This paper describes the development of a new version of the UK global ocean and sea-ice modelling system based on the NEMO ocean model. The paper is very comprehensive in the details on what has been changed and goes into details with some of the choices made and the rationale behind them.

The paper is very well written and has a nice level of details without it being too much. Besides a few minor comments below I think that it is very close to being publishable.

Minor comments:

Line 28: Not sure why Earth is not earth.

Line 143: The 2*10 does not make sense to me. Is the exponent missing?

Line 205: It would be good to have more details on the machine used here since the speed of the cores will depend on which CPU’s is installed.

Line 268: The figure reference is missing. I suspect that this is the figure in the supplement.

Line 440: Westernn should be western.

Figure 13: This figure is very fuzzy when I print it and the difficult to read. The caption only mentions the sea ice area (top 2 plots) and not the sea ice volume (bottom 2 plots). There are also no references to the PIOMAS used for comparison of sea ice volume.

Line 544: I don’t understand the 3 after chlorophyll.

Line 544: Likewise, the 14 after GOSI9.

Review of “GOSI9: UK Global Ocean and Sea Ice configurations” by Guiavarac’h et al. 

Recommendation: Minor revisions

Summary and overall impression: 

The manuscript presents a new version of the UK global ocean and sea ice model. As with the last version, the same model code is used to produce three model configurations which only vary in horizontal resolution and a few parameters that scale with resolution (time step, diffusivity etc). 

The main difference from the previous version (GO6) is the upgrade of the NEMO model and the sea-ice code, and unifying the ocean and sea-ice configurations. 

Sections 1 and 2 are excellent introductions and descriptions of the model and each parameter choice is very well motivated. The later sections show the results from the forced model integrations which are also compared to results from the older version, GO6. There are also some sensitivity analysis of the impact of each parameter choice. Put together, the improvements from GO6 to GOSI9 are shown in a clear way and can usually be associated with a particular change in the code or parameters. It seems that the switch from 2nd order to 4th order advection is responsible for a large part of the improvements. 

The authors mostly discuss the global oceans but dive into the details for the Southern Ocean, the North Atlantic Ocean and the North Pacific Ocean in some more detail. Each of these seems to have been the focus of a smaller team among the authors. 

Overall, this is a very well written manuscript where motivation, results and conclusions are clearly presented. I recommend the manuscript for publication after the authors have considered the minor comments below. 

Minor comments: 

Line 87: The bathymetries have been derived in very different ways, using different sources and post processing methods. A motivation for this is needed. For example, why was the ORCA025 bathymetry smoothed but not the others? Why could the authors not derive the ORCA12 bathymetry and then coarse-grain it down to ORCA025 and ORCA1? 

Table 1: I understand that one does not simply give NEMO a viscosity anymore but rather a velocity and length scale from which NEMO will compute a viscosity. But this Table makes it very hard to compare the values between configurations and to GO6 and GC3. I strongly recommend the authors to present the time step, lateral diffusivity, and lateral viscosity for the 1, 1/4, and 1/12 configurations in GO6 and GOSI9. This was done in Storkey et al. 2018. Since parameters vary horizontally, you could choose the value at the equator or some other reference latitude. 

Line 178: The ocean will call SI3 each time step, meaning that the ice model time step is the same as the ocean model. Why was this choice made? One could also use the same ice-model time step for all configurations, which would likely make the higher resolutions (1/4, 1/12) faster, but perhaps that violates the CFL criterion in SI3? 

Line 205: The purpose of the upgrade GO6 -> GOSI9 seems to have been two-fold: 1) Make a new model with smaller biases and 2) make a model that is faster. The authors discuss (1) a lot, but leave (2) out almost entirely. I would like to see a Table with the throughput and approximate cost for each model configuration, i.e. simulated years per day and core-hours per simulated year. If the numbers are also available for GO6, then we can judge if GOSI9 is faster as well as better, or if it’s just better but not faster. 

Line 207: One key to NEMO throughput can sometimes be the XIOS output server, which I am sure the authors have used, and I would guess that some effort went in to finding optimal settings to maximise throughout or minimise cost for 1/4 and 1/12 models. Some mention of this work would be of great interest to the community here. 

Line 247: “partial slip condition” is written twice

Line 268: “Figure ??” refers to Figure S1. 

Line 345: There seems to be a big change in the temperature due to the upgrade to NEMO 4.0. I understand its near impossible to know exactly which part of the upgrade caused this, but some speculation would be welcome. Did the formulation of advection or diffusion change? Or is it mostly the upgrade from CICE to SI3?

Line 390: There seem to be a lot of changes in MLD but they are hard to see in Fig 12. I would recommend the authors to show zonal-mean MLD in the supplement. Perhaps that would reveal the shift around 40°S and the North Atlantic better?

Line 405: The improvement in Antarctic sea ice (Fig 13) is very modest and within natural variability. Storkey et al. 2018 similarly showed very little change in ice extent but much more in ice volume. Is there a difference in ice volume between GO6 and GOSI9? 

Fig 13: This figure was hard to see on a printed A4. Suggest to make this larger when its time for publication! 

Fig 16d: This figure is never referenced to, but I think it should be. ORCA1 produces a reasonable MLD while the higher resolutions overdo it. This is not uncommon in NEMO. Are 1/4 and 1/12 in GOSI9 better than in GO6, i.e. is the Labrador MLD less excessive? And what causes this bias? The convection becomes less excessive in 1/20° (Biastoch et al. 2021, doi: 10.5194/os-17-1177-2021, Fig 9 b,c) so perhaps its just a matter of having enough horizontal resolution to produce eddies in the Lab Sea? On the other hand, both 1/4 and 1/12 in GOSI9 include some scale-aware GM which one would hope reduces this problem a bit. The authors say that the GM had some impact on the NAC position. Have they looked into whether the inclusion of GM in 1/4 or 1/12 reduced the Lab Sea MLD? 

Line 472: “Compared to GO6” but we don’t know what the position was for GO6. A number in the text or horizontal line in Fig 16 would help. 

Line 540 and Line 543: Figure references are wrong. 

Code availability: The zenodo repository with namelists etc is zip file which I was unable to open on my Mac and a Windows machine. I suspect the file is broken.