This study examines the relationship between the QBO and tropical high clouds in observations and CMIP6 models. Unlike previous studies which typically used simple regression or composite analysis, this study applied a cloud-controlling factor (CCF) framework (with modifications) and ridge regression. Observational analysis revealed a significant reduction in tropical high-cloud fraction during the QBOW, due to warmer upper-tropospheric temperature (TUT), increased static stability at 150 hPa (S150), and lower upper-tropospheric relative humidity (RHUT). While this result is partly consistent with that of Chen and Thompson (2026), the combined effects of the three factors are discussed in more detail within the CCF framework. The models systematically underestimate the response of high-cloud fraction to the QBO, confirming previous studies. They suffer from mislocated cloud sensitivities and underestimated S150 signals. This result highlights that improving high-cloud sensitivities to thermodynamic parameters is essential to more accurately capture QBO-related high-cloud changes.

Observational results are not surprising, as previous studies have reported the similar findings. However, the identification of model biases and possible sources of inter-model spread is novel. The model-observation comparison was conducted carefully, and the conclusions are well founded. I am confident that this study will contribute to a better understanding of QBO downward influence in the tropics and improving climate models. I recommend publishing this paper with minor revisions.

Major

Seasonality: This study does not consider seasonality. It would be helpful to quantify the seasonal dependency of the QBO-high cloud relationship, as shown in Sweedy et al. 2023.

Minor

L28 Reference: You can include Haynes et al. 2021 (https://doi.org/10.2151/jmsj.2021-040)

L137 TUT: Is TUT computed in the same way as RHUT, vertically averaged within the 200 hPa below the tropopause?

L147 Eq. 1: What is r? What is the data resolution? Does 21*11 indicate latitude grids within 30S-30N and pressure levels?

L152 Ridge regression: I am not familiar with this method. How does it minimize the correlation between variables. TUT, S150, and RHUT are closely related with each other.

L169 Zonal wind: A 3-month lag is applied but it still shows a maximum signal at 50 hPa. Is this correct?

L245 Reference: You can include Son et al. 2017.

L367 Model error: Models show weaker or opposite TTL cloud sensitivities to S150 and RUHT. Why? I know this is a very difficult question. Any speculation would be helpful.

L393 Model error: TUT signals are not consistent across models. Could this be related with model vertical resolution, lowermost level of the QBO signal in the model, or something else? Any speculation would be helpful.

Overall Comment: 

Many previous publications have shown observational links between tropical high clouds and the QBO. This study evaluates these connections in CMIP6 class models which can reproduce the QBO with some fidelity. Findings show several deficiencies in the model simulation of QBO influence on high clouds, particularly those in the TTL. By using a cloud-controlling factor analysis, the study also investigates the environmental pathways by which the QBO influences high clouds in both observations, and models. The study suggests that the largest reason for differences between models and observations is the modeled sensitivity of cloud fraction to the cloud controlling factor. 

I found this paper enjoyable to read and generally found it easy to understand. However, there are a few locations and descriptions where the paper should be further refined. I would really enjoy a further explanation of the significance of the TUT results, especially due to recent papers which have argued for the importance of S150 (Chen and Thompson, 2026). The paper as it currently feels like a lot of background before getting to any of the actual results. I thought the results were straightforward, so you may be able to cut a significant amount of the intro, data/methods, and preamble to Fig 1 in the results to have a more succinct story. 

Major Comments:

1. This paper would benefit greatly from more discussion/analysis related to the statistical significance. Given that we only have 14 years of calipso data, and we are trying to test something related to the QBO, we only have ~6 independent samples (not accounting for complications due to the QBO disruptions). I know this partly motivates ridge regression as opposed to linear regression, but I still wonder if you can really fit 6 predictors to the model. 
1. When showing significance, are you accounting for autocorrelation? 
2. Line 124: How is ENSO regressed out? Is it regressed out of the QBO index and all the latxlon datasets? Did you regress it out with a lag in the ENSO index? ENSO impacts on TTL clouds have been found to maximize after a few months (Tseng and Fu, 2017). 
3. How is TUT defined? Consider more clearly showing how you calculate TUT, especially given its relevance in the later results. 
4. Line 139: You change upper tropospheric stability to S150 related to the Chen and Thompson results. I am wondering if you should also change w300 and u300 to things more relevant for the TTL. For example, you might change w300  to w100 or u300 the shear between 300 and 100 hPa. Given that many of the thermodynamic changes driven by the QBO operate via changes in the vertical velocity in the upper-troposphere and lower-stratosphere, the w100 might be interesting to look at. 
5. Equation 1: I found it difficult to understand what each term/symbol represented. Consider defining everything in the sentence immediately following. For example: “... where AMT represents the cloud fraction, r represents the individual grid cell, Θi represents the sensitivity of high cloud fraction to the ith cloud-controlling factor (Xi) and M is the total number of cloud-controlling factors (M=6)”.
1. Writing the equation for Θi as d(high_cloud)/dXi would have been helpful for me.
6. Fig. 1: I liked the comparison of models and observations here. It may be helpful to provide a difference row as well as the obs and modeled results. 
1. Consider reducing the hatching thickness as the current version feels too crowded
2. Consider adding thin black lines as another contour at fixed levels to highlight changes in the magnitude
3. For Fig S2, consider using the same colormap and colorbar as done in Fig 1 for ease of comparison
4. For Fig S3, consider using same colormap and colorbar as Fig 1
7. L224: I noticed this at this point, but this is true throughout the paper. Consider reducing the emphasis on QBO westerlies reducing high clouds. While this is of course true, some studies have shown that the impact of the QBO on the TTL is actually much stronger during QBO easterlies (Tegtmeier et al., 2020). This is more subtle in this study because it is regression based, but may confuse the results. For example in L227-229, you state that the zonal asymmetries in the cloud impact are tied to QBO westerlies, when in reality, these are just the regression coefficients and may be dominated by easterly QBO variability 
8. L294-295: This is a very interesting statement. I would like to understand this better. If this suggests that clouds respond separately to these 3 predictors, could you give more intuition or examples on how? More importantly, how do we square these results with the recent results of Chen and Thompson (2026) which seemed to argue that previous studies overstate the QBO-cloud coupling, and that studies should only look at the S150 variable? Is the flaw in that study that they only consider S150? If so, it would be helpful to have some physical intuition on why TUT is important. I know you expand on this slightly in L300-305, but having more discussion would highlight that stability alone is insufficient to describe cloud changes, and highlight the discrepancy with the Chen and Thompson study. 
1. I think Fig S6b shows that including all predictors with TUT (Fig. 5b) gives a larger response than all predictors with TSURF (Fig. S6b). Please correct me if I am wrong. 
2. How much of the variance between S150 is shared with Tut?
9. Fig 4: Given the importance of TUT here, it should be included in this figure. 
1. Fig S7: I don't understand the TUT sensitivity. Is it true that the reconstruction in Fig. 3b should be obtained by multiplying the pattern in Fig. S7aa? If I understand correctly, the positive values in Fig. S7a would suggest that an increase in temperature leads to an increase in cloud fraction. This seems opposite of what I would expect, and also contradictory to the results of Fig. 1. Please let me know if I am reading this incorrectly. 
2. Also, the TUT sensitivity in S7 seems spatially disorganized compared to S150. I think more interpretation of the results here would be helpful
10. L345-355: Could some of the zonal discrepancies between obs and models here be explained by differences in tropopause between obs and models? In observations, the QBO influence is much stronger above the tropopause, and much weaker below. If the tropopause is incorrect in models, this might prohibit QBO anomalies from extending into the troposphere. In observations the tropopause has longitudinal dependence, and this may be true for models as well. 
11. Fig 5: Should the observation total in Fig. 5b be the same as the observational value in 5a? I am not familiar with the regression techniques used here, but if not, then does this change the interpretation of the role of TUT from Fig S6? 
12. L370-380: I noticed this here, but I think it is relevant throughout the paper. I am having trouble interpreting the %/std(QBO) units here. Sweeney et al. (2023) show variations in cloud fraction of about 2-5%. I know this is a composite difference compared to a regression coefficient here, but it still seems that there might be a discrepancy. Please correct me if I am misunderstanding.

Minor Comments:

1. L87: “...and so on” → “among other reasons (cite relevant paper)”
2. L98: “...underrespresentation of very thin clouds in CALIPSO-GOCCP compared to standard CALIOP-NASA products” → This seems like it could be important given that many of the TTL clouds are very thin. Is this something that may impact the results?
3. L159: Discussion on the QBO definition might be better for section 2.
4. L231: consider citing Tegtmeier et al. (2020)
5. L255: “with a proposed mechanism…” → “with one of many proposed mechanisms involving…”
6. L257: “Figure 2 indicates the strongest QBO-cloud relationship in observations occurs over the maritime continent…”
7. L257: Consider also citing Lin and Emanuel (2023)
8. Fig 3: Given that RHUT is one of the three main contributors, consider moving it to row 3.  
9. L311: “...do not align spatially in the MMM…” Do you mean that they do not align spatially with the observations or that the 3 predictors have less similar spatial patterns than those of the observations?
10. L322-323: My understanding is that this has been known from previous papers, consider citing them here. 
11. L91–92: "(Swales et al., 2018**). has been developed" should read "…2018) has been developed."
12. L92–93: "generates outputs that is comparable" → "that are comparable."
13. L155: "20 years of MODIS data are also used from comparison" → "for comparison."
14. L257: "the QBO–cloud relationship in observations peak occur over maritime continent" → "…peaks over the Maritime Continent"
15. L355: "Figure S6 (first raw)" → "first row."
16. L401–402: "both cloud sensitivities and QBO modulation play part" → "play a part."
17. L403: "the cloud sensitivities to TUT and S150 varies" → "vary."
18. L104: "22 years of MODIS dataset are analysed" → "22 years of MODIS data are analysed." Also consider avoiding starting the sentence with a number. 
19. L111: "global circulation models (GCM)"  I am familiar with GCM meaning general circulation models, not global. 
20. L24, L122 and throughout: inconsistent unit spacing ("10hPa," "150hPa," "440hPa") vs the spaced form used elsewhere ("150 hPa"). Standardize.
21. L53: "1 W/m2" → "1 W m⁻²" (and superscript).
22. L124: "Niño3.4" → "Niño 3.4."

References:

1. Tegtmeier, S., Anstey, J., Davis, S., Ivanciu, I., Jia, Y., McPhee, D., et al. (2020). Zonal asymmetry of the QBO temperature signal in the tropical tropopause region. Geophysical Research Letters, 47, e2020GL089533. https://doi.org/10.1029/2020GL089533
2. Lin, J., and K. Emanuel, 2023: Stratospheric Modulation of the MJO through Cirrus Cloud Feedbacks. J. Atmos. Sci., 80, 273–299, https://doi.org/10.1175/JAS-D-22-0083.1.
3. Sweeney, A., Fu, Q., Pahlavan, H. A., & Haynes, P. (2023). Seasonality of the QBO impact on equatorial clouds. Journal of Geophysical Research: Atmospheres, 128, e2022JD037737. https://doi.org/10.1029/2022JD037737

Summary

This paper applies a cloud-controlling factor and ridge-regression framework to diagnose the relationship between the QBO and TTL high-cloud in observations and CMIP6 models. Whilst the results reported for the QBO/high-cloud relationship in observations is replicated in the literature, they act as a useful baseline for the CMIP6 model analysis that is novel and of great interest. The authors show that the CMIP6 multi-model mean reproduces the sign of the tropical-mean QBO/high-cloud response from observations, but underestimates its magnitude and struggles to replicate the zonal structure. Consistent with other studies, TTL static stability is identified as a key mediator of the QBO/high-cloud relationship, as well as upper troposphere temperatures (TUT) and relative humidity (RHUT). The discrepancy between CMIP6 models and observations is attributed to issues with cloud sensitivities to the cloud-controlling factors. 

The results have important implications for improving downward influence of the QBO in climate models, as well as emphasising the importance of TUT and RHUT alongside static stability as a pathway for coupling between the QBO and convective systems such as the MJO via cloud-radiative effects. However, the results would be strengthened by clearer uncertainty quantification and robustness testing given the short timespan of some datasets and potential sensitivity to the choice of cloud-controlling factors. Moreover, the study is at present difficult to reproduce given a lack of details on the methodology. 

Major comments

1. Introduction/conclusion: The authors cite studies such as Sweeney et al. (2023) and  Chen & Thompson (2026) which are closely related, but could more clearly state what the original contribution that their research represents. What are the main points of difference? What is the significance of the emphasis on TUT in the authors’ results?
2. Robustness testing and details of CCF analysis (section 2.1): 
1. The authors give a useful explanation of the replacement of surface temperature with TUT and the use of 150 hPa static stability instead of a mean upper-troposphere value. However, these variables are not defined in sufficient detail, especially TUT which is central to the conclusions.
2. The authors claim that the changes to the CCFs “better represents” QBO-related pathways (L138) but without justification. Can you quantify the reconstruction skill, for example the fraction of variance explained or a spatial map of residuals? Can the authors show these metrics for a few different sets of CCFs to convince the reader that the chosen set really is the best choice?
3. Methodological details.
1. The methods section is already reasonably long (in that it takes a long time to get to what are relatively succinct and neatly presented results) but there remains a lack of detail in the methodology. For example, many of the terms in the CCF definition (equation 1) are not defined, and it is difficult to interpret given the choice of variable names and symbols. The authors might consider comparing-and-contrasting their methodology with other closely-related studies such as Chen & Thompson (2026) – the main point of differences appears to be the use of area-averages rather than grid-point values; can the authors elaborate on why?
2. Similar to above, there is no information given on the ridge regression methodology or the way in which ENSO is removed. Did the authors consider removing other important modes of variability? The authors may consider adding an Appendix with the fine details of the methodology included.
3. The authors use a standard QBO index (L162) at 50 hPa with a three-month lag. Since the QBO/high-cloud pathway operates in the upper-troposphere/lower-stratosphere, have the authors considered using a QBO index at a lower pressure level with no lag? Are their results robust to changing the QBO index?
4. The CALIPSO record used is short, containing only a few QBO cycles, and importantly the record includes QBO disruptions. This is an important caveat that should be more clearly dealt with by the authors.
5. The authors state (L294-295) that clouds respond separately to TUT, RHUT, and S150. I am not sure I follow their arguments. Can the authors clarify this statement?
6. The authors state (L112) that 7 models generate a “reasonable QBO”. What is the selection criteria used?
7. Presumably the multi-model mean is an even-weighted mean of the 7 models (L113). Is that appropriate? Are the models all independent in the sense that they use distinct dynamical cores? If not, the authors might consider weighting the average.

Minor comments

1. L53: you mention the seasonality of the QBO impact on high clouds in the introduction, but do not explore this in your own study. Is this a simple extension of the results that the authors might consider including? Particularly since they mention consequences for the QBO-MJO connection (L255) which is strongest in boreal winter.
2. L97-98: “results in an underrepresentation of very thin clouds in CALIPSO-GOCCP compared to standard CALIOP-NASA products” -> what is the consequence of this in the context of this study?
3. L332, 432: I found use of the term “mislocated” confusing. It is hard to discern whether the authors mean that the cloud sensitivity to the CCFs is wrong itself, or whether the erroneous sensitivities result in erroneous responses to the CCFs that produce a spatially distinct response in the CMIP6 MMM compared to observations. 
4. Figure improvements
1. Fig. 1: the stippling and hatching makes it difficult to see the underlying plots. Either reduce the stippling/hatching or use a colourbar with more contrast. Can the authors provide some guidance to the reader on how to interpret the units (e.g. m/s/sigma)?
2. Fig. 3: should share the same colourbar as fig 2 to aid comparison. There is also a lot to absorb from this figure: could it be simplified to help with interpretation somehow? Note that a lot of the results shown in this figure are not even mentioned in the text, so arguably not relevant to the central story. 
5. Typos:
1. L91: “mimic” -> “mimics” and remove the . after the citation
2. L156: “from comparison” -> “for comparison”
3. L195: “The reduction … in CALIPSO (Fig. 1i)” -> Fig. 1e?
4. Fig. 2 caption refers to MODIS but figure subtitles say CALIPSO
5. Fig. 2 caption says hatching in “(b, d, e)” -> (b, d, f)?
6. L257: “peak occur” -> “peaks”
7. L297: “with S6). And” -> join to form one sentence?
8. L355: “first raw” -> “first row”
9. L454: “CMIP5/6 data” -> was CMIP5 data used?
10. L416-417: “but systematically underestimating” -> “underestimate”