Importance of Pre-Storm Morphological Factors in Determination of Coastal Highway Vulnerability
Round 1
Reviewer 1 Report
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Overall a well written manuscript that elaborates on prior research on barrier island morphological factors affecting roadway risks to storm impacts. The paper depends rather highly on a prior paper that documents the dataset created. The paper is also slightly ambiguous or conflating the model prediction of risks versus machine learning applied is primarily a classification process. The study area specific extent is somewhat vague and figure 1 is unclear if it exactly matches (was HWY 12 full extent studied or just Pea I. Natl. Wildlife Refuge?) See specific comments. Nonetheless, the representative risks of different storm types and morphological factors make this a valuable study in the literature.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 3 Report
Thank for the opportunity to review this interesting and well-written manuscript. Pesantez et al present the results of machine learning algorithms on post-storm impacts to NC Highway 12 along the northern Outer Banks, and use these results to compare machine learning to other methods of assessing post-storm damages.
The paper does a good job of explaining the study rationale and quantitative methods, but could do a better job of interpreting results between this and other studies as well as between elements of this study. I recommend this paper be published with some revisions, which I will list below. First I will include some major commentary, which is mostly concerned with how the results are described and presented, with line-by-line edits following.
MAJOR COMMENT 1:
The manuscript is very thorough in explaining the setup and implementation of the machine learning algorithms, but does not really get into the implications of these results. To that point, the discussion section is rather short and there are no final conclusions. Additionally, the methods section is thorough in its coverage of statistical methods but does not necessarily translate these concepts in a way that is easily digested by a coastal scientist. As a result, the manuscript reads a bit more like a technical communication than a full-length article. Not to worry, however - I’d imagine this could be remedied with a few relatively easy changes.
First, including more detail comparing these results to those presented by Behr et al (2022). As a reader, I am interested to know how methodological differences between their approach and yours may have impacted the difference in results. You do a good job of discussing the differences in biasing between various methods, I recommend continuing that sort of language to include comparisons between your methods and those of Behr et al.
Second, providing more qualitative interpretation would be useful for readers unfamiliar with machine learning and/or storm impacts. The methods section is quite thorough and technically proficient, but discussion of the practical meaning of these statistics and parameters is relatively scarce. Even a cursory explanation of what, for example, the F-1 score really means would be helpful. Including this sort of nuanced language in your methods and/or results may help the reader understand, for instance, how significant the F-1 difference between Behr et al and this study really is.
Third, there is virtually no discussion of why certain geomorphic variables represent a more predictor of post-storm damages. While the distance from road EOP to dune toe parameter is an understandable candidate for more impactful, dune crest elevation placing last is a bit surprising…why do you think that is the case? How do these findings align or contrast with empirical observations of storm impacts? Fortunately, there is a wealth of information in both scientific and popular literature on storm impacts and the Outer Banks, so you shouldn’t have much trouble finding good resources for this type of information.
LINE-BY-LINE EDITS:
Line 86: please define what is meant by ‘traditional optimization techniques’
Lines 109-111: while I understand these methods are described in greater detail in Behr et al, perhaps providing a visual example of what a ‘road impact’ represents would be useful. Or, defining a specific threshold for counting an impact.
Lines 137-139: this sentence is a bit confusing, I think you could move ‘The two inputs required to train…” to the beginning and it would read more smoothly.
Line 337: this is related to my major comment, but I’d be curious to know how much of difference these ranges in F-1 values represent. I understand a quantitative improvement is what it is, but presenting the nuance and meaning of these differences could add to the quality of the paper.
Line 339: again, this is related back to my major comment, but some more interpretation and/or information on the qualitative meaning of these results would be helpful to determine whether or not EOP to dune toe is in fact a ‘powerful and reliable predictor’. Obviously the results here speak for themselves, but including some language exploring these statements would add to the quality of the paper in my opinion.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
It would be handy to make a sketch with a cross-section of a coastal profile indicating the predictors used (e.g. the ones indicted in fig 6). For example it is not clear from the paper if the road is on the on the seaward side of the dune toe, or somewhere in the dune. Also dune toe elevation is measured in respect to NAVD83. But for non-American readers this is not clear. What is the height of MSL and the storm surge above NAVD83? Ocean Shoreline is HW line or LW line?
After also reading your paper [5] it became more clear. It might be a good idea to copy fig 5 from paper [5] into this paper. From that paper it became clear to me that your road is on the bay-side of the dune, and not on the ocean side.
What I miss in general is a relation between the load (storm intensity, storm level, waves) and the damage (e.g the work of Hanson and Larson). Als much research has been done by many authors about breaching and overwash of barrier islands.
The present paper focuses only on statistics based on a limited number of cases. What is called “machine learning” in this paper is in fact a clever curve-fitting of a limited dataset. But because there is no relation made between the risk of occurrence of a given storm, also nothing can be said about the probability of serious damage to the road. A relation to e.g. fig 7 in your paper [5] would be useful.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report
No further commeng
Author Response
This reviewer has provided no further comment therefore we have not included an additional point by point response. We appreciate the reviewers' helpful input in the first round.
