The paper is describing a relatively simple method for the point-wise estimation of skew surge from the wind data at several locations. The manuscript is essentially a well-documented and transparently described model development. The model is a more sophisticated variation of earlier developed regression models and effective wind concept with quite convincing statistics for the estimated skew surge results. The topic is relevant, especially in the scope of ever-growing amount of climate change scenarios and a need for fast and reliable tools, which translate regional atmospheric scenarios into local water level extremes.

Below are several technical questions and minor comments.

Eq. (1): Are the mean wind speed (U_k) and mean zonal/meridional winds calculated from all timesteps of the reanalysis or only those related to the skew surges from the corresponding hwss? If all timesteps are considered, then the normalized mean wind (uas_k/U_k, vas_k/U_k) is an averaged normalized wind at this grid cell independently on whether it causes storm surge or not and thus effwind is a projection on average wind direction, not the direction favorable for the generation of surges. In this case it is unclear why it is called “effective wind”, effective for what? If only winds associated with skew surge higher than a certain threshold are considered to construct the normalized mean, please specify which timesteps exactly were taken (e.g. at the moment of the skew surge or within 12 hours prior or something else)?

L145-153: I’m trying to understand the shape of the individual model (for each grid cell, time lag, hwss). From this passage I would assume it looks like log(skew_surge)=a*(ef_wind²)+b*(ef_wind)+c. If this is not what was used, please reconsider the description. If this is what was used, then (1) just for the sake of terminology, this is not a multiple linear regression as stated in Line145, but rather a simple quadratic regression (2) I find the sentence “In this way, we ensure the effect of the negative sign” misleading here. Firstly, for high skew surges (hwss) effective winds will be positive anyway. Yes, consideration of only positive coefficients may be helpful later when the skew surges for the whole year are reconstructed and negative effective winds are well possible, as explained in Sect. 2.3.2, but to ensure influence of negative effective winds a special treatment is necessary and positive coefficients alone do not ensure the effect of negative sign. I suggest to refer here directly to the Sect. 2.3.2 for more coherent explanation of this constrain for those interested.

General comment: I wonder how sensitive is the model to the selected training dataset? That is, how much the models (and the estimated skew surge) change when different years are excluded, as it has been done during the validation procedure? If, for example, years Y1 and Y2 are excluded to generate Model1, Y1 and Y3 are excluded to generate Model2, would the reconstructions of skew surges for the year Y1 from these two models be identical? More generally, especially if the model to be used for scenarios, does the size or selection of the training dataset matter?

L20: “Coastal protection institutions”  - Meant are those organizations who plan and construct protection structures? Is this an established term?

L22: “continuing rise of sea level” -> continuing rise of mean sea level

L24: “as sea level pressure” -> as atmospheric sea level pressure

L28: “… these events: The storm surges studied include…” -> … these events. The storm surge studies include …

L36: The most widely used and reliable method of such translation is a hydrodynamic model. It is clear, that this study is about simple fast methods beyond classical models, still I think the dynamical models should be mentioned somewhere in the text. Maybe within a short explanation why they are not always the best choice and where the alternatives are needed.

L57: “They find that the external surge…” -> They find that thus considered external surge and…

L64: “excluded in the model setup” -> excluded from the model setup

L274: “the track consistently follows a northwest to southeast orientation” – I presume here the track refers to the path of grid points with maximum R^2 and not the storm track. Maybe choose another word because “track” has a certain connotation. The position of crosses also doesn’t help to identify the storm track itself, maybe only hints to it, so the relation to the next sentence about prevailing northern storm tracks is not obvious.  

L352-356: On what dataset the bias correction (quantile mapping) was trained before it was applied to the omitted years? Would be helpful to specify in the text what was the training and what was the validation datasets.

Review: Development of a wind-based storm surge model for the German Bight; by Schaffer et al.

This study develops a simple yet effective storm surge model for the German Bight using multiple linear regression with 10 m effective wind as the predictor. Trained on historical skew surge data (1959–2022), the model achieves a high predictive skill. Such models can complement storm surge prediction in climate simulations and aid in evaluating future risks under a changing climate. Overall, this is valuable work and I recommend that it be published in Natural Hazards and Earth System Sciences after revision. Below, I outline a few main comments along with several minor ones for consideration.

Main comments:

1) The manuscript is difficult to read due to excessive over-explanation, which does not significantly enhance the understanding of the methods. To improve readability, I suggest streamlining the text by focusing on essential details in the main body and moving less critical or overly detailed sections to the appendix. For instance, Section 2.2.1 and 2.2.2 could be relocated to the appendix. In the comments below, I will highlight specific passages that can either be omitted or moved. Overall, the authors should aim to make the paper more concise and accessible.

2) Regarding the use of the effective wind 12 to 1 hour before the skew surge event, I understand the logic explained in the paper, but am not sure of its broader purpose. If this method is intended to detect surges in climate model simulations, wouldn't it be sufficient to use e.g., the 1-hour or 0-hour time step, unless the inclusion of 12-, 6- and 2-hour time steps (as in Equation 8) demonstrably improves model skill? Could the authors comment on the extent to which these additional time steps improve the predictive ability of the model? Furthermore, given that most model outputs are available at 3 or 6 hour intervals, wouldn't the 2 and 1 hour time steps be impractical for such applications?

3) The bias correction (Figures 5 and 6) does not significantly improve the model performance for values above 80 cm. If the focus is on values below 80 cm, a threshold of 50 cm or lower could be considered instead of 80 cm. Overall, this additional step might not be necessary.

Other comments:

Title: I would suggest replace “wind-based” with statistical or something like that

Line 71: replace “However, as all these” with → all mentioned 

line 86: remove “of relevant … positions of the”

line 88-89: “(c) … training the model… (d) training and …” Are “training” in c and d different?

Line 94-95: “...time series of high water skew surge (hwss)… 1959-2022” could you show a figure? Is it figure 8?

Figure 2: panel (b) instead of “identifying relevant predictor position” with something like→ locating the effective wind with the highest value of ?? for every time step prior to the event

Table 1: These data points are over which period? Could you include 250 cm as well?

Please improve the writing in Section 2.1.2 for clarity and readability:

Line 199-120: remove “a grid cell wise”. For composite analysis, normally you describe them as e.g., composite map of uas and vas based on thereshold of hwss for the period of 1959-2022…

Remove details of ERA5 or mention it earlier

Reomve line 127-129. 

Bring the equation after line 132

Section 2.2:

Please move Sections 2.2.1 and 2.2.2 to the appendix, retaining only the most important parts. The section could then be renamed to 'Statistical Model Development for …' or a similar title. Alternatively, if the purpose of this section is to explain the methodology, the text needs to be improved for clarity and focus. 

Section 2.3:

Most of the content can be moved to the appendix. Some information, such as the F1 score definition, is repeated later in the text.

Line 199-203: repeats information already mentioned earlier in the text. 

Line 264: “...spatial and temporal… ” please change e.g., location and time lag

Figure 3: would it be possible to show the same for both hwss>50 and hwss>150cm?

Line 270: remove “the grids cells in”. Similarly in other part of the paper

Line 305: Referring to my comment 3 above, you could use 50 cm as the threshold as well, correct? From Figure 5, it’s difficult to determine whether 50 cm is better than 80 cm. However, if you opt for 50 cm, would bias correction still be necessary?

Figure 5: please show the chosen method (elastic net) but instead show a similar plot for both hwss>50 and hwss>150. Please also include the fitted line for the data.

Figure 7: Please include F1 score for panel a plot.

Table 2: Please show the values without bias correction as well. Also mention the number of events from observations.

Line 376-378: Is this statement for bias corrected case? Then you can use as an argument to my comments above.