Comments/edits are provided in the enclosed PDF.

Review of AMT-2023-214

Summary:   During the 6-deployment ACTIVATE campaign over the Atlantic, the high-flying King Air supported the HSRL-2 (lidar) and RSP (polarimeter).  While both can retrieve AOD, which is better? There are few ground-measurements, so retrieved AODs were triple-collocated with MODIS (satellite) -retrieved AODs to suggest that standard RSP retrievals are the outlier.  Adding a stricter cost-function filter helps improve the RSP accuracy (reduce the offset compared to HSRL-2).   Physical reasons for the differences between HSRL-2 and RSP includes cloud contamination, aerosols near surface, multiple aerosol layers, absorbing aerosols, non-spherical aerosols, and simplified retrieval assumptions.  Neglecting stratospheric aerosol contributions appears to not be a major source of RSP bias (compared to HSRL-2).   A conclusion was that “These results demonstrate the pathway for optimal aerosol retrievals by combining information from both lidar and polarimeter for future airborne and satellite missions”

Evaluation:  This was one of those papers that seemed like a straightforward intercomparison between 2 products.  On my first reading, I was willing to suggest only minor revisions, but then I read it again and gained major concerns.  Not so much about what the authors did do (use MODIS as a proxy for sanity checking, and accept a higher level of QA filtering as cost function thresholds), but a lot of questions about what was not attempted or even suggested (examine the physical reasons for differences).    A few examples:

• The authors mention theoretical ±1σ accuracy of the MAPP algorithm (RSP) for AOD is ±0.02. They also mention practical accuracy of HSRL-2 as 0.01. Are there studies of the practical accuracy of RSP (compared to AERONET, AATS/4-STAR or other sunphotometer)?   I don’t know the RSP literature, but I know it was flown often and there must be such information available?
• The RSP standard retrieval assumes a near-surface, non-absorbing coarse spherical sea salt. What are the sensitivities to assuming something else? Would that help improve the retrievals? Same questions about SSA of the fine modes.   Would that reduce the cost function χ from 0.15 to 0.05, and thus lead to a more “successful” retrieval?   I am not satisfied simply to change the cost function threshold.
• Since you are collocating with MODIS, what does MODIS suggest about cloud-cover/fraction and all the issues that might make collocation (and retrieval) difficult? Also, what about gradients of aerosol nearby the collocation?
• My biggest issue I think is the lack of attempting to look at the rest of the ACTIVATE data. There were 162 coordinated lower-level Falcon flights measuring in-situ properties. While not many, there are AERONET sites nearby (Bermuda, LaRC, Hampton, Wallops). What do they tell you about clouds? About aerosol layering? About the ocean surface? Etc?
• This is one of those studies where doing case studies (one collocation at a time) could be instructive. Pick one, learn something. Pick another interesting case, do it.  Maybe one more.  Then you can derive statistics and attempt to generalize.
• Finally, I am unconvinced by the conclusion statements regarding the need to combine information from lidar and polarimeter.  Well, yes, I believe it, but right now I see the RSP retrieval being subject to poor assumptions, which I can tell are poor even before bringing the HSRL-2 retrievals to confirm.   Pick one the cases, show us how the lidar information helps the polarimeter (and possibly vice-versa). Then I would accept that statement.

Questions and comments related to text:

Lines 78 then line 84:  Introduces χ2 and then χ′. Are these supposed to be the same terms?  

Line 104: Reference for the 10-s HSRL product?

Line 176: Eq 10.  Think subscript j should be replaced by subscript 1.  Should there be a ±sgn function too?

Line 221:  “introduce MODIS AOD as third dataset.”  As you clearly say, MODIS is not a ground-truth.  Maybe this is a dumb question, but how “good” does this 3^rd dataset need to be to be useful the RSP vs HSRL-2 sanity check?  The authors of the MODIS dataset claim something like ±(0.04 + 10%) for accuracy over ocean. Is this good enough?

Line 222: “we combine both datasets to maximize the number of files for collocation.” What if Terra or Aqua are biased differently?

Line 225:  Do you do any checking of the MODIS imagery to determine if there appears to be a rapid gradient nearby the 3-way collocation?

Line 242 – 260: I find this super confusing in terms of its presentation. I can understand why you might think of restricting threshold for “fitting error/cost function” χ, but don’t see the logic of attempting to restrict by any of the retrieved fine or coarse AOD parameters.

Line 255 – 257: By reducing the cost function threshold to 0.05, it is at the expense of reducing available points by 50%.   Is this acceptable?

Figs 5 and 6: Not authors fault, but really frustrating that the figures are shown >1 page after the text describing them.

Line 272: Don’t understand phrase “which indicates that the bias is possibly lagging behind the cloud influence.”

Line 274:  Don’t understand what this means:  “There is only one year of below-aircraft cloud mask data but it includes other properties such as dynamic cloud fraction which also considers the influence from glint.”  Which year? A summer and a winter? Please explain “dynamic cloud fraction” and the “influence from glint”.

Line 276: I must be missing something here: Isn’t it of course obvious that: “Higher cloud mask fraction is associated with higher below-aircraft cloud cover (Figure 7c).”  Is there a difference between cloud cover fraction and cloud mask fraction?

Line 277:  “This implies that there is not a linear relationship between the cloud cover and bias.”  I am wondering if there is a relationship between cloud cover (or mask) and the fitting error/cost function?   Note that RSP-HSRL “bias” is plotted (Fig 7) in absolute terms, not in percentage terms, so maybe we don’t care about small biases if the AOD is large.

Line 290: “Practically dusty mix is a mixture…”  This makes no sense. 

Line 292: This is backwards: “An SSA value from 0.90 to almost 1 represents a weakly to moderately absorbing aerosol”  (0.90 is absorbing,  approaching 1 is non-absorbing)

Line 305:  Suggest insert particle to read “the shape of a sea salt particle depends”.

Line 306: “reported that sea salt may cause high depolarization” .  Why is that?   Because of shape? It becomes non-spherical?  

Lines 320-325: For someone not as familiar with aircraft measurements, I am thinking a diagram might be helpful. What are each of these instruments observing above and below the aircraft, and what layering is “missing” because of the way these sensors must work?   

Lines 329-331: The chronology doesn’t make sense: “The peak AOD in 2020 is very likely due to the stratospheric eruption at Raikoke (48◦ N, 153◦ E) in June 2019 (Kloss et al., 2021). Subsequently the stratospheric AOD increased up to 0.027 in October 2019 (not shown) and gradually decreased in the ensuing months.”  Nonetheless, 0.01-0.02 stratospheric AOD doesn’t explain magnitude of 10 higher biases.

Lines 359 (conclusion 4.):  “It proves that the quality of AOD retrievals can be optimally improved by combining information from both lidar and polarimeter.”   I definitely believe it, but I don't see how this study proved that statement.   

Lines 360-365: To me this discussion is disjointed from the rest the paper.   How does this study lead to desire to use 2-dimensional imaging? (which I don’t even know what that is).