General comments

The article looks into differences in estimates of significant wave height extremes due to tropical and extra-tropical storms at two offshore locations in the eastern coast of the United States.

The article is poorly written and some of the analyses are not sound. For example:

• There is no motivation being given for the type I tail assumption being made when fitting the Gumbel instead of the Generalized extreme value distribution to the annual maxima. Furthermore, and as can be read in a wealth of extreme value theory publications, for instance the cited book of Coles, the Weibull and the Gumbel distributions are not the asymptotic distributions of data sampled using the peaks-over-threshold approach. This does not mean that they cannot be used but their use should be justified.
• Also, the authors confuse the directional spreading of a wave system or sea state with the variability of the mean wave direction during a storm.

Specific comments  

Lines 82-83: What is the rational for fitting the Weibull (of minima, I assume) and the Gumbel distribution to POT data? Can you justify why you are deviating from the Generalized Pareto distribution?

Lines 136-138: Please indicate whether there is corresponding between the storms leading to the annual maxima in both GF and hindcast datasets.

Lines 159-161: Please rephrase of remove. The estimates are obtained using the likelihood method? If so, it suffices to state it.

Section 2.2.1: Please motivate why the Gumbel instead of the Generalized extreme value distribution is being fitted to the data.

Lines 170-171: Why is p called “probability period”? Please add that when computing the return values p is substituted by 1/n with n being the return period in years. In the tables and text only n is being given, not p.

Line 178: Why does the storm list given in Appendix C only starts in 1991?

Figure 3: 1)The data to which the distributions were fitted need to be added to the figure. (If not possible in absolute scale, then in relative scale as in Figure 4.) 2) Preferably also the 95% confidence intervals of the estimates should also be given. 3)The legend should contain for each of the lines the periods covered by the data or the sample size (number of considered annual maxima).

Line 91: There are only 15 samples in the ‘GF-EC. Trop.’ fit? Please comment on the uncertainty of the estimates.

Section 3.1.2: In my opinion this section can be removed. What is its purpose? Why are the plots of the significant wave height (even if normalised) not shown?

Section 3.1.3: The contents of this section are incorrect. First, how can the authors not be aware that the waves in the roses in Figure 7 are from the coast and therefore not realistic. Second, the authors present the variation in the mean wave directions during the consider storms (may wave systems, sea states) and analyse with reference to the article of Forristall and Ewans on directional spreading of a wave system or sea state.

Section 3.2: This section needs also to be completely redone. When comparing statistical estimates the sample sizes and confidence intervals should be given. Furthermore, when making assumption in terms of the tail of the data these should be justified.

Technical corrections

Line 46:  Please specify which are the variables being considered in the univariate and bivariate analyses you are referring to. Why is this relevant for this article?

Line 69: Specify which 3 models and models of what?

Line 78: What does “Return period results” mean? Should it be “Return value estimates”?

Line 96: You mean Appendix C instead of B?

Line 104: Is the magnitude of  Cds correct? Please introduce the meaning of the symbols it wanting to give the values.

Line 110 and elsewhere: Explain what you mean with “most “at-risk” turbine location” and how this has been defined.

Lines 114-118, …: Provide references for SWAN, Delft3D, Westhuysen, WAM, OWI3G,...

Line 117: Define acronyms throughout the text. For instance, what does YSU mean?

Lines 124: State also model depth for the considered output location.

General Comment

This article proposes a methodology to find extreme wave values in regions where both tropical and extra-tropical cyclones occur. The motivation given is for the more accurate prediction of design values for offshore wind turbine farms. Long term return values (up to 10,000 year return periods) are calculated using statistical methods. Overall, I find that this article has scientific merit; however, the findings/methods are somewhat obfuscated/unclear. I believe that by clarifying the methodologies used and providing more context for why specific scientific decisions were made the article would be significantly improved.

Specific Comments

Line 73: You mention "validated and calibrated" models. Have these calibrations/validations been published elsewhere? If so, please include the citation to the appropriate papers. If not, please include a subsection with an overview of both the calibration (what was calibrated, how parameters were selected, etc.) and some of the validation data.

Line 80: Why was the Block Maxima with Gumbel Fit selected over, for example, a Generalized Pareto DIstribution with peaks over threshold? Just below this line you mention that you performed a sensitivity analysis using different methods but you do not explain why you ended up highlighting the BM with Gumbel results.

Line 89:  Could you explain more about why a POT method is not appropriate for "only storm events"? Given that POT assumes the events are independent (which I would say applies to individual storm events) and that threshold selection, whether through graphical or automated methods, relies upon the fact that for any threshold that produces an adequate fit a threshold larger than that should produce the same fit (when using a generalized Pareto distribution), I fail to see why the lack of "normal sea states" precludes the use of a POT methodology.

Section 2.1: I find the description of the numerical models to be lacking in detail. As you mention in multiple locations, the location and derivation of boundary conditions can greatly change the results of a numerical model. Despite this, there is no description of the model domains, i.e., does the model cover the entire North Atlantic basin? Does it only cover the insets from Figure 1? You also mention again that the models are "locally validated". Where can I find this validation data? You mention the Commonwealth Wind metocean report (Wrenger, 2022) at line 108. Using the information in your works cited, I was unable to locate this report. There is the same issue with the Georgas (2023) report you cite for the Mid-Atlantic model (line 122). For the GROW-Fine East Coast model we are simply referred to Oceanweather inc. Please either provide the validation statistics in your work or, if possible, provide open-source and easily accessible reports showing why we should trust these models.

Section 2.1.4: I see here some mention of model validation. Consider moving some part of Appendix C into the body of the text. Especially given you specifically refer to the figures and error values in the appendix it seems appropriate that it would be part of the main text.

Figures 11 and 12: What are  the confidence intervals of the return periods you calculate here? Given the use of such a short time series for the estimation of very long return periods I would expect to see relatively large confidence intervals.

There are other locations where the methodology could be clarified and greatly improve this manuscript. At the moment, I find that the experiments herein would be very difficult for another researcher to reproduce, greatly limiting the usefulness of the findings.