the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 23 Sep 2026
Intercomparison of Liquid Cloud Droplet Size Retrievals from Different PACE Instruments
Abstract. This paper presents an intercomparison of liquid cloud droplet size retrievals from the three instruments onboard the Plankton, Aerosol, Cloud, and ocean Ecossystem (PACE) mission: SPEXone, HARP-2, and OCI. SPEXone and HARP-2 provide retrievals of cloud droplet effective radius (CER) and cloud effective variance (CEV) from respectively hyperspectral and hyper-angular polarization observations. OCI provides CER retrievals using radiometric measurements in the visible and shortwave infrared. In the one-year (June 2024 - May 2025) intercomparison, closest agreement is found between SPEXone and HARP-2 CER, while OCI generally retrieves larger CER than both polarimeters. The CER retrieval comparison between SPEXone and HARP-2 yields a mean difference (MD) of 1.45 µm (0.81 µm) over land (ocean) and a mean absolute difference (MAD) of 2.53 µm (2.58 µm). The MD between SPEXone and OCI is -2.06 µm (-2.67 µm) over land (ocean) and the MAD is 3.15 µm (3.60 µm). For the comparison between HARP-2 and OCI, the MD is -3.51 µm (-3.47 µm) over land (ocean) and the MAD is 3.89 µm (3.89 µm). Globally, annual CER statistic distribution peaks near 7 µm for SPEXone and HARP-2 and around 10 µm for OCI. CEV comparison between SPEXone and HARP-2 yield an MD of -0.003 both over land and ocean and an MAD of 0.039 (0.037) over land (ocean). The CEV global statistic distribution for both instruments peaks near 0.07. The results in this paper suggest that polarized retrievals consistently yield a substantially smaller CER than conventional bi-spectral retrievals, even for two polarized retrievals from different instruments, using different retrieval algorithms. These differences may be caused by differences in vertical sensitivity or biases caused by cloud inhomogeneity or 3D radiative transfer effects.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Atmospheric Measurement Techniques.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(6963 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 29 Oct 2026)
- RC1: 'Comment on egusphere-2026-5631', Anonymous Referee #1, 24 Sep 2026 reply
-
RC2: 'Comment on egusphere-2026-5631', Anonymous Referee #2, 05 Oct 2026
reply
This is a review on the manuscript entiteled “Intercomparison of Liquid Cloud Droplet Size Retrievals from Different PACE Instruments”, which compares the independently derived liquid cloud droplet effective radius from the three instruments (SPEXone, HARP2 and OCI) onboard the PACE satellite for the observational period between June 2024 and May 2025. Further, the derived effective variances from the two independent polarimetric retrievals of SPEXone and HARP2 are compared. Since the instruments operate on the same satellite, a large statistical evaluation is possible, which is an advantage compared to existing studies, which are usually constrained to individual cases due to collocation limitations. Hence, the results of the study are certainly interesting and relevant for the field and fit well into the scope of AMT. However, the study lacks a bit of detail and some questions arise from the comparisons. I suggest to accept the paper, after the following general and specific comments were addressed.
General comments:
-
To what extend can the found differences in the MD and MAD be explained by known uncertainties of the different retrievals? For example, the uncertainties in the polarimetric CER and CEV retrieval were extensively studied and multiple comparisons to bispectral retrievals were made and the authors also cite some of those papers. However, from my point of view, the paper would benefit from a more detailed discussion, implicitely comparing numbers to previous studies. E.g. a typical value is an overestimate of MODIS derived CER by about 2 µm compared to POLDER measurements in stratocumulus regions (Bréon & Doutriaux-Boucher, 2005). How does such a finding compare to the values found here? Further, the polarimetric cloudbow retrieval is thought to derive cloud droplet size distributions with a high accuracy, and there is hardly any discussion on the found differences between the two different polarimetric approaches. How are the uncertainties in the SPEXone retrieval compared to the polarimetric retrieval? Please give numbers and discuss further.
-
Due to the difference in spatial resolution between OCI and HARP2/SPEXone, the OCI retrieval results are averaged onto the HARP2/SPEXone grid. This is a natural approach to compare differently resolved retrievals on a common resolution. However, the effective radius is a nonlinear quantity, hence the averaging of the individual pixel effective radii to the larger footprint does not necessarily resemble the effective radius of the cloud droplet size distribution seen by the polarimeters. This should be taken into account in the interpretation of the results.
-
The introduction is also rather short. In my opinion, the paper would benefit if the state of the art of the determination of cloud droplet effective radius (and variance) from space would be discussed, in particular, the two common and established retrieval methods (bispectral and polarimetric cloudbow retrieval) should be mentioned, while it can also be emphasized, that they are evaluated compared to a new NN retrieval method.
-
The scientific questions addressed by the different analyses could be outlined in more detail (general comparison, comparisons in dependence of COT and cloud fraction, with respect to regional (←→ meteorological differences), ...). I.e. they sould be mentioned in the introduction and then picked up in the discussion of the individual figures, such that it becomes clear to the reader, what the authors try to invest with the individual analyses.
-
What are general limitations of the individual retrievals? E.g. what are the ranges of the CER and CEV that can in principle be derived by the individual retrievals? Some additional information would be desirable in Sec. 2.
-
Some information about the spread of the difference distributions (e.g. the standard deviations) would be beneficial in addition to the MD and MAD.
Specific comments:
Line 33: To what extend do the CER and CEV play a central role in the ACI quantification?
Line 45: “In this paper, a comparison is performed for liquid CER and CEV retrievals (..) among the instruments on PACE.”
Line 62f: “Note that such retrievals require the detection of the cloudbow, which provides a highly effective filter for the presence of liquid cloud droplets.” Next to the grammatical corrections, I would be more precise here: a cloudbow is also detected in mixed-phase clouds, it only requires liquid droplets at the cloud top.
Line 64: Is there a reason why CER > 30 µm are discarded?
Figure 1:
- The caption says the instruments were collocated at “1°x1° grid boxes” which have not been defined before. Are those lat/lon coordinates? What about the footprints and the averaging mentioned in Sec. 2, does that matter at all then? Or are those along-/across-track coordinates?
- What is the basis of the cloud mask applied? That would also be worth mentioning beforehand, e.g. in Sec. 2.
- The figure labels are quite small, please increase the size of the title, labels etc.
Line 90: “bi-spectral retrievals penerate deeper into the cloud than polarimetric retrievals”?
Line 91f.: That sentence stays in contrast to the results found in this work. I would suggest to argue the other way around: While lower effective radii might be expected for the bispectral retrieval compared to polarimetric retrievals with regard to the adiabatic cloud model theory, the results of the presented study are well in line with the mentioned previous studies. In particular, possible biases due to 3D effects in the bispectral retrievals should be mentioned here.
Figure 2:
- Line 95f.: but that is not necessarily the case for small cloud fractions, right? E.g. the MAD decreases slightly for low cloud fractions and large optical thicknesses in the SPEXone-HARP2 comparison, and this also holds for the MD. The SPEXone-OCI MAD and MD plots look significantly different and in my point of view, there is no clear constant error for COT <= 7. Hence, for small COT, the cloud fraction seems to be important with regard to the different retrieval results, while it seems to be less important for larger COT. Please verify and discuss further to make the relevance clear to the reader. What conclusions can be drawn with regard to the different retrieval sensitivities?
- What about the comparison between HARP2 and OCI? This is not mentioned in the text.
- For clarity, I would suggest to plot either the absolute value of the MD such that increases/decreases of the differences become clear, reverse the colorscale or to use a diverging colorscale, e.g. plot positive differences in one (e.g. red) and negative ones in another colorscale (e.g. blue)
- The authors could further think about joint limits for the colorbars, which would emphasize differences in the instrument comparisons
Figure 3:
- Please add the different regions as titles to the figure.
- Why did the authors choose these particular regions? What about continents?
- Line 103: is there a reason why SPEXone is slightly more heavy-tailed?
- The figure also suggests a clear tendency towards smaller CER over land compared to ocean. This is a nice consistency check due to more aerosols over land and could be mentioned in the discussion of the figure.
- Line 109: Is there a reason why the bispectral CER is overestimated in cirrus contaminated cases? And do the authors have an idea how often cirrus cases are classified as “liquid cloud fraction > 0.9” by the OCI cloud mask?
- Currently, the text only refers to the peaks of the distributions and not the median values marked by the dashed lines. This is a little confusing. If the dashed lines are kept, I would also suggest to add their positions to the legend.
Line 114f.: Is that because those are the stratocumulus regions?
Line 120f.: The particularly stated agreements between SPEXone and HARP2 over the Southern Ocean and land are quite hard to see without plots of the differences.
Line 123f.: Why should SPEXone retrieve smaller CER over most NH oceans compared to SH oceans? Is that expected for some reason?
Fig. 5 caption: “… from the two instruments …”
Line 141: “… may be contributed by …”
Fig. 5: Is there a reason why SPEXone does not retrieve low effective variances? What are the ranges of the LUTs?
Fig. 6: Please set the titles as for Fig. 3
Line146f.: Can the authors think of a reason why HARP2 shows flatter distributions?
Fig. 7: I would suggest decimal notation in stead of scientific for the colorbars.
Fig. 7 caption: Please move the last sentence to the main text and the discussion of the figure.
Line 151: The fact that SPEXone retrieves lower CEV in the stratocumulus regions is a bit surprising with regard to Fig. 5 and 6 as in general HARP-2 finds more small effective variances compared to SPEXone. Are the small CEV found by HARP-2 in general located in different regions (the polar regions?) or is that an averaging artefact, i.e. is the spread of the HARP-2 CEV larger in the stratocumulus regions and in general?
Line 152: “higher” instead of “lower”? This is as I would expect as the clouds are in general more convective in the tropics than in the mid-latitudes
Line 152f.: The fact that the SPEXone CEV is lower west of Australia is similar to the lower effective variances in the stratocumulus regions and the not very pronounced feature might potentially be averaged out by HARP-2? Here the question arises, whether the CEV is underestimated by SPEXone or whether HARP-2 potentially finds more extremes and why so?
Line 153f.: do the authors have an idea why the CEV would be lower in the SH in the summer months? What about the NH?
Line 168f.: check statement again, see Sec. 3!
Line 176: clarify “broader distributions” as this might also be related to larger CEV values
Line 176f.: “However, it has to be noted that CEV is sensitive to measurement resolution and cloud field heterogeneity.”: Please clarify as this has not been discussed before. To what extend is CEV sensitive to one or the other?
Line 183f.: I disagree with the given statement. Large rain droplets usually occur only in small numbers and the optical thickness of rain is small. Hence, rain/drizzle should not significantly affect bispectral retrievals. In contrast, the presence of only a few precipitation droplets near cloud top would be enough to largely influence the polarized radiance and the cloudbow region, as the rain droplets would produce an additional rainbow signal. Therefore, particularly the cloudbow retrieval might be biased high.
Line 189f.: the consistency of the global variation of the retrieved CER from OCI with respect to MODIS/VIIRS has not been discussed in the text before. This should be done in more detail beforehand, including citations.
Line 208: Acknowledgement
Line 219: the “Arias” citation should include that this is the IPCC Technical Summary
Citation: https://doi.org/10.5194/egusphere-2026-5631-RC2 -
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 320 | 113 | 36 | 469 | 43 | 35 |
- HTML: 320
- PDF: 113
- XML: 36
- Total: 469
- BibTeX: 43
- EndNote: 35
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This paper presents a one-year (June 2024–May 2025) intercomparison of liquid cloud droplet effective radius (CER) and effective variance (CEV) retrievals from the three PACE instruments SPEXone, HARP-2, and OCI. Data were collocated on a common 5.2 km grid and aggregated to 1°×1° cells globally. The authors find that the two polarimetric retrievals (SPEXone and HARP-2) agree most closely with each other while OCI's bi-spectral retrieval consistently yields systematically larger CER than both polarimeters with the discrepancy growing as cloud fraction decreases and largest in the tropical belt and broken-cloud regimes; CEV comparisons between SPEXone and HARP-2 show broadly consistent global distributions (peaking near 0.07) despite differences in shape and regional detail.
The authors attribute the polarimetric-versus-bi-spectral CER offset to differing vertical sensitivity within the cloud layer and known 3D radiative-transfer/inhomogeneity biases in bi-spectral retrievals. The analysis is a useful, well-organized validation exercise for a new mission's cloud products and appropriately caveats known limitations (e.g., CEV's sensitivity to resolution and cloud heterogeneity, OCI's coarser resolution and lack of sub-pixel cloud detection), though the explanations for the systematic CER differences remain largely qualitative I wish that the authors had more rigorously explored reasons behind the differences in retrieved CER and CEV. Overall, I think this paper is useful with minor revisions because it provides a nice reference for these very new cloud products and their discrepancies, but it leaves a lot to be desired in terms of explanatory power.
Specific Comments:
Line 74: It would be good to know how the data were filtered. For example how were liquid clouds identified, Were certain cloud conditions filtered out, was there any scattering geometry imposed, etc…? I am particularly curious of certain scattering geometries were used and whether this could introduce a sampling bias.
Line 99: If I understand the figure 2 correctly the disagreement (MAD and MD) between the two polarimetric methods is as large as the differences between the polarimeter and bi-spectral methods, especially when cloud fraction is low. (1) I’m struggling to relate this finding to figure 1, (2) This is a bit unexpected to me, do you have reason in mind as to why? (3) It seems this is worth some discussion in the text.
Line 109: Do you not have an indication of cirrus contamination from the hyper angle polarimetric channel that can test this?
Line 122-124: This hemispheric difference between the two polarimeters is notable. I’m left wanting some sort of exploration as to why this happens. What varies systematically between the SH and NH datasets that could explain this?
Line 137: logarithmical -> logarithmic
Lines 144 – 147: Some more discussion of Figure 6 is warranted. (1) The tendency of the HARP retrieval to produce lots of near 0 CEV values seems problematic. This speaks to measurement errors. (2) I’m not seeing the type of variation amongst regions that I would expect, that is larger CWV in clean (oceanic) regions and smaller CEV in polluted (land) regions. We see some of this later in Figure 8 but the signal appears very noisy. Overall, in this section I think you need to mention uncertainty and there should be some citations of the literature here regarding CEV calculated from in-situ data and what we expect it to depend on.