The authors make an important contribution to how to approach modeling adiabatic processes that can impact Rossby waves for three different case studies. The work is excellent but could include a more in-depth look at why the errors and improvements occur as the model resolution and treatment of convective processes change.  

Review of ‘Grid-spacing sensitivity of Rossby wave breaking to mesoscale diabatic processes’ by Rixen et al.

General Comments

In this article the authors present a systematic resolution comparison using the ICON model and ensemble simulations of three different case studies in which Rossby wave breaking is involved. The topic is definitely in scope to be published in Weather and Climate Dynamics, and it falls in a research domain which is currently very active, making it a timely contribution. However I think the article would benefit from more in-depth discussion of at least four aspects as outlined in the ‘Specific comments’ section below, before it is considered for full publication in the journal.

Specific Comments

• Forecast error v chaotic behaviour

The article would benefit from discussion on the interpretation of ensemble spread and low ACC even in those ensemble members that exhibit improved results. To what extent are we talking about forecast error and to what extent are we talking about inherent unpredictability of the system. The following are some comments relevant to this aspect:

Abstract: L18-19: ‘ensemble spread and forecast error’ Are these to forecast features at the same level, i.e. are both desirable or undesirable?

L106: An alternative is that forecast busts are related to inherent low predictability of the system. How can we distinguish between the two alternative possibilities?

L473: Comment on situations where ensemble spread signals an inherently unpredictable situation. How do we distinguish between those and one in which the ensemble spread is erroneously large? Would it be useful to incorporate comparisons against the ensemble mean error in the analysis?

L279-280: I agree that at 40km resolution the ACC is generally low but some ensemble members yield very high ACC. How do you interpret this?

L291-292: I disagree. They are outliers when taken as part of this ensemble, but they are still able to capture the actual system's evolution.

L410: Are these two clusters pointing to inherent unpredictability in the system, which is also reflected in the ACC plot (Fig. 13).

L458: Here or elsewhere, comment on the level of ACC reached as some of the cases do not reach particularly high levels. What is the mean ACC and how do the improvements compare to that?

• Results in the context of the operational IFS and ERA5

The article should include more comments on how the ICON simulations compare to the IFS and ERA5. For example, it should discuss what processes are expected to be resolved by ERA5 and what processes are not resolved by the reanalysis.

L349-351: In this case the 20km resolution is the one that performs the worst, similar to the IFS even though the IFS had at least double resolution…? It would be informative to comment on the results in the context of the operational forecasts and the reanalysis.

Figure 9: These simulations can be grouped as with and without parametrised convection. For both groups the ACC improves as resolution increases, but both reach similar levels at 10km the first and 2.5 km the second. It's interesting that some members at 10km seem to perform much better than all those at 2.5 km and that the ACC achieved in this case is generally lower than that in the first case.

L426: Is this due to IFS and ICON having similar resolution at 10km?

L455-456: Here or elsewhere, more discussion is needed on the ability of ERA5 to represent these features itself as its resolution is in the middle in comparison with those used here and it's low in comparison with that of the IFS.

• Physical phenomena v their model representation

L38-43: It would be useful to make a distinction between the physical phenomena and their representation in models. It is the latter that leads to subsequent forecast error. Tropical cyclones are sustained sources of diabatic heating. Are we systematically under- or over-estimating this heating in models?

• Physical phenomena as error vectors

L263-264: I'm not convinced this makes sense. A physical phenomenon, in this case a WCB, can't advect/propagate error. A more objective way of putting this would be to consider the initial discrepancy as a difference in initial conditions at the start of the WCB.

Technical Comments

L100: Should it read ‘investigating’ or ‘improving’ rather than ‘investing’?

L114: ‘Which diabatic processes are better simulated in fine grid spacings?’ This will depend on the model and its associated parametrisation schemes therefore it would be helpful at this point to have a list of the processes that are being considered here.

L112-115: The way in which the questions are written here is an implicit distinction between large-scale forecast errors and forecast busts? Can this differentiation be made explicitly, e.g. are they different in terms of lead times?

L121: What is the IFS’s native resolution? In general, the results from the ICON simulations should be put in context by considering the resolution of the IFS and ERA5.

L125: ‘as it is based from data assimilation using IFS’ I’m not clear on what the message here is. Do you mean it is based on the same forecasting system including data assimilation?

L126: Delete ‘perturbed medium range’

L127: Why are these ensemble members characterised as perturbed? Are they not simply the IFS ensemble members, or are there additional perturbations applied to them?

L130: It should read ‘Multi-SatellitE’. Delete ‘satellite’ after ‘(IMERG)’.

L135-136: Unless I’m misunderstanding the idea, ‘...perturbations taken from the 50 perturbed IFS ensemble.’ should read ‘...given by the 50 IFS ensemble perturbations.’

L136-138: Would it be better to use the SSTs that the operational IFS actually takes as boundary conditions? Otherwise, there are multiple factors that will make the results confusing.

L153: ‘In order to compare the same ensembles…’ I thought the whole idea was to generate new ensembles at different resolutions, or do you mean the same ensemble members? Even so, in what sense would they be the same?

Equation (1): Possibly not necessary but the equation can be simplified. More importantly, should the expression include a correction for latitude as grid points at higher latitudes represent smaller areas than those at lower latitudes? This same comment is valid for (2).

L183-184: ‘... ensuring that the deterministic forecast does not represent an outlier within the ensemble distribution.’ Why is this argument not deemed necessary for the first case? And how do you decide that the evolution of an ensemble member is an outlier and not a physically plausible realisation of the system?

L189-190: Move the references next to ‘LAGRANTO’, otherwise they appear to document the ICON simulated data.

L203: The subscript ‘obs’ is misleading as it actually labels reanalysis data.

L219: ‘The storm produced the third-highest daily rainfall total in the UK since 1891.’ Do you have a reference to back this claim?

L221-222: An MCS would be associated with negative PV anomalies and therefore with the formation of a ridge rather than a trough. The development needs rewriting to make it clearer.

Throughout the text: Discuss the figures and panels in order.

L229: ‘... highlighting differences in the WCB outflow.’ How do you know these differences are caused by differences in WCB outflow?

L233: Repetition: Delete ‘performed at horizontal grid spacing ranging from 40km to 2.5km’.

L234: Repetition: Delete ‘The simulations are used to analyse the sensitivity of forecast skill to grid spacing and to investigate key contributing processes’.

L238: Here and other places where times are quoted. To improve the paper’s readability, in addition to quoting the validation time (1200 UTC 10 February) indicate forecast lead time. Double check the journal’s notation requirements.

L241-242: The standard deviation has grown but its link to MCS representation is not obvious. Could MCS variability be shown explicitly?

Figure 1: The two panels could be merged into one.

L245: Leave conclusion to the end or indicate that this will be justified: Remove ‘A clear sensitivity to horizontal grid spacing is observed’.

L245-247: This statement is true but it could be affected by environmental factors such as vertical wind shear.

L248: However, ERA5 native resolution is much lower than 2.5km grid spacing… How can this be reconciled?

Figure 3 (and others): Improve figures so that legends do not obscure plots.

Figure 3 caption: Why do you use 3 PVU and not the standard 2 PVU contour which marks the position of the dynamical tropopause. At what level is the negative PV area calculated?

Figure 6: Language: An ensemble is a collection of ensemble members. I don’t think this is the meaning intended in the legend of panel (a). Panel (b): Difference plots are easier to interpret if the reference (i.e. ERA5) is subtracted from the field you want to evaluate, i.e. IFS - ERA5.

L311: ‘positive temperature anomalies’ I would have expected a negative anomaly as warm air fails to reach higher latitudes.

L336: Could you give more details on how this was calculated? How the jet stream wind speed is determined and then what does the 99th percentile represent?

Figure 10(b) and Fig. 11(b): Are these panels discussed anywhere in the text?

L382: Delete ‘that’.

L383: Change ‘exhibiting’ to ‘exhibits’.

L292: ‘wave activity energy’ can be deleted as WAE has already been defined.

L401: I assume it should read ‘Figure 10b’.

L403: Change ‘ensembles’ for ‘ensemble members’.

Figure 12 colorbar: Do you mean ‘percentage of ensembles’ or ‘percentage of ensemble members’? By showing only this variable it is difficult to assess whether the WCB outflow in ERA5 is actually being captured by any members in the various ensembles?

L443-444: ‘divergent-flow anomalies’ What do you mean by anomalies in this context? What is the reference against which the anomalies are calculated? Do you actually mean forecast error?

L447: ‘...that resolving these upstream processes largely improves the forecast skill downstream’. You do show that the forecasts improve at shorter and longer lead times but I’m not convinced that the link to processes is fully shown in the paper.

L453: It should read ‘Fine’ rather than ‘Fin’.