The manuscript is overall well written, logically structured, and presents a useful and innovative nudging strategy with clear relevance to atmospheric modeling and constrained simulations. The methodological development is meaningful, and the results are generally convincing. I only have several minor comments and requests for clarification that I believe will further strengthen the manuscript.

Equation3. It seems that “1 Pa” in the equation may actually refer to 1 hPa. Please verify whether this is a typographical error.

In addition, how sensitive are the results to the choice of P_0 ? It would be beneficial if the authors could provide guidance, or at least discuss strategies, for choosing  P_0 , especially for readers who might apply the method in different models.

Line 133-135. The nudging tendency term is calculated and applied at different locations in the model. Could the authors explain the reasoning behind this choice?

Is there a specific numerical or physical advantage to computing and applying the tendency at different places?

Line 304-307. The speculation here is interesting. Could the authors be more specific about what type of microphysics tuning that is applied in the free-running configuration but cannot be to the nudged simulations? Relatedly, this raises a broader question about other sub-grid tuning (e.g., CAPE relaxation time in the convection scheme). The authors may wish to expand their discussion on how such tuning parameters might interact with or influence nudging behavior.

Figure 1. For DNDG-UVQ and DNDG-UVTQ, the reported PCC values (0.79 and 0.82) are already quite high. It may help to comment on whether such high correlations are expected or what they imply about the baseline model behavior.

Figure 2. What does the “STRESS MAG” stand for?Moreover, it does not appear to be listed in Table B1. please clarify or include the relevant information.

Figure 5. Why is the analysis limited to December 1, 2010 to November 30, 2011?

A short justification would help the reader understand this choice.

Figure 6. The NDG-UVQT_SRF1 experiment does not seem to substantially improve PRECT. Could the authors explain why this might be the case?

Figure 10 seems to indicate that nudging temperature and humidity does not significantly change the result. Could the authors elaborate on why the impact is relatively small here?

The manuscript proposes an improved nudging strategy for EAMv2 based on vertically modulated tendencies, aiming to mitigate distortions of physical processes caused by improperly designed nudging of temperature and humidity. Overall, this study addresses a practical problem in constrained atmospheric modeling and demonstrates a certain degree of innovation. Despite these positive aspects, several important elements of nudging implementation are insufficiently documented. In addition, some key conclusions would benefit from additional sensitivity experiments to better demonstrate the robustness and general applicability of the proposed method. In summary, this work represents a contribution with practical value and some novelty. I recommend that the manuscript undergo major revision prior to publication.

Major Comments:

1. Nudging is intended to steer the model state toward the observed state and typically involves weighting parameters in both space and time. However, the methodological description in Section 2.2 only addresses vertical spatial weighting. In the authors’ experiments, each model grid point obtains an interpolated value at every integration time step. For observations with longer temporal intervals, is  also derived through spatial and temporal interpolation in order to perform nudging? If so, could this introduce additional errors for highly nonlinear variables such as water vapor? The authors are encouraged to provide a more detailed explanation of the horizontal spatial weighting and temporal weighting schemes. In general, observations should exert a smooth and continuous influence before and after the observation time to avoid excessive shocks to the model. Furthermore, it is currently unclear whether a relaxation coefficient is included to control the rate at which the nudging tendency is introduced into the model state. If such a parameter exists, it should also be explicitly specified and described.
2. According to the authors’ description, the proposed nudging configuration appears conceptually similar to the grid-nudging approach used in WRF, in which ERA5 is treated as pseudo-observations and interpolated to each model grid point to compute the local nudging tendency. If this is indeed the case, the physical meaning and practical role of the spatial influence weighting become difficult to interpret, since each grid point already possesses its own “true” value derived from ERA5. The authors should clarify how the spatial influence weighting operates within this framework and explain its necessity and impact when the nudging target field is already defined at every grid point.
3. For the third group of experiments, it is recommended to include additional nudging experiments using the original formulation in Eq. (2). I noticed in Fig. 3 that, particularly for temperature (c3, c4) and water vapor (d1–d4), the errors near the lower and upper boundaries of the atmosphere are noticeably larger than those in the mid-troposphere. I would like to know how the assimilation results would appear when the original vertical weighting scheme is applied, and to quantitatively assess the improvement achieved by the revised vertical weighting approach.

Minor Comments:

1. Table 1: The variables for the second group of experiments are incorrectly labeled.
2. Line 165: ERA5 data are interpolated to a 30-minute interval, while the temporal coefficients for the upper atmosphere and surface are set to 6 hours and 1 hour, respectively. What is the rationale for choosing these specific settings?
3. Line 185: Could the authors further explain why the precipitation magnitude becomes smaller after assimilating Q?
4. Figure 2b: Compared with DNDG-UV, DNDG-UVT shows reduced RMSE in LWCF, CLDTOT, and PRECT. However, why does DNDG-UVTQ exhibit a noticeably larger RMSE in these variables compared to DNDG-UVQ? Could the authors provide some discussion for this contrasting behavior?
5. Line 205: The relatively large errors in cloud, precipitation, and radiation variables may be partly attributed to the low-pass filtering nature of nudging, which can smooth out local temperature and moisture gradients. This filtering effect may be particularly detrimental to diagnostic variables with strong nonlinear characteristics, such as cloud, precipitation, and radiative fluxes, leading to amplified errors in their representation.
6. Line 210: I remain cautious about the authors’ conclusion that overly constraining thermodynamic fields necessarily damages the model’s physical consistency. Although ERA5 is designed to be as physically consistent as possible, it does not guarantee strict consistency across all scales and variables. ERA5 is still a product of data assimilation under model constraints, and different variables are not assimilated simultaneously, which can lead to residual imbalances and physical inconsistencies. Therefore, conclusions based solely on comparison with ERA5 may not be fully convincing.
7. A more reliable approach would be to conduct OSSE experiments, in which pseudo-observations are generated from EAMv2 forecasts, followed by nudging hindcast experiments and systematic evaluation. This would provide a more robust basis for assessing the physical consistency of the proposed nudging strategy.
8. Figure 3: The errors at different vertical levels should be analyzed in conjunction with the prescribed vertical weighting scheme, as discussed in Major Comment 3.
9. Section 4.2: add description of the parameter settings used for nudging the surface variables
10. Figure 9: Full-variable nudging (red line) appears to suppress the peak of the seasonal cycle of the GPI, please provide an explanation