This manuscript by Ern et al uses data from the SABER and HIRDLS atmospheric limb sounders to quantify and characterise the intermittency of gravity wave momentum fluxes and potential energies in the terrestrial stratosphere and mesosphere. 

The paper is well-written, clear and technically accurate, and I would have no objections to it being published in something very close to the current form. I have a few comments, but none of these are major criticisms and are intended to help make the paper stronger.

Main Comments

-----------------------

[1] I do have a probable answer to one question raised in the manuscript. In several places (e.g lines 532-543, lines 657-664, line 730) the authors highlight that their results appear quantitatively inconsistent with a previous study (W2013a, reference below). Specifically, the wave intermittencies measured in the current study are consistently higher than those seen in W2013a. However, I believe this arises from a key methodological difference. In W2013a, we used an approach (described by WG2013, reference below) which selects for multiple waves in a given HIRDLS measurement, and which typically (see W2015, reference below) identifies four discrete waves at any one measurement location. In contrast, the current study identifies at most a single wave at each point (L173). These 'additional' waves in W2013a tend to be lower-amplitude and have smaller momentum fluxes (WG2013, W2015) than the 'main' waves measured by the method used here, and will hence tend to strongly pull intermittency values down. I think this methodological difference is likely to explain most if not all of the differences between the current study and W2013a.

I emphasise that this difference is not an mistake in experimental design in the current manuscript and I have no objections to the authors using a more cautious approach and identifying at most one wave as they do here - it is a perfectly valid choice. The WG2013 method has the advantage that it detects more smaller waves from the same data, but in particularly for HIRDLS can be negatively affected by a known fault with the instrument which will introduce a small height-varying population of nonexistent waves into the data (W2015) and hence would have to be treated cautiously for a study of this type. We had not identified this problem at the time of writing W2013a, but it shouldn't affect the results presented there too much as the effect is very small at the low altitudes that study focuses on, and even at high altitudes is only a few percent of the measured waves - i.e. I suspect that the differences between the current study and W2013a will be almost entirely methodological rather than due to this data issue.

[W2013a] Wright, Osprey and Gille, JGR 2013: 10.1002/jgrd.50869

[WG2013] Wright and Gille, GRL 2013: 10.1002/grl.50378

[W2015]  Wright, Osprey and Gille, ACP 2015: 10.5194/acp-15-8459-2015

[2] The methodological choice to normalise the distributions (L238-247) does make sense when the results are considered, but probably needs a little more justification. For QBO regions, where filtering varies strongly from year to year, the logic is clear and coherent, but I am less clear on the justification for doing so at extratropical latitudes (at least outside SSW periods). If this was a minor point I would be happy with the current presentation, but since this choice underpins most of the results presented I think it needs to be justified a bit more strongly.

[3]  In section 5.1, I was still confused after several readings as to exactly how the gradient effect (line 444 onwards) was being compensated for - if the normalisation is taking place at the level of the bin, then how does this reduce spatial biasing due to gradients within the bin? I suspect I am misunderstanding something here, and as such would appreciate this section being made clearer.

[4] A minor concern I do have is that the manuscript feels very long and could probably benefit from trimming, but this is not a critical problem and the paper does contain a lot of data which does justify this. If the authors do choose to trim it, I think it would be better to do so by slimming down each section rather than removing some parts entirely, and by reducing repetition between sections. 

Additional Comments

---------------------------------

L131: note that the resolution of HIRDLS drops sharply above 60km, averaging ~2km above this level - see e.g. Figure 5.1.1 of the HIRDLS Data Quality Document

HIRDLS DQD: https://docserver.gesdisc.eosdis.nasa.gov/repository/Mission/HIRDLS/3.3_Product_Documentation/3.3.5_Product_Quality/HIRDLS-DQD_V7.pdf

L148: what kind of high-pass filtering is being applied here? Some types could introduce small waves, which may pull measured intermittencies down.

L201 and other places: while it's reasonable clear that by 'average' the authors mean 'mean', the mode, median and mean are all types of average - I would suggest switching from 'average' to 'mean' throughout to avoid any confusion.

L273: additionally, presumably SABER cannot access the smallest horizontal wavelengths due to the inter-profile spacing being about double that of HIRDLS. This is mentioned in s4.2.2 later, but this is the first mention where that effect is relevant so it might help to put it here instead.

L339 onwards: similar high intermittencies are seen in the open Southern Ocean using AIRS data in Hindley et al 2019 (their figure 9). While this has a very different observational filter, it may be relevant to the discussion here too.

Hindley et al, ACP 2019: doi:10.5194/acp-19-15377-2019

Section 4.2.4 - I like this!

L470: I understand here that you mean individual profiles (i.e. single profiles against altitude), but the text as-written could be read as referring to single levels *within* a profile (i.e. a single altitude level of a given profile) - it might help to clarify this.

L484: This normalisation depends on the averaging method used and more detail about this would help - for example, are the averaged values being stored at the grid-corner or grid-centre before being interpolated to the profile location?

L515 onwards: could the lack of shorter vertical wavelengths and/or different normalisation areas cause this higher Gini coefficient estimate?

L544: this is remarkable given the very different observational filters involved - do you think this is a real similarity and (e.g.) that the same intermittency effects are being observed uniformly across the GW spectrum despite the very different physical scales, do you think the methods are actually observing the same waves, or do you think it's just a coincidence?

The manuscript investigates the global distribution of intermittency of gravity waves and its change with altitude using two limb sounding satellite measurements. The analysis is based on the probability distributions (and their fitting functions) and the Gini coefficient for the wave potential energy and momentum flux. An obvious strength of the analysis in this study is its broad spatial coverage (nearly global in the stratosphere and mesosphere). The author also introduced the importance of a proper normalization in the methodology for deriving the intermittency estimates. The results of the analysis are well presented with reasonable interpretations and with useful comparisons to previous observational and/or high-resolution modeling studies. The manuscript is also clarifying some limitations of the investigation due to characters of measurements (e.g., sampling issue; spectral coverage; noise range). I can strongly recommend this paper for publication after a minor revision (which will be rather technical). The following list includes only minor comments. Particularly the comments with '*' may be considered.

[ Specific Comments ]

L3: 'effect' of what ?

Eqs. (4)-(5): Would this part be fit more in Section 3.2 than here ?

L122: 'every about 60 days for about 60 days': Would the second part be removable ?

L134: One of the two 'instrument' should be removed.

L146: 'location': Would it be replaced with 'latitude, height' ? (if I understand correclty)

L156: 'different': between what ?

* L198: What are the displacements between overlapping bins ? (i.e., spacings of final estimates in the map)

L201: 'and' must be removed.

L240: 'single' values: What does this mean ?

* L240: 'normalized by the global distribution of median values': I would suggest clarifying more, with (for instance) 'spatially and temporally varying medians'.

Moreover, given the context, it seems that this median field varies even within a bin of single estimation (so that the local gradient of median is taken into account). However it is not clear how this inner-bin field is made. Later in L456, it is first mentioned that an interpolation has been used. I suggest introducing this method around here (or earlier).

L245: 'sensitivities': instrumental sensitivities ?

L247-250: (just a question) In other words, momentum flux distributions show an interannual difference only in their overall magnitudes but not in the shapes of distributions. What would this mean ? Is this a statistical nature of gravity waves ?

L253: 'contribute to ... for the unnormalized values': It would be more clear to say 'cause an overestimation of ... when the unnormalized values are used.'

L256: 'very close to unity': Why not exactly the unity ?

L256: 'the reciprocal of': This might be removable in the context.

L270: Here I had wondered about what the meaningful range in the left-end of the distribution would be, considering some measurement noises. Then later this information was found in Section 4.2.4 (it is so nice to have this).

I would suggest referring to that section here and, in case it is possible, providing the meaningful range in values (from about -2 in most cases ?) briefly.

L272: 'the SABER PDFs in Figs. 6c and 6d': These panels are for HIRDLS.

L276: '90th and': This should be removed, as the 90th percentile shows a slight increase in SABER.

L287: 'regions. The locations of ... are illustrated' can be shortened by 'regions illustrated'.

* L299-300: '... wider': How can they be compared with different units ?

* Table 3: '0.61 mPa' (1st row, 1st column): This value differs from Fig. 7a (0.51 mPa).

L302-305: Here the heights of the distribution tails relative to the lognormal functions (fitted for each distribution) are used for the comparison of the intermittency in the tropics and that in winter high latitudes. However, it should also be considered that the fitted lognormal function in the tropics has a width (~2 at 10^-5) that is not larger than that in the high latitudes (~2.4 at 10^-5). If this were not the case, the statement L302-305 would not simply hold.

L306: '90th percentiles': of unnormalized fluxes ?

L325: 'balloon' is missing.

L366: 'jet-related gravity source processes act more continuously' (in the northern hemisphere): Would this be contradictory to the fact that the mean momentum flux (unnormalized) is only less than half that in Southern Ocean in austral winter (Table 3) ?

L390: 'the regional difference in the potential energy PDFs ... to the corresponding difference in ...' ?

L408: 'HIRDLS and SABER' can be deleted (repetitive).

L457-458: Has the reason for this been mentioned before ?

L567: 'introduce additional': I would suggest changing this to 'alter' (or 'increase/decrease'), as the temporal changes of the atmosphere can also reduce the intermittency depending on the situation (Kim et al., 2021, JAS).

L580: '... with altitude, mainly in the mesosphere.' ? (In the stratosphere, the increase seems to be very small: 0.01--0.02?)

L614: 'We will ...': Section 6.1 also focussed on SABER. Please place this sentence to the earlier part.

L620: 'seen'

L642: 'gravity wave' (the first-appearing one) can be deleted.

L734: 'flat' at around what value (could you please include this) ?

The manuscript presents an extensive study of gravity-wave momentum fluxes and potential energy

intermittencies as observed by HIRDLS and SABER spaceborne instruments. Using these limb sounding instruments

enable the authors to provide a broad geographical and vertical coverage: only polar regions are excluded, and

intermittencies are reported from 30 km to 90 km of altitudes. This observational study gathers a lot of

information on gravity-wave intermittency, and may thus provide very helpful guidance for parameterizations in

GCMs. The article is well-written, easy to follow, references to previous litterature is abundant, and the figures are of excellent quality. I would therefore recommend publication with only a few minor comments that may be addressed before.

Minor comments

• Convective waves: The paper mentions quite thoroughly the limitations of the satellite dataset, notably in
  
  terms of spectral characteristics of the waves that can be observed. I would appreciate though a further
  
  discussion regarding gravity waves generated by convective systems in the tropics. Like convection, the
  
  activity of those waves likely presents a strong diurnal cycle (cf. e.g., Corcos et al., 2021, reference
  
  already cited). I wonder how the HIRDLS/SABER observation characteristics (e.g., local time of passage) might 
  
  alter the retrieval of the intermittency. My impression is that undersampling the diurnal cycle would probably
  
  lower the observed intermittency... I also wonder whether this might be one possible reason for the higher
  
  intermittencies reported in SABER observations around lines 515 onward.
• PDF Normalization process: I agree with the justification of PDF normalization... but I am still unsure how
  
  the normalization is actually applied. Actually, this is explained in only one sentence (l 239-240), which is
  
  repeated in line 247-248. I would appreciate some further details. In particular, my current understanding is
  
  that, in every gegraphical box used to obtain gravity-wave momentum flux PDFs (and for every calendar month),
  
  one normalization factor is computed: this would explain that the authors refers to the "global distribution
  
  median values", which are used as normalization factors. What I feel confusing is the application
  
  of this process that is made in Figure 6 (namely Figure 6a => 6b). A natural choice (at least for me!) would
  
  have to use a single normalization factor per year for the whole 65S-50S region. But this seems at odds with
  
  several of your results/remarks, so I have inferred that this is not what was applied.

Technical remarks

- l 122: remove "for about 60 days"

- l 153: flying northward rather than "southward", right?

- l 184: discarded rather than "neglected"?

- Figures 17 and 18: I would recommend putting the highest altitudes at the top of the figures and the lowest

altitudes at the bottom.