Cloud Phase Recognition Based on Oxygen A Band and CO₂ 1.6 µm Band

Round 1

Reviewer 1 Report

BRIEF SUMMARY

Cloud phase recognition is an important part of many remote sensing algorithms. This paper studies the possibility of retrieving the cloud phase based on radiances observed by passive satellite instruments at the weakly-absorbing edges of the oxygen A band and the CO₂ 1.6 µm band. This is based on the finding that for a given cloud optical thickness,
the ice cloud reflectance is typically higher than that of a water cloud
in the oxygen A band, but lower in the CO₂ 1.6 µm band. An algorithm
is suggested in which (i) the presence of cloud is evaluated based on
the magnitude of the reflected radiances in the two bands, and (ii)
the cloud phase is evaluated based on the ratio of the reflected
radiances. This cloud-phase retrieval algorithm is tested using radiances
measured by the OCO-2 instrument and cloud-phase retrieval products by
the CALIOP lidar.

BROAD COMMENTS

I think this paper reports a mostly reasonable study on a topic relevant
for Remote Sensing. However, unfortunately, the paper is somewhat difficult to read due to numerous problems with the English language (in both grammar and vocabulary). There are too many issues to be listed individually in the comments, but some are nevertheless mentioned below. A careful proof-reading by a person native/fluent in the English language is very strongly recommended. There are also some other issues with the presentation, included in the specific comments below.

Regarding the science, my most important comment concerns surface albedo. The authors' statement regarding this ("the threshold alpha is different under different surface albedo" (on line 213)) is basically true, but it actually downplays the issue. I think that the cloud-phase recognition algorithm works more or less properly over ice-free ocean, for which the surface albedo does not vary very much and also depends only weakly on wavelength. (This also seems to be the surface type over which the authors test their algorithm, although even this is not stated very clearly). For other surface types, most notoriously snow and ice, the performance is probably going to be worse. See specific comment 9.

SPECIFIC COMMENTS

1. The title as it stands now is not clear. It is not obvious to me, what
"CO₂ weak absorption zone" refers to. "CO₂ 1.6 µm band" would be clear.
Also, "Oxygen A band" would be more standard terminology than "Oxygen A zone". So I suggest "Cloud phase recognition based on Oxygen A band and CO₂ 1.6 µm band".

2. line 95: specify what is meant with the CO₂ weak absorption band.

3. lines 132-135: The fact that reflected radiance is larger for the ice cloud than for the water cloud in the oxygen A band, but smaller in the CO₂ 1.6 µm band, is so central for this paper that it should be explained briefly why this is the case. I think this is because of (i) the difference in scattering phase function (more sideward/backward scattering for ice crystals than water droplets) makes the reflectance larger for the ice cloud in the O₂ A band, and (b) the absorption is stronger for the ice cloud than for the water cloud in the 1.6 µm region (both because of a larger imaginary part of refractive index, and larger particle size), which suppresses the reflection.

4. Line 147: "the radiances ... generally increase". Perhaps it would be worth adding that the (relative) increase is larger in the water cloud case.

5. Figure 3 fails to convey the message it is intended to convey. Judging by the text (lines 160-161), the idea is to show that if you pick the wavelengths with weakest gas absorption in these channels, the impact of cloud top height is small. However, what captures the reader's attention is that the dependence on cloud top height is very large at the wavelengths with strongest absorption (which are not useful for the proposed cloud phase retrieval).  Using a logarithmic scale for K could help somewhat.

6. lines 160-161: the term "continuous absorption band in oxygen A zone (758-759 nm)" gives an impression that there is a separate absorption band in addition to the absorption lines in the oxygen A band. Is this really true (or is it just caused by the wings of the strong lines in the middle of the O₂ A band, centered at about 761-762 nm)? In the latter case, it would be more appropriate to speak of the "region with weak continuous absorption" at the edge of the oxygen A band. The same question/comment also goes for CO₂ absorption at 1594-1595 nm.
(Also elsewhere in the paper, starting from the abstract, line 11).

7. As a follow-up to the previous "semantic" comment, you should state
explicitly that you select this pair of wavelengths (758-759 nm vs. 1594-1595 nm) for studying cloud phase recognition in the rest of the paper.

8. lines 179-180: please refer explicitly to Fig. 4(d).

9. lines 196-213: Some discussion should be added regarding the impact
of surface albedo. The basic challenge is that when the surface albedo is high, or the cloud is optically thin, the surface reflection contribution to the top-of-the atmosphere reflected radiance may be comparable to or even larger than the contribution from the cloud. In such cases your approach will most probably not work properly.

Another concern is that especially over land, the surface albedo depends
on wavelength (which is not considered in your tests). For example, the
albedo of snow is very high in the oxygen A band and very low in the CO₂
1.6 µm band, which would make separating ice clouds and snow very
difficult. Even water clouds might not be identified properly, unless they are very thick optically. For snow-free land surface the situation is less extreme, but still some of the same problems are present (the albedo depends on wavelength, and may vary as a function of time), which could make the use of predefined threshold curves for alpha (as suggested on line 213) challenging.

So my expectation is that the suggested cloud-phase recognition algorithm would not work equally well (if well at all) over other surface
types than ice-free ocean.

10. lines 223-227: could you define the "judgment accuracy rate"  explicitly (i.e., with equation(s))?

11. line 230: these thresholds presumably depend on surface albedo.

12. line 234: Figure 10 is not discussed at all in the manuscript (at least it is not referred to anywhere in the text?). Please mention how you determined the water and ice cloud fits and the threshold curves. And most importantly, which data is used (presumably the same dataset as
in Fig. 8) and how was the cloud phase determined here (I suppose it was
taken from CALIOP data)?

13. line 235: which geographic region does this OCO-2 data track represent?

14. In Fig. 11, the labels are painfully small to read. Also, what is "3-HO"?

15.  In Fig. 12(a), it is difficult to discern visually the different red symbols from each other, and the different blue symbols from each other. And to add to the confusion, the same symbols have different meanings in Fig. 12(b). Use of different colors (up to 6 per one figure panel) might help.

16. Figure 13. It would be most logical to organize also Fig. 13(b)
according to the cloud-phase retrieval of the CALIOP instrument. That is, I suggest to have two columns, one for "CALIOP ML, ice top" and another for "CALIOP ML, water top", with colors indicating how your cloud-phase retrieval algorithm for OCO-2 classifies these cases.

LANGUAGE ERRORS AND TYPOS:

1. Line 24: this should be "clouds consist of ..."

2. lines 27-28: it is not clear what the "cloud change process" means.

3. lines 47, 60, 92: replace "recognition effect" with "recognition skill"?

4. lines 52, 58, 69, 80: Author surnames (Jin, Wouter) should not be written
with all caps.

5. lines 54-55: a subject is missing in this sentence. Start with "this algorithm", or "they".

6. line 86: remove "with" in "is with cloud".

7. lines 96-97: This sentence should start with a subject. "We"?

8. lines 109-110: "which also belongs to the A-Train satellite sequence".
Delete. This is said three lines later.

9. line 117, 273 (and elsewhere): "simulation calculations". Why not simply
"calculations" (or "simulations")?

10. lines 130-138: I think you should speak of "reflected spectral radiances" here. Not "radiation values" (line 133) or "spectral of" (line 136).

11. lines 197-202: Honestly, this (multi-)sentence is intractable! Please
try to clarify it. 

12. lines 203-204: "the alpha of ice cloud and water cloud has changed
greatly". Compared to what? The case with alpha=0.02?

13. line 216: Replace "... data are the results of survey over" e.g. with
the "...data represent transects over ...".

14. line 226: This should be "Figure 8".

15. line 242: You can remove "... RO₂ is the same, if ...".

16. lines 261-266: I cannot quite follow what you are trying to say here.
Do you mean that out of those CALIOP multi-level clouds that are identified as ice cloud by OCO-2, the top-level cloud is an ice cloud in more than 70% of the cases? Also, "...there is no obvious corresponding relationship between the changes of alpha and RO₂ of multi-layer cloud." Do you mean that the relationship between alpha and RO2 is not different for the multi-level clouds with ice vs. liquid water in the top layer?

17. line 278: add "when" before "the solar zenith angle is less ...".

18. lines 282-283: Suggestion: "... recognition of the phase of the uppermost cloud layer of multi-layer clouds".

Reviewer 2 Report

The objective of this study is to set up an algorithm to retrieve the cloud phase based on OCO data only.

Figure 1 clearly shows the impact of the cloud phase on the simulated radiance spectra. But this case is ideal. Several factors affect this behaviour. COT and cloud height parameters are discussed in section 3 while sun angle and surface albedo parameters are discussed in section 4.

Due to the large variations of the ratio (alpha) between the two reflectances, the choice of a simple ratio is not demonstrated.

Also, the threshold line (Figure 10) seems to be set by the comparison with CALIOP data. Therefore, it is not an actual validation of the method and the method accuracy is biased. (Figure 10 is slightly different than Figure 4d, supposed to be the expected behaviour).

The method cannot be applied to the multi-layer cloud scenes. It is a strong limitation since the objective of the study was to apply the method to other sensors, and so not in combination with CALIOP data.

Some detailed comments:

• Homogeneity on the use of COT or tau for the  cloud optical thickness in the figures legends
• Explain what is reflectivity? I prefer the term Reflectance, but I am not sure that you are really using reflectances.
• Figure 2 and 3: The ratio... of which terms? I understand that it refers to the case COT=1, but it should be explained in the legend.
• Figure 3: not so readable. If you choose a spectra region outside the absorption peaks, it is not necessary to plot all the wavelengths range.
• Why don't you use the information on the CTH retrieved with the oxygen band?
• The test case over the ocean is a very favourable case since the surface reflectance is almost null at these wavelengths outside of the glitter area.
• Quid of aerosol effect on the measurement?

Reviewer 3 Report

The paper describes work that looks promising in the estimation of cloud phase from the 760 nm O2 and 1600 nm CO2 bands, found on OCO-2. However, the analysis does not seem to be developed clearly. Results are given for clouds at inconsistent heights. Results are given as radiances, reflectivities or ratios without clear explanation of the relationships.

Most of the analysis is carried out for different cloud top height for ice and water clouds. In reality, the challenge of phase estimation is most serious for cloud top temperatures where either or both liquid and ice can occur; ie between 0 and about -30 C. There seems to be no recognition that mixed-phase cloud is not uncommon.

Figures tend to repeat results, but do not optimally show key results.

The atmospheric model is fixed with 1km cloud depth, with droplets at 12um and ice as hexagonal prisms (50x100um). What is the sensitivity of the results to microphysics and cloud depth (not just tau)?

Evaluation is carried out on just one satellite track. The preliminary nature of the results should be made clear.

Specific comments

l1. The title is a little unusual: ‘research on..’ is valid for every paper. Perhaps ‘On cloud phase recognition…’ implies a contribution to the general problem, rather than the overall solution.

I also wonder whether ‘zone’ would be better replaced by ‘band’.

L14. under → Under

l20. Is ‘inversion’ right? Perhaps ‘estimation’

l24. Clouds have more types than droplets and crystals. Perhaps ‘Clouds can contain water in both liquid and ice phases…’

l31. Biggest → most

l32. What is ‘it’?

L33. Either ‘..size; among…’ or ‘… size. Among…’

l35. CTH is often estimated from an external source, based on CTT.

L52. JIN → Jin

l54. ‘Set a …’ is not really a sentence.

L58. JIN again

l68. Arking sp.

L69. WOUTER → Wouter; not in references.

L72. Verifies → to verify

l80 JIN

l85. ‘inverting’ → estimating

l86. Delete ‘with’

l96. The CALIOP cloud phase product is used….

Figure 1 caption. ‘spectral’ → spectral radiance

l141-147. I don’t understand figure 2. Figure 1 shows the sensitivity of radiance to particle phase in the 760 and 1600 nm bands for a cloud with tau=10 and CTH=4 km. Figure 2 takes a different CTH for a water (4 km) and ice (9 km) and plots the absorption in each band with different values of tau, normalised to the absorption for tau=1. For the liquid cloud, we can take the curve in figure 2 with tau=10 to be the curve in figure 1. But, how do we relate the curves for the ice clouds to figure 1 (with different CTHs)? Why not plot the actual radiances and why not use the same CTH? Moreover the challenge is when both liquid and water can occur, so why not use the same CTH for liquid and ice? At least have some overlap.

L151-156. Similarly for figure 3, we have radiances normalised for ice clouds at 9 km and water clouds at 2 km. These plots are even more difficult to interpret because the normalising curves are barely visible.

L159-161. The stated sensitivity to CTH is not apparent from the curves as drawn in figure 3 – most of the curves are hidden by the 8-km curves.

L162-166. Why not define Ro2, Rco2 and alpha earlier? They should arise naturally from figure 1, and could perhaps be plotted there, as well as in figures 2 and 3. Indeed, figure 4 essentially puts figures 2 and 3 into perspective (or replaces them). The relationship between radiances (in figs 1-3) and reflectivities (in fig.4) needs to be made clear.

L167-171. Once again we have in figure 4 different CTH for ice and water clouds. The key challenge is for the phase of clouds with CTT warmer than about -30 C; ie overlapping ice and liquid CTH is essential for any useful analysis.

L174. COT and tau need to be specified, and used consistently.

Figure 4. The caption does not clearly describe each sub-plot. Fig.4d summaries 4b and 4c, so why include 4b and 4c? Once again, the cloud heights for ice and water clouds are different. Do the differences in alpha remain when CTH is the same for ice and water?

Figure 4a may be more useful with a log scale for COT. How significant is the difference for small COT?

Section 3 on method actually gives results of the analysis, but it does not give a method for an algorithm to use alpha and Ro2 is clearly determine water phase and COT under all likely scenarios; especially for CTT between 0 and -30 C.

l185-191. Figure 5 gives an indication of the sensitivity of alpha and Ro2 to zenith angle by considering zenith angles of 30 and 70 degrees, but not 0 degrees which is the base case in earlier figures. Moreover the values of other parameters such as CTH or CTT are not specified.

The discussion of figure 5 notes the sensitivity to COT, but it is not clear how COT is varied in the figure.

L196-204. Figure 6 considers the sensitivity to surface albedo. Figs a and b are not labelled, and the symbols are swapped for albedo=0.3 in a and b. The overlapping symbols makes it hard to interpret the results; as colour is used, it would be better to have different colours in each figure.

Once again, figure 6 seems to incorporate the results of figure 5: why two figures?

What are the values of other cloud parameters in figure 6?

l203-204. The meaning of this sentence is not clear. The results of figure 7 are in figure 6.

l218. SZA is defined here, but used earlier (fig. 6 and 7) in the text.

L219. Is there a difference between the subscripts CO2 and WKCO2?

Figure 8 is labelled as figure 9.

l221-224. Figure 8 does not show a clear separation of radiances for cloud and no-cloud.

L228-232. What is ‘judgment accuracy’, as shown in figure 9? Is it a probability? The caption of figure 9 should clearly state the variables. How do you manage two different thresholds: either or both need to be exceeded?

L235-238. There is no a or b in figure 10. How were the curves in figure 10 computed? What was the variation in cloud properties along the (one) track?

Given all the small values of Ro2 in figure 10, perhaps it should be logarithmic in Ro2.

L239-245. The algorithm starts to be described after the results, rather than before. The threshold curve is not specified.

L243-245. The text and captions do not describe the results shown in Figure 11, 12 and 13. Is the cross-section in fig 11 the same as for fig 11? What is in figure 11 a and b? I assume 11a is from CALIOP and 11b is from the present analysis. Given the described threshold method, why does not every cloud pixel get a phase; ie why ‘unknown’? Why can’t you relate the pixels to a cloud temperature (if not a height)? What is HO in fig 11a?

Figure 12 is a bit misleading. I guess the purpose is to show that mis-recognition of phase is mainly for multi-layer cloud, as identified by CALIOP. This result would be clearer if (a) showed only single-layer cloud and (b) showed multi-layer cloud (as now). However, the use of colour is misleading at present, especially in b. In (b) with fewer points, colour should be used to highlight the mis-matches.

A further point to make is that the authors give no means for them to identify multi-layer cloud, and so the mismatches for these cases cannot be avoided – this should be noted.

Figure 13a continues to separate single-layer from multi-layer clouds, although the current approach cannot make that distinction. The fraction of each case from CALIOP should be specified in Figure 13a.

The caption of figure 13b is not clear: what are the blue and red bands? I assume they are the CALIOP phases for the top-level cloud.

L266-267. Stating that the algorithm is not applicable to multi-layer cloud is not useful unless there is a means to identify multi-layer cloud. Results should really be given for all cloud.

L280-281. “...shows that the recognition effect is better”. What does this mean? Better than what?

L282-283. What is the basis of the statement about upper stratospheric cloud?