the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Development of damage curves for buildings near La Rochelle during storm Xynthia based on insurance claims and hydrodynamic simulations
Manuel Andres Diaz Loaiza
Jeremy D. Bricker
Remi Meynadier
Trang Minh Duong
Rosh Ranasinghe
Sebastiaan N. Jonkman
Download
- Final revised paper (published on 08 Feb 2022)
- Preprint (discussion started on 18 Jun 2021)
Interactive discussion
Status: closed
-
CC1: 'Comment on nhess-2021-161', Xavier Bertin, 30 Jun 2021
In this paper by Loaiza et al., the authors apply the Delft3D model to hindcast the sea state, water levels and flooding depths associated with the storm Xynthia, to further derive damage curves in the region of La Rochelle, France. Although this topic is relevant in a context of increasing flooding risks due to sea-level rise and increase in population along the coasts, the authors by-passed several studies explaining the key mechanisms that drove the large surge associated with Xynthia. As a consequence, their model underestimates the maximum water levels reached during Xynthia by almost half a meter in la Rochelle (and not 0.36 m a stated in their paper, there is also an error in the vertical referencing). Below is a list of suggestions that could contribute to improve this paper:
-Replacing Xynthia in the context of other storms in the Bay of Biscay would be useful. Breilh et al. (2014) reviewed the major flooding events that affected this region over the last centuries while Bulteau et al. (2015) performed a detailed statistical analysis of the return period associated with the water level reached during Xynthia in La Rochelle.
- In addition to the phasing between between the surge peak and the high spring tide, the key point of Xynthia was that the particular track of the storm from SW to NE induced a young sea state, which strongly enhanced the surface stress and drove a surge abnormally high with respect to the wind speed. In Bertin et al. (2015), one can see that using a bulk parameterization to compute the surface stress (e.g. Pond and Pickard, 1983) results in an underestimation of the peak surge by 0.4 m, as in the present study. As Delft3D is already coupled with the SWAN model, the authors could easily use a wave-dependent parameterization to compute the surface stress, such as the one proposed by Donelan et al. (1993).
- In Bertin et al. (2014), we performed a high resolution hindcast of the flooding associated with Xynthia, using a unique unstructured grid covering the whole NE Atlantic Ocean with a grid size locally reaching 3 m at the location of the dikes and natural barriers. In this study, we showed that such a fine resolution was required to represent the coastal barriers adequately. We further showed that the major flooding associated with Xynthia lowered the water level seaward by up to 1 m in estuaries, compared to a simulation where the flooding would not be represented. This important result suggests that one-way nesting approaches would result in pessimistic flooding predictions. The authors should better explain their nesting procedure and possibly discuss the limitations of using a one-way nesting if it is the case.
I hope that the authors will find these comments useful, sincerely,
Xavier Bertin
Cited references:
Bertin, X., Li, K., Roland, A., Zhang, Y.J., Breilh, J.F., Chaumillon, E., 2014. A modeling-based analysis of the flooding associated with Xynthia, central Bay of Biscay. Coast. Eng. 94, 80–89.
Bertin, X., Li, K., Roland, A., Bidlot, J.R., 2015. The contributions of short-waves in storm surges: two case studies in the Bay of Biscay. Cont. Shelf Res. 96, 1–15.
Breilh, J.-F., Bertin, X., Chaumillon, E., Giloy, N., Sauzeau, T., 2014. How frequent is storm-induced flooding in the central part of the Bay of Biscay? Global and Planetary Change 122, 161–175.
Bulteau, T., Idier, D., Lambert, J., Garcin, M., 2015. How historical information can improve estimation and prediction of extreme coastal water levels: application to the Xynthia event at La Rochelle (France). Natural Hazards and Earth System Science 15, 1135–1147.
Donelan, M.A., Dobson, F.W., Smith, S.D., Anderson, R.J., 1993. On the dependence of sea surface roughness on wave development. Journal of physical Oceanography 23, 2143-2149.
Pond, S., Pickard, G., 1983. Introductory dynamical oceanography. Butterworth Heinemann.
Citation: https://doi.org/10.5194/nhess-2021-161-CC1 -
AC1: 'Reply on CC1', Manuel Andrés Díaz Loaiza, 08 Jul 2021
In this paper by Loaiza et al., the authors apply the Delft3D model to hindcast the sea state, water levels and flooding depths associated with the storm Xynthia, to further derive damage curves in the region of La Rochelle, France. Although this topic is relevant in a context of increasing flooding risks due to sea-level rise and increase in population along the coasts, the authors by-passed several studies explaining the key mechanisms that drove the large surge associated with Xynthia. As a consequence, their model underestimates the maximum water levels reached during Xynthia by almost half a meter in la Rochelle (and not 0.36 m a stated in their paper, there is also an error in the vertical referencing). Below is a list of suggestions that could contribute to improve this paper:
Response: The reviewer brings up good points, which we respond to point by point. Thank you for pointing out the vertical referencing error. This has been corrected in Figure 4. The corrected maximum value is 4.089 m.
- Replacing Xynthia in the context of other storms in the Bay of Biscay would be useful. Breilh et al. (2014) reviewed the major flooding events that affected this region over the last centuries while Bulteau et al. (2015) performed a detailed statistical analysis of the return period associated with the water level reached during Xynthia in La Rochelle.
Response: The purpose of the work was to develop damage curves by correlating hydrodynamic model results with the insurance claims data available to us. These claims data are only available to us for Xynthia, not the other storms cited. Nonetheless, the reviewer brings up excellent points about previous work done for this region, which we regret omitting in our first draft. Here, we add discussions of the cited works in our introduction.
- In addition to the phasing between between the surge peak and the high spring tide, the key point of Xynthia was that the particular track of the storm from SW to NE induced a young sea state, which strongly enhanced the surface stress and drove a surge abnormally high with respect to the wind speed. In Bertin et al. (2015), one can see that using a bulk parameterization to compute the surface stress (e.g. Pond and Pickard, 1983) results in an underestimation of the peak surge by 0.4 m, as in the present study. As Delft3D is already coupled with the SWAN model, the authors could easily use a wave-dependent parameterization to compute the surface stress, such as the one proposed by Donelan et al. (1993).
Response: This is a good point, and would indeed make the simulation more accurate for high resolution storm surge simulation. However, our simulation was accurate within 20cm of the La Rochelle tide gauge, which is sufficient for the purpose of developing damage functions. Nonetheless, we discuss this drawback to our storm surge simulation as compared to the reviewer’s highly accurate simulations.
- In Bertin et al. (2014), we performed a high resolution hindcast of the flooding associated with Xynthia, using a unique unstructured grid covering the whole NE Atlantic Ocean with a grid size locally reaching 3 m at the location of the dikes and natural barriers. In this study, we showed that such a fine resolution was required to represent the coastal barriers adequately. We further showed that the major flooding associated with Xynthia lowered the water level seaward by up to 1 m in estuaries, compared to a simulation where the flooding would not be represented. This important result suggests that one-way nesting approaches would result in pessimistic flooding predictions. The authors should better explain their nesting procedure and possibly discuss the limitations of using a one-way nesting if it is the case.
Response: this is a very good point. In our simulation, we represent barriers as thin weirs, as indeed they are too narrow to be resolved by the topography data itself. These thin weirs are defined on the flux faces of grid cells, and parameterize overflow of structures of a defined height and discharge coefficient (typically sharp-crested weir) using the weir equation. We also carry out a sensitivity analysis between simulations with and without these structures included. The domain decomposition used is 2-way nesting, not one-way, so the problem cited with 1-way nesting should not occur in our simulations; in the revised paper, we explicitly explain that Delft3D domain decomposition is 2-way. In addition, we discuss the paper suggested by the reviewer in our introduction.
I hope that the authors will find these comments useful, sincerely,
Xavier BertinResponse: In the new version it will be included the aforementioned discussion. Thank you for the observations,
Andres and Jeremy.
Citation: https://doi.org/10.5194/nhess-2021-161-AC1
-
AC1: 'Reply on CC1', Manuel Andrés Díaz Loaiza, 08 Jul 2021
-
RC1: 'Comment on nhess-2021-161', Anonymous Referee #1, 15 Jul 2021
Dear authors,
This is an interesting research on damage curves development based upon insurance damage data and hydrodynamic model results. The methods are clearly presented in the paper. The validation results of hydrodynamic models seems reasonable. The only concern of myself is the proposed standard normal distribution of damage curves because the results of all the damage curves developed in this paper is based on this hypothesis. I was wondering how to validate these damage curves? How will the insurance company utilize the damage curves for further risk analysis?
Other comments:
- a thorough discussion of literature review is missing in the ‘Introduction’. what is the common way of developing damage functions? what do the previous researchers have done? what are the main conclusions of their works? what is the current research gap? what is the scientific contribution of this research? please also explicitly explain the significance of this work.
- Line 49-55 could be moved to the section of ‘Introduction’. These literatures are the damage functions developed by other data and researchers.
- Line 56: ‘In’ à ‘in’
- Line 60: I suggest add the units of these parameters. e.g. Hsig is the significant wave height [m];
- Line 65: Figure 23àFigure 3
- Caption of Table 1 could be ‘Description of three scenarios of topography and bathymetry data used in the model : low resolution (a), high resolution (b), high resolution + structures (c)’.
- I suggest zoom in the study area of Ille du Re and La Rochelle to show the water depth and Hsig. Otherwise the readers cannot get useful information from Figure 6. This figure currently didn’t convey clear information on water depth and wave height.
- Both figure 7 and Table 2 show that the damage curves for water depth and total water depth have good and very similar fitting curves for coarse (GEBCO) and fine (IGN+structure) data. It seems that damage curve is not that sensitive to the topography data for the variable of water depth. I recommend to discuss it in the section of ‘Discussion’.
- I suggest reorganize the conclusion section. The paragraph of uncertainty analysis should be moved to the section of ‘Discussion'.
Citation: https://doi.org/10.5194/nhess-2021-161-RC1 -
AC2: 'Reply on RC1', Manuel Andrés Díaz Loaiza, 19 Sep 2021
Dear authors,
This is an interesting research on damage curves development based upon insurance damage data and hydrodynamic model results. The methods are clearly presented in the paper. The validation results of hydrodynamic models seems reasonable. The only concern of myself is the proposed standard normal distribution of damage curves because the results of all the damage curves developed in this paper is based on this hypothesis. I was wondering how to validate these damage curves? How will the insurance company utilize the damage curves for further risk analysis?
Response: Dear reviewer, thank you for your comments and your observations. I will answer them in the next paragraphs. For the concern related with the use of standard normal distribution for the damage function development we decided to use this function since it is one of the most common statistical distributions used in this scope. Nevertheless, now we decided to apply another statistical distribution in order to compare with the current results. The results will be displayed in the appendix for the newer version together with the comments of Xavier Bertin. With regard to how insurance companies can use the damages curves for future risk analysis the answer is that we are helping to determine which physical quantities (flow velocity, water depth, significant wave height etc..) generate the best correlation with damage data, and this is something insurance companies are willing to know.
Other comments:
- a thorough discussion of literature review is missing in the ‘Introduction’. what is the common way of developing damage functions? what do the previous researchers have done? what are the main conclusions of their works? what is the current research gap? what is the scientific contribution of this research? please also explicitly explain the significance of this work.
Response: In the beginning we decide to be the most direct with the structure and content of the paper, since the structure is intended for people who already have experience in the topic. But based on this observation we decided to include two paragraphs on this matter. Thank you for the observation
- Line 56: ‘In’ à ‘in’
Response: yes, in the new version line 56 is changed
- Line 60: I suggest add the units of these parameters. e.g. Hsig is the significant wave height [m];
Response: we will guarantee that the correspondent units appear in the tables and in all the figures
- Line 65: Figure 23àFigure 3
Response: yes, the figure title is changed in the text of line 65
- Caption of Table 1 could be ‘Description of three scenarios of topography and bathymetry data used in the model : low resolution (a), high resolution (b), high resolution + structures (c)’.
Response: We feel the current caption is explicative enough of the content in table 1.
- I suggest zoom in the study area of Ille du Re and La Rochelle to show the water depth and Hsig. Otherwise the readers cannot get useful information from Figure 6. This figure currently didn’t convey clear information on water depth and wave height.
Response: yes, a better image with a zoom over Ille du Re and La Rochelle will be included
- Both figure 7 and Table 2 show that the damage curves for water depth and total water depth have good and very similar fitting curves for coarse (GEBCO) and fine (IGN+structure) data. It seems that damage curve is not that sensitive to the topography data for the variable of water depth. I recommend to discuss it in the section of ‘Discussion’.
Response: yes, that behaviour was initially detected although the real explanation for this is not clearly understood. Although only IGN for this variable have worst RRSE, RMSE and Pearson coefficient compared to IGN+Structures and GEBCO, these values are good in between the rest of the variables, indicating water depth and total water are good descriptive variables for the damage curves in this case study. Nonetheless, for all the variables the best goodness of fit indices are in the case of IGN+structures (better bathymetry/topography compare to GEBCO).
- I suggest reorganize the conclusion section. The paragraph of uncertainty analysis should be moved to the section of ‘Discussion'.
Response: Thank you again for your comment and all the previous, we will consider this last for the new version.
Citation: https://doi.org/10.5194/nhess-2021-161-AC2
-
RC2: 'Comment on nhess-2021-161', Bret Webb, 17 Aug 2021
The authors present the results of a coupled Delft3D+SWAN hindcast simulation of extratropical storm Xynthia and present damage curves derived through analysis of the hindcast results in combination with insurance claims data near La Rochelle, France. Through their results, the authors demonstrate that grid/mesh resolution can impact the shape of the resulting damage curves, and that the best explanatory variables for damage are water depth and total hydrodynamic force. The authors suggest in their concluding statements that their work may have broad application to assess damage from future events along the French Atlantic coast, but subsequently provide numerous qualifiers on their work that contradict the preceding claim.
This certainly is an interesting piece of work and I believe it has a strong foundation that can be improved upon in subsequent revisions. While the technical focus is appropriate, I found the current version of the manuscript lacking in a few substantial ways. The deviation from standard practice when developing damage curves notwithstanding, the work shows promise and will be an excellent contribution to the published literature with a few substantial improvements to the analyses and manuscript.
My general recommendations for improving the manuscript are as follows:
1) Improve the organization of the manuscript, especially the early sections of the text. There should be a clear and distinct progression from the introduction to the methods. The methods section contains information about the study area and the storm, which would be better presented in the introductory section of the manuscript. The organization of the methods section is inconsistent and could be improved to flow more logically. For example, there is discussion of the models and model setup in multiple places of 2.1, 2.2, and 2.3. Furthermore, this section begins with (cf. 2.1) a detailed discussion of the particular storm event without first describing the storm or the models. Section 2.2 could be combined with another section in the reorganization. One may also argue that the validation results belong in the “Results” section, not in the method section. Section 2.4, albeit brief, is appropriately placed and contains helpful information. I will, however, note that use of the term “damage level x” in line 124 is somewhat inconsistent with your chosen approach and terminology. Also, that “x” is not the damage ratio but rather the value of the conditional variable (the hazard) for a specific damage ratio increment. Therefore, you likely need a subscript on P such that P_i(x)=… gives the probability of experiencing hazard value “x” for damage ratio value “i” and so on. I’m sure that is what you did in the analysis, but the typesetting of Equation 2 and the corresponding text should be improved.
2) The analyses, while well intentioned, are not particularly robust in their presentation. For example, there is no quantitative assessment of model errors or bias in the prediction of either water levels or wave heights. Simply plotting predictions and measurements and saying the agreement is “good” does not inspire confidence, particularly when the disagreement between the two for wave heights appears to be quite substantial. As a second example, there really is not enough information provided relative to the development of the damage curves given its prominence in the title of your manuscript. So while the content of the existing manuscript is strong, it is simply short on details and could benefit from an expanded discussion in many places (a few of which are noted below).
3) There is a duality in the manuscript that I am having a hard time reconciling, particularly given point #2 above (lack of detail). There is a significant emphasis placed on the influence of grid resolution on the resulting damage functions. However, there is not enough supporting detail provided for these grids/meshes. Given that there is similarly a lack of detailed information regarding the development and application of the damage curves (additional comments below), this leaves the manuscript lacking in technical details as mentioned earlier. While the impact/influence of the grid resolution is noteworthy, it does not appear to be the focal point of the paper (not in the title) so I would suggest minimizing its relevance and adding much more detail to the damage curve discussion. Alternatively, if the authors would prefer not to expand the discussion of the damage curves and instead reorient the focus of the paper to one associated with the grid resolution, then consider greatly expanding details regarding the features and characteristics of those grids and perhaps modify the manuscript title accordingly.
4) I have some reservations about your analysis methodology. Not to say that it is in any way “wrong” but it does suffer from a lack of explanation (again, just my opinion). I would like to see some detailed description of the building archetypes considered in this analysis. Are all buildings considered to be of the same archetype (I assume so because there is no differentiation in the results)? Can you provide more details beyond “stone masonry” such as number of floors/heights, foundation types, age of structures, roof types/materials, etc.? Without the qualifiers, I think it would be very easy for someone to misapply your methodology. Also, I would like to see a better presentation of the explanatory variable (hazard) values for the damage curves. I know that you have presented them graphically in the appendix, but it would be valuable to also list the means and standard deviations (likely for only one grid) of those variables/variable groups. Finally, can you add some discussion regarding potential weaknesses of your chosen “damage ratio” approach to representing damage? There are many weaknesses with using this as a substitute for the more common “damage state” because the damage ratio does not correct for valuation based on location among other weaknesses. As another example, “insured value” is often a personal/elective choice made by the homeowner and there is bound to be substantial inconsistency in what one chooses to insure their property for. To expand a bit further, a low damage ratio value may be the result of minimal damage or a very high insured value. Therefore, the damage ratio is sensitive to two metrics, one of them choice-based, as opposed to a traditional damage state classification which, while somewhat objective, focuses only on the severity of damage to the property. My primary concern here is founded upon the fact that nearly one-half of your 423 reported claims have an assigned damage ratio <0.1 (cf. Figure 1). Finally, in a traditional damage/fragility analysis one would also consider structures with no damage. I do not recall any mention of non-damaged structures in your analysis. Therefore, the resulting damage curves may very well be biased.
Here are some additional comments that address specific items in the manuscript…
Line 30/Figure 1: recommend normalizing the ordinate values by the total number of claims so that you can report these in terms of their true “frequency” instead of simply counts. If not, please edit the axis title as these are not frequencies.
Line 38: the introduction in its current form is significantly lacking in terms of a thorough review of pertinent literature on damage functions derived from coastal hazard models (e.g., Masoomi et al., 2019 and many others), lacks an orientation to the study area, and does not thoroughly describe the storm event. I would recommend adding:
-much more background on relevant literature
- a detailed description of study area with location map, exposure/vulnerability to extreme events, hydrodynamic setting, etc.
- more information on the history and characteristics of Xynthia
Masoomi, H., van de Lindt, J.W., Do, T.Q., Webb, B.M. 2019. Combined wind-wave-surge hurricane-induced damage prediction for buildings. Journal of Structural Engineering 145(1).
Line 41: minor comment but use consistent typesetting of “Delft3D” throughout the document.
Line 48/Figure 2: any reason why there is a font change in this graphic? Was that intentional?
Lines 49-69: I find it odd that you are interjecting more literature review here as opposed to providing it earlier in the document.
Line 63: missing comma… “… storm characteristics, a regional model…”
Line 65: can a resolution of 80 meters accurately capture terrain features and individual homes?
Line 69/Figure 3: There is not enough contrast in this image to make out the details. The date/time codes for every storm report make the figure unnecessarily busy.
Line 73: what data sources did you use for land cover / land use to assign friction coefficients? (nb. Ignore this comment, I see the answer on line 99).
Lines 75-79/Table 1: how do these relate to the 80-meter resolution mentioned previously?
Line 83: “… spatial resolution and temporal every 3hrs.” awkward phrasing
Line 99: use of a constant Manning’s “n” value for the entire grid is a significant technical weakness in this study. While it “may” be appropriate for some open water conditions, it is certainly not reflective of the terrain where the subject structures were likely found. Could you please provide a justification and suitable citation to support the use of a constant friction factor? I have read numerous papers in the past ten years that point to the importance of accurate representation of terrain roughness through the assignment of proper friction coefficients.
Line 105: what is SHOM?
Line 105: “… during the whole simulation is good (Figure 4).” Is good relative to what? There is no quantitative basis for this statement.
Line 110/Figure 4: Can you please explain the spurious oscillation at the beginning of your simulation results? I incorrectly assumed that this was the surge event, but it appears to be a numerical instability associated with model spinup.
Line 115/Figure 5: there is absolutely no assessment or narrative to accompany these results. Since wave properties are highlighted as one of your preferred explanatory variables in the damage analysis, and since wave processes contribute to coastal flooding (your other explanatory variable), can you provide some commentary on the disagreement between the modeled and measured waves? Why are the observations listed as “swell height”? Are you comparing two different wave statistics in this figure (i.e., swell height and significant wave height)?
Line 121: “… specifically, relates the…” typo and awkward phrasing
Line 127: missing commas after “paper” and “way”
Lines 132-138: Damage curves are often given for different levels/magnitudes of damage. Here it appears that you are integrating the damage results across all of the discrete damage ratio increments. Was this a specific choice/preference, an artifact of your damage indicator scheme, or something else entirely? Was there no interest in disaggregating the damage data as is traditionally done in these types of analyses? For example, developing unique damage curves for different damage states?
Line 133: I don’t think “Box-Whisker” is capitalized, not proper nouns
Line 136: “Damage” should not be capitalized
Line 142: “Where” should not be capitalized since you are using the equations as the subject of your sentence.
Line 145/Figure 6: I find these figures to be less helpful than I had hoped. The size (small) and contouring scheme do not allow for much interpretation of the results.
Line 150: “Similarly” -> “Similar”
Line 152/Figure 7: at the scale provided it is difficult to discern details in these figures. Also, there is no explanation of the symbols in these figures.
Line 157: “related with the” -> “related to the”
Line 160/Table 2: For explanatory values (cf Table 2) used in Eq 2, how were means and standard deviations evaluated for combinations of variables that do not necessarily vary consistently in time? In other words, did you estimate the time-variation of each variable group/combination and then take the mean and standard deviation of the entire time-series? Or, did you evaluate the mean/stdev of each individual parameter and then form the variable groups?
Line 164: typesetting of “hsig” -> “Hsig”
Line 207: delete comma after point
Line 215: “… as thin or concrete structures like flood walls at typically only a few 10’s of centimeters thick, and so do not appear in digital elevation models.” Awkward phrasing.
Lines 215-220: what about errors/uncertainty in your model predictions?
I commend the authors on a very strong first draft of what I'm sure was a very challenging manuscript to prepare. The authors are absolutely on track towards having a very strong publication that will productively add to the body of literature on damage to coastal structures during extreme events.
Sincerely,
Bret Webb
Citation: https://doi.org/10.5194/nhess-2021-161-RC2 -
AC3: 'Reply on RC2', Manuel Andrés Díaz Loaiza, 19 Sep 2021
The authors present the results of a coupled Delft3D+SWAN hindcast simulation of extratropical storm Xynthia and present damage curves derived through analysis of the hindcast results in combination with insurance claims data near La Rochelle, France. Through their results, the authors demonstrate that grid/mesh resolution can impact the shape of the resulting damage curves, and that the best explanatory variables for damage are water depth and total hydrodynamic force. The authors suggest in their concluding statements that their work may have broad application to assess damage from future events along the French Atlantic coast, but subsequently provide numerous qualifiers on their work that contradict the preceding claim.
This certainly is an interesting piece of work and I believe it has a strong foundation that can be improved upon in subsequent revisions. While the technical focus is appropriate, I found the current version of the manuscript lacking in a few substantial ways. The deviation from standard practice when developing damage curves notwithstanding, the work shows promise and will be an excellent contribution to the published literature with a few substantial improvements to the analyses and manuscript.
Response: Dear Bret, thank you for your comments, we will certainly improve the manuscript with your comments and the other reviewers/participants on the interactive discussion forum. Down below I will answer point by point your comments.
My general recommendations for improving the manuscript are as follows:
- Improve the organization of the manuscript, especially the early sections of the text. There should be a clear and distinct progression from the introduction to the methods. The methods section contains information about the study area and the storm, which would be better presented in the introductory section of the manuscript. The organization of the methods section is inconsistent and could be improved to flow more logically. For example, there is discussion of the models and model setup in multiple places of 2.1, 2.2, and 2.3. Furthermore, this section begins with (cf. 2.1) a detailed discussion of the particular storm event without first describing the storm or the models. Section 2.2 could be combined with another section in the reorganization. One may also argue that the validation results belong in the “Results” section, not in the method section. Section 2.4, albeit brief, is appropriately placed and contains helpful information. I will, however, note that use of the term “damage level x” in line 124 is somewhat inconsistent with your chosen approach and terminology. Also, that “x” is not the damage ratio but rather the value of the conditional variable (the hazard) for a specific damage ratio increment. Therefore, you likely need a subscript on P such that P_i(x)=… gives the probability of experiencing hazard value “x” for damage ratio value “i” and so on. I’m sure that is what you did in the analysis, but the typesetting of Equation 2 and the corresponding text should be improved.
Response: We think that the best way to structure the paper is to start mentioning the economic damages and the damage functions development, instead of focusing on the storm Xynthia and the hydrodynamic simulation. Indeed, it is mentioned below that we should give more importance on the damage function development than the hydrodynamic simulations. Nevertheless, due to your comment and those from reviewer 1, we decide to include another paragraph about “the common way of developing damage functions? what do the previous researchers have done? what are the main conclusions of their works?” and move some part from the conclusions to the discussion. A short sentence with reference on the introduction related with the Xynthia storm was added too
Related with the explanation of the damage function, we agreed that now is not clear the explanation. In the new version, the text will be as follows: “where P(x) is the cumulative probability of the damage ratio with values between 0 and 1, and x is the hydrodynamic variable, Φ is the standardized normal distribution, μ is …”.
2) The analyses, while well intentioned, are not particularly robust in their presentation. For example, there is no quantitative assessment of model errors or bias in the prediction of either water levels or wave heights. Simply plotting predictions and measurements and saying the agreement is “good” does not inspire confidence, particularly when the disagreement between the two for wave heights appears to be quite substantial. As a second example, there really is not enough information provided relative to the development of the damage curves given its prominence in the title of your manuscript. So while the content of the existing manuscript is strong, it is simply short on details and could benefit from an expanded discussion in many places (a few of which are noted below).
Response: In the newer version we are including a goodness of fit indexes (RMSE, Pearson coefficient and RRMSE), between the observed tide and waves values and we compare these values with others values obtained in the literature.
- There is a duality in the manuscript that I am having a hard time reconciling, particularly given point #2 above (lack of detail). There is a significant emphasis placed on the influence of grid resolution on the resulting damage functions. However, there is not enough supporting detail provided for these grids/meshes. Given that there is similarly a lack of detailed information regarding the development and application of the damage curves (additional comments below), this leaves the manuscript lacking in technical details as mentioned earlier. While the impact/influence of the grid resolution is noteworthy, it does not appear to be the focal point of the paper (not in the title) so I would suggest minimizing its relevance and adding much more detail to the damage curve discussion. Alternatively, if the authors would prefer not to expand the discussion of the damage curves and instead reorient the focus of the paper to one associated with the grid resolution, then consider greatly expanding details regarding the features and characteristics of those grids and perhaps modify the manuscript title accordingly.
Response: Yes, indeed the idea of the paper is not to focus on the mesh resolution. Instead, in the current paper we investigate three different bathymetry/topography data sets, each with its own resolution and quality.
4) I have some reservations about your analysis methodology. Not to say that it is in any way “wrong” but it does suffer from a lack of explanation (again, just my opinion). I would like to see some detailed description of the building archetypes considered in this analysis. Are all buildings considered to be of the same archetype (I assume so because there is no differentiation in the results)? Can you provide more details beyond “stone masonry” such as number of floors/heights, foundation types, age of structures, roof types/materials, etc.? Without the qualifiers, I think it would be very easy for someone to misapply your methodology.
Response: For this initial part of point 4 unfortunately no more information was available (such as the types of structures, number of floors and damage stages), even more the location of the claims due to the new European data protection policy were masked, making the extraction of the hydrodynamic variables an iterative process between the insurance company and TU-Delft.
Also, I would like to see a better presentation of the explanatory variable (hazard) values for the damage curves. I know that you have presented them graphically in the appendix, but it would be valuable to also list the means and standard deviations (likely for only one grid) of those variables/variable groups. Finally, can you add some discussion regarding potential weaknesses of your chosen “damage ratio” approach to representing damage? There are many weaknesses with using this as a substitute for the more common “damage state” because the damage ratio does not correct for valuation based on location among other weaknesses. As another example, “insured value” is often a personal/elective choice made by the homeowner and there is bound to be substantial inconsistency in what one chooses to insure their property for. To expand a bit further, a low damage ratio value may be the result of minimal damage or a very high insured value. Therefore, the damage ratio is sensitive to two metrics, one of them choice-based, as opposed to a traditional damage state classification which, while somewhat objective, focuses only on the severity of damage to the property. My primary concern here is founded upon the fact that nearly one-half of your 423 reported claims have an assigned damage ratio <0.1 (cf. Figure 1). Finally, in a traditional damage/fragility analysis one would also consider structures with no damage. I do not recall any mention of non-damaged structures in your analysis. Therefore, the resulting damage curves may very well be biased.
Response: Regarding the means and the standard deviations we believe that in order to maintain an adequate extension of the paper we should not include another table since they are already displayed in the box-whisker plots. For the newer version, we add a paragraph commenting on the uncertainty that the quality of the claims add to the analysis, since unfortunately cannot be assessed. As you commented, the insured value, or the economic claim can be exaggerated by the owners. About the 423 reported claims and the distribution of their damage ratio, as you mentioned, is a matter that involves the quality of the claims itself, but also the hydrodynamics conditions, that are particular for every case. In the case of La Rochelle and more noticeable for Ille du Re, there are structures at the foreshore, but usually the more densely populated areas are some hundreds of meters inland where elevation is higher and then less damages is expected. Finally, including non-damaged structures indeed will add value into this kind of analysis; however we do not possess this data.
Here are some additional comments that address specific items in the manuscript…
Line 30/Figure 1: recommend normalizing the ordinate values by the total number of claims so that you can report these in terms of their true “frequency” instead of simply counts. If not, please edit the axis title as these are not frequencies.
Response: Yes, the figure is changed in the new version.
Line 38: the introduction in its current form is significantly lacking in terms of a thorough review of pertinent literature on damage functions derived from coastal hazard models (e.g., Masoomi et al., 2019 and many others), lacks an orientation to the study area, and does not thoroughly describe the storm event. I would recommend adding:
-much more background on relevant literature
- a detailed description of study area with location map, exposure/vulnerability to extreme events, hydrodynamic setting, etc.
- more information on the history and characteristics of Xynthia
Masoomi, H., van de Lindt, J.W., Do, T.Q., Webb, B.M. 2019. Combined wind-wave-surge hurricane-induced damage prediction for buildings. Journal of Structural Engineering 145(1).
Response: Yes, this reference and others, both together with a better explanation for the damage functions and Xynthia storm is added.
Line 41: minor comment but use consistent typesetting of “Delft3D” throughout the document.
Response: Yes, thank you is amended.
Line 48/Figure 2: any reason why there is a font change in this graphic? Was that intentional?
Response: Fonts on the figures are checked and corrected
Lines 49-69: I find it odd that you are interjecting more literature review here as opposed to providing it earlier in the document.
Response: As commented before more references are added in the beginning of the document
Line 63: missing comma… “… storm characteristics, a regional model…”
Response: Yes, Thank you is amended.
Line 65: can a resolution of 80 meters accurately capture terrain features and individual homes?
Response: Homes and small terrain features are not included in the grid resolution but flood walls were included by means of thin weirs in the model
Line 69/Figure 3: There is not enough contrast in this image to make out the details. The date/time codes for every storm report make the figure unnecessarily busy.
Response: Yes, some labels were removed and the figure was modify.
Line 73: what data sources did you use for land cover / land use to assign friction coefficients? (nb. Ignore this comment, I see the answer on line 99).
Response: See comments on line 99.
Lines 75-79/Table 1: how do these relate to the 80-meter resolution mentioned previously?
Response: As commented above, the intention is to compare different bathymetry/topography information data. Resolution of this information is variable from 500m (GEBCO) to 5m (IGN topography). The mesh of 80m can produce different results depending on the type of the information.
Our analysis consists of three domains because we implemented the model in domain decomposition mode, but this is not related to our analysis of the data resolution. For each DEM (GEBCO, IGN, IGN +Structures) we implemented that DEM in all of our domains.
Line 83: “… spatial resolution and temporal every 3hrs.” awkward phrasing
Response: Ok, the sentence has been changed
Line 99: use of a constant Manning’s “n” value for the entire grid is a significant technical weakness in this study. While it “may” be appropriate for some open water conditions, it is certainly not reflective of the terrain where the subject structures were likely found. Could you please provide a justification and suitable citation to support the use of a constant friction factor? I have read numerous papers in the past ten years that point to the importance of accurate representation of terrain roughness through the assignment of proper friction coefficients.
Response: We do agree that the roughness friction coefficients have an influence on the hydrodynamic results, particularly in places of shallow waters, but in order to simplify the analysis as was done in Tomiczek T., 2017, we decide to keep it constant. A small paragraph on the discussion is added regarding this topic.
Line 105: what is SHOM?
Response: A web link is added. Is an institute in charge of the monitoring of the north Atlantic bouys in front of the French and Spanish coast.
Line 105: “… during the whole simulation is good (Figure 4).” Is good relative to what? There is no quantitative basis for this statement.
Response: A table with the goodness of fit indicators is added. Also a small explanation on the reason by which the significant wave height is compared to the swell height is done.
Line 110/Figure 4: Can you please explain the spurious oscillation at the beginning of your simulation results? I incorrectly assumed that this was the surge event, but it appears to be a numerical instability associated with model spinup.
Response: Since at the time of the beginning of the simulation the water levels in the whole domain are unknown, it is common to set as 0 m.a.s.l the water elevation as initial condition. This produces the need to extend the simulation and start it before the day from which we want to make the analysis. How long is this period (or spin-up time) depends on the size of the domain and time interval of the simulation. In our case we start the simulation from the 20 of February of 2010 guaranteeing enough time for the model to adjust the tide values to the real ones.
Line 115/Figure 5: there is absolutely no assessment or narrative to accompany these results. Since wave properties are highlighted as one of your preferred explanatory variables in the damage analysis, and since wave processes contribute to coastal flooding (your other explanatory variable), can you provide some commentary on the disagreement between the modeled and measured waves? Why are the observations listed as “swell height”? Are you comparing two different wave statistics in this figure (i.e., swell height and significant wave height)?
Response: Yes, thank you for this observation. An explanation is added both together with the goodness of fit indices for the waves and tides are added.
Line 121: “… specifically, relates the…” typo and awkward phrasing
Response: Ok, the sentence has been changed.
Line 127: missing commas after “paper” and “way”
Response: Ok, the sentence has been changed.
Lines 132-138: Damage curves are often given for different levels/magnitudes of damage. Here it appears that you are integrating the damage results across all of the discrete damage ratio increments. Was this a specific choice/preference, an artifact of your damage indicator scheme, or something else entirely? Was there no interest in disaggregating the damage data as is traditionally done in these types of analyses? For example, developing unique damage curves for different damage states?
Response: Unfortunately we do not have enough data to disaggregate.
Line 133: I don’t think “Box-Whisker” is capitalized, not proper nouns
Response: Thank you for the observation.
Line 136: “Damage” should not be capitalized
Response: Thank you for the observation.
Line 142: “Where” should not be capitalized since you are using the equations as the subject of your sentence.
Response: Thank you for the observation.
Line 145/Figure 6: I find these figures to be less helpful than I had hoped. The size (small) and contouring scheme do not allow for much interpretation of the results.
Response: Figure 6 is changed modifying the colours, the legend and making a zoom over the simulation domain.
Line 150: “Similarly” -> “Similar”
Response: Thank you for the observation
Line 152/Figure 7: at the scale provided it is difficult to discern details in these figures. Also, there is no explanation of the symbols in these figures.
Line 157: “related with the” -> “related to the”
Response: Thank you for the observation
Line 160/Table 2: For explanatory values (cf Table 2) used in Eq 2, how were means and standard deviations evaluated for combinations of variables that do not necessarily vary consistently in time? In other words, did you estimate the time-variation of each variable group/combination and then take the mean and standard deviation of the entire time-series? Or, did you evaluate the mean/stdev of each individual parameter and then form the variable groups?
Response: For every single claim at the Ille du Re and surroundings the maximum hydrodynamic variables were extracted at every single location. In this way the time series of the simulations are restricted for the maximum values that reach every single variable. A short line explaining this is added in the damage curve section.
Line 164: typesetting of “hsig” -> “Hsig”
Response: Thank you for the observation.
Line 207: delete comma after point
Response: Thank you for the observation.
Line 215: “… as thin or concrete structures like flood walls at typically only a few 10’s of centimeters thick, and so do not appear in digital elevation models.” Awkward phrasing.
Response: Yes, the phrase is adjusted.
Lines 215-220: what about errors/uncertainty in your model predictions?
Response: Thank you for the comment. It was added along the document at the meteorological and damage functions some paragraphs about the uncertainty coming from the hydrodynamic modelling and from the claims data itself in the document.
I commend the authors on a very strong first draft of what I'm sure was a very challenging manuscript to prepare. The authors are absolutely on track towards having a very strong publication that will productively add to the body of literature on damage to coastal structures during extreme events.
Sincerely,
Bret Webb
Response: We certainly acknowledge your comments on the document, thank you for reading the document and give your feedback. We hope the present version is more accurate.
Andres et al.,
Citation: https://doi.org/10.5194/nhess-2021-161-AC3
-
AC3: 'Reply on RC2', Manuel Andrés Díaz Loaiza, 19 Sep 2021