New In Situ Aerosol Hyperspectral Optical Measurements over 300-700 nm , Part 2 : Extinction , Total Absorption , Water-and Methanol-soluble Absorption observed during the KORUS-OC cruise

1National Institute of Aerospace, Hampton, Virginia, United States of America 2NASA Langley Research Center, Hampton, Virginia, United States of America 3NASA Goddard Space Flight Center, Greenbelt, Maryland, United States of America 10 4City University of New York, New York, New York, United States of America 5Science Systems and Applications Inc., Lanham, Maryland, United States of America 6Science Systems and Applications Inc., Hampton, Virginia, United States of America 7California State University, San Bernardino, California, United States of America 8Springfield College, Springfield, Massachusetts, United States of America 15

UV-Vis-IR range and may not exhibit absorption in the Vis-IR at all. Non-carbonaceous aerosols, such as dust, are also known to weakly absorb light as a function of their composition, principally according to their proportion of iron oxides (e.g., Sokolik andToon, 1999: Moosmüller et al., 2012;Shi et al., 2012). Two forms of iron oxide, hematite and goethite, exhibit very different wavelength dependence in their absorption spectra (Schuster et al., 2016).

100
Given the importance of composition on the absorption of aerosols, it is not uncommon to treat sabs as a function of composition, while treating sscat (and by extension sext, since scattering is the dominant term in extinction) as a function of aerosol size distribution. This separation is not strictly true as mixing state and size distributions have been shown to influence the spectral behavior of sabs (Schuster et al., 2016), while absorption has been shown to influence the spectral behavior of sext (Eck et al., 2001). Hence, the ability to simultaneously measure hyperspectral sabs and sext of in situ aerosols along with their chemical 105 composition and size distributions is anticipated to improve our understanding of the interaction of differing spectral dependencies due to various microphysical aspects of ambient aerosols.
For KORUS-OC, the in situ aerosol measurement package did not include size distribution measurements or on-line composition, however, there is extensive information about the ambient aerosols observed throughout the KORUS-AQ campaign that can be 110 used to assess how the ambient aerosol sizes and composition influenced the observed spectral characteristics observed aboard the R/V Onnuri. As will be discussed in Sects. 2.1 and 3.2, there were three distinct synoptic meteorological periods during the KORUS-OC cruise  that gave rise to aerosol populations that differed in size and composition (Jordan et al., 2020a;Jordan et al., 2020b). These different ambient populations provide the framework that will be used here to assess shifts in the optical properties related to size distribution (sext, following the discussion in Part 1) and composition (sabs, sDI-abs, and sMeOH-115 abs).

Ship deployment and high-frequency, ambient aerosol measurements
A detailed description of the ship deployment and the high frequency measurements made as part of the in situ aerosol package deployed on the Korean Institute of Ocean Sciences and Technology (KIOST) R/V Onnuri is provided in the companion paper 120 (Jordan et al., 2020b). In brief, the KORUS-OC cruise of the R/V Onnuri sailed first along the east coast of South Korea, then transited to the west. The in situ aerosol measurement package measured fine mode aerosols with a 50% size cut of 1.3 µm diameter. Filter samples were collected over 3 hour daytime intervals, along with a 12 hour overnight sample, each day from May 22 nd to June 4 th (local Korea Standard Time, KST, = UTC + 9) outside of the South Korea's territorial waters (> 12 nautical miles, 22.2 km, from the coast) with mean sampling locations shown in Fig. 1 (see Table S1 for details as some filter samples overlap in 125 the figure). The locations shown in Fig. 1 are color coded according to the meteorological regimes during the cruise as described by Peterson et al. (2019): Stagnant (May 17 th -22 nd ), Transport/Haze (May 25 th -31 st ), and Blocking (June 1 st -7 th ) with periods nominally defined from midnight (00:00) through midnight (24:00) local time. Note, in Fig. 1, there is a group of filter samples classified as "transition" collected on May 23 rd and 24 th as the Stagnant period meteorological conditions gave way to those that defined the Transport/Haze period. Following the discussion in the companion paper, the Transport/Haze filter set is divided into 130 two subsets, those downwind along the east coast and those upwind to the west of the peninsula with the one sample along the https://doi.org/10.5194/amt-2020-318 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License. south coast assigned to the west group. Details on the specific meteorological conditions and aerosol characteristics will be discussed in Sect. 3.2.
Means for each filter sampling interval (Table S1) were calculated from high temporal resolution data sets described in Jordan et 135 al. (2020b): 3 visible wavelength scattering (450, 532, and 632 nm sscat, AirPhoton IN101 nephelometer, Baltimore, MD) and absorption (467,528,and 652 nm sabs,Tricolor Absorption Photometer,TAP,Brechtel,Hayward,CA), and hyperspectral sext (300-700 nm) from the custom built SpEx instrument. Filter mean sext calculated at 450, 532, and 632 nm (see Jordan et al., 2020b for details) from the sscat and sabs data is denoted NT sext here, in contrast to SpEx sext. The high-resolution data set was filtered to remove interceptions of the R/V Onnuri's own ship stack emissions (Jordan et al., 2020b). Similarly, here it was also used to 140 assess potential ship plume contamination of the filter set (see the discussion in the supplementary information). Of the 53 pairs of filters collected, no ship plume interceptions occurred for 15. The degree of contamination depended on the duration of the interception as well as on the aerosol property of interest (e.g., Fig. S1). Here, single scattering albedo calculated from IN101 and TAP data provided the most conservative delineation between ambient and ship plume aerosol (Figs. S1 and S2) resulting in 13 of the filter pairs being rejected (Table S2) from most of the analyses presented in Sect. 3 (except as noted). These data were retained 145 for the assessment of the correction factor for the spectral absorption measurement discussed in Sect. 2.3.

Filter preparation and sampling procedures
Two sets of filters were collected for subsequent spectral analyses in the laboratory: Glass fiber filters (Whatman GF/F 47 mm filters 0.7 µm pore size) for total absorption spectra (sabs) measured in an integrating sphere attached to a spectrophotometer (see Sect. 2.3 below) and Teflon filters (Fluoropore PTFE 47 mm filters 1 µm pore size) for soluble absorption spectra from deionized 150 water (sDI-abs) and methanol (sMeOH-abs) extracts from the filters (see Sect. 2.4 below). For each filter set, a blank filter was collected each day during the cruise by briefly putting it in the sampling line without any air flow through the system. The Teflon filters were taken directly from their package and placed in a filter cassette for sampling. GF/F filters were pre-baked at 450°C for 6 hours in the laboratory prior to the cruise. They were then individually stored in polystyrene petri dishes sealed 155 with Teflon tape, wrapped in aluminum foil, and enclosed in Ziploc bags until placed in a filter cassette for sampling aboard ship.
The two filter cassettes sampled in parallel off the common inlet used for the measurement suite. The flow rates through the Teflon and GF/F filters were ~24 and ~21 lpm, respectively. After sampling, filters were individually placed in polystyrene petri dishes, sealed with Teflon tape, wrapped in aluminum foil to limit exposure to light, placed in Ziploc bags, and kept frozen until they were analyzed in the laboratory.

Total absorption coefficient spectra from GF/F samples
Total absorption coefficient (sabs) spectra from 300-700 nm at 0.2 nm spectral resolution were obtained from the GF/F set by placing each filter in the center of an integrating sphere (Labsphere DRA-CA-30) attached to a dual beam spectrophotometer (Cary 100 Bio UV-Visible Spectrophotometer). This methodology is well established in the OC community for obtaining particulate sabs spectra from filtered water samples (e.g., Stramski et al., 2015;Neeley et al., 2015;IOCCG Protocol Series, 2018). The GF/F 165 set were collected during KORUS-OC to test the applicability of this approach to atmospheric aerosol samples. https://doi.org/10.5194/amt-2020-318 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.
The absorbance spectra of the GF/F filter set were measured in the laboratory over the course of 4 days with regular air scans (spectra obtained with the filter holder in the integrating sphere without a filter in place) measured to account for any drift in the spectrophotometer measurement. Absorbance spectra from field blank filters were consistent with laboratory blank filters 170 indicating that handling in the field did not contribute to the measured absorbance of the sample filters. Blank corrections for the sample filters were calculated by taking the mean of the air scans and subtracting it from the mean of the field blank spectra.
Multiple scans (2-6) for each filter sample were performed, rotating the filter position between scans to assess variability in the measurement by sampling different parts of the filter with the narrow, collimated beam. The mean spectrum was then blank corrected. Two lamps are used in the spectrophotometer to cover the full spectral range. The UV portion (< 370 nm) was noisy 175 at the measurement resolution of 0.2 nm, so a 10 nm boxcar smoothing algorithm was used in processing the data set to eliminate that noise. The lower limit of detection (LLOD) for the sample data set was calculated from the mean + 3 standard deviations of the field blank filters. Measurement error for the absorbance scans was estimated to be about 15%.
The dimensionless blank-corrected absorbance (Abs, also known as the spectral optical density of the filter (Stramski et al., 2015)) 180 is then used to calculate the sabs via where fa is the area of the filter (in m 2 ), Vair is the volume of air sampled through that filter area (in m 3 ), b is a term used to correct for pathlength amplification of the signal by the filter and collected particles (Butler, 1962;Kiefer and Soo Hoo, 1982;Kishino et al., 1985), and the ln(10) term converts the base 10 logarithm absorbance measurement (= -log10 T, where T is transmittance) to a 185 natural logarithm as used for ambient environmental observations. These terms result in units of m -1 for sabs(l), which are inconvenient for atmospheric data. Hence, an additional factor of 10 6 is also used to convert the units to Mm -1 for the data set herein. The b correction accounts for the scattering enhancement by the fibers within the filter itself, as well as by the particles collected on and within the filter medium and is defined as the ratio of the measured filter absorbance to the true absorbance of the sample.

190
b has been determined empirically with the standard protocol (IOCCG Protocol Series, 2018) adopting the parameterization determined by Stramski et al. (2015), denoted here as bs, The ocean optics and biogeochemistry protocol (IOCCG Protocol Series, 2018) notes that "understanding ... the types of particle 195 assemblages for which the formulated pathlength amplification correction is representative" is important. Here, those assemblages included nearshore mineral-dominated and red tide samples, mixtures of phytoplankton species, and other coastal and offshore particle assemblages spanning a size range of about 0.7 to 50 µm (Stramski et al., 2015). A similar study that used a different instrument configuration to measure the absorption of particles filtered from snow and ice meltwater (Grenfell et al., 2011) noted that the calibration standard used for soot was carefully filtered to remove particles that were too large to be representative of the 200 ambient samples. Figure S3 shows the relationship between sabs calculated using Eqs. (1) and (2) from the filter measurements in the integrating sphere compared to sabs measured by TAP (values at 467, 528, and 652 nm wavelengths are all plotted together here). Using no correction at all (b = 1) results in values much higher than the TAP values (with a slope of 2.361), while the bs correction leads to values that are much lower (slope of 0.593).
Fresh soot particles from diesel engines are small, typically on the order of tens of nm (Bond et al., 2013), hence the size range measured by the in situ aerosol package aboard R/V Onnuri (≤ 1.3 µm diameter) likely included smaller particles (as well as particles with a very different composition) than those used to derive bs. Given that TAP is a commercial instrument with a well characterized correction (Ogren et al., 2017) suitable for atmospheric BC aerosols, along with the known importance of using a b correction determined from particles representative of the sample data, we have greater confidence in the magnitude of the TAP 210 values than the bs correction for this data set. Empirically determining an appropriate b for the samples obtained in this study was beyond the scope of this project. However, taking advantage of the TAP data set, b for the KORUS-OC data (bK-OC) was obtained by scaling bs to fit the TAP data resulting in quantitative sabs spectra. Error propagation was used to estimate the uncertainty in the sabs spectra (Eq. (1)) from the estimated 15% error of the absorbance spectra.

215
The spectrometers used to measure sext and sabs had different spectral resolutions. Hence, in order to calculate single scattering albedo (w) these spectra sets were averaged into 2 nm bins over 300-700 nm. Then spectral w was calculated via

Soluble absorption coefficient spectra from Teflon samples
The Teflon filters were cut in half in order to extract water-soluble aerosol components from one half that could be further analyzed 220 via ion chromatography and aerosol mass spectrometry (see Sect. 2.5), and non-water-soluble components from the other half using an organic solvent. Here, methanol (MeOH) was used for the organic solvent as both water and MeOH solutions were compatible with the waveguide used to measure the absorbance of the dissolved aerosol components in the extracts. Half of the filter was placed in a clean 15 ml polypropylene centrifuge tube (Corning 430052; triple rinsed, soaked overnight, and triple rinsed again with Milli-Q 18 MW deionized (DI) water) and extracted in 10 ml of DI water by hand shaking the tube for 60 s. Tests were 225 performed that showed this method extracted the soluble chromophores equally well as with 60 min of sonication, hence the faster approach was employed. Extracts were filtered to remove any insoluble particles using polypropylene Soft-Ject disposable 12 ml syringes (Henke Sass Wolf) and 0.2 µm pore size PTFE-membrane filters (Cole-Parmer).
The other half of the filter was placed in a clean 15 ml glass vial with a Teflon cap (rinsed with spectrophotometric grade (≥ 99.9%)

230
MeOH (Sigma Aldrich product number 154903) and dried in a fume hood) and extracted in 10 ml of MeOH by hand shaking the vial for 60 s. Extracts were filtered to remove particles using a glass syringe with a 0.2 µm pore size PTFE-membrane filter (Cole-Parmer). Unfortunately, some of the MeOH extracts vials were contaminated during handling and are marked as "missing" in Table S1. As a result, there are fewer MeOH-soluble absorption spectra in the data set than DI-soluble absorption spectra. In all cases where there are both spectra, greater absorption was observed from the MeOH extracts.

235
A liquid waveguide capillary cell (LWCC; World Precision Instruments LWCC-3100, 100 cm pathlength) attached to an SM240 spectrometer (Spectral Products, Putnam, CT, ~0.4 nm spectral resolution) with a DH-2000-BAL light source (Ocean Optics, Dunedin, FL) operated with a 2 s integration time was used to measure the absorbance of the soluble aerosol chromophores in each filter extract. Note, absorbance (Abs) is the measured quantity, it is not the same as the absorption coefficient (ssol-abs, where the 240 notation sol-abs is used to distinguish soluble from total absorption coefficients). ssol-abs, the relevant atmospheric quantity, is derived from Abs. The relationship between these values is explained below. Each set of measurements in the laboratory started https://doi.org/10.5194/amt-2020-318 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License. with a dark count spectrum followed by an alternating sequence of a reference spectrum from the solvent alone, then a sample spectrum from a filter extract, then another reference spectrum, etc. Periodically, the LWCC would be cleaned with 0.5 N HCl and acetone (for HPLC, ≥ 99.9%, Sigma Aldrich product number 270725).

245
The Abs of the dissolved materials in the extract measured in the LWCC was calculated via where, T is the transmittance, i.e., the fraction of the total light (I0) that passes through the sample (IS). To account for any drift in the light intensity from the lamp, the reference intensity (I0) before (I01) and after (I02) each sample (IS) was averaged and used as 250 the reference for that sample. Correcting each term for the dark counts of the spectrometer (ID), Eq. (4) can then be expressed as (wavelength dependence not shown for clarity) For each extract the intensity spectra were acquired for 40 -60 s. The mean and its standard deviation for each ID, I01, I02, and IS were calculated. The uncertainty for each absorbance spectrum was obtained using error propagation from the standard deviations 255 of the intensity spectra. Extracts from field blank filters were also measured and used to calculate the lower limit of detection (LLOD) in absorbance (using the mean + 3 standard deviations of the blanks). The LLOD is wavelength-dependent and was found to be higher for the MeOH set than the DI set. In contrast, the upper limit of detection (ULOD) was constant and found to be 0.8 Abs for both solvents.

260
ssol-abs for the soluble aerosol chromophores in the extracts were then calculated from the blank-corrected Abs spectra via where Vs is the volume of solvent used to extract the filter (in liters), Va is the volume of air sampled by the filter (in liters), l is the absorbing path length of the capillary cell (in meters), and ln(10) converts the base 10 log used for absorbance to natural log typically used for atmospheric quantities (Hecobian et al., 2010). Again, the units were converted from m -1 to Mm -1 per the 265 convention commonly used for atmospheric data sets. ssol-abs spectra are reported for the wavelength range of 300-700 nm with problematic wavelengths (due to saturation of the detector) removed resulting in evident gaps in the spectra (e.g., Fig. 2). The uncertainty in ssol-abs(l) was assessed using error propagation and found to be about ± 30%, principally due to the uncertainty in the volume of the solvent used to extract the filters.

270
Previous authors (e.g. Hecobian et al., 2010;Zhang et al., 2013) have calculated ssol-abs(l) at 365 nm (ssol-abs(365nm)) by averaging over the 360-370 nm range of the spectrum in order to use it as a single value proxy for soluble organic aerosol chromophores.
This avoids contributions from inorganic nitrate absorption that occurs at wavelengths < 330 nm. We adopt this proxy for the discussion herein.
specifically an HR-ToF-AMS, Aerodyne Research, Inc., Billerica, MA). Lower limits of detection for the IC data set were calculated from the mean plus 3 times its standard deviation of the field blanks. Most of the below detection IC data were among the sea salt and dust ions (Na + , Cl -, Mg 2+ , and Ca 2+ ), with a few occurrences found among NO3 -, K + , and C2O4 2-. No samples were 280 below detection in NH4 + and SO4 2-. These results are consistent with the expectation that anthropogenic pollution sources dominate the fine fraction of aerosols, the population sampled by this measurement suite with a 50% size cut of 1.3 µm diameter (Jordan et al., 2020b). No sea salt corrections were applied to any of the reported concentrations.
The water-soluble extract was aerosolized using a nebulizer supplied with particle-free air and subsequently dried followed by 285 sampling with the AMS. The mass spectrum measured by the AMS reflects the composition of the low volatility (e.g. particulate) but non-refractory, water-soluble fraction of components collected on the filter. Unlike the IC data, it is not possible to quantify the atmospheric mass concentrations of the aerosol components measured here with the AMS. The nebulizers used to disperse the DI extracts into aerosols do so with an unknown liquid flow rate such that the original atmospheric aerosol mass concentrations in ambient air cannot be calculated. Hence, in order to make comparisons across the data set, the AMS data is presented in terms of 290 ratios, either the ratio of the major chemical groups (sulfate, ammonium, nitrate, chloride, and organic) to their summed total mass (Total) or the ratio of individual m/z groups to the total organic mass (Organics). Note, since organic sulfate and nitrate compounds can contribute to the sulfate and nitrate measured by AMS the notation used for ions is not used here. Throughout this paper the overlapping IC and AMS measurements are distinguished by using the ionic notation for the former (i.e., SO4 2and NO3 -) and the terms (i.e., sulfate and nitrate) for the latter.

295
The summation over all m/z ≥ 12 used to calculate Organics included negative values that arise from the subtraction of a reference spectrum (filtered airflow to remove the particles) from the unfiltered airflow containing aerosols in the AMS. Typically, the negative values were a minor contribution. However, in some cases they were large. Hence, the fractional contributions of each m/z to the sum (denoted here as "f_m/z") were normalized across only positive values. The normalized value was used to 300 approximate above detection contributions for each m/z to Organics. Note, in the assessment of individual contributions to Organics the range of values examined was truncated to the m/z 12-73 range. For all but 6 filters this captured ≥ 0.9 of Organics.

Filter spectra set overview
Each panel of 2 shows one filter from each meteorological period as shown in Fig. 1  Key features to note in 2 include the spectral variability in sabs, where sometimes it is smoothly varying (e.g., Blocking) and 315 sometimes there are spectral features evident (e.g., the UV portion of the Stagnant spectrum). Also note the changing relationship between sMeOH-abs and sDI-abs, where they can be very similar across the spectrum (e.g., Transport/Haze -East) or diverge substantially at longer wavelengths (e.g., Blocking). Finally, note the partial spectra for sMeOH-abs and sDI-abs, where only a portion of the spectrum is above detection. These partial spectra vary considerably in the above detection wavelength range with the Transport/Haze -West example showing an extreme case where all of the sMeOH-abs is above detection, while little of the sDI-abs is 320 above detection. In contrast, for the Stagnant example both of these spectra are partial, but span most of the wavelength range.
The implications of the spectral variability will be discussed further in subsequent sections.

KORUS-OC
The synoptic meteorology of the KORUS-AQ campaign described by Peterson et al. (2019)   began following a frontal passage that brought clean air from the north. As discussed in the companion paper (Jordan et al., 2020b), the fine mode aerosols sampled aboard R/V Onnuri during this period were likely due to small absorbing aerosols from ship emissions in the West Sea.

340
The aerosol composition measured via the AMS from the nebulized water-soluble filter extracts (Fig. 3)  The upwind vs. downwind shift is also evident in the IC analyses of these DI extracts (Fig. 4). NO3was up to an order of magnitude 355 greater downwind than upwind during Transport/Haze, while peak SO4 2nearly doubled, consistent with previous reports of local South Korean heterogeneous secondary production of these constituents (Jordan et al., 2020a). Throughout the cruise, SO4 2was generally fully neutralized by NH4 + , i.e. in the form (NH4)2SO4 (Fig. 4). K + was linearly related to NH4 + throughout the cruise with an r 2 = 0.78 (Fig. 4). In contrast, C2O4 2measured by IC exhibited elevated values both for the Stagnant (and transition) and downwind Transport/Haze samples (Fig. 4). Organic aerosol was elevated during the Transport/Haze period (Jordan et al., 2020a),

360
but it did not increase as rapidly or dramatically as SIA, hence the decreased percent contribution to the total aerosol mass in the AMS samples (Fig. 3). During the Blocking period organics again dominated the composition but not to the extent observed during the Stagnant period (64% vs. 85%). For all aerosol components (organics and inorganics) concentrations during this period tended to be relatively low (Fig. 4).
This is not unexpected as extinction is related to aerosol mass loading and size distributions, while soluble absorption is a function 370 of soluble aerosol composition. Total aerosol absorption was not well correlated with any of the water-soluble aerosol components (as expected, since BC dominates total absorption in the atmosphere and is not soluble). However, as shown in the companion paper, peak sabs tended to coincide with peak sext leading to moderate r 2 values of ~0.5 depending on wavelength (Fig. 5). This is likely a reflection of mass loading, such that when pollutant aerosol concentrations downwind of South Korea are high, both scatterers and BC are elevated in the air mass. Single scattering albedo clarifies an interesting feature evident in sabs, i.e., the total 375 absorption is sensitive to organic absorbers at 365 nm (i.e., BrC). There is little difference in w(365 nm) and w(532 nm) for the Transport/Haze and Blocking periods, however, there is a substantial reduction in w(365 nm) for the Stagnant (and transition) period samples (Fig. 5).
This is a striking observation due to the fact that the Stagnant period aerosols were dominated by local photochemically produced 380 SOA. K + mass concentrations were relatively low during this period, while elevated C2O4 2concentrations at this time were independent of K + (Fig. 4). Both K + and C2O4 2have known biomass burning sources and have been used as tracers for biomass burning (e.g., Park et al., 2013;Park and Yu, 2016;Szép et al., 2018;Johansen, et al., 2019). However, they also have other sources that can complicate their usage for such a purpose. For example, C2O4 2can be formed by aqueous-phase cloud chemical processes (e.g., Huang et al., 2006;Ervens et al., 2011) and fine fraction K + can arise from fertilizer use (e.g., Szép et al., 2018;  Levoglucosan has also been used as a biomass burning tracer, however, its relative contribution has been shown to decrease with smoke plume age (Cubision et al., 2011) and recent work indicates that it is shortlived in the atmosphere such that it is useful for fresh emissions, but not for aged biomass burning aerosol (Wong et al., 2019). 3 390 shows m/z 60 from the nebulized water-soluble extracts was negligible throughout the campaign. m/z 60 is typically attributed to the C2H4O2 + fragment from levoglucosan (Schneider et al, 2006) and is enhanced for the biomass burning organic aerosol PMF factor (Cubison et al., 2011). All of these data suggest that during the Stagnant period photochemically produced SOA contributed https://doi.org/10.5194/amt-2020-318 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.
to absorption at 365 nm without any evident indication of biomass burning. However, the Wong et al. study also reported that absorbing high molecular weight biomass burning aerosol is relatively stable and much longer lived in the atmosphere than low 395 molecular weight biomass burning aerosol components such as levoglucosan. The implications of this will be discussed further in Sect. 3.3.
The difference between w(365 nm) and w(532 nm) for the absorbing aerosols of the Stagnant period and Transport/Haze period show that the elevated C2O4 2and elevated sDI-abs do not fully capture differences across these populations of absorbing organic 400 aerosols. That is, decreases in w from 532 to 365 nm are small for the Transport/Haze population, in contrast to the large decrease for the Stagnant (and transition) aerosols. The molecular structures that absorb light differ between the two groups such that the photochemistry of the Stagnant period appears to have produced molecularly different SOA than was produced under the cloudy hazy conditions of the Transport/Haze period.

405
It must be emphasized that the chemical contributions discussed in this section only include aerosol components that were soluble in water and hence, only represent a portion of the ambient chemical composition influencing the sext, sabs, and sMeOH-abs measurements. sDI-abs contributed only a fraction to sabs, and sabs were typically an order of magnitude smaller than sext. Hence, relationships that may be explored within this data set between chemical composition and optical properties are constrained by these limitations. Future work would benefit from concurrent in situ aerosol AMS sampling, as well as size distribution 410 measurements, to more fully relate aerosol composition to optical properties. Even so, chromophores may constitute only a small fraction of the aerosol mass (e.g., Bones et al., 2010). Hence, changes in sabs may or may not be tightly coupled to aerosol composition measurements. The development of in situ hyperspectral techniques to measure the optical properties of in situ aerosols, along with the chemical and microphysical properties of those aerosols is expected to reduce some of the uncertainties in assessing the relationships among the full suite of ambient aerosol characteristics.

Characterizing spectral shapes with 2 nd order polynomial coefficients
While isolating individual wavelengths from the spectral data can be informative as illustrated in Sect. 3.2, the anticipated value of these hyperspectral measurements comes from examining the full spectra. In Part 1 (Jordan et al., 2020b) it was shown that the logarithmically transformed extinction spectra were better fit with a 2 nd order polynomial than a line, due to curvature of the spectra in log space. Similarly, for the filter spectra sets 2 nd order polynomials, 420 ( ( )) = ( + ' ( ) + < ( ( )) < yielded better fits to the measured spectra than linear fits (i.e. representing a power law), of the spectra as shown by the improvement in the fit residuals (Fig. 6). An example set of spectra (Fig. S4) is provided in the SI to illustrate the difference in the residuals from these two fits. Note, that all three absorption spectra tended to exhibit positive 425 curvature in contrast to extinction spectra that are all negative for the filter mean spectra set (e.g., Fig. S4). In Part 1, it was shown that the expanded wavelength range in sext led to smaller values of aext due to the negative curvature of the spectra. Here, the positive curvature of the absorption spectra leads to the opposite result, i.e., larger values of aabs compared to those shown in Part be discussed further below.
The residuals from the 2 nd order polynomial and linear fits to each spectra set were binned into 20 bins spanning the full range of differences between each measured spectrum and fit spectrum (Fig. 6). The range of differences is smaller leading to narrower bins for the 2 nd order polynomial fits than the linear fits. The normal distribution around zero of the 2 nd order polynomial fits 435 shows they provide a better fit to the data. Following the approach used in the companion paper, the additional information that may be gained from the coefficients of the 2 nd order fits (a1 and a2, Eq. (7)) than from the Ångström exponent, a (= -b, Eq. (8)) provided by linear fits is examined here.
As discussed in detail in Part 1, the coefficients between the two fits are related to each other by the derivatives of Eqs. (7) and (8),

440
where, Here, the notation lch is used to denote that there is one wavelength for which a and (a1,a2) yield equivalent results for any given spectrum. -2ln(lch) can be thought of as the slope of a line that describes that relationship in (a1,a2) space. This is illustrated in Fig. S5, where for each spectra set lch was calculated from the pair of fits to each spectrum. In the case of sabs, only a very narrow 445 range of values was found such that lch = 0.47. Using this value in Eq. (9), one can map the set of parallel a lines into (a1,a2) space using the range values obtained for these coefficients from the two sets of fits. For any given a, spectra with differing curvature will map into different pairs of (a1,a2) along that line (Fig. S5). A plot of the individual a values obtained from the sabs set ( Fig.   7) shows those points are aligned according to the map in Fig. S5. Note, the color bars for a used in Fig. 7 match those in Fig. S5 (right panels). For sabs, the range of a is 1.08 to 2.80. These are larger values than the range shown in Part 1 from the visible 450 wavelength TAP data where aabs ~ 0.5 -1 (Jordan et al., 2020b). As noted above, positive curvature and a wavelength range extending further into the UV leads to the higher aabs values here. Previous studies have used aabs to distinguish absorbing aerosol composition (e.g., aabs = 1, fresh urban-industrial BC, ≥ 2, desert dust, with intermediate values indicative of biomass burning aerosols (Russell et al., 2010)). However, Schuster et al. (2016) caution that mixing state and size distribution play a role in aabs along with composition.

455
The (a1,a2) mapping can be used to examine the distribution of samples according to the meteorological regime (Fig. 8) as described in Sect. 3.2. It is interesting to note that for the Stagnant period aabs ≥ 2, suggesting desert dust, but that is not consistent with what is known about the ambient aerosol population that was dominated by fine mode SOA at the time. As discussed in Sect. 3.2, the high ratio of m/z 44:43 in the DI extracts for these samples suggest these aerosols were more photochemically aged OOA with 460 lower volatility organic components (Ng et al., 2010). This is a somewhat confounding observation given the significant production of SOA during this period. It is possible that non-BC contributions to sabs were decoupled from the composition of the bulk of the SOA mass concentrations as has been suggested by Bones et al. (2010). Another possibility could be photochemically produced chromophores among the SOA. Or as Schuster et al. (2016) suggest factors related to the mixing state may be important here.

470
While aabs values were lower, closer to expectations of BC dominated absorbing aerosol downwind of the peninsula during Transport/Haze (Figs. 7 and 8), there was not clear separation in fit coefficient values across the meteorological periods. The separation observed in the extinction coefficients ( Figs. 7 and 8) is less distinct than that of the hourly mean sext spectra fits reported in the companion paper. This may be due to the longer sampling intervals of the filter spectra set, which may also be a contributing 475 factor to less distinct separation in fit coefficients to sabs across meteorological regimes (Fig. 8). Note, the mapping of aext isn't strictly parallel in (a1,a2) space as it was for aabs, due to a wider range in lch values (= 0.41 -0.47, Fig. S5) that results in a fanned out spread of aext (Figs. S5 and 7).
A similar separation in the distribution of a for the soluble absorbers leads to what appear to be 2 distinct branches in (a1,a2) space 480 with lower a values clustered over a narrower range of (a1,a2) values and larger a values spanning a broad range in (a1,a2) extending to both positive and negative extremes in the observed a2 range (Fig. 7). This mapping arises from the relatively broad range of lch = 0.34 -0.47 for both sDI-abs and sMeOH-abs (Fig. S5). The differing ranges of lch across the spectra sets is due to the presence (or absence) of partial spectra, i.e., spectra for which longer wavelengths are below detection and hence, not included in the fit.
The fit coefficients are sensitive to the wavelength range of the fit, such that shorter spectra can exhibit more extreme values over 485 the (a1,a2) range. Further, the limited wavelength range shifts the value of lch to shorter wavelengths. For the individual measured sext spectra set reported in the companion paper lch ranged from 0.36 -0.46. Mean sext spectra over the filter sampling intervals narrowed and shifted that range to slightly longer wavelengths, 0.41 -0.47 as noted above. There were no partial spectra for sabs leading to lch = 0.47, while for the soluble absorbers there were many partial spectra as illustrated in Fig. 9. Here, all of the above detection sDI-abs at 0.315 µm are shown, with below detection values at each increment in wavelength across the spectrum absent 490 in subsequent panels (Fig. 9). The magnitude of the soluble absorption coefficients were not separated in (a1,a2) space ( Fig. 9), however, the a values and meteorological regimes (Figs. 7 and 8) exhibit clear separation. These results are consistent with expectations given the spectral dependence of absorption as a function of the number of conjugated bonds, heteroatoms, and functional groups present in the organic component of aerosols (see Sect. 1). Hence, it is not unexpected to find a variable range of wavelengths over which above detection soluble absorption is found for ambient particles. (a1,a2) mapping may therefore be 495 useful for relating optical properties to chromophores.
All of the Stagnant sDI-abs and sMeOH-abs spectra were partial, while none of the Transport/Haze -East were partial leading to separation in (a1,a2) space for these meteorological regimes (Fig. 8) (Fig. 7) within a narrow range of (a1,a2), but there were a few upwind samples with high aDI-abs values on the partial spectra branch of the (a1,a2) distribution (Figs. 7 and 8). All of the A detailed analysis of the organic composition of the DI extracts and the (a1,a2) mapping of sDI-abs is beyond the scope of this work, but the evident separation in the maps shown here suggest future studies to relate composition to sDI-abs may benefit from this approach.
3.4 Variability and spectral structure of w(l)

510
While 2 nd order polynomial coefficients provide more information than linear fits, additional spectral structure is evident in the complete spectra that cannot be captured by simple fits as in Sect. 3.3. In particular, w spectra exhibited diverse shapes with a range of features throughout the cruise (Fig. 10). The top 4 panels in Fig. 10 show w(l) calculated from the sext and sabs spectra shown in Fig. 2. Two additional (bottom) panels are included to further illustrate the observed variability. The Transport/Haze -East 2 sample illustrates a relatively flat and featureless w spectrum. The other two Transport/Haze examples show greater 515 curvature in the UV portion of the spectrum. The Stagnant spectrum shows distinct UV features. The peak value of w tends to be in the vicinity of 400 nm, but this is not always the case as evident in the Blocking 2 example (Fig. 10). The Blocking 2 example is particularly intriguing for its mid-visible features, as well as its UV features. The variability and spectral features found in the sabs (Fig. 2) and w (Fig. 10) spectra suggest that spectral analysis tools such as peak fitting, as well as curvature coefficients as in Sect. 3.3, may be useful in deriving new relationships between ambient aerosol composition and optical properties.

4 Conclusions
Parts 1 and 2 of this work have explored the information content of in situ hyperspectral measurements of ambient aerosols over 300-700 nm. Here in Part 2, the analyses focused on filter-based measurements of aerosol total absorption coefficients (sabs) and the soluble absorption coefficients measured from DI and MeOH extracts (sDI-abs and sMeOH-abs, respectively). sabs together with the extinction coefficient (sext) spectra (as reported in Part 1) averaged over the filter sampling intervals enabled calculation of 525 hyperspectral single scattering albedo (w). Transforming the measured coefficient spectra into logarithmic space, it was found that all were better fit with 2 nd order polynomials than with lines as would be expected from power law representations. The derivatives of the two fits are equivalent to each other such that any given Ångström exponent, a, maps into a line in (a1,a2) coefficient space with a slope of -2ln(lch), where lch is the wavelength in any given spectrum where the two fits yield the same result. This twodimensional mapping space allows for separation among aerosol populations that otherwise exhibit the same a. Schuster et al.

530
(2006) explored the implications of this for bimodal aerosol size distributions with differing proportions of fine and coarse mode aerosols using ambient total column AERONET retrievals. Here, this method was applied to in situ measurements of the fine fraction only. Although size distributions were not measured with the in situ aerosol package aboard the R/V Onnuri, the published literature on the synoptic meteorology and its role in the observed regional characteristics of PM2.5 Jordan et al., 2020a) permitted an assessment of the optical attributes reported here as a function of the varying particle sizes and 535 composition encountered during the cruise. Clear separation of these ambient aerosol populations in (a1,a2) maps of hyperspectral optical properties indicate that such mapping offers more detailed information on the linkages between ambient aerosol optical properties and their chemical and microphysical characteristics than a alone can provide.
The May-June 2016 KORUS-OC cruise was a collaborative research campaign involving atmospheric composition and ocean 540 color scientists within the broader umbrella of the KORUS-AQ mission. That collaboration across disciplines inspired the application of an established technique in the ocean color community to measure hyperspectral sabs(l) over the 300 -700 nm range https://doi.org/10.5194/amt-2020-318 Preprint. Discussion started: 19 August 2020 c Author(s) 2020. CC BY 4.0 License.
for in situ ambient aerosols. At the outset, there was some uncertainty as to the concentrations of aerosols likely to be encountered in the marine boundary layer throughout the cruise. That, along with limited space for personnel aboard ship, led to the decision to use relatively long filter sampling times (3 hr daytime, 12 hr nighttime) to ensure above detection samples were obtained on a 545 reasonable sampling schedule. The separation in filter mean sext in (a1,a2) space for the differing aerosol populations was somewhat less distinct here than that shown in the higher temporal resolution spectra set of Part 1. This result suggests that improved temporal resolution in sabs sampling may also lead to more distinct separation of different aerosols in (a1,a2).
Evident separation in the soluble absorption coefficients arose largely from the different parameterization of partial and complete 550 spectra. The data were blank corrected and only those data above the lower limit of detection were reported and analyzed. All of the sabs(l) spectra were complete (i.e., spanned the full 300 -700 nm wavelength range), but the soluble absorption spectra depended on the wavelength-dependent absorption of the soluble chromophores, such that for some samples only partial spectra of the soluble extracts (sDI-abs(l) and sMeOH-abs(l)) were above the detection limit. Partial spectra led to larger values in a and more extreme values in (a1,a2), both positive and negative.

555
The DI extracts were also analyzed for water-soluble aerosol composition allowing for direct comparison of water-soluble composition to the optical properties. Examples of relationships for the optical properties at 365 nm were examined and found to differ such that sext(365 nm) was best correlated with anthropogenic pollution tracers (SO4 2and NH4 + ), both sDI-abs and sMeOH-abs were best correlated with oxalate (C2O4 2-), while sabs(365 nm) was not well correlated with any water-soluble ion. These are not 560 unexpected results. More interestingly, elevated C2O4 2could be separated into 2 groups, one accompanied by elevated K + and one that was not. Both groups were observed downwind of the Korean peninsula, with the former group present during the Transport/Haze period and the latter present during the Stagnant period. m/z 43, 44, and 60 from AMS measurements of the nebulized DI extracts were examined to shed more light on these populations. m/z 60 revealed no evidence of biomass burning in the data set as might be expected from elevated K + and C2O4 2-. Higher ratios of m/z 44:43 observed for the Stagnant group 565 suggested this group was more photochemically aged compared to the Transport/Haze group, consistent with what is known of the synoptic meteorology of those periods . The soluble absorption spectra for the Stagnant group were partial which may indicate chromophores that might have absorbed at longer wavelengths had degraded below detection, another potential indication of photochemical aging. Further examination of these two groups of samples using w(365 nm) and w(532 nm) showed the Stagnant group exhibited a much larger decrease in w from 532 nm to 365 nm than the Transport/Haze group. This result 570 suggests that elevated C2O4 2can represent different chromophores within the soluble organic aerosol population. Hence, it ought not to be inferred that the two groups differed only due to aging, the underlying chromophores may have arisen from different sources and/or atmospheric processes entirely, but photochemical aging may have been one factor contributing to the differences.
Finally, it was shown that spectral fit parameters did not fully capture spectral features that were observed in sabs and w. These 575 features highlight the need for hyperspectral measurements and indicate that spectral analysis tools such as peak fitting may be useful in further exploration of the wavelength-dependence of chromophores.
From the perspective of in situ ambient sampling alone, it is anticipated that hyperspectral sext, sabs, sDI-abs, and sMeOH-abs coupled to commensurate composition and size distribution information will lead to advances in deriving the physicochemical properties (TEMPO, Zoogman et al., 2017) and Phytoplankton, Aerosol, Cloud, Ocean Ecosystem (PACE, Werdell et al., 2019). TEMPO splits its hyperspectral range into two parts: 290-490 nm and 540-740 nm each with a spectral resolution of 0.57 nm (Zoogman et al., 2017). The Ocean Color Imager (OCI) planned for PACE will cover 340-890 nm at 5 nm resolution (plus 7 additional bands 585 extending further into the IR range) and it will be joined by the SPEXone polarimeter spanning 385 -770 nm at 2-4 nm resolution for that satellite (Werdell et al., 2019). It will be necessary to have in situ measurement capabilities for validation of such remote sensors. The in situ spectra reported in this work illustrate how hyperspectral information can provide new approaches that may be explored to expand the suite of aerosol products that can be retrieved from hyperspectral remote sensors.

590
All data presented here are available under the R/V Onnuri Ship tab in the KORUS-AQ archive (DOI: 10.5067/Suborbital/KORUSAQ/DATA01).

Author contribution
CEJ led the experiment, analyzed the data, and wrote the manuscript.   Table S1 for details. Fig. 1 of the spectra set obtained from each filter: mean SpEx sext (red line, shaded with ± 1 standard deviation), sabs (black line), sMeOH-abs and sDI-abs (purple and blue lines, respectively). Measurement error for all of the absorption coefficient spectra estimated using error propagation (shading, on log scale this is more difficult to discern for sabs than for sMeOH-abs and sDI-abs). Mean NT sext (red symbols ± 1 standard deviation) and mean TAP sabs (gray symbols ± 1 standard deviation) are shown for comparison. All in units of Mm -1 .