Supporting Information : Key factors affecting single scattering albedo calculation : Implications for aerosol climate forcing

School of Earth and Environmental Science, Seoul National University, Seoul, Republic of Korea Department of Physical and Chemical Sciences, Center of Excellence for the Forecast of Severe Weather (CETEMPS), University of L'Aquila, L'Aquila, Italy Department of Civil, Environmental, and Architectural Engineering, University of Colorado, Boulder, CO, USA now at Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, CO, USA now at Department of Chemistry and Biochemistry, University of Colorado, Boulder, CO, USA now at School of Earth and Environmental Science, Seoul National University, Seoul, Republic of Korea


Introduction
Aerosol Optical Depth (AOD), Single Scattering Albedo (SSA), and asymmetry parameters are important inputs used for calculating the Direct Radiative Forcing (DRF) of aerosols (Iacono et al., 2008).Among aerosol optical properties, the SSA is the factor that most contributes to the uncertainty of the aerosol DRF calculations.Previous studies have 5 suggested that a small change in the SSA can lead to a significant change in the aerosol radiative forcing based on the analysis of in-situ observations, chemical transport and radiative transfer model results, and satellite data (Loeb and Su, 2010;McComiskey et al., 2008;Srivastava et al., 2011).For example, a decrease in the SSA from 0.9 to 0.8 (an 11 % decrease) can often change the sign of the DRF from negative (cooling), to positive 10 (warming) values (Jethva et al., 2014).
Generally, the aerosol SSA is calculated using the Mie theory by assuming the particles are spherical.This spherical assumption may not be applicable to aerosols that have diverse geometrical shapes such as Black Carbon (BC) and soil dust (Mishchenko, 2009).For this reason, previous studies have suggested alternative methods such as the Rayleigh-Debye-15 Gans theory for aggregates (Farias et al., 1996), and the T-matrix method for non-spherical particles (Mishchenko, 2009).However, the Mie theory is still a powerful tool for global model calculations by virtue of its brevity, computational efficiency, and applicability to radiative transfer models.As a result, in a recent Aerosol Comparisons between Observations and Models (AeroCom) study (Myhre et al., 2013b), which was the basis of the fifth 20 assessment report from the Intergovernmental Panel on Climate Change (IPCC) (Myhre et al., 2013a), Mie theory was used in all of the participating models.
The application of Mie theory for aerosol SSA calculations requires precise information about the size distribution, refractive index, particle density, and hygroscopic Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.
-4-growth factors of aerosols.However, it is difficult to determine these parameters from measurements because the characteristics of aerosols vary widely depending on the region, source, aging and mixing state, and measurement technique.As a result, the input parameters used for aerosol SSA calculations are variable and depend on the corresponding measurements, which can cause a wide range of calculated aerosol SSAs.For example, some 5 previous global modeling studies found that no significant bias existed between calculated SSAs versus Aerosol Robotic Network (AERONET) observations (Chin et al., 2009;Dai et al., 2015), while other studies showed positive biases in the estimates (Feng et al., 2013;Jo et al., 2016;Lin et al., 2014;Myhre et al., 2009).We surmised that the SSA bias could be reduced when models use relatively high BC emissions and a low BC density.For example, 10 Chin et al. (2009), used a BC emission of 10.2 TgC yr -1 , which is 50 % higher than the 6.8 TgC yr -1 used by Jo et al. (2016) and a BC density of 1.0 g cm -3 , which is the lowest value used in the literature (Koch et al., 2009).Since there is limited information available on the emission and physical parameters of the BC values used in previous studies, we cannot reach a conclusion that explains the range of global SSA values that have been estimated by 15 previous modeling studies.Instead, we mainly focused on the physical parameters affecting the SSA calculation, which are discussed below.
Choosing the appropriate input parameters for Mie calculations is a greater challenge for BC than for other aerosols, because BC particles usually exist as agglomerates of primary spherules rather than individual spherical particles (Alexander et al., 2008).As a result, the 20 physical parameters for BC used in the AeroCom models are variable for the mass median diameter (0.020 -0.852 μm), density (1.0 -2.0 g cm -3 ), and refractive indices (1.75 -0.44i -2.0 -1.0i) (Koch et al., 2009).Furthermore, the SSA may vary significantly depending on the mixing state.Both theoretical and observational studies showed that an SSA value of an -5-internally mixed BC aerosol is lower than that of an externally mixed BC (Bond et al., 2006;Drury et al., 2010;Shiraiwa et al., 2008).
Although BC is a major species contributing to aerosol absorption, soil dust is also an important contributor (Yang et al., 2009).The global mass concentration of soil dust is the largest amongst the other aerosols.Thus, soil dust could have a significant impact on the 5 global SSA, especially since the size distributions of dust are poorly evaluated in global models (Kok, 2011;Kok et al., 2017).Previous studies showed a significant change in the observed SSA (Khatri et al., 2014) and simulated aerosol DRF (Mahowald et al., 2014) as a function of the coarse/fine fractions of soil dust.
Finally, few models consider the absorbing fraction of Organic Aerosols (OA), which 10 are typically assumed to be a scattering aerosol, although recent studies have confirmed the presence of this component (Alexander et al., 2008;Kirchstetter and Thatcher, 2012;Laskin et al., 2015).The inclusion of the absorbing OA (known as Brown Carbon, BrC) may have a significant effect on the SSA calculation (Feng et al., 2013;Jo et al., 2016;Lin et al., 2014).
In this study, we focused our analysis on four factors, the physical parameters of BC, -6-insights in the climatic implications of current Global Climate Models (GCMs).

Model description
We used the GEOS-Chem (version 10.1) global 3-D chemical transport model (Bey et al., 2001), which is driven by the Modern Era Retrospective-analysis for Research and 5 Applications (MERRA) assimilated meteorological data from the Global Modeling and Assimilation Office (GMAO) (Rienecker et al., 2011).Meteorological data including temperature, wind, humidity, planetary boundary layer height, and other meteorological variables at 0.5° x 0.667° horizontal resolutions were re-gridded to 2° x 2.5° for this study.
We conducted an NOx-Ox-hydrocarbon-aerosol full chemistry simulation, including 10 BC, OA, SO4 2-, NO3 -, NH4 + , soil dust, and sea salt aerosols.BC and OA were simulated based on the work published by Park et al. (2003).The model simulated BC and Primary OA (POA), with a hydrophobic and a hydrophilic tracer for each.For a Secondary OA (SOA) simulation, we used the scheme developed by Pye et al. (2010), which considers the parent hydrocarbons of monoterpenes, sesquiterpenes, isoprene, and aromatics with NOx dependent 15 yields.SOA was considered as hydrophilic, and further details are described in Pye et al. (2010).
For the sensitivity study described in Section 3.4, we additionally considered BrC by using the BrC/OC ratio calculated by Jo et al. (2016).We applied the BrC/OC ratio to the simulated POA and SOA of each grid box in order to calculate the BrC from both primary 20 and secondary sources.The ratio of BrC to OC burden was 0.19 across the globe, but could be regionally variable as shown in Figure 7 in Jo et al. (2016).The detailed BrC estimation method and the global distributions are described in Jo et al. (2016).
The model simulated inorganic aerosols following the method of Park et al. (2004).-7-Nitrate aerosol formation was computed using ISORROPIA II (Fountoukis and Nenes, 2007;Pye et al., 2009), which calculates the H2SO4-HNO3-NH3-H2O thermodynamics.Dust aerosols were simulated using the Dust Entrainment And Deposition (DEAD) scheme of Zender et al. (2003), along with the source function used in GOCART (Ginoux et al., 2001), as presented by Fairlie et al. (2007).The sea salt emission was estimated as a function of dry 5 particle size, local 10 m wind speed, and sea surface temperature (Alexander et al., 2005;Jaeglé et al., 2011).
We used the anthropogenic emission values of BC, OC, CO, NOx, NH3, and SO2 from the Hemispheric Transport of Air Pollution (HTAP) v2 inventory with a horizontal resolution of 0.1° x 0.1° (Janssens-Maenhout et al., 2015).HTAP v2 includes ships, energy, industry, 10 transport, residential, and agriculture sector data.For NH3, we applied the diurnal variation factors from Zhu et al. (2015) in order to reduce the high bias of nitrate aerosol simulations.
Aircraft emission values were considered by using the Aviation Emissions Inventory Code (AEIC) database (Stettler et al., 2011).Anthropogenic emissions of hydrocarbons including benzene, toluene, and xylene were taken from the Reanalysis of the Tropospheric chemical 15 composition (RETRO) inventory (Schultz et al., 2007).Biogenic VOC emissions were taken from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.1 (Guenther et al., 2012).The emissions of the VOCs were computed on the basis of emission activity factors, soil moisture, leaf age, and canopy environment.The Global Fire Emission Database version 4 (GFEDv4) inventory (Giglio et al., 2013)

FlexAOD
We calculated the AOD, SSA, and asymmetry parameters using the Flexible Aerosol Optical Depth (FlexAOD) tool, which is a GEOS-Chem post-processing tool for aerosol optical property calculations (Curci et al., 2015).FlexAOD uses Mie theory (Mishchenko et 5 al., 2002) with input data including size distributions, refractive indices, particle density, and hygroscopic growth factors included for each aerosol species.Aerosol mixing state and its effects on the AOD, SSA, and asymmetry parameter were also considered.We assumed three different mixing states -external mixing, homogeneous internal mixing, and core-shell internal mixing.Each aerosol was assumed to be formed by a single chemical species in the 10 case of external mixing.Homogeneous internal mixing assumes that all aerosols are well mixed, and the core-shell internal mixing assumes that an insoluble well-mixed core is coated by a concentric well-mixed soluble shell.A detailed description of the aerosol optical property calculation with different mixing states can be found in Section 2 of Curci et al. (2015).

Input parameters for FlexAOD
Table 2 summarizes the input parameters used in the Mie calculation of FlexAOD.
The values in Table 2 are the same as the input parameters used in online aerosol optical property and photolysis rate calculations in the standard GEOS-Chem v10.1.Input parameters in GEOS-Chem were initially taken from the Optical Properties of Aerosols and 20 Clouds (OPAC) database (Hess et al., 1998) and Chin et al. (2002), as described in Martin et al. (2003).
Herein, we briefly discuss the significant updates to the method of Martin et al. (2003) that we used in this study.The log-normal distribution was assumed for aerosols except for field campaign (Drury et al., 2010).Inorganic aerosols followed the refractive index and 5 hygroscopic growth factors (HGFs) of water-soluble component (WASO), which were changed from those of sulfate droplets (SUSO) in OPAC.HGFs of OA were updated to 1.35 at 95 % RH based on the chamber and field observations of Jimenez et al. (2009).The geometric mean radius of BC was 0.020 μm, which was increased from 0.012 μm.The density of BC was increased to 1.8 g cm -3 (Bond and Bergstrom, 2006) from 1.0 g cm -3 by 10 OPAC (Hess et al., 1998).These BC-related parameter changes led to significant changes in aerosol absorption, which are discussed in the following section.

Effects of size distribution and refractive index of BC on absorption
As discussed in Section 1, BC particles typically form aggregates (Alexander et al., 2008), which makes it difficult to choose the appropriate BC radius in the Mie calculation 15 (Koch et al., 2009).Here we examined the sensitivity of BC absorption with respect to the choice of geometric mean radius (rg) and geometric standard deviation (σ) using Mie theory.

FlexAOD simulations
Using the aerosol mass concentrations from the GEOS-Chem simulations above, we carried out a number of FlexAOD simulations by changing the input parameters.Here we focused on BC because of its high contribution to light absorption and the diverse morphologies found in the atmosphere (Alexander et al., 2008).We examined the sensitivity 15 of the calculated aerosol absorption to the assumed input parameters shown in Table 3.For example, the GEOS case indicates a FlexAOD simulation using GEOS-Chem default values as of v10-01.The OPAC database was used for the OP sensitivity case.Refractive indices between 1.95 -0.79i were used for BB, BBR, and BBHR cases with different geometric radiuses of BC (Table 3).The geometric mean radius of 0.065 μm was adapted from Bond et 20 al. (2013).We additionally assumed a geometric mean radius of 0.1 μm, which seemed to be higher than that used by previous studies (Bond et al., 2013;Wang et al., 2014), however, the value was still within the range of the AeroCom study (Koch et al., 2009).
After the sensitivity simulations shown in -11-mixing states, the effects of brown carbon, and also the observationally constrained dust size distributions.We added tag names for representing these factors.As discussed in Section 3.1, we considered the three mixing states of external, homogeneous internal, and core-shell internal and the corresponding "_E", "_H", and "_C" tag names were added to base case names in Table 3, respectively.If the sensitivity case includes BrC absorption, the "_BR" tag 5 name was added.We also implemented a new dust size distribution in the model as suggested by Zhang et al. (2013).Zhang et al. (2013), developed the dust size distribution with constraints from in-situ measurements and the work of Kok (2011), who suggested new dust size distributions based on the physics of the scale-invariant fragmentation of brittle materials.The "_DI" tag name was applied when we used the dust size distribution by Zhang

Global aerosol mass concentration
We used the AMS dataset (Zhang et al., 2007) and the SPARTAN observations (Snider et al., 2015) to evaluate the simulated mass concentrations of aerosols globally.Since considered in this study similar to previous studies (Jo et al., 2013;Spracklen et al., 2011;Yu, 2011).In contrast to inorganic aerosols, the model showed a low correlation coefficient (0.36) and a high RMSE (4.77 µg m -3 ) for OA.This result is consistent with the evaluation of 10 OA against AMS found previously in the literature (Jo et al., 2013;Yu, 2011).The low bias (-37 %) in the model was mainly caused by poor SOA simulations, which still showed a considerable gap from observations as pointed out by the recent AeroCom inter-comparison study for OA (Tsigaridis et al., 2014).
Figure 3 shows the scatter plots of BC, sulfate, nitrate, ammonium, dust, and have been decreasing (Curier et al., 2014;Granier et al., 2011;Simon et al., 2014), and the United States and Europe are the main regions covered by the AMS observation network (Zhang et al., 2007).Therefore, there is a possibility that these two different global -13-observation networks (AMS and SPARTAN) represent different nitrate concentration levels.
We found that the nitrate overestimation mainly occurred over the United States (7 of 19 points above 10:1 line in Figure 3) and South Asia (7 of 19 points above 10:1 line in Figure 3).The nitrate overestimation of GEOS-Chem is consistently reported in previous studies for the United States (Heald et al., 2012;Schiferl et al., 2014;Zhang et al., 2012;Zhu et al., 5 2013), South Asia (Gu and Liao, 2015), and East Asia (Wang et al., 2013).There have been some efforts to find the possible cause of the nitrate overestimation (e. g. the uptake efficiency of N2O5 on aerosols, the reaction rate for the formation of nitric acid, dry deposition velocity, boundary layer height, and the uptake of nitric acid on coarse mode dust or sea salt), but this issue is not still resolved (Heald et al., 2012;Zhang et al., 2012;Zhu et 10 al., 2013).Photolysis of nitrate also can reduce the overestimation as recently observed by Ye et al. (2017).An investigation of nitrate overestimation is beyond the scope of this study, and will be conducted in future studies.
Compared to nitrate, the BC values showed a relatively better result with a regression slope of 0.94.However, the BC concentration was underestimated by 37 %, and the 15 underestimation occurred mainly in Manila (10 of 13 points below 10:1 line in Figure 3).We found that the model grid box corresponding to this site was mostly oceanic (Figure S1).The spatial resolution of the model (2° x 2.5°) was too coarse to capture local sources on the island.
For the comparison of dust, we inferred the observed dust mass concentrations

AOD and SSA
The model appeared to appropriately capture the spatial and temporal (monthly) 10 variability of the observed PM2.5 described above.This may result from the combination of the nitrate overestimation and the OA underestimation.To ensure the reliability of the model for the AOD and the SSA evaluation in this section, we also selected simulated results at AERONET sites, which met the following criteria: (1) Differences between the monthly mean observed and simulated concentrations of each aerosol species (inorganics, OA, BC, 15 and dust) should be less than a factor of two at surface networks (AMS and SPARTAN).( 2) AERONET and AMS (or SPARTAN) sites should be in the same model grid or adjacent model grids.We could have applied stricter criteria for our data selection, but too few data were available for the analysis.Our conclusions, however, were found not to be sensitive to the criteria used as discussed below.-15-respectively.From this result, we concluded that the aerosol optical property calculation in FlexAOD was properly performed.When we filtered out data at AERONET sites using the criteria above, the model showed improved statistics against the observations.The NMB was reduced to -10 %, and the correlation coefficient was increased to 0.82 (Figure 4(b)).sites for 440 nm and 870 nm.Here we excluded SSA values when the AOD was less than 0.4 because the AERONET SSA retrieval showed high uncertainty for low AOD conditions (Dubovik et al., 2002).The model appropriately simulated the SSA at 870 nm in common with the evaluation of the AOD at 500 nm, but significantly overestimated the observed SSA at 440 nm.The SSA of the model (0.949) was higher than that of the AERONET (0.897) by 10 0.052.In other words, the model underestimated the observed absorption (1-SSA) by -50 %.
Although the model underestimated BC by -37 %, the model also underestimated scattering aerosols (PM2.5 minus BC) by -23 %.Therefore, the SSA overestimation could be caused by other factors than the underestimation of the BC mass concentration.This overestimation is similar to the result by Jo et al. (2016), who also found significant SSA overestimation of the 15 GEOS-Chem version 9.1.2against AERONET.The reasons for this overestimation will be investigated in following sections.The correlation was increased to 0.5 from 0.34, which is analogous to the evaluated result of 20 the AOD at 500 nm.From these results, we concluded that the model simulation capability of the AOD and SSA was improved when we selected AERONET sites where the model showed good results against the AMS and SPARTAN network.Hereafter, we used these selected AERONET sites for the sensitivity studies below.-16-

SSA sensitivity
In this section, we present the sensitivity analysis of the calculated SSA values for three factors including the aerosol mixing states, the inclusion of BrC, and dust size distributions.The sensitivity of the BC physical properties were also investigated but did not show significant differences (Figure S2).We compared the calculated values against the 5 observed SSA at 440 and 870 nm from AERONET sites across the globe.We thoroughly examined the statistics of the simulated values versus the observed SSA for the ensemble of AERONET sites in this section in order to determine the best parameters for the global SSA estimation.
The effects of the BC internal mixing on the SSA calculation are shown in Figure 6.We found that both homogeneous (Figure 6(a)) and core-shell (Figure 6(b)) internal mixing assumptions led to a significant decrease of the SSA (0.842 -0.851) at 440 nm.Furthermore, the NMBs of the AOD were decreased to -52 % with the internal mixing cases (GEOS_H and GEOS_C) from -10 % with the external mixing case (GEOS_E) (Table S1).The decreased internal mixing AODs compared to the external mixing decreased by -34 --37 % because of 15 the decrease of the aerosol numbers with the increased size of each aerosol, which is consistent with previously published results by Curci et al. (2015).-17--1 %) with respect to the GEOS_E case.

The inclusion of the
Figure 7 shows the combined effects of both the BrC absorption and the dust size distribution on the 440 nm SSA calculation.All four sensitivity cases showed improved results although the overestimation of the SSA still existed in the model.We speculate that this remaining gap could be reduced when we consider strongly absorbing BrC as some 5 studies have reported (Alexander et al., 2008;Chung et al., 2012).However, it is too uncertain whether the strongly absorbing BrC is dominant globally.The reported BrC absorptions vary by two orders of magnitude (Jo et al., 2016) and further studies are needed to clarify the presence of strongly absorbing BrC.
On the other hand, we found that the correlation coefficients and regression slopes 10 decreased as the radius of BC increased.We surmised that the use of a spherule size radius (~0.02 μm) seemed more appropriate than the aggregate size radius (~0.1 μm) for the calculations using Mie theory.However, the result can be affected by many factors, and therefore it is difficult to generalize the use of the radius in this study.
Contrary to the comparisons of the SSA at 440 nm, the model showed different results

15
for the SSA at 870 nm (Figure 8).Aerosol absorption at 870 nm is mainly affected by BC rather than BrC and dust (Yang et al., 2009).Therefore, the differences are mainly attributed to the size distributions and the refractive indices of the BC.Since BC is the only contributor to aerosol absorption at 870 nm, we found that the effects of the BC refractive index on the SSA were more important for the 870 nm than for the 440 nm results.The 870 nm SSA 20 decreased by the use of higher refractive index (BB_BR_DI_E -GEOS_BR_DI_E) was -0.017, the absolute magnitude of which is larger than the 440 nm SSA decrease of -0.011 (Table 6).
Considering all the evaluations for the AOD and SSA above, we found that the -18-GEOS_BR_DI_E case showed the best performance against AERONET observations.In the following section, we used this case for the DRE calculation and calculated the effects of the sensitivity factors on the DRE change.

Implication for global DRE
5 Here we calculated the global clear-sky DRE of selected sensitivity simulations in Figure 9 by using the Rapid Radiative Transfer Model for General Circulation Models (RRTMGs) (Iacono et al., 2008).MERRA reanalysis data were used for albedo and other meteorological variables.-19-of the reference case.
The inclusion of the BrC absorption (-0.015) led to a more effective decrease of the SSA at 440 nm than the change of dust size distribution (-0.008) as shown in Table 5.
However, the effects on the DRE were different.The relatively higher DRE change (0.17 W m -2 ) shown in Figure 9(f) is mainly caused by high dust burden (14.4 Tg), which was 10 -5 100 times higher than the BC (0.1 Tg), or OA (1.4 Tg) burden.Note that the DRE is calculated by using the AOD (~proportional to burden) as well as the SSA.Therefore, we conclude that the size distribution of dust plays an important role in the modelled DRE estimation, as also reported by Kok et al. (2017). 10

Conclusions
Although we found that the GEOS_BR_DI_E case showed the best performance against the AERONET AOD and SSA, it is difficult to confirm whether this result can be applied to other studies.There were many factors simultaneously affecting the AOD and SSA calculation such as the hygroscopic growth factor, which was not discussed in detail.
15 However, we were able to identify some notable findings that could apply to future modeling studies when using Mie theory for the calculation of aerosol optical properties: 1.There were no significant differences between the calculated BC absorptions using input parameters from the OPAC database and data from Bond and Bergstrom (2006).The refractive indices used by Bond and Bergstrom (2006) (1.95 -0.79i) were higher than those 20 used by OPAC (1.74 -0.44i), and the resulting BC absorption of Bond and Bergstrom (2006) was 49 % higher than that of OPAC.However, the low BC particle density of OPAC (1.0 g cm -3 ) increased the BC number concentration and the AOD, which enhanced the BC absorption.We found that the mean SSA of OP_E (0.927) was slightly lower than that of -20-BB_E (0.934) (Table 5).Although many previous studies did not provide the BC density they used for the Mie calculation (Feng et al., 2013;Jo et al., 2016;Lin et al., 2014), an assumed BC density was critical to the SSA calculation and should be provided in future intercomparison studies.
2. The model using the external mixing assumption showed better performance than the 5 model using internal mixing assumptions in a global modeling sense.Drury et al. (2010), also showed similar results against aircraft observations.The internal mixing assumptions caused very high absorption and an underestimation of the AERONET SSA.The effect of internal mixing on absorption could be overestimated in the model calculations; Cappa et al. (2012) reported that the observed ambient BC absorption enhancements obtained by internal mixing 10 were small (~6 %), which is less than predicted from theoretical calculations and observed from laboratory measurements.Furthermore, BC absorption enhancement by internal mixing could vary by region (Liu et al., 2015).Coating thickness could also be different due to aging timescales and the distance of source regions.Further observational and modeling studies are needed especially for considering region-specific absorption enhancement.3. The modelled SSA calculations from previous studies were mostly evaluated between 440 -550 nm (Dai et al., 2015;Jo et al., 2016;Lin et al., 2014).However, SSAs at both shorter and longer wavelengths should be evaluated together for the model evaluation.The calculated SSA could show different performances at different wavelengths because they are affected by different aerosols and wavelength-dependent physical characteristics.-29-Figure 2. Point-to-point comparisons of the simulated sulfate, nitrate, ammonium, and organics against the AMS network observations.We multiplied the organic matter to organic carbon ratio of 2.1 to organics to account for the non-carbon mass.The 1:1 line (solid), 2:1 lines (dashed), 10:1 lines (dotted) are indicated.
15 mixing state, dust size distributions, and the presence of BrC.We conducted global simulations for the years 2008 -2010 using a 3-D chemical transport model (GEOS-Chem).First, we extensively evaluated the simulated aerosol mass concentrations and aerosol optical properties by comparing the model with observations from global aerosol observation networks such as the global Aerosol Mass Spectrometer (AMS) dataset, the Surface 20 Particulate Matter Network (SPARTAN), and AERONET for the years 2008 -2010.Multiple sensitivity simulations were also carried out to examine the effects of individual factors on the calculated SSA as discussed above.Finally, using the simulation results, we estimated the aerosol Direct Radiative Effect (DRE) and its sensitivity to the assumed factors, to provide Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.

15
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.-9-soil dust, which follow the Gamma distribution.The geometric standard deviations of the inorganic aerosols, OA, and BC were reduced from 2.0 (OPAC) to 1.6 based on measurements from the optical particle counter instrument during the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT) 2004 (a).Values were generally within the reported ranges of global mean BC AAODs (0.8 X 10 -3 -3.5 X 10 -3 ) 20 from the AeroCom phaseⅠ models, although we did not consider hygroscopic growth.Red star and blue rectangle symbols indicate OPAC values (rg = 12 nm and σ = 2.0) and GEOS-Chem default size distributions (rg = 20 nm and σ = 1.6).We found that the use of these two size distributions led to very similar AAODs (< 1 % difference), resulting from the combined Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.-10-effects of the increased rg and decreased σ.

Figure 1
Figure 1(b) represents the calculated BC AAOD values as a function of real and imaginary refractive indices.BC AAOD values increased as the imaginary refractive index increased and the real refractive index decreased.Red star and blue triangles indicate BC AAOD values using the refractive indices from OPAC (1.74 -0.44i) and Bond and 5 10 et al. (2013).For instance, "GEOS_BR_DI_H" indicates that the model with GEOS-Chem default input parameters, brown carbon, dust size distribution by Zhang et al. (2013), and the homogeneous internal mixing assumption were used.
the observations were available for multiple years(AMS: 2000~ 2008, SPARTAN: 2013~ 2015), we focused our evaluation on the observed monthly variation using the monthly mean 20 value of the simulation averaged for the years 2008 -2010.Year to year variability was not

Figure 2
Figure2shows point-to-point comparisons of the simulated monthly mean sulfate,

15 Particulate
Matter with a radius of 2.5 m (PM2.5) between SPARTAN observations and the model results.Statistical parameters are also shown in Table 4.We found that the model performance for inorganic aerosols was degraded, especially for nitrate aerosols.Differences in values between observation years is one of the reasons for the discrepancy between the AMS (2000 -2008) and the SPARTAN (2013 -2015) dataset model evaluations.During the 20 past decade, NOx emissions and column concentrations over the United States and Europe Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.

20 following
the SPARTAN speciation ofSnider et al. (2016).They calculated the dust mass by multiplying [Al + Fe + Mg] by 10 based on the elemental composition of natural dust presented byWang (2015).Although this calculation seems oversimplified, we found that the method ofSnider et al. (2016) was similar to the method ofMalm et al. (1994) for estimating Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.-14-dust mass concentrations based on the observed trace metal concentrations (regression slope of 1.0 and correlation of 0.98).Simulated PM2.5 concentrations were generally in good agreement with the observations (R = 0.76) with a slight low bias observed for the mean concentrations (-24 %).The performance of the model was comparable to that of global PM2.5 estimates constrained 5 by using satellite AOD observations (R = 0.73 -0.81 and regression slopes of 0.68 -0.96)(van Donkelaar et al., 2015), which provides confidence in the model's ability to reproduce the AOD and SSA values.

20 Figure 4
Figure4(a) shows the scatterplot of the simulated versus observed AOD at 500 nm using the base case (GEOS_E).We found that the statistics of the AOD evaluation were very similar to that of the PM2.5.The slopes were 0.68 and 0.70, the correlation coefficient was 0.76, and the Normalized Mean Biases (NMBs) were -20 % and -24 %, for AOD and PM2.5,

Figure 5 (
Figure 5(a) and 5(b) are point to point comparisons of the SSA with all AERONET

Figure 5 (
Figure 5(c) and 5(d) represent comparison results similar to Figure 5(a, b) but use only selected data.The model showed an improved result especially for the SSA at 440 nm.

10
BrC absorption (GEOS_BR_E) and the use of the dust size distribution presented byZhang et al. (2013) (GEOS_DI_E) also reduced the SSA overestimation (Figure6(c-d)).The mean SSAs were decreased by 0.008 -0.015.The 20 regression slope also increased from 0.38 to between 0.55 -0.67.The correlation slightly increased for the GEOS_BR_E case (0.56), but decreased for the GEOS_DI_E case as a result of the more widely scattered points that were affected by high dust concentrations.In contrast to the internal mixing results above, the AOD was not significantly decreased (-7 % Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.

20 4.
BrC absorption and observationally constrained dust size distributions should be considered for the SSA calculation in the model, especially for the shorter wavelengths.These factors reduced the SSA at shorter wavelengths, resulting in a decrease of the positive SSA bias at 440 nm generally found in global models.Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.-21-5.The changes of dust size distribution led to a significant increase of the global DRE, whose value (0.17 W m -2 ) was more than two times higher than the global DRE increase by BrC absorption (0.07 W m -2 ).Global models likely underestimate the global dust DRE because of the overestimation of fine mode dust and the underestimation of the coarse mode dust as shown by Kok et al. (2017), who constrained the global dust size distribution using global 5 observations and models.from the Natural Sciences and Engineering Research Council of Canada.We thank the principal investigators and their staff for establishing and maintaining the AERONET sites used in this study.We also want to thank Q. Zhang and D. V. Spracklen for providing the global AMS data.This study was supported by the Korea Ministry of Environment as "Climate Change Correspondence Program" and "Public Technology Program based on 15 Environmental Policy" (2017000160001).Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License. .Phys.Discuss., https://doi.org/10.5194/acp-2017-1104Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 December 2017 c Author(s) 2017.CC BY 4.0 License.-23-Table 2. Input parameters used in the Mie calculation for FlexAOD.The log-normal distribution with geometric mean radius (rg) and geometric standard deviation (σg) are assumed for aerosols except for dust, which follows the gamma distribution with gamma A and gamma B. Particle density (ρ), refractive index (RI), and hygroscopic growth factors (HGF) are also presented.

10 15Figure 5 .
Figure 4. Scatter plot of simulated versus observed AOD at 500 nm from 2008 -2010 for (a) all AERONET points and (b) AERONET points constrained by model evaluation against the AMS and SPARTAN network.Reduced major axis regression is shown along with the regression equation and R. Each point indicates monthly averaged AOD when the number of 5

10 15 10 15
Figure 7. Scatter plots of simulated versus observed SSA at 440 nm for sensitivity simulations of (a) GEOS_BR_DI_E, (b) BB_BR_DI_E, (c) BBR_BR_DI_E, and (d) BBHR_BR_DI_E.GEOS, BB, BBR, and BBHR represent different BC input parameters as described in Table3.BR_DI_E indicates the external mixing assumption with brown carbon absorption and 5

Figure 9 .
Figure 9. Calculated DREs at the top of the atmosphere.Left panels are the DREs of the (a) GEOS_E, (b) GEOS_BR_E, (c) GEOS_DI_E, and (d) GEOS_BR_DI_E cases.The changes compared with the GEOS_E case are shown in right column.5

Table 5 .
Statistical parameters for the comparison between the simulated and the observed SSA at 440 nm for AOD > 0.4.The observed mean SSA was 0.888.