This is a report on an ECMWF contribution to the nextGEMS project consisting of a 4.4 km and 2.8 km version of IFS coupled to two eddy-permitting ocean models, NEMO and FESOM. Accurate coupled 5-year simulations, especially with IFS-FESOM, are a nice achievement on their own and a step toward the nextGEMS goal of multidecadal km-scale ocean-coupled simulations using seamless models that can skillfully span weather and climate timescales. A variety of improvements over ECMWF’s operational IFS-NEMO model (9 km atm/0.25 degree ocean) are implemented, including global water and energy fixers, cloud tuning for radiation balance, a scale-aware parameterization of convective mass flux, ocean cooling from snow melt, Antarctic meltwater runoff, better land, urban and snow schemes, etc., in combination producing an impressively realistic climate simulation.  Some phenomena that require km-scale to simulate are highlighted, including atmosphere-ocean interaction around leads, urban heat islands, and a QBO of realistic period driven primarily by resolved-scale gravity waves.

Overall, the paper is well written and the results are state-of-the-art. I appreciated the illuminating discussion of the model development process, proceeding through three major iterations.  It would have been nice to see a bit more analysis of the simulated ocean state, including mesoscale eddy statistics and vertical structure.  The 2m temperature bias over ocean in the 5th year of the IFS/FESOM simulation looks remarkably small in Fig. 10, which is encouraging, but were other characteristics such as eddy kinetic energy, mean currents, thermocline depth, etc. examined?  These characteristics do evolve on sub 5-year timescales and are relevant to the 30-year performance of a coupled model, so they seem in scope for this paper. 

It is sometimes claimed that km-scale climate modeling is fundamentally simpler than 100 km grid climate modeling due to less reliance on poorly constrained parameterizations of subgrid variability, e.g. orographic gravity waves/drag, deep convection, subgrid cloud heterogeneity, etc.  The IFS experience seems to be that this is only partly true - the simulations are best if those subgrid parameterizations are still active at this scale, even though resolved-scale motions are doing more of the work.  This point might be made a bit more clearly in the conclusion to the paper.

Introduction

The authors document the development of a new IFS-FESOM atmosphere-ocean coupled simulation, alongside discussion of the same atmosphere model coupled with the NEMO ocean model. The novel element of wide interest is discussion of the experiences of running this system at km-scale resolution for global domains over multi-annual period. This represents a substantial achievement, at the forefront of weather and climate model development, and therefore the work merits publication.

Given the significance of this development, however, I would recommend that the manuscript should be revised prior to acceptance. In the view of this reviewer, the resulting paper would have considerably more impact, if the following general remarks can be considered and addressed.

=====

A) Paper length:

The manuscript represents a considerable development (as illustrated by substantial number of authors in the team), with multi-component codes brought together within a common modelling framework alongside supporting infrastructure. Even accounting for this, the paper is overly-long in my view, running to 59 pages in draft form with 16 main Figures and a further 5 Supplementary Figures within Appendices A-E. Whilst it is valuable to capture many of the details, a more rigorous review in general across all sections to assess where extraneous details are described might encourage wider readability. Some more specific suggestions are captured in remainder of this review, but I also suggest the authors should identify any opportunities to further streamline the text (including use of more summary Tables where relevant), reduce number of Figures etc.

====

B) Paper focus:

Given the range of simulation experiments conducted (Table 1), it is difficult to maintain a clear sense of the key research focus of the paper. For example, the abstract highlights a “focus mainly on….IFS-FESOM…with atmospheric resolution of 2.8km and 4.4km”, while much of the analysis (e.g. Fig 5, 6, 9, 10, 11, 12, 13, 14, 15?, 16) seemed to focus more on comparisons between 4.4km and 9km systems, and at times more on comparisons between IFS-NEMO and IFS-FESOM. Greater clarity is therefore required as to the key focus for the documentation here. There is merit for GMD in focus on:

1. discussion of system innovations and improvements between Cycle1, Cycle2 and Cycle3 for a particular configuration,
2. discussion of the sensitivity of simulations to model resolution (across 9km, 4.4km and 2.8km if a common model framework can be used for this, or focus more deliberately on 9km vs 4.4km if simulation data better suited to this),
3. discussion of the impact of changing ocean model between FESOM and NEMO.

At present the paper attempts to capture certain aspects of all 3 approaches highlighted above, but this arguably makes for a relatively confused (and long) narrative for readers.

====

C) Paper structure:

Linked to comment B, the paper structure and sign-posting would merit a review and improvement by authors.

Introduction of Table 1 (and Section 2.4 in general) earlier in the discussion of Section 2 might both help to reduce the overall length and make the discussion easier to follow. Section 2.1 is then rather exhaustive when not yet set in context of the nextGEMS runs, including perhaps some discussion on operational IFS which could be omitted where/if not directly related to the results presented (e.g. introduction of IFS-NEMO with ORCA12 as specific but not only example). Given its importance to help orientate readers on the scope of the following discussion, I suggest improving Table 1 to include more details as easy reference, including some of the text from caption (e.g. translation of atmosphere resolution) into the main table, along with any other relevant details from the text.

Section 3 highlights specific model developments within Cycles, alongside a discussion of the impact on results. This was not always straightforward to follow given the variety of topics discussed, so at least improved sign-posting and sub-sectioning would be useful (e.g. distinction between sub-parts of 3.1.1 and other discussion under 3.1.2+ not clear). Results presented then begin to introduce elements and discussion of the comparisons across development Cycle as well as to model resolution and model components highlighted under my comment B). In general, I would suggest that Section 3 and thereby also Section 4 be reorganized to document more cleanly between these 3 aspects - or indeed be clearer in focus if only 1 or 2 of these considerations are to be discussed here.

Given concerns over paper length and clarity, I ask the authors to consider and provide a clear defence of the value of keeping discussion of Cycle 1 results within scope of this paper, in preference to a cleaner focus on Cycle 2 vs Cycle 3 developments here. Have Cycle1 to Cycle2 improvements been documented in other papers to date? 

In general for all Figures from Fig. 3 onwards, a different blend of simulation results are plotted, but not always as a full consistent set, and thereby at times highlighting impact of Cycle changes, sometimes resolution, sometimes system differences, and sometimes a combination of some of these. Figures 6 and 7 for example, and accompanying discussion in 3.1.4 would sit better as a discussion of model characteristics for precipitation as a key variable of interest and sensitivity to treatment of convection at 9km and 4.4km than it being somewhat lost among other parts of system development discussion. In passing, I will note again that Figure 6 is focussed on 4.4km vs 9km IFS-NEMO (rather than 4.4km and 2.8km IFS-FESOM as per abstract stated focus?).

By Figure 9, the results in Section 3 appear less concerned with impacts of Cycle developments, but back to discussion of comparative impact of ocean model and/or model resolutions.

I suggest that with some reorganisation, the “selected examples” documented in Section 4 might then appear to be less standalone collection of somewhat ad-hoc topics, with sections 4.2 and 4.3 relatively weaker in terms of depth of analysis presented (and again reflecting some blend between a discussion of Cycle impacts and model resolution and codes, rather than being consistently focused on one specific strand). As highlighted by Reviewer 1, the lack of more detailed discussion of Ocean results is an omission, but given my concerns over paper length and focus, I might recommend being more deliberately focussed on only some aspects of atmosphere model performance here for example (i.e. Section 4.2 is rather ‘random’ in context of rest of the discussion), and encourage discussion of ocean and sea ice aspects to a second accompanying paper for example?.  

====

D) Figure quality:

In general the paper is well written, and Figures of a good quality for publication. However, I would also suggest to ensure:

1. Can any figures be provided with fewer simulations included (e.g. if discussion relates more specifically to impact of model Cycle, is this more clearly identifiable with fewer lines – see Figure 4 for example)
2. Aim for more consistent inclusion/focus on certain experiments (e.g. choice of IFS-NEMO in Fig 6 rather than IFS-FESOM seemed somewhat arbitrary?).
3. Figure 7 – this is an important plot in context of km-scale global model development, but the detail is a bit lost with smaller sub-panels and number of lines. What is the key sensitivity to be drawn out here? 6 model experiments are listed in the legend for example, but I’m not sure I can differentiate more than 5 lines (maybe Cycle 3 9km IFS-NEMO and IFS-FESOM overlapping??).
4. Figure 8 – similar comment on difficulty capturing results beyond “all in the bunch”. Suggest plotting as differences relative to GPM rather than absolute zonal average precipitation may be preferable here to focus on sensitivities.
5. Keeping consistency in sub-panel layout will help readability (e.g. a = IFS 9km, b = IFS 4.4km in Fig 9, but opposite way round by Fig 10). Similarly, keeping consistency of model framework – Fig 9 and Fig 10 both representing comparison of results with both changed resolution and changed ocean model – is it ocean model or atmosphere resolution that dictates here, and if known therefore present the direct comparison.
6. Figure 15 – I personally found the ‘hatching’ across land areas more confusing/distracting than helpful on first reading. Combining this discussion more directly with/into 3.2.3 would help make the paper less disjointed.

Finally, Figure 16 provides a cleaner comparison of impact of Cycle change with common model resolution and framework (so arguably more of a ‘Section 3’ item in the current paper structure?). Are the authors able to clarify (isolate) whether the differences shown are a function of the change in LULC, or the addition of the urban scheme, or some useful combination of both changes? Given the substantial change between panels c) and d), it is tempting to think it is mostly the LULC update that drives the improvement?

=====

Summary

In summary, this is a substantial and noteworthy paper documenting considerable technical and scientific progress. However, in its current form, I do not think the paper is reporting that achievement as well as it could, and hope the reflection above highlights some areas that might support further improvement prior to publication.

I would be happy to provide further review if useful, including potential for more specific review where a shorter and more focussed paper might make this more tractable.

This manuscript provides an overview of multi-year km-scale global coupled simulations using the IFS coupled to two different ocean-sea ice models. In addition to a description of the various model components, the manuscript highlights the existing issues in the previous version of the model and the major model developments that address those issues for the new version of the model. In addition to highlighting the improvements in this version of the model, the authors also point out existing biases, which are just as valuable to document and share with the community in this burgeoning area of km-scale coupled simulations. The authors conclude the manuscript by providing the context in which the development of these models exist and how such high resolution climate simulations open the avenue for 

These 5-year coupled simulations with two different ocean-sea- ice models at these resolutions is a considerable accomplishment, and a manuscript that documents these simulation results well warrants publication. The manuscript is clearly organized and written. The scope of this manuscript is large, covering the atmosphere, land, and ocean-sea ice models and that leads to quite a long manuscript, but considering that it is an overview paper, I believe the length is warranted. I only have the following minor comments that I believe will help provide more context to the baseline Cycle 3 simulations that are presented in this manuscript. 

Minor comments

More details about radiative fluxes in the Cycle 3 experiments

Based on Section 3.1.1, one of the key issues fixed in Cycle 3 appears to be the radiation balance of the model. While the Figure 4 is helpful in seeing the time evolution of the global mean top of atmosphere radiation bias with respect to CERES, it doesn’t give us a sense of the partitioning of longwave and shortwave and geographic distribution. It is mentioned in Section 3.1.1 that changes in both low-clouds and high-clouds were made to reach a more balanced radiative balance. A geographic map of the change in the radiative fluxes would be useful in seeing how those change between Cycle 2 and 3 manifest themselves geographically and in the shortwave and longwave fluxes. 

Thermodynamic and dynamic impact of the model developments

I would have expected that the fixes to the water and energy imbalance and changes to the deep convective scheme to have an impact on the thermodynamic (temperature and humidity) and dynamic structures within the atmosphere. Were they not reported here because the impacts were negligible? If so, a sentence or two describing the lack of impact would be informative to show that indirect impacts are small compared to the direct impact that are reported in this manuscript. If it did have an impact, I believe it would also be useful to show those changes, because this is an overview of these simulations and the changes in other large-scale features from the changes are already reported (radiative fluxes and zonal precipitation). 

Specific (minor) comments

L394: qv, ql and qi are listed but they are referred to as water vapour, cloud ice and snow. Based on the equation, it should be qv, qi and qs.

L402: typo with Wm--2

L461-463: Do the authors have any more information of whether these weakly active convection events are deep events with little mass flux or whether they are shallow events?

L762-L764: I realise it might be difficult to explain, but do the authors have a hypothesis for why at higher resolution the eastward inertia-gravity waves and mixed Rossby-gravity waves might be better represented?