Supplement of Source-specific light absorption by carbonaceous components in the complex aerosol matrix from yearly filter-based measurements

Abstract. Understanding the sources of light-absorbing organic (brown) carbon (BrC)
and its interaction with black carbon (BC) and other non-refractory
particulate matter (NR-PM) fractions is important for reducing uncertainties
in the aerosol direct radiative forcing. In this study, we combine multiple
filter-based techniques to achieve long-term, spectrally resolved, source-
and species-specific atmospheric absorption closure. We determine the mass
absorption efficiency (MAE) in dilute bulk solutions at 370 nm to be equal
to 1.4 m2 g−1 for fresh biomass smoke, 0.7 m2 g−1 for
winter-oxygenated organic aerosol (OA), and 0.13 m2 g−1 for other less absorbing OA.
We apply Mie calculations to estimate the contributions of these fractions
to total aerosol absorption. While enhanced absorption in the near-UV has
been traditionally attributed to primary biomass smoke, here we show that
anthropogenic oxygenated OA may be equally important for BrC absorption
during winter, especially at an urban background site. We demonstrate that
insoluble tar balls are negligible in residential biomass burning
atmospheric samples of this study and thus could attribute the totality of
the NR-PM absorption at shorter wavelengths to methanol-extractable BrC. As
for BC, we show that the mass absorption cross-section (MAC) of this
fraction is independent of its source, while we observe evidence for a
filter-based lensing effect associated with the presence of NR-PM
components. We find that bare BC has a MAC of 6.3 m2 g−1 at 660 nm and an absorption Ångström exponent of 0.93 ± 0.16,
while in the presence of coatings its absorption is enhanced by a factor of
∼ 1.4. Based on Mie calculations of closure between observed
and predicted total light absorption, we provide an indication for a
suppression of the filter-based lensing effect by BrC. The total absorption
reduction remains modest, ∼ 10 %–20 % at 370 nm, and is
restricted to shorter wavelengths, where BrC absorption is significant.
Overall, our results allow an assessment of the relative importance of the
different aerosol fractions to the total absorption for aerosols from a
wide range of sources and atmospheric ages. When integrated with the solar
spectrum at 300–900 nm, bare BC is found to contribute around two-thirds of
the solar radiation absorption by total carbonaceous aerosols, amplified by
the filter-based lensing effect (with an interquartile range, IQR, of 8 %–27 %), while the IQR of the contributions by particulate BrC is 6 %–13 %
(13 %–20 % at the rural site during winter). Future studies that will
directly benefit from these results include (a) optical modelling aiming at
understanding the absorption profiles of a complex aerosol composed of BrC,
BC and lensing-inducing coatings; (b) source apportionment aiming at
understanding the sources of BC and BrC from the aerosol absorption
profiles; (c) global modelling aiming at quantifying the most important
aerosol absorbers.


Similar to Corbin et al. (2019), MWAA measurements were carried out also upon extraction (using small petri dishes) of 4 Magadino 2014 and 5 Zurich 2013 PM2.5 filters in 5 mL water and then in 5 mL methanol (1 h, 70 rpm). Results were blank-subtracted in all cases. The repeats of original or water-washed punch measurements showed very good reproducibility in the AAE values, but not always accurate in terms of absolute absorbance. This is likely due to mechanical removal of insoluble particles, as we could confirm through SEM analysis of evaporated aerosol extracts (Fig. S13). Therefore, here we used only the AAE values as a measure of the wavelength dependence of the remaining aerosol, under the assumption that particle mechanical removal with extraction is not selective to certain insoluble aerosol.

S3.2: Absorption of aerosol extracts
The experimental setup and detailed analytical protocol for absorption measurement in extracts were described in our earlier publication (see SI of Moschos et al., 2018). Briefly, attenuation spectra of aerosol extracts in ultrapure water or methanol were measured from 280 nm to 600 nm using a UV-visible spectrophotometer (Ocean Optics) coupled to long-path detection cell (length l = 50 cm). The instrument records the light attenuation in solution, ATN sol 10 (λ), for a sample j as a function of wavelength, as the decadic logarithm of the ratio of signal intensities of the reference (solvent, I0) and the sample (I), both corrected for background S2 signals with the light source off. From ATN sol 10 (λ), the absorption coefficient in solution (in Mm -1 ), = Abs UV−vis, (λ), can be quantified (Hecobian et al., 2010) based on the Lambert-Beer law (Eq. (S3)), assuming that additional light scattering by the solute can be neglected upon dissolution: where Vair and sol are the volumes of air sampled through the filter (punch) and solvent used during the extraction (both in m 3 ), respectively. In this work, when the solvent was methanol, the extract volume was 3 or 6 mL and PTFE syringe filters of same dimensions (0.45 μm) were used (0.22 vs. 0.45 μm filtering did not have any noticeable effect on the recorded spectra).

S4. UV/Vis-PMF applications
We performed an exploratory analysis (Table S1) to determine which factors from each AMS-PMF solution drive the measured absorbance in methanol. This analysis was based on the correlation between the factor mass and the absorbance, for the full dataset of each factor solution, and for selected time points where each factor dominates in terms of mass compared to other absorbing fractions. The outcome of this analysis pointed to the inclusion of all OA factors in the UV/Vis-PMF model, except for SCOA and PBOA that contribute significantly to the coarse-mode (PM10-PM2.5) mass. Also, the absorbance of the coarse fraction in Magadino during 2014 contributed significantly less to PM10 absorbance than its respective contribution to PM10 mass (Fig. S8). HOA, COA, fOOA and SOOA were combined into a single (for each AMS-PMF solution) "other OA" factor before the UV/Vis-PMF analysis, because initial model runs showed that their individual contributions to absorbance were consistently lower than those of BBOA and WOOA. It was also confirmed, by considering all OA factors in the model, that the scaled residuals were not further reduced by including SCOA and PBOA (Fig. S9), which further demonstrated that these fractions did not contribute significantly to the absorbance of the PM10 aerosol fraction. UV/Vis-PMF (Moschos et al., 2018) was then constrained using three factors (BBOA, WOOA, other OA) and was applied to both methanol-soluble ( MeOH, (λ)) and water-insoluble ( MeOH, (λ) − H2O, (λ)) absorbance matrices, where H2O, (λ) was corrected for the water/methanol solvent effect (Fig. S5). The application of UV/Vis-PMF on the water-insoluble absorbance data showed that, unlike BBOA, WOOA and other OA did not explain a quantifiable portion of the absorbance (too low or noisy contributions). For example, for case (g) mentioned in this section, water-insoluble BBOA could explain the totality of the input water-insoluble absorbance at 370 nm. For the calculation of k(λ) for methanol-soluble BBOA and water-insoluble (but methanol-soluble) BBOA, we multiplied the total AMS-PMF BBOA mass with 0.93 to account for the average extraction efficiency of BBOA in methanol (Fig. S6). Also, the estimated solubility of WOOA and BBOA in water was up to 90 % and 66 %, respectively, based on the calculated recoveries by AMS-PMF analysis (Daellenbach et al., 2017;Vlachou et al., 2018).
WOOA can thus be considered fully water-soluble. We examined the uncertainties in (λ) (or (λ)) by varying the sample size of the input matrices, using the respective factor mass concentrations from the offline AMS-PMF Solutions 1 and 2. Specifically, the spatial coverage and temporal coverage or resolution were reduced compared to the full dataset runs for each PM10 AMS-PMF factor solution, as follows (the retrieved factor-specific k at 370 nm from the different sensitivity runs is shown in Table S2

S5. Calculation of MAC/MAE370nm
The factor contributions to particulate BrC absorption were obtained by considering the following cases for the source-specific We considered an OOA particle size of 200-400 nm ("larger size range") and a lower limit of ~120 nm for primary (non-OOA) BrC, based on available SMPS data for Zurich (Fig. S7). Fresh BBOA is known in certain cases to have smaller diameter, e.g. from SP2 measurements in Paris (Laborde et al., 2013). Therefore, we have considered the above cases where BBOA occurs in smaller sizes. HOA and COA were treated as externally mixed small size particles except for "case 5". Furthermore, past size distributions from Zurich (Wolf et al., 2017) indicate that (W)OOA most likely resides in the accumulation-mode. PBOA and SCOA were not considered in these cases, because these fractions reside mainly in the coarse-mode and thus are not expected to be internally mixed with other aerosol fractions.
For each case, we obtained the effective k, kmix, of each particle type, p (Eq. (S4)), by using the median wavelength-dependent (λ) of methanol-soluble WOOA, BBOA and other OA estimated by UV/Vis-PMF, multiplied with the mass fraction (f) of each aerosol component X in the different particle mixtures (volume mixing rule). The total BBOA mass was assumed to have similar absorptivity to the methanol-soluble BBOA.
We converted the resulting mix, (λ) to mass absorption cross-sections (MAC) at four AE33 wavelengths (370 nm, 470 nm, 520 nm, 590 nm) with Mie code programmed in the software package Igor Pro (WaveMetrics). The resulting mass absorption crosssections of each particle type, MAC ( , ), were multiplied with the total mass concentrations of its constituents to obtain the respective absorption coefficients, abs Mie, BrC−p (λ), at the different AE33 wavelengths, using Eq. (S5): The retrieved abs Mie, BrC− (λ) values for each particle type were summed to obtain the total absorption coefficient from all BrC-  S1. Exploratory analysis for AMS-PMF factor selection in the UV/Vis-PMF model. The individual factors eventually considered to constrain the model were based on specific correlation criteria; no fully transparent OA fractions were included in the model, i.e. R 2 between factor mass time series and bulk methanol solution absorbance (PM10 or coarse) time series < 0.4, for selected time points where each OA factor (PM10 or coarse) mass dominates compared to that of other absorbing factors (coarse data existed only for data from 2014, so the correlations with the coarse fraction could be tested only for Solution 1; only positive values for both the coarse fraction mass and absorbance were used). Note that the Pearson's r was positive in all cases except for the coarse PBOA (the respective R 2 value are indicated with an asterisk).

Factor
Mass ratio criterion    2) and each factor mass from the respective AMS-PMF factor solution. Note that for "case 5" (fully internally mixed BBOA + OOA + inorganics), the conversion factor is identical for all absorbing AMS-PMF source components. The min. and max. (from the seven cases) MAC/MAE were used to convert the UV/Vis-PMF-based median k (Fig. 2) Table S4. Different proxy results tested for the determination of bare BC MAC at 660 nm for three different cases, following a stepwise multiple linear regression approach. Only data points in the linear proxy (NR-PM:EC) range were considered for the linear fits. (OOA+BBOA+m-SIA):EC was the most consistent of the tested proxies for the three cases, in terms of the obtained MAC for bare BC, slope and Pearson's r, indicating that OOA, BBOA and inorganics likely partitioned to coatings on EC leading to its absorption enhancement by acting as a lens.  Figure S1. Flowsheet of how to obtain the various (mass) absorption coefficients, EC mass concentration and other aerosol properties from the raw filter-based data delivered by different types of instruments and methods. These were then used to obtain multi-wavelength optical closure for the different absorbing species.  . Calculated absorbance for a winter and summer filter extracted in different solvents. Acetone and acetonitrile have a larger dipole moment than methanol, no hydrogen bonds and can dissolve charged species. An absorption hump around 470 nm was observed at longer wavelengths for water and acetone. Acetone, which is less polar than methanol, provided comparable results but had the disadvantage of a cut-off wavelength of 330 nm, whereas the absorbance in water was significantly lower than in methanol for the Magadino winter sample. The absorbance in dichloromethane (the least polar of the tested solvents) was negligible for the summer sample. Methanol was selected in this study for optical closure purposes considering the obtained absorbance in the full wavelength range, as well as the extraction efficiency of BBOA-dominated samples in methanol (Fig. S6). Figure S4. Linearity of ATN sol 10 obtained at three different wavelengths for a Magadino 2014 PM2.5 filter sample extracted in methanol three times at five different concentrations, covering the full range of our solvent extraction (x axis) and UV-vis measurement (y axis) conditions. Figure S5. Assessment of the water vs. methanol solvent effect on UV-vis measurements, performed by comparing the attenuation spectra for ambient PM extracted in water and then diluted in methanol or in water in a 10/90 ratio. The blue curve and grey shading indicate the average and 1 SD, respectively, from 5 Magadino 2014 PM2.5 samples. The ratio at 280-470 nm was higher than that calculated in previous study (Phillips and Smith, 2017), whereas above 470 nm we considered a ratio of 1.0 because the signal was noisy. The water-insoluble (but methanol-soluble) BrC absorbance used in UV/Vis-PMF, = − , was corrected for the solvent effect by multiplying with the average wavelength-dependent absorption ratio below 470 nm.     Table S1. The Q/Qexp values for the 3-factor (BBOA, WOOA, Comb) solutions were 5.9, 2.6 and 4.8 for Magadino winter, Magadino summer and Zurich, respectively. Figure S10. Upper panels: highly spectrally resolved cumulative relative contributions to absorbance of methanol-extracted PM10 aerosol, apportioned by UV/Vis-PMF analysis (Moschos et al., 2018) to various OA source components resolved by offline aerosol mass spectrometry (Fig. S2)  Figure S12. a) Correlation between the average time resolved MACBrC,370nm and the EC-to-BrC mass ratio for ambient filter samples from Magadino and Zurich in both seasons. BrC is considered as the total OA without the contributions of the non-absorbing SCOA and PBOA (i.e. BBOA + WOOA + other OA). No significant trend is observed, which limits the applicability of the respective parameterisation (Saleh et al., 2014) to ambient (mixed-source, aged and processed) samples. b) Same as (a) but with MAE values.

NR-PM
(a) (b) (a) (b) Figure S13. FE-SEM-EDS images of selected untreated filters from Zurich (a and b) and Magadino (c and d) winter. Carbon assemblies exhibit scarce adhesion onto the filter fibres and accumulate onto deeper, thinner and rougher fibres and their intersections (a). While the only spherical particles observed in untreated Zurich samples were non-carbonaceous, either Fe-bearing or containing K, Mg, Ca, Al and S (b), in Magadino both bare -including both lacey and compact/collapsed-BC (c) and drop-shape bigger pseudo-spherical (likely deformed upon collection on the filter) (d) carbonaceous particles (with traces of K and S, based on EDS analysis) were observed, the latter disappearing after washing with water (e; in-depth magnified view), as observed in many different fields. Note that most of the organic particles/coatings, especially the more volatile ones, have not survived under vacuum and thus are not observable with this technique. f) Alumina filter (Anodisc 25 Whatman, pore size 0.1 μm), on which the supernatant was deposited after sonication of the original quartz microfibre filter in water (conditions: carbon coated, high vacuum; voltage: 10 kV, signal: SE in-lens), and g) polycarbonate filter (isopore membrane track-etched, pore size 0.2 μm), on which the supernatant was filtered after water washing of the original quartz filter, showing that the washing with water seemed to remove (mechanically) a noticeable quantity of carbon nanoparticle aggregates, whereas quartz fibres were also detached.
(a) (b) (c) (d) (e) (f) (g) Figure S14. Calculation of MACbareBC as the intercept of uncertainty-weighted linear fits (95 % confidence interval for the slope and intercept shown) of MACBC,660nm vs. the selected proxy for BC coating thickness, upon obtaining no significant evidence of a source-dependent MACBC (Fig. 4b).