-transfer-reaction mass spectrometer (PTR-MS) (Bruns et al., 2016(Bruns et al., , 2017).These high rates and enhancement ratios indicate the rapid production of SOA.Based on the bulk gas phase measurements of SOA precursors (Bruns et al., 2016), the obtained enhancements are consistent with high bulk SOA yields of ~50%.These high values are not surprising, considering the nature of these gases (e.g.PAH and phenol derivatives), the low temperatures (263 K), and the relatively high concentrations (Aged OA ~100 µg m -3 ) at which the experiments have been conducted (Bruns et al. 2016).
Combining these calculated enhancements with the average contributions of POA in primary emissions, the evolution of f OA with aging was determined and is shown in Figure 8B.The uncertainties in Figure 8B (dotted lines) represent one standard deviation on f OA obtained by a Monte Carlo propagation of uncertainties due to experimentto-experiment variability, fitting errors and wall loss correction errors (see SI).While this calculation represents a simplification of the SOA production mechanisms (the dependence of SOA yields on OH exposures/multigeneration chemistry and OA mass concentrations was neglected), it results in residuals much smaller than the experiment-toexperiment variability.We therefore used these calculations to assess the relative contribution of OA to the total carbonaceous absorption.We show in Figure 8C that below 400 nm and upon aging, the absorption coefficient of the total organics was at least as high as the one of BC.
Using the MAC values of the different components (in m 2 g -1 ), their abundance (in g m -3 ) and the solar irradiance data (S, in W m -2 nm -1 ) calculated at sea level for a cloudless day, the fractional energy transfer due to the BrC light absorption relative to that due to the total carbonaceous aerosol absorption ,  OA ( exp ), in air masses dominated by residential burning emissions can be determined as Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.
OA , approximated as the integral of AMS-measured OA mass concentration times sample flow rate over the filtersampling period.Accordingly, the resulting  OA values represent lower limits for the true values, as the OA extraction efficiency was not accounted for.Higher  OA values based on online absorption coefficient measurements Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.
where the right hand side follows the general definition of MAC along the lines of Equation 4. M i and MAC i are the mass concentration and MAC, respectively, of the i th species, with n absorbing species in total.By considering that the light absorption at ref  = 880 nm is exclusively due to BC, and by defining BC to be the n th species, Equation 11 can be written as For the data at time t 0 before the start of photo-oxidative aging, Equation 12 simplifies to: ( 0 , , 880) =  BC+POA ( 0 , , 880) Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 1 :
Figure 1: Determination of (A) MAC BC (880nm) and (B) aethalometer C value using MWAA absorption measurements, thermal/optical EC (EUSAAR2 protocol) and aethalometer attenuation measurements.MWAA absorption measurements at 880 nm is determined by extrapolating the absorption coefficients at 850 nm using an α determined from the ratio between the absorption coefficients at 850 nm and 635nm.The aerosols were either primary (no OH exposure), Aged 1 (~1x10 7 molec OH cm -3 h), or Aged 2 (~4x10 7 molec OH cm -3 h).No difference in MAC or C value was discernable with aging (see also Figure S2).Also shown is the MAC of pure BC recommended by Bond et al. (2006) (dotted line in A).The C value derived from   recommended by Drinovec et al. (2015) = 2.6 compares well with the value derived in Figure 1B.

Figure 2 :
Figure 2: Evolution during photochemical aging of  ++ (, ) (two-wavelength Ångström exponent calculated using total absorption data at 370 nm and 880 nm), where the different symbols denote individual experiments.Data are colored by the OA mass fraction   =   /(  +   ).The black line is a fit to guide the eye.

Figure 3 :
Figure 3: (A) Relationship of  + (, ) to   for seven wavelengths.Lines are fits of Equation 13 to the data.(B) Relationship of  ++ (, ) to   for several experiments.Data in (A) and (B) are colored by the wavelength and OH exposure, respectively.

Figure 4 :
Figure 4: MAC SOA (λ) calculated from several smog chamber experiments plotted as box-whiskers as a function of wavelength (also shown by the color of the bars).The thick black lines, the boxes and the whiskers mark the medians, the quartiles and the 10 th and the 90 th percentiles, respectively.Also shown are the MAC BC (λ) and MAC POA (λ) reported in

Figure 5 :
Figure 5: MAC SOA (370nm) and  SOA,fit (370nm, 660nm) calculated from several smog chamber experiments plotted as a function of OH exposure.MAC SOA (370nm) was obtained using Equation 19. SOA,fit (370nm, 660nm) was obtained from fitting the MAC SOA values in the range 370-660 nm for the different experiments against the wavelength. SOA,fit (370nm, 660nm) is the slope of the linear fit applied after log transforming the data.MAC SOA (λ) for higher wavelengths are shown in figure S10.

Figure 6 :
Figure 6: Comparison of the MAC OA (370nm) of aged aerosols determined from online and offline measurements of absorption.The offline filter extraction method directly quantified properties of total OA (ordinate), while the average of MAC SOA and MAC POA from the online measurements weighted with respective mass concentrations is shown on the abscissa.(A) offline measurements of water-soluble OA, (B) methanol-soluble OA.

Figure 8 :
Figure 8: Impact of BrC absorption on total primary and secondary wood-burning-aerosol absorption.(A) MACs of different particle components (BC, POA and SOA) along with their corresponding standard deviations plotted as a function of wavelength based on smog chamber data and compared to the solar irradiance spectrum.(B) Species average relative abundance in the smog chamber (f OA ) plotted as a function of the OH exposure.(C) Image plot showing the OA absorption coefficient relative to the total aerosol absorption as a function of wavelength and OH exposure.(D) Rate of energy transfer due to BrC light absorption relative to the total carbonaceous aerosol absorption (W OA ) estimated as a function of aging using the solar flux, the fractions of the different components and their MACs.

837wavelengths.
Uncertainties were obtained from fits of Equation 13 for MAC BC, MAC POA , while for MAC SOA uncertainties 838 GSD values are geometric standard deviation values on the MAC SOA average values from all experiments.These 839 uncertainties do not include uncertainties related to the determination of MAC BC (880nm).By definition, BrC absorbance 840 at 880 nm is zero.
Equation 2 only equals the true BC mass concentration,  BC , if the applied  ATN is identical to the true attenuation cross-section of BC,  ATN,BC , and if light attenuation at 880 nm is exclusively due to BC.  ATN,BC (880 ) can be inferred from the true MAC of BC, MAC BC , and the true C value: Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.
abs,POA+BC (, 880 ) ≡  abs(, 880 ).The second assumption is that the two-λ α values of primary emissions do not change during aging  POA+BC (, , 880 ) ≡  POA+BC ( 0 , , 880 ).The latter approximation is based on the underlying assumptions that the MAC of POA is not altered by aging and that the proportions of POA and BC mass lost to the wall are identical.Under these assumptions  abs,POA+BC becomes:  abs,POA+BC (, ) =  abs (, 880 ) ( aerosol, which contains the absorbing species BC, POA and SOA,  abs,SOA is the difference of the total absorption minus the absorption by POA and BC:  abs,SOA (, ) =  abs (, ) −  abs,POA+BC (, ) (15) Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscriptunder review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.The absorption by POA and BC in the aged aerosol is a priori unknown, but can be calculated under certain assumptions.The first assumption is that SOA does not contribute to absorption at 880 nm:

4 Mie calculation to relate k OA with MAC OA The
Equation 19 are available from either the aethalometer or AMS measurements or are otherwise known.It can be seen from Equation19that the imaginary part of the refractive index of an aerosol component is an intensive material property.However, the MAC of such an aerosol component additionally depends on the size and morphology of the aerosol (except for the Rayleigh regime).The online aerosol absorption measurements provide estimates for MAC values, while the UVvisible absorbance measurements of filter extracts provide the imaginary part of the refractive index.We used Mie calculations in order to compare the two quantities.The  OA ()obtained from the filter extracts is converted to a MAC OA,bulk (,  OA ,  OA ,  OA ) = mass concentrations used to calculate MAC SOA solely originate from AMS data, thus being consistent with the calculation of MAC POA (see above).Equation 19 is based on the assumption that POA is "chemically inert", i.e. no chemically induced changes of  POA and MAC POA occur.Such chemically induced changes of absorption by POA, Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscriptunderreview for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.iftheyoccur, are assigned to the absorption by SOA, thus resulting in a corresponding adjustment of the inferred MAC SOA .3.MAC OA,bulk by assuming that all OA is present in homogeneous spherical particles with a diameter distribution identical to the mobility diameter distribution measured by the SMPS.In this manner, MAC OA,bulk becomes equal to the mass-weighted average (=volume-weighted average) of the diameter dependent MAC: The decrease of M BC /M OA caused by Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.this equation,  SOA,WLC (),  POA,WLC () and  SOAP,WLC are the wall loss corrected mass concentrations of SOA, POA and the SOA potential (the maximum SOA formed upon the consumption of all precursors).OH represents an estimation of reaction rate of SOA precursors towards OH based on SOA production rates.By fitting the observed OA ( exp ) =  OA ( exp )  tot ( exp ) ⁄ = ∫ {  POA ( exp )× POA ()+ SOA ( exp )× SOA ()}×()× 880 300 ∫ {  BC ( exp )× BC ()+ POA ( exp )× POA ()+ SOA ( exp )× SOA ()}×()× 880 300 (22) Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-159Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 13 April 2018 c Author(s) 2018.CC BY 4.0 License.