This work by Garcia-Franco et al. looks at the relationships between 

the QBO and tropical climate in observations and centennial pre-industrial 

CMIP6 simulations with one coupled climate model.

The connections are difficult to diagnose from observations so long simulations are useful.

The paper is overall interesting and covers many topics, but the authors

should check the consistency of the symbols and names used (see comments by line).

It can be confusing to read different acronyms for the same quantities.

The units reported in the plots should be verified.

Given the central role of model simulations, more information on its

skill at simulating QBO and ENSO should be provided. For example, how realistic 

is the QBO amplitude at 70 hPa for this specific model?

Apart from composite differences, some climatologies should be discussed.

In the introduction reference to Geller et al., 2016 on gravity wave changes would fit.

Model-dependence of the results should be stressed, since different 

configurations of a single model are analysed and QBO/SST biases may play a big role.

The causality analysis on how the QBO influences ENSO is not very convincing as it stands. 

I guess the authors should also say something about the frequency of LN/EN

events during neutral QBO (QBO-N).

The section about monsoons should be revised and maybe shortened, since QBO 

surface impacts may be very dependent of any QBO bias. For example, Giorgetta 

et al. 1999 (cited) nudged to QBO, so it was realistic in their case.

The data description should be modified to provide pertinent information.

Specific comments by line

L52, maybe 'on the convective process'?

L55, define 'CMIP', rephrasing L62

L63, are GWs tied somehow to sources?

L76, both monthly means?

L82, is there a reason for not using the standard 0.25x0.25?

L83, it is a bit strange to put the (generic) link only for ERA5; 

I would move to data availability with direct links for all datasets

(for ERA5 https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-pressure-levels-monthly-means?tab=overview) and proper citation https://confluence.ecmwf.int/pages/viewpage.action?pageId=197704114

L90seq, define 'N' and 'ORCA' for the components resolution

L91, UKESM or UKESM1?

L94, So 3 simulations in total? Would be good to state that you have two models

with lower resolution and one with better resolution, which is the main interest. 

L96, I would move to 'data availability' or similar

L98, Here or later I would add something about some relevant model properties (e.g.

both models have more spectral power in 2-3 years compared to observations).

Also ocean resolution seems to be important for mean biases, and the

realism of the ITCZ should be mentioned as well.

L105, What about UKESM1? Not sure why only HadGEM is mentioned.

L111, years or months?

L117, Which levels? Above you just mention 70 hPa.

L119, '1' and '2' are subscripts

L124, The product you use (GPCP?) for this index is providing convective and stratiform

precipitation separately? Or is it a total precipitation? If not, remove convective

(here and also in all instances following). 

Can you explain why using a precip-based IOD index rather than the standard SST-based one?

Please add a reference if it was used before.

L125, I'd use same style for EN3.4, with []

L133seq, This symmetry seems strange (given the ENSO asymmetry and QBO stalling) can you provide numbers?

L135, This is 'observed' for ERA5? Can you provide the values for HadSST? It is useful to compare model/observation statistics.

L140, Maybe start with 'When estimating correlations, they are...'

Fig 1, 'mm day-1' in brackets, or move 'pr' to title

L159, Please comment on the ITCZ realism.

L162, Add reference

L206, I guess would be useful to have a table in the method section with the 

different numbers for ENSO and QBO. Why 120, does it have a special meaning?

L209, But the wet anomaly in the Pacific and dry in the Atlantic are more marked with ENSO included. 

This is also seen in Fig5.

Fig5, If regression coefficients are re-scaled (caption), then a prime is missing in a&d titles.

See Supplement as well.

L214, (1) -> (Fig. 1)

L216, it was EN3.4 before

L221, why no significance in FigS3?

L225, mention Gray et al., 1992

Fig6, I'd use E and W for QBO in (b). Moreover I would define once all the acronyms

(EN, LN, E, W) in the methods and be consistent throughout (no 'ea', EN3.4, etc.).

Suggest NE or NN for Neutral ENSO. Moreover, would it be easier to read the plot ordering 

the boxplots as LN/NE/LN ? Why not showing E and W phases separately for the amplitude?

L238, Have you stated which level are descent rates for? From the methods I got that the 

amplitude is integrated in the 10-70 layer, but descent rate is by level.

L246, See Geller et al 2016 about GW variations.

L252, So the frequency would be for example (# months EN) / (# months W)?

Maybe mention that IOD will be considered later?

L260, ENSO3.4 -> EN3.4 (or maybe ENSO)

Fig 7, [] missing around mm day-1 (check other plots as well). I guess IOD-prc is same as IOD?

L266, write months in full. Can you elaborate on how the difference model/obs

depends on the ENSO evolution in the model (e.g. Lengaigne et al., 2006)?

Also worth nothing how the model index amplitudes are 2-3 times smaller than obs.

Fig8, as before, why 'convective'? Why now using a higher confidence level?

L275, but could this be model-dependent?

L280, Please avoid the mix of abbreviations and months in full

L286, Maybe the Indian Ocean sector, rather than IOD?

L293, why '.'?

L295, atmospheric circulations. However, the model biases should be noted.

L300, How are these longitudes selected?

Fig9, Only convective, stratiform rainfall removed? Is panel (b) indicating a double ITCZ bias?

Can you comment in the text?

L317, remove 'rate'

Fig10, define acronyms MSD, NAM. For more direct comparison you could mask values

over oceans? Do you know why the regions show very net boundaries in some cases?

Compare with Lee and Wang, 2012 their Fig4

Fig11, I am confused by vector sizes. They are 3 or 0.3 10-2 Pa s-1,

 but their lengths do not differ by a factor 10. Please clarify.

Also the plots are quite busy, can you try improving them?

L330, Mention the QBO biases which may be important

L335, If you integrate to the top, then the integration bounds are swapped 

and 0->p_top (or p_surf)? Gravitational acceleration (g) rather than constant (G)?

How do you compute the divergent component of zonal wind?

L351, To me some QBO/ENSO superposition can also be seen from the plots.

L406, or role of QBO bias...

L416, have you ever mentioned TRMM in the text?

L420, 'observations' -> 'variables'

L422, revise. you speak about days, I understand the input data is monthly mean, 

so it this weighting already built in? Does the MOHC model have 360_day calendar?

L435, I guess the 'i' subscript is redundant with one predictor? Same in Fig S3

L440, State that summation is 'j=1...N', as X_0 appears already

L446, Is there a stray A3?

L513, why uppercase?

Additional references

Geller et al JGRA 2016 https://agupubs.onlinelibrary.wiley.com/doi/10.1002/2015JD024125

Gray et al., JMSJ, 1992 https://www.jstage.jst.go.jp/article/jmsj1965/70/5/70_5_975/_article

Lee and Wang, CD, 2012 https://link.springer.com/article/10.1007/s00382-012-1564-0

Lengaigne et al., JC, 2006 https://journals.ametsoc.org/view/journals/clim/19/9/jcli3706.1.xml

The tropical route of QBO teleconnections in a climate model

Jorge L. Garcia-Franco, Lesley J. Gray, Scott Osprey, Robin Chadwick, and Zane Martin

Recommendation: May be publishable after major revisions

The paper seeks to understand the link between the tropical stratospheric QBO and variability elsewhere in the tropical atmosphere (and on ocean SST). The tools used include observational/reanalysis output since 1979, and several preindustrial control runs from various versions of the Met Office model. This is an interesting subject and the foundation of a paper that could eventually be publishable in WCD is clearly present. Many of the arguments are not convincing in their current form however, and while I don’t think these critiques are insurmountable, addressing them will require some major rethinking and rewriting.

My main criticism is that the authors argue there is a connection between the QBO and ENSO, but the evidence provided is not strong enough (and in this reviewer’s opinion the authors’ claims are actually incorrect). First, the observational period covered by this paper only begins in 1979, however high-quality radiosondes have tracked the QBO since ~1953, and reliable information on the ENSO state is available even earlier. Studies that have used the entirety of the observational record have reached an opposite conclusion of that reached by the authors. Specifically, in the period before 1979, there were more easterly QBO events simultaneous with El Nino. This has been noted by at least three papers (Garfinkel and Hartmann 2007, Hu et al 2012, Domeisen et al 2019), none of which were cited in this paper. The net effect is that the observed connection between ENSO and the QBO is non-stationary, and (cherry-) picking a limited subset of the full observational record can lead to misleading (and erroneous) conclusions.  Over the entirety of the observational record (at least until 2018, the last year considered by Domeisen et al 2019), the correlation was essentially zero.

Second, the modeling evidence presented by the authors for a relationship between ENSO and the QBO is also misleading and perhaps wrong. The authors consider several different simulations from one model, however Rao et al 2020 (not cited) recently considered the connection between ENSO and the QBO in ~17 different CMIP5/6 models. Rao et al found that some models simulated a connection of the same sign as that found in this paper. However other models simulated an opposite effect. Notably, the two MetOffice CMIP6 models considered by Rao et al 2020 had opposite responses (their Figure 11n and 11r). The multi-model mean effect was essentially null in Rao et al 2020. Thus, it is conceivable that the MetOffice models examined in this study do indeed simulate a connection between ENSO and the QBO, however this relationship does not appear to be generic, and future work is needed to unravel the causes of model disagreements.

The net effect of these criticisms is that I don’t think it is particularly informative or meaningful to study the tropical atmospheric response to the QBO unless and until the ENSO signal has been regressed out. The authors indeed do perform such a regression, and they also additional examine ENSO neutral years only, which is great! But the analysis earlier and also later in the paper is suspect to this reviewer. Stated another way, the authors themselves note that the observed response to the QBO depends sensitively on whether neutral ENSO only is examined, so why even show the observed response before removing the ENSO influence?

My suggestion is to focus on the results where ENSO is “removed” much more or exclusively (as they do indeed contribute to the scientific discourse), and significantly shorten the rest of the paper.  At the very least, the discussion of the figures 1,2, 3, 7 needs to be rewritten.

Finally, Rao et al 2020 also consider the response of OLR and precip to the QBO with the ENSO signal regressed out, and find a wide range of responses across the models. Particularly perplexing to this reviewer is that Figure 7n/8n and 7r/8r of Rao et al consider the OLR and precip response in two different versions of the Met Office models, and find if anything opposite results. The present paper focuses mainly on the higher resolution runs which were not analyzed by Rao et al, however the authors should include in the supplemental material additional figures for the other model versions for most of the figures in the paper.

Specific comments:

1. My general comments mentioned four very relevant papers that appear to have not been cited. Please add them as appropriate throughout the manuscript.

2. The authors attempt to remove an ENSO influence throughout by forming eQBO and wQBO composites during ENSO neutral years only. Note that this doesn’t guarantee that the mean ENSO index during the wQBO and eQBO composite are actually identical. Can the authors compute the mean of the Nino3.4 index for these composites, in order to confirm that any ENSO influence is removed?

An ENSO index can be removed also by linear regression, e.g., linearly regressing out variability associated with the Nino3.4 index, as done in figure 5. The authors seem to prefer to examine neutENSO conditions instead. There are pros and cons for both methods, and it would be worth noting in the text if results are different for either method of attempting to remove the ENSO influence.

Comments on specific lines/figures/tables (mainly on the first half of the paper, as I will likely review the revised version again):

Line 55 add Rao et al

Line 69 the the

Figure 1, 4, and 8: This figure looks fairly different from figure 8a, 8n, and 8r of Rao et al. Particularly perplexing is that 8n and 8r of Rao et al, which focus on two different versions of MetOffice models, do not agree with each nor with any of the panels here as best as I can tell. There are certainly many methodological differences between the studies (whether/how ENSO is removed, the season analyzed, historical vs. PI control), but if the results are so sensitive to these choices then the overall effect may not be particularly robust.

Line 184-185, 250-270 see my general comments about the ENSO-QBO relationship. These sentences are not representative of the entirety of the published literature or other runs of the model used in this paper.

Figure 3: please use as much as possible of the 1953-2022 period for the observational composites. I expect the resulting figure to be rather different to what is shown here, which will necessitate a rewrite of the accompanying text.

Table 1: please add the other model versions to this table

Section 3.4: Garfinkel and Hartmann 2011 (already cited) discuss changes in convective precipitation and OLR over monsoon regions and the ITCZ in response to the QBO with fixed SSTs. Note that Garfinkel and Hartmann 2011 also performed some targeted experiments in which the QBO profile nudged towards was modified (line 409).

Section 3.4: Hu et al 2012 discuss Walker circulation changes in response to the QBO. Please include in your discussion.

Figure 11 caption discusses panels g and h, which don’t appear to exist.

Rao, J., Garfinkel, C. I., & White, I. P. (2020). How Does the Quasi-Biennial Oscillation Affect the Boreal Winter Tropospheric Circulation in CMIP5/6 Models?, Journal of Climate, 33(20), 8975-8996.

Hu, ZZ., Huang, B., Kinter, J.L. et al. Connection of the stratospheric QBO with global atmospheric general circulation and tropical SST. Part II: interdecadal variations. Clim Dyn 38, 25–43 (2012). https://doi.org/10.1007/s00382-011-1073-6

Domeisen, Daniela IV, Chaim I. Garfinkel, and Amy H. Butler. "The teleconnection of El Niño Southern Oscillation to the stratosphere." Reviews of Geophysics 57, no. 1 (2019): 5-47.

Garfinkel, C. Iꎬ, and D. Lꎬ Hartmann. "Effects of the El Niño–Southern Oscillation and the quasi‐biennial oscillation on polar temperatures in the stratosphere." Journal of Geophysical Research: Atmospheres 112, no. D19 (2007).

The paper analyses the QBO in 3 long preindustrial control experiments with different versions of the HadGEM climate models. The emphasis is  on the equatorial teleconnections and the possible coupling to the  ENSO. The results are compared to observations and reanalyses. The long model experiments make it possible to get more statistically significant results than in the much briefer observational records.



The paper is in general well written and the subject is interesting. However, many aspects of the methodology is not explained in enough details. I also find that the authors sometimes over-interpret the differences they find in the composites. I find that the paper needs some major improvements before it can be accepted for publication.





Specifc comments:



l110. I guess this description also is valid for the models and not just ERA5.



l130: I don't understand this weighting. Why is this important and how important is it?  In particular the weighting with the number of days in each month cannot be important.  The annual-mean composites seem to long-term means.



l135: I don't think I understand these counts. For example, in observations you have 62 QBOW Elnino months in a 40 years period while you have 376 months in 500 years for the model.  But 62/40 is very different from 376/500. Does the model behave differently from observations or have I misunderstood something?



l140:  More details should be given here. Is it individual years or months that are resampled? The time-scale of both the QBO and the ENSO are much   longer and this should be reflected in the resampling procedure. If this is not done, the significance will be overestimated.



l165: The signal in the model seems weaker and more confined than in observations.



Figure 3: I would say that the signal is in general weaker in the model than in observations and that there are considerable differences between model and observations.



l224: It is not correct that multiple linear regression assumes that the independent variables are orthogonal. But they cannot be linearly dependent.



Figure 6: The authors should briefly mention what the box plot shows: median, std. dev. etc.



l240-250: How is the statistical significance of the differences in Fig. 6 estimated?  The spread seems very large.



l249: Christiansen et al. 2016 (doi:10.1002/2016GL070751) suggests that strong warm ENSO events change the phase of the QBO. Is there evidence for this in the model?



Table 1 Why are the errors smaller for the observations than for the model?



l250-255: I don't understand this paragraph. Why do you look at the pdfs? The K-S test tests if the pdfs are different, and not necessarily if the averages are different.



l260: Does this significance refer to the * in Fig. 7? There are very few *.



l268:  They are very often opposite in sign. Can you say that the numerical values are different within the error-bars?



Figure 8: The hatching is hard to see.



l275: The plots in Fig. 8 seem very similar to me. Are you sure the difference of differences are statistically significant?

l370: What is the difference between the sentence 'When only ..' and the following sentence?