reviewer

(550 nm) retrieved by the combined CERES-MODIS algorithm for the five years of 2014-2018. Data are averaged for different Abstract. Single Scattering Albedo (SSA) is a leading contributor to the uncertainty in aerosol radiative impact assessments. Therefore accurate information on aerosol absorption is required on a global scale. In this study, we have applied a multi-satellite algorithm to retrieve SSA (550 nm) using the concept of ‘critical optical depth.’ Global maps of SSA were generated following this approach using spatially and temporally collocated data from Clouds and the Earth’s Radiant Energy System (CERES) and Moderate Resolution Imaging Spectroradiometer 5 (MODIS) sensors on board Terra and Aqua satellites. Limited comparisons against airborne observations over India and surrounding oceans were generally in agreement within ± 0.03. Global mean SSA estimated over land and ocean is 0.93 and 0.97, respectively. Seasonal and spatial distribution of SSA over various regions are also presented. The global maps of SSA, thus derived with improved accuracy, provide important input to climate models for assessing the climatic impact of aerosols on regional and global scales.

against airborne data is good to have but this is only a small part of the picture, and definitely not enough by itself.
Response: Thank you for this suggestion on strengthening the uncertainty discussion. We have included a new section on uncertainty analysis (section 5) where retrieval uncertainties due to possible perturbations in various parameters have been calculated and presented.
Additions to the revised manuscript: Table 1 identifies the major sources of error in the retrieval and summarizes their individual contribution. Uncertainty in the retrieved SSA was estimated by calculating retrieval sensitivities to perturbations in the possible error sources. The range of perturbation was based on published literature or reasonable assumptions for possible variations. Uncertainty in shortwave integrated surface albedo from CERES results in the maximum uncertainty in SSA of ±0.03. MODIS retrieved aerosol optical depth contains considerable uncertainties due to assumed aerosol models (Jeong et al., 2005). The MODIS aerosol optical depth uncertainty is 20% ±0.05 over land  and 5% ±0.03 over the ocean . The corresponding error in our retrieval is ±0.02. For a typical variation of angstrom exponent (±0.4) and refractive index (±0.01), the uncertainties vary depending on the surface albedo and are mostly around ±0.01.

Uncertainty Analysis
Changes in aerosol height can vary the TOA radiances due to Rayleigh scattering interactions, which depend on pressure. Sensitivity to aerosol height was estimated by conducting a synthetic retrieval of SSA over a range of aerosol height values and perturbations from those heights. The average uncertainty observed for an aerosol height variation of ±1 km was ±0.01. Many methods have been developed for detecting aerosol type, especially smoke vs. dust, to improve the uncertainties of various AOD and SSA retrievals.
Uncertainties due to possible variations on scales of the regions used for linear fitting were estimated as residuals of the fit. The uncertainty on the linear intercept is spatially dependent and is mostly around ±0.02, with higher values for those combinations having a slope close to zero during the regression. For highly correlated cases (i.e., correlation coefficient | r | > 0.5), the probability of obtaining a slope close to zero is ~20% over the ocean and <5% over land. These cases are mostly formed over regions where AOD variations are less. Regions having large variations in AOD values have lower uncertainty due to residual fit.
Overall, the algorithm is most sensitive to variations in surface albedo, followed by higher sensitivity towards AOD values used in the linear fit. The uncertainties are higher for scattering aerosols over bright surfaces and absorbing aerosols above dark surfaces. Sensitivity to water vapor is almost negligible, except in very few cases where the uncertainty is + 0.008. The CERES-MODIS algorithm is most effective over regions with large AOD variations and less surface albedo variations.

Comment:
It is not clear to me if the derived SSA data are publicly available; I did not see a link in the paper. They ideally should be somewhere.
Response: These datasets were generated as part of the author's ongoing Ph.D. research. All the datasets generated as part of the thesis work will be published online on the department's website on successful completion of the degree. For now, as mentioned in the data availability section, it will be available on request.

Comment: Specific comments:
experiments was as described by Nair et al., 2009. Uncertainties in the scattering coefficient measurement by nephelometer are ~±10%, as reported by Anderson et al., 1996. As stated by Babu et al., 2016 uncertainties in the columnar SSA values estimated from RAWEX aircraft measurements depend mainly on instrumental uncertainties, sampling errors, and large spatial averaging.

Response continued:
Initially, while making the aircraft data comparisons, we had looked into other aircraft data available from various field campaigns such as ORACLES (South Eastern Atlantic), ACE-ENA (Northeastern Atlantic), DISCOVER-AQ (USA), etc. These flight datasets are available in the ASDC and ESPO data archives. They provide the raw data collected during the flightssuch scattering coefficient measured by nephelometer at various latitude, longitude, and altitudes. These raw datasets need to be carefully processed considering the various instrument calibrations and experimental setup to obtain the scattering coefficient profiles over the flight track, from which the SSA profiles are computed. Further, these profiles need to be vertically integrated (also considering the flight's lat-lon variations) to obtain the columnar SSA required for comparison with the CERES-MODIS dataset. This entire work in itself would be an extensive experimental-data processing of the flight data. Carrying out these detailed computations would only provide a few datapoints for comparison with CERES-MODIS values over that region for the period.
The SWAAMI, RAWEX, and SWAAMI-RAWEX campaigns were organized and conducted by our research group. Hence the processed-datasets generated from the raw data by the experimentalists were readily available to us for comparison with the CERES-MODIS satellite data. -For the study period of 2014-18, the CERES-MODIS SSA has also been compared to monthly AERONET SSA data (440 nm) for the corresponding period for various AERONET sites. As suggested by the reviewer, we have chosen AERONET sites based on the type of aerosols as given by Giles et al., 2012. These results have been incorporated as the new Section 7 in the revised manuscript.
-Following the reviewer's comment, we intended to compare with the MERRAero reanalysis dataset. It would have been a valuable addition to the paper. Unfortunately, the OpenDap server was not accessible for downloading the data. Monthly SSA data files downloaded from the GEOS-5 data server were missing data values in it. We tried with the GrADS data server, but the download link was unavailable. We had also got our manuscript response deadline extended with ACP, hoping the data server would be up running by then. But we couldn't get to download the files.
Additions to the revised manuscript: 4 Results and discussion  The retrieved SSA dataset (500 nm) was compared with other widely used global SSA datasets -OMI SSA (500 nm) and climatological POLDER SSA (565 nm). OMAERUVd V3 (Torres et al., 2007;Torres et al., 2013;Ahn et al., 2014) for the corresponding period are shown in panels a, c, e, and g in  leading to large data gaps. In comparison, we can notice that CERES-MODIS and POLDER have better data coverage on a global scale. In the CERES-MODIS maps, the absence of data is mostly due to the unavailability of MODIS AOD.
-The Global Ocean, a relatively dark surface covering more than 70% of the Earth's surface, plays a significant role in determining global aerosol radiative forcing effects. Therefore, the better data coverage over oceans by the CERES-MODIS and POLDER provides better input for radiative forcing calculations.   Table S1.

Comparison with AERONET data
The Aerosol Robotic Network is a ground-based worldwide federated network of Cimel Sun photometers that measure extinction AOD from direct Sun measurements (Holben et al., 1998). The spectral diffuse sky radiations measured at different angles are inverted in conjunction with direct Sun measurements to derive the spectral SSA (440, 675, 870, and 1020 nm) and size distribution (Dubovik and King., 2000). The estimated uncertainty in retrieved SSA is largely attributed to the uncertainties in instrument calibration and is within 0.03 for AOD (440 nm) larger than 0.4. (Dubovik et al., 2000.   Comparison with SSA from 15 AERONET sites showed an acceptable agreement between AERONET and CERES-MODIS SSA, within their uncertainties.
 Overall, the combined CERES-MODIS algorithm provides global SSA maps with improved accuracy and better spatial coverage. These global maps provide valuable input for models to make assessment of aerosol-climate impacts on both regional and global scales. Global maps of SSA were generated following this approach using spatially and temporally collocated data from

Comment
Clouds and the Earth's Radiant Energy System (CERES) and Moderate Resolution Imaging Spectroradiometer 5 (MODIS) sensors on board Terra and Aqua satellites. Limited comparisons against airborne observations over India and surrounding oceans were generally in agreement within ±0.03. Global mean SSA estimated over land and ocean is 0.93 and 0.97, respectively. Seasonal and spatial distribution of SSA over various regions are also presented. The global maps of SSA, thus derived with improved accuracy, provide important input to climate models for assessing the climatic impact of aerosols on regional and global scales. 10

Introduction
Atmospheric aerosols play a significant role in the Earth's radiation budget (IPCC, 2013). The climatic impact of aerosols depends on their absorption and scattering properties, quantified by Single Scattering Albedo (SSA).
Even a slight reduction in SSA can change the aerosol radiative forcing from cooling to warming, depending on the underlying surface albedo Chand et al., 2009). However, the lack of an accurate global 15 aerosol absorption database has led to SSA being the largest contributor to the total uncertainty in aerosol radiative impact assessment (IPCC, 2013).
The high spatio-temporal variability in aerosol properties entails the need for observations on a global scale Levy et al., 2007;Remer et al., 2008;Hammer et al., 2018). Satellite data, despite its inherent limitation associated with an inverse problem, can provide the global perspective required in analysing 20 Deleted: The method has been validated using the data from aircraft-based measurements of various field campaigns. The retrieval uncertainty is +0.03 and depends on both the surface albedo and aerosol absorption. 2 spatio-temporal aerosol characteristics (Torres et al., 2002;Lenoble et al., 2013). However, it is difficult to quantify the absorption over bright surfaces (Kaufman and Joseph, 1982;Ahn et al., 2014;Jethva et al., 2018).
Hence, quantifying the aerosol absorption over land regions using satellite-based remote sensing remains a challenge even now (Torres et al., 2013;Jethva and Torres, 2019).
Fraser and Kaufman., 1985 developed a critical surface reflectance method to retrieve SSA using satellite data. 5 Their method is based on radiative transfer simulations, which showed a particular surface reflectance for which the top of atmosphere albedo is independent of AOD. Upward radiances between a clear and a hazy day over a varying surface reflectance region are used, along with radiative transfer simulations, to derive SSA. This method has been widely applied to data from various satellites to derive SSA over particular regions (Kaufman, 1987;Kaufman et al., 1990Zhu et al., 2011;Wells et al., 2012). Seidel and Popp., 2012 have done extensive 10 studies on the method's sensitivity to various parameters.
Various studies have ascertained the inadequacy of single-sensor data in the accurate retrieval of aerosol absorption Zhu et al., 2011). Dawn of the A-Train satellite constellation (Anderson et al., 2005) with spatially and temporally near-collocated observations facilitates multi-satellite retrieval of aerosol absorption (Eswaran et al., 2019;Hsu et al., 2000;Hu et al., 2007Hu et al., , 2009Jeong and Hsu, 2008;Narasimhan and 15 Satheesh, 2013;Satheesh et al., 2009) However, all these multi-sensor retrievals are in the Ultra Violet (UV) wavelengths, and SSA is extrapolated to visible wavelengths using spectral dependence of assumed particle size distribution. Satheesh and Srinivasan (2005) defined the concept of "critical optical depth" (τc) and introduced a method to retrieve SSA in the visible region by combining ground-based and satellite measurements. The method was validated/demonstrated over many locations, including the desert location of Solar Village in Saudi Arabia, 20 using Aerosol Robotic Network (AERONET) data.
In this paper, we have utilized the concept of τc and further extended the methodology to develop the combined CERES-MODIS retrieval algorithm to derive regional and global maps of aerosol absorption (550 nm) using multi-satellite data. The "critical optical depth" method developed in this research paper shares a similar concept to the critical surface reflectance method (Fraser and Kaufman., 1985). For a particular parameter (such as surface 25 reflectance or optical depth), there exists a critical value at which the top of atmosphere albedo can be considered independent of variations in that parameter. Both the methods retrieve SSA by parameterizing the critical value as a function of SSA using radiative transfer simulations. The critical reflectance method requires two-days data and large variations in surface reflectance over the region. It's suitable for retrieving daily SSA for a particular 3 region. Whereas the critical optical method developed in this paper is suitable for retrieving mon thly or seasonal global maps of SSA.
The concept of τc, which forms the scientific basis for the development of this retrieval algorithm is illustrated in Section 2. The various steps involved in the retrieval algorithm are detailed in the Section 3, data and methodology.
Section 4 presents the results and comparison with other satellite datasets. Uncertainity analysis is studied in 5 Section 5. Comparison with aircraft measurements from various field campaigns are shown in Section 6.
Comparison with AERONET data from 15 sites are shown in section 7. Summary and conclusions are provided in Section 8.

Critical optical depth
Let Δα be the difference between the top of the atmosphere (TOA) albedo and surface albedo. Then, for a 10 particular location, with a given surface albedo, Δα variations are only due to changes in TOA albedo. The presence of absorbing aerosols over a bright surface decreases the TOA albedo. In contrast, scattering aer osols over a dark surface increase the TOA albedo. Thus, the increase (decrease) in aerosol loading due to scattering (absorbing) type of aerosols leads to an increase (decrease) in Δα. The rate of change in Δα with aerosol loading is dependent on SSA. 15 Satheesh and Srinivasan (2005) utilized this concept to retrieve SSA in the case of absorbing aerosols over a bright surface. In a pristine atmosphere (Aerosol Optical Depth = 0) over a bright surface, the Δα is positive for solar zenith angle (SZA) = 0. Here, when absorbing aerosols become dominant, Δα decreases with an increase in aerosol optical depth (AOD) and eventually turns negative. The AOD at which Δα equals zero is defined as τc. For a given surface albedo, τc is the AOD at which the scattering and absorbing effects of the aerosol cancel each other. The 20 rate of decrease in Δα with the increase in AOD is higher when SSA is high and consequently lowers the resulting values of τc. A radiative transfer (RT) model was then used to calculate the SSA that reproduces the same τc, given atmospheric conditions. Deleted: Section 4 presents the validation of SSA derived using this approach using aircraft measurements from various 25 field campaigns.

Deleted:
The global maps of SSA thus retrieved, its comparison with SSA from Ozone Monitoring Instrument (OMI), and the seasonal distribution of SSA over many regions are presented in Section 5. An increase in aerosol loading by absorbing (scattering) type of aerosol leads to decrease (increase) in TOA albedo .
(a) Absorbing aerosols above a dark surface; (b) Absorbing aerosols above a bright surface; (c) Scattering aerosols above a dark surface; (d) Scattering aerosols above a bright surface.
In this paper, the concept of τc is extended to retrieve SSA for all scenarios of surfaces (dark and bright) and aerosols (absorbing and scattering). For AOD less than 1, Δα is almost linearly dependent on AOD. Then τc is mathematically the x-intercept when parameterizing the linear relationship. Figure 1 shows the estimation of τc for four different scenarios. Details of these RT simulations are given in 5 Section 3.2. Unlike Satheesh and Srinivasan (2005), where simulations were carried out for SZA = 0, here the Δα is diurnally averaged. Therefore, it is possible to have negative Δα for AOD = 0 over relatively bright surfaces. It is difficult to retrieve SSA where the slope of regression line is close to zero.

3 Data and methodology
The Combined CERES-MODIS retrieval algorithm consists mainly of two steps: (1) determining τc using MODIS and CERES data for a location, and (2) estimation of SSA that reproduces the same τc for the associated atmospheric conditions and surface albedo of that particular location. Figure 2 shows the flowchart illustrating the combined CERES-MODIS retrieval algorithm. 5 TOA and surface fluxes, used to determine Δα, are obtained from CERES SYN1deg-day (Edition 4.1) (Wielicki et al., 1996;Rutan et al., 2015). To avoid angular dependence of fluxes, the diurnally averaged flux data product from CERES is used, which is available only at 1° resolution. Hence, other satellite data sets in this study are also used at the same spatial resolution. AOD and total columnar water vapor are obtained from the MODIS Daily Global Product (MxD08_D3 version 6.1). MODIS retrieves columnar AOD at 550 nm using two different types 10 of algorithms -"Dark Target" (Levy et al., , 2013 and "Deep Blue" (Hsu et al., 2004(Hsu et al., , 2006Sayer et al., 2013). Dark target retrieves AOD over both land and ocean, whereas deep blue retrieves only over land. In this study, we have used a combined dark target and deep blue product.

Determining the critical optical depth
The first step for retrieval is to determine τc by linear regression analysis between Δα vs. AOD as shown in Fig.   3. The x-intercept of the resultant line of best fit (i.e., the AOD at which Δα = 0) provides the value of τc. CERES and MODIS daily data are at 1° resolution, and SSA is retrieved for each 1° × 1° grid. In order to have adequate number of points for a meaningful regression analysis, it was required to use data over a larger interval (temporal and spatial) -whose extent is large enough to get a statistically significant fit but small enough to ensure insignificant variations in SSA. Thus, to determine τc for a given pixel, seven days of data from its surrounding 5° × 5° region has been considered. This data is further constrained based on surface albedo and water vapor. 5 Only those pixels in this region having surface albedo within + 0.025 and water vapor within + 0.25 cm of the given pixel are considered for regression analysis. These constraints ensure that the τ c determined from the best fit is dependent only on SSA and not affected by changes in surface albedo and water vapour. Figure 3a shows an example of regression with a positive correlation coefficient over the Arabian Sea. This can happen over regions of low surface albedo and the dominance of scattering aerosols. Figure 3b is an example of regression analysis 10 with a negative correlation coefficient obtained over Sahara in the presence of dust aerosols.
The above procedure is repeated for all pixels, where data from the surrounding 5° × 5° region is used to determine τc for each pixel. For the regression analysis, points which are outside one standard deviation are considered as outliers. Line of best fits with a slope close to zero yields extreme τc values (very high positive/very low negative).
In such cases, we did not attempt a retrieval. A significance test on the correlation coefficient between AOD and 15 ΔAlbedo is performed with a 0.05 significance level. Only those τc values obtained through regressions that are statistically significant at 95% confidence level are utilized further to retrieve SSA. The final product of this step is a 360 × 180 matrix that stores τc value corresponding to each 1° pixel. In these matrices, not all points would have a τc value owing to the insufficient number of points available for regression, either due to cloud-masking or large variations in surface albedo over the land. At least seven days of data is required to perform a statistically significant fit to compute τc and retrieve SSA The next step in the procedure is 5 to estimate SSA from these τc values using an inverse lookup table (LUT) approach.

Retrieval of SSA
Since the objective of this study is to retrieve SSA globally, look-up-tables (LUTs) were developed to reduce the computation time and avoid repeated RT simulations. The aerosol models from OPAC (Optical Properties of Aerosols and Clouds), developed by Hess et al., (1998), are given as input to SBDART (Santa Barbara DISORT 10 Atmospheric Radiative Transfer) model (Ricchiazzi et al., 1998) to simulate TOA fluxes. Specifications of the model used are shown in Table S5, S6 and S7.
The RT computations were carried out to obtain the diurnally averaged (SZA: 0° to 84°) TOA and surface f luxes using 16 radiation streams and spectrally integrated over the shortwave region (0.3 to 5 μm). For a particular case of surface albedo, water vapor, and SSA, AOD is varied from 0 to 1 in steps of 0.2 to generate its corresponding 15 diurnally averaged Δα. Then a linear fit is performed between AOD and simulated Δα to determine τc. A threedimensional LUT that stores τc for different combinations of surface albedo, water vapor, and SSA have been developed. The LUT is indexed by 11 values of surface albedo (0 to 0.5, increments of 0.05), 17 values of water vapour (0 to 8 cm, increments of 0.5 cm) and 10 values of SSA (0.8,0.83,0.85,0.87,0.9,0.92,0.95,0.97,0.99,and 1). A total of 89760 RT simulations were performed in the present study. 20 The next step is to estimate SSA from τc using the LUT. For a given surface albedo and water vapor of that pixel, we find the SSA associated with its determined τc. An inverse lookup operation is performed on LUT by linear interpolation between the nearest two indices. SSA is estimated for each available τc values of a pixel and then averaged to compute the seasonal mean SSA.  The retrieved SSA dataset (500 nm) was compared with other widely used global SSA datasets -OMI SSA (500 5 nm) and climatological POLDER SSA (565 nm). OMAERUVd V3 (Torres et al., 2007;Torres et al., 2013;Ahn et al., 2014) for the corresponding period are shown in panels a, c, e, and g in Fig 5. And POLDER 1-2 Level 3 climatological seasonal mean SSA maps are shown in panels b, d, f, and h in Fig 5. For a generalized qualitative comparison, we can assume that SSA does not vary much for the small 50 nm spectral difference between CERES -MODIS and OMI SSA. (Zhu et al., 2011;Jethva et al., 2014). -Over the ocean, OMI retrieves SSA only for regions with high values of UVAI, leading to large data gaps.
In comparison, we can notice that CERES-MODIS and POLDER have better data coverage on a global scale. In the CERES-MODIS maps, the absence of data is mostly due to the unavailability of MODIS AOD.

5
-The Global Ocean, a relatively dark surface covering more than 70% of the Earth's surface, plays a significant role in determining global aerosol radiative forcing effects. Therefore, the better data coverage over oceans by the CERES-MODIS and POLDER provides better input for radiative forcing calculations.  Table S1. 15 Table 1 identifies the major sources of error in the retrieval and summarizes their individual contribution.

Uncertainty Analysis
Uncertainty in the retrieved SSA was estimated by calculating retrieval sensitivities to perturbations in the possible error sources. The range of perturbation was based on published literature or reasonable assumptions for possible variations. 20 Uncertainty in shortwave integrated surface albedo from CERES results in the maximum uncertainty in SSA of ±0.03. MODIS retrieved aerosol optical depth contains considerable uncertainties due to assumed aerosol models (Jeong et al., 2005). The MODIS aerosol optical depth uncertainty is 20% ±0.05 over land  and 5% ±0.03 over the ocean . The corresponding error in our retrieval is ±0.02. For a typical variation of angstrom exponent (±0.4) and refractive index (±0.01), the uncertainties vary depending on the 5 surface albedo and are mostly around ±0.01.
Changes in aerosol height can vary the TOA radiances due to Rayleigh scattering interactions, which depend on pressure. Sensitivity to aerosol height was estimated by conducting a synthetic retrieval of SSA over a range of aerosol height values and perturbations from those heights. The average uncertainty observed for an aerosol height variation of ±1 km was ±0.01. Many methods have been developed for detecting aerosol type, especially smoke 10 vs. dust, to improve the uncertainties of various AOD and SSA retrievals.
Uncertainties due to possible variations on scales of the regions used for linear fi tting were estimated as residuals of the fit. The uncertainity on the linear intercept is spatially dependent and is mostly around ±0.02, with higher values for those combinations having a slope close to zero during the regression. For highly correlated cases (i.e., correlation coefficient | r | > 0.5), the probability of obtaining a slope close to zero is ~20% over the ocean and 15 <5% over land. These cases are mostly formed over regions where AOD variations are less. Regions having large variations in AOD values have lower uncertainty due to residual fit.
Overall, the algorithm is most sensitive to variations in surface albedo, followed by higher sensitivity towards AOD values used in the linear fit. Seaonal mean maps of surface albedo are shown in Fig S3. The uncertainties are higher for scattering aerosols over bright surfaces and absorbing aerosols above dark surfaces. Sensitivity to 20 water vapor is almost negligible, except in very few cases where the uncertainty is + 0.008. The CERES-MODIS algorithm is most effective over regions with large AOD variations and less surface albedo variations.

Comparison with airborne observations
For the comparison of columnar SSA values thus retrieved, we have used aircraft-based measurements of SSA from three campaigns: South West Asian Aerosol Monsoon Interactions (SWAAMI), Regional Aerosol Warming 25 Experiment (RAWEX), and SWAAMI-RAWEX, to obtain column-integrated SSA. Available data points over India and adjoining oceanic regions (Arabian Sea and Bay of Bengal) from these field campaigns were compared with the retrieved SSA. Instrument design and calibration were based on Anderson et al., 1996 and its application for Indian field experiments was as described by Nair et al., 2009. Uncertainties in the scattering coefficient measurement by 10 nephelometer are ~±10%, as reported by Anderson et al., 1996. As stated by Babu et al., 2016 uncertainties in the columnar SSA values estimated from RAWEX aircraft measurements depend mainly on instrumental uncertainties, sampling errors, and large spatial averaging.
Retrieved SSA, for the same period as the campaign, over a 2°×2° region around the campaign location was utilized for comparison. Figure  could be due to frequent cloud cover during the monsoon season, leading to fewer SSA points retrieved over the 5 ocean and land. SSA estimated over Nagpur in Central India during RAWEX is ~0.8, while CERES-MODIS retrieves ~0.85. This inconsistency is due to the large surface albedo variations (standard deviation >0.05) over Central India, which leads to fewer points available for retrieval. Except for few such cases, most of the other points lie within an absolute difference of 0.03.
For comparison purposes, many previous studies have used ground -level SSA data from AERONET obtained 10 through inversion methods (Zhu et al., 2011;Jethva et al., 2014). Even in this study, only very few points were available for comparsion due to the limited number of direct measurements of columnar SSA. Despite this limitation, this comparison exercise provided confidence to generate global maps of SSA following this method. The Aerosol Robotic Network is a ground-based worldwide federated network of Cimel Sun photometers that measure extinction AOD from direct Sun measurements (Holben et al., 1998). The spectral diffuse sky radiations measured at different angles are inverted in conjunction with direct Sun measurements to derive the spectral SSA (440, 675, 870, and 1020 nm) and size distribution (Dubovik and King., 2000). The estimated uncertainty in 5 retrieved SSA is largely attributed to the uncertainties in instrument calibration and is within 0.03 for AOD (440 nm) larger than 0.4. (Dubovik et al., 2000. AERONET version 3, level 2.0 monthly average values from selected sites were compared with corresponding CERES-MODIS SSA data. Sites were chosen to represent various types of aerosols following that of Giles et al., 10 2012. The location of the sites is shown in Fig S2 and Table S3 (Giles et al., 2012)

Summary and Conclusions
 Global maps of aerosol absorption have been generated following the concept of "critical optical depth".  Overall, the combined CERES-MODIS algorithm provides global SSA maps with improved accuracy and better spatial coverage. These global maps provide valuable input for models to make assessment of 15 aerosol-climate impacts on both regional and global scales.

Data Availability
MODIS and CERES data used in this study are available at https://asdc.larc.nasa.gov/. The combined CERES-MODIS datasets are available upon request from the corresponding author.

Author Contributions
SKS conceptualized the method. AD developed the algorithm, carried out the simulations, and analyzed the data. 5 AD wrote the manuscript with revisions from SKS.

Competing interests
The authors declare they have no conflict of interest.