Synopsis:  This is a pretty straightforward paper comparing passive Aerosol Layer Height (ALH) retreivals generated from TROPOMI, GEMS and EPIC over two Asian domains for signciant dust and smoke events.  Comparisons are made to AERONET for AOD, CALIPSO CALIOP products for the vertical centroid, and intercomparring heights between each other. Also included are comparisons of the nature of diurnal retrievals and are two example cases.  As the authors note, the oxygen band retrievals are coming into their own, and while they have limited measurement degrees of freedom, they do have coverage that cannot be achieved with lidar systems.   Overall, the paper is certainly appropriate to AMT and can have importance to the community. This said, I do think the paper requires major revisions.  These revisions are in two prime areas.   First, they authors delicately ignore a number of sampling considerations throughout the paper.  Notably, when they compare the passive products to CALIOP, implicit in that comparison is excellent viewing conditions.   If there was say thin cirrus, there would not be a CALIOP retrieval to compare.   For the intercomarison between retrievals without CALIOP, there clearly is a wider and higher distribution of heights.   Notable are little “isolate” retrievals with very high height retrievals.   The authors need to examine this data population closely. This may require a lot of hand analysis.

A second concern is the reported diurnal cycle.  Certainly we expect the diurnal cycle to be important to retrievals such as GEMS and EPIC that have widely varying scattering angles for retrieval physics if not reality, and thus can be important to include in a paper like this.  Indeed, all of the major AOD retrievals from geostationary have significant diurnal biases.  However, the authors gloss through potential artifacts to explain diurnal differences, and are highly suggestive that the diurnal cycle from EPIC (their retrieval mind you) is real.  However their explanation of the nature of diurnal aerosol height is flatly wrong-they suggest that the maximum aerosol height should be around solar noon due to PBL mixing.   First, mixing will sustain itself through the afternoon and thus we would not expect a solar noon peak.  Second, their product uses aerosol index and they are not as sensitive to the PBL anyways. Thus physically their explanation does not make sense.    As EPIC stares at the sunny hemisphere of the earth, they may have an unaccounted for diurnal scattering angle issue in their RT code that causes symmetry along solar noon, or they may have a cosine response/resolution issue and perhaps also an associated resolution-based cloud mask bias.  Now in full transparency, I am a co-author on a paper from this group that reported a strong diurnal height cycle of dust off of Africa.  This said, as part of my contribution I pushed for softer language and examination of biases.   As we see the exact same behavior here, I am thinking even more about diurnal bias in their EPIC retrievals. They have to come clean up front, and say they have no current way to evaluate which is correct, noting the possible artifacts, or remove the section all together. I strongly encourage the authors to look at some ground based lidar data in the region to verify their algorithm.  There are plenty of Japanese and NASA lidars in this region that they can use.

In addition to these two science concerns, the paper has many minor copy editing typos, language irregularities, verb tense, dropping articles etc., Just an example, starting a paragraph on line 84 “The light travels longer path when aerosols locate at lower altitude than those at higher altitude, leading to more absorption from more O2 molecules in longer path (Ding et al., 2016; Xu et al., 2019).” The paper has sentences like this in almost every paragraph, and frequently drops articles (the/a/an).  I suggest the authors utilize university copy editing programs before the paper can proceed.

Specific notes

Abstract Line 15 (And Intro lineS 72-76).  I think from the beginning the authors should be a little more modest about what an aerosol layer height retrieval means.  It does indeed provide information, but it is only a single degree of freedom in what can be a complex aerosol vertical structure. From the beginning, Aerosol Layer Height (ALH throughout) can be a bit misleading, and largely based on a false presupposition-that there is one layer.  I am ok with language like “scale height”, or “aerosol centroid”, but honestly ALH has always bugged me as a labeled variable. Indeed, as noted in line 23 they have to make adjustments in definitions from the different product lines. I only ask they authors be mindful of this throughout the paper.

Abstract line 24-25, you may want to mention how the products cross correlate.

Abstract line 30 “EPIC and TROPOMI tend to overestimate AOD by 0.33 km and 0.23 km, respectively, in dust cases”  I am not sure where the km fits into this, as AOD is unitless.

Introduction. Line 52 “Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) on board

with the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) platform detects aerosol backscattering extinction profile with fine vertical resolution (Winker et al., 2013).”  This is a bit of unusual language.  What CALIOP measures is attenuated backscatter.  It retrieves aerosol extinction profiles, but these are retrievals and can have uncertainties of their own.

Introduction. Line 55.   “With the retirement of CALIPSO in August 2023, passive remote

sensing will become the only routine technique from space for filling the data gap of measuring aerosol vertical distribution before next lidar dedicated to measure aerosols are launched into space.”  Technically this is not true, as the Chinese have an HSRL in space.  But they don’t release the data.  We will see if they ever do.

Methods, Section 2.3.  I think throughout the paper, it needs to be emphasized that “golden days” are being used in this evaluation.  I have no objection to this per say, as long as they make it clear in the abstract and introduction that these results are for ideal viewing conditions, and indeed day to day “mileage my vary considerably” Indeed, the very use of the CALIPSO as a verification dataset implies that they don’t have to worry so much about things like cirrus or other cloud contamination.

Results

Line 295: I am not entirely sure how meaningful the stats are here “For dust cases, both TROPOMI and EPIC AOD exhibit a positive bias compared to AERONET AOD, with values of 0.23 and 0.33 for TROPOMI and EPIC, respectively.” As these are for a distribution of AODs with different densities across AOD values.    You can calculate bias as a function of AOD, or if it is linear enough a slope bias.

Line 320-323:  The authors should probably use more direct language, that based on Figure 3, the TROPOMI and EPIC dust AOD products were quite high biased (presumably as stated line 323 due to surface reflectance), and GEMS was low perhaps due to the optical model.  Thus, when you cross compare, of course there is a massive bias between them. But can you also check this by looking only at data over water?

Line 345-352: AI can be a can of worms.  There are a host of other issues than those noted, including pressure assumption differences, not just land reflectance but altitude models, resolution differences resulting in different cloud effects, etc. probably you need to calculate it yourself consistently.

Line 354-Figure 5.  Why is there no TROPOMI vs EPIC plot?  I think it would be good to cross characterize everything.

Line 383-Figure 6. Why do the axis go much further than where there is data?   Maybe set to lesser spread and put the results in a table? Also, why not do all of the passive sensors for the different AI ranges?

Line 408-Figure 7. It would be good to verify what is going on with AI cases bigger than 4, and what role clouds may have in increasing these values.   There are no CALIPSO verification cases with AOCH values over 4-5 km. I think it will be necessary to show that these cases where the retrieved AOCH are above such values are real, and not due to some artifact, such as thin cirrus above or unmasked low clouds below. In their case study (e.g, Figure 9) spot retrievals of very high AOCH are visible.   The authors should look in detail as to what is going on there.

Line 430-Figure 8 discussion.   The diurnal version of EPIC  looks pretty symmetrical around noon.  We have talked about this before, that this may be indicative of a retrieval bias. I don’t think you can quickly dismiss this as being part of the PBL cycle, especially since the focus is on absorbing aerosol layers above the PBL.  Regardless, the effect of the PBL should maintain its height through the afternoon. The authors really need to incorporate a ground based lidar into these diurnal analyses.  In fact, they might consider dropping this section until that can be done.

Line 485-Figure 11.  Please add a local solar time to the x-axis.

Line 518-Figure 13(b).  The crosses for GEMS, TROPOMI and EPIC are very hard to read. The authors may want to change the color scheme and line thickness here and back on Figure 10 to make it consistent.

In this study, Kim et al compared aerosol layer heights products from several satellite instruments such as GEMS, EPIC and TROPOMI using retrieval algorithms all based on oxygen absorption bands. O2-A and B bands are used for TROPOMI and EPIC, O2-O2 band is used for GEMS. To have consistent comparisons, the aerosol layer heights are converted with a similar definition. Cases studies including dust and smoke over several regions in Asia are also discussed. Discrepancies between the products of the three instruments may reveal limitations in assumed aerosol and surface models and shed new lights to improve ALH retrievals. In general, this work fits the scope of AMT, and provide detailed and thorough analysis with useful results. I have several suggestive comments which may help improve the clarity of this work.

General comments:

1. There are many acronyms, and many of them are not defined when mentioned the first time. For example: UVAI (the ultraviolet aerosol index), AEH ( aerosol effective height) have been used several times, but only defined in Page 5, Line 146 and 156. I would recommend having a table defining those acronyms.
2. Most conclusions are made by comparing with AERONET and inter-comparisons. However, each product also has uncertainties which relate to the measurement uncertainties and retrieval algorithms. I didn’t find discussions on the accuracy of the measurements from these instruments, and corresponding ALH uncertainties. In principle, the ALH uncertainty from each instrument can be validated using the AERONET data too.
3. Due to the information content, the ALH uncertainty should have strong dependency with the aerosol loading (or AOD). I don’t have a clear understanding on what the AOD range is used in the discussion of ALH from this study, and how that impact the conclusion.
4. The authors suggested that the aerosol and surface model used in the ALH retrievals may cause the discrepancy between different products. It would be useful to add more discussion on how such models impact the ALH retrievals, and how the authors would recommend to improve based on the results from this study.

Detailed comments (some may be duplicated and related to the general comments):

Page 1, line 24 “In comparison with CALIOP ALH, both EPIC and TROPOMI ALH display a high correlation coefficient (R) higher than 0.7 and an overestimation by ~ 0.8 km, whereas GEMS ALH exhibits minimal bias (0.1 km) but a slightly lower correlation with R of 0.64.”

Why there is larger bias with higher correlation? Does this indicate limitation to use correlation as a metric?

Page 1, line 25: UVAI not defined, what is its meaning?

Page 2, Line 63: “…the degrees of freedom for signal (DOFS) increase from 2.1 to 2.8, which becomes sufficient for three parameter retrievals (AOD, aerosol peak height, and aerosol layer thickness)…”

How do you know that the three parameters are the right set of parameters, not other ones, such as SSA, aerosol size, etc?

Page 4. Line 115, although O2-A, O2-B and O2-O2 bands all have sensitivities, I didn’t find any discussion on the measurement uncertainties from the three sensors using those bands?

Page 4, line 125, “Accurate retrieval of ALH requires reliable retrieval of AOD, and past studies have shown that ALH and UVAI relationship can change with AOD (Xu et al., 2017).”

Can you elaborate how the relationship will change? And how that applied to this study?

Page 5, Line 135, “all algorithms assume quasi-Gaussian distribution described by two parameters including centroid height and half width (fixed at 1 km) at half maxima”

It would be useful to show the formula, which can help explain what is a quasi-Gaussian distribution, and half width at half maxima. I feel FWHM (full width at half maximum) is more commonly used. (I saw the formula in later section, you may need to add a reference).

Page 5, Line 156, “aerosol types are classified by the ultraviolet aerosol index (UVAI) and visible aerosol index derived from GEMS observations”

How UVAI is used to classify aerosol types?

Page 5, Line 159. “For LUT generation, aerosols are assumed to be spherical and their particle size distribution, refractive index and fine mode fraction for each aerosol types are derived from global AERONET inversion climatology.”

Can you confirm that whether AERONET aerosol inversion already considered non-spherical aerosols? I believe there are products used non-spherical aerosol model.

Page 6, Line 172, what O2AB-UI algorithm stands for?

Page 6, Line 180, UVAI is defined in previous page.

Page 6, Line 183, “only those pixels covered by lofted layer of absorbing aerosols with UVAI larger than 1.5 and AOD larger than 0.2 (at 680 nm) are analysed.”

Is this the case for all following analysis? Fig 3, seems include AOD as small as 0.1 for all sensors.

Page 6, Line 191, “However, the hyperspectral measurements from TROPOMI, unlike the EPIC measurement in narrow channels, prevent us to applying the EPIC AOCH algorithm in TROPOMI L1B data directly”

So what is the band width for TROPOMI?

Page 7, Line 210, Eq (1), what is the reason not choosing a Gaussian distribution but choose the current form? If a Gaussian distribution is used, FWHM or half width at half maximum can be easily expressed by the standard deviation.

Does the choice of 1km as half width at half maximum impact the ALH results?

Page 8, Line 265 “Hence, the accuracy of each AOD product also influences corresponding ALH retrieval, which will be validated here by the ground-based Aerosol Robotic Network (AERONET) inversions as well.”

Similar to a few previous comments, the aerosol loading itself also impacts ALH retrieval. One example can be found from polarimetric retrievals, such as Gao et al 2023, (https://doi.org/10.5194/amt-16-5863-2023). It would be useful to make it clear how AOD impacts the conclusion in this study.

Page 8, Line 274 “Since TROPOMI and EPIC AOD products are retrieved at the wavelength of 680 nm whereas GEMS AOD products are retrieved at 354, 443, and, 550 nm,…”

Is there any estimation of the AOD accuracy derived from these bands? Which one is more accurate?

Page 8, Line 290 “The observed underestimation of GEMS AOD at 680 nm can be in part due to an overestimation of the Angstrom Exponent (AE), which can be affected from inaccurate particle size or refractive index in the wavelength-dependent aerosol model.”

What aerosol model is used?

Page 11, Line 302, what is the surface model used here?

Page 12, Fig 3. “Satellite data points only with a standard deviation less than 0.3 are shown for spatial consistency.”

How is the standard deviation derived?

Page 13 Line 318, how the dust and smoke cases are separated?

Page 13, Line 321, what dust aerosol model is used, in terms of size, refractive index etc?

Page 14, Fig 4, it seems GEMS AOD has a boundary constraint which make its less than 0.5 most of the time, at least for dust? But it seems smoke case don’t have such constraint.

Page 16, 374-375, UVAI are used to categorize GEMS aerosol retrievals. What is the meaning for UVAI? Does different category relate to different aerosol types?

Page 19, Line 425, how the boundary layer height relates to the ALH? Is there any quantitative relationship?

Page 19, Line 443, are these UTC time? Can you also provide local time?

Page 20, Fig 9, can you provide local time too?

Page 25, Line 513, “…This suggests that EPIC and TROPOMI ALH retrievals exhibit a systematic positive bias for aerosols over Southeast Asia, indicating the potential need for tuning in the related smoke model, including surface reflectance and aerosol properties.”

Can you elaborate what can be tuned in the smoke model?

Page 27, Line 554, “Both EPIC and TROPOMI consistently overestimates ALH in comparison to CALIOP, with an approximate bias of 0.8 km.”

At what AOD range this conclusion is made?