Ice and mixed-phase cloud statistics on the Antarctic Plateau

Cossich, William; Maestri, Tiziano; Magurno, Davide; Martinazzo, Michele; Di Natale, Gianluca; Palchetti, Luca; Bianchini, Giovanni; Del Guasta, Massimo

doi:https://doi.org/10.5194/acp-21-13811-2021

Articles | Volume 21, issue 18

https://doi.org/10.5194/acp-21-13811-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/acp-21-13811-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 21, issue 18

Research article

|

17 Sep 2021

Research article |

| 17 Sep 2021

Ice and mixed-phase cloud statistics on the Antarctic Plateau

William Cossich, Tiziano Maestri, Davide Magurno, Michele Martinazzo, Gianluca Di Natale, Luca Palchetti, Giovanni Bianchini, and Massimo Del Guasta

Download

Final revised paper (published on 17 Sep 2021)
Preprint (discussion started on 23 Mar 2021)

Interactive discussion

Status: closed

RC1:
'Comment on acp-2021-97', Xianglei Huang, 31 Mar 2021
Review of ACP-2021-97 “Ice and Mixed-Phase Cloud Statistics on Antarctic Plateau” by Cossich et al.

Based on in-situ measurements, this manuscript documents ice and mixed-phase cloud statistics at the Concordia Station in the Antarctic. Various aspects of cloud statistics are discussed. A comparison with satellite L3 data product is also described.

This is a valuable study. Given the scarcity of in-situ measurements and the need to validate satellite retrievals in the polar regions, such study is also much needed by the observational community. Including far-IR, the portion of spectrum rarely used by other observational studies is another shining point of this study. However, some discussions and depictions in the text are not accurate, which I shall describe below in detail. I recommend acceptance after these issues are addressed.

Major comments:

It is well known that cloud fraction statistics from satellites hinges on multiple factors, such as the size of the field of view, the frequency used in the observation, and the detection method (active vs. passive). Besides, passive remote sensing has significant challenges in distinguishing clouds from snowy or icy surfaces over the polar region. It is good to see the authors attempted to compare REFIR-PAD cloud fraction with satellite L3 products (Figure 10) but to thoroughly and correctly interpret this figure is not trivial at all. The discussions related to Fig. 10 ignore a couple of key points: (1) cloud fraction statistics from satellite L3 products are related to footprint size, and none of the L3 products used here has the same field of view as REFIR-PAD; (2) active sensors usually can give more accurate results than passive sensors in terms of cloud occurrence, but their footprint sizes are so different that no way a real “apple-to-apple” comparison can be made. An enormous amount of effort has to be invested for data subsetting and collocation in order to make a fair comparison, which cannot be done with the L3 product directly.

Figure 10 is useful and informative, but the interpretation here has to be conservative, with all caveats well described before the discussion. For example, the abstract stated, “A comparison of monthly mean 15 results with cloud occurrences/fractions derived from level 3 satellite products, from passive and active sensors, emphasizes the difficulties of satellite observations in the Antarctic region and highlights the ability of the CIC/REFIR-PAD synergy to identify multiple cloud conditions and studying their variability at different time scales.”, which I think is not a fair statement given all the reasons mentioned above.

Other comments:

Figure 2c, red spectra, I am surprised to see the clear-sky spectrum here has a peak at CO2 band center (~667 cm-1) as low as 150 K BT. Since this is a surface measurement looking up, the BT at the CO2 band center should be close to the temperature in the lower troposphere or even in the boundary layer. Thus, it cannot be so low. The cloudy spectra here, as well as the clear-sky spectra in Figure 5, look all reasonable to me. Thus, this 150K BT peak at the CO2 band center in Figure 2c needs to be examined and explained.

Line 130: The impact of multiple scattering within liquid clouds on the depolarization ratio can be included as justification for the 15% depolarization threshold.

Line 154: These classes are not explicitly mentioned in the abstract, please list them there.

Table 3: Misclassifications are in the hit rate column. Please make it clearer that these are misclassifications through labeling or reformatting the table.

Figure 7: The caption mentions unclassified spectra, but there does not seem to be any in the figure.

Line 345: The authors mention that a subset of the long-term data is used for training and testing. Is this different from the training data and testing data mentioned in the preceding text? If so, please state. If not, then please indicate why the algorithm is retrained and reoptimized.
Citation: https://doi.org/10.5194/acp-2021-97-RC1
- AC1: 'Reply on RC1', Tiziano Maestri, 22 Apr 2021
  
  Replies in the attached file
  
  Citation: https://doi.org/10.5194/acp-2021-97-AC1
RC2:
'Comment on acp-2021-97', Bryan A. Baum, 31 Mar 2021

Review of ACP-2021-97: Ice and mixed-phase cloud statistics on Antarctic Plateau by Cossich et al.

This is a very interesting study of cloud type and frequencies of occurrence observed from MIR and FIR spectra over the Antarctic Plateau. The study is fairly well written and informative; I recommend acceptance after these comments have been considered by the authors.

Relatively minor questions:

1. It would be quite interesting to provide an analysis of the surface-based lidar measurements for cloud occurrence, thermodynamic phase, cloud heights and layers, even if the measurements are limited to a height of about 7km. The lidar analysis would be complementary to the REFIR-PAD analysis, if there was enough data collected.

2. I have two questions about the lidar data:

a. does multiple scattering above a liquid layer or at the cloud top impact the interpretation of the results? There is some evidence for this in the upper panel of Figure 7.

b. Is there evidence of ice particles falling through the base of the liquid layer? How often does this happen? This question arose when I read lines 290-296, and studied Figure 7 and related text.

3. In the paragraph beginning on line 437, I am puzzled by the lack of cloud fraction information in the winter (dark) months for the combined Terra and Aqua MODIS cloud product. If the information is available for the MODIS product from each of the Terra and Aqua platforms, there must be a problem with the combined data product. This seems to be something that the MODIS cloud team has to resolve. Suggest leaving out the combined Terra and Aqua data product until it has been resolved.

Other comments:

Title: “on Antarctic Plateau” —> “on the Antarctic Plateau”

Line 114: does 1928 refer to the number of REFIR-PAD spectra that are collocated with LiDAR measurements?

Line 126: I think there’s a problem with the reference “Sassen and yu Hsueh”. Should be Sassen and Hsueh.

Line 142: Radiosondes Vaisala RS92 —> Radiosondes (Vaisala RS92)

Line 176: are arranged —> are prepared

Line 188: generally small cloud —> generally low cloud

Line 199: that can be different —> which can be different

Line 228: “window wavenumbers, that results in a very”—> window wavenumbers, and the measurements can have very

Line 286: “if for the 8.3” —> if for 8.3

Line 302: (c) —> or (c)

Line 345: as TS —> for training

Lines 361-363: A third possibility is suggested by the authors but not included in the sentence: The temperature and mixed-phase cloud correlation could indicate that warm temperatures are favorable for mixed-phase clouds formation or that the presence of warm liquid clouds implies a stronger cloud forcing at the surface and, consequently, an increase in the temperature values near the ground. The third possibility is warm air advection of moisture. If the authors agree on this point, this third possibility should be included here and also in the Conclusion section.

Line 391: in correspondence of cloud sky conditions —> when clouds are present

Line 412: in presence of different —> for different

Lines 409-410: please mention where the winds from the NE originate to provide some potential insight as to the origination of the moist layer that is being advected over the Plateau.

Line 434: by both satellites platforms —> by both satellite platforms

Line 439-440: Thus —> For some reason... Note: the MODIS cloud mask should always have a result regardless of solar illumination because it includes infrared measurements.

Line 456: in case of —> in the case of

Line 460: In months in which —> In the months where

Line 460: CALIOP products in green —> CALIOP products as shown in green

Line 461: maximum of —> maximum in

Line 497: in presence of —> in the presence of

Lines 499-500: Potential explanations for this are this could be due… —> Reasons for this could include...Actually, this entire sentence is a bit awkward and should be reworked.

Line 512: classification —> classifications

Line 515: set up —> optimized

Line 525: sets in two —> sets into two

Lines 545-547: could include warm air advection as a third option

Line 564: intense insolation —> higher insolation

Line 565: CALIOP collocates the —> CALIOP data indicates that the

Line 566: similarly to what derived —> similar to what is derived

Line 572: a hourly —> an hourly

Line 572: with maximum —> with a maximum

Line 573: and minimum —> and a minimum

Line 579: but it is reduced —> but smaller

Citation: https://doi.org/10.5194/acp-2021-97-RC2
- AC2: 'Reply on RC2', Tiziano Maestri, 22 Apr 2021
  
  In the attached file the one to one replies to the Rerviewer
  
  Citation: https://doi.org/10.5194/acp-2021-97-AC2
RC3:
'Comment on acp-2021-97', M. G. Mlynczak, 15 Apr 2021

This paper presents a comprehensive analysis of four years of far-infrared and mid-infrared spectral measurements of downwelling radiation at the Dome C site in Antarctica. Nearly 88,000 spectra comprise the database. These spectra are analyzed with a machine learning code to identify scenes as clear or comprised of ice-phase or mixed-phase clouds. The method of training the machine learning algorithm using coincident lidar data is described.

The infrared spectral observations are automated and are taken throughout the year. Analyses of the occurrence of the different scene types are presented in various ways (by year, by time of year (season)). Comparisons against observations of cloud type made by satellites are also presented. The paper also presents an analysis of the occurrence of cloud type against the meteorological conditions.

The paper is very exhaustive in its analyses and I would recommend publication after these minor comments are addressed.

Line 101 – are the effects of the air in the 1.5 m chimney significant at any wavelength? My team found it necessary to account for the “chimney effect” in analysis of our ground-based, uplooking data. See https://doi.org/10.1016/j.jqsrt.2015.10.017 and http://dx.doi.org/10.1016/j.jqsrt.2017.04.028

Line 116 – what is meant by (1928)?

Figure 2 – The 150 K brightness temperature in the red (clear sky) curve is interesting. This is at the very core of the 15-um band, the strongest part of the band, so the line should be saturated almost immediately within the atmosphere, and thus, 150 K appears to be too small. Is there a suspected reason for this anomalous feature? Or is it perhaps a microwindow saturating in the polar summer mesosphere where the temperatures can be well below 150 K? These are January (summer) spectra, so the polar mesosphere is quite cold at that time.

Figure 5 – Similarly, the 300 K brightness temperature seem a bit high for such a saturated part of the spectrum. In the summer season the polar stratopause temperature can approach 300 K, but in the winter season this seems too warm. In addition, the brightness temperatures near 1300 cm-1 and the standard deviations approaching 350 K are non-physical. To what extent does the cloud classification system (CIC) depend on these regions of the spectrum in which the brightness temps appear to be incorrect?

Figures 8 and 11. The legends in these figures are difficult to read as the font is very small. I am looking at the figures on my computer screen that projects each manuscript page at full size. The labeling of the axes of these figures is also difficult to read. Please re-plot these figures with larger axis labels and please use a larger font on the figure legends.

Citation: https://doi.org/10.5194/acp-2021-97-RC3
- AC3: 'Reply on RC3', Tiziano Maestri, 22 Apr 2021
  
  Attached the file containing the detailed replies to the reviewer's comments
  
  Citation: https://doi.org/10.5194/acp-2021-97-AC3
CC1:
'Comment on acp-2021-97', Penny Rowe, 19 Apr 2021

Overall: This paper presents results from a unique and valuable dataset. The two main contributions are cloud classification of this dataset into clear skies, ice cloud, and mixed phase cloud, and an algorithm that can quickly do this classification (which is presumably applicable elsewhere). However, we believe there are a number of serious problems with this paper. Most importantly, there is insufficient evidence that the authors are classifying cloud phase. Instead, it appears likely that they are grouping views into 3 types: 1) clear sky, 2) colder, optically thinner clouds, and 3) warmer, optically thicker clouds. Given this, it is not clear what value the algorithm adds to the literature, given that other methods exist that can classify phase that also classify optical depth and hydrometeor effective radius. The authors need to determine and report what they are actually classifying views into, e.g. using simulated data. More details follow.
A better review of the literature and comparison to existing methods are needed

Referencing of the lidar instrument and how phase is determined is insufficient.
Referencing of recent work on Antarctic cloud properties and similar cloud property retrievals is insufficient. Reading this paper, it would seem that there have been no surface-based studies of Antarctic clouds after 2012. The authors should reference recent papers by Lachlan-Cope et al 2016, Silber et al 2018, Lubin et al 2020, etc.
Machine learning concepts need to be referenced. Due to a complete lack of such references, it is unclear what are established methods (PCA, confusion matrix, hit rate, etc) and what was invented by the authors. E.g. are there references for the method of using a test set and an extended test set? Summing subtracted eigenvectors? Such references would be very helpful to fill in gaps and help understand what is novel.
The paper should compare this new method to existing methods for retrieving cloud phase from infrared radiances. For example, they reference Cox et al 2014, who retrieve cloud properties from Arctic infrared radiances, but do not compare to this work. They should also reference and discuss comparison to Rowe et al 2019 & Lubin et al 2020, which includes development and application of a cloud property retrieval, including phase, to clouds over McMurdo, Antarctica. Simulated datasets exist which could be used for an inter-comparison of methods. See, e.g. Cox et al 2016, Earth System Science Data, 8(1), 199–211.
Examination of the data in a real-world context is needed

The authors report the common occurrence of cloud with a liquid base and an ice layer at the top, which is contrary to what has been reported previously, both in the Arctic and Antarctic. This difference from previous work calls for some justification. This also underscores the need for a better explanation of the lidar design and methodology for determining cloud phase. What is meant by determining cloud layers from lidar by “human intervention?” Is this objective and repeatable? Why can’t it be automated? Overall, using lidar as truth is not properly justified.
The authors use Principal Component Analysis (PCA), but they never explore, plot, or discuss the associated eigenvalues and eigenvectors. The retrieval is blind in the sense that it does not take into account the atmospheric state in terms of temperature, humidity, CO2 concentration etc. This would be ok if it was shown that the retrieval works without taking these into consideration, including some exploration of how it works, but this has not been done. It should be noted that almost all the variance, and thus the strongest PCs, will be associated with cloud temperature and optical depth, not phase. Which PCs are associated with phase? Why use all PCs believed to be above the noise level? It seems likely that the classification is not based on cloud phase at all, but rather that scene views are subdivided into: 1) clear sky, 2) colder, optically thinner clouds, and 3) warmer, optically thicker clouds. They call category 2 “ice” and category 3 “mixed phase.” It is possible these classifications are often correct, since ice clouds tend to be optically thinner and colder, and liquid clouds tend to be optically thicker and warmer on the Antarctic Plateau. However, this needs to be characterized, addressed and discussed, including errors and caveats.
Several lines of evidence support the idea that they are not classifying cloud phase but rather optically thick and warm vs optically thin and cold clouds. First, looking at Fig. 2, it is unlikely that it is possible to determine phase from the green spectrum. This spectrum looks saturated, which means phase will have no influence on it - that is, there is no information about phase. It does, however, indicate that the cloud is optically thick. The authors could assess for which cases phase cannot be retrieved, using simulated spectra. Instead, are all such cases classified as “mixed phase” by the algorithm? Second, as the authors point out, it has been shown that the far IR is critical for determining phase. Yet Fig. 6 suggests that a wavenumber range that excludes the far IR altogether would be equally good as one that includes it: the threat score is close to 1 for a range of just above 560 cm-1 to ~1020 cm-1. Indeed, the authors find the best range to be 540-1020 cm-1 for mixed phased clouds (it is unclear how they determine this), excluding essentially all of the far IR. Third, in the cold macro-season the algorithm does not retrieve cloud phase at all; instead all clouds are assumed to be ice.
Given the above, the authors should report the results of testing their method on simulated data, as has been done for other methods in the literature. This would allow them to test whether they truly have a cloud phase categorizer or if they are categorizing by cloud temperature / optical thickness. They could also determine and define characteristics of each category in terms of temperature, optical depth and phase ranges. This would also allow exploration of how errors propagate.
The authors ignore previous work on the temperature dependence of the single-scattering parameters (SSPs) of liquid water, which indicate that the SSPs of supercooled liquid water are intermediate between those of liquid and ice (Rowe et al 2013 and 2020 and references therein). In particular, Rowe et al (2020) indicates that uncertainties are large in the far IR.
Questions and Concerns about Methodology and use of Machine Learning

The authors need to justify why they used the method they developed. It is not clear why PCA is used, or why the SID is used. Why isn’t a simpler method tried, or at least compared to, to justify the more complicated method used? Fig. 5 suggests that only one wavenumber is needed to distinguish cloudy from clear skies. Such a cloud mask has been reported in the literature but is not referenced or noted here (e.g. Weaver et al 2017, Atmos. Meas. Tech., 10, 2851–2880, 2017, Appendix). Classification using a single wavenumber would be sufficient for all of the cold macro-season data. Why is a considerably more complicated method used? To distinguish ice cloud from mixed-phase cloud, how many PCs are needed? Is PCA justified? Also, it seems odd to first divide cases into clear sky vs ice cloud and confusing that these each include mixed-phase. Why not divide first to clear and cloudy? Then subdivide cloudy into ice and mixed-phase. Such important details are left unexplored by the authors.
The authors use PCA to remove noise (Eqns 3-4) using an established method. However, Antonelli et al (2004, J Geophys Res 109, D23102), who should be referenced, state that the size of the training set should be greater than the number of spectral elements (M>N) to most accurately reconstruct the atmospheric signal and most efficiently remove noise. Here it appears that M<<N. How does this affect the noise reduction and signal reconstruction?
Antonelli et al (2004) also state that if some spectra are not well-represented by the set of spectra used for noise reduction, a larger number of PCs may be needed to properly represent those spectra. This seems likely to be the case when the input spectrum is not a member of the training set in Eq. (6). How is this handled and how does it impact the results?
The authors reduced the dimensionality of the observations by modifying the spectral interval of the test set members and re-running the algorithm. Given that the authors are already using PCA, and that PCA is typically used for dimensionality reduction, why isn’t PCA used for this dimensionality reduction?
Furthermore, it is not good practice to use the test set to select features (wavenumbers) to use. Using the test set to optimize the algorithm exaggerates the accuracy of the method and can lead to overfitting. Model development should be done using training or validation sets. See, e.g. Ripley, B.D. (1996) Pattern Recognition and Neural Networks, Cambridge: Cambridge University Press, p. 354. The data with known labels should be split into training, validation, and testing sets. The testing set should be held apart and only used to estimate the accuracy of the method. None of the training, testing, or validation data should then be included in the analysis. The authors need to clarify which data is being used in each step and ensure they are following established practice.
More detail is needed to allow the analysis to be repeated.

It is not clear how the authors handle erroneous data points. One of the reviewers pointed out that a data point at the center of the CO2 band is erroneous, at 667 cm-1. This is typical with such instruments because calibration is impossible at such wavenumbers (see Rowe et al 2011, Optics Express, 19(6), 5451–5463, and Optics Express 19(7), 5930-5941). There are many other erroneous brightness temperatures evident - for example, none of the BTs below 200 cm-1 appear useable, as well as many between 200 and ~350 cm-1, where BTs are very high. How did the authors handle such points in their analysis? Were they included or omitted? The authors should briefly explain the instrument error characterization and point to a reference with more detail.
The algorithm description could use some clarification. The development should proceed linearly from training to testing to implementation. It seems that what is meant by the input spectrum on line 164 varies; this needs to be clarified. For example, it seems the SIDs and the CSIDs are developed from the training set first (to get Fig. 4)? How is this done?
Finally, the utility of this algorithm seems likely to be specific to the unique conditions on the Antarctic Plateau. The authors should discuss whether it would be applicable elsewhere.
Penny Rowe and Steven Neshyba

Citation: https://doi.org/10.5194/acp-2021-97-CC1
- AC4: 'Reply on CC1', Tiziano Maestri, 14 May 2021
  
  Our comments and answers in yellow.
  
  Citation: https://doi.org/10.5194/acp-2021-97-AC4
CC2:
'Comment on acp-2021-97', Robert Stillwell, 30 Apr 2021

Please find attached a comment on this manuscript.

Citation: https://doi.org/10.5194/acp-2021-97-CC2
- AC5: 'Reply on CC2', Tiziano Maestri, 17 May 2021
  
  Our replies in yellow
  
  Citation: https://doi.org/10.5194/acp-2021-97-AC5

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Tiziano Maestri on behalf of the Authors (03 Jun 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (07 Jul 2021) by Michael Schulz

RR by Bryan A. Baum (23 Jul 2021)

RR by Xianglei Huang (17 Aug 2021)

ED: Publish subject to minor revisions (review by editor) (20 Aug 2021) by Michael Schulz

AR by Tiziano Maestri on behalf of the Authors (24 Aug 2021) Author's response Author's tracked changes Manuscript

ED: Publish as is (26 Aug 2021) by Michael Schulz

AR by Tiziano Maestri on behalf of the Authors (26 Aug 2021) Manuscript

Short summary

The presence of clouds over Concordia, in the Antarctic Plateau, is investigated. Results are obtained by applying a machine learning algorithm to measurements of the infrared radiation emitted by the atmosphere toward the surface. The clear-sky, ice cloud, and mixed-phase cloud occurrence at different timescales is studied. A comparison with satellite measurements highlights the ability of the algorithm to identify multiple cloud conditions and study their variability at different timescales.