This is a nice piece of study that addresses mismatches of water level extremes in the Baltic Sea simulated using different high-quality modelled wind data sets. Although such mismatches are known for some time, it is the first systematic analysis of the discrepancies, associated uncertainties and ways towards a reasonable forecast of extreme water levels for moderately long (30 years) return periods based on the application of an ensemble of hindcasts.

It is encouraging that the average of the resulting ensemble represents well return periods extracted from modelled data. The analysis of spatial variations in the trends of annual (storm-season) water level extremes (both modelled and extracted from measurements) is also interesting and valuable even though there are also major uncertainties and discrepancies between the ensemble members.

The analysis is sound and generally well presented. The use of English is appropriate. The presentation still needs some polishing as the text contains numerous examples of unnecessary field-specific jargon that would be nice to remove (unless specifically justified, e.g., for brevity). In particular, the use of “default wind speed simulations” to denote the unadjusted results of such simulations is misleading. Consider using “unadjusted” or similar instead. 

The results are generally presented in proper context, so I have only a few remarks (see below). The most intriguing conjecture is that there has been effectively no increase in the Baltic Sea water level extremes over last decades even though the average water level has been increasing. One the one hand, this outcome is basically consistent with the general perception of the Baltex/Baltic Earth community that wind speed extremes and the number of strong storms have not substantially increased in the Baltic Sea region. On the other hand, it calls for further analysis of the reaction of Baltic Sea water masses to changing wind forcing both in terms of waves and (extreme) water levels as several analyses (e.g., Jaagus, J., Suursaar, Ü., 2013. Long-term storminess and sea level variations on the Estonian coast of the Baltic Sea in relation to large-scale atmospheric circulation. Est. J. Earth Sci. 62, 73–92. https://doi.org/10.3176/earth.2013.07) have identified rapid increase in water level extremes in the eastern Baltic Sea.

It might be mentioned that the most extreme water levels in the Baltic Sea often represent a different population of extremes (because of preconditioning) and thus the classic approaches such as the GEV technique (that presumes the presence of one population) do not necessarily work as noted in (Suursaar and Sooäär, 2007) and mentioned several times later. This specific population is visible in Figures S1 and S2 as a few highest extremes that do not match the theoretical curve. 

Line 156: “On average, the simulations underestimate the maximum water levels” and line 161 “However, the wind speed increase is necessary to capture the correct ESLs in the Western Baltic Sea” are very strong results/claims. Please comment whether this is a typical issue in other studies and/or provide supporting references.

The obtained predominantly negative trends of annual sea level maxima obtained in Section 3.3 radically differ from large positive ones reported in (Pindsoo and Soomere, 2016) // (Soomere and Pindsoo, 2020) for the entire Baltic Sea. Could the difference stem from different time periods? Or different forcing and ocean models? Or the adequacy of replication of water exchange via the Danish straits?

The Discussion section provides some claims that seem to be overly simplified reflections of the results. For example, the sentence on page 16, line 234–235 “Nevertheless, the ensemble agrees well with observed return levels and trends” might be augmented to mention that in quite many locations the trends from measured data have even different signs compared to trends evaluated from simulations.

Minor comments:

Page 2, lines 25–26: the meaning of the sentence “However, the representation of ESLs in numerical models adds another source of uncertainty” remains unclear (or at least ambiguous); please explain what is meant.

Line 41: please indicate the year for reference (IPCC).

Line 43: “their Tab. 2.” – probably it is meant Table 2 of (Bamber et al., 2019); please make the link clear.

Line 44–45: the claim “With this rise, the exchange of water masses with the North Sea is expected to increase, which could affect preconditioning” is not fully justified. I guess that after water level rise in the North Atlantic there will be simply a new balance of water masses of the North Sea and Baltic Sea. However, as the Danish straits and the Baltic Sea will be deeper, the wind-driven slope of the Baltic Sea water could be smaller than today and thus the water exchange in the context of the current study may even be reduced. This change, however, will affect preconditioning but – as mentioned – does not necessarily enhance this process. If my reasoning is not correct, please provide additional evidence or arguments to support the claim.

Line 48: “singular” seems not appropriate as this word has clear meaning in mathematics and physics; consider saying “exceptional” or similar.

Lines 47–49: the estimate needs a supporting reference.

Line 51: must be “therein”.

Line 51: probably “spatially heterogeneous” is meant and not temporal heterogeneity.

Line 54: please adjust wording for clarity.

Line 55­56: “return periods … become more frequent” is nonsensical; please reformulate.

Lines 56–57: “This inhomogeneity in sea-level rise and atmospheric changes has led to more frequent and longer ESL events” does not make much sense unless an area is indicated.

Page 3, Line 60: shifts were detected in (Kudryavtseva et al., 2021) for the Gulf of Riga while most changes were basically smooth on the Latvian seashore of Baltic proper.

Line 64: style of references is inconsistent.

Line 77: says “On the other hand …” but “One the one hand …” is missing.

Line 88: should be “Baltic Sea” or “Baltic region” or similar.

Page 4, Figure 1: there is no need to use expanded West-East scale.

Lines 101–103: I agree with both claims (and “However” should be removed) but you lose interaction of tides with water exchange through the Danish straits. Please either provide an explanation why this is negligible or mention this assumption when describing limitations of the study.

Lines 108–110 and Page 5, caption to Table 1: please minimize repeated information.

Line 114: “land uplift” – Was land uplift included in the GETM model? Or how/elsewhere?

Lines 117–119 and caption to Fig. 1 provide repeated information; please adjust for clarity and brevity.

Line 125: it is recommended to refer here also to (Kudryavtseva et al., 2018. Non-stationary modeling of trends in extreme water level changes along the Baltic Sea coast. Journal of Coastal Research, Special Issue No. 85, 586–590, doi: 10.2112/SI85-118.1) that signals non-stationarity of GEV parameters for extreme water levels in several locations of the Baltic Sea while (Kudryavtseva et al., 2021) focuses on Latvian waters.

Page 6, Figure 2: consider indicating also original names of water level gauge locations (e.g., Warnemünde, Pärnu).

Page 7, line 129: “using the time series of annual storm season (July to June) maxima, such as block maxima” is a good idea; however, there should be either a direct justification of this choice or a reference to, e.g., (Männikus et al., 2020. Variations in the mean, seasonal and extreme water level on the Latvian coast, the eastern Baltic Sea, during 1961–2018. Estuarine Coastal and Shelf Science, 245, Art. No. 106827, https://doi.org/10.1016/j.ecss.2020.106827) where this justification is provided. The issue is fundamental as water level maxima over calendar years are not necessarily independent owing to preconditioning, and then the GEV technique is not applicable. Also, remove “, such”.

Line 133: consider replacing “subfamilies are defined” by saying that the GEV is reduced to one of the following distributions depending on the sign of the shape parameter.

Eqs. (3), (4): there is no need to introduce $\alpha$ but if you do so, please give the link to $\xi$ (even though this link is basically trivial).

Line 146: I guess that a Python script is used also for this fitting.

Page 8, Lines 152–153: use simply “ESLs (Fig. 3), where” and add “(Fig. 1)” to the end of sentence, so that the reader is directed to the relevant map.

Line 155: “the default wind speed simulations” is nonsensical as mentioned above, most likely something like “the simulations forced with wind properties from the atmospheric models” or similar is meant.

Line 156: “On average, the simulations underestimate the maximum water levels” is a very strong result/claim as mentioned above. Please comment whether this is a typical issue in other studies. For example, (Pindsoo and Soomere, 2020) who use another model and Rossby Centre wind fields do not complain about that.

Line 158: “the mean bias of the model results improves significantly” does not make sense as the bias can become smaller but cannot be improved; please reformulate.

Lines 161–162: please reformulate “default simulations” to make clear that simulations forced with unadjusted winds from atmospheric models are meant. It is strongly suggested to use “unadjusted” or similar instead of “default” that could be misleading also in what follows. The same applies to legends of Fig. 4 and to numerous occasions below.

Line 169: use simply: “stations (Fig. 4) shows”.

Line 179–180: “the ensemble is close to the observed return levels” is heavy jargon and ambiguous; please explain what exactly is meant.

Page 9, Caption to Fig. 3: explain STD; reformulate “with and without the increased wind speed”, the rest of the caption repeats information given in the text and should be removed or radically shortened.

Line 196: one can understand what “variability between the 30-year return levels” means but a classic way of saying this is “differences between the estimates of 30-yr return levels”.

Page 11, lines 201–203: “In short, comparing the 30-year return levels for each station, the patterns are similar except for the generally lower estimates for the 30-year return levels.” – This sentence says nothing.

References to Fig. 6 appear before references to Fig. 5.

Line 207: probably “pattern obtained using” is meant.

Line 208: “the 95% confidence interval of the return levels from the GPD method is similar to the results” – please reformulate for clarity.

Lines 211–212: please reformulate for clarity.

Page 12, caption to Fig. 5: remove “Spatial”.

Page 14, lines 214–215: “the comparison of ESLs with observations and the estimated 30-year return levels differ significantly” – a ‘comparison’ cannot ‘differ’; please reformulate.

Lines 224–225: “However, how the size and shape of the different areas of positive and negative trends differ, e.g. the extent of the area in the Western Baltic Sea that shows a positive trend” seems unfinished, please reformulate for clarity.

Lines 237–238: something is missing in “The variability of the atmospheric high wind speeds, i.e. storms, in the different atmospheric datasets forward into the ESLs.”

Line 259: “To reduce the uncertainty, a large regional ensemble is needed to minimise the errors” contains two basically independent claims.

Page 18, line 278: “∂_tX should denote the trend” seems jargon; consider saying “the value of partial derivative ∂_tX characterizes the slope of the trendline” or similar.

Line 280: it is indeed true that “directional changes and preconditioning must play a critical role”; however, it might make sense to even more strongly discuss with (Soomere and Pindsoo, 2016) and (Pindsoo and Soomere, 2020) who derive a fairly rapid trend for annual maxima of preconditioning using the RCO model with a specific (gust-adjusted) wind fields. This kind of manipulation with wind data might be a good additional member of similar ensembles in the future compared to straightforward increase in the wind speed.

Line 309: “strongly modified by the default increase in wind speeds” remains unclear.

Lines 320 and 321: “distribution” needs not to be capitalized.

Line 325; “This uncertainty can be reduced by using much larger ensembles” is probably true but still it is a conjecture (=hope), not a conclusion.

Page 22, line 322: Andree et al. (2022) has been published as Nat. Hazards Earth Syst. Sci., 23, 1817–1834, 2023, https://doi.org/10.5194/nhess-23-1817-2023

Page 23, line 380: reference Grinsted, A.: is incomplete.

DOI is missing from most of references and the references contain several smaller issues (e.g., capitalization of journals’ titles).

This paper presents extreme sea level analysis for the Baltic Sea, based on a GETM simulation of the water elevation dynamics for the years 1979-2018. The paper is well written and addresses an important topic. However, certain aspects of the research should be clarified, particularly related to the ocean model configuration, before the manuscript can be recommended for publication.

Major comments:

The model configuration, presented in Section 2.1 should be elaborated. The authors should provide sufficient information about the model configuration such that the simulations can be reproduced by independent researchers. In particular, as atmospheric pressure and wind stress are key drivers for extreme water levels, used forcing methods should be presented in more detail. Which wind stress formulation was used? Why that particular one was chosen? Can it be argued that the existing and widely-used formulations are suitable for the Baltic Sea (to the best of my knowledge, many wind stress formulations are designed for global ocean simulations). These details should be added and their impact also discussed in the manuscript.

In this study, the wind speeds have been artificially increased by 3 to 11 percent to take into account the fact that atmospheric models have a tendency to underestimate extreme wind speeds. Altering the wind stress parametrization would be another way of achieving the same, potential benefit being that one can only alter the wind stress for high wind speeds.

In order to obtain reliable extreme value estimates, the model should be sufficiently accurate in representing the tides, emptying and filling of the Baltic Sea basin, seiche waves, and atmospherically-driven effects - extreme SSH values are formed as a superposition of all of these. (While the tides might not play a major role in the Baltic Sea itself, they may affect the volume flux to/from the North Sea.) The manuscript focuses mostly only on the atmospheric effects. It would be good to discuss the other effects and their impact as well.

Section 3 shows a comparison of extreme sea levels. To better assess the skill of the model, I recommend including a comprehensive statistical analysis of the reference model SSH in the Baltic Sea (and perhaps the Danish waters). It would then be clearer whether we are only testing the effects of different atmospheric forcings using a poor SSH model, or whether the model reproduces SSH dynamics well in general (giving more confidence that combined SSH effects can be reproduced).

The model uses a constant bottom roughness length value (z0 = 1 mm) for all the simulations. Is this a realistic value? How was it chosen? Was the same value used of all nesting levels? Tuning the bottom friction for North Sea-Baltic Sea simulations is not a trivial task (e.g. Kärnä et al. (2023) and Kärnä et al. (2021)) if one wishes to represent SSH dynamics well across the domain - it also affects the attenuation of seiche oscillations in the Baltic.

I highly encourage the authors to share the GETM source code, input files, and also post-processing scripts for reproducibility. Quite often some data (e.g. forcings) cannot be shared due to licence restrictions, but even under such circumstances, the authors should provide links to where the data can be accessed.

In the light of the given results, it seems that understanding the sensitivity of extreme sea level values to wind forcing is essential. Such sensitivities can be computed directly with adjoint models (e.g. Kärnä et al. (2023) and references therein). As such adjoint/inverse modeling could be an important asset in this research in addition to more traditional ensemble methods.

Minor comments:

Which GETM version was used to run the simulations?

line 98: Please elaborate. Are you only using the local atmospheric pressure information to calculate an additional offset to boundary SSH value? This would not include any atmospheric pressure-driven waves.

line 105: The atmospheric forcing is not the same for the North Atlantic and the Baltic Sea models. How can you ensure that SSH at the nesting boundary agree, e.g. if a low pressure system is located in a different place in ERA5? This could generate spurious strong wave fronts in the model and skew the ESL analysis.

line 154: For the sake of clarity, please emphasize how you calculate ESLs from model/observation SSH. E.g. add an equation.

Figure 4: I would not use a line plot here. There's no continuity between the stations (at least in the sense the figure suggests). A bar or violin plot for each station would be better.

Figure 7 caption: if you mention a) mention also b)

line 288: typo FThe

line 322: remove "therefore", there's no causality here

line 330: The reviewers should have access to the data/source code during the review.

References:

Kärnä T., Wallwork J. G., Kramer S. C. (2023). Efficient optimization of a regional water elevation model with an automatically generated adjoint. Journal of Advances in Modeling Earth Systems (under review). Preprint: https://arxiv.org/abs/2205.01343

Kärnä T., et al. (2021). Nemo-Nordic 2.0: operational marine forecast model for the Baltic Sea. Geoscientific Model Development. https://doi.org/10.5194/gmd-14-5731-2021