In this contribution, the authors seek to validate horizontal line of sight (HLOS) winds from Aeolus against ground-based radar measurements in the Arctic and Antarctic. Compared to MARA in Antarctica and ESRAD in the Arctic, the Aeolus HLOS product generally compares favourably, with a significant bias documented in austral summer over Antarctica.

This well-motivated study seeks to address whether bias-corrected Aeolus data validates well against ground-based remotely sensed data in the polar regions. It represents a significant contribution in the effort to characterize Aeolus measurements in data sparse regions that are important for numerical weather prediction applications and for scientific understanding. The work is generally clearly presented and laid out and within the scope of AMT and the Aeolus special collection. My comments do not suggest any major changes are necessary to bring the paper to publishable form. I therefore suggest minor revisions to the current ms.

The main analysis recommendation is to see if more insight can be gained by digging further into the origin of the differences seen between ESRAD and Aeolus, in particular to what extent biases in ESRA might account for those differences (versus the lack of coincident data, particularly in the Mie channel, which is clearly described). Regarding ESRAD biases, taking a quick look at Belova et al. (2020) AMTD (https://amt.copernicus.org/preprints/amt-2020-405/amt-2020-405.pdf), the differences of ESRAD from radiosonde, HARMONIE-AROME, and ERA5 are complex, and I couldn't find a clear statement in it about recommended bias correction to ESRAD (apologies if I missed it).  I was looking for this because the nature of and rationale for the bias correction made to ESRAD in the current paper (e.g. p.7, line 85-87) is not entirely clear. At p.4, ll.14-16, what does it mean to say that there is a 'systematic underestimate of wind speed by about 8% in zonal wind and 25% in meridional wind'. Wind speed is a scalar, so is the issue that ESRAD winds are weaker (lower speed) than the other products? From the other Belova et al. (2020) paper in AMTD, it seems like these statistics refer to separate linear regressions carried out for U and V. Is this suitable to apply when comparing ESRAD to an HLOS product like Aeolus? Again, why separately correct zonal and meridional winds rather than wind speed and direction? Have the authors done a separate analysis of the comparison of ESRAD to Aeolus without the Belova et al. 2020 bias corrections on ESRAD? If so, what does this reveal?

Also regarding ESRAD, it is interesting that the fcx_aeolus was implemented as a special effort for the calibration/validation effort (p.7,l.78, and Table 1). But apart from the mention on p.7, a separate analysis of this data does not appear. Were systematic differences were found for this mode?

The main textual revisions I recommend are to clarify in the abstract and elsewhere that the analysis is based on a single six-month period of 1 July-31 December 2019, to clarify what the nature of the 'winter' and 'summer' seasons are here, and to discuss the implications of the use of a single season to characterize these errors. The reader might assume from the abstract that the analysis would take place over the entire Aeolus period instead of just when the homogenized and reprocessed 2B10 data was available. The authors could expand on their justification of only including this data in their analysis at p.3, line 84. In addition, while it is ok to characterize 1 July- 24 September  (24 September - 31 December) as boreal (austral) 'summer', the complementary periods are shoulder seasons (boreal autumn/austral spring) and not 'winter'. This nomenclature is used when the authors interpret some of the wind biases in terms of winter-versus-summer seasonality (p.12, ll.63-65; p.16, ll13-14). This interpretation should mention that results could be influenced by a small number of weather systems that happened to occur at these sites during the six-month analysis period.

It is clear that the amount of sunlight distinguishes the two periods (p.8, ll.04-05) and the reason to separate the periods in this way is to focus on the role of insolation backscatter in controlling errors. So could the authors call the 'winter' period something like the 'non-summer' or 'non-sunlit' period?

Specific comments:

* p.2, l.43: Clarify what is meant by 'hot pixels'.

* p.2, l.55: Is this comment necessary for this paper? Perhaps it would be better placed in the discussion. The description of the 'limitation' of the Aeolus orbit design is distracting. The sun-synchronous orbit presents a challenge for calibration/validation but it is a reasonable strategy for capturing free tropospheric/lower stratospheric winds whose diurnal cycle is relatively weak, especially on the typical horizontal measurement scales of 10-100km achievable by this technology.

* p.3, l.78: to about *an* 87 km horizontal

* p.3, l.79: for better impact on weather prediction -> to improve the impact of the retrieved Aeolus winds on numerical weather prediction

* p.6, l.38: ">8 m s^-1": While it seems reasonable, how was this rejection threshold chosen, and what impact did it have on the results?

* p.6, starting l.45: Replace "Mie winds" with "Mie wind measurements"

* p.7, Table 1: Is there a typo in the end heights (104km, 100 km) - and if not is this consistent with the descriptions in Section 2?

* Figures 3 and 6 show strong negative values for ascending HLOS in Aeolus but not in MARA; can the authors comment on these outliers? These extreme differences might be worth pointing out. Is it possible that there are ranges of horizontal wind speeds that aren't captured by MARA's retrieval?

* p.12, l.63: The wording is confusing here, suggest "... do not vary between the seasons and weater systems are more variable in winter rather than in summer, which would lead to more spatial variability in winter than in summer". But again, as pointed out above, it is speculative to make this kind of generalization when only analyzing a single season, especially since there is a lot of synoptic variability in Antarctica year-round.

* p.16, l.10: "For winter the random differences are higher than in the comparison with radiosondes." Could this be quantified?

This manuscript presents a comparison of Aeolus HLOS wind observations by two ground-based wind profilers in Antarctica and Arctic that takes into account the uncertainty statistics of the collocation observations. It contributes to characterize Aeolus measurements in data sparse regions. This paper therefore is of added value to the community and it is expected that such analysis will be used in future comparisons of Aeolus instrument and impact study for numerical weather prediction in polar area. This work is relevant and timely, considering the spare measurements in polar area. The manuscript is well written, logically organized, and contains clear figures and tables that support the presentation of the analysis. I suggest minor revisions of current manuscript before publication.

Comment / questions:

1. The agreement between Aeolus and radar winds is generally very good considering the spatial/temporal differences between two measurements. Some exceptions may need further investigation or discussions if it is not easy to clarify with limited comparisons. For instance, the systematic bias for Mie cloudy wind at MARA for ascending passes in summer. Scattered sunlight from ice-cap in summer can increase the background noise leading to the large random noise but not playing high impact on bias. Other point is why the systematic bias happens for ascending passes? Does the observation angle of the satellite affect it? Aeolus's off-nadir angle and the sun's altitude angle form an angle that approximately matches the incidence and reflection, which is the opposite of the situation when the orbit is descending, causing the sun background light scattered into the telescope to be stronger?
2. Figure 4 also shows that the bias descending Rayleigh measurements at MARA is larger than that of ascending orbit. It would be appreciated to see the analysis in discussion together with Mie winds.
3. For ESRAD measurements, there is a larger bias for Mie/ascending/Winter that cannot be easily explained by the above-mentioned reasons. The authors explain a small negative bias for all Rayleigh winds, on average -1 m/s that might be the bias from the systematic offset for ground-base radar. According to P4. L14: “These show a systematic underestimate of wind speed by about 8% in zonal wind and 25% in meridional wind at ESRAD, most likely due to nonrandom noise which cannot be easily removed”,
4. For Table 2 MARA wind comparison, why the intercept and bias are larger in winter than winds in summer when skylight background is much lower? It is also interested to see the small random error in summer measurement, which is contrary to the above law. Could it be an instrument offset/drift of radar local oscillator due to the extreme low temperature in Antarctic?

Minor points:

Figure 2. It is better to indicate the ascending and descending orbit for Rayleigh wind as well, even though it can be guessed from Mie plots.

On 22 August 2018, ESA launched the first Doppler wind lidar onboard the satellite Aeolus into space. Circling the Earth on a sun-synchronous dawn-dusk orbit at a mean altitude of 320 km, global wind profiles are obtained within 7 days. The dual-channel receiver design allows acquiring Rayleigh wind measurements in aerosol-free regions as well as Mie wind measurements in aerosol or cloud layers. In order to use Aeolus data in NWP models and to improve the data quality in newer processor versions, the systematic and random errors of Aeolus must be understood. 

In previous papers and analysis, it was already shown, that the systematic error of Aeolus shows a latitudinal and longitudinal behavior - results which could only be obtained by global validations. Numerical weather prediction models were essential to provide such findings but could be affected by potential model biases. Hence, a validation of Aeolus observations with independent remote sensing or in-situ measurements distributed over the globe are a crucial part of the Aeolus Cal/Val activities. 

In this regard, this manuscript provides a useful contribution for the validation of Aeolus wind observations by using collocated measurements from radar wind profilers covering particle-free regions as well as regions with particles in the troposphere and lowermost stratosphere with high temporal resolution. The instruments are located in Arctic Sweden and Antarctica, regions where observational wind data is sparse. 

The manuscript is well structured and written, presenting the obtained results with adequate figures. The paper deserves publication after some minor revisions.

General comments

1. In the manuscript the reprocessed data set from 1 July 2019 until 31 December 2019 is validated. First of all, this is an important information which should be mentioned in the abstract. The second point is, did you also analyze the operational data set from this time period in order to estimate the improvements of the new processor versions especially at the locations near the poles, where this information could be of interest for the processor developers?
2. Aeolus wind observations are filtered for an error estimate threshold of 8 m/s for Rayleigh-clear and 4.5 m/s for Mie-cloudy winds. On which facts is this QC-criteria based? Did you try other values and how this affects the number of available data points and the determined random error of Aeolus wind observations?
3. As horizontal collocation criteria a radius of 100 km around the radar wind profiler sites was chosen. The location of the two RWP sites close to the poles provide the opportunity to cover many Aeolus orbits at this latitudes where the satellite tracks are closer together. Did you try to increase the radius to a larger value (120 or 130 km) to see if this could improve the statistics by including more data points although the representativeness error would increase?
4. How does the inclusion of all Rayleigh-clear observations within the horizontal collocation radius affect the validation instead of using only the closest observations? This would be consistent with the approach which is applied for Mie-cloudy wind measurements and would provide more data points.
5. Have the authors included the random errors of the RWP measurements as well as an estimate of the representativeness error in the determination of the Aeolus wind observation error? Otherwise the determined random error of Aeolus would be a combination of the different errors.
6. An ERA5 model comparison which was performed in addition was mentioned several times in the manuscript. Did the authors consider to include this analysis in this work also in view of the possible imperfect bias correction of the ESRAD system? 

Specific comments

1. P.1 L.20: Please include the data set time period and mention that reprocessed data was used in the analysis.
2. P.3 L.65: Please include a short overview of the manuscript structure
3. P.3 L.76: "20 laser pulses": Until January 2019, 19 laser pulses were accumulated for one measurement. Afterwards it was only 18 pulses. See Weiler et al. 2020 or Lux et al. 2021 ("ALADIN laser frequency stability...")
4. P.3 L.79: Mie winds were horizontally averaged up to 14 km as of March 2019. Also horizontal averaging lengths smaller than 14 km are possible.
5. P.4 L.14: How was the ESRAD bias correction done? Is this correction stable over all 6 months and the covered heights? From Belova et al. 2020, it seems not to be a constant offset over all heights. 
6. P.5 Fig.2: What is plotted here? Are these the locations of the single wind observations in the L2B product? How does the number of Aeolus wind observations which are used in the comparison fit to the numbers (N points) from Table 2 and 3? For me it looks like there were much more Mie winds used than Rayleigh winds which is not represented by the numbers in the tables. Please clarify.
7. P.6 L.37: Is the QC-criteria based on an error estimate threshold applied before choosing the closest Rayleigh-clear wind observation or afterwards? Please clarify.
8. P.6 L.36: To better understand the statement "24 or 27 km" the authors could mention that the Aeolus range bins are following a digital elevation model (DEM) and within a radius of 100 km around the MARA site a strong change in topography from 0 to around 3000 m is covered.
9. P.6 L.40: Between mid October and December there was a so called Strateole range bin setting (RBS) with a maximum height of 17 km (17 km or 19-20 km for the MARA site) and an AMV setting with maximum height of 14 km for 2 weeks in between.
10. P.6 L.42: The height resolution is only possible as a multiple of 250m. Please check if 1130 m should be changed to 1000 or 1250 m.
11. P.6 L.43: The AMV RBS with 14 km maximum height was in operation for 2 weeks.
12. P.6 L.44: See comment P.3 L.79
13. P.6 L.44: The Mie wind RBS are not the same as the Rayleigh RBS. For the sake of completeness, the Mie wind RBS could be added.
14. P.6 L.49: Why did the authors create a single profile out of all Mie wind observations and did not compare single Mie wind observations with RWP wind measurements? Are the obtained Mie wind profiles free of gaps? Since the Mie wind RBS are not the same as the Rayleigh RBS settings, additional errors could be introduced when averaging Mie winds to the vertical Rayleigh wind resolution.
15. Subsection 3.1: Could the authors please comment if blacklisted data was excluded?
16. P.6 L.60: As already mentioned in comments above, the Mie and Rayleigh RBS are not the same. Are you also averaging to the Mie wind profiles?
17. P.8 L.16: To be consistent with the bias value in table 2 the value here should be changed to -1.9 m/s.
18. Fig.4, 5, 10 and 11: Please increase the font size to the same value as used in Fig.2.
19. Table 2 and 3: Can the authors please comment on the reason for the big difference in available data points for "Summer" and "Winter"? Are there less valid Aeolus or less valid RWP winds in winter?
20. P.10 L.35: -2 m/s -> -1.9 m/s to be consistent with values in table 2
21. P.13 Fig.10 b): Do you have any explanation for the strong negative bias for descending tracks at around 9 km altitude? Why is the std difference between ascending and descending orbits that large? 
22. P.16 L.21: As mentioned above: up to 14 km