First review of “A method to retrieve mixed phase cloud vertical structure from airborne lidar” by Crosbie et al.,

January 15, 2025

Reviewer Recommendation:

Somewhere between Minor and Major revisions required

Summary:

The submitted manuscript introduces an empirical model in an attempt to separate the origin of depolarization observed with a lidar. In my understanding, the authors take observations that are expected to be from liquid only clouds, as screened by temperature, and tune an empirical model to tightly match the depolarization profile observed. This model accounts for depolarization due to multiple scattering. When taking the same model and applying it to clouds that are not necessarily expected to be liquid only, deviations are then attributed to ice. In general, this approach seems reasonable and caveats and assumptions are fairly well documented.

In a world that is often filled with algorithms that over promise and under deliver, this manuscript is, in my opinion, candid with limitations and expectations. I don’t often get to write that and I think it is important to note. In general, the paper is well written; it has generally excellent clarity and citations are appropriate in my opinion. It fits well within my understanding of the scope of AMT and I believe it is a contribution that should be published.

I note some elements that I find unclear, and I believe some revisions will improve the draft. That said, it is my recommendation that this manuscript be accepted with revisions (somewhere between major and minor).

Major Comments:

1. My prime concern is that I am unclear how far to trust the validation of this technique. In general, I am sympathetic to the fact that you are using airborne measurements and that they are generally infrequent (especially because in this case you require 2 planes in the same patch of sky simultaneously) and you are trying to see into a region where few technique exist. However, in Figure 6 & 7 the flight track of the Falcon and your measurements seem to be worryingly far away for a worryingly large fraction of time. This seems to be the crux of the problem: lidar has poor penetration depth into clouds but you need coincident measurements to validate the technique. My specific comments/questions are:
   1. It might be best to specify how much total data overlap there is (in terms of hours or kilometers or measurements or something else), i.e. when the Falcon is within the realm of your measurements that you trust (to my novice eye, 15-20 km range on Figure 6 seems to be reasonable overlap but there seems to be almost none from 30-40 km on the same figure).
   2. Lines 428-429: It is clear to me that you are picking +-300 meters as a threshold because you need to have some reasonable sample size. However, it is unclear to me how sensitive to this limit (+- 300 meters) your results are? 300-meter penetration into liquid clouds seems optimistic. Your Figure 3 & 4 suggests that your penetration depth into clouds may reach this threshold but can also be limited to 50-60 meters. As I read this, I think either: 1) your results need to be shown to be robust for various depths of observation or 2) the sensitivity to that change needs to be established. At minimum, if you use a more restrictive threshold, I believe you need to comment on what it does to your interpretation of results.
   3. Line 380-382: Given that you are using these examples as validation for your method, it seems like a non sequitur to specify that the method works and that the Falcon’s data can’t be used because it was too far away.
   4. I presume the GRD case from Table 1 has no flight validation data. Is that true? If that is true…is it a fair control case? If not, you should more clearly say how that data is compared/validated. For example..where is your temperature data coming from? 
2. Section 2.3.1 and Lines 412-414: My understanding of the observation of oriented ice crystals is that their properties are highly dependent on the angle of observation. For example looking at Noel and Chepfer 2010, it is my understanding that a change from 0.3 degrees from nadir and 3 degrees from nadir basically removed CALIOP’s sensitivity to oriented ice crystals. With that being said, are your observational legs done at nadir exactly or is there some angle of attack? It the plane pitching/rolling in any way that would change your sensitivity? Does that affect your measurements of oriented ice crystals? A comment clarifying your sensitivity would be warranted in my opinion.

Noel, V., and H. Chepfer (2010), A global view of horizontally oriented crystals in ice clouds from Cloud‐Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), J. Geophys. Res., 115, D00H23, doi:10.1029/2009JD012365

Minor Comments:

1. Lines 122-124: You mention cross talk in the molecular channel due to imperfect extinction by your HSRL filter. This physical effect is absent your equation 2. I would suggest including that term.
2. Equations 10/11/13: It is unclear to use the same variable (z) as both the integrand and the upper bound of integration. Suggest integrating over z’ or some other stylized version to differentiate your two variables.
3. Line 192: I would expect this value to be lower (like maybe 8 to 10). I think therefore that a citation for the lidar ratio used here is needed.
4. Table 1: What is “Cloud Temperature”? Is it really cloud-top temperature or average temperature or something else? Please clarify.
5. Table 2: I would suggest adding the RMSE value that would be calculated if using your “Final” coefficient values as another row. It would be helpful to see how much your averaged coefficients alter the fit to each test case.
6. Table 2: Why is your GRD case included in Table 1 and not here?
7. Figure 2: I would suggest changing your colors for each case to perhaps use different symbols (diamonds, triangles and so forth). For example, I am having a hard time differentiating ACTHIGH and GRD. Additionally, I am not Red-Green colorblind but CPEX and ACTLOW1 might be difficult to differentiate for someone that is.
8. Figure 3: Why is there a huge gap in measurements from 60 meters to 120 meters? Is that last data point truly trustworthy or are your data filters kicking in with just one point serendipitously falling into the end?
9. Figure 3: It might be helpful to add an ordinate axis for optical depth instead of just using physical depth. 
10. Figure 3 & 4: I would include in the caption what the difference between gray and blue does is (specifically looking at panels C/F/I). I presume it is for observed and modeled depolarization, respectively, but it would be good to clarify.

The manuscript presents an empirical methodology for identifying ice and liquid water features within atmospheric clouds. Hydrometeor characterization is a well-documented challenge in cloud research and has been addressed in prior studies. However, this work is the first to develop and test a method specifically designed for an airborne lidar system. The study leverages extensive measurement campaigns, incorporating lidar observations and in situ data, adding significant value to the analysis. 

While the manuscript outlines the general approach effectively, certain methodological details require further elaboration. Key geometrical considerations, such as the field-of-view and distance to the cloud, were addressed, and the results support the reliability of the approach. Nevertheless, questions remain regarding the quality of the lidar data, particularly the reported depolarization values, which appear anomalously high.

Additionally, the manuscript suggests that the system is configured to measure only in a downward direction. This setup has significant implications and needs to be explicitly addressed. Unlike ground-based lidars, the downward-facing configuration could lead to depolarization increases due to multiple scattering, which may at some point merge with the single-scattering depolarization of ice crystals. This is contrary to the typical behavior observed in ground-based systems. Moreover, strong attenuation of the signal in liquid layers near the cloud top might obscure ice or liquid water signatures beneath these layers.

To enhance the robustness of the method,  synthetic scenes or ground-based observations might be beneficial in the future. 

Specific Comments:

• Line 100: Using the HSRL, backscatter and extinction coefficients can be unambiguously derived. However, how are these parameters affected in liquid clouds? Could crosstalk introduce significant errors in the measurements? Additionally, why was the extinction from the molecular channel not considered in this approach?
• Equations 1, 3, and 5: These equations assume an idealized polarization system. How do the authors ensure the absence of polarizing effects, given the potential inefficiencies in each channel? (Mattis et al. 2009, Freudenthaler 2016). Furthermore, the reported depolarization values for liquid layers appear unusually high.
• Equations 1 and 2: The use of the same constant for both channels requires justification. Since the detectors and optical paths differ, how is it ensured that the constants are identical? And shouldn’t the cross-polarized channel in Equation 3 have a different constant? A polarization calibration might be necessary.
• Equations 8 and 9: These equations describe the attenuated backscatter from the system to range z, not from the cloud base to z. The left-hand equality seems incorrect as written. Shouldn’t the left-hand side be divided by (T_norm^2 * T_m^2)?
• Equations 10–13: These equations do not appear essential for introducing integrated quantities. Please consider removing them to streamline the manuscript.
• Line 198: The explanation regarding larger lidar ratios for ice crystals underestimating extinction more than liquid clouds is unclear. While multiple scattering contributes to extinction underestimation, this applies to both cases, ice or liquid. Please clarify. What reference supports Equation 16, and what are its limitations? The equation seems to assume a constant extinction coefficient profile, so additional reasoning is required.
• Equation 19: With five constants to retrieve, how good is the performance of this retrieval? Given the potential for multiple solutions, how do the authors ensure an optimal and unique solution?
• Figure 3: The differences among the three cases are difficult to discern. Extending the vertical range in the figure could enhance interpretability.
• Section 3.3: The utility of the retrieved microphysical information is unclear. The primary instrument, the 2DS optical array, only detects hydrometeors larger than 90 μm. This limits its ability to observe cloud droplets, which are typically smaller than this threshold.
• Sections 3.5–3.7: When introducing the retrieval scheme, it would be beneficial to first present the lidar-observed variables (e.g., attenuated backscatter, depolarization).
• Figure 5: Depolarization values up to 50% in warm liquid clouds seem unusually high. Such values are typically associated with ice crystals. What depolarization values are observed for ice crystals?
• Finally, the manuscript would benefit from a detailed author contribution statement. Given the technical complexity of the work and the number of coauthors, such a statement would clarify the specific roles of each contributor.

References:

Mattis, I., Tesche, M., Grein, M., Freudenthaler, V., and Müller, D.: Systematic error of lidar profiles caused by a polarization dependent receiver transmission: quantification and error correction scheme, Appl. Optics, 48, 2742–2751, 2009.

Freudenthaler, V.: About the effects of polarising optics on lidar signals and the Δ90 calibration, Atmos. Meas. Tech., 9, 4181–4255, https://doi.org/10.5194/amt-9-4181-2016, 2016.