Review of the paper “Regional analysis of multivariate compound flooding potential: sensitivity analysis and spatial patterns” submitted to NHESS by Camus et al.

This is a well-written manuscript that analyses the compound flooding potential in Europe considering 4 drivers: precipitation, river discharge, storm surge and waves. To my knowledge, it is the first study that analyses these 4 drivers altogether at European scale, therefore providing meaningful insight about compound flooding potential. This study also provides a uselful sensitivity assessment to different choices that can be made when analysing compound events, such as sampling method, time window, etc. Lastly, they derive a severity index that combines the effects of all drivers into one index to explain potential coastal flooding, which is later analyzed with self organizing maps and kmeans cluster analysis to identify hotspots of potential flooding.

I recommend its publication after addressing the comments below (minor revision).

Conceptual aspects:

Line 17 “with homegeneous forcing” (also in line 97). I think the term “homogeneous” is not correct here. Modern reanalysis are subject to temporal inhomogeneities due to increasing amount and type of assimilated data over time. I suggest to

replace it by “with the same forcing” or “with coherent forcing”, I would also remove the “temporally consistent” part in line 105.

Line 74-83. It should also be mentioned that, if information about the impact is available, another option is conditional sampling on the impact variable, ie analyse the behaviour of the drivers/predictors when the impact is extreme. 

Line 180: “ignoring the astronomical tidal component of sea level, as it is deterministic”.  I am not sure this is a valid argument. It is true that astronomical tides are deterministic but the timing in which large tide might occur respective of, for example, large storm surges is not. I suggest removing the deterministic part or change it to “which is deterministic” or similar.

Line 185-186 I am a bit confused about these pairs. I thought that waves were characterized in terms of SW (simplified wave contribution to water level), and WL (wave set up including the effect of Tp + surge). First, why do you consider surge together with wave set up when is calculated considereing Tp, but then wave set up and surge are separate when using the simpler formula? Then why WL does not appear as part of the pair of driver combinations analysed? Maybe I got something wrong but I suggest clarifying this in the text.

Line 204-205: I suggest listing threshold as the first parameter introduced in POT. Then I would keep the other factor more general, such us “criteria used to select independent storms” (it is not always just one time window between peaks, alternative or complementary times can be used to determine if two storms can be considered independent, such as the time below the threshold between storms) 

Section 4.1.1 What is the time window used to assess the sampling sensitivity?

Figure 3. As noted by the authors, it is expected that the number of co-ocurring events are larger for POT6 than POT3, and POT3 than AM, as the number of total events differs for each sampling method. I suggest to show a normalized number of co-ocurring events, which might provide more interesting information.

Line 447. Why is W multiplied by 0.2 if afterwards is normalized?

Section 5. I acknowledge the authors provide an extended discussion of many of the results. However, I would discuss a bit more the results obtained, such as why correlation with AM is larger than with POT. Could it be because the sample is shorter? Also, why the correlation is lower when using a wave set up formulation that includes Tp? Influence of remote swells that are not correlated with local storms?

Formal aspects:

Line 48: add space ; and space between Couasnon and Ridder citations.

Line 53: This paragraph is quite long. I suggest dividing it in two, for example just before “High-dimensional systems can be modelled using…” in line 60.

Figure 1 is hard to see (specially panel a). I suggest making the figure larger.

Line 182: I suggest adding in the manuscript the exact formula used to calculate wave set up.

Figure 2. Figure is too small, and I also suggest making the legends smaller so there is not overlap with the histograms bars.

Figure 6. I suggest to clarifying the difference of what is substracted from what. Perhaps add subpanels titles.

Figure 7. Isn’t there overlapping of red and blue colours when the join occurences (expressed as circle size) are similar for both combinations of pairs?

Section 4.2.2. I believe Figure S8 should be Figure 8 (line 455, 457 and 462)

Line 606: remove “found” after Coausnon et al. (2020)

This study on compound floods in Europe has two main aspects, one is investigating how sensitive are the compound flood estimates to the choice of method, and second is finding a synthetic measure of their severity. In both cases the paper provides meaningful results and I think will be an important work of reference. The manuscript is rather clear and concise, even though the presented findings are quite technical and not easy to synthesize. Overall I didn’t really see any major issues with the study beyond those indicated by Referee #1.

The only thing I really found missing is that no reference is made to potential bias or overarching inaccuracy in the models in reproducing compound events, method used notwithstanding. As Ganguli et al. (2020) and Paprotny et al. (2020) have shown, the dependency between compound flood drivers produced by models differ considerably from dependencies computed from observations. For north-western Europe, where compound events are potentially most frequent and most severe there was a positive bias in the models used in the studies. I think the authors should comment on this aspect in the discussion in context of their calculation of the severity index and clusters.

Below I list a few minor textual points. I’m looking forward to the authors’ revision of the paper.

Minor points:

L86: Paprotny et al. (2020) considered four drivers as well.

L180-185: I concur with reviewer #1 that this paragraph is very unclear as to how storm surge and waves are combined, and reading 3.1.4 doesn’t help much, until actually reading the results. I suggest to use a single letter to define the combined surge-wave height (e.g. H or L) and then differentiate the two methods for computing the joint height as e.g. H1 and H2, so that the relevant pairs are referred as e.g. Q-H [H-Q].

L226: “between 1 and 10 days or 1 and 3 days”. I think would be clearer to write “of ±10 and ±3 days”.

L485 & L510: in “JQ(Q-P-SW)I” it should be “JO”, I guess?

L540: the ‘large’ labels of the vertical axis, indicating the pairs do not clearly connect with their variants in smaller fonts, and further are not aligned with the caption which has some undefined abbreviations e.g. “Co”. Please check that and synchronize the naming with the text.

Supplement: Figure S7’s caption indicates that Kendall’s correlation is used, yet the panels are labelled with “rho”, suggesting Spearman’s correlation. Please check what data is actually shown.

References:

Ganguli P., Paprotny D., Hasan M., Güntner A., Merz B. 2020. Projected changes in compound flood hazard from riverine and coastal floods in Northwestern Europe. Earth’s Future 8(11):e2020EF001752, doi:10.1029/2020EF001752

Paprotny D., Vousdoukas M.I., Morales Nápoles O., Jonkman S.N., Feyen L. 2020. Pan-European hydrodynamic models and their ability to identify compound floods. Natural Hazards, 101(3), 933–957, doi:10.1007/s11069-020-03902-3