This article provides a fairly comprehensive description of the exchange grid method used in a regional ocean-atmosphere model of the Baltic Sea region.

The exchange grid developed at GFDL is a relatively old one. But it is not widely used. And it is very little described in the literature, and in that sense the description and fairly complete analysis in this article is very welcome, with an interest that extends beyond the Baltic modellers. The analyse of the different exchange grids is interesting. The article is well suited to the scope of GMD in the model description paper category.

The flux calculator allows to take into account heterogeneity in a model grid box. It still make strong assumptions on what is homogeneous : 

1/ In this coupling scheme, there is the assumption that even above surfaces of variable temperature, the atmosphere is homogeneous, on the scale of a model grid box, from the first level of the model to the top. This is a very strong assumption, and in fact a priori false. You only have to look at the clouds above a free water-sea ice front to that at least the planetary boundary layer may have very different vertical structure within one atmosphere grid box.

2/ The transfer coefficients c_h and c_m are sent by the atmosphere (Fig. 4). They are not calculated by the flux calculator. It is quite possible in reality to have stable air above the ice and unstable air above the open water. This yield very different transfert coefficients above the two surfaces. This should lead to very different flux compare to what the flux calculator computes. Note that if the calculations of these coefficients involve iterations in the atmosphere, this can indeed be complicated to compute them in the flux calculator.

This two limitations should be explained to the reader, with if possible some descriptions of the potential impact, and technical constraints that explained the chosen algorithm.

Major concerns.

I didn't really succeeded to catch the time pattern of exchanges. Fig. 5 is supposed to address this issue, but it doesn't explain how exchanges are synchronised with models time stepping, and the frontiers of the time steps. We need a time sketch showing when the different components (models and flux calculator) are working, when they are waiting, etc ...

In Fig. 2, Fig. 3 and Fig.4, an ocean grid cell is either ice free or fully ice covered. But line 145 reads that fluxes are computed for free ocean and each ice category. Do you consider partially ice covered ocean grid boxes, with several ice categories ? Is so, please show ice fraction and ice categories (by thickness classes ?) in the figures, and give more details in the text about the flux computation over different surfaces.

Radiation is not computed by the flux calculator. But each ocean grid cell may have different albedos, especially if there is sea ice. Do all ocean grid cells receive the same short wave flux ? Or do you use the albedos of each cell and surface type (water, ice) to redistribute the solar flux ? With no redistribution, the flux can be very unrealistic (for instance very low solar flux toward ocean when the ice fraction is large, very high solar flux on ice when ice fraction is low). Please details that, and address the potential impact of your procedure. 

Minor concerns

Line 59 "by (Balaji et al. 2006)" -> "by Balaji et al. (2006)"

Line 174, 183, 188. The atmospheric wind speed is used in the bulk formulae. In Fig. 5, exchanged variables include Uair, Vair, U10m, V10m. What is the difference, and what variables are used in which flux computation ?

Line 189 : "It is noteworthy, that the horizontal velocity components of the ocean’s water body are negligible compared to the atmospheric ones and are thus omitted". That's true if the model does not resolve the tides. Is it the case ?

Line 200 : "Moreover, the presented formulas might be updated to more elaborate schemes using more sophisticated theories e.g. a TKE-based ansatz for the calculation of transfer coefficients (Doms et al., 2011)." As the transfert coefficients are computed in the atmosphere, and not in the flux calculator, it seems not feasible to implement a TKE formulation in the flux calculator. 

Line 219 : time steps are 600s and 150s for the oceanic and atmospheric model, respectively. For the atmosphere I suppose that this is the 'physics' time step, and advection is called more frequently ? A question important for the coupling is the time step of the computation of solar radiation. Is it called at the same pace than the coupling ? Or with a longer time step ?

The Exchange-grid approach for the IOW ESM by Karsten et.al. explores the usage of the exchange grid introduced by Balji et al in 2006 for proper calculation of fluxes at the interface between ocean and atmosphere. It seems to fill a gap in regional modelling by demonstrating the potential of the exchange grid in combination with flux calculation. The manuscript is well written and well structured, describes technical advances with IOW Earth system model. Apart from some minor clarifications and corrections the manuscript is in my opinion ready for publication. If the authors wish to further improve the manuscript I have made some suggestions below.

Minor clarifications needed

l 189: I am surprised that ocean velocity is not coupled. Could you give some more justification for this. Adding these should not have a big impact on the run time of the model. Could you provide some numbers to explain why the expected error is negligible.

l 197-199: It is not clear to me what is happening with the radiation fluxes w.r.t. the exchange grid. I stumbled over the phrase “do not simply depend on”. Perhaps you could rephrase this paragraph to make it better understandable.

l 220: Please clarify in the text if atmosphere coupling fields are averaged in time over one coupling interval.

Either as part of the introduction or if you decide for adding a discussion section: Could you indicate how your flux calculator approach differs from the “Flux Coupler” in CESM2.1: https://www.cesm.ucar.edu/models/cpl which is also programmed to calculate fluxes.

Technical corrections

l 63 and elsewhere in the text: suggestion to a use some other word than “automatically” or simply drop it.

l 199: only → Only

Figures

Fig 1: increase size if axis labels

Fig 4: increase font size for text in the figure

Fig 5: increase font size for text in the figure

Fig 8: increase size of figures and font size of axis labels

Fig 9: increase size of figures and font size of axis labels

Some suggestions and thoughts (optional) for a discussion section

Could “Flux Coupling” find its way into the title as both the exchange grid and the flux coupling are somewhat key here?

Couldn’t you finalise the conclusions with some strong point, e.g. what you gained and what has been significantly improved in your model? I find it a bit disappointing being told in the final message only what you would like to do in future.

I am missing some kind of discussion section where you put your new approach into some perspective:

• While the improvements you show are convincing I wonder how this relates to a model configuration where the atmosphere iand ocean are run on identical grids – at the same resolution as your ocean?
• As you couple albedo how would small scale processes, in particular clouds, would change the game if better resolved with a higher horizontal resolution of the atmosphere model. On the coarse atmosphere grid some horizontal averaging is done by construction.
• How would implicit coupling change the game. See e.g. Kang et al., 2021: Mass-conserving implicit–explicit methods for coupled compressible Navier–Stokes equations. https://doi.org/10.1016/j.cma.2021.113988 or Balaji et al., 2016: Coarse-grained component concurrency in Earth system modeling: parallelizing atmospheric radiative transfer in the GFDL AM3 model using the Flexible Modeling System coupling framework. https://doi.org/10.5194/gmd-9-3605-2016
• How does your approach relate to the one in the Bergen Model described by Furevik et al, 2003: Description and evaluation of the Bergen climate model: ARPEGE coupled with MICOM. https://doi.org/10.1007/s00382-003-0317-5