I find the approach proposed in the manuscript to be of interest for the journal. It follows the now discussed road of separation of concerns, whereby the code part dealing with numerical algorithms and the part dealing with the infrastructure (mesh, parallelization) are separated and treated in different ways. However, the manuscript in its present form misses the goal: one expects that the material of the manuscript describing a modeling approach will be sufficient for a reader to learn how to use the approach. I do not see that this goal is reached.

The text on lines 165 - 190 shortly describes how the JASMIN is involved, but I doubt a reader can get any understanding of what and why is done. Moreover, it is not at all clear how to use JASMIN in conjunction with the updated code. How the JASMIN environment can be installed, how code is compiled, etc. The description should be essentially extended and be such that those who are willing to follow author's approach can do it. 

The manuscript devotes more than a half of its volume to the description of simple test configuration, going into too much detail, which is hardly optimal. The test case remains the test case, and one can only learn that the approach proposed by the authors is working, yet not without drawbacks related to one-way nesting (the development of errors on fine-coarse boundary). I do not think that this test case is well suited to demonstrate the need of adaptivity. Figure 11 shows clearly that small-scale turbulence occupies 3/4 domain on full mesh, and it occupies only pieces where the resolution is refined in b and c. It is different from the initial phases in Fig. 9 and 10, but 5 or 20 days is a too short time for turbulence to equilibrate, and this transient phase is of no interest (it depends on coarse initial conditions, and does not model any reality). So the conclusion here is that dynamic adaptivity is an interesting, but perhaps not very needed possibility as concerns eddying flows. Static refinement might be doing the work, and one will take a decision where to resolve based on one interest. I foresee, however, one direction, where dynamic refinement still might be of interest -- the simulations of seasonal course of variability. Submesoscale eddies might be suppressed in warm seasons, and a coarser mesh will be sufficient for mesoscale. My recommendation are to make the experimental part more compact. One can hardly learn anything from detailed description of particular eddy features or the comparison of transects (Fig. 13, 14), and there is very little sense in Fig. 9 and 10. 

Please also check the text: there are numerous cases when plural/singular and some terms (e.g. kinematics)  are used inappropriately.

The manuscript by Zhang et al describes an implementation of a nested regional ocean modeling capability using the parallel computing framework “J Parallel Adaptive Structure Mesh Applications Infrastructure” (JASMIN). They demonstrate the application of the implementation with simulations in an idealized basin double gyre configuration. The presentation is generally clear and the authors provide a realistic assessment of the remaining challenges with this particular implementation as well as for nested ocean modeling generally. My only significant concern with the study is that the entire work is based on a proprietary software stack, making the reproducibility and accessibility of the effort to the broader community questionable.

I suggest below a few additional points and questions that I believe would be helpful to a reader. The grammatical construction (e.g.  singular/plural usage) of the text is a little rough in places and could use some polishing by a technical editor. I have no substantial concerns with the manuscript as written and recommend publication after minor revision. 

Line 31: “Other ensuing practice …” ?? Not right words here

Line 37: “In the following up part of the paper …” -> In this paper ….

Line 58: suite -> suit

Line 70 ASMIN -> JASMIN

Line 114-119: It would be helpful to actually state the man-months of effort required to complete the refactorization.

Line 128: 422 patches … Might be helpful to state where this number comes from .. (# of variables) * (# of grid hierarchy levels) * (???)

Line 131, 133, 175 : Figure 3 should be Figure 1 or Listing 1?

Line 180: does level here refer to depth level or grid hierarchy level?

Line 295, Figure 5. Label the figures with the case name or include identifiers in the caption.

Line 296: “Besides, L-M-I covering more area in the subtropical gyre “ This seems backwards from the figure. Poor sentence construction.

Section 3.3: A key conclusion of this section is that the mis-representation of APE->KE transfer in the eddy-parameterized parent-grid leads to boundary forcing biases in the child grid. Later in the paper, the authors allude to the need for a GM-type parameterizations in this regime. Does the implementation support different parameterization methods at different levels of the hierarchy (they show that different parameter values can be used)? An additional case with GM turned on in the 0.5 degree grid would be informative of the suitability of this approach with a scale aware parameterization strategy.

Line 346: “proxies” Not sure this is the correct word here. I believe you mean something more like “metrics of grid quality”

Line 398: bu -> by

Line 423: LaTex typo for circle symbol

Line 424 : “following up part” -> following partLine 442: We pick the a -> We pick a 

Line 474: This is indicated by that both … -> This is indicated by the fact that both …

Line 505-514: Could alternatively provide an estimate of the man-months of effort here.

Line 529: (Section on refinement criteria) : Some description of how the current implenmexation aggregates points into individual patches would be helpful.

Line 597 ( Section on realistic cases): Some discussion of challenges expected with realistic topography and coastlines would be helpful. Would topography or the coastline be refined along with the grid? How would mismatches in land-ocean boundaries across hierarchy levels at the lateral boundary be handled?

Other: I presume some data on scalability and performance will be presented in the companion manuscript, but a very brief statement about this would be helpful to make the present manuscript self-contained.