In the manuscript titled “The DOE E3SM Version 2.1: Overview and Assessment of the Impacts of Parameterized Ocean Submesoscales”, the authors develop the version 2.1 of U.S. Department of Energy’s Energy Exascale Earth System Model (E3SM) with new updates, by adding Fox-Kemper mixed layer eddy parameterization. The improved eddy parameterization simulates the submesoscale instabilities, especially the mixed-layer eddies slump the fronts. As it is triggered by the mechanism of amplitude of lateral density fronts in the weakly stratified surface mixed layer. After adding the mixed layer eddy parameterization, the simulation for energy exchanges, radiation heat transfer, and adiabatic motion of dry and moist air in the mixed layer are improved.

General comments:

This paper is comprehensive and well-supported by evidence, providing an accurate assessment of the effectiveness of mixed layer eddy parameterization in global ocean model of E3SMv2. It presents significant advancements relevant to optimizing the overturning streamfunction. This article is well-written, logically coherent, and structurally sound. Therefore, the following discussion focuses only on specific issues and does not affect the overall scientific validity of the paper. For these reasons, I believe that the manuscript can be accepted for publication by the GMD after minor revision. Below, I have some specific suggestions for the authors.

Specific comments:

• Line #307: The link of a full list of all code changes made from v2 to v2.1 should be updated, the hyperlink in the PDF does not direct to the correct section. After manually copying the link, there are only two tags, the Atmosphere and Land tags on Github. There is no way to find the corresponding code based on Appendix B, such as River, Sea Ice etc. Reading Fortran code directly is something that often happens when running meteorological models. It is recommended to update the tags or provide an index, then reader could find scripts after reading Appendix B. Additionally, is the repository only open to project developers, or does the entire Github community have permission to modify the code? It seems visitors can leave comments on each line of code.

• To prevent readers from constantly switching between this paper and Bodner et al., 2023 to reference the key equations in this study, it is recommended to further explain these two variables. The following section is a reference. In equation (1), the physical meaning of the variable Ψ is streamfunction, the gradient of the streamfunction can indicate the bolus velocity. μ(z) vertical fluxes at height z. Equation 2 is vertical structure function, so the vertical fluxes will be vanished when height is below the mixed layer (z< -H), and approach zero when z is almost zero.

• 1, This SSH result is contrary to expectations. Is it possible that there is an issue with reduced deep water formation in the North Atlantic that should feed into the AMOC process (line #134)? Or air-sea interaction? Overlapping lines make it difficult to determine how many lines there are. It is recommended to use different symbols instead and slightly separate the lines in x-axis at overlapping positions to show the location of each point and indicate this in the figure caption.

• 4, if the discussed region, "western boundary current in the North Atlantic around the region of the RAPID array," and “the west transect of the Overturning in the Subpolar North Atlantic Program array” are highlighted with a box in the figure, they would be easier to understand. If the decrease in MLD is caused by a very small increase in the magnitude and extent of the northward limb, then what is the direct cause of the "very small increase"? Is it because the mixed layer eddy parameterization enhanced the bolus velocity of eddies? In E3SM version 2, is the bolus velocity represented by a constant? Could the authors explain this in more detail?

• 6, this is just a suggestion: could the shaded areas in CMIP6 be slightly enhanced, for example by reducing transparency or using borders? The colors are too faint and hard to distinguish. This requires locating the Python script used for plotting in PMP. If it is too difficult, the authors may decide to update it or not.

• Line #163, what could be the possible reason for the lack of significant changes in large-scale extratropical modes here? In the discussion section (line #301), the reason is also not explained here.

• 14 and 15, it seems that a strong stratification buffer has been enhanced and interior stratification has been eroded in the v.2.1 model compared with v2. Is this solely due to the addition of the mixed layer eddy parameterization?

"The DOE E3SM Version 2.1: Overview and Assessment of the Impacts of Parameterized Ocean Submesoscales" has two aspects: it is the report detailing the new version of this model (highly appropriate for GMD) and it is a scientific report on the consequences of including the 2011 version of the MLE parameterization into this model.  The latter is not novel in application, but instead compares the sensitivity of the E3SM vs. that of other models incorporating this scheme. The basic result is that upper ocean biases and RMSEs are improved, which then improves many climate metrics. I have only minor suggestions for the authors to consider.

1) Using a constant L_f is not a common choice, although it is not unprecedented.  The authors should review the results of extensive testing of different choices of L_f as constants and as variations with mixed layer deformation radius in Calvert et al. (http://dx.doi.org/10.1016/j.ocemod.2020.101678)

2) In Fig. 4, which MLD observations are used?  Are they consistent with the definition of MLD used?  The new SEANOE MLD product is very nice in its incorporation of much more Argo data than previous versions (http://dx.doi.org/10.5194/gmd-16-3849-2023, https://www.seanoe.org/data/00806/91774/).

3) I recognize that there are many centuries of simulation data reported here, but I wonder the extent to which the MLE parameterization is responsible for these changes as opposed to the many other changes made.  Are there any runs available for comparison with the v2.1 that differ only in turning off the MLE parameterization?