the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 11 Dec 2023
Validated probabilistic approach to estimate flood direct impacts on the population and assets on European coastlines
Abstract. This work presents the approach used to estimate coastal flood impact, developed within the EU H2020 European Coastal Flood Awareness System (ECFAS) Project, for assessing flood direct impacts on population, buildings, and roads along the European coasts. The methodology integrates object-based and probabilistic evaluations to provide uncertainty estimates for damage assessment. The approach underwent a user-driven co-evaluation process, it was applied to 16 test cases across Europe and validated against reported impact data in three major reference cases. A comparison with grid-based damage evaluation methods was also conducted. The findings demonstrate that the ECFAS Impact approach offers valuable estimates for affected populations, reliable damage assessments for buildings and roads, and improved accuracy compared to traditional grid-based approaches. The methodology also provides information for prevention and preparedness activities, facilitates further evaluations of risk scenarios and cost-benefit analysis of disaster risk reduction strategies. The approach is a tool suitable for large-scale coastal flood impact assessments, offering improved accuracy and operational capability for coastal flood forecasts. It represents a potential advancement of the existing EU-scale impact method used by the European Flood Awareness System (EFAS) for riverine flood warnings. The integration of object-based and probabilistic evaluations, along with uncertainty estimation, enhances the understanding and management of flood impacts along the European coasts.
- Preprint
(5601 KB) - Metadata XML
- BibTeX
- EndNote
Interactive discussion
Status: closed
-
RC1: 'Comment on nhess-2023-197', Anonymous Referee #1, 19 Jan 2024
The manuscript “Validated probabilistic approach to estimate flood direct impacts on the population and assets on European coastlines” aims to develop and validate the ECFAS Impact model, a comprehensive approach for assessing the impacts of coastal flooding on buildings and infrastructure. The model integrates probabilistic evaluations, flood damage curves, and ensemble methods to simulate damage to residential, commercial, and industrial structures. The objectives include validating the model by comparing its simulated impacts with reported data from historical storm events in e.g. France, the United Kingdom, and Spain. The study assesses the model's performance in accurately representing damage to buildings, considering factors such as spatial representation, content damage, and the impact of different storm characteristics. Overall, the paper seeks to advance understanding and reliability in predicting the impacts of coastal flooding on various assets through a robust and validated modeling approach. It is a nice piece of work and I commend the authors with the manuscript, however It could benefit from some minor improvements. Therefore, I propose to reconsider this manuscript for publication upon revision of the following issues.
Minor comments
- Line 33: Indirect impacts also have short-term consequences
- Line 35/36: Sentence a bit vague, not clear why you use the word “instead”
- From line 86: It of course inevitable for this ECFAS project to have multiple associated names, but for the reader this is a bit unclear. What is the difference between ECFAS Pan-EU Impact Catalogue and ECFAS Flood impact layers generated for ECFAS Flood Catalogue. What is the ECFAS CEMS framework?
- Section 2.1: On which basis were the extreme weather events and cities selected?
- Section 3.2: The correction of interpolated values using the ratio between the cell areas of the flood map and the datasets is mentioned. It would be beneficial to explain the rationale behind this correction in more detail, as it could be a critical step in ensuring accuracy.
- Section 3.2: The section mentions that the datasets were interpolated to match the spatial resolution of the flood model. Please provide more information which interpolation method is used and provide elaboration on why despite which limitations the upscaling has been selected.
- Section 3.3: The manuscript mentions the use of an ensemble approach based on FDCs, citing Figueiredo et al. (2018) and Duo et al. (2020). It would be helpful to briefly explain how this ensemble approach works and its advantages in the context of building damage evaluation.
- Table 5: at Xaver road impact range is not presented
- Section 5: The manuscript adequately acknowledges the alignment of simulated damage with reported ranges for residential buildings in the Xaver storm, emphasizing the potential for overestimation and attributing it to uncertainties inherent in the probabilistic approach. However, it would greatly benefit from a more detailed elaboration on the specific sources of uncertainty, such as the assumptions made in vulnerability modeling or the variability in reported data. Similarly, while the underestimation of damage for commercial buildings and the mixed category is acknowledged, a deeper exploration into the nuanced factors contributing to this discrepancy and their implications for the model's reliability in diverse urban environments is warranted. Additionally, the discussion on the significant gap between simulated and reported road damage is informative, yet a more thorough analysis of whether the model adequately captures the diverse characteristics of different road types and the potential reasons behind this observed difference would enhance the manuscript's comprehensiveness.
Citation: https://doi.org/10.5194/nhess-2023-197-RC1 -
AC1: 'Reply on RC1', Juan Montes, 08 Apr 2024
Dear referee, many thanks for your time and feedback, that will contribute to the improvement of this manuscript. Keeping the structure of your comments, the authors- answers are as it follows:
1. Line 33: Indirect impacts also have short-term consequences
As the way it is written may lead to confusion, the paragraph will be corrected in L32-34 to clarify that indirect impacts also have short-term consequences.
2. Line 35/36: Sentence a bit vague, not clear why you use the word “instead”
It will be corrected to improve the connection between sentences and make the paragraph easier to understand, as the sentence was a bit vague. Modifications will be made on L34-37.
3. From line 86: It of course inevitable for this ECFAS project to have multiple associated names, but for the reader this is a bit unclear. What is the difference between ECFAS Pan-EU Impact Catalogue and ECFAS Flood impact layers generated for ECFAS Flood Catalogue. What is the ECFAS CEMS framework?
The ECFAS Pan-EU Flood Catalogue consists of flood maps covering most of the European coast, describing 15 flood scenarios of maximum TWL and duration for each of the defined coastal sectors (see Le Gal et al. (2023)* for more details). The ECFAS Pan-EU Impact Catalogue collects layers of impacts to population and other assets, such as buildings, roads, etc, and it was produced using the flood maps from the ECFAS Pan-EU Flood Catalogue.
The paragraph will be edited on L86-91 to avoid confusion between the different products and the methodology presented in this paper.
*Le Gal, M., Fernández-Montblanc, T., Duo, E., Montes, J., Cabrita, P., Souto Ceccon, P., Gastal, V., Ciavola, P. and Armaroli, C. (2023). A new European coastal flood database for low–medium intensity events. Natural Hazards and Earth System Sciences, 23(11), 3585-3602.
4. Section 2.1: On which basis were the extreme weather events and cities selected?
The extreme coastal events and locations were retrieved from the ECFAS database of extreme events (Souto Ceccon, P., Duo, E., Ciavola, P., Fernández Montblanc, T., Armaroli, C., 2021. Database of extreme events, test cases selection and available data, Deliverable 5.1 – ECFAS Project). The database contains information of extreme coastal events in the period 2010-2020 that were identified based on information collected through publicly available resources, Copernicus Emergency Management Service activations, and from other flood impact databases. The ECFAS database contains events that generated significant flooding and impacts along EU coastlines, and therefore it was used to retrieve coastal flood impact data necessary to perform the analysis and to build the impact tool and catalogue of impacts at pan-EU scale. See Souto Ceccon et al. (2021) for more details.
Information about the database used to retrieve the extreme weather events and affected cities will be added to Section 2.1 in order to clarify this aspect.
5. Section 3.2: The correction of interpolated values using the ratio between the cell areas of the flood map and the datasets is mentioned. It would be beneficial to explain the rationale behind this correction in more detail, as it could be a critical step in ensuring accuracy.
The number of people affected by coastal flooding was carried out using the Global Human Settlement - Residential Population (GHS-POP) and ENACT layers, with a spatial resolution of 250 m and 1 km respectively. For each cell of the layer, the value represents the absolute number of inhabitants of the cell, and it is therefore dependent on the cell area. Given that the spatial resolution of the flood layers used in this study (100 m) is better than the population datasets, the population layers were interpolated (nearest neighbor) using as reference the center of the cells of the flood maps. The interpolated values were corrected by multiplying them by the ratio between the cell areas of the flood map and the dataset to take into account the different cell resolutions. For example, the ratio between the cell areas of the flood maps and the GHS-POP layer is 0.16: if the interpolated value is 100 people (in a cell with a resolution od 250 m), the corrected value is 16 people (in a “flooded” cell with a resolution of 100 m). This is reported in the manuscript at L186-191. The text will be edited in order to improve the paragraph and facilitate its understanding.
6. Section 3.2: The section mentions that the datasets were interpolated to match the spatial resolution of the flood model. Please provide more information which interpolation method is used and provide elaboration on why despite which limitations the upscaling has been selected.
The interpolation used was the nearest neighbor method, which, after several tests, proved to be the most reliable in comparison to the linear interpolation or other common methods. Please, see the previous answer (Comment 5) for details on the correction applied to match the resolution of the flood maps.
7. Section 3.3: The manuscript mentions the use of an ensemble approach based on FDCs, citing Figueiredo et al. (2018) and Duo et al. (2020). It would be helpful to briefly explain how this ensemble approach works and its advantages in the context of building damage evaluation.
The model ensemble approach is a probabilistic-based assessment that relies on the combination of flood damage curves from different impact models. Most impact models are deterministic, but different studies have shown that the use of multi-models produces better results. The result of these types of approaches, like the one from Figueiredo et al. (2018), provides reliable probabilistic damage estimates that are more useful results for interpretation and decision making. For model ensembles, results improve as more models are considered.
Additional information will be added to the section of the general aspect (3.1) to briefly describe the ensemble approach and how it works.
8. Table 5: at Xaver road impact range is not presented
This was an oversight from an older version of the manuscript. In the meantime, results were rerun and refined. The current version (Appendix D for details) reports 24.4 – 45 k€/km (32.9 k€/km) as road impact for Xaver. The missing value in Table 5 will be amended.
9. Section 5: The manuscript adequately acknowledges the alignment of simulated damage with reported ranges for residential buildings in the Xaver storm, emphasizing the potential for overestimation and attributing it to uncertainties inherent in the probabilistic approach. However, it would greatly benefit from a more detailed elaboration on the specific sources of uncertainty, such as the assumptions made in vulnerability modeling or the variability in reported data. Similarly, while the underestimation of damage for commercial buildings and the mixed category is acknowledged, a deeper exploration into the nuanced factors contributing to this discrepancy and their implications for the model's reliability in diverse urban environments is warranted. Additionally, the discussion on the significant gap between simulated and reported road damage is informative, yet a more thorough analysis of whether the model adequately captures the diverse characteristics of different road types and the potential reasons behind this observed difference would enhance the manuscript's comprehensiveness.
Validating impact models for large-scale applications is a difficult process due to several factors, such as the limited availability of reliable data and that data are often provided in an aggregated form. In the present work, an extensive effort has been made to collect data on the impacts generated for different reference cases along the European coasts, performing the validation with impacts generated by 3 historical events that impacted coastal areas with different characteristics.
In this type of analysis, the selected flood model, the flood damage curves and the reported damage can introduce uncertainty into the study. An under- or overestimation of the flood extension or of the flood depth could lead to an under- or overestimation of the damage. In Section 2.2 the characteristic of the used flood maps is explained. In the case of the flood damage curves, this paper uses an ensemble approach. In the case of commercial buildings, 4 different flood damage curves were used to build the ECFAS Impact Model. These flood damage curves may have discrepancies at certain values of flood depth. In section 3.3 the models used are explained. Finally, in section 3.5 (table 4) an analysis of the reliability and the representativeness of the different resources used for the validation of the modelled results is presented, an analysis of the confidence of the data used for validation is presented in section 5.1.4, and a description of the limitations of the methodology is presented in section 5.3.
Citation: https://doi.org/10.5194/nhess-2023-197-AC1
-
RC2: 'Comment on nhess-2023-197', Anonymous Referee #2, 14 Feb 2024
This article outlines a method and presents results for computing flood damages in European coastal regions, supporting a broader European initiative aimed at assessing flood-related damages from rivers and storm surges. While the paper is engaging and well-written, demonstrating the potential for impact assessment alongside acknowledging inherent uncertainties, several issues prompt me to suggest a major revision.
My concerns are as follows:
- Conclusions: The concluding remarks read more like an endorsement of a specific EU project and prospects for future model advancements rather than a reflection on the key findings of the paper itself. It's essential to rephrase them to emphasize the main takeaways from the study. What insights were gained? Avoid introducing new elements; those can be addressed in the discussion section.
- Validation: The comparison between computed damages and observed damages in the three reference cases reveals disparities, which is understandable given the incomplete input and damage data. While the authors attempt to account for these differences, the explanations seem more like attempts to justify them. It's crucial to maintain a factual approach.
- Scope of Analysis: Given the discrepancies observed in the three reference cases and the explanations needed to analyze these, I advise not presenting the additional sites. What additional value does it offer? The results for these cases are inadequately presented and explained (only in the context of grid-based vs object-based). I strongly advise to focus solely on the three reference cases in this paper. There is adequate scope to write a paper on this.
- Clarity on Probabilistic Approaches: The frequent use of the term "probabilistic" lacks immediate clarity regarding the specific stochastic processes employed. It would be beneficial to provide a concise overview of the stochastic methodologies used. Which stochastic variables do you consider? Which not, and why not?
- Treatment of Population Data: In one instance, it appears that population data is treated as a stochastic variable. However, given the deterministic nature of population demographics, this approach seems unwarranted. You know the month and time that an event impacted a particular coast. It is more appropriate to explore the sensitivity of the results to different population datasets by comparing the two data sets (thus one of the 24 elements of the second data set)
- Sensitivity Analysis of Vulnerability Curves: While the vulnerability curves significantly impact the results, they are not inherently probabilistic. Reframing the discussion around sensitivity analysis might be more accurate.
- Wind: how is damage due to wind treated and isolated in the damage reports?
- Estimation of Cost Data: The utilization of probabilistic methods for estimating cost data requires clarification on the approach employed.
- Assessment of Flood Maps: It seems that the flood maps are considered as given, obtained from an external source. It would be valuable to evaluate the accuracy of these maps, particularly in the context of the three reference cases. What was the bias between the model results and the observations in terms of high water marks and flood extends?
- Explanation of Methodology: The paper contains numerous references to methods described in other works, making it challenging for readers not familiar with the referenced EU project to follow. Providing more detailed explanations of the methodologies employed would enhance comprehension.
- Reference Clarification: There is a discrepancy where section 3.1 in Line 230 is referenced, but no description of the probabilistic evaluation is provided therein.
- Inclusion of Damage Maps: Please show maps depicting computed and observed damages . Incorporating such visual representations, including building footprints and damage extents, would provide valuable context for interpreting the aggregated results.
- Line 266: What is a semi-quantitative, holistic comparison?
- Figure 8: the uncertainties seem very small. Is this because the variations in Figure 10 are so small?
- Table 5 Xaver/Roads still has xx’s.
Overall, addressing these concerns through a major revision would significantly strengthen the clarity and rigor of the paper.
Citation: https://doi.org/10.5194/nhess-2023-197-RC2 -
AC2: 'Reply on RC2', Juan Montes, 09 Apr 2024
Dear referee, many thanks for your time and feedback, that will contribute to the improvement of this manuscript. Keeping the structure of your comments, the authors- answers are as it follows:
1. Conclusions: The concluding remarks read more like an endorsement of a specific EU project and prospects for future model advancements rather than a reflection on the key findings of the paper itself. It's essential to rephrase them to emphasize the main takeaways from the study. What insights were gained? Avoid introducing new elements; those can be addressed in the discussion section.
The conclusion section (L526-542) will be reviewed and edited to better highlight the key findings of the paper.
2. Validation: The comparison between computed damages and observed damages in the three reference cases reveals disparities, which is understandable given the incomplete input and damage data. While the authors attempt to account for these differences, the explanations seem more like attempts to justify them. It's crucial to maintain a factual approach.
The scarcity of reliable data, usually in aggregated form, makes validation of impact models difficult. In this study, a validation process was carried out for 3 historical events that impacted different coastal areas (Section 5.1), together with an analysis of the reliability and the representativeness of the different resources used for the validation of the model output (Section 5.1.4). Although the authors agree that a more analytical way would be more appropriate for comparison with the results, due to the type of data used it is necessary to make this comparison based on an "expert judgement" approach.
3. Scope of Analysis: Given the discrepancies observed in the three reference cases and the explanations needed to analyze these, I advise not presenting the additional sites. What additional value does it offer? The results for these cases are inadequately presented and explained (only in the context of grid-based vs object-based). I strongly advise to focus solely on the three reference cases in this paper. There is adequate scope to write a paper on this.
Although the authors agree that the information for the 3 reference cases provide key information for the scope of this study, the test cases represent important additional information because they have been used to compare the object-based methodology presented in the study with a grid-based analysis that is widely used in similar studies by the scientific community. In fact, the test cases represent sites that have experienced coastal storms that generated remarkable impacts in the period 2010-2020. The validation of these results is not possible due to the lack of data and that the direct comparison between the object-based and the grid-based approaches is not meaningful.
4. Clarity on Probabilistic Approaches: The frequent use of the term "probabilistic" lacks immediate clarity regarding the specific stochastic processes employed. It would be beneficial to provide a concise overview of the stochastic methodologies used. Which stochastic variables do you consider? Which not, and why not?
Additional information regarding the model ensemble approach and the ability it has to provide an uncertainty estimation was added to the section of general aspects (3.1) to clarify this aspect (See Answer to Reviewer 1, Comment 2).
The probabilistic resample was applied to the number of people affected by the flood, to the financial damage to buildings and to roads. Given the different approaches, the resampling was different for each damage sector. The general description, common to population, buildings and roads, was described in Section 3.1, while the specific information was detailed in sections 3.2, 3.3 and 3.4. Some clarifications are presented below:
- Population (number of affected people): the method generates an ECDF based on a set of values of the number of affected people in each flooded cell. The set is built using multiple sources of population density. Then, for each flooded cell, a large number of values (1000) is randomly generated from the ECDF, representing the probabilistic estimation (distribution) of the number of affected people in that flooded cell. The process is applied to each flooded cell and the probabilistic estimate of the total affected population is based on the resampling set of each cell. This process is described in L186-195 and in Figure 2.
- Buildings (relative damage): the method generates an ECDF based on a set of values of relative damage factor for each flooded building. The set is built using a set of damage curves. Then, for each flooded building, a large number of values (1000) is randomly generated from the ECDF, representing the probabilistic estimation (distribution) of the damage factor for that flooded building. The distribution is then multiplied for the (deterministic) maximum damage to retrieve the financial damage. The process is applied to each flooded building and the probabilistic estimate of the total building damage is based on the resampling set of each building. Additional information will be added to section 3.3 to clarify this aspect.
- Roads (relative damage and maximum damage): the probabilistic resampling is applied to both the relative damage (using multiple damage curves) and the maximum damage (using an empirical set). As before, in both cases the method generates an ECDF based on a set of values of relative damage factor for each flooded road, and a set of values of maximum damage, and the ECDFs are resampled (n=1000). The resampling of the relative and maximum damage are combined generating a set of n x n values of financial damage for each road that represent the probabilistic estimation. This is described in Section 3.4 and in Figure 5. The caption of Figure 5 will be modified to include some of the information in the main text.
Note that other variables were not chosen for the probabilistic estimates mainly because of the lack of available data or or of the lack of multiple models available for model ensembles.
5. Treatment of Population Data: In one instance, it appears that population data is treated as a stochastic variable. However, given the deterministic nature of population demographics, this approach seems unwarranted. You know the month and time that an event impacted a particular coast. It is more appropriate to explore the sensitivity of the results to different population datasets by comparing the two data sets (thus one of the 24 elements of the second data set)
The number of people affected by the coastal flood is evaluated by both the Global Human Settlement - Residential population (GHS-POP) and ENACT datasets. There are different sources of uncertainty from the selected datasets: related to the temporal reference of the datasets (ENACT: 2011; GHS: 2015) in comparison with the date of the flood event, and related to the differences between flood maps and datasets spatial resolutions. The seasonal and night/day variability is accounted for by applying the probabilistic resampling of the datasets. At the operational level, an evaluation of the affected population based on the timing of the coastal extreme is expected to be more appropriate. Certainly, it would represent a refinement of the assessment, from a deterministic point of view. By evaluating the number of people on the test case AoIs, the results of the comparisons between the GHS-POP and the yearly average night/day and summer/winter seasons showed minor variability in terms of magnitude of the people's presence. In general, all evaluations identify a similar number of affected people, if variations within the same magnitude are considered acceptable. For these reasons, a probabilistic implementation was preferred using the combination of the different datasets.
6. Sensitivity Analysis of Vulnerability Curves: While the vulnerability curves significantly impact the results, they are not inherently probabilistic. Reframing the discussion around sensitivity analysis might be more accurate.
The authors agree that an FDC is not a probabilistic model itself. However, using multiple curves as multi-model ensemble is recognized to generate probabilistic estimates, as long as the result is represented as a distribution (mean-dev.stand, quantiles, etc…) generated by combining the curves. Considering the aim of the paper, the authors believe that there is no need to reframe the discussion as sensitivity-based. However, a sentence will be added to the discussion clarifying this aspect.
7. Wind: how is damage due to wind treated and isolated in the damage reports?
The effect of wind is not isolated from the reported damage, so the reported damage was interpreted with caution. Although the authors would like to have disaggregated data, it is in most cases impossible to access it and it is a limitation of this type of study. Nevertheless, the selected reference cases have been thoroughly analysed, and although wind can play an important role in the damage, the impacts generated by the selected events have mainly been caused by flooding.
8. Estimation of Cost Data: The utilization of probabilistic methods for estimating cost data requires clarification on the approach employed.
A brief description of the ECDF-based approach of cost estimation is present in the manuscript at L248-252, which includes: references to the dataset used from Van Ginkel et al. (2021), reference to the details in the Appendixes, and a figure (Figure 5) that shows the representativeness of the ECDFs applied in comparison to the original dataset. We will edit this part clarifying these aspects.
9. Assessment of Flood Maps: It seems that the flood maps are considered as given, obtained from an external source. It would be valuable to evaluate the accuracy of these maps, particularly in the context of the three reference cases. What was the bias between the model results and the observations in terms of high water marks and flood extends?
The flood maps were retrieved from the European flood catalogue implemented in the ECFAS project and described in the paper by Le Gal et al. (2023)*. The process of flood map generation and validation can be found in Le Gal et al. (2023)*. The calibration and validation of the numerical model used for the generation of the flood maps (LISFLOOD-FP) can be found in Le Gal et al. (2022)**.
*Le Gal, M., Fernández-Montblanc, T., Duo, E., Montes, J., Cabrita, P., Souto Ceccon, P., Gastal, V., Ciavola, P. and Armaroli, C. (2023). A new European coastal flood database for low–medium intensity events. Natural Hazards and Earth System Sciences, 23(11), 3585-3602.
**Le Gal, M., Ciavola, P., Gastal, V., Fernández-Montblanc, T. and Delbour, S. (2022). Validated LISFLOOD-FP model for coastal areas, Deliverable 5.2 – ECFAS Project (GA 101004211), www.ecfas.eu (Versión 2). Zenodo. https://doi.org/10.5281/zenodo.7488694
10. Explanation of Methodology: The paper contains numerous references to methods described in other works, making it challenging for readers not familiar with the referenced EU project to follow. Providing more detailed explanations of the methodologies employed would enhance comprehension.
In this paper several damage models from previous studies are used to build the ECFAS impact approach based on model ensemble. Although providing more detailed information could clarify some concepts, the authors consider that adding more details can add confusion and make the paper longer and difficult to read. However, the authors recognise that so many references can make the paper hard to understand, so brief statements/key messages about the methods of cited papers in the methodology will be added.
11. Reference Clarification: There is a discrepancy where section 3.1 in Line 230 is referenced, but no description of the probabilistic evaluation is provided therein.
In section “3.1 General aspects”, paragraph 2 (lines 175-179), reference is made to the probabilistic evaluation. As this aspect was common for population, buildings and roads, the authors decided to add it in a general section to avoid repetition.
12. Inclusion of Damage Maps: Please show maps depicting computed and observed damages . Incorporating such visual representations, including building footprints and damage extents, would provide valuable context for interpreting the aggregated results.
Due to the current length of the document and the fact that it goes beyond the scopus of this study, the authors consider that it was not necessary to add this figure in the paper. However, an attempt will be made to include an impact map for the 3 reference cases in a specific annex.
13. Line 266: What is a semi-quantitative, holistic comparison?
We will review and simplify the specific sentence (L268-270) avoiding reference to the mentioned text to avoid confusion.
14. Figure 8: the uncertainties seem very small. Is this because the variations in Figure 10 are so small?
The small uncertainty for affected population values are due to the fact that the used datasets (ENACT and GHS-POP) provide similar information, as shown in Fig.10 and discussed in Section 5.3.1. To note that in both figures (Figure 8 and 10) the representation of affected people follows a logarithmic scale.
15. Table 5 Xaver/Roads still has xx’s.
This was an oversight from an older version of the manuscript. In the meantime, results were rerun and refined. The current version (Appendix D for details) reports 24.4 – 45 k€/km (32.9 k€/km) as road impact for Xaver. The missing value in Table 5 will be amended.
Citation: https://doi.org/10.5194/nhess-2023-197-AC2
Interactive discussion
Status: closed
-
RC1: 'Comment on nhess-2023-197', Anonymous Referee #1, 19 Jan 2024
The manuscript “Validated probabilistic approach to estimate flood direct impacts on the population and assets on European coastlines” aims to develop and validate the ECFAS Impact model, a comprehensive approach for assessing the impacts of coastal flooding on buildings and infrastructure. The model integrates probabilistic evaluations, flood damage curves, and ensemble methods to simulate damage to residential, commercial, and industrial structures. The objectives include validating the model by comparing its simulated impacts with reported data from historical storm events in e.g. France, the United Kingdom, and Spain. The study assesses the model's performance in accurately representing damage to buildings, considering factors such as spatial representation, content damage, and the impact of different storm characteristics. Overall, the paper seeks to advance understanding and reliability in predicting the impacts of coastal flooding on various assets through a robust and validated modeling approach. It is a nice piece of work and I commend the authors with the manuscript, however It could benefit from some minor improvements. Therefore, I propose to reconsider this manuscript for publication upon revision of the following issues.
Minor comments
- Line 33: Indirect impacts also have short-term consequences
- Line 35/36: Sentence a bit vague, not clear why you use the word “instead”
- From line 86: It of course inevitable for this ECFAS project to have multiple associated names, but for the reader this is a bit unclear. What is the difference between ECFAS Pan-EU Impact Catalogue and ECFAS Flood impact layers generated for ECFAS Flood Catalogue. What is the ECFAS CEMS framework?
- Section 2.1: On which basis were the extreme weather events and cities selected?
- Section 3.2: The correction of interpolated values using the ratio between the cell areas of the flood map and the datasets is mentioned. It would be beneficial to explain the rationale behind this correction in more detail, as it could be a critical step in ensuring accuracy.
- Section 3.2: The section mentions that the datasets were interpolated to match the spatial resolution of the flood model. Please provide more information which interpolation method is used and provide elaboration on why despite which limitations the upscaling has been selected.
- Section 3.3: The manuscript mentions the use of an ensemble approach based on FDCs, citing Figueiredo et al. (2018) and Duo et al. (2020). It would be helpful to briefly explain how this ensemble approach works and its advantages in the context of building damage evaluation.
- Table 5: at Xaver road impact range is not presented
- Section 5: The manuscript adequately acknowledges the alignment of simulated damage with reported ranges for residential buildings in the Xaver storm, emphasizing the potential for overestimation and attributing it to uncertainties inherent in the probabilistic approach. However, it would greatly benefit from a more detailed elaboration on the specific sources of uncertainty, such as the assumptions made in vulnerability modeling or the variability in reported data. Similarly, while the underestimation of damage for commercial buildings and the mixed category is acknowledged, a deeper exploration into the nuanced factors contributing to this discrepancy and their implications for the model's reliability in diverse urban environments is warranted. Additionally, the discussion on the significant gap between simulated and reported road damage is informative, yet a more thorough analysis of whether the model adequately captures the diverse characteristics of different road types and the potential reasons behind this observed difference would enhance the manuscript's comprehensiveness.
Citation: https://doi.org/10.5194/nhess-2023-197-RC1 -
AC1: 'Reply on RC1', Juan Montes, 08 Apr 2024
Dear referee, many thanks for your time and feedback, that will contribute to the improvement of this manuscript. Keeping the structure of your comments, the authors- answers are as it follows:
1. Line 33: Indirect impacts also have short-term consequences
As the way it is written may lead to confusion, the paragraph will be corrected in L32-34 to clarify that indirect impacts also have short-term consequences.
2. Line 35/36: Sentence a bit vague, not clear why you use the word “instead”
It will be corrected to improve the connection between sentences and make the paragraph easier to understand, as the sentence was a bit vague. Modifications will be made on L34-37.
3. From line 86: It of course inevitable for this ECFAS project to have multiple associated names, but for the reader this is a bit unclear. What is the difference between ECFAS Pan-EU Impact Catalogue and ECFAS Flood impact layers generated for ECFAS Flood Catalogue. What is the ECFAS CEMS framework?
The ECFAS Pan-EU Flood Catalogue consists of flood maps covering most of the European coast, describing 15 flood scenarios of maximum TWL and duration for each of the defined coastal sectors (see Le Gal et al. (2023)* for more details). The ECFAS Pan-EU Impact Catalogue collects layers of impacts to population and other assets, such as buildings, roads, etc, and it was produced using the flood maps from the ECFAS Pan-EU Flood Catalogue.
The paragraph will be edited on L86-91 to avoid confusion between the different products and the methodology presented in this paper.
*Le Gal, M., Fernández-Montblanc, T., Duo, E., Montes, J., Cabrita, P., Souto Ceccon, P., Gastal, V., Ciavola, P. and Armaroli, C. (2023). A new European coastal flood database for low–medium intensity events. Natural Hazards and Earth System Sciences, 23(11), 3585-3602.
4. Section 2.1: On which basis were the extreme weather events and cities selected?
The extreme coastal events and locations were retrieved from the ECFAS database of extreme events (Souto Ceccon, P., Duo, E., Ciavola, P., Fernández Montblanc, T., Armaroli, C., 2021. Database of extreme events, test cases selection and available data, Deliverable 5.1 – ECFAS Project). The database contains information of extreme coastal events in the period 2010-2020 that were identified based on information collected through publicly available resources, Copernicus Emergency Management Service activations, and from other flood impact databases. The ECFAS database contains events that generated significant flooding and impacts along EU coastlines, and therefore it was used to retrieve coastal flood impact data necessary to perform the analysis and to build the impact tool and catalogue of impacts at pan-EU scale. See Souto Ceccon et al. (2021) for more details.
Information about the database used to retrieve the extreme weather events and affected cities will be added to Section 2.1 in order to clarify this aspect.
5. Section 3.2: The correction of interpolated values using the ratio between the cell areas of the flood map and the datasets is mentioned. It would be beneficial to explain the rationale behind this correction in more detail, as it could be a critical step in ensuring accuracy.
The number of people affected by coastal flooding was carried out using the Global Human Settlement - Residential Population (GHS-POP) and ENACT layers, with a spatial resolution of 250 m and 1 km respectively. For each cell of the layer, the value represents the absolute number of inhabitants of the cell, and it is therefore dependent on the cell area. Given that the spatial resolution of the flood layers used in this study (100 m) is better than the population datasets, the population layers were interpolated (nearest neighbor) using as reference the center of the cells of the flood maps. The interpolated values were corrected by multiplying them by the ratio between the cell areas of the flood map and the dataset to take into account the different cell resolutions. For example, the ratio between the cell areas of the flood maps and the GHS-POP layer is 0.16: if the interpolated value is 100 people (in a cell with a resolution od 250 m), the corrected value is 16 people (in a “flooded” cell with a resolution of 100 m). This is reported in the manuscript at L186-191. The text will be edited in order to improve the paragraph and facilitate its understanding.
6. Section 3.2: The section mentions that the datasets were interpolated to match the spatial resolution of the flood model. Please provide more information which interpolation method is used and provide elaboration on why despite which limitations the upscaling has been selected.
The interpolation used was the nearest neighbor method, which, after several tests, proved to be the most reliable in comparison to the linear interpolation or other common methods. Please, see the previous answer (Comment 5) for details on the correction applied to match the resolution of the flood maps.
7. Section 3.3: The manuscript mentions the use of an ensemble approach based on FDCs, citing Figueiredo et al. (2018) and Duo et al. (2020). It would be helpful to briefly explain how this ensemble approach works and its advantages in the context of building damage evaluation.
The model ensemble approach is a probabilistic-based assessment that relies on the combination of flood damage curves from different impact models. Most impact models are deterministic, but different studies have shown that the use of multi-models produces better results. The result of these types of approaches, like the one from Figueiredo et al. (2018), provides reliable probabilistic damage estimates that are more useful results for interpretation and decision making. For model ensembles, results improve as more models are considered.
Additional information will be added to the section of the general aspect (3.1) to briefly describe the ensemble approach and how it works.
8. Table 5: at Xaver road impact range is not presented
This was an oversight from an older version of the manuscript. In the meantime, results were rerun and refined. The current version (Appendix D for details) reports 24.4 – 45 k€/km (32.9 k€/km) as road impact for Xaver. The missing value in Table 5 will be amended.
9. Section 5: The manuscript adequately acknowledges the alignment of simulated damage with reported ranges for residential buildings in the Xaver storm, emphasizing the potential for overestimation and attributing it to uncertainties inherent in the probabilistic approach. However, it would greatly benefit from a more detailed elaboration on the specific sources of uncertainty, such as the assumptions made in vulnerability modeling or the variability in reported data. Similarly, while the underestimation of damage for commercial buildings and the mixed category is acknowledged, a deeper exploration into the nuanced factors contributing to this discrepancy and their implications for the model's reliability in diverse urban environments is warranted. Additionally, the discussion on the significant gap between simulated and reported road damage is informative, yet a more thorough analysis of whether the model adequately captures the diverse characteristics of different road types and the potential reasons behind this observed difference would enhance the manuscript's comprehensiveness.
Validating impact models for large-scale applications is a difficult process due to several factors, such as the limited availability of reliable data and that data are often provided in an aggregated form. In the present work, an extensive effort has been made to collect data on the impacts generated for different reference cases along the European coasts, performing the validation with impacts generated by 3 historical events that impacted coastal areas with different characteristics.
In this type of analysis, the selected flood model, the flood damage curves and the reported damage can introduce uncertainty into the study. An under- or overestimation of the flood extension or of the flood depth could lead to an under- or overestimation of the damage. In Section 2.2 the characteristic of the used flood maps is explained. In the case of the flood damage curves, this paper uses an ensemble approach. In the case of commercial buildings, 4 different flood damage curves were used to build the ECFAS Impact Model. These flood damage curves may have discrepancies at certain values of flood depth. In section 3.3 the models used are explained. Finally, in section 3.5 (table 4) an analysis of the reliability and the representativeness of the different resources used for the validation of the modelled results is presented, an analysis of the confidence of the data used for validation is presented in section 5.1.4, and a description of the limitations of the methodology is presented in section 5.3.
Citation: https://doi.org/10.5194/nhess-2023-197-AC1
-
RC2: 'Comment on nhess-2023-197', Anonymous Referee #2, 14 Feb 2024
This article outlines a method and presents results for computing flood damages in European coastal regions, supporting a broader European initiative aimed at assessing flood-related damages from rivers and storm surges. While the paper is engaging and well-written, demonstrating the potential for impact assessment alongside acknowledging inherent uncertainties, several issues prompt me to suggest a major revision.
My concerns are as follows:
- Conclusions: The concluding remarks read more like an endorsement of a specific EU project and prospects for future model advancements rather than a reflection on the key findings of the paper itself. It's essential to rephrase them to emphasize the main takeaways from the study. What insights were gained? Avoid introducing new elements; those can be addressed in the discussion section.
- Validation: The comparison between computed damages and observed damages in the three reference cases reveals disparities, which is understandable given the incomplete input and damage data. While the authors attempt to account for these differences, the explanations seem more like attempts to justify them. It's crucial to maintain a factual approach.
- Scope of Analysis: Given the discrepancies observed in the three reference cases and the explanations needed to analyze these, I advise not presenting the additional sites. What additional value does it offer? The results for these cases are inadequately presented and explained (only in the context of grid-based vs object-based). I strongly advise to focus solely on the three reference cases in this paper. There is adequate scope to write a paper on this.
- Clarity on Probabilistic Approaches: The frequent use of the term "probabilistic" lacks immediate clarity regarding the specific stochastic processes employed. It would be beneficial to provide a concise overview of the stochastic methodologies used. Which stochastic variables do you consider? Which not, and why not?
- Treatment of Population Data: In one instance, it appears that population data is treated as a stochastic variable. However, given the deterministic nature of population demographics, this approach seems unwarranted. You know the month and time that an event impacted a particular coast. It is more appropriate to explore the sensitivity of the results to different population datasets by comparing the two data sets (thus one of the 24 elements of the second data set)
- Sensitivity Analysis of Vulnerability Curves: While the vulnerability curves significantly impact the results, they are not inherently probabilistic. Reframing the discussion around sensitivity analysis might be more accurate.
- Wind: how is damage due to wind treated and isolated in the damage reports?
- Estimation of Cost Data: The utilization of probabilistic methods for estimating cost data requires clarification on the approach employed.
- Assessment of Flood Maps: It seems that the flood maps are considered as given, obtained from an external source. It would be valuable to evaluate the accuracy of these maps, particularly in the context of the three reference cases. What was the bias between the model results and the observations in terms of high water marks and flood extends?
- Explanation of Methodology: The paper contains numerous references to methods described in other works, making it challenging for readers not familiar with the referenced EU project to follow. Providing more detailed explanations of the methodologies employed would enhance comprehension.
- Reference Clarification: There is a discrepancy where section 3.1 in Line 230 is referenced, but no description of the probabilistic evaluation is provided therein.
- Inclusion of Damage Maps: Please show maps depicting computed and observed damages . Incorporating such visual representations, including building footprints and damage extents, would provide valuable context for interpreting the aggregated results.
- Line 266: What is a semi-quantitative, holistic comparison?
- Figure 8: the uncertainties seem very small. Is this because the variations in Figure 10 are so small?
- Table 5 Xaver/Roads still has xx’s.
Overall, addressing these concerns through a major revision would significantly strengthen the clarity and rigor of the paper.
Citation: https://doi.org/10.5194/nhess-2023-197-RC2 -
AC2: 'Reply on RC2', Juan Montes, 09 Apr 2024
Dear referee, many thanks for your time and feedback, that will contribute to the improvement of this manuscript. Keeping the structure of your comments, the authors- answers are as it follows:
1. Conclusions: The concluding remarks read more like an endorsement of a specific EU project and prospects for future model advancements rather than a reflection on the key findings of the paper itself. It's essential to rephrase them to emphasize the main takeaways from the study. What insights were gained? Avoid introducing new elements; those can be addressed in the discussion section.
The conclusion section (L526-542) will be reviewed and edited to better highlight the key findings of the paper.
2. Validation: The comparison between computed damages and observed damages in the three reference cases reveals disparities, which is understandable given the incomplete input and damage data. While the authors attempt to account for these differences, the explanations seem more like attempts to justify them. It's crucial to maintain a factual approach.
The scarcity of reliable data, usually in aggregated form, makes validation of impact models difficult. In this study, a validation process was carried out for 3 historical events that impacted different coastal areas (Section 5.1), together with an analysis of the reliability and the representativeness of the different resources used for the validation of the model output (Section 5.1.4). Although the authors agree that a more analytical way would be more appropriate for comparison with the results, due to the type of data used it is necessary to make this comparison based on an "expert judgement" approach.
3. Scope of Analysis: Given the discrepancies observed in the three reference cases and the explanations needed to analyze these, I advise not presenting the additional sites. What additional value does it offer? The results for these cases are inadequately presented and explained (only in the context of grid-based vs object-based). I strongly advise to focus solely on the three reference cases in this paper. There is adequate scope to write a paper on this.
Although the authors agree that the information for the 3 reference cases provide key information for the scope of this study, the test cases represent important additional information because they have been used to compare the object-based methodology presented in the study with a grid-based analysis that is widely used in similar studies by the scientific community. In fact, the test cases represent sites that have experienced coastal storms that generated remarkable impacts in the period 2010-2020. The validation of these results is not possible due to the lack of data and that the direct comparison between the object-based and the grid-based approaches is not meaningful.
4. Clarity on Probabilistic Approaches: The frequent use of the term "probabilistic" lacks immediate clarity regarding the specific stochastic processes employed. It would be beneficial to provide a concise overview of the stochastic methodologies used. Which stochastic variables do you consider? Which not, and why not?
Additional information regarding the model ensemble approach and the ability it has to provide an uncertainty estimation was added to the section of general aspects (3.1) to clarify this aspect (See Answer to Reviewer 1, Comment 2).
The probabilistic resample was applied to the number of people affected by the flood, to the financial damage to buildings and to roads. Given the different approaches, the resampling was different for each damage sector. The general description, common to population, buildings and roads, was described in Section 3.1, while the specific information was detailed in sections 3.2, 3.3 and 3.4. Some clarifications are presented below:
- Population (number of affected people): the method generates an ECDF based on a set of values of the number of affected people in each flooded cell. The set is built using multiple sources of population density. Then, for each flooded cell, a large number of values (1000) is randomly generated from the ECDF, representing the probabilistic estimation (distribution) of the number of affected people in that flooded cell. The process is applied to each flooded cell and the probabilistic estimate of the total affected population is based on the resampling set of each cell. This process is described in L186-195 and in Figure 2.
- Buildings (relative damage): the method generates an ECDF based on a set of values of relative damage factor for each flooded building. The set is built using a set of damage curves. Then, for each flooded building, a large number of values (1000) is randomly generated from the ECDF, representing the probabilistic estimation (distribution) of the damage factor for that flooded building. The distribution is then multiplied for the (deterministic) maximum damage to retrieve the financial damage. The process is applied to each flooded building and the probabilistic estimate of the total building damage is based on the resampling set of each building. Additional information will be added to section 3.3 to clarify this aspect.
- Roads (relative damage and maximum damage): the probabilistic resampling is applied to both the relative damage (using multiple damage curves) and the maximum damage (using an empirical set). As before, in both cases the method generates an ECDF based on a set of values of relative damage factor for each flooded road, and a set of values of maximum damage, and the ECDFs are resampled (n=1000). The resampling of the relative and maximum damage are combined generating a set of n x n values of financial damage for each road that represent the probabilistic estimation. This is described in Section 3.4 and in Figure 5. The caption of Figure 5 will be modified to include some of the information in the main text.
Note that other variables were not chosen for the probabilistic estimates mainly because of the lack of available data or or of the lack of multiple models available for model ensembles.
5. Treatment of Population Data: In one instance, it appears that population data is treated as a stochastic variable. However, given the deterministic nature of population demographics, this approach seems unwarranted. You know the month and time that an event impacted a particular coast. It is more appropriate to explore the sensitivity of the results to different population datasets by comparing the two data sets (thus one of the 24 elements of the second data set)
The number of people affected by the coastal flood is evaluated by both the Global Human Settlement - Residential population (GHS-POP) and ENACT datasets. There are different sources of uncertainty from the selected datasets: related to the temporal reference of the datasets (ENACT: 2011; GHS: 2015) in comparison with the date of the flood event, and related to the differences between flood maps and datasets spatial resolutions. The seasonal and night/day variability is accounted for by applying the probabilistic resampling of the datasets. At the operational level, an evaluation of the affected population based on the timing of the coastal extreme is expected to be more appropriate. Certainly, it would represent a refinement of the assessment, from a deterministic point of view. By evaluating the number of people on the test case AoIs, the results of the comparisons between the GHS-POP and the yearly average night/day and summer/winter seasons showed minor variability in terms of magnitude of the people's presence. In general, all evaluations identify a similar number of affected people, if variations within the same magnitude are considered acceptable. For these reasons, a probabilistic implementation was preferred using the combination of the different datasets.
6. Sensitivity Analysis of Vulnerability Curves: While the vulnerability curves significantly impact the results, they are not inherently probabilistic. Reframing the discussion around sensitivity analysis might be more accurate.
The authors agree that an FDC is not a probabilistic model itself. However, using multiple curves as multi-model ensemble is recognized to generate probabilistic estimates, as long as the result is represented as a distribution (mean-dev.stand, quantiles, etc…) generated by combining the curves. Considering the aim of the paper, the authors believe that there is no need to reframe the discussion as sensitivity-based. However, a sentence will be added to the discussion clarifying this aspect.
7. Wind: how is damage due to wind treated and isolated in the damage reports?
The effect of wind is not isolated from the reported damage, so the reported damage was interpreted with caution. Although the authors would like to have disaggregated data, it is in most cases impossible to access it and it is a limitation of this type of study. Nevertheless, the selected reference cases have been thoroughly analysed, and although wind can play an important role in the damage, the impacts generated by the selected events have mainly been caused by flooding.
8. Estimation of Cost Data: The utilization of probabilistic methods for estimating cost data requires clarification on the approach employed.
A brief description of the ECDF-based approach of cost estimation is present in the manuscript at L248-252, which includes: references to the dataset used from Van Ginkel et al. (2021), reference to the details in the Appendixes, and a figure (Figure 5) that shows the representativeness of the ECDFs applied in comparison to the original dataset. We will edit this part clarifying these aspects.
9. Assessment of Flood Maps: It seems that the flood maps are considered as given, obtained from an external source. It would be valuable to evaluate the accuracy of these maps, particularly in the context of the three reference cases. What was the bias between the model results and the observations in terms of high water marks and flood extends?
The flood maps were retrieved from the European flood catalogue implemented in the ECFAS project and described in the paper by Le Gal et al. (2023)*. The process of flood map generation and validation can be found in Le Gal et al. (2023)*. The calibration and validation of the numerical model used for the generation of the flood maps (LISFLOOD-FP) can be found in Le Gal et al. (2022)**.
*Le Gal, M., Fernández-Montblanc, T., Duo, E., Montes, J., Cabrita, P., Souto Ceccon, P., Gastal, V., Ciavola, P. and Armaroli, C. (2023). A new European coastal flood database for low–medium intensity events. Natural Hazards and Earth System Sciences, 23(11), 3585-3602.
**Le Gal, M., Ciavola, P., Gastal, V., Fernández-Montblanc, T. and Delbour, S. (2022). Validated LISFLOOD-FP model for coastal areas, Deliverable 5.2 – ECFAS Project (GA 101004211), www.ecfas.eu (Versión 2). Zenodo. https://doi.org/10.5281/zenodo.7488694
10. Explanation of Methodology: The paper contains numerous references to methods described in other works, making it challenging for readers not familiar with the referenced EU project to follow. Providing more detailed explanations of the methodologies employed would enhance comprehension.
In this paper several damage models from previous studies are used to build the ECFAS impact approach based on model ensemble. Although providing more detailed information could clarify some concepts, the authors consider that adding more details can add confusion and make the paper longer and difficult to read. However, the authors recognise that so many references can make the paper hard to understand, so brief statements/key messages about the methods of cited papers in the methodology will be added.
11. Reference Clarification: There is a discrepancy where section 3.1 in Line 230 is referenced, but no description of the probabilistic evaluation is provided therein.
In section “3.1 General aspects”, paragraph 2 (lines 175-179), reference is made to the probabilistic evaluation. As this aspect was common for population, buildings and roads, the authors decided to add it in a general section to avoid repetition.
12. Inclusion of Damage Maps: Please show maps depicting computed and observed damages . Incorporating such visual representations, including building footprints and damage extents, would provide valuable context for interpreting the aggregated results.
Due to the current length of the document and the fact that it goes beyond the scopus of this study, the authors consider that it was not necessary to add this figure in the paper. However, an attempt will be made to include an impact map for the 3 reference cases in a specific annex.
13. Line 266: What is a semi-quantitative, holistic comparison?
We will review and simplify the specific sentence (L268-270) avoiding reference to the mentioned text to avoid confusion.
14. Figure 8: the uncertainties seem very small. Is this because the variations in Figure 10 are so small?
The small uncertainty for affected population values are due to the fact that the used datasets (ENACT and GHS-POP) provide similar information, as shown in Fig.10 and discussed in Section 5.3.1. To note that in both figures (Figure 8 and 10) the representation of affected people follows a logarithmic scale.
15. Table 5 Xaver/Roads still has xx’s.
This was an oversight from an older version of the manuscript. In the meantime, results were rerun and refined. The current version (Appendix D for details) reports 24.4 – 45 k€/km (32.9 k€/km) as road impact for Xaver. The missing value in Table 5 will be amended.
Citation: https://doi.org/10.5194/nhess-2023-197-AC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 445 | 141 | 43 | 629 | 34 | 38 |
- HTML: 445
- PDF: 141
- XML: 43
- Total: 629
- BibTeX: 34
- EndNote: 38
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1