In general this is an interesting topic and a very valuable effort which I think warrant publication. However, the organization and writing of the paper is quite confusing and could really benefit from a good overhaul, making the paper and the code more useful and reproducible. There is also a lot of discussion on the results, but since this is a model description paper, I would have liked a lot more discussion on the modelling choices. Though some emphasis is put on the limitation of process parametrization and a global approach, the model is still quite complex and little to no time is spent discussing the additional value of this approach against something of lower complexity and with faster computational time. I would very much like to see some discussion, both of how much compute time this module adds to a typical MAGICC run, if some of the complexity could have been shaved off (maybe this is negligible and totally worth it, but from the paper I have no idea). Here is a list of additional questions that I feel should be addressed in a discussion section:

1. The models being emulated show a big spread in their predictions and there is uncertain observational data. Is this the right setting and time to make an emulator of this kind? (The answer might be that this is the perfect time to have such an emulator to explore and push for better data, and be ready.)

2. The model though simple in a way, is really quite complex with a really large number of parameters. The fact that it can emulate model behaviour relatively well using so many parameters is not very surprising. I think a more important question is whether there is enough input data to constrain such a large parameter space.

3. Context. I imagine this will be used as part of MAGICCs default setup, or at least a fairly accessible version of it. How does this number of parameters add to the current number of free MAGICC parameters? Can they be fit in an online fit? Do they affect other MAGICC parameters? How easy is it to fit MAGICC parameter ensembles with this? What does it add to computational time for the full MAGICC? Etc. I see that you plan to discuss some of these issues in a different paper, but I expect at least some discussion of this here. Also can this module easily be coupled to other simple models? Perhaps the offline use demonstrated in this paper is just as interesting as online use?

One more overall point before going over the paper from top to bottom: The structure and "story" of the paper is confusing to me. I often don't know where I am and feel like I'm scrambling for an overview when reading it. I think this can be solved without too much work using the following few principles: 1. Spoilers are great. Tell me what is going to happen and give me more of an overview on top, and on top of every section and subsection. Forcing yourself to write such miniature summaries might also help you understand how you've structured your text and see whether it makes sense. Sometimes it might not... 2. Tables are great, long lists in sentences are less so. Tables of models used, input data used to calibrate, parameters to be calibrated, experiments used etc. Also, there are so many variable abbreviations that a dictionary in the supplement would actually be useful. 3. What does MAGICC mean? It is highly unclear when you mean the full model, just this coupled carbon-nitrogen module, the new version of MAGICC the old version of MAGICC? This confusion is present in nearly every reference to the model. Please find a way to distinguish between these three things (MAGICC old version, MAGICC new version and the Coupled Carbon-Nitrogen Cycle model (MAGICC-CCNC?))

Specific comments from text:

Title: Is the version number correct? Is this version number for an upcoming python version of MAGICC? Isn't MAGICC on a much higher version number. This is already confusing, and it shouldn't be.

Line 68: "Section 2 presents a detailed descritpion ... in MAGICC". The new version of MAGICC? Also, as far as I can understand it is only the land carbon-nitrogen cycle and carbon-coupling which is described and not how they actually feed in to the wider MAGICC code, though I'd prefer the text in the section to change to reflect that rather than this sentence.

Line 72: "In future work...". Is this specific planned work or just an aspirational statement? Both are fine, but clarity is preferable.

Line 75: Add a sentence or two on what the model description will involve. This is a great opportunity to prepare the reader for the overall model structure. Mass balance equations including pools for plants, soil litter and mineral nitrogen with feedbacks from temperature and CO2 etc...

Line 75/ section 2 overall: The Figure 1 Flowchart shoud appear much earlier, it should be explained, and it should be annotated with equations and sections. It should also be much clearer how the code flows through the various equations. Does each section refer to a method or function or are they not built that way? This is a model description paper, I expect to get some idea about this from reading it.

Line 76: I would like to see a flowchart of the workings of MAGICC which shows me where the new coupled carbon-nitrogen model fits in. Also this section does very little in the way of giving me an overview of the workings of MAGICC, a sentence or two to explain a flowchart would really improve this.

Line 85: "intial design..." give some reference to equations and sections coming up that the describe the design that you landed on, if I want to flip over and have a look from here. Also since this is an update to MAGICC, what did MAGICC have to treat this before? A sentence on that would be helpful before the initial design of this model.

Lines 95-110: Please be much more explicit about which equation or equations you are explaining when. Also referring to the different parts of the flowchart of figure 1 might be helpful. Again I'd very much like the supplement to include a vocabulary for reference as the amount of shorthand used is daunting. Also giving more overview would be helpful. For instance is the sentence from line 99 to 100 describing equation 4? Is equation 4 a sum of equations 1 to 3? Is equation 9 the sum of equations 5-8? Is the sentence from line 104 to 105 explaining equation 9?

Lines 110-123: Can the equations have an explanatory text to the side which would make them easier to come back to and review, such as Plant, Litter, Soil, Atmosphere to land?

Lines 162-165: In my opinion this should be a single equation with two domains or something like that \epsilon_CO2 = { and then two lines with different conditions. I find that easier to read.

Line 172: That is a mouthful... A table of free parameters with a longer descriptive name/ explanation in the supplement would be a really good thing. Perhaps with possible ranges and ranges in the calibration set? Mean + std for ESMs and actual values in the OCN and CABLE calibrations?

Line 226-228: I would like a reference either external or internal (i.e. to a specific upcoming section/figure) for this statement.

Lines 264-265: I assume recurring parameters are the same as presented before, but if there was a supplemental table to peruse, that would be much easier to understand.

Lines 277-282: To me these look like they are more or less the same equation. That could be used to explain it in a schematic way as in equation 37, and these individual equations could be moved to the supplement. Even equation 36 is the same equation, only it doesn't include a carbon-nitrogen coupling because there is no corresponding carbon mineral pool. This also becomes such a long list of independent parameters, where a table would be useful, and I would like a discussion on whether so many separate parameters can be meaningfully calibrated from the available data (maybe they can and it's no big deal, then tell me).

Line 289: "PU and atmospheric deposition" this statement mixes a shorthand and a full writing out of something that has a similar shorthand. Please do one or the other, at least in a single term like this. With a supplementary vocabulary, I think using the shorthands systematically would make a lot of sense. When you mix terms like this I also get confused on how to interpret "atmospheric deposition", does it now mean something else than the "atmospheric deposition = AD" shorthand? What would that be?

 Figure 1: I'd like to reiterate that this figure should come sooner, and that the caption, and possibly the arrows themselves could do with references to sections and equations that they depict.

Line 352: Before you start the subsections, give an overview of the calibration process and datasets involved so I know what I am getting into in this section.

Lines 355-360:  Are these datasets publicly available? Where can I get them? If I wanted to reproduce your paper how would I do that? This should be very clear either here or in the data availability statement. I find it in neither.

Line 364-363: "as results from experiments without the nitrogen effect are unavailable". Is this a wish for future experiments? How useful would these be? How could they aid in work similar to this? I think this might be worth revisiting in the outlook section.

Line 369-370: I suspect the reference is not in the right place here "calibration in this paper" refers to the current paper I am reviewing while "the original paper" is Meyerholt et al 2020. Please move it if my assumption is correct.

Line 381-384: This long listing of models, land models and references are perfect for a table. They are considerable less perfect for this sentence format. 

Line 392-417: This section left me with a lot of questions: 1. What experiments are used for the calibration? All? Or do you reserve some for testing? Which parameters are you fitting and which variables are you fitting to (I'd like a table or two for that in the supplement)? How do you weight the different variables when you fit? Do you fit entire timeseries? Again how do you measure the fitness of a run? Do you scale them? You say offline calibration, please define? In fact, is anything in this paper online? I could not reproduce your procedure from this text.

Line 403: I assume "imputed" should be "inputed"

Line 412-416: These three sentences read to me like a three step process. Numbering it as such might aid the understanding.

Line 419- 463: I am very confused by this section. It reads mainly as a discussion of differences between CABLE and OCN with occasional references to MAGICC (which MAGICC, BTW? the offline CCNC I presume, but I would bet on it...) being able to capture them. If this is calibration results maybe this makes sense, but then the headline should reflect that. Also some summary table of how well the fits are doing would make sense to have here. It is also not clear to me whether you've used the same calibration for all experiments (I think so, but I shouldn't have to figure that out from gathering scattered clues in the text...)

Lines 472- 528: This section too seems to be about calibration results rather than the detailing the calibration. I also expect some sort of error or performance summaries. I am also still confused about what MAGICC we are talking about when and whether the results are all "offline" (whatever that means) and if so why no "online" results are included. Maybe they wouldn't change anything? Maybe it's not ready for that? I'm fine with either, but make it clear.

Lines 485-500: This discussion on mismatching is interesting, but I'd wish you'd take the discussion a bit further, to raise questions like: Does this mean you are perhaps fitting to the "wrong" variables? Are the data/observations/models to uncertain for the type of exercise you've done here (that doesn't mean what has been done isn't very useful, in fact in my opinion it might make it all the more interesting in fact)? Is your underlaying parametrization what has an issue? Also, some of this could have been discussed before the calibration and when you did your data selection.

Lines 530-536: There is so much going on in the axis here that I had trouble initially understanding the x- and y-axis setup here. Please make that clearer also in the caption.

Lines 534-535: "Diagonal dashed lines represented points where the emulation equals the target". The use of the past tense here is confusing, maybe also rewrite overall "Diagonal dashed lines represent the line where model and emulation are the same" or something like that.

Line 437: Is this discussion or results? What are we discussing? Please give me a few sentences of summary of what discussion are upcoming, and maybe think about whether they are really discussions or just results. (Results are fine too, but they shouldn't be called discussions)

Line 539: "remain considerably different" from what? Each other? Reality? The emulation here?

Lines 549-551: I smell a table here... Also are these model means and spread? Again, this could be more easily communicated in a table.

Line 559: I think you should probably drop "is" here.

Line 589: This would be a great please to discuss whether the model presented here is perhaps too complex. Maybe it isn't, but you've done nothing to convince me.

Line 620-621: Is this emulation also consistent with this for CLM?

Line 590-636: This whole discussion is interesting, but it is not entirely clear whether the results discussed includes information gleamed specifically from the emulation or whether it is just model comparison which you could do without it. If it is some mix of those it should be clearer what insights have actually arisen from the emulation.

Line 691: I want a summary of what's to come... Somewhere here I also want a real and open discussion on whether this level of complexity is the right one. How important is it versus how expensive is it? How well do we believe in the tuned parameters? Are there too many tuned parameters? Are some of the parameters very constrained overall (i.e. maybe they don't need to be free)? Are some all over the place (completely unconstrained)?  Also I'm not even sure if this model needs specific inputs or if it can be run online with MAGICC for something like an AR6 scenario database member? Has it been tested with online MAGICC? How fast is it? How compute intensive was the calibration? For a future online run with this for impacts, what would you recommend calibrating to? All of these are questions that you don't have to have an answer to, but I expect you to acknowledge them, and say something about how they may be addressed in the future.

Line 748: Will this  prescribing of biological nitrogen fixation make the model less easy to run "in the wild" so to speak, with just scenario information online or offline?

Figure A1: Maybe state explicitly that this figure is like figure 2 and what the difference between them are. Also the plots here are so close that ylabels go into each other making them hard to read.

Figure A4: Again maybe like figure 2 but with differences blahblah..

Lines 859-887: Text A1 - It is entirely unclear to me what this section has to do with this model and its calibration.

Line 879: indented -> intended

Line 904: model's or models'? (It says the former, I think maybe you mean the latter?)

Line 925-928: Be more clear and explicit here, especially regarding the CABLE and OCN datasets.

I have had a look at the code, though not in great detail. In an ideal world, I'd like to see it as an importable library function, but I guess the reason why it isn't is that it will be part of a not yet publicly available MAGICC codebase. Maybe say that explicitly if that is the case. In an ideal world I'd also like more function doc-strings.

All comments are provided in the attached pdf.

See attachment