Review for Atmospheric Measurement Techniques

Title: Retrieval of UV-Visible aerosol absorption using AERONET and OMI-MODIS synergy: Spatial and temporal variability across major aerosol environments

Authors: Vinay Kayetha, Omar Torres, Hiren Jethva

General Comments:

This is a paper that addresses a new approach to the remote sensing (RS) of aerosol absorption by satellite, and with comparison to the same ground-based RS retrievals that are a major input to the algorithm/technique. The use of all AERONET retrieved parameters except for imaginary refractive index is justified for the retrieval of SSA in the UV wavelengths which is currently not an AERONET product (440 nm is the shortest wavelength in V3). The concept of the paper is somewhat new in scope and has potential to add significantly to the current knowledge on global aerosol absorption.  However, there are numerous significant shortcomings to this paper, especially related to the estimated uncertainty of these satellite retrievals of aerosol single scattering albedo (SSA). The uncertainty of the satellite retrievals is somewhat inconsistent in the text and the values in Table 2. You need to clearly state that the uncertainty in SSA from these retrievals due to the complete ensemble of the known sources of error. These known sources of error are uncertainties in all of these factors: AOD, BRDF, aerosol layer height (ALH), size distribution, real refractive index and satellite sensor calibration uncertainty. Full uncertainty calculations and a complete discussion seems to be missing from this paper, and it is essential. Your assumed uncertainty in surface albedo of 0.01 is quite small and is somewhat hard to believe, except in the UV where the reflectivity is quite low for most surface types. References are missing to support this level of surface reflectance uncertainty for the visible wavelengths.  Additionally surface albedo is not what is important here but it is BRDF that is a function of both view and solar zenith angles. Additionally, since you use a climatology of size distribution and refractive index from AERONET how much does that affect the retrievals at high and/or low AOD when there are departures from the presumed averages used in your climatological values? See comments below in ‘Specific Comments’ about the dynamic nature of size distributions as a function of AOD and relative humidity. Also, you make no mention of the magnitude of the uncertainty in satellite sensor calibration and yet this needs to be included in your calculation of overall uncertainty of the satellite retrievals. Your computed uncertainty of +-0.03 in the UV is as good as AERONET at 440 nm (see Sinyuk et al. 2020) and is somewhat hard to believe since for AERONET the effect of uncertainty in surface reflectance and ALH is minimal at AOD(440)>0.4, while the factor of ALH is quite a large of error in the OMI satellite algorithm retrievals in the UV. Table 2 does seem to include AOD uncertainty calculations as +- 0.02 but you fail to discuss this in the text. Also the header of Table 2 only mentions the effects of ALH, so this header needs to be re-written to include the other two factors.

Another problematic aspect of this paper is classification of some sites as biomass burning sites (ex. Arica, Chile) when that this clearly erroneous and other sites as urban aerosol types when they are clearly biomass burning sites (ex. Cuiaba, Brazil). I give details below (in Specific Comments) including several other sites plus some references in the peer-reviewed literature regarding the aerosol types at these AERONET sites.

Additionally, the referencing of published scientific literature is sometimes lacking and sometimes completely erroneous in this manuscript. I provide numerous examples below in my Specific Comments. The authors need to follow normal standards for correct referencing for a journal of high standards such as AMT, and in this manuscript these standards have not been met.

As a result, my opinion is that this manuscript requires extensive revisions and corrections before it may possibly be suitable for publication in AMT.

Specific Comments:

Line 40: It would be nice to quantify here the SSA value corresponding to 50% less UV flux. Also, it seems that you meant ‘decrease’ instead of ‘increase’ in this sentence.

Lines 51-53: It should also be mentioned that most in situ measurement techniques make several assumptions and have significant 'corrections' made to the data to overcome instrumental shortcomings or issues. Therefore the community in general does not consider these data to be a 'gold standard' for validation and there currently is no such gold standard data product available.

Lines 60-63: This statement is not entirely accurate. The Hybrid scan has been taken for several years now from newer model Cimel instruments utilized by AERONET. These Hybrid scan retrievals are robust down to 25 degrees solar zenith angle as compared to 50 degrees for the almucantar scans (see Sinyuk et al. 2020). Therefore the high SZA limitation no longer applies. Also these same instruments have taken 380 nm sky radiance data which is now being analyzed by the AERONET group for a future SSA retrieval at this wavelength. Although this data product is not yet available it should be mentioned that the data exits at many sites for a 380 nm retrieval of SSA, which is an ongoing research topic within the AERONET group. Additionally the recently published paper of Sinyuk et al. (2020) provides uncertainty values of SSA for all AOD levels (figures and tables) so that a data user can assess the level of uncertainty of these retrievals at the 4 retrieval wavelengths for a compete range of AOD for four different aerosol types.

Lines 76-77: SSA is sometimes assumed constant for a region in some MODIS retrievals, but not one constant value globally. You should reference the relevant papers on this regionally varying assumed SSA by Remer et al. 2005 and Levy et al. 2007.

Line 90: ‘Sahara belt’ is an odd terminology. Maybe you mean the Saharan and/or Sahelian regions?

Line 102: This is confusing, as MODIS does not measure at a local noon overpass time. Do you mean 'local time' here instead of 'local noon'?

Line 120: You need to reference Giles et al. (2019) here for the AERONET Version 3 Level 2 AOD data that are cloud screened and QA checked.

Line 121: You say nine nominal wavelengths and then you only give 8 wavelengths, omitting 1640 nm.

Line 122: You say the AERONET measurement interval is 15 minutes, but for many instruments the current interval is 5 minutes and has been for a few years. Please convey this in the text.

Line 122: You need to say how you interpolated the spectral AOD measured by AERONET to the OMI measurement wavelengths. A second order polynomial fit of AOD versus wavelength in logarithmic coordinates is much more accurate than using the Angstrom or linear fit (Eck et al., 1999).

Line 123: Dubovik and King (2000) is the retrieval algorithm paper and not did analyze the AOD measurement accuracy. Therefore, this is a totally inappropriate reference here. The AERONET paper that provides an analysis and estimate of AOD measurement uncertainty is Eck et al. (1999).

Line 128: The Dubovik et al. 1998 paper is a totally inappropriate reference here. Please replace this with Dubovik et al. (2006).

Line 129-130: This is a completely obsolete reference for the uncertainty of SSA from AERONET. For Version 3 retrievals there are spectral estimates of SSA uncertainty in Sinyuk et al. (2020) for each of the 4 wavelengths that are also provided as a function of AOD. 

Lines 130-132: This is confusing since below and in Table 1 you say that you use L2 retrievals, yet in this sentence the mention of L1.5 data implies that possibly some of this data may also have been used in this paper. Please clarify this issue in the text.

Lines 135-137: Note that you have an error in Table 1 where you say that for AERONET retrievals Version 2 is used, while in the text you say Version 3 is analyzed.

Lines 164-167: In section 2.3 MODIS, it seems that at least one reference for MODIS data is warranted in the text.

Line 177: Did you create a size distribution climatology that is a function of AOD? It is well known that fine mode size increases significantly as AOD increases at many sites due to aerosol aging and/or hygroscopic growth. This has been shown in many AERONET papers, see Eck et al. 2012 where the fine radius increases by over 50% over the range of AOD at the GSFC site. Additionally, Eck et al. 2010 presents examples where the fine mode radius doubles over a wide range of fine mode fraction.

Line 211: This sentence is very confusing, please clarify in the text here. Do you average all AERONET measurements of AOD within the +-2 hour interval? Also average all retrievals of size distribution and refractive index within the +-2 hour interval? What specifically do you mean by 'keep it intact'?

Line 221-222: How accurate is this aerosol typing? How many types are there to choose from? Please give a short summary of the typing procedure and it's accuracy or reliability.

Lines 224-226: Please provide a short description of the accuracy of this aerosol height climatology in the text, as this is a critical factor in the overall uncertainty of the satellite retrievals of SSA in the UV wavelengths.

Line 237: What is the wavelength of the AOD in the x-axis of Figure 4? This should be added to the figure.

Lines 239-241: It should be noted that these are significantly different SSA values than those retrieved by AERONET in the visible wavelengths for the GSFC site. See Giles et al. (2012) that gives SSA at 440 nm = 0.96 and SSA at 675 nm = 0.95. These are VERY large differences (~0.06 to 0.10), especially when it is considered that the SSA parameter range is only ~0.80 to 0.99 for >99% of AERONET retrievals. However you show a large trend as a function of AOD and the lowest AOD bin may be dominating these averages. Why give equal weight to low AOD retrievals that you are aware have very large uncertainty and likely large biases? It is good to see that the Table 3 values for SSA are for AOD(440)>0.4 and that these are much closer to the AERONET retrieval values of the Level 2 database.

Lines 252-253: This sentence is somewhat confusing and suggests that you may possibly have used different surface reflectances from the MAIAC product that you documented above.

Line 260: This site name has been mis-spelled, it is Mauna Loa, the Langley calibration site for AERONET. Please note that the AOD is also VERY low at these 3 sites, well below the threshold for accurate retrievals of SSA. Additionally, the Mauna Loa site is on a high mountain (3400 meter elevation for the site) so the AOD from AERONET would not correspond to the aerosol signal of the total atmospheric column as measured by OMI or MODIS. You should have filtered out all sites on mountains for this reason before attempting this analysis.

Line 270: Angstrom must always be capitalized as it is the name of a scientist.

Line 275: Why not reference the Angstrom 1929 paper, as it is the origin of this parameter.

Line 282-284: The selection of AE<0.2 for coarse aerosol cases is somewhat extreme as there are relatively few cases of desert dust with AE<0.2 in the AERONET database. Utilizing AE<0.4, which is still coarse mode dominated, would have resulted in many more dust cases to analyze. There are many papers based on in situ measurements that have identified fine mode dust thus the AE of airborne desert dust does not often equal zero or have a negative value which would be the case if there were only coarse mode particles. The 0.2<AE<1.2 bin for mixed mode cases encompasses a very wide range of fine mode fraction of AOD, ~30% to 70% at 500 nm (see Eck et al. 2010).

Lines 287-294: The factors you have identified here (a-d: (a) aerosol extinction measurements, (b) estimation of particle size distribution, (c) real part of refractive index, (d) calibration of satellite measured TOA radiances) however are significant sources of uncertainty in the satellite retrieval of SSA. How much do these factors affect the uncertainty of your SSA retrievals with this algorithm?

Lines 309-311: How did you arrive at an uncertainty estimate of 0.01 for surface reflectance? This is a very small uncertainty in my opinion since surface reflectance varies seasonally (vegetation phenology), and also as a function of view and solar zenith angle within a day.  Additionally, how did you arrive at +-1 km for uncertainty in Aerosol Layer Height?

Lines 312-313: Why is the SSA satellite retrieval uncertainty expected to increase with increasing wavelength, just because AOD is less at larger wavelengths or some other reason? If it is related to AOD only then the dust cases would not show a decrease in SSA uncertainty for the longer wavelengths. Please clarify in the text.

Lines 316-318: “Our analysis shows that for small τ440 (~0.2), the error in retrieved SSA is much higher (> +-0.05) for visible wavelengths, while that in the near-UV region reaches up to +-0.03.”  Again, is this because the AOD is higher in the UV wavelengths? I am surprised that the ALH effect which is much greater in the UV does not counter the AOD wavelength dependence. If it is AOD wavelength dependence that is the main factor then this statement is not true for cases with low values of the Angstrom Exponent.

Lines 326-328: The way you have written this section it appears to me that you have only considered the factors of uncertainties in ALH and surface reflectance combined in your estimates of SSA uncertainty. It seems that you have not considered the effects of 0.01 uncertainty in AOD as measured by AERONET and this does not even factor in the spatial variance of AOD over the OMI pixel size. Satellite sensor calibration is also not mentioned in your written description of these estimates of SSA uncertainty. What value of satellite sensor calibration uncertainty did you use? Did you include all of these sources of uncertainty in your calculations but just failed to document them in the text of this paper, or vice versa?

Line 344: The aerosols at the Arica site are definitely not biomass burning aerosols as you suggest here. They are dominated by sulfate emissions from copper smelters and therefore non-absorbing.  See Eck et al (2012) for a discussion of fine mode size and SSA for this site as follows: "...typical of most retrievals at Arica, where the average SSA is 0.98 for all wavelengths, from nearly 400 retrievals from 1998 to 2000, where AOD (440 nm) >0.4. This is consistent with the principal aerosol sources in the Arica region, as the SO2 emissions from copper smelting create sulfate particles that are non-absorbing"

Line 345: Note that Sao Paulo is a major urbanized region (one of the largest on Earth) and therefore the primary aerosol type is urban-industrial, not biomass burning.

Lines 357-359: Why do you even include a discussion of urban aerosols in this section since this section is titled 5.1 Biomass Burning? Have you averaged the Arica retrieval results into the urban or into the biomass burning category in this confusing paragraph? Note that the Arica site is neither in an urban region nor is it a biomass burning site. In my opinion the Arica, Sao Paulo and CEILAP-BA sites should be dropped from this subsection or you should rename the title of section 5.1.

Lines 360-362: Please note that the higher absorption at the CUIABA sites are due to biomass burning of Cerrado vegetation (similar to wooded savanna) which exhibits more flaming phase combustion therefore more BC than the predominately smoldering combustion at the tropical rain forest sites (see Schafer et al 2008). Your lumping the CUIABA site aerosol type into the urban category of Sao Paulo is erroneous.

There are many errors in Table 3 for South America in my opinion. You list the Rio Branco site as urban when it is located in a rural region dominated by biomass burning emissions. The Arica site is missing from Table 3 so why even include it in the averaging in your analysis in this section? You also list a high percentage of retrievals for all of the biomass burning sites in South America as urban which is erroneous. Why do you classify some smoke cases as carbonaceous and some as urban in Table 3? This apparent mis-classification needs to be discussed in the text.

Line 365: Again, it is an odd choice to mix this Pretoria site (urban) in with two rural sites that are dominated by biomass burning aerosols, in a section titled Biomass Burning.

Lines 365-366: There are very few natural forest fires in this southern Africa region. Most biomass burning emissions are from savanna burning initiated by farmers or livestock grazers with minor contribution of crop residue burning. Please refer to Eck et al. (2001 & 2003) for discussion of biomass fuel types in southern Africa.

Line 374-375: It seems very doubtful that these relatively high AOD cases in southern Africa are urban aerosol dominated although there may be some mixtures. This urban versus carbonaceous type classification is very confusing and needs to be addressed before these regions are analyzed/introduced in these sections.

Lines 377-378: Please give the spectral range of AAE here for these values and also note that for the visible to NIR spectral region the AAE of urban or biomass burning types is never as high as 2.2 in the published literature.

Lines 384-385: In northern Australia it is not just savanna but also woodland and forests.

Lines 396-397: Please be clear here, are these values AAE. The text should be easier to read without constantly referring to the Figure.

Line 397: Note that neither of these 2 sites (Jabiru and Lake_Argyle) are strongly influenced by urban aerosols therefore this classification seems erroneous.

Line 402: Note that the Saada site is adjacent to the city of Marrakesh (~1 million pop.) therefore influenced by urban emissions while the Tamanrasset site is in a rural region in the middle of the Sahara.

Lines 422-425: This is a completely erroneous reading/interpretation of Eck et al. (2003). There have never been smoke particles in Africa with AE ranging from 0.2 to 0.5 from AERONET measured AOD spectra. You need to eliminate this false statement. Smoke/dust mixtures may have a wide range of AE but these mixtures cannot be called ‘smoke particles’. Also note that Eck et al. (2003) reports no AOD measurements or retrievals of smoke properties from West Africa. This paper is focused on Southern Africa biomass burning aerosols, a completely different region. The inaccurate and misleading use of references in this paper is somewhat disturbing.

From Eck et al. (2001), from page 3442 of the ZIBBEE paper: “Liousse et al. [1995] found AE (computed for similar wavelengths: 450, 650, and 850 nm AOD) for savanna burning smoke to range from 0.84 for aged smoke to 1.42 for fresh smoke at Lamto, Ivory Coast. However, in that West African location it is possible that these relatively low AE values may be influenced by the presence of Sahelian/Saharan coarse mode dust as a second aerosol type.”

Lines 429-433: See Eck et al. (2010) for data and discussion of the highly absorbing smoke cases at the Ilorin site, from grassland and treed savanna biomass burning aerosols in the winter season (DJF). Include some comparison in the text.

Line 438: This section should be renamed to "Middle East/North Africa/Arabian Peninsula' since only 1 site out of the 4 sites you have listed is actually geographically located on the Arabian Peninsula (per Wikipedia).

Line 459: Classifying the aerosol at Missoula and Rimrock sites in the urban/industrial category (section 5.3 here) is somewhat absurd. These are far from having significant sources of urban/industrial pollution and just about all of the cases with moderate to high AOD (>0.4 at 440 nm) could readily be attributed to biomass burning sources.

Line 473: No, there is no similarity between the aerosol sources of the sites you analyze in your section 5.3.1 Western North America with the aerosol sources / types in Eastern North America. As mentioned above 2 of the 3 sites you list for western NA are dominated by biomass burning aerosol sources.

Line 477-478: You should mention that 4 of the 5 sites in this subsection are in a relatively small area, the central mid-Atlantic US (3 in central MD including GSFC), while one site is in the mid-West US (Bondville, Illinois).

Line 482-484: This statement is not clear, please elaborate here to better explain what you are referring to.

Line 487: There are 17 sites that are averaged together here for Europe, much more than for any other region. Some discussion regarding differences between sites is warranted. Also it seems probable that the somewhat noisy looking wavelength dependence of SSA for the SON season is likely due to a small sample size. The number of days of observations should be included in these plots, Figures 6-9, and when the sample size is small and therefore statistically weak it should be noted.

Line 492-493: However this organic carbon absorption that you suggest is inconsistent with the AAE values in the same plot, as these AAE values which are close to 1 are typical values associated with black carbon absorption. Please discuss this apparent discrepancy and explain why you interpret such an AAE value as being associated with organic carbon absorption. Also include references to support your interpretation.

Line 504: This section is very odd, in that you only include one site and this site is one of the largest urbanized regions in the world (Mexico City) that also happens to be located in Central America. Mid-Atlantic North America which is your title of this subsection is Maryland, Virginia etc. Mexico City is equally distant from the Pacific Ocean and Caribbean Sea therefore using mid-Atlantic to describe its location is beyond strange. It seems that the authors are either extremely careless with geographical labeling (or very careless in writing) or need to become more familiar with commonly used regional/geographical names.

Line 509-510: “It is clearly evident that such absorption curve and seasonal variation is a result of prevailing mixture of aerosols.” This is a confusing sentence. Please elaborate what you mean here by a mixture of aerosols since elsewhere you define an AE range as mixtures due to fine/coarse particle size mixtures.

Line 512-514: Please discuss the uncertainty in your computed values of AAE somewhere in this paper. The AAE parameter is highly susceptible to small errors in AOD and SSA and the uncertainty in both AAE  and AE increases as the wavelength range decreases due to the resulting small differences in AOD that approach the uncertainty level in the AOD itself. The uncertainty in the AAE for 354 to 388 nm is therefore very high for this reason of relatively small differences in AOD between these two close wavelengths (only 34 nm apart).

Line 518: This reference makes no sense as there is no mention of Mexico City in this particular paper (Eck et al. 1998). The use of inappropriate or plain wrong references in this paper is disconcerting at best. The authors need to follow normal standards for correct referencing for a journal of high standards such as AMT.

Line 521-522: Please note in the text that for cases of AOD(440)>0.4 as shown in this figure the two sites in China (Beijing and Xianghe, only ~60 km apart) completely dominate the statistics of these 4 site averages. The AOD in these China sites are very high while the AOD levels for the two sites in Japan are much lower such that there are relatively few cases in Japan that exceed the AOD threshold of 0.4. Averaging multiple sites in these types of plots can sometimes be justified but in this case it is very misleading.

Line 522: If you want to label this Figure 9b plot ‘Eastern China’ then leave the 2 sites in Japan out of the data averaging. In fact, you write this entire section as though the Japanese sites are not included so why did you average these data in with the Chinese sites data at all?

Line 525-526: Note that the variation of AAE as a function of fine mode fraction (FMF) for both the Xianghe and Beijing sites are shown in Eck et al. (2010). For low FMF which is equivalent to your dust category, the AAE for both sites was ~2.5 which is consistent with most AAE values in for dust aerosol in the published literature. However your dust value of AAE is much lower than that and therefore you should explain why it is anomalously low for this aerosol type.

Line 534-535: “The spectral behavior of urban aerosols is similar to carbonaceous aerosols with decrease in magnitude of average SSA, AOD, and UV-Vis AAE.” This sentence seems to be somewhat incomplete, as what decrease in these 3 parameters are you referring to? A spectral decrease? Please clarify what you mean here.

Lines 559-561: I have not been convinced that your separation of urban versus carbonaceous aerosol types is robust, or has much relation to other classifications in the published literature. There is a lack of discussion of the accuracy of the separation of these two fine mode aerosol types.

Lines 578-581: However you inexplicably have averaged the rainforest burning dominated sites with the cerrado vegetation burning dominated site (CUIABA). Note that cerrado vegetation is much like wooded savanna in southern Africa and that is why the SSA values for Cuiaba are lower than for the other South American biomass burning sites, see Schafer et al. 2008. You also have made a distinction between carbonaceous and urban types for these SA biomass burning sites which does not make any sense except for the Sao Paulo site (which has very few retrievals with AOD(440)>0.4). In short Figure 6 is quite misleading since you have mixed apples and oranges so to speak and then average the whole ensemble of sites.

Line 585: This would make more sense to add 'the savanna grasses' to ‘in the central region’ here.

Line 588-589: Biomass burning is always a mixture of flaming and smoldering combustion, but when the fraction of flaming combustion increases then the black carbon production increases and the SSA decreases. Please correct this sentence since you imply that this Australian vegetation burns entirely in the flaming phase, which is false.

Line 595: Again the uncertainty in AAE for such a narrow wavelength range is much greater than for the 358 to 646 nm range. You should (as a computational exercise) vary the AOD by +0.02 at 354 nm and by -0.02 at 388 nm and see how different the AAE is for this expected  AOD uncertainty alone. Add to this the uncertainty in SSA and it may not be very clear if the AAE is really that different in the UV alone from the UV-Visible wavelength range values, or just within the uncertainty error bars.

Lines 597-598: “flaming combustion prevails” is too strong here as there is just a higher fraction of flaming combustion, while smoldering still produces more than half of the smoke aerosol.

Line 599: Please note that for the Ilorin site, Eck et al. (2010) found AAE to vary from 1.37 for fine mode dominated to 2.1 for dust dominated as a function of fine mode fraction. Please put your results in that context of variation as a function of Angstrom Exponent or FMF.

Line 603-605: Mixing with coarse mode aerosol is only a part of the explanation of lower AE in these sites. Another factor that you should mention is that the fine mode particle size is larger at these sites due to aging processes of coagulation, condensation, hygroscopic growth and cloud processing.

Lines 610-612: I would suggest that you change 'often found' to 'sometimes found' since this varies greatly depending on the continent, region and season.

Lines 612-614: The reference of Torres at al. (2002) for this phenomenon (change in desert dust SSA during transport over the Atlantic from north Africa) is not very robust since the very large TOMS pixel size is very susceptible to partial cloud contamination which would be much more of an issue over the Atlantic Ocean than over the desert source regions and would thus yield higher SSA over the ocean. Additionally the aerosol height utilized in these retrievals may have introduced significant uncertainty that may also differ for these two regions.

Lines 615-618: Your suggested explanation seems very unlikely to be true. See in Eck et al. (2010) the section on the Ilorin site where the fine mode (biomass burning) aerosols are highly absorbing in the Sahel/Sudanian zones since this is primarily grassland and savanna burning with a relatively high contribution from flaming combustion (more BC produced). The SSA decreased at this site as more fine mode smoke was mixed with the dust.  Additionally some desert dust aerosol sources that advects into the Sahel region are relatively weakly absorbing. The Bodele Depression which is perhaps the largest single dust source is an example of weakly absorbing mineral dust since some of the material is diatomaceous sediment which does not contain iron oxides. See Eck et al. (2010) and Di Biagio et al. (2019) for more information on the SSA of this specific source and other dust sources.Lines 618-619: These dust SSA values from your retrievals should be compared to the values in the literature such as Di Biagio et al. (2019).

Line 620: Please include in the Figure 11 cation an explanation of the 5.2 and 6.2 with the X inside the circle symbol.

Lines 625-628: It should be noted that this is such a wide range of AAE for dust (1.5 to 3.5) that it could be argued that even with very large retrieval uncertainties you still can fall within this range. Please defend the value in these dust AAE retrievals including estimates of the uncertainties.

Lines 630-632: Same comment as immediately above: It should be noted that this is such a wide range of AAE for dust in the literature that it could be argued that even with very large retrieval uncertainties you still can fall within this range.

Line 638: Please include in the Figure 12 cation an explanation of the 5.1 with the X inside the circle symbol.

Line 665: Please provide the number of days of data for each site in Figure 13 (for both the AERONET and satellite retrievals) so that the relative statistical robustness may be evaluated.

Line 666-667: Please note in the text that the uncertainty in the AERONET retrieved SSA is ~0.03 at AOD(440)~0.4 but that this uncertainty decreases for higher AOD levels (see Sinyuk et al. 2020: Fig 22 and Tables 14-17 ).

Line 667-668: This must be an error, or else you need to change the value of 0.05. Why only include the effect of surface reflectance in your estimate of retrieval uncertainty of SSA from satellite when there are several other significant sources of uncertainty such as AOD, aerosol layer height, aerosol size distribution, refractive index and satellite sensor calibration.

Line 671-673: However, please note in the text that the overlapping error bars comprise a wide range of +-0.08 which is a large fraction of the expected parameter space for aerosol single scattering albedo (~0.8 to 0.99).

Lines 676-678: Please be aware that surface reflectance is a second order source of uncertainty for AERONET since the instrument is upward scanning, and the primary sources of uncertainty for AERONET retrievals of SSA are sky radiance calibration (which is independent of the direct sun cal), solar flux and AOD. For downward viewing satellite retrievals however the surface reflectance is a major source of uncertainty and therefore the way this sentence is written might possibly be quite misleading.

Lines 679-681: “For absorbing aerosols, the SSA differences observed here is likely a result of different surface reflectances data employed by the two data sets.” No your statement here is not true at all. See my comment immediately above.

Lines 691-692: It should be noted that the uncertainty in SSA for AERONET is somewhat higher at 675 nm than at 440 nm for fine mode aerosol cases (for a given value of 440 nm AOD; see Sinyuk et al. (2020); Figure 22). This is due to the lower AOD at 675 nm therefore lower absorption signal in the data. A similar increase in satellite uncertainty at the longer wavelength is inevitable for the same reason.

Lines 703-705: It should be noted here in the text that the uncertainty in AERONET retrieved SSA decreases as AOD increases (Sunyuk et al. (2020) and that the same is true for the satellite retrievals since at high AOD the aerosol signal overwhelms the sources of uncertainty such as surface BRDF, calibration and AOD.

Lines 710-712: This statement is only partially true and needs to be revised. For dust aerosol at low Angstrom Exponent (AE) the AERONET retrieval imposes very weak constraint on the spectral variation of the imaginary refractive index since the AOD is high at all wavelengths and the absorption signal is therefore sufficient at all 4 retrieval wavelengths (440, 675, 9870 and 1020 nm). The sky radiances are fit well at all wavelengths for dust cases and therefore the retrieval is robust at all wavelengths. For fine mode dominated aerosol at high AE values however AERONET version 3 imposes a constraint on the spectral variation of imaginary refractive index. This constraint for high AE retrievals is based on the fact that black carbon exhibits minimal wavelength dependence in imaginary refractive index, plus the fact that for large AE the AOD at the longer wavelengths is quite low and therefore the aerosol absorption signal is insufficient for a robust retrieval at the long wavelengths.

Lines 745-746: This statement is misleading as it suggests that the SSA is spectrally flat for the entire wavelength range. However in Table 3 the values for Mongu show significantly higher SSA at 466 nm than at 646 nm.

Line 750-752: Again you are exhibiting a tendency to ignore the fact that in your Table 3 the SSA retrievals at 466 nm are significantly higher than at 646 nm for both the Alta Floresta and Missoula sites. Please explain the lack of consistency in your retrieval data versus your interpretations in the text of the paper.

Line 758-759: You call the Arabian Peninsula a biomass burning region? Please back up this interpretation with published references and some substantial analysis. Also regions such as eastern China and Northern India do have biomass burning in specific seasons however they are more strongly dominated by other emission sources for more months of the year and therefore are not considered biomass burning regions in the literature (except for specific ~1 month periods).

Lines 761-762: Why did you only mention 466 nm SSA for this region in this Conclusions section? This section has been inconsistently written.

Line 774-776: It is somewhat absurd to categorize the CUIABA site as representative of the urban aerosol type. The cases where the AOD at 440 nm exceed 0.4 at Cuiaba are almost always dominated by biomass burning smoke from cerrado vegetation plus some long-range transport of smoke from rain forest burning to the north (see Schafer et al., 2008).

Line 779-781: Please provide some references to back up this interpretation, including discussion of the fuel types in most of Europe you are alluding to.

Line 795: This does not make a lot of sense and needs to be supported with additional references and analysis. The uncertainty of the retrievals at low AOD levels is inherently greater therefore this does not have much basis in rigorous analysis.

Lines 806-807: “Given the lack of aerosol absorption information at near-UV wavelengths in the existing AERONET record…” This is not a completely accurate statement. There is currently no AERONET product available on absorption at 380 nm but the sky radiances measurements have been made for years at 380 nm from many instruments in the global network. A retrieval that includes the 380 nm imaginary refractive index and SSA is currently under development by the AERONET project. This statement needs to be revised and expanded to reflect this additional information.

Review of Retrieval of UV-Visible aerosol absorption using AERONET and OMI-MODIS synergy: Spatial and temporal variability across major aerosol environments by Kayetha et al.

This manuscript presents a method for deriving aerosol absorption from a combination of satellite and ground-based remote sensing measurements. The single scattering albedo is derived in five wavelengths in the UV-Visible range. The method can be applied over locations with co-located satellite and ground-based measurements. The method adds UV information on top of the existing data from AERONET. The method is applied to a data record over more than 100 sites globally for the period 2005 – 2016. Results are discussed per region.

Overall, I think this is interesting work which has a lot of potential. However, the way that it is presented can be much improved. I recommend splitting this manuscript in two parts, where part I describes the method, sensitivity study, case studies and validation, and the part II describes global and regional results. Part I would fit for AMT, whereas part II would better fit in ACP or a similar journal.  Part I should answer questions like, what is the added value of this method with respect to the standard AERONET retrievals of SSA? How can the retrieved spectral behavior of the SSA be explained by the expected refractive index? That is why I encourage the authors to withdraw the current manuscript and resubmit it in two improved parts. Because of this recommendation I will not provide detailed textual comments on the current manuscript, but rather indicate where the work needs further improvements.

Section 3.

Before section 3.1 a text needs to be added that introduces the methods physical background. E.g. what determines the radiation measured at ground-based and satellite level. A diagram would be useful for this.

Section 3.1.

The selected wavelengths in the visible are affected by NO2 and ozone absorption. However, these absorptions are apparently not part of the LUT design. These should be included, or the authors should justify why these absorptions can be neglected.

In the end, the method uses reflectances from the satellite instruments and derives the SSA using the LUT. Therefore, the SSA is the depended parameter for the method. For this reason, the axes of Figure 3 should be switched (x-axis reflectance, y-axis SSA). This will immediately visualize the problem for low AOD, where the method will be very noisy.

Section 3.3.

It is unclear where the surface pressure information is coming from.

Section 4.

The sensitivity analysis is incomplete. The approach to only perform a sensitivity analysis for parameters which are controlled in the retrieval is clearly not acceptable and also not true, because the real part of the refractive index and other aerosol model parameters are also selected as part of the algorithm. So, all identified parameters should be included in the sensitivity analysis, along with the surface pressure, signal to noise of the instruments, the (tropospheric) ozone column and the NO2 column. This should be presented in graphical way to convince the reader that the method is sound.

While the analysis of the GSFC method is good, I am left with a number of questions. First of all, the SSA should significantly lower values at 646 nm compared to the other wavelengths. Is this realistic. Is this in line with our knowledge about the refractive index in the visible? Is it possible that this is related to ozone absorption in the visible?

I propose that on top of the GSFC analysis, the authors present 3-6 cases studies of single retrievals over different sites, where for a given day for which both the AERONET SSA and the combined SSA retrievals are available. These cases should cover both good and bad comparisons and discuss the reasons for these results. This gives an opportunity to demonstrate the added value of the satellite method.

Section 7

Logically, the next section would be the validation (currently section 7).

The authors have chosen to compare only the 466 and 646 nm SSA to AERONET. Also comparison between 388 nm retrievals and 440 nm AERONET shall be included. Although the spectral distance is larger than for 466 nm (MODIS), it is the best way of also including the OMI retrievals in the validation.

The validation data should also be split by AOD bin. In this way I hope that some correlation can be demonstrated for the medium and high AOD values. Alternatively, times series of data sets over sites with large variability in SSA could be presented to convince the reader that the retrievals add value wrt to the results from AERONET. The current Figures 14 and 15 (top plots) are not very convincing, however the representation of Figure 15 (bottom) is much better.

Section 5

In section 5 regional results are presented.

There are different ways of computing the average SSA. If I understand it correctly, the unweighted average SSA is presented. Alternatively, given the dependence of the accuracy on the AOD, the AOD could be used as weights. This would be equivalent to computing the mean SSA as (1- mean(AAOD)/mean(AOD). Also, mean and standard deviations can be significantly affected by outliers, whereas the median and percentiles are more robust statistics. How would the presented results be affected if other methods of computing statistics are used?

The results are presented per region. However, given the poor spatial sampling the statistics will not be representative for the whole region. It is therefore questionable if this analysis is useful at all. For example, there is a huge difference between aerosols on the Californian coast and those of continental Canada, however they are in the region. The same is true Mediterranean sites and sites in Northern Europe as well as other regions. The authors should rethink how these data can be best presented, beyond the current split in regions. I suggest starting with some global maps where the data is plotted per season and aerosol type. Maybe as a circle of which a quarter is used per season or something similar.

Section 5 is very hard to read, as it mixes observations with speculations. Also results from large regions are discussed in terms of very local phenomena. Here a clear choice should be made by authors to either discuss the global distribution of the SSA, or to dive into the details of one or more regions. Now the scope is somewhere in between and that doesn’t work for me. Furthermore, it should be clearly identified when the other claim that the data prove something, or when they speculate about possible explanations.

In many cases in Figure 6/7/8/9 a significantly lower SSA at 648 nm is reported as compared to the other wavelengths (e.g. 6a/b, 8 a/b/c, 9 b/c). Do the authors have an explanation of this, in terms of the spectral behavior of the refractive index? Is this reported in other studies? I am not convinced that this is not caused by measurement errors.

Section 6

I am not convinced by the analysis based on the mean AAE. Overall, satellite retrievals of the AE are difficult and the AAE is much more difficult. Also note that AAE is a combination of the spectral behavior of the AOD and that of (1-SSA). Before concluding jumping to conclusions on the AAE, the authors should first provide that there is any value in these mean AAE results.

Comments on Tables and Figures:

Tables 3. I propose remove this table to the supplemental material and make it available as complete data set (e.g. HDF5 file(s) or excel files(s) ) for all the sites, containing both individual retrievals and statistics.

Figure 1. Move this to supplemental material.

Figure 10-12, replace the bars by violin plots or box-whisker plots.

Figure 13, replace the plots by box and whisker plots using the spread of the points instead of the assumed uncertainty.

Review of "Retrieval of UV-Visible aerosol absorption using AERONET and OMI-MODIS synergy: Spatial and temporal variability across major aerosol environments."

General Comments

This work attempts to use the OMI-MODIS synergetic data and the measured and retrieved products from the AERONET measurements from different Earth regions. This work used UV-Vis spectral band measurements and retrieved the SSA separately for each bands using a LUT approach. Seasonal variability of the SSA in different regions worldwide is studied using the AERONET products and retrieved SSA and AAE from this work. 

In general, the attempt to quantify the aerosol absorption using the retrieved SSA in UV-Vis multi-year multi-dataset is interesting and useful for the scientific community. However, the methodology and the manuscript need to be improved before considering for publication. 

Specific comments

• AERONET measurement-based climatology is used for representing the particle size distribution in the retrievals. However, no discussion based on it was found in the manuscript. It would be ideal to provide the details of the PSD used in the retrievals. You could create similar plots like the figure 6,7,8, and 9; instead of SSA, you could plot the mean PSD distribution with the error bar as the SD (This can go to the supplemental material).
• Authors should discuss the criterion for classifying AERONET stations into biomass burning, dust, Urban/Industrial, and mixed aerosol in detail. E.g., the Sao Paulo station classified as biomass burning is wrong since it is a megalopolis. 
• Section 5 is not very clear. Please consider rewriting it to avoid ambiguity. 
• Consider moving Table 3 to supplemental material.
• A discussion on the correction of atmospheric gas absorption before the aerosols retrievals is needed

L37: What models are authors discussing here? GCM? Global Earth System models? Please specify it.

L55: Did you mean by SSA retrievals?. Because it is clear that there is a long term measurement of AOD at UV bands from AERONET stations.

L71-73: Specifically mention these retrieval algorithms with citation. Are you specifically talking about MODIS operational algorithms? 

L164: Provide geometry information. Like SZA, VZA, RAA used for simulating TOA observation for the GSFC site.

L165-167: This is calculated for a particular satellite-sun geometry, and it can vary considerably in the analysis used in this study. How can you generalize this sensitivity of SSA and AOD to other satellite-sun geometries?

L177-180: Can you calculate the uncertainty in SSA due to the assumption made here?. It will be important, especially since retrieved SSA differs for regions with different spectral signatures and magnitude.

L184-186: Did you use the closest observation as collocated data? 

L203: Why the inversion is done independently? You could use the multiple wavelength information in minimization. What is the advantage of doing inversion independently over different spectral bands?. In the discussion section, you are using only the data points with retrievals for all bands. It makes sense to use all the information together to do a retrieval. Please specify the rationale behind not using this method. 

L212-214: This is not correct for the blue band. The Rayleigh signal will mainly dominate the signal in this band when the aerosol loading is low.

L270-271: How did you come up with these numbers? Provide references for this. It would be best to use the uncertainty specified by the surface albedo product you are using in the retrieval.

L282: This achievable accuracy depends on the accuracy of surface reflectance products used in this study. It should be calculated based on the accuracy mentioned for the surface product used and should differ based on surface type, and it will become dominant in the longer wavelengths. 

L325-327: I can't see a plot for the DJF season. Is this a typo?

L366-367: This can be verified using the plots of PSD from AERONET retrievals.

L369-370: Is this just the author's opinion? What are the pieces of evidence for this?

L395-396: For the case of JJA, the SSA is increasing with the wavelength for the region 340-388 nm. It is contradicting to the sentence in these lines.

L405: Typical urban spectrum based on what work? Cite the literature!

L422-423: This has to be verified using PSD.

L518-521: The SSA difference for those two regions can be due to the error from surface reflectance estimation. 

L521-522: This is just another hypothesis; there is no proof here.

L555: What kind of interpolation?

L560: In figure 13, why there is no STD for the UV wavelengths?

L561-563: Show the SSA as a box plot with error bars. It will give us an idea of the spread of the SSA values for the averaging AERONET station.

L576: Move the annotations to the lower right corner of the plots in Figure 14.

L601: You could cite Dubovik et al., 2006 to describe the smoothness parameter imposed in the retrieval.

L601-603: Another difference is the use of multi-angular- multispectral information in the AERONET retrieval, Whereas the work presented here used the PSD and real refractive index from those retrieval and basically, the imaginary part of the refractive index is varied to retrieve the SSA.

L616-618: Define the range of parameters used for this sensitivity study.

L661-662: These two AERONET stations is in the biomass burning aerosols category. Then why you have it here in Urban?

L687: Specify that no SSA and IRI retrievals available at the moment.

Technical corrections

L40-41: Is it a 50% decrease?

L59: Citation required for this statement.

L72: 'observations in the visible assume'. It should be 'visible spectrum' instead of just 'visible'.

L124:In Table 1 it is mentioned that version 2 is used. Which one is correct?

L198-199: Did you mean an exponential distribution?. Because the peak is on the surface.

L290: Is it τ₄₄₀ ≥0.2?

L305: The values given for SSA are mean; specify it explicitly with the SD as uncertainty.

L441: In figure 9b, instead of 'northeastern china', it is mentioned as 'eastern china.'