the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 Jun 2025
Identifying MBL cloud boundaries and phase over the Southern Ocean using airborne radar and in-situ measurements during the SOCRATES campaign
Abstract. The Southern Ocean Clouds, Radiation, Aerosol Transport Experimental Study (SOCRATES) was an aircraft-based campaign (Jan 15 – Feb 26, 2018) using in-situ probes and remote sensors, targeting low-level clouds over the Southern Ocean (SO). A novel methodology was developed to identify cloud boundaries and classify cloud phases in marine boundary layer (MBL) clouds using airborne HIAPER Cloud Radar (HCR) and in-situ CDP+2D-S measurements. Cloud boundaries were determined using HCR reflectivity and spectrum width gradients. Single-layer low-level clouds accounted for ~85 % of observed cases. HCR-derived boundaries showed decent agreement with the Ceilometer and Micropulse lidar (MPL)-measurements during the Measurement of Aerosols, Radiation, and Clouds (MARCUS) ship-based campaign, with mean base and top differences of 0.04 km and 0.29 km. Additionally, HCR-derived cloud base heights correlated well (R = 0.78) with HSRL observations. A reflectivity–liquid water content (Z-LWC) relationship, LWC = 0.70Z0.29, was derived to retrieve LWC and liquid water path (LWP) from HCR profiles. The estimated LWP closely matched MARCUS microwave radiometer (MWR) retrievals, with a mean difference of 9.24 g/m². Cloud phase was classified using HCR-measurements, temperature, and LWP. Among single-layered LOW clouds, 48.8 % were classified as liquid, 23.3 % mixed-phase, and 6.9 % ice, with additional categories identified: drizzle (16.2 %), rain (3.4 %), and snow (1.5 %). The classification algorithm demonstrated over 90 % agreement with established phase detection methods. This study provides a robust framework for boundary and phase detection of MBL clouds, offering valuable insights into cloud microphysical processes over the SO and supporting future efforts in satellite algorithm development and climate model evaluation.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(4113 KB)
-
Supplement
(392 KB)
-
This preprint has been withdrawn.
- Preprint
(4113 KB) - Metadata XML
-
Supplement
(392 KB) - BibTeX
- EndNote
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-874', Anonymous Referee #1, 18 Jun 2025
- AC1: 'Reply on RC1 and RC2', Anik Das, 23 Aug 2025
-
RC2: 'Comment on egusphere-2025-874', Anonymous Referee #2, 08 Jul 2025
The authors use in-situ observations and remote sensors to classify hydrometeors observed during the SOCRATES field campaign over the Southern Ocean. While the objective of the study is of high interest, unfortunately, it is not very well carried out, as outlined in the General Comments below. The proposed algorithms are based on interesting ideas but the results are not convincing. A much more rigorous evaluation of the proposed algorithms is needed before possible publication. Also, the authors do not sufficiently explain why their method is superior to previously published methods, i.e., why this new method is needed.
General Comments:
1. The authors claim in the conclusion that "Both the cloud base detection and phase classification methodologies were rigorously evaluated against existing methods.". Unfortunately, I have to disagree with this statements. For the most part, the authors only compare bulk statistics of cloud base height or particle phase, which is not particularly meaningful. A bulk comparison does not say much about the accuracy for individual cases. It only compares averages, and even those averages are skewed, given that the radar-derived quantities are time-height cross sections while the in-situ probes measure only along the flight path. The only rigorous evaluation is the pixel-by-pixel evaluation against the PID method - which reveals significant differences. Without a pixel-by-pixel comparison with the cloud probe based methods, the value of the algorithms cannot be confirmed.
2. The cloud base detection algorithm is strongly based on spectrum width, which is a very noisy quantity. I find it unlikely that it would actually give the desired results and unless the authors provide a convincing rigorous evaluation showing the opposite, I would be very surprised if it were possible to derive cloud base from HCR spectrum width.
3. For the phase classification, the authors rely heavily on the derived LWC-Z relation. I do not find the arguments that such a relation should exist very convincing. If I understand the method correctly, it attributes all variations in Z along the radar beam to variations in LWC. What is the basis for this assumption? Again, I could be convinced otherwise, but only with a better evaluation of the algorithm.
4. To make this study publishable, the authors would need to do the following:
a) Evaluate the cloud base algorithm on a ray-by-ray basis under precipitating conditions against, e.g., the GV-HSRL. If the results don't agree, the algorithm is likely flawed and should not be used.
b) Evaluate the phase classification on a pixel-by-pixel basis (near the aircraft in in-cloud flight legs) against the particle probe derived results from previous studies. The evaluation against PID is good and necessary but only an evaluation against a different sensor will show the true potential of the method.
Specific comments:
5. Title: It is not ideal to use an acronym in the title (MBL).
6. Line 23: Why is LOW capitalized?
7. Line 38: Define SLW.
8. Line 55: The statement "onboard radar and lidar experience less 56 attenuation than ground-based sensors" is a bit misleading and should be clarified.
9. Line 62: MBL has already been defined.
10. Line 65 (and later): These instruments are are generally known as HCR (not GV-HCR) and GV-HSRL (not NCAR-HSRL). Please be consistent throughout the paper.
11. Line 81: At W-band and at the high vertical resolution of HCR, surface clutter is insignificant.
12. Line 97: It is not clearly stated why yet another cloud classification algorithm is needed. What are the expected improvements over the existing ones?
13. Line 127: The temporal resolution of HCR and the GV-HSRL is 2 Hz in the combined dataset.
14. Line 131: This statement is a bit misleading. The GV-HSRL actually detects more clouds, especially thin clouds, because of sensitivity limitations of HCR. However, as you state, it is highly attenuated.
15. Line 133: 3D -> 2D.
16. Section 2.2: One major issue with the Z-LWC relation is attenuation of the radar signal, which is significant at W-band. How does this method deal with attenuation?
17. Section 3.3: I am not convinced by the identification method for Hbase. The difficulty is the distinction between cloud echoes and echoes from precipitation shafts underneath the clouds. The histograms comparing with MARCUS observations do not sufficiently show that the method is working because they do not distinguish between precipitating and non-precipitating clouds. The fact, that GV-HSRL derived Hbase values are significantly higher than the HCR derived values likely shows that the HCR-based algorithm mis-identifies precipitation shafts as clouds. It would be helpful to focus only on radar beams with surface precipitation and carefully compare those with HSRL and MARCUS observations. Some plots demonstrating the method under precipitation conditions would also be helpful (Figure 3 is not detailed enough to provide insight into the quality of the algorithm). I am very doubtful that a method base on spectrum width can work because many different phenomena contribute to the observed spectrum width values and it is very difficult to tease the different contributions apart. Overall, I am wondering if the identification of Hbase is even needed for the current study? Given all the question marks regarding the method, I am wondering if it would be better to exclude it from the manuscript.
18. Line 326: How are "noisy pixels" defined and identified?
19. Figure 7: Please list the exact date and time of this example. RF 13 took place on February 20th but I could not find this example at the given time.
20. Sections 4.3.1 and 4.3.2: I am not sure what this comparison is supposed to tell us. Comparing bulk fractions of different phases is not meaningful given the differences in sampling (1D vs. 2D). For a true comparison, the authors would need to compare actual data points from in-cloud flight legs with HCR observations in close proximity. Then they could calculate some kind of hit/miss table or similar.
21. Section 4.3.3: The comparison with the PID method is interesting but shows significant differences. Only a pixel-by-pixel comparison with a different instrument (as was done for the PID method) will reveal which method actually works better.
22. Line 584: The authors claim that the HCR-phase method has strong capabilities for detecting mixed-phase clouds. However, this has not been shown or verified. Just because the method classifies many pixels as "mixed" does not mean that this represents reality.
Citation: https://doi.org/10.5194/egusphere-2025-874-RC2 - AC1: 'Reply on RC1 and RC2', Anik Das, 23 Aug 2025
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-874', Anonymous Referee #1, 18 Jun 2025
1. Overview of the paper
This study analyzes low-level marine boundary layer (MBL) clouds over the Southern Ocean using airborne radar and in-situ data from the SOCRATES campaign. It introduces a method to detect cloud boundaries with radar reflectivity and spectrum width, derives an empirical relationship between reflectivity (Z) and liquid water content (LWC), and classifies cloud phases using radar, temperature, and LWP. Single-layer low clouds (<3 km) made up ~85% of cases. The Z–LWC relation enabled accurate LWP estimates, validated against observations. The cloud phase classification identified 48.8% liquid, 23.3% mixed-phase, and 6.9% ice clouds, and showed good agreement with other established methods. Overall, the study offers a robust radar-based approach for cloud boundary and phase detection, supporting better satellite retrievals and climate modeling over the Southern Ocean.
This resubmitted version of the study presents clear improvements both in writing quality and in the validation of the proposed methodology. The structure is more refined, and the explanations are clearer, which enhances the overall readability of the manuscript. One of the major additions is the comprehensive comparison with multiple existing methods for cloud boundary characterization. This significantly strengthens the robustness of the authors' approach. Based on these elements, I would recommend only minor revisions before final publication.
2. General suggestions
The scientific objective is clear, and the methodological approach is generally robust, particularly the development of a radar-based classification for cloud phase. The Supplement is clear and provides important supporting information, especially concerning the in-situ datasets and instrumentation details.
However, several issues hinder the manuscript’s clarity and accessibility. First, the text suffers from a lack of consistency in terminology and definitions—key terms and acronyms such as SLW or LOW are introduced without prior explanation. Secondly, there are redundancies and repetitions across the sections, especially in the presentation of the classification results, which could be streamlined for conciseness. Additionally, while the discussion around uncertainties is both important and informative, the writing would benefit from reformulation for clarity and structure. On the methodological side, the rationale for excluding certain cloud types (e.g., fully glaciated clouds) should be more explicitly justified, especially in the context of deriving a Z-LWC relationship. Finally, more attention should be given to harmonizing visual elements (e.g., figure fonts and legends) and providing clearer context in tables and figures, including the comparison of cloud boundaries with in-situ data.
Overall, while the work is of clear interest to the cloud microphysics and remote sensing communities, the revised version appears well-refined, with notable improvements in structure, clarity, and scientific rigor.
3. Specific comments and technical corrections
Abstract.
L.26 : If you are referring to your “LOW” class, you need to define the acronym before using it. Otherwise, you can remove the capital letters or use “low-level clouds”.
Introduction.
L.35 : This might not be necessary. However, the references are neither in chronological order nor in alphabetical order.
L.38 : "SLW" is not defined before being used.
L.53–60 : This paragraph on uncertainties is important and interesting but not very clear. The authors could try to rephrase it.
L.63 : This is a long sentence (6 lines); please split it into two.
L.73 : It’s a bit vague—what were these inconsistencies?
Results.
L.131 : Maybe merge this sentence with the one on L.127.
L.149 : The ice phase (> 200 µm), although minor, is not used in your method? Is it only useful for comparing your classification with Alessandro’s?
L.174 : When you refer to a “subset of the dataset”, does this correspond to the “5th to 95th percentile interval of the dataset”?
L.178 : “FIGURE” → “Figure”
L.244 : No space between ~ and 22.2%.
L.256 : This is essentially what you say in L.226 — consider avoiding the repetition.
L.265 : You should add the term “mean” as in your table, otherwise it’s confusing — it sounds like you’re referring to maximum values.
L.295 : How do you explain a larger mean difference for cloud top altitude?
L.314 : “FIGURE” → “Figure”
L.319–328 : Formatting issue in the label of Figure 4 and repetition in the text. Please correct this paragraph.
L.324 : I understand the importance of stating that you focus on low-level clouds, but this is repeated too many times.
L.324 : Repetition — please revise.
L.336 : The authors refer to drizzle in both the text and Figure 5. You mention both “drizzle” and “liquid drizzle” — what is the difference? When you say “drizzle”, are you referring to freezing drizzle?
L.330–376 : In this paragraph, the authors mention several thresholds for classifying phases and for Figure 6. Some thresholds are cited with references (Wu et al., 2020a; …) and others are not — how were these chosen?
L.377 : Does this kind of “misclassification” occur frequently? Can the authors quantify this? Does a cloud with LWP ≥ 20 g/m² and LWCs > 0.2 g/m³ necessarily exclude the presence of ice?
L.448 : The authors could add a sentence at the beginning of paragraph 4 stating that only low-level clouds are used, in order to avoid repeated reminders later.
L.481 : The concept of scale plays a crucial role in method comparison — the authors should emphasize this point more clearly in the text.
L.525 : The authors should describe Table 2 in more detail. Otherwise, it may not be necessary to keep it in the main text; consider moving it to the supplementary materials, since you only mention it once.
L.574 : The percentages are calculated on classes that are not present in both methods (Mix % and Melting %), which could explain the larger differences.
L.582 : Exactly — you could perhaps add that the mixed-phase strongly depends on the observation scale (microphysical <--> macrophysical).
Figures.
Figure 3 : The font size for y-ticks is not the same in panels f and h. You could also increase the x-ticks size in panel h to match the one in subplot d.
Figure 4 : There is an issue in the label: “4 Low-level cloud phase classification method and discussion”.
Citation: https://doi.org/10.5194/egusphere-2025-874-RC1 - AC1: 'Reply on RC1 and RC2', Anik Das, 23 Aug 2025
-
RC2: 'Comment on egusphere-2025-874', Anonymous Referee #2, 08 Jul 2025
The authors use in-situ observations and remote sensors to classify hydrometeors observed during the SOCRATES field campaign over the Southern Ocean. While the objective of the study is of high interest, unfortunately, it is not very well carried out, as outlined in the General Comments below. The proposed algorithms are based on interesting ideas but the results are not convincing. A much more rigorous evaluation of the proposed algorithms is needed before possible publication. Also, the authors do not sufficiently explain why their method is superior to previously published methods, i.e., why this new method is needed.
General Comments:
1. The authors claim in the conclusion that "Both the cloud base detection and phase classification methodologies were rigorously evaluated against existing methods.". Unfortunately, I have to disagree with this statements. For the most part, the authors only compare bulk statistics of cloud base height or particle phase, which is not particularly meaningful. A bulk comparison does not say much about the accuracy for individual cases. It only compares averages, and even those averages are skewed, given that the radar-derived quantities are time-height cross sections while the in-situ probes measure only along the flight path. The only rigorous evaluation is the pixel-by-pixel evaluation against the PID method - which reveals significant differences. Without a pixel-by-pixel comparison with the cloud probe based methods, the value of the algorithms cannot be confirmed.
2. The cloud base detection algorithm is strongly based on spectrum width, which is a very noisy quantity. I find it unlikely that it would actually give the desired results and unless the authors provide a convincing rigorous evaluation showing the opposite, I would be very surprised if it were possible to derive cloud base from HCR spectrum width.
3. For the phase classification, the authors rely heavily on the derived LWC-Z relation. I do not find the arguments that such a relation should exist very convincing. If I understand the method correctly, it attributes all variations in Z along the radar beam to variations in LWC. What is the basis for this assumption? Again, I could be convinced otherwise, but only with a better evaluation of the algorithm.
4. To make this study publishable, the authors would need to do the following:
a) Evaluate the cloud base algorithm on a ray-by-ray basis under precipitating conditions against, e.g., the GV-HSRL. If the results don't agree, the algorithm is likely flawed and should not be used.
b) Evaluate the phase classification on a pixel-by-pixel basis (near the aircraft in in-cloud flight legs) against the particle probe derived results from previous studies. The evaluation against PID is good and necessary but only an evaluation against a different sensor will show the true potential of the method.
Specific comments:
5. Title: It is not ideal to use an acronym in the title (MBL).
6. Line 23: Why is LOW capitalized?
7. Line 38: Define SLW.
8. Line 55: The statement "onboard radar and lidar experience less 56 attenuation than ground-based sensors" is a bit misleading and should be clarified.
9. Line 62: MBL has already been defined.
10. Line 65 (and later): These instruments are are generally known as HCR (not GV-HCR) and GV-HSRL (not NCAR-HSRL). Please be consistent throughout the paper.
11. Line 81: At W-band and at the high vertical resolution of HCR, surface clutter is insignificant.
12. Line 97: It is not clearly stated why yet another cloud classification algorithm is needed. What are the expected improvements over the existing ones?
13. Line 127: The temporal resolution of HCR and the GV-HSRL is 2 Hz in the combined dataset.
14. Line 131: This statement is a bit misleading. The GV-HSRL actually detects more clouds, especially thin clouds, because of sensitivity limitations of HCR. However, as you state, it is highly attenuated.
15. Line 133: 3D -> 2D.
16. Section 2.2: One major issue with the Z-LWC relation is attenuation of the radar signal, which is significant at W-band. How does this method deal with attenuation?
17. Section 3.3: I am not convinced by the identification method for Hbase. The difficulty is the distinction between cloud echoes and echoes from precipitation shafts underneath the clouds. The histograms comparing with MARCUS observations do not sufficiently show that the method is working because they do not distinguish between precipitating and non-precipitating clouds. The fact, that GV-HSRL derived Hbase values are significantly higher than the HCR derived values likely shows that the HCR-based algorithm mis-identifies precipitation shafts as clouds. It would be helpful to focus only on radar beams with surface precipitation and carefully compare those with HSRL and MARCUS observations. Some plots demonstrating the method under precipitation conditions would also be helpful (Figure 3 is not detailed enough to provide insight into the quality of the algorithm). I am very doubtful that a method base on spectrum width can work because many different phenomena contribute to the observed spectrum width values and it is very difficult to tease the different contributions apart. Overall, I am wondering if the identification of Hbase is even needed for the current study? Given all the question marks regarding the method, I am wondering if it would be better to exclude it from the manuscript.
18. Line 326: How are "noisy pixels" defined and identified?
19. Figure 7: Please list the exact date and time of this example. RF 13 took place on February 20th but I could not find this example at the given time.
20. Sections 4.3.1 and 4.3.2: I am not sure what this comparison is supposed to tell us. Comparing bulk fractions of different phases is not meaningful given the differences in sampling (1D vs. 2D). For a true comparison, the authors would need to compare actual data points from in-cloud flight legs with HCR observations in close proximity. Then they could calculate some kind of hit/miss table or similar.
21. Section 4.3.3: The comparison with the PID method is interesting but shows significant differences. Only a pixel-by-pixel comparison with a different instrument (as was done for the PID method) will reveal which method actually works better.
22. Line 584: The authors claim that the HCR-phase method has strong capabilities for detecting mixed-phase clouds. However, this has not been shown or verified. Just because the method classifies many pixels as "mixed" does not mean that this represents reality.
Citation: https://doi.org/10.5194/egusphere-2025-874-RC2 - AC1: 'Reply on RC1 and RC2', Anik Das, 23 Aug 2025
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 428 | 57 | 13 | 498 | 32 | 7 | 26 |
- HTML: 428
- PDF: 57
- XML: 13
- Total: 498
- Supplement: 32
- BibTeX: 7
- EndNote: 26
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Xiaojian Zheng
Xiquan Dong
This preprint has been withdrawn.
- Preprint
(4113 KB) - Metadata XML
-
Supplement
(392 KB) - BibTeX
- EndNote
1. Overview of the paper
This study analyzes low-level marine boundary layer (MBL) clouds over the Southern Ocean using airborne radar and in-situ data from the SOCRATES campaign. It introduces a method to detect cloud boundaries with radar reflectivity and spectrum width, derives an empirical relationship between reflectivity (Z) and liquid water content (LWC), and classifies cloud phases using radar, temperature, and LWP. Single-layer low clouds (<3 km) made up ~85% of cases. The Z–LWC relation enabled accurate LWP estimates, validated against observations. The cloud phase classification identified 48.8% liquid, 23.3% mixed-phase, and 6.9% ice clouds, and showed good agreement with other established methods. Overall, the study offers a robust radar-based approach for cloud boundary and phase detection, supporting better satellite retrievals and climate modeling over the Southern Ocean.
This resubmitted version of the study presents clear improvements both in writing quality and in the validation of the proposed methodology. The structure is more refined, and the explanations are clearer, which enhances the overall readability of the manuscript. One of the major additions is the comprehensive comparison with multiple existing methods for cloud boundary characterization. This significantly strengthens the robustness of the authors' approach. Based on these elements, I would recommend only minor revisions before final publication.
2. General suggestions
The scientific objective is clear, and the methodological approach is generally robust, particularly the development of a radar-based classification for cloud phase. The Supplement is clear and provides important supporting information, especially concerning the in-situ datasets and instrumentation details.
However, several issues hinder the manuscript’s clarity and accessibility. First, the text suffers from a lack of consistency in terminology and definitions—key terms and acronyms such as SLW or LOW are introduced without prior explanation. Secondly, there are redundancies and repetitions across the sections, especially in the presentation of the classification results, which could be streamlined for conciseness. Additionally, while the discussion around uncertainties is both important and informative, the writing would benefit from reformulation for clarity and structure. On the methodological side, the rationale for excluding certain cloud types (e.g., fully glaciated clouds) should be more explicitly justified, especially in the context of deriving a Z-LWC relationship. Finally, more attention should be given to harmonizing visual elements (e.g., figure fonts and legends) and providing clearer context in tables and figures, including the comparison of cloud boundaries with in-situ data.
Overall, while the work is of clear interest to the cloud microphysics and remote sensing communities, the revised version appears well-refined, with notable improvements in structure, clarity, and scientific rigor.
3. Specific comments and technical corrections
Abstract.
L.26 : If you are referring to your “LOW” class, you need to define the acronym before using it. Otherwise, you can remove the capital letters or use “low-level clouds”.
Introduction.
L.35 : This might not be necessary. However, the references are neither in chronological order nor in alphabetical order.
L.38 : "SLW" is not defined before being used.
L.53–60 : This paragraph on uncertainties is important and interesting but not very clear. The authors could try to rephrase it.
L.63 : This is a long sentence (6 lines); please split it into two.
L.73 : It’s a bit vague—what were these inconsistencies?
Results.
L.131 : Maybe merge this sentence with the one on L.127.
L.149 : The ice phase (> 200 µm), although minor, is not used in your method? Is it only useful for comparing your classification with Alessandro’s?
L.174 : When you refer to a “subset of the dataset”, does this correspond to the “5th to 95th percentile interval of the dataset”?
L.178 : “FIGURE” → “Figure”
L.244 : No space between ~ and 22.2%.
L.256 : This is essentially what you say in L.226 — consider avoiding the repetition.
L.265 : You should add the term “mean” as in your table, otherwise it’s confusing — it sounds like you’re referring to maximum values.
L.295 : How do you explain a larger mean difference for cloud top altitude?
L.314 : “FIGURE” → “Figure”
L.319–328 : Formatting issue in the label of Figure 4 and repetition in the text. Please correct this paragraph.
L.324 : I understand the importance of stating that you focus on low-level clouds, but this is repeated too many times.
L.324 : Repetition — please revise.
L.336 : The authors refer to drizzle in both the text and Figure 5. You mention both “drizzle” and “liquid drizzle” — what is the difference? When you say “drizzle”, are you referring to freezing drizzle?
L.330–376 : In this paragraph, the authors mention several thresholds for classifying phases and for Figure 6. Some thresholds are cited with references (Wu et al., 2020a; …) and others are not — how were these chosen?
L.377 : Does this kind of “misclassification” occur frequently? Can the authors quantify this? Does a cloud with LWP ≥ 20 g/m² and LWCs > 0.2 g/m³ necessarily exclude the presence of ice?
L.448 : The authors could add a sentence at the beginning of paragraph 4 stating that only low-level clouds are used, in order to avoid repeated reminders later.
L.481 : The concept of scale plays a crucial role in method comparison — the authors should emphasize this point more clearly in the text.
L.525 : The authors should describe Table 2 in more detail. Otherwise, it may not be necessary to keep it in the main text; consider moving it to the supplementary materials, since you only mention it once.
L.574 : The percentages are calculated on classes that are not present in both methods (Mix % and Melting %), which could explain the larger differences.
L.582 : Exactly — you could perhaps add that the mixed-phase strongly depends on the observation scale (microphysical <--> macrophysical).
Figures.
Figure 3 : The font size for y-ticks is not the same in panels f and h. You could also increase the x-ticks size in panel h to match the one in subplot d.
Figure 4 : There is an issue in the label: “4 Low-level cloud phase classification method and discussion”.