The manuscript submitted by Viet Le et al. presents long-term measurements with several Doppler lidars in Finland for the purpose of depolarization measurements of atmospheric aerosol.

It is one of the first studies (if not the first) presenting a long-term study of depolarization ratio of aerosol observed by a Doppler lidar – which is very valuable. The remote locations in Finland make the result rather representative for less populated areas (even if station is characterized as semi-urban) and thus I see a high potential for follow-on studies in other regions. With the techniques and methods described in the manuscript, also other HALO Doppler Lidar in different region of the world can hopefully be utilized for future studies and thus collect a valuable data set. Thus, I consider this study pioneering and paving the road for other groups to do so similar.

The paper is relevant for the scientific community and well written and mostly well organized. Argumentation is logical and reasonable and it is worth publishing. The content would have been sufficient to for 3 publications, and sometimes is very technical, but I consider all these information as very valuable.

However, I have some minor comments, which are needed to make the excellent results better understandable (and reproducible) for the broader community of AMT and not only (Doppler) lidar experts. Sometimes, after addressing these points, the manuscript can be published.

General comments:

• The authors should properly introduce the variables they use, i.e., the attenuated backscatter coefficient and the volume (particle) depolarization ratio. And then explain, why in case of the HALO lidar, the molecular contribution can be neglected (i.e., why the authors can state that they measure the particle depol ratio directly and not the volume depol ratio). I consider this as important, because what the authors present would be not valid for a lidar with similar technique but operating at a shorter wavelength.:
  
  Therefore I recommend, either to show the formula for the particle depol, indicating why you can neglect molecular contributions, or a proper referring to other publications. This starts already in the abstract, while you are stating the bleed trough is sufficient to measure particle depol
• Furthermore, please make clear in the beginning what is the difference between depol_particle and depol_aerosol.
• As for the depol ratio, it should be also briefly stated, why the attenuated backscatter can be used instead of the particle backscatter. No need for long formulas, but a proper referring or short introduction is needed, i.e., why the transmission is close to 1 at this wavelength and therefore the att. bsc. is close to particle bsc. in non-cloud regions.
• In addition, please correct the unit for the attenuated backscatter throughout the manuscript (Mm-^1 sr^-1 instead of Mm^-1 only).
• The bleed-through is determined at cloud base, with a lot of efforts which are really appreciated (specially to avoid saturation). However, could you add a short statement, that the determined “B” is valid for the whole profile, i.e. also in conditions with much less SNR?

Figure 2a:

The depolarization ratios below the cloud base are significantly scattered, with values of -0-4 and + 0.07. Giving this plot, I could doubt the depolarization measurements in low SNR regimes. Can you comment on this?

Aerosol indicator description:

As you refer many times to Figure S3, I would prefer to include in in the manuscript. On the other hand, you never refer to Fig. 3 a – c. Please find a compromise! I think it would be helpful, to use the plots 3a-c for explaining the typing methodology together with plot S3.

Post processing:

Is mainly very technical and hard to understand as sometimes further information is missing. Is it possible to have a more detailed part of the description in the appendix and shorten the technical part in the manuscript? For example, move lines 212 to 229 to the appendix and extent the description a bit, so that other groups with HALO lidars can use similar approaches.

Specific:

line 43: Seems to be wrong reference for narrow Calipso swath?

Line 46: Please rephrase sentence.

108: Remove brackets from citation

109: Why do you need the focal length to determine the attenuated backscatter, please briefly explain or give proper reference.

166: I find this abbreviation “AI” quite challenging, giving the fact that it is often used for artificial intelligence. Consider to change to AIA?

178: Here you use a threshold of 10^-5.5 for aerosol cloud discrimination. Later you use 10^-7 for precipitation. This is a bit in contradiction. Can you explain better?

179: Please also briefly describe what you mean with a median and maximum kernel. It is yet ambiguous.

210: “….were averaged for 1 hour before calculating delta_aerosol.”

So basically delta_aerosol is just the average of delta_particle in aerosol only regions, correct? Please write exactly what you mean.

213: “…a weak 2nd order polynomial shape (c.f. Manninen et al., 2016) appears in SNRco and SNRcross…” : From the given reference, it is not clear what is meant with second order polynomial shape appearing in SNR. Please describe in more detail. Does it mean electronic oscillations in the signal? It becomes more clear later, but should be clearly described here.

215: I do not understand this part of the sentence: “…2021), this component of the noise  floor has been accounted for through a fit to SNR profiles, …” Please rephrase!

220: What is “with wavelet bior2.6.” ??? is this a fragment or a reference missing?

221: “Next, the variance of the noise in the SNR is removed by applying a hard threshold shrinkage function using universal thresholding (Donoho and Johnstone, 1994) to the approximation and detail coefficients from level 1 to 4.” Please explain in more detail. This is not understandable. For example, what is level 1 to 4? Was never introduced.

Line 234: A reference to table 3  would be already great here.

Figure 4d: Please enlarge the scale of x-axis so that one can see the depol values in the ice cloud (is it an ice cloud?). I.e. from 0 to 0.6

Figure 4b-d are never referenced. Please do so at the appropriate places in the text.

244: 2.3.1 subsection is introduced, but there is no 2.3.2 Please check sectioning in whole manuscript.

261: Sentence not understandable without further background. What “chains” shall converge? What are sampling chains? Please expand description here (or more in appendix)

Sec 3.2.: very interesting and impressive results!

421: Any idea what kind of aerosol this can be in September with high depol?  I consider for Pollen it’s too late, is it?

431: The minimum range of the lidar is stated later, but it would be good to give the values already here and maybe even in Tab. 2.

438: 3.4.1 subsection is introduced, but there is no 3.4.2 Please check sectioning in whole manuscript.

465: The case studies come a bit out of nothing, but are very interesting. However, please consider to move them to before the long-term analysis....Furthermore, do you have one case study with low depol in summer?

472: To compare the delta_aerosol to dust values at other wavelengths is a bit critical. Especially, as you later show similar values for Pollen. Furthermore, some studies show that depol ratio for dust is highest in the visible and is decreasing towards longer wavelengths (1064 nm). Can you comment on this? Furthermore, do you have any more information/proof for dust presence in this case study? E.g. DREAM or CAMS model prediction? Evidence from nearby lidar stations?

This also reminds me, that it would be interesting if  you ever have observed a lofted aerosol layer in summer with low depol? Smoke or so? Or is your depol always high?

500: Please also state here already somewhere that you developed a complete new AIA for HALO systems. This is really remarkable.

515: “The overall delta_aerosol are 0.13 ± 0.08 in Hyytiälä, 0.11 ± 0.07 in Vehmasmäki, 0.09 ± 0.07 in Utö and 0.07 ± 0.08 in Sodankylä.”: Interesting! This means that in contrast to many other places in the world, you are mostly prone to Pollen (continental background) in summer in case of clear sky. Maybe worth stating

542: Code availability: It would be great if you could make a release on github and publish this frozen version (referred to in this publication) on zenodo or similar for getting an DOI.

545: Data availability: Lidar data are available upon request from the authors. Such statements are not anymore acceptable. Please consider to provide the data via a data portal.

Supplement:

Caption Table S2 need to be expanded: More information needed! What is shown? The increase in data points? Not clear now.

Caption Table S4 need to be expanded: More information needed: Linear regression between what? I.e. what is y, what is x? What about an offset? What is p what is R squared?

Please check all captions in supplement!

The manuscript demonstrates the capability of Halo doppler lidars to retrieve the particle linear depolarization ratio (δ_aerosol) at 1565 nm and investigates the stability of the background noise levels and the performance of the internal polarizer, both of which have to be considered for the retrieval of δ_aerosol. The retrieved particle linear depolarization ratio is used to perform a seasonal characterization of the suspended aerosols at four different sites of Finland of different environments (e.g. marine, forest, rural, sub artic regions) using four-year lidar measurements from 2016 to 2019. For the seasonal analysis on the δ_aerosol, a new aerosol identification (AI) algorithm has been developed to separate the aerosol from the cloud layers. The performance of the AI algorithm has been compared against the Cloudnet classification algorithm showing adequate performance on identifying and separating the aerosol layers from the clouds. Moreover, two case studies, one Saharan dust transport event and one pollen from continental areas event, have been selected to be presented in the study while using complimentary data for the air mass origin from FLEXPART model.

To sum up, the manuscript presents new methods and new developments falling within the scope of the journal. It is well-structured and well-written even though I have the feeling that some parts could be further explained and/or discussed in order to be easier to follow for a reader non-relevant to the study. The scientific significance makes the manuscript suitable for publication in AMT, after some minor revisions have been considered from the authors.

Comments:

A general comment for the authors is to explain better the difference between δ (is it the volume linear depolarization ratio?) and δ_aerosol (particle linear depol. ratio?) in the text as it might be confusing for the readers

• Line 53: “…δ_aerosol is measured using Raman lidar (Engelmann et al., 2016; Baars et al., 2016)”

The depolarization ratio profile can only be retrieved when a lidar is equipped with depolarization channels, thus from a depolarization lidar (not all Raman lidars have depolarization channels). PollyXTs are Raman and depolarization (and water-vapor) lidar systems.

Please rephrase and add also older studies since depolarization in lidar measurements is being used even before 2016 (see for example Sassen, K.: Polarization in Lidar, in Lidar: Range-Resolved Optical Remote Sensing of the Atmosphere, edited by C. Weitkamp, pp. 19–42, Springer New York, New York, NY., 2005)

• Line 124: “…estimate the internal polarizer performance, or bleed-through (Vakkari et al., 2021)”

Here you could add a short description about what the bleed-through is, in order to be easier for an independent reader to follow. Then if the reader wants more details, can read the Vakkari et al., 2021.

• Line 147: “… The mean and standard deviation of δ at the cloud base were then calculated and used to determine the bleed through”

Here the authors could also mention that the B term they are using in eq. 3 is the calculated mean value of δ.

Moreover, the values of δ at the cloud base do they correspond to an averaged δ inside the range gate that is identified as the cloud base?

• Lines 159 – 263:

These sections are too technical and refer to readers with specific expertise in HALO wind lidar technologies and data handling. Therefore, I would suggest to move these sections in an appendix and maybe lines 170-208 where the steps of AI are described to be moved at the supplement below figure S3. Instead of this too technical description you could add a more qualitative description of the AI and the signal post-processing sections to achieve a smoother transition to section for the investigation of the air mass origins. However, Figure 3 is a good way to demonstrate the results on the target classification from AI and thus could remain in the main text but with further discussion.

• Line 166: “…2D Kernel manipulation…”

What is this 2D-Kernel manipulation? Maybe a description about the basic idea in a sentence along with a reference or link would be helpful. This way the reader can understand that this method is bout image processing.

• Lines 299 – 300: “...This is due to the Streamline Pro models were configured to utilize only half of the bandwidth, i.e half the Nyquist velocity…”

Could the authors discuss a bit further how the half of bandwidth can affect the noise floor levels? Maybe you could add an extra explanatory sentence along with a reference.

• Line 307: “...Figure 6 displays the time series of δ at liquid cloud base for each instrument in the network.”

Here the authors could also add that they are using the time series of δ in order to calculate the bleed through (B) and its uncertainty as the mean value and std of δ over time (as you state in lines 318–320).

• Figure 6:

Data from the same stations are used in Figures 5 and 6. A suggestion that could help the reader understand faster Figure 6 is to use the same colors for each station of Figure 5 in Figure 6, too.

• Section 3.4.1: The effect of relative humidity

In this section the authors mention the diurnal pattern of δ_aerosol and RH below 300 m a.g.l.. How these RH patterns obtained if not measured? Please clarify this also in the caption of Figure 10.