The manuscript by Kirkwood et al. reports on the validation of the Aeolus wind product in the period from July 2019 to May 2021 by means of two radar wind profilers located in Arctic Sweden and coastal Antarctica, respectively. While the Rayleigh-clear and Mie-cloudy wind biases are found to be of the order of 1 m/s in both regions when considering the whole datasets, the random errors differ for the two detector channels. Subdividing the datasets into summer/winter and ascending/descending satellite passes reveals substantial differences in the Aeolus wind accuracy and precision.

General comments:

The work is certainly of interest to readers working in the field of atmospheric remote sensing, and particularly to the Aeolus community including the many different Cal/Val teams assessing the data quality of the Aeolus wind product. In this context, it is appreciated that the authors applied a comprehensive quality control to reject outliers from the dataset while keeping as many valid winds as possible for their statistical comparison. This is especially important since the number of collocated wind observations is rather small, given the fact that the satellite passes the two ground stations only 3 to 4 times a week. The QC method based on the estimated error and the modified Z-score is a useful approach to yield meaningful statistics despite the limited dataset.

However, while the statistical analysis used for the Aeolus wind validation is well-founded, the study fails to properly put the results in the context of other validation studies. In the final section of the manuscript, it is briefly mentioned that “biases are similar in magnitude to results from other locations”. Nevertheless, I think that it is necessary here to elaborate on the bias differences observed for ascending and descending orbits and for different seasons and to discuss whether these findings are in accordance with other Aeolus wind validations. Moreover, the evolution of the wind errors over the two-year period should be addressed and compared to the error assessments from other Cal/Val teams. In general, the manuscript cites only very few articles that are relevant with regard to the Aeolus wind validation and that might help to understand the found discrepancies in the wind errors.

Specific comments:

1. In the introduction there should be several more references to Aeolus validation studies based on wind comparisons against both model data and ground and airborne instruments.
2. The acronyms MARA and ESRAD should be introduced in the text. Although the details of the two wind profilers are given in Belova et al. (2021), it would be useful for the reader if the authors provided a brief description and some key specifications of both instruments (horizontal, vertical resolution, accuracy, precision, etc.). In particular, the measurement precision of the two profilers should be mentioned in order to justify that the bias and random error of the respective instrument are considered negligible in the statistical comparisons against the Aeolus wind data. In the abstract, the “combination of spatial differences, and the random errors in both radar winds and Aeolus winds” (l. 20 f.) is mentioned, but not quantified later in the manuscript. Please add a short estimation of the respective contributions to the determined Aeolus wind random errors.
3. The restrictions on the radar winds (l. 104 f.) should be explained.
4. During the discussion of Fig. 1, the fraction of comparison points depending on the EE threshold should be addressed, too. Due to the relatively low number of points (a few hundreds according to Tables 1 and 2), this fraction has a significant influence on the statistical parameters and their uncertainty. For this reason, I also suggest to indicate the total number of valid wind comparisons (corresponding to 100%) in Figs. 1 and 2.
5. The summer and winter periods are not consistent in the text (l. 110) and the captions of Tables 1 and 3. Please check.
6. What are the reasons for the increasingly skewed wind error distribution for the Mie winds at ESRAD, as described in Sect. 5? Is this effect also observed for the wind comparison at MARA? If not, could the authors give reasons why?
7. I am missing a comparison of the validation results with those from previous studies. In particular, the differences in the systematic and random errors of the Aeolus winds for ascending and descending passes, different seasons and the two locations should be discussed in the context of similar investigations.
8. As the analysis covers a period of nearly two years, it would be interesting to see how the wind error has evolved over this time frame, especially since the Aeolus performance had degraded between 2019 and 2021 due to signal loss, as mentioned in l. 193ff. Given the relatively small amount of data points available for comparison, I would propose to split the two-year dataset into four sub-periods instead of combining the summer and winter months from different years. In this manner, the statistical results become more meaningful with a view to the signal degradation.

Technical corrections:

Line 17: The sentence: “For the Mie wind at ESRAD there is a clear problem with a distribution of wind differences which is skewed to positive values” should be paraphrased to, e.g. “The Mie wind error with respect to the wind data measured at ESRAD shows a skewed distribution toward positive values.”

Line 24: “doppler” should be changed to “Doppler”

Line 31: “effects on the mirror” should be changed to “effects on the primary telescope mirror”

Line 59: HLOS should be printed with capital letters throughout the manuscript.

Lines 73/93: “gaussian” should be changed to “Gaussian”

Line 76: The sentence should have a colon at the end before Eq. (4).

Line 65: “estimated random error (EE)” should be changed to “estimated error (EE, also included in the Aeolus Level_2B product)”

Tables 1-4: The units are missing for bias, SD, ScMAD.

Fig. 3.: The units are missing for bias, SD, ScMAD in the insets.

Figs. 7/8: The label of the x-axis should be changed from “Data” to “Rayleigh (Mie) wind error with respect to MARA (ESRAD), respectively.

Description:

The manuscript on “Extended validation of Aeolus winds with wind-profiling radars in Antarctica and Arctic Sweden” by Kirkwood et al. reports on the quality of winds measured by the ESA Aeolus Doppler wind lidar mission. Aeolus wind bias and random errors are derived from comparisons to ground based wind measurements performed by two radars, ESRAD in the NH, and MARA in SH arctic regions in the time-frame of July 2019 to May 2021. As such it is building on previous work by Belova et al. (AMT, 2021) by extending the period of interest and additionally applying a refined quality control for the Aeolus data as published by Lux et al. (AMT, 2022). The almost two years data set was also subdivided into separate analysis on summer and winter periods as well as ascending and descending orbits on the two sites.

General comments:

The work is of interest for the validation community in general, as it delivers an example of successfully applying the refined QC including the modified Z score to a comparison of Aeolus winds with Radar data. The long time period and the arctic position of the ground radars in the northern and southern hemisphere provide a rare dataset, thus the results are of high value for the Aeolus validation. The reported values for the Aeolus wind bias in the order of 1 m/s and random errors in the order of 5 – 6 m/s for Rayleigh and 4 – 6 m/s for Mie winds depending on site, season and orbit are in the range of what has been reported by other validation studies. The manuscript is well structured and clear in language. However, a detailed discussion and comparison to appropriate other studies should be added. There is also no reference to specific other high latitude comparisons like e.g. Chou et al. (AMT, 2022). A discussion on how the specific radar measurement errors contribute to the statistical analysis results should be added, addressing the sensitivity, measurement geometry and representativity.

Specific comments:

1. Already in the abstract it is stated that the derived random differences “… represent an upper bound for Aeolus wind random errors since they are due to a combination of spatial differences, and random errors in both radar winds and Aeolus winds.” In l 189 it is stated that ESRAD provides higher accuracy for the wind product. Both is not further specified in the manuscript. The radars description and performance should such be shortly introduced in addition to referencing to previous work. In Belova et al. standard deviations in the order of 4 m/s for MARA and 5 m/s for ESRAD 1h averages were derived from radiosonde comparisons, which need to be explained in the context of deriving Aeolus random errors as low as 4 m/s (in the example of Mie winds compared to ESRAD). As up to 100 km distance between radar and Aeolus measurements are accepted in the comparison, the representativity error could be estimated from comparing e.g. Mie winds in the proximity to longer distance winds. It should be clearly stated in section 2 or the Table captions that the statistical comparison results are not corrected for the radar measurement errors. Alternatively, the Aeolus errors should be also given after correction e.g. by assuming Gaussian errors for both instruments, if the radar error knowledge allows.
2. Based on the expansion on the radars as described in 1. above, please also provide a number of available valid Aeolus winds (Rayleigh and Mie) that would in principle be available for a comparison within the radar volume, but did not find a radar measurement counterpart because of different measurement principle and sensitivities. Without going into details which would be outside the scope of this study, such a number compared to the numbers given in the analysis would at least give a rough estimate for the comparison of both technologies. For the example of a single overpass, there could be valid Rayleigh winds at 4 km, but the radar did not produce a wind output at this altitude during this overpass, because of the meteorological or technological constraints. E.g. the number of compared winds in Fig. 6 decreases with altitude for the Mie case above 8 km for meteorological reasons, whereas the Rayleigh winds decrease above 12 km for range limitations of ESRAD. Expanding this to the subsets and full dataset would provide total available valid Aeolus winds that would be a valuable addition to the (a) and (d) plots of Fig. 4 and 6. So far these figures are restricted to the compared valid winds. Please add this fact in the caption. It is so far not clear what is the share of radar wind results w.r.t. Aeolus wind results within the region of focus. Please also expand on the radar error’s altitude dependency.
3. Refer to Belova et al. in l 61/62 for the projection to HLOS and please use capital letters throughout the manuscript for this abbreviation.
4. In Fig. 1 and 2 please state weather this is for the example of the full data set and not one of the subsets (this becomes clear not earlier than page 10). Also provide the total numbers in addition to the % as this has implications for the validity of the statistical results. Please also highlight the EE case which you selected as baseline for the subsequent analysis as described in l 103-104.
5. In l 105, please be more specific concerning “previously” e.g. refer to Belova et al. or other work.
6. These chosen EE thresholds are also applied to the data subsets, although the choice could be different when judging from charts provided for the subsets. Please elaborate on this issue.
7. Comparing Rayleigh and Mie random errors for MARA and ESRAD the Mie winter and ascending cases for MARA stand out, because the random errors are unexpectedly high or even higher compared to those for the same case Rayleigh winds. This could be an issue of the smaller data set (but the QC seems to act effectively according to Fig. 3), the measurement method and data overlap, the Aeolus performance state, analysis bugs or others. Please expand on this in the discussion. It would be a valuable addition to also show a wind speed difference histogram example with/without QC and to provide a discussion on how the results relate to the Aeolus performance state, which significantly changed in the analyzed period. Please connect the latter to what was published in the Belova corrigendum.
8. In l 111 the seasonal influence on the satellite performance should be added as a justification. This and the latitudinal effects are described as important performance drivers by Weiler et al. 2021, especially for bias.
9. In l 114 “except for bias” should be added, as there is a significant difference (bigger than the Aeolus bias specification) in table 1.

Technical corrections

1. In Fig. 3 caption please correct “in and” in the sentence: “Scatter plots of Aeolus HLOS Rayleigh-clear winds (a-c) and Mie-cloudy winds (d-f) vs MARA winds. (a),(d) show all orbits together, (b),(e) in and data show ...” and add units in the insets (also for Fig. 5).
2. In Tables 1 and 3 please add the used EE thresholds in the caption, rather than referring to the text. This eases inter-study comparisons within the validation community.
3. In Tables 1 and 3 please add SH/NH to the respective summer/winter cases and “full data period” or alike to ascending/descending column descriptions. Please also add units and the wind speed range measured for the different cases.
4. Focusing on bias values between +/- 1 m/s, the bias axes’ ticks are too coarse in Fig. 4 and 6. Please widen the bias plots and go to 1 m/s ticks. The number plot can be narrower to compensate.
5. In Fig. 7 please add a unit to the “Data”-axis and rename to wind-speed difference or alike