the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 26 Feb 2025
Convection-generated gravity waves in the tropical lower stratosphere from Aeolus wind profiling, GNSS-RO and ERA5 reanalysis
Abstract. The European Space Agency's Aeolus satellite, equipped with the Atmospheric LAser Doppler INstrument (ALADIN), provides global provides near-global wind profiles from the surface to about 30 km altitude. These wind measurements enable the investigation of atmospheric dynamics, including gravity waves (GWs) in the upper troposphere and lower stratosphere (UTLS). This study analyzes ALADIN wind observations and ERA5 reanalysis, by deriving GWs kinetic energy (Ek) distributions, examining their temporal and spatial variability throughout the tropical UTLS. A prominent hotspot of enhanced GW activity is identified by Aeolus, migrating from the Indian Ocean in Boreal Summer to the Maritime Continent in Boreal Winter, closely matching outgoing longwave radiation minima and thus highlighting convective origins. Results show that ERA5 consistently underestimates Ek in convective regions, especially over the Indian Ocean, where conventional wind measurements are sparse. Additional comparisons with Global Navigation Satellite System Radio Occultation (GNSS-RO) measurements of GW potential energy (Ep) corroborate these findings and suggest significant underrepresentation of convection-driven wave activity in reanalyses. A multi-instrumental exploratory analysis also allows to verify the empirical grounding of the established Ek to Ep ratio, as well as the wavelength of the waves retrieved by Aeolus. By filling critical wind data gaps, Aeolus emerges as a key tool for improving the representation of GWs, particularly in remote tropical regions. When combined with GNSS-RO measurements, Aeolus data provides new insights into how convective processes drive GW generation, revealing opportunities to refine reanalysis products and model parameterizations, as well as improving the energy ratio.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(2638 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2638 KB) - Metadata XML
- BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-394', Anonymous Referee #1, 28 Feb 2025
- AC2: 'Reply on RC1', Mathieu Ratynski, 06 Jul 2025
-
CC1: 'Comment on egusphere-2025-394', Brian Mapes, 20 Mar 2025
The work described here flows from a unique and novel data source, a wonder of the age, and seems fairly sound. However, the figures are not all they could be, and not fully explained, lacking any statistical analysis or mentions of evident artifacts. Many (incomplete) data handling details and other science issues are sprinkled though a rather smooth vague narrative, with too little scientific circumspection. With a modest amount more care, perhaps aided by senior author input, this paper could become as excellent as the data deserve.The manuscript suffers from three problems common in the dissertation-to-literature translation, detailed further below but listed thematically here:1. Key technical details for understanding the results are too buried. In place of crisp exposition and taut circumspection are meandering vague mentions of issues around methods, defensive at times and elsewhere a sales pitch for the arbitrary trade-offs settled on. This may be how a committee explained a complex recipe to a student, but is not ideal for a paper addressing peer researchers.2. Meandering threads are also present as a means for a student to telegraph to a committee their awareness of a reading list, cited often in vague mentions rather than claim-supporting paraphrases of the content. This may be good for dissertations written to a captive readership, but is not ideal for peer researchers. As just one instance, not all winds on larger scales than the filter belong in a named list of Mode Types: the wind is the wind.3. Findings and interpretations are a bit vague, unobservant of the actual figure set, and repetitive (for instance the “especially Indian Ocean” trope appearing several times). False color scales are unhelpfully distorting, and questions raised by the figures and results are not pursued with a scientifically committed vigor. This may embody the rationally bounded commitment of a student to the physical problem at hand, but ideally a senior coauthor might exert the leadership to bring more depth of inquiry and perspective.Issue 1. Clarifying key technical details.Fig. 1 was very helpful but the text not so clear.SUMMARY:The variance of u and temperature T fluctuations, sub-weekly in period and Fourier bandpassed to 1-9km wavelengths in the vertical, are averaged over a UTLS layer (about 11-25 km after vertical smoothing). The 500m common grid is mentioned far from the other data details (line 124) requiring a second read and search. Half the variance (with rescaling factor for T') is energy. Was the averaging density-weighted (like a physical energy interpretation in J/kg should be), or is it just 1/2 a height-averaged variance? The text is silent.Lidar Wind profiles:A side-looking spaceborne lidar measures u(z) along its line of sight, which is almost the zonal direction (that wasn’t clear to this reader without web searches). Profiles were processed to isolate deviations from weekly 20x5-degree Lon-lat averages. Then Fourier filtering passed shorter than 9km vertical wavelengths (on a 500m grid so 1km is the shortest). The square of that filtered deviation profile was vertically smoothed with a 7km boxcar, then averaged over a layer (line 203 is ambiguous, why "49 points?” of 500m depth?) The layer is summed from 1km below the tropopause (about 14km in tropics) to 22km (is this mass weighted?), to make seasonal maps (smoothed how? not mentioned) and about 3-weekly (looks like, from figures? not mentioned) longitude-time sections. Artifical slow trends due to (squared) instrument noise variance increasing were estimated and subtracted.Identical processing was applied to RO T(z) and ERA5 winds and T (interpolated on the same grid, again please mention in the data section not introduction). This makes an excellent baseline of comparisons and opportunity for interpretation!How many pages did the above take to describe in the manuscript? Too many, a tedious read to fish out key details in order to bring a skepticism the somewhat slick text seemed to lack. Lines 172-174 are a good example of sales tone taking over: “making it possible”…”without introducing significant biases”…”configuration mitigates errors”…”ensuring reliable… and robust…” Declarations of success are not very scientific ways to express understandings of trade-offs. The real strength is *identical* processing of comparison datasets. Tossaway adverbs (e.g. “strongly”, “specifically") also set a slick tone in places, undercutting reader trust in the self-skepticism of those best positioned to see problems.Issue 2: Clarifying results2.1 Maps of KE and PE in the tropics and subtropics (Fig. 4) are clearly smoothed, but no mention is made of how and how much and why. Was raw data really too rough for scientific readers’ eyes, or is the paper trying to be too smooth? It is surprising to this reader how strongly confined to the equatorial belt the energy is even in solstice seasons. Is there some kind of conditioning or weighting behind this feature, or is it truly an aspect of convection as a source of signal (’the ITCZ’?)? Or is there an effect of the Coriolis force somehow suppressing sub-weekly sub-7km layer fluctuations? Silence about the smoothing undercuts reader confidence.2.2 Why no mention of the obvious artifacts in DJF2019-JJA2020, with zero in weak areas, or JJA2019 with 5+ in those areas? Are we all looking at the same figure here and describing its characteristics from the most obvious to the most subtle? Reader confidence is again at stake.2.3. The color scale for positive-definite variance, especially when discussed in a linear meaning like energy, should not have perceptual jumps like this one. Gray shading would be the honest choice, or a single color. This perceptual nonlinearity may be the source of the several-times-repeated “especially Indian Ocean” trope which otherwise seemed inscrutable to this viewer. Seychelles, Diego Garcia, Indonesia; are the other equatorial oceans really so much better covered with “conventional” wind soundings? And anyway, does ERA really get its UTLS wind variability from assimilating rawinsonde data into its imbalanced flow manifold? Did a desire to say something and move on override a thoughtful scientific assessment of the differences between reanalysis and observations, differences fluffed by the redness of a (not accessibility recommended) color map, and perhaps a misinterpretation of u fluctuations in the upper troposphere in the Maritime Continent wet season (detailed below)?2.4. Fig. 2 is about the anisotropy of waves in ERA5. Is this the second thing a reader wants in a paper about new observations? A value of 2 means isotropic waves, <2 suggests E-W elongation of the variance ellipse, >2 a N-S elongation. Here is a case where a color scale with a perceptual steep part could make sense, but 2 belongs there. Here a less meaningful choice was made. The focus is estimating something like the KE “missing” from the zonal (or LOS) component only, rather than the issue of isotropy which implies something interesting about sources. But all only in the reanalysis, before the first comparisons to even make that dataset as a relevant one. Might a better choice for a second figure be continuing raw obervations (Figs. 3-4) with discussions of the obvious artifacts? Let the ERA5 comparisons wait.2.5. Fig. 5 is a nice comparison, although again distorted by the use of false color for a linear positive quantity. Variance is a jumpy quantity from squaring the data (Fig. 1 shows this nicely) so statistical significance is tricky and surprising. It is customary to use a RATIO rather than a DIFFERENCE of variance (which has no meaning), subject to the F-test for significance. Many students are surprised by how hard it is to pass the F-test with quite a few degrees of freedom. The student should consult a table and appreciate the issue. No statistical testing is evident in the work, surprisingly. That can be overdone or a distraction, but none at all seems weak, again undermining reader confidence.2.6. Here is an actual scientific error I suspect: In the wet season over the Maritime Continent, convection is strong and localized and organized on island-strait and diurnal mesoscales that models struggle to represent. The intense divergent outflow of convection in the upper few km of the troposphere (squared) is not UTLS gravity wave energy! The 14-point smoother smears squared wind from a 4.5km layer of the upper troposphere into the averaging layer. Might the pat, recipe-like data analysis choices (Issue 1 above) and an insufficiently critical success-declaring mindset be combining here in a genuine misinterpretation? Sensitivity to this vertical smearing of non-wave wind variance should be assessed.2.6b Figure 7, again a ratio (F tested) would be more meaningful than a difference. The lack of RO signal in the Maritime Continent wet season further strengthens my belief that the Ek is a misinterpretation of convective outflows.2.7. The discussion of wave sources seems shallow. There are various mechanisms including temporally varying convective heat sources (which might set vertical wavenumber and frequency), mountains and transient mountains of lofted air in shear (which might set horizontal wavenumber and phase speed), and more. Would they be anisotropic? The phrase “trade winds” appears in the context of Fig. 2 (anisotropy), as if the surface wind direction has something to do with anisotropy at the tropopause level. Does it? Might one of the senior authors add a little depth?2.8 Likewise the meaning and source of the strong, anisotropic (more meridional) “waves” (isotropy ratio >2) on the midlatitude edges could be thought about more deeply. Are all subweekly meridional wind fluctuations (squared), from 4km below the tropopause, really UTLS gravity waves, or was that just a tidy story for students?2.9 Figure 8: at last a variance ratio, but only to be taken at face value (with no credible-interval estimation from the F test) in light of some vague gestures at theory whose linearity is considered an easy target. Line 478 says “Fig. 8 presents a detailed analysis” but literally it is just a data plot, with no analysis at all. Too much sales and not enough product for this reader.2.10 Figure 9: Panel a: Here might be the misinterpretation of convective outflow again, further exaggertated by the false color scheme. Panel b: what is a “dominant” wavelength? Anyone who looks at spectra knows that peak detection is far from trivial and every spectrum is always broad and usually red (more variance falls in eachoi of the wider bins at the low frequency end). What does geometric wavelength really signify over a layer whose stratification goes from upper tropospheric (almost neutral) to 22km (highly stratified)? How does the range here (6500-12000m) relate to the filter which supposedly excludes >9km? Is the spectrum basically red like all geophysical spectra, such that the widest bin near the longest permitted wavelength at the edge of the filter’s passband has the most variance? Does that deserve the word “dominant”? Is this figure worth including, or just a thesis figure looking for a place? Is the red exaggeration here the source of the “especially Indian Ocean” trope repeated several times? It's not exactly over the Indian ocean. The authorial prose should reflect a close look, as a reader brings.3. Discussion should be rewritten with care and thought, in light of all the above. A celebration of this amazing dataset, a technological marvel from such long hard efferts by so many, deserves more science value than a too-easy critique of reanalysis and/or underlying voids in data sources (common over all the equatorial oceans), and some vague words about how nature is not linear. Some senior author voice could help, if a bit of leadership can be mustered from a committee. Congratulations to so so many people contributing to make this possible! Wonderful data.Citation: https://doi.org/
10.5194/egusphere-2025-394-CC1 - AC1: 'Reply on CC1', Mathieu Ratynski, 06 Jul 2025
-
RC2: 'Comment on egusphere-2025-394', Anonymous Referee #2, 09 Apr 2025
The premise of this study is very promising and the results, if robust, are of high significance in comparing Aeolus, ERA5 and GNSS-RO gravity wave energy parameters in the tropical UTLS.
Still, the more I go through the manuscript I find too many details in the methodology unjustified, or their interpretation too stretched. Almost half of the text belongs to data and methods section which should be streamlined a lot. The text overall lacks an organized and concise structure, and some method details or datasets (e.g. NCEP reanalysis or the OLR datasets) seem to appear out of the blue.
I have several major comments about methodology that need to be clarified, because some of the results do not look very robust to me from the beginning, and this cascades then to the rest.
Figures could be improved a lot, and the authors should make a big effort in the text to avoid repetitive sentences, unnecessarily long explanations / verbose in methods or results (a lot of examples in minor/technical comments).
Also I feel that in many instances things are presented in a rather bombastic way, e.g. without really specifying where and how these valuable results have applications.
In section 5-6, some of the conclusions might change if some small tweaks in methodology were applied -- the authors make many choices and assumptions in the method -- and many grand statements with what comes out of it. Unless one shows very convincing and robust results (which would require a fair amount of supplement material), in the plots provided in this manuscript I see some inconsistencies that make me remain a bit skeptical.
To be clear, I'd very much like to see this study on such relevant topic published, and I hope the large amount of comments I assembled below are helpful for this. I recommend a major revision, and at least an additional round of reviews will be needed after that since the required changes are very substantial.
#
#
# Major comments
#
##
# 1: vertical grid and filtering choice
#-l.124-126: this grid penalizes ERA5 and RO a lot more than Aeolus, and 'acceptable middle ground' does not really justify your choice in my opinion. Is there any other literature doing this kind of middle-ground approach with other datasets?
-There are undesirable sources of uncertainty if you sub-sample or interpolate onto your 0.5km vertical grid: this might affect the resulting profile if a wave is not well aligned with your 0.5km vertical grid. Also ERA5, Aeolus and RO have each a very different vertical (original) grid alignment with your 0.5km grid.
In my opinion one should err on the side of caution and interpolate to a finer grid that retains all dataset's vertical structures as much as possible, and then filter out the scales that the coarser dataset cannot see, I explain below:-In section 2.2 you specify that you apply vertical high-pass filter to the data. Why not use a finer vertical grid of e.g. 0.1km, and apply bandpass between e.g. 1km and 9km? This way the uncertainty with sub-sampling is gone, and you remove the shortest vertical scales that Aeolus cannot see to even the field among all datasets. To me, this would be the fairest way to make the comparison by taking the vertical scales resolved by all datasets.
#
# 2: NCEP reanalysis and smoothing (l.198-204)
#A lot of things appearing out of the blue here without proper justification.
--> Is this the NCEP-DOE Reanalysis 2? It is not referenced either. How come this dataset is not mentioned in section 2.1?
--> Just because it's easier to integrate does not justify using it. I just don't understand why ERA5 is not used with its own tropopause.
--> Also, you don't show anywhere how similar are the results compared to ERA5. It certainly has poorer vertical resolution than ERA5, and this choice just adds an unnecessary layer of uncertainty. Not even some comparison material in a supplement?
"The profile is then smoothed using a 14-point moving average over the 49-point profile"
--> No justification given anywhere for this. Any other studies doing similar things that you could reference here?
--> You should explain what the purpose of this smoothing is. My impression is that it's not even necessary (see last part of my Major Comment #1 for a better option to compare what's resolved by all datasets).#
# 3: treatment of GNSS-RO data, details not properly justified
#l.206-211: details are very vague, e.g. which windowing is used for both in the end? Please state clearly what settings are applied to RO data and Aeolus.
l.213-214:
"Where the Brunt-Vaisala frequency squared (N2) is smoothed using binomial (Gaussian) smoothing of 10th order."
--> This is not justified, where is this coming from? Any reference for this?"Consequently, the data treatment across various instruments, whether wind or temperature remains consistent"
--> I strongly disagree!#
# 4: Most figures are low quality
#Fig.2:
--> please use the degree sign ° and not "DegN/E" (also present in Figs.4,5,6,7... all figures with lon or lat dimension...)
--> color scale is not the best for visibility, a color every 0.1 would improve guessing the exact value by eye.
--> It appears a bit pixelated if one zooms in just a bit.Fig.3
--> unreasonably large to show only four lines
--> way too many labels on the x-axisFig.5
--> label sizes too small
--> too many labels on y-axis
--> odd alignment of a)b)c) with subpanel titles
--> b) panel title size mismatch with the others, looks like put there by hand unlike a) and c)
#
# 5: pages 11-14
#-whole pages 11-12: this can be briefly summarized in the main manuscript and all the details moved to a supplement, including Fig.3
-The noise correction makes a quite long list of assumptions, could the authors provide some results/comparison of Aeolus results without noise correction for reference?
-Fig.4: the stark contrast of MAM 2019 and MAM 2020 does not give a reader a lot of confidence in your method. I am skeptical of how realistic the evolution of the left column is (the noise-corrected AEOLUS HLOS*).
-l.351-352: "The geographical distribution and evolution of energy hotspots are largely similar between the two datasets"
--> I disagree, the evolution of their strength, even in relative terms, seems quite different: e.g. compare the last 3-4 rows.
#
#
# Minor / technical comments
#
## Abstract
"revealing opportunities to refine reanalysis products and model parameterizations, as well as improving the energy ratio."
--> too vague, is there any specific recommendation here?
-l.2: cite ERA5 reference here-l.28-30: but the Podglajen study is from before ERA5 was around, please rephrase sentence for consistency.
-l.45-46: "short-wavelength waves are primarily lower frequency gravity waves, as dictated by the dispersion relation"
--> To avoid confusion please specify that it's short vertical wavelength, and give a ballpark number of the range of vert. wavelengths you are referring to.
--> Also in the next sentence, specify what vertical wavelenghts can be captured by Aeolus.-l.59-61: calling it "climatology" from 3 years sounds a bit stretched...
--> perhaps simply state this as an observational estimate for Jun.2019-Aug.2022
--> also this is an example of a repetitive sentence. E.g. 'and its link with deep convection' could be removed without any loss of information-l.70: I would support sub-subsections for each separate dataset and methods.
-l.72: range bin settings and other specifications should have an earlier reference.
--> Also please update the Rennie and Isaksen 2020 reference to the 2024 ESA contract report (which includes all information from the 2020 TM). Check throughout the manuscript.
--> https://www.ecmwf.int/en/elibrary/81546-nwp-impact-aeolus-level-2b-winds-ecmwf-l.87-88: please confirm whether you got the data on that native resolution?
-l.94: best candidate (by far in my opinion), especially when compared to other reanalysis products.
-l.95: "standard" --> you mean the 137 hybrid levels? Standard is usually associated with the 37 standard pressure levels, I recommend not using this term here to avoid confusion.
-l.99: GNSS-RO datasets --> please list which missions are included + their references, and I presume COSMIC-2 dominates the overall data amount? If so, mentioning a bit
-l.111-113: perhaps merge with l.97-98 at the beginning of the paragraph, otherwise to me feels a bit repetitive.
-l.103-104: defined by bending angle gradient, which increases near inversion layers / humidity gradients
--> I recommend to refer to Kursinski et al. 1997 here --> https://doi.org/10.1029/97JD01569-l.115-118: feels very repetitive and could be streamlined
-l.130-131: overselling and too vague, remove or specify recommendations to enhance reanalyses and models from the results of your study.
-l.131-133: just say they are independent datasets, this sentence can be streamlined and toned down.
############### 2.2 Methods and limitations-l.139-150: regarding the trickiness of background state removal, I miss a discussion about research that used GNSS-RO and Aeolus to study Kelvin waves, their vertical scales and (in the case of Randel et al., 2021) the behavior of the small-scale residual.
These references are very relevant to your study's methodology, the more so since they use the same datasets as you.--> Randel and Wu (2005) --> https://doi.org/10.1029/2004JD005006 (using GPS-RO)
"Vertical wavelengths of ∼6–8 km are observed near and above the tropopause in December 2001 to January 2002 (Figures 6a and 6b), while shorter vertical wavelengths (∼4–5 km) are observed in May and August–September 2002 (Figures 6c and 6d). "--> Randel et al. (2021) --> https://doi.org/10.1029/2020JD033969 (using COSMIC-2)
"strong residual variance occurs in the longitudinal shear zones of Kelvin waves" and this small-scale residual T variance is associated with GWs.--> Zagar et al. (2021) --> https://doi.org/10.1029/2021GL094716 --> "Aeolus assimilation modifies the representation of vertically propagating Kelvin waves in the tropical UTLS" (Aeolus)
################
-l.160-161: "Aeolus now provides the necessary tools to apply the same approach for GW Ek."
--> Sorry to be picky here, but what tools does Aeolus bring now that it didn't before. You use the same approach (your tool) to calculate GW Ek from Aeolus (data, not a tool). Such phrasing is just unnecessary verbose.- Fig.1: please include the U(z) notation in the labels, and specify which datasets you take U from.
-l.216-219: you should state all this when introducing Ek{hlos}.
--> Fig.2 belongs in a supplement
--> And wouldn't it be fairer to compare EK from ERA5 U to EK_HLOS??-l.228-231: belongs also in a supplement in my opinion
-l.245-249: a lot of verbose here, show the figure in a supplement and move the text there
-l.343-344: another example of verbose.
Citation: https://doi.org/10.5194/egusphere-2025-394-RC2 - AC3: 'Reply on RC2', Mathieu Ratynski, 06 Jul 2025
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2025-394', Anonymous Referee #1, 28 Feb 2025
The manuscript presents an analysis of gravity wave (GW) kinetic energy distributions, derived from new Aeolus satellite wind profiles, that shows great promise in pushing the needle forward in the construction of observational constraints of gravity waves and their impacts on upper troposphere/lower stratosphere circulation. A methodology is presented for deriving the kinetic energy associated with small-scale GWs in regions of deep convection in the tropics over a period spanning June 2019 to August 2022. Comparisons with ERA5 suggest that the reanalysis product underestimates GW-associated kinetic energy; conversely, GW-associated potential energy comparisons between ERA5 and temperature-profiles from an independent instrument (GNSS-RO) show much more consistency, suggesting that the use of kinetic energy highlights a distinct feature of the GW energy spectrum that is not typically assessed (and, incidentally, is not well represented in ERA5). The authors further speculate that this underestimate may reflect lack of assimilated direct wind observations, in contrast to temperatures, which are assimilated. All in all, the manuscript does a good job of presenting a new dataset with all necessary caveats, while also making a generally convincing case that this new data will be valuable. To this end, I recommend acceptance, pending that minor revisions be made to address the following concerns:
#1. Page 4: There is no description of the GW drag parameterization employed in ERA5. In particular, does the model have an explicit parameterization for non-orographic GW drag due to parameterized convection? If so, what is it and how has it been evaluated/performed in past assessments? This will be important in terms of interpreting the dearth of kinetic energy in the model, relative to the Aeolus-derived energy.
#2. Page 7: Presumably the definition of "background" based on "the arguments presented in Alexander et al. (2008b)" apply to past analysis of temperature, not wind, profiles, no? More generally, it would be good for the reader to have a better sense of the sensitivity of the profiles depicted in Figure 1a to choice of grid box averaging domain, the temporal period over which profiles are averaged (currently set to 7 days, etc.), etc. I imagine the authors have already done this sensitivity analysis, so they could consider showing in an appendix figure.
#3. Equation (1): This notation becomes slightly confusing/counterintuitive as the text moves on, since the meridional component often goes to zero due to the pointing vector retaining its approximate angle at ~100 degrees. In other words, V_HLOS would be more intuitively referred to as U_HLOS (or something similar) since, indeed, it primarily reflects the zonal component of the flow. Is there any particular reason why "v" is used instead of something more generic? I suggest changing.
#4. Figure 14, lines 354-355: The first sentence of this paragraph does not make sense to me. In particular, the bit referring to "ERA5 shows a considerable reduction" is vague. Reduction relative to what? Please clarify.
#5. Figure 5: The temporal resolution labeled on the y-axes of these hovmoller plots is too high/unnecessary as it crowds the figures. Please show only every other two or three months. Same comment applies to Figure 7.
#6. Figure 16, Discussion concluding Section 3.2: The discussion here seems weak and understates the disagreement between the Aeolus and ERA5 Ek temporal patterns. The second-to-last paragraph highlights the common features between Aeolus and ERA5, but I think the plots look very different. In particular, the hotspots coincident with low OLR are totally missing in ERA5 (Fig. 5b). The phrasing in the text, however, seems to suggest that the differences are only minor. Please rephrase.
#7. Section 4: Doesn't the ratio of Ek/Ep (shown for ERA-5 in Fig. 8a) suggest that these two quantities are extremely different and not meaningful to compare with each other? I appreciate that the authors want to move beyond traditional (conservative) analysis and attempt to do a bit more, but Figure 8a suggests that the two quantities are in much more disagreement than the discrepancy predicted by llinear wave theory (i.e., factor of 4, not factor of 2). My suggestion here is to introduce Figure 8a earlier as a way to more directly address the concerns with comparing potential and kinetic energy (within a self-consistent product like ERA-5).
#8. Last paragraph on page 19 (lines 467-470): How do you know it's the failure to assimilate the winds directly that's causing the poor representation of GW-associated EK? In principle, one might be able to capture these features using a convective non-orgraphic gravity wave drag parameterization within the ERA-5 model, no? In other words, the assimilation is one way to correct the problem, but an alternative approach is to tackle the model bias directly. However, without having more knowledge about the underlying GW drag parameterization in the model it's hard for the reader to know how many degrees of freedom are afforded to the modeler. Can the authors please comment on the role played here by model bias? And how this is/is not handled by the GW drag parameterization?
#9. Discussion: No mention is made of how these observations might be used to develop constraints on the momentum fluxes (which is what modelers seek most). Is that something that the author has considered? This is a challenging question, so I am not seeking any complete answers here; I am just wondering if the author can speculate in a sentence or two how to potentially bridge V_HLOS with the momentum fluxes.
Citation: https://doi.org/10.5194/egusphere-2025-394-RC1 - AC2: 'Reply on RC1', Mathieu Ratynski, 06 Jul 2025
-
CC1: 'Comment on egusphere-2025-394', Brian Mapes, 20 Mar 2025
The work described here flows from a unique and novel data source, a wonder of the age, and seems fairly sound. However, the figures are not all they could be, and not fully explained, lacking any statistical analysis or mentions of evident artifacts. Many (incomplete) data handling details and other science issues are sprinkled though a rather smooth vague narrative, with too little scientific circumspection. With a modest amount more care, perhaps aided by senior author input, this paper could become as excellent as the data deserve.The manuscript suffers from three problems common in the dissertation-to-literature translation, detailed further below but listed thematically here:1. Key technical details for understanding the results are too buried. In place of crisp exposition and taut circumspection are meandering vague mentions of issues around methods, defensive at times and elsewhere a sales pitch for the arbitrary trade-offs settled on. This may be how a committee explained a complex recipe to a student, but is not ideal for a paper addressing peer researchers.2. Meandering threads are also present as a means for a student to telegraph to a committee their awareness of a reading list, cited often in vague mentions rather than claim-supporting paraphrases of the content. This may be good for dissertations written to a captive readership, but is not ideal for peer researchers. As just one instance, not all winds on larger scales than the filter belong in a named list of Mode Types: the wind is the wind.3. Findings and interpretations are a bit vague, unobservant of the actual figure set, and repetitive (for instance the “especially Indian Ocean” trope appearing several times). False color scales are unhelpfully distorting, and questions raised by the figures and results are not pursued with a scientifically committed vigor. This may embody the rationally bounded commitment of a student to the physical problem at hand, but ideally a senior coauthor might exert the leadership to bring more depth of inquiry and perspective.Issue 1. Clarifying key technical details.Fig. 1 was very helpful but the text not so clear.SUMMARY:The variance of u and temperature T fluctuations, sub-weekly in period and Fourier bandpassed to 1-9km wavelengths in the vertical, are averaged over a UTLS layer (about 11-25 km after vertical smoothing). The 500m common grid is mentioned far from the other data details (line 124) requiring a second read and search. Half the variance (with rescaling factor for T') is energy. Was the averaging density-weighted (like a physical energy interpretation in J/kg should be), or is it just 1/2 a height-averaged variance? The text is silent.Lidar Wind profiles:A side-looking spaceborne lidar measures u(z) along its line of sight, which is almost the zonal direction (that wasn’t clear to this reader without web searches). Profiles were processed to isolate deviations from weekly 20x5-degree Lon-lat averages. Then Fourier filtering passed shorter than 9km vertical wavelengths (on a 500m grid so 1km is the shortest). The square of that filtered deviation profile was vertically smoothed with a 7km boxcar, then averaged over a layer (line 203 is ambiguous, why "49 points?” of 500m depth?) The layer is summed from 1km below the tropopause (about 14km in tropics) to 22km (is this mass weighted?), to make seasonal maps (smoothed how? not mentioned) and about 3-weekly (looks like, from figures? not mentioned) longitude-time sections. Artifical slow trends due to (squared) instrument noise variance increasing were estimated and subtracted.Identical processing was applied to RO T(z) and ERA5 winds and T (interpolated on the same grid, again please mention in the data section not introduction). This makes an excellent baseline of comparisons and opportunity for interpretation!How many pages did the above take to describe in the manuscript? Too many, a tedious read to fish out key details in order to bring a skepticism the somewhat slick text seemed to lack. Lines 172-174 are a good example of sales tone taking over: “making it possible”…”without introducing significant biases”…”configuration mitigates errors”…”ensuring reliable… and robust…” Declarations of success are not very scientific ways to express understandings of trade-offs. The real strength is *identical* processing of comparison datasets. Tossaway adverbs (e.g. “strongly”, “specifically") also set a slick tone in places, undercutting reader trust in the self-skepticism of those best positioned to see problems.Issue 2: Clarifying results2.1 Maps of KE and PE in the tropics and subtropics (Fig. 4) are clearly smoothed, but no mention is made of how and how much and why. Was raw data really too rough for scientific readers’ eyes, or is the paper trying to be too smooth? It is surprising to this reader how strongly confined to the equatorial belt the energy is even in solstice seasons. Is there some kind of conditioning or weighting behind this feature, or is it truly an aspect of convection as a source of signal (’the ITCZ’?)? Or is there an effect of the Coriolis force somehow suppressing sub-weekly sub-7km layer fluctuations? Silence about the smoothing undercuts reader confidence.2.2 Why no mention of the obvious artifacts in DJF2019-JJA2020, with zero in weak areas, or JJA2019 with 5+ in those areas? Are we all looking at the same figure here and describing its characteristics from the most obvious to the most subtle? Reader confidence is again at stake.2.3. The color scale for positive-definite variance, especially when discussed in a linear meaning like energy, should not have perceptual jumps like this one. Gray shading would be the honest choice, or a single color. This perceptual nonlinearity may be the source of the several-times-repeated “especially Indian Ocean” trope which otherwise seemed inscrutable to this viewer. Seychelles, Diego Garcia, Indonesia; are the other equatorial oceans really so much better covered with “conventional” wind soundings? And anyway, does ERA really get its UTLS wind variability from assimilating rawinsonde data into its imbalanced flow manifold? Did a desire to say something and move on override a thoughtful scientific assessment of the differences between reanalysis and observations, differences fluffed by the redness of a (not accessibility recommended) color map, and perhaps a misinterpretation of u fluctuations in the upper troposphere in the Maritime Continent wet season (detailed below)?2.4. Fig. 2 is about the anisotropy of waves in ERA5. Is this the second thing a reader wants in a paper about new observations? A value of 2 means isotropic waves, <2 suggests E-W elongation of the variance ellipse, >2 a N-S elongation. Here is a case where a color scale with a perceptual steep part could make sense, but 2 belongs there. Here a less meaningful choice was made. The focus is estimating something like the KE “missing” from the zonal (or LOS) component only, rather than the issue of isotropy which implies something interesting about sources. But all only in the reanalysis, before the first comparisons to even make that dataset as a relevant one. Might a better choice for a second figure be continuing raw obervations (Figs. 3-4) with discussions of the obvious artifacts? Let the ERA5 comparisons wait.2.5. Fig. 5 is a nice comparison, although again distorted by the use of false color for a linear positive quantity. Variance is a jumpy quantity from squaring the data (Fig. 1 shows this nicely) so statistical significance is tricky and surprising. It is customary to use a RATIO rather than a DIFFERENCE of variance (which has no meaning), subject to the F-test for significance. Many students are surprised by how hard it is to pass the F-test with quite a few degrees of freedom. The student should consult a table and appreciate the issue. No statistical testing is evident in the work, surprisingly. That can be overdone or a distraction, but none at all seems weak, again undermining reader confidence.2.6. Here is an actual scientific error I suspect: In the wet season over the Maritime Continent, convection is strong and localized and organized on island-strait and diurnal mesoscales that models struggle to represent. The intense divergent outflow of convection in the upper few km of the troposphere (squared) is not UTLS gravity wave energy! The 14-point smoother smears squared wind from a 4.5km layer of the upper troposphere into the averaging layer. Might the pat, recipe-like data analysis choices (Issue 1 above) and an insufficiently critical success-declaring mindset be combining here in a genuine misinterpretation? Sensitivity to this vertical smearing of non-wave wind variance should be assessed.2.6b Figure 7, again a ratio (F tested) would be more meaningful than a difference. The lack of RO signal in the Maritime Continent wet season further strengthens my belief that the Ek is a misinterpretation of convective outflows.2.7. The discussion of wave sources seems shallow. There are various mechanisms including temporally varying convective heat sources (which might set vertical wavenumber and frequency), mountains and transient mountains of lofted air in shear (which might set horizontal wavenumber and phase speed), and more. Would they be anisotropic? The phrase “trade winds” appears in the context of Fig. 2 (anisotropy), as if the surface wind direction has something to do with anisotropy at the tropopause level. Does it? Might one of the senior authors add a little depth?2.8 Likewise the meaning and source of the strong, anisotropic (more meridional) “waves” (isotropy ratio >2) on the midlatitude edges could be thought about more deeply. Are all subweekly meridional wind fluctuations (squared), from 4km below the tropopause, really UTLS gravity waves, or was that just a tidy story for students?2.9 Figure 8: at last a variance ratio, but only to be taken at face value (with no credible-interval estimation from the F test) in light of some vague gestures at theory whose linearity is considered an easy target. Line 478 says “Fig. 8 presents a detailed analysis” but literally it is just a data plot, with no analysis at all. Too much sales and not enough product for this reader.2.10 Figure 9: Panel a: Here might be the misinterpretation of convective outflow again, further exaggertated by the false color scheme. Panel b: what is a “dominant” wavelength? Anyone who looks at spectra knows that peak detection is far from trivial and every spectrum is always broad and usually red (more variance falls in eachoi of the wider bins at the low frequency end). What does geometric wavelength really signify over a layer whose stratification goes from upper tropospheric (almost neutral) to 22km (highly stratified)? How does the range here (6500-12000m) relate to the filter which supposedly excludes >9km? Is the spectrum basically red like all geophysical spectra, such that the widest bin near the longest permitted wavelength at the edge of the filter’s passband has the most variance? Does that deserve the word “dominant”? Is this figure worth including, or just a thesis figure looking for a place? Is the red exaggeration here the source of the “especially Indian Ocean” trope repeated several times? It's not exactly over the Indian ocean. The authorial prose should reflect a close look, as a reader brings.3. Discussion should be rewritten with care and thought, in light of all the above. A celebration of this amazing dataset, a technological marvel from such long hard efferts by so many, deserves more science value than a too-easy critique of reanalysis and/or underlying voids in data sources (common over all the equatorial oceans), and some vague words about how nature is not linear. Some senior author voice could help, if a bit of leadership can be mustered from a committee. Congratulations to so so many people contributing to make this possible! Wonderful data.Citation: https://doi.org/
10.5194/egusphere-2025-394-CC1 - AC1: 'Reply on CC1', Mathieu Ratynski, 06 Jul 2025
-
RC2: 'Comment on egusphere-2025-394', Anonymous Referee #2, 09 Apr 2025
The premise of this study is very promising and the results, if robust, are of high significance in comparing Aeolus, ERA5 and GNSS-RO gravity wave energy parameters in the tropical UTLS.
Still, the more I go through the manuscript I find too many details in the methodology unjustified, or their interpretation too stretched. Almost half of the text belongs to data and methods section which should be streamlined a lot. The text overall lacks an organized and concise structure, and some method details or datasets (e.g. NCEP reanalysis or the OLR datasets) seem to appear out of the blue.
I have several major comments about methodology that need to be clarified, because some of the results do not look very robust to me from the beginning, and this cascades then to the rest.
Figures could be improved a lot, and the authors should make a big effort in the text to avoid repetitive sentences, unnecessarily long explanations / verbose in methods or results (a lot of examples in minor/technical comments).
Also I feel that in many instances things are presented in a rather bombastic way, e.g. without really specifying where and how these valuable results have applications.
In section 5-6, some of the conclusions might change if some small tweaks in methodology were applied -- the authors make many choices and assumptions in the method -- and many grand statements with what comes out of it. Unless one shows very convincing and robust results (which would require a fair amount of supplement material), in the plots provided in this manuscript I see some inconsistencies that make me remain a bit skeptical.
To be clear, I'd very much like to see this study on such relevant topic published, and I hope the large amount of comments I assembled below are helpful for this. I recommend a major revision, and at least an additional round of reviews will be needed after that since the required changes are very substantial.
#
#
# Major comments
#
##
# 1: vertical grid and filtering choice
#-l.124-126: this grid penalizes ERA5 and RO a lot more than Aeolus, and 'acceptable middle ground' does not really justify your choice in my opinion. Is there any other literature doing this kind of middle-ground approach with other datasets?
-There are undesirable sources of uncertainty if you sub-sample or interpolate onto your 0.5km vertical grid: this might affect the resulting profile if a wave is not well aligned with your 0.5km vertical grid. Also ERA5, Aeolus and RO have each a very different vertical (original) grid alignment with your 0.5km grid.
In my opinion one should err on the side of caution and interpolate to a finer grid that retains all dataset's vertical structures as much as possible, and then filter out the scales that the coarser dataset cannot see, I explain below:-In section 2.2 you specify that you apply vertical high-pass filter to the data. Why not use a finer vertical grid of e.g. 0.1km, and apply bandpass between e.g. 1km and 9km? This way the uncertainty with sub-sampling is gone, and you remove the shortest vertical scales that Aeolus cannot see to even the field among all datasets. To me, this would be the fairest way to make the comparison by taking the vertical scales resolved by all datasets.
#
# 2: NCEP reanalysis and smoothing (l.198-204)
#A lot of things appearing out of the blue here without proper justification.
--> Is this the NCEP-DOE Reanalysis 2? It is not referenced either. How come this dataset is not mentioned in section 2.1?
--> Just because it's easier to integrate does not justify using it. I just don't understand why ERA5 is not used with its own tropopause.
--> Also, you don't show anywhere how similar are the results compared to ERA5. It certainly has poorer vertical resolution than ERA5, and this choice just adds an unnecessary layer of uncertainty. Not even some comparison material in a supplement?
"The profile is then smoothed using a 14-point moving average over the 49-point profile"
--> No justification given anywhere for this. Any other studies doing similar things that you could reference here?
--> You should explain what the purpose of this smoothing is. My impression is that it's not even necessary (see last part of my Major Comment #1 for a better option to compare what's resolved by all datasets).#
# 3: treatment of GNSS-RO data, details not properly justified
#l.206-211: details are very vague, e.g. which windowing is used for both in the end? Please state clearly what settings are applied to RO data and Aeolus.
l.213-214:
"Where the Brunt-Vaisala frequency squared (N2) is smoothed using binomial (Gaussian) smoothing of 10th order."
--> This is not justified, where is this coming from? Any reference for this?"Consequently, the data treatment across various instruments, whether wind or temperature remains consistent"
--> I strongly disagree!#
# 4: Most figures are low quality
#Fig.2:
--> please use the degree sign ° and not "DegN/E" (also present in Figs.4,5,6,7... all figures with lon or lat dimension...)
--> color scale is not the best for visibility, a color every 0.1 would improve guessing the exact value by eye.
--> It appears a bit pixelated if one zooms in just a bit.Fig.3
--> unreasonably large to show only four lines
--> way too many labels on the x-axisFig.5
--> label sizes too small
--> too many labels on y-axis
--> odd alignment of a)b)c) with subpanel titles
--> b) panel title size mismatch with the others, looks like put there by hand unlike a) and c)
#
# 5: pages 11-14
#-whole pages 11-12: this can be briefly summarized in the main manuscript and all the details moved to a supplement, including Fig.3
-The noise correction makes a quite long list of assumptions, could the authors provide some results/comparison of Aeolus results without noise correction for reference?
-Fig.4: the stark contrast of MAM 2019 and MAM 2020 does not give a reader a lot of confidence in your method. I am skeptical of how realistic the evolution of the left column is (the noise-corrected AEOLUS HLOS*).
-l.351-352: "The geographical distribution and evolution of energy hotspots are largely similar between the two datasets"
--> I disagree, the evolution of their strength, even in relative terms, seems quite different: e.g. compare the last 3-4 rows.
#
#
# Minor / technical comments
#
## Abstract
"revealing opportunities to refine reanalysis products and model parameterizations, as well as improving the energy ratio."
--> too vague, is there any specific recommendation here?
-l.2: cite ERA5 reference here-l.28-30: but the Podglajen study is from before ERA5 was around, please rephrase sentence for consistency.
-l.45-46: "short-wavelength waves are primarily lower frequency gravity waves, as dictated by the dispersion relation"
--> To avoid confusion please specify that it's short vertical wavelength, and give a ballpark number of the range of vert. wavelengths you are referring to.
--> Also in the next sentence, specify what vertical wavelenghts can be captured by Aeolus.-l.59-61: calling it "climatology" from 3 years sounds a bit stretched...
--> perhaps simply state this as an observational estimate for Jun.2019-Aug.2022
--> also this is an example of a repetitive sentence. E.g. 'and its link with deep convection' could be removed without any loss of information-l.70: I would support sub-subsections for each separate dataset and methods.
-l.72: range bin settings and other specifications should have an earlier reference.
--> Also please update the Rennie and Isaksen 2020 reference to the 2024 ESA contract report (which includes all information from the 2020 TM). Check throughout the manuscript.
--> https://www.ecmwf.int/en/elibrary/81546-nwp-impact-aeolus-level-2b-winds-ecmwf-l.87-88: please confirm whether you got the data on that native resolution?
-l.94: best candidate (by far in my opinion), especially when compared to other reanalysis products.
-l.95: "standard" --> you mean the 137 hybrid levels? Standard is usually associated with the 37 standard pressure levels, I recommend not using this term here to avoid confusion.
-l.99: GNSS-RO datasets --> please list which missions are included + their references, and I presume COSMIC-2 dominates the overall data amount? If so, mentioning a bit
-l.111-113: perhaps merge with l.97-98 at the beginning of the paragraph, otherwise to me feels a bit repetitive.
-l.103-104: defined by bending angle gradient, which increases near inversion layers / humidity gradients
--> I recommend to refer to Kursinski et al. 1997 here --> https://doi.org/10.1029/97JD01569-l.115-118: feels very repetitive and could be streamlined
-l.130-131: overselling and too vague, remove or specify recommendations to enhance reanalyses and models from the results of your study.
-l.131-133: just say they are independent datasets, this sentence can be streamlined and toned down.
############### 2.2 Methods and limitations-l.139-150: regarding the trickiness of background state removal, I miss a discussion about research that used GNSS-RO and Aeolus to study Kelvin waves, their vertical scales and (in the case of Randel et al., 2021) the behavior of the small-scale residual.
These references are very relevant to your study's methodology, the more so since they use the same datasets as you.--> Randel and Wu (2005) --> https://doi.org/10.1029/2004JD005006 (using GPS-RO)
"Vertical wavelengths of ∼6–8 km are observed near and above the tropopause in December 2001 to January 2002 (Figures 6a and 6b), while shorter vertical wavelengths (∼4–5 km) are observed in May and August–September 2002 (Figures 6c and 6d). "--> Randel et al. (2021) --> https://doi.org/10.1029/2020JD033969 (using COSMIC-2)
"strong residual variance occurs in the longitudinal shear zones of Kelvin waves" and this small-scale residual T variance is associated with GWs.--> Zagar et al. (2021) --> https://doi.org/10.1029/2021GL094716 --> "Aeolus assimilation modifies the representation of vertically propagating Kelvin waves in the tropical UTLS" (Aeolus)
################
-l.160-161: "Aeolus now provides the necessary tools to apply the same approach for GW Ek."
--> Sorry to be picky here, but what tools does Aeolus bring now that it didn't before. You use the same approach (your tool) to calculate GW Ek from Aeolus (data, not a tool). Such phrasing is just unnecessary verbose.- Fig.1: please include the U(z) notation in the labels, and specify which datasets you take U from.
-l.216-219: you should state all this when introducing Ek{hlos}.
--> Fig.2 belongs in a supplement
--> And wouldn't it be fairer to compare EK from ERA5 U to EK_HLOS??-l.228-231: belongs also in a supplement in my opinion
-l.245-249: a lot of verbose here, show the figure in a supplement and move the text there
-l.343-344: another example of verbose.
Citation: https://doi.org/10.5194/egusphere-2025-394-RC2 - AC3: 'Reply on RC2', Mathieu Ratynski, 06 Jul 2025
Peer review completion
Journal article(s) based on this preprint
Data sets
Pertubation Profiles Dataset Mathieu Ratynski https://doi.org/10.5281/zenodo.8113261
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 787 | 66 | 28 | 881 | 26 | 45 |
- HTML: 787
- PDF: 66
- XML: 28
- Total: 881
- BibTeX: 26
- EndNote: 45
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Mathieu Ratynski
Sergey Khaykin
Alain Hauchecorne
Joan Alexander
Alexis Mariaccia
Philippe Keckhut
Antoine Mangin
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2638 KB) - Metadata XML
The manuscript presents an analysis of gravity wave (GW) kinetic energy distributions, derived from new Aeolus satellite wind profiles, that shows great promise in pushing the needle forward in the construction of observational constraints of gravity waves and their impacts on upper troposphere/lower stratosphere circulation. A methodology is presented for deriving the kinetic energy associated with small-scale GWs in regions of deep convection in the tropics over a period spanning June 2019 to August 2022. Comparisons with ERA5 suggest that the reanalysis product underestimates GW-associated kinetic energy; conversely, GW-associated potential energy comparisons between ERA5 and temperature-profiles from an independent instrument (GNSS-RO) show much more consistency, suggesting that the use of kinetic energy highlights a distinct feature of the GW energy spectrum that is not typically assessed (and, incidentally, is not well represented in ERA5). The authors further speculate that this underestimate may reflect lack of assimilated direct wind observations, in contrast to temperatures, which are assimilated. All in all, the manuscript does a good job of presenting a new dataset with all necessary caveats, while also making a generally convincing case that this new data will be valuable. To this end, I recommend acceptance, pending that minor revisions be made to address the following concerns:
#1. Page 4: There is no description of the GW drag parameterization employed in ERA5. In particular, does the model have an explicit parameterization for non-orographic GW drag due to parameterized convection? If so, what is it and how has it been evaluated/performed in past assessments? This will be important in terms of interpreting the dearth of kinetic energy in the model, relative to the Aeolus-derived energy.
#2. Page 7: Presumably the definition of "background" based on "the arguments presented in Alexander et al. (2008b)" apply to past analysis of temperature, not wind, profiles, no? More generally, it would be good for the reader to have a better sense of the sensitivity of the profiles depicted in Figure 1a to choice of grid box averaging domain, the temporal period over which profiles are averaged (currently set to 7 days, etc.), etc. I imagine the authors have already done this sensitivity analysis, so they could consider showing in an appendix figure.
#3. Equation (1): This notation becomes slightly confusing/counterintuitive as the text moves on, since the meridional component often goes to zero due to the pointing vector retaining its approximate angle at ~100 degrees. In other words, V_HLOS would be more intuitively referred to as U_HLOS (or something similar) since, indeed, it primarily reflects the zonal component of the flow. Is there any particular reason why "v" is used instead of something more generic? I suggest changing.
#4. Figure 14, lines 354-355: The first sentence of this paragraph does not make sense to me. In particular, the bit referring to "ERA5 shows a considerable reduction" is vague. Reduction relative to what? Please clarify.
#5. Figure 5: The temporal resolution labeled on the y-axes of these hovmoller plots is too high/unnecessary as it crowds the figures. Please show only every other two or three months. Same comment applies to Figure 7.
#6. Figure 16, Discussion concluding Section 3.2: The discussion here seems weak and understates the disagreement between the Aeolus and ERA5 Ek temporal patterns. The second-to-last paragraph highlights the common features between Aeolus and ERA5, but I think the plots look very different. In particular, the hotspots coincident with low OLR are totally missing in ERA5 (Fig. 5b). The phrasing in the text, however, seems to suggest that the differences are only minor. Please rephrase.
#7. Section 4: Doesn't the ratio of Ek/Ep (shown for ERA-5 in Fig. 8a) suggest that these two quantities are extremely different and not meaningful to compare with each other? I appreciate that the authors want to move beyond traditional (conservative) analysis and attempt to do a bit more, but Figure 8a suggests that the two quantities are in much more disagreement than the discrepancy predicted by llinear wave theory (i.e., factor of 4, not factor of 2). My suggestion here is to introduce Figure 8a earlier as a way to more directly address the concerns with comparing potential and kinetic energy (within a self-consistent product like ERA-5).
#8. Last paragraph on page 19 (lines 467-470): How do you know it's the failure to assimilate the winds directly that's causing the poor representation of GW-associated EK? In principle, one might be able to capture these features using a convective non-orgraphic gravity wave drag parameterization within the ERA-5 model, no? In other words, the assimilation is one way to correct the problem, but an alternative approach is to tackle the model bias directly. However, without having more knowledge about the underlying GW drag parameterization in the model it's hard for the reader to know how many degrees of freedom are afforded to the modeler. Can the authors please comment on the role played here by model bias? And how this is/is not handled by the GW drag parameterization?
#9. Discussion: No mention is made of how these observations might be used to develop constraints on the momentum fluxes (which is what modelers seek most). Is that something that the author has considered? This is a challenging question, so I am not seeking any complete answers here; I am just wondering if the author can speculate in a sentence or two how to potentially bridge V_HLOS with the momentum fluxes.