Overall comments

This is a well-written paper and was a pleasure to read. The contents are not groundbreaking, but they are a nice quantification/comparison of the relative impacts of certain input parameters. The inclusion of the convergence of the sensitivity with respect to number of starting points and number of random seeds was a nice expansion. That being said, I have some significant reservations about the treatment of fatigue loads. Furthermore, I think there is perhaps some gap in the understanding of the extreme-load sensitivity, and I have some specific small comments on the structure/analysis of the paper.

Specific comments

First and foremost, I have significant concerns about the treatment of fatigue loads. As you write in the Conclusion (but not the abstract), “The standard deviation of the load was used as a proxy for the fatigue.” I understand that you did not want to complicate the post-processing or start adding material uncertainties to your input parameters, but to actually draw conclusions on fatigue loads, you need to calculate fatigue loads. The aggregation of fatigue loads is very nonlinear and not treating this rigorously is likely impact your results.

If you do not want to post-process your results into fatigue loads, then I would recommend removing references to “fatigue loads” and instead referring to “load variability”. But in either case, you should revisit Eq. 4. If you have fatigue loads, you should not be aggregating them with a mean operator but rather with a damage-equivalent-load-like operator. If you keep the standard deviation, then that should be aggregated by the standard deviation, not by the mean. Either way, the fatigue sensitivity calculation in Eq. 4 should be revisited. I don’t expect this would impact the ranking, but it might impact the relative importance of certain parameters.

Second, your paper does not reference existing literature that considers loads sensitivities of floating wind turbines. This seems like a rather unfortunate oversight, as the conclusions of the existing literature could be compared with the conclusions that are drawn in your paper. Here are three references I found that seem relevant, but there may be more:

1. Doubrawa, Paula, et al. "Load response of a floating wind turbine to turbulent atmospheric flow." Applied Energy 242 (2019): 1588-1599.
2. Nybø, Astrid, Finn Gunnar Nielsen, and Marte Godvik. "Sensitivity of the dynamic response of a multimegawatt floating wind turbine to the choice of turbulence model." Wind Energy 25.6 (2022): 1013-1029.
3. Müller, Kolja, and Po Wen Cheng. "Application of a Monte Carlo procedure for probabilistic fatigue design of floating offshore wind turbines." Wind Energy Science 3.1 (2018): 149-162.

Third and final, I’m not sure the paper adequately explains the connection between the sensitivity of certain input parameters and the shift in UEE caused by the shift in mean. for example, in Line 297 you write

“The ultimate values shown in Figure 4 are strongly stratified by wind condition for most outputs. The only outputs without this strong separation are the global system motion related quantities of mooring line tensions, watch circle, and heel angle”

This to me seems logical, as many of the load channels have a substantially different mean value at different wind speeds. But the text does not connect this to Yw in Eq. 3, which is of course the value that can change a lot. I expect also that this is why the anchor and fairlead ultimate tension are more sensitive to the current than to the wave conditions, because the current shifts the mean load.

I have a few other comments that I will mention here purely as food-for-thought for your future work. First, although the NREL 5 MW is, as you note, a well-characterized model, it is becoming an outdated design, so you might consider upgrading to the IEA 15 MW UMain semisub. Second, you are using ElastoDyn, which is of course a linear structural model. It could be interesting to see the sensitivity of the loads to modelling with BeamDyn, to see if there are any significant changes. But these changes might be insignificant in a stiff blade design like the NREL 5 MW.

Lastly, I want to re-emphasize the high quality of the writing and figures. I have some relatively small suggestions on the writing level (see “Technical corrections”), but the overall quality is good. I appreciate the time and effort put into this submission – thank you.

Technical corrections

• Line 69: It could be nice to include a description of the upcoming sections, especially mentioning Sections 6 and 7, which came a bit unexpectedly for me as a reader.
• Line 103-107: This is redundant, repeating the lines in L128-L133. I recommend removing Lines 103-107 and simply referring to the more in-depth description in L180-L133.
• Line 134: Consider explicitly saying how many seeds were chosen and including a reference to the seed convergence in Section 6.
• Line 146: “partial derivative type” -> “partial-derivatike-like”
• Line 144 – 151: This entire paragraph is confusingly worded, in my opnion, and thus difficult to follow. What is capital delta? And how do you calculate u_iw,range from your inputs? Consider revising and clarifying.
• Line 156: You are mixing notations in this equation, including both an overbar and vertical lines for the mean operator. Consider revising.
• Line 184: Here you could reference some of the existing literature FOW loads-sensitivity studies to motivate your choice of input parameters (or verify that your list is inclusive of previously analysed parameters).
• Line 190: Consider reiterating here that the variation of the input parameters was 10% of the range – I forgot this because it was only mentioned once previously and thought you were varying the parameters in the entire range.
• Line 203: “Justifications” should not be capitalized
• Line 209: This is not expected, considering Fig. 6 in [1]. Consider reviewing after you update your fatigue-load calculations.
• Line 300: If you are going to present Figures 4 and 5 in the main text, then I think they deserve a more thorough treatment than what is given here. Consider expanding this paragraph to not only include descriptions of the plots but also analysis of (a) why the trends make sense and (b) the important conclusions of the work. If there are no important conclusions, consider placing them in an appendix.
• Figure 4 and 5: Consider using a solid edge but a transparent face for the bar charts, to better prevent the bar on top from obscuring the bars plotted beneath.
• Figure 5: Consider using a logarithmic y-scale to better highlight the differences in the tails of the histograms.
• Figure 6 and 7: I really like these plots. A lot of nice information in a very concise way. Well done.
• Line 324: Section 6 comes very unexpectedly, and it was not immediately clear to me what the objective of the section was. Consider rewriting the section slightly and adding a brief introduction that explains that the previously presented sensitivities are depending on the number of seeds, and that this section will go into the convergence of the sensitivities based on seed.
• Line 331 – 338: This paragraph was confusing to me, especially because you are analysing plots presented in the appendix. If it’s important enough to deserve treatment in the main text, then the plots to be discussed should be in the main text. If they are not relevant to the main text, then they should be removed. My opinion is that Figures 8 and 9 are the “meat” of this section, and thus Appendix B can be removed entirely.
• Line 356: This section ends abruptly. What is the final conclusion of the section?
• Line 360: You can relate your conclusions to Fig. 3 in [1], although that was for an onshore turbine.
• Line 385: Why does adding more Sobol points increase the relative imfluence of sigma_u on UEE?
• Appendix A: If you plot in log y-scale, you might not need this appendix and you can move this discussion to the main text, where I think it belongs.

[1] Mozafari, Shadan, et al. "Effects of finite sampling on fatigue damage estimation of wind turbine components: A statistical study." Wind Engineering (2023): 0309524X231163825.

Summary of requested changes

I have written a lot of comments, so let me summarize the most important ones that I think must be addressed before acceptance:

1. Revisit the calculations of the “fatigue” loads, either by replacing the standard deviation with actual fatigue loads or updating Eq. 4 to be relevant for standard deviation.
2. Add missing references and discussion of existing literature mentioned in “Specific comments”.
3. Revise the entire text considering the dependence of the ultimate load on the mean value of the load.

All my other comments are purely informative – you are welcome to update the text based on those comments, but it is not a requirement.

The work presented here addresses an interesting topic, and the work is generally presented clearly. Studies investigating sensitivity are naturally sensitive to choices made regarding the range of inputs and how these are interpreted. In my opinion, the present study, while interesting, has limited relevance for design because 1) many of the input ranges considered are not related to *design* values of each variable, but rather related to temporal variations in these variables (see explanation from Sørum et al.) and 2) the environmental conditions are not necessarily relevant for extreme loads, yet ultimate loads are considered. Furthermore, the standard deviation is used as a stand-in for damage. Although this may be justified because of the short simulations (which would likely cause problems with unclosed cycles), only including the standard deviation and probability of occurrence is not very convincing. Overall, the results can be used to comment on how responses vary for different inputs, but not really to discuss uncertainties in design.

The choice of platform/turbine can also be interpreted as a weakness in the study, as the turbine is starting to be outdated and the platform is extremely heavy. (Consider the displacement of 14000 tonnes, compared to IEA VolturnUS 15 MW of 20000 tonnes or INO-WINDMOOR 12 MW of 14000 tonnes). Furthermore, the turbine controller is simply detuned (assuming that the controller has not been modified since OC4), which is an unrealistic strategy regarding power tracking for a  modern turbine.

I don’t find it surprising that the wave conditions appear relatively unimportant when:

• the platform is heavy
• the wind speed standard deviation variation is taken as a range between a very low (common) turbulence intensity and a conservative 90^th percentile design value taken from the standards
• the variation in significant wave height is taken as observed (i.e. not a conservative design value).

The time duration of each simulation is chosen to be 10 minutes, and it is noted that 1 minute is removed to account for transients. How was it determined that 1 minute would be sufficient, especially given that the platform has natural periods which are longer than 1 minute?

Specific comments:

P4: Please define wave stretching more clearly. I suspect that the authors mean that wave loads are integrated to the mean free surface (while wave stretching could also refer to specifically Wheeler stretching).

P5: I’m not sure that Robertson et al. 2018 is the correct original reference for the radial EE.

P6: “a local partial derivative type calculation” – maybe this could be rephrased as a local partial derivative approximation using finite differences?

Is it impossible to combine tables 2 and 3 for space reasons?

P10: The “maximum wave height” should be rephrased as the “maximum significant wave height” because the maximum wave height is a different quantity.

P12: The choice of ranges should be discussed more clearly with respect to what would be design values. Even though the ranges for some parameters are taken from previous work, the reader would benefit from being able to compare the reasoning behind these choices and the additional parameters (new to this study) in greater detail. 

P12: Capitalization error “Justification”

P13: Over what time period or with what probability are the maximum significant wave heights determined? These statistics need to be clearly defined.

P13: Is the breaking wave limit given for 200 m water depth? What water depth is considered for the buoy data?

P13: “site”  - “sites”

Fig 3: Perhaps the unphysical wave conditions should be removed from the figure (or shaded over)?

P14: It is interesting that the current is included, as this effect can be quite important. However, I’m not sure that the load models account for this effect completely. Vortex shedding is not included in the hydrodynamic load models, and there is no discussion of how the current is assumed to affect the waves (in particular the wave period).

P14: For some platforms, current in the opposite direction from the wind could actually give a larger pitch angle, depending on the shape of the platform. That might not be the case here, though.

P15: Line 271 – subject-verb agreement

P20: I’m not convinced that aerodynamic damping is the right explanation for the response to waves. Damping is important when resonance is involved. Maybe aerodynamic damping is important for the resonant responses in the tower that are excited by waves, but those are high frequencies where there is little wave energy. At the main wave frequencies, there are no natural frequencies (by design). I suspect that the massive platform and the choice of ranges in the inputs have a greater effect here. Similarly, P22, it is note that “wind turbulence has high energy at low frequencies” – where we also have resonance, which makes these variations particularly important.

P27: I would distinguish between watch circle and device footprint. These are two different things, as the footprint on the seabed will reach to the anchors regardless.