Review of “Evaluation of satellite retrievals of liquid clouds from the GOES-13 Imager and MODIS over the midlatitude North Atlantic during NAAMES campaign”, by Painemal et al.

The authors compare effective radius (r_e) and optical depth (τ) retrievals in low liquid water clouds from in situ and remote aircraft observations with retrievals from satellite observations. The aircraft retrievals are r_e from a Cloud Droplet Probe (CDP) and both r_e and τ from the NASA GISS Research Scanning Polarimeter using observations at 2.26 µm (r_e) and 0.865 µm (τ). RSP r_e is retrieved using a polarimetric technique while τ is retrieved using a standard reflectance-based method. The satellite cloud retrievals are from GOES-13 and MODIS (Aqua and Terra) using CERES Edition 4 algorithms which use a visible channel and a near IR channel (3.79 µm for MODIS, 3.90 µm for GOES-13). The differences between the various datasets are presented and discussed. Both satellites r_e are larger than aircraft r_e. GOES-13 r_e is notably larger than MODIS r_e, and the authors show that the difference increases towards the backscattering direction. The hypothesis is that for this geometry of observation, retrieved r_e depends on the prescribed effective variance of the size distribution used in the satellite algorithms, which might not be well adapted.

The results presented in this manuscript are of interest for the scientific community.  However, I think that the main comments below should be addressed before the manuscript can be published, and I also suggest considering the specific comments.

Main comments

1) The authors state that the RSP r_e retrievals are at 2.26 µm (line 105, page 4). I could not find information about retrievals at 2.26 µm using the polarimetric technique in the various references. The authors should give more details about the retrieval technique, and perhaps explain why this channel was chosen to retrieve r_e.

2) How do the authors justify that the RSP r_e retrievals at 2.26 µm are mostly sensitive to the cloud top (optical depth ≈ 1) (lines 176-177, page 6)?  Based on the work by Platnick (2000), they state that 3.79-3.9 µm satellite r_e is representative of about 2 optical depths down from the cloud top (line 161, page 5). Stating that the RSP retrievals at 2.26 µm are weighted higher in the cloud than the satellite retrievals at 3.79-3.9 µm seems inconsistent with Platnick (2000). Please explain.

3) I believe that a discussion regarding the expected differences between retrievals at 2.26 µm (RSP) and 3.70-3.90 µm (satellites) is missing.

4) The authors state that post-deployment evaluation of the CDP probe showed that there was an overcounting of droplets for all bins (lines 94-96, page 3) and that as a result, CDP could provide only r_e, but not water content, extinction coefficient, and cloud droplet number concentration. Therefore, I don’t understand how the authors could determine the τ =2 altitude level from the top to determine CDP cloud-top r_e (lines 166-167, page 5). Please explain.

5) The authors mention the presence of supercooled liquid water clouds during the cold months when they present the airborne observations (line 85, page 3), and in the conclusion, they state that both supercooled and warm boundary layer clouds are a climatological feature (lines 331-332, page 10). However, I do not see any discussion on this topic in the presentation of the results.  Why is this important? Are the comparisons different for supercooled and water clouds? Please develop.

Specific comments

Abstract, line 18: I suggest specifying in the abstract which GOES-13 and MODIS (Aqua and Terra) products are used for the study.

Introduction or where relevant: please define “effective radius”.

Lines 70 and 84, page 3: I found that Fig. 1 was not very informative because too small. I would suggest 1 panel per campaign with the associated mean Aqua "cloud cover". It looks like the caption should actually say “low cloud fraction”.

Line 85, page 3: please explain how RSP data could confirm the presence of supercooled cloud tops during the cold months. During which campaign(s)?

Line 150, page 5: Please describe the CERES SSF product and provide a reference.

Line 193, page 6: The authors give an overview of the cloud vertical structure during the campaign. Is it for only one of the 3 campaigns? If yes, which one? Please clarify. Should results for warm clouds and supercooled clouds be shown separately?

Line 195, page6; Fig. 3: the authors state earlier in the text (lines 86-87) that cloud sampling was limited to boundary layer liquid clouds with a mean cloud top height of 1376 m±602 m (± standard deviation). What about cloud base?

Lines 287 to 290, page 9: this discussion is difficult to follow without an illustration or at least a reference.

Lines 321-322, page 10: please explain how the effective variances shown in Fig. 12 (not Fig.11) were retrieved. Is it from RSP or CDP?

Technical comments:

Line 140, page 5: “east-west”  => could be rephrased.

Line 321-322, page 10: Fig.11 should be Fig. 12

References: the format does not seem compliant with AMT specifications.

This paper presents a comparison of satellite derived cloud optical thickness (tau) and droplet effective radius (re) for liquid clouds using airborne in situ observations and remote sensing retrievals obtained during NASA’s NAAMES field campaign (2015-2017). Specifically, Terra and Aqua MODIS tau and re retrievals from the CERES Edition 4 algorithms, and similar GOES-13 retrievals from their SatCORPS algorithm counterparts, are compared against re derived from droplet size distributions measured by the CDP cloud probe and re and tau retrieved from the RSP polarimeter (re from polarimetric observations, tau from NIR reflectance). The comparisons show that the satellite retrievals of re are larger than those obtained from both the CDP and RSP, with GOES-13 having the largest differences; comparisons of tau show better agreement. Biases in both re and tau are further investigated as a function of scene heterogeneity and viewing/scattering angles as possible explanations of the differences.

The paper is really well written – clear, concise, and easy to comprehend. The methodology is sound, and I think it will be a good addition to the body of literature on evaluating satellite cloud optical/microphysical retrievals using airborne observations. My comments are mostly on the discussion of the comparisons and potential causes of disagreement, which I find to be too narrowly focused on sensor spatial resolution (along with the inherent impacts of heterogeneity and 3D effects) and angular sampling at the exclusion of other important players – particularly potential forward model deficiencies beyond the plane-parallel assumption that I identify below. I suggest the authors do a bit more legwork on these before this paper is accepted for publication.

Comments:

• Page 2, lines 56-57: I suggest citing the Witte et al. (2018) paper here, along with implications on this study. (Witte, M. K., Yuan, T., Chuang, P. Y., Platnick, S., Meyer, K. G., Wind, G., & Jonsson, H. H. (2018). MODIS Retrievals of Cloud Effective Radius in Marine Stratocumulus Exhibit No Significant Bias. Geophysical Research Letters, 45(19), 10,656–10,664, doi:10.1029/2018GL079325.)
• Page 3, line 86: Just to clarify, the C-130 only flew 1 to 1.5 hour (60 to 100 minute) flights? Seems short to me, so perhaps I’m misunderstanding.
• Page 3, lines 96-97: This statement on overcounting being thought to equally affect all size bins is an assertion without evidence. Was this verified to be the case? It might be an acceptable assumption, but there’s nothing here that makes that case.
• Page 3 line 102: Why is the CIP sampling not the same as the CDP? My understanding is that these two in situ instruments are supposed to complement each other to resolve the full width of the droplet size distribution, and are flown together for that reason.
• Page 4, lines 115-117: Yes, the RSP polarimetric re retrievals may be accurate, at least for the synthetic LES cases considered in Alexandrov. But the key question here, given their use as a benchmark for satellite re retrievals, is whether these retrievals should be consistent with those from total reflectance approaches considering their different vertical weighting functions (e.g., Platnick (2000)). The polarimetric signal is a single-scattering phenomenon and thus is sensitive to the very top of the cloud. Looking at the re profiles in Fig. 3 (and from knowledge of similar profiles from other field campaigns), there is a decrease in re at the very top of the cloud. This decrease may in fact be too small to matter, but the authors don’t fully address this other than later in the paper stating that using different tau thresholds (1 and 3) in their averaging of “cloud top” CDP measurements yields only a roughly 0.1µm re change. The single-scattering polarimetric signal may be in large part from the portions of the cloud above even 1 optical depth into the cloud. Please comment on this.
• Page 4, lines 118-121: Radiometric calibration, and relative radiometry between two imagers, can have a big impact on tau retrievals and their agreement between two sensors (see, e.g., Meyer K, Platnick S, Holz R, Dutcher S, Quinn G, Nagle F. Derivation of Shortwave Radiometric Adjustments for SNPP and NOAA-20 VIIRS for the NASA MODIS-VIIRS Continuity Cloud Products. Remote Sensing. 2020; 12(24):4096. https://doi.org/10.3390/rs12244096). While the tau retrieval agreement is quite good later in the paper, did the authors assess the relative radiometry between RSP and MODIS/GOES? It’s possible that the good agreement is fortuitous and may be masking larger heterogeneity effects.
• Page 6, lines 177-178: See my comment above on the vertical weighting functions of polarimetry versus total reflectance. I guess for 3.7/3.9µm, the difference in weighting with respect to polarimetry is reduced compared to, say, 1.6µm, but this is a little hand-wavy and there may still be differences.
• Page 6, lines 204-205 and Fig 4: I suggest adding error bars to this plot similar to those in Fig. 5. For the MODIS vs CDP plot, can you stratify these results by the MODIS 250m heterogeneity index (Liang et al. (2009), again similar to what is done in Fig. 5)? Also, what about sensitivity to the width (effective variance) of the observed droplet size distribution? The satellite retrievals are making an assumption on veff (later on defined as 0.1) – how do these results stratify as a function of divergence of that veff assumption from the observations? Veff can be calculated from the observed DSDs, so I suggest doing that analysis.
• Page 7, lines 213-215: Using RSP to define the heterogeneity index only provides information in one direction, i.e., along the flight track. Both satellite imagers have footprints much larger than the width of the RSP footprint, so across-track heterogeneity may be missed. Using the MODIS 250m heterogeneity, as I suggest above, would be helpful. Also, following my previous comment, what is the veff retrieved by RSP for these comparisons? Are the RSP veff generally consistent with CDP, at least where the two can be reasonably compared? I see the RSP veff are shown later in Fig. 12, but there is no stratification of CER retrieval differences as a function of veff deviation similar to what was done for VZA and scattering angle, or even heterogeneity. Veff sensitivity should be a no-brainer to add here.
• Page 7, lines 230-243: Perhaps the MODIS vs GOES re retrieval differences are tied to the rather large central wavelength difference (3.75 vs 3.9µm) and may point to a different forward model issue? Specifically, the liquid index of refraction assumed in the calculation of the cloud single scattering properties – see Platnick et al (2020) for a discussion of re sensitivities to refractive index and temperature (Platnick S, Meyer K, Amarasinghe N, Wind G, Hubanks PA, Holz RE. Sensitivity of Multispectral Imager Liquid Water Cloud Microphysical Retrievals to the Index of Refraction. Remote Sensing. 2020; 12(24):4165. https://doi.org/10.3390/rs12244165). Note that paper shows MODIS 3.75µm vs VIIRS 3.7µm re differences on the order of those shown here, though I admit the impacts of heterogeneity are difficult to disentangle. Can the authors at least comment on the implications of this on their MODIS vs GOES results?
• Page 8, lines 255-258: This may be more challenging, but I think you can at least plot the MODIS scattering angle distributions within each GOES scattering angle bin (perhaps as an accompanying box plot). That should indicate scattering angle sampling differences. You should also plot Terra and Aqua MODIS separately, since the scattering angle sampling may be quite different.
• Page 8, line 264-267: While the precipitation likely isn’t aliasing into the satellite retrievals, how do the DSDs observed by CDP itself change between precipitating and non-precipitating clouds? If it’s significant, it’s possible that there may be a correlation with re differences given the assumed veff may deviate more/less from reality.
• Page 9, line 278-279: Besides spatial resolution differences between GOES-13 and 16, what about scattering angle differences? This was pointed to as a key player in the MODIS vs GOES-13 differences, and GOES-13 and 16 weren’t viewing from the same orbital location.
• Page 10, lines 323-325: I don’t think investigating veff impacts needs to wait for future work, nor does it require using veff as an additional input to the satellite retrievals (i.e., using various veff in the forward models). As I suggested above, you can simply look at re retrieval differences as a function of RSP veff (or CDP veff). You already have these data from RSP, and can calculate veff quite easily from CDP, so the hypothesis at least can be partially tested here. I suggest the authors do this analysis.