General comments:

This manuscript applies multiple linear regression (MLR) to monthly mean zonal wind data in the stratosphere, mesosphere, and lower thermosphere obtained from SABER observations, MF and meteor radar observations, and MERRA2 meteorological reanalysis to examine the effects of QBO, ENSO, and solar activity as well as seasonal changes and long-term trends. Although many similar studies based on the MLR analyses have been conducted using long-term meteorological reanalysis data, there have been few research above the stratopause due to the difficulty of observing winds. In this sense, the efforts in this manuscript are commendable. On the other hand, the method of MLR analysis and statistical significance are not well documented, and the consideration of the short data period is not sufficient. In addition, English grammar check by a native speaker is also recommended. Therefore, I think that this manuscript needs substantial revision before publication. Detailed comments are given below.

Major comments:

• Time interval of the data

It seems that 18 years are too short to fit the 11-year solar cycle. Although the authors evaluated its impacts by changing the time interval, half or more of the data periods overlap, which does not seem very meaningful. Rather, a comparison using 40 years of MERRA2 data would be more meaningful. As the authors say, the MLS has been assimilated since 2004, but its effect appears to be strong only for the vertical structure of temperature, not so much for the meridional gradient of temperature and the distribution of zonal wind that is related to the meridional gradient of temperature.

• Method of MLR analysis

From the explanation in section 2.2, it appears that eq. (2) is applied to data for 216 months over 18 years, in which case only one regression coefficient is obtained for the entire period. On the other hand, section 3 shows that regression coefficients were obtained for each month, suggesting that eq. (2) without including the seasonal variation term was actually applied to 18 years of data for each month. In that case, I do not know how the seasonal variation was estimated. The authors need to properly explain the MLR method.

• Multicollinearity

In the MLR analysis, multicollinearity often leads to meaningless results. The authors need to evaluate and indicate whether the correlations between regressors are sufficiently small before performing the MLR analysis.

• Statistical significance

In this manuscript, the regression coefficient is considered statistically significant if it is greater than 1σ. However, there is no description of how σ is calculated. In addition, when determining whether a regression coefficient is statistically significant in the MLR analysis, it is common practice to use the p-value of each regression coefficient. Unless there is a special reason to use σ, the p-value should be used (e.g., Mitchell et al. (2015)).

• Impact of SSW

In general, if the effect of SSW is large, it should occur that the regression coefficient is not statistically significant despite its large value. The authors should first check to see if this is the case, especially in the high latitudes of the winter northern hemisphere.

Furthermore, it is questionable whether it makes sense to apply the MLR analysis to spline interpolated data. Also, it should be explicitly stated which latitude bands were replaced by spline interpolation. Looking at Fig. 10c, it appears that all winters were replaced by spline interpolation, but major SSW does not occur every year. It should be explicitly stated by what criteria SSW is defined.

Minor comments:

• L. 143-145

Is it safe to consider data from a single point observation as the same as the zonal average, even though it is a monthly average? For example, how does this compare to the data of Smith et al. (2017)?

• Fig. 2

It is hard to see the phases from the arrows. I recommend to show the amplitudes by contours and the phases by colors.

• L. 237

Please clarify how the annual mean response was calculated. Is it an annual mean of the regression coefficient for each month? Or did you apply the MLR to the data including whole months (216 months)?

• L. 275

higher southern (northern) latitudes in summer (winter) → higher latitudes in the winter hemisphere

• L. 275-277

I cannot see the signal at 50S/N at z=50-80 km.

• L. 381-420

Trend fitting is sensitive to the values at both edge points. The authors need to mention this point.

• L. 455-456

I think that the seasonal asymmetry is explained by semiannual and terannual components to some extent.

References:

Mitchell, D.M., Gray, L.J., Fujiwara, M., Hibino, T., Anstey, J.A., Ebisuzaki, W., Harada, Y., Long, C., Misios, S., Stott, P.A. and Tan, D. (2015), Signatures of naturally induced variability in the atmosphere using multiple reanalysis datasets. Q.J.R. Meteorol. Soc., 141: 2011-2031. https://doi.org/10.1002/qj.2492

Smith, A. K., Garcia, R. R., Moss, A. C., & Mitchell, N. J. (2017). The Semiannual Oscillation of the Tropical Zonal Wind in the Middle Atmosphere Derived from Satellite Geopotential Height Retrievals, Journal of the Atmospheric Sciences, 74(8), 2413-2425. Retrieved Jan 18, 2023, from https://journals.ametsoc.org/view/journals/atsc/74/8/jas-d-17-0067.1.xml

The authors derived the global balance wind (BU) in the height range of 18-100 km and latitudes of 50 S- 50 N from 2002 to 2019 using the gradient wind approximation and SABER temperatures and modified by meteor radar observations at the equator. Using this data set, the authors examined the responses of zonal wind to QBO, ENSO and solar activity. MERRA2 zonal wind is used to validate BU and its response below 70 km. The manuscript is well organized and easy to follow. The interannual response as a function of month is an interesting results. My main concern is the dataset length and the significance test of the presented coefficients. BU has 18 years of data, to fit interannual response as a function of months, this means there are only 18 data points available for the fit. The manuscript did not present any significance test of fit. Rigorous significance test(s) should be added to make the results trustworthy to readers. 

More detailed comments:

1. More details should be added to explain the tidal aliasing at the equator. I assume tidal aliasing is not only an issue for 0 degree, it should be the equatorial region. How can a meteor radar station represent the zonal wind in the whole equatorial region? Is there any potential aliasing from semidiurnal tide in the mid latitudes?

2. More details should be added to the MLR model especially on how the monthly coefficients for ENSO, QBO, and solar are obtained. How do you deal with the AO, SAO, and TAO when obtaining the monthly coefficients for interannual variability. Line 170, how does 42 parameters come about?

3. Rigorous significance test should be added. I would suggest a Monte Carlo method. Other methods involve equations have underlying assumptions. If possible, multiple significance test methods should be used. 18 years data is short to study solar cycle, and 18 January is short to get interannual variability on the time scales of 2-5 years (ENSO and QBO) response on a monthly basis.

4. Figure 1 g, the R^2 value is 0.98. This number is somewhat misleading since I assume the goodness of fit is mainly coming from seasonal fit. How good is the fit if only interannual variability is considered?

5. In the discussion of the effects of the data interval, I share another reviewer's point of view: all the data intervals are overlapped somewhat. At least two totally separate data intervals should be used. Significance tests should be added here as well.