This manuscript proposes and applies Z-score-based QC procedures for the evaluation of Aeolus L2B wind retrievals. It is very well written and easy to follow. While the degree of novelty is a bit limited, the paper is certainly interesting and useful, and fits well into the scope of the journal. The results and methods are sound, and despite my best mental efforts I have only a few very minor comments, which are listed below. Finally, I want to mention that I especially appreciate the effort the authors have put into a clear language and high-quality figures, which makes the manuscript very accessible.

General comment:

The paper has a bit of a gift wrap structure. The methods and goals are only rather vaguely touched in the beginning of the introduction, and become clear more and more only as one reads on. I recommend to give a more specific outline for what the reader is to expect. 

 

Somewhat related, I find the naming / separation of sections 3 and 4 a bit confusing. Both sections are structured pretty much identically, Sec. 3 showing results for the model evaluation against ECMWF, Sec. 4 showing results for the AVATAR-T evaluation. The methods are indeed introduced in Sec. 3 as the name suggests, but it is somewhat of a results / methods hybrid. 

 

I can see the benefit of introducing the methods on the examples with ECMWF, but I think a dedicated methods-only section that gives a summary (perhaps even a schematic figure?) of all methods that are going to be used could help the reader to have a clear picture on what to expect. As mentioned earlier, right now the exact methodology reveals itself only over long stretches of the manuscript. 

Specific comments:

L25: I find the term "biased gross errors" a bit unusual and unclear.

L68: Is a Gaussian distribution actually appropriate for wind retrievals? Since wind speed is a bound variable, shouldn't the error distribution get more and more skewed as the retrievals get closer to the limit? I appreciate this is a common assumption for many variables that has also some practical reasons, but since such a strong focus is put on forcing the data into a normal distribution, some words on that might be helpful.

L91-93: This statement is distracting and unnecessary here. Also, I have the impression that "Rayleigh-clear" and "Mie-cloudy" are used more often than not later on, so perhaps just delete this statement altogether (or stick with the simple notation consistently, which I'd actually prefer).

L103: Where does this strong signal decrease over time actually come from?

L121/L321: "Whereas" is a rather unusual conjuction to start a sentence with and sounds a bit awkward to me. Perhaps better use "while"?

L297: "the distribution is far from normal" sounds a bit funny to me. Perhaps better "far from Gaussian"? 

L298: The value of 3.5 seems purely empirical. What was the decision criterion of Iglewicz and Hoaglin (1993), and is it likely to make it a good choice for your study as well?

L335: remove the comma after "Gaussian distribution"

L345: ECMWF are considered as absolute "truth" in the presented analyses. I'm wondering how likely it is that some of the supposed "gross errors" are actual rare extreme wind occurences that were not modelled properly but captured correctly by the observations? 

L352: Perhaps change to "it was *assumed* that"?

L386: Is 10 ms-1 a commonly used threshold? Perhaps add a rationale / reference?

L429/Fig 6.: The graph contains a whole lot of information which is hard to grasp as a whole. Perhaps add a more high-level description of what the purpose of the plot is before describing the axes and lines, etc. specifically?

L499: Perhaps change to "1:1 line"?

L573: There are no orange bars. I assume this should be "black bars"?

L599: Replace "On the contrary" with "In contrast"?

L639: The presented graph is a great way to summarize and visualize a lot of information, but I, personally, find it a bit exaggarated to call it "developing a new graph". I think it just draws the attention a bit away from the actually interesting part, which is the proposed systematic approach for selecting an EE threshold in a somewhat more objective manner. 

L654-657: I'm wondering how the statistics against ECMWF predictions would look like when using only the exact same locations as those in the AVATAR-T campaign? That is, how much of the mentioned deviations are just related to spatial representativeness?

L670-- I find the concluding remarks straying a bit off-topic. 

Review of Lux et al., 2022 “Quality control and error assessment of the Aeolus L2B wind results from the Joint Aeolus Tropical Atlantic Campaign”

The manuscript by Lux et al. addresses the methods of quality control and error assessment of the Aeolus ALADIN L2B wind measurements. The study investigates the influence of different QC approaches on the key statistical parameters through comparison of L2B data of both Rayleigh and Mie channels with ECMWF analysis and airborne lidar measurements within dedicated validation campaigns. The authors point out that the estimated error provided with the L2B wind product does not fully incorporate all relevant error sources and propose a two step QC, involving a modified Z-score with individually derived thresholds. The proposed method is intended for ensuring consistency and comparability among diverse validation studies.

The paper is carefully written, the experimental setup and validation methodology are thoroughly and comprehensively described, the graphical material is very high quality, whereas the conclusions and recommendations are well substantiated. Although the paper’s scientific significance and the breadth of potential implications are somewhat limited, I find this study useful and fitting well the scope of the Aeolus inter-journal special issue.

General remarks:

The abstract appears to be a bit too verbose and I would suggest to make it more focused on the key results and their implications

The discussion of the potential implications of the results presented is mostly missing. It would be useful to develop this aspect in the abstract and in the concluding section.

Specific remarks:

The title of the last section should probably be Discussion and summary instead of Summary and conclusions

L.573: should it be “Black bars”?

Review of Lux et al., 2022 “Quality control and error assessment of the Aeolus L2B wind results from the Joint Aeolus Tropical Atlantic Campaign”

The main objective of the Aeolus mission is the improvement of Numerical Weather Prediction (NWP) models by filling an observational gap in the global wind data coverage, especially over the oceans, the poles and in the tropics.

Two components are very important to help achieve that objective: i) providing wind estimates with proper Quality Control (QC) of the retrieved winds; ii) providing accurate estimation of the uncertainty of these retrievals to help with the Data Assimilation (DA) into the models.

The paper addresses the impact of different QC schemes on the results from validation against other sources of truth – airborne observations or collocated model fields.

Random Wind Error is defined as the standard deviation of a Gaussian error distribution around a truth value.  And so, the procedure for removal of outliers is a critical component.  Most QC schemes relay on the validity flags and on the Estimated Error (EE), both being part of the L2B data.  However, the maximum EE thresholds above which data are excluded are determined mostly subjectively and are not consistent among the different studies.

Furthermore, the Rayleigh EE is mostly determined by the Signal-to-Noise Ratio (SNR) of the atmospheric return signal, which has been varying over the course of the mission. The Rayleigh-clear EE also depends on the geographical location, as it considers the noise caused by the solar background radiation, which varies over the orbit. Importantly, other noise sources are not considered. The evolution of the Mie-cloudy EE is less affected by the signal trend and rather driven by the data processing algorithms and relating setting which were modified several times during the mission.

All these make the results obtained from different validation studies and campaigns difficult to compare, thus necessitating a more detailed treatments of the QC schemes and how they affect the statistical results (the estimation of the uncertainty).  This detailed treatment of the QC is, indeed, the goal of this paper.

The paper provides careful analyses of Quality Control schemes and their impact on statistical results from the evaluation of ALADIN winds, based on different campaigns and comparisons to models.

The authors developed a two-step QC based on the EE and on a modified Z-score for each observation.  This Z-score uses scaled median absolute deviation (MAD) which is more resilient to outliers and so a more robust measure of the wind error variability than the standard deviation.

Employing the two-step process resulted in error distribution with high degree of normality while retaining a large portion of the original dataset.  This resulted in significant improvement, especially in the Mie winds, decreasing the systematic error for both Rayleigh and Mie channels to 0.3 m/s.  Random errors were evaluated against two sources of truth ECMWF and airborne Doppler wind observations taken by the DWL onboard the Falcon during the JATAC which took place in September 21 in Cabo Verde. (Mie vs. model: 5.3 m∙s-1, Mie vs. DWL: 4.1 m∙s-1; Rayleigh vs. model: 7.8 m∙s-1; Rayleigh vs. DWL: 8.2 m∙s-1).

The value of the paper is in developing an improved QC scheme which results in the rejection of only the worst outliers.  Maybe even more importantly, the new QC scheme allows the development of better error characterization of the retrieved winds.  Both of these (rejection of outliers and improved error characterization) would have a very important impact on the outcome of future data assimilation into operational models.

The paper is written very well, the presentation is clear and the figures are well designed and contribute significantly to the understanding.

General comments:

I agree with the comment of RC-2 regarding the mostly missing discussion on the implications of the improved QC.  While, in my mind, the paper presents results with high scientific value (understanding the instrument characteristics, improving the outcomes of validation and the outcomes of DA efforts), the importance of the improved QC are not strongly highlighted either in terms of motivation, or in terms of impact on the scientific or applications efforts (e.g. operational DA).

Specific comment:

• When using the ECMWF winds as the truth against which the Aeolus winds are compared, up to a 12-hour ECMWF forecast is used so that the model data are independent from the Aeolus winds (they have not been assimilated yet). However, Aeolus data have been assimilated in the previous model runs (cycles). What is the possible impact of the fact that the previous cycles have already assimilated the Aeolus winds?
• The text on P. 23 that describes Fig. 10 has several inconsistencies with the figure - e.g. Figure 10 does not have orange bars; it is said that the 1-step QC statistics are given in a black inset while it is gray.