Dear referee,

Thank you for your time and for your feedback. The concern raised is understandable knowing that the validation of the flood model was one of the most difficult aspects of the catalogue’s creation due to the lack of qualitative data. In this context, it is important to note that twelve test cases were used, instead of one or two as usually performed in literature. In addition, the satellite image derived flood maps were coupled with highly precise flood markers, for which satisfactory results were obtained. 

Concerning the validation of the Total Water Level, the development and validation of the ECFAS hindcast was done separately to the present work and is presented in the ECFAS project report by Melet et al. (2021)¹, which concluded that there was a good level of accuracy for both average and extreme total water level. The accuracy of the hindcast total water level was verified using tide gauge data from 2010 to 2020. The validation results indicate that overall, 75% of the Root Mean Square Errors (RMSE) are below 0.15 m and 90% of relative RMSE values are below 15%. Furthermore, for extreme events, more than 75% of the RMSE values are below 0.20 m, and more than 80% of the relative RMSE values are below 20%. 

Currently, this information is not included in the manuscript and will be added for the next version before publication. Unfortunately, there was no TWL comparison per test case.

While there is a possibility to over-predict the flood, the representation of flooded data is globally satisfactory by comparison with observed data and supported by a validated total water level model.

¹Melet, A., Irazoqui, M., Fernandez-Montblanc, T., & Ciavola, P. (2021). Report on the calibration and validation of hindcasts and forecasts of total water level along European coasts, Deliverable 4.1 - ECFAS project (GA 101004211), www.ecfas.eu (Version 2). Zenodo. https://doi.org/10.5281/zenodo.7488687

Dear referee, many thanks for your time and feedback.

The concern about the validation section is noted. While the results are described and briefly commented in section 4.1 Validation against observed data, the main discussion about the validation is detailed in the section 5.1 Validation of the flood modelling method and evaluation of the synthetic storm approximation. Following your comment some clarifications and adjustments will be made in section 4.1 to make it clearer. On the same line, section 4.2 aimed to validate the relevance of synthetic storm scenarios to build the catalogue by comparing maps from realistic events with those from the catalogue. Here again, some clarifications will be added.

About the concern raised on the validation results and method, the overestimation witnessed in the results should be considered in context of the partial satellite based map. This is clearly visible on the maps (Figure 3) such as for Vaia 2018 at Lido Nazioni, Xaver 2013 at Norfolk and Gloria 2020 at Castellon, for which flooding waters were detected far from the coastlines but not in between. It then assumes that the observed data represent some parts of the flooded area but may not be the full extent of the flooding as explained in section 5.1. This is the reason why the validation process is a difficult aspect of this study as highlighted in section 5.1 and also why observed local markers were also used to produce additional validation aspects. These observed markers gather precisely geo-referenced data of observed flood, meaning they only show what was flooded and there is no marker of non-flooded areas. While the concern of bias representation due to the definition of the validation parametric is valid, the absence of non flooded markers limits us on the definition of a marker parameter that could go beyond 100% and thus showing possible overestimation of the model. At the end, the observed data in the present study showed points and areas that were flooded, and thus the validation analysis is limited to the validation of flooded areas which could favour over-estimating models as acknowledged in section 5.1.  Following the referee's feedback, additional comments will be added in the manuscript (sections 2.2, 4.1, 4.2).

Dear referee,

Many thanks for your feedback and comments that will greatly contribute to the improvement of the present manuscript. Keeping the structure of the points addressed by the referee (in bold) , below are our answers to the different concerns and additional information that will also be integrated into the manuscript when missing.

1. The information in the paper is in some cases insufficient for understanding what methods exactly have been used. For example, I have not seen a table with the roughness factors that the authors have used for the different land-use types and whether they have, e.g., merged classes. Also the information on the generation of synthetic events is very basic and should be better described in the paper, at least briefly (a reference is not sufficient).
   
   A table in the appendix will be added with the different Manning’s coefficients that were used in this study. Some classes were merged depending on the available information. Concerning the method used for the synthetic events, the section 3.3 ECFAS catalogue is dedicated to the definition of the synthetic scenarios and the presentation of the methods used to generate the key parameters (Storm duration and TWL Return Level).

1. The authors should clearly state what the flood catalogue contains and what is openly accessible (hydrographs, scenarios, floodplains etc.). My understanding is that currently only the maps are available? However, these maps would be rather useless if not combined with detailed metadata on how they were produced.
   
   The authors understand this point and ideally metadata and data would be gathered. At the current stage and part of the agreement made for the ECFAS project, only the flood extent maps with maximum water depth and velocity (not presented as not validated) are available in the flood catalogue. Some of the metadata used are not under the Public licence such the DEM and thus can not be shared. Other metadata were created for the ECFAS project in support for the current flood catalogue, such as the Coastal LU/LC layer, the ECFAS TWL Hindcast and Return Levels. They are available on their own through the ECFAS project deliveries, however, their presentation in the current article and inclusion in the catalogue is beyond the scope of the present work. Information where to find them is detailed in the manuscript. 
2. Lines 45-60: I feel that the authors are confusing databases and assessments as many of the papers cited refer to assessments rather than databases.
   
   This paragraph will be adjusted to follow the terminology employed by the respective authors.
3. Line 77: lower is confusing – maybe use “better”? Term will be adjusted.
4. I understand the limitations related to validating coastal flood maps. However, the description of the validation process employed in this work is insufficient. For example:

• comparison to satellite images: it is not clear if the authors simulate the full event and extract the flood plain at the exact same time when the flood was reported by the satellite images. It would be helpful for the reader to better understand the model performance if this was clearly stated in the methods. My understanding is that satellite images were collected after the events, however, for some of the floodplains there were satellite data available during the event.
  
  The full event was modelled and the maximum flooded area extension of each test case was compared to the flood maps extracted from satellite images. The maximal flooded area corresponds to the area of the cumulative flood extent of the event, and the satellite derived flood maps were generated by the comparison between before and after satellite images. The delays of acquisition for the post-event image are listed on Table 1.
  
  The satellite image selection process first listed all available images in public missions and the main private missions. The exploitable images closest to the event were selected: in a few cases the images were acquired during the event, but in most shortly after the event, for storm events also impact satellite images (cloud cover in optical images and wind conditions affecting the rugosity of water surfaces in case of SAR).
• I am a bit concerned about taking local flood markers as validation points for a 100 m model grid and considering the full cell as a hit when the flood marker is enclosed by the modelled flood.
  
  An effort was made to include the maximum observed data available to apprehend the validation of the flood models. The local flood markers are completing the acquisition of the satellite derived flood maps, attesting the presence of water in very local points. Considering a hit when the enclosing cell is flooded just attest that the model correctly predicts flood where it was observed.
• the authors use three metrics for validating the model. However, there is extensive discussion on H, which does not really offer much. For example, a model that would predict ALL pixels flooded would have a H of 100%, which is clearly not a good indicatoin. I am also unsure regarding the C metric and its usefulness. Kiesel et al. (2023) (see https://nhess.copernicus.org/preprints/nhess-2022-275/nhess-2022-275.pdf) use similar indices to discuss their validation. It might be useful for the authors to have a look. Further, I find the comparisons in section 4.2 rather confusing as this section does not really refer to an accuracy assessment regarding flooding per se but rather evaluates the representativeness of the catalogue. Maybe the authors can qualitatively report this comparison and simply focus on differences in flood extent when using the hindcasts.
  
  H represents the percentage of correctly predicted cells, therefore a 100% H means that the model correctly predicts the flooding of observed flood areas. Indeed if the model floods the entire domain, H will also be 100%, that is why generally H is presented along with the F, percentage of overpredicted cells, and C, global assessment of the model, indicators. Ideally, C will be a good indicator of the modelled flood map correctness. However, in the present study, while H comes with quite a certainty of flooding as presence of waters was detected in the satellite images, F and C are sensitive to the uncertainties relative to the non-flooded areas and the absence of detected water. Indeed, if there is a delay between the event and the post storm image, an underestimation of the flooded area could be expected from the observation, therefore falsing the F indicators and biassing the C estimation. Among others, this is the case of the test-case of Warnemunde during the storm Axel (2017) in the present study. The observed map generated with images 2 days after the event, does not detect water on the quay of the city while being reported. Consequently, the false indicator is very high (>400%) as the model predicts flooding in these areas and C very low (0.7%). This is why, in the presence of partial flood maps, the present study based its discussion on the H indicators, showing that the model correctly predicts the observed flooding, while emphasising for a possible over-estimation of the model (section 5.1).
  
  Thank you for indicating the interesting work performed by Kiesel et al. (2023)¹. In their study, the validation of the coastal flood extent is supported by satellite derived maps and the estimation of the similar indicators used in the present study, as H represents the percentage of correctly predicted cells and F the percentage of overpredicted cells. It is interesting to see that both works conclude with similar limitations of the use of this kind of observed maps.
  
  Concerning section 4.2, it is indeed not about the assessment of the flooding areas, but how close are the catalogue synthetic maps to realistic flood maps generated from TWL hindcast data, thus assessing the representativeness of the catalogue as detailed in section 3.4. In this case, the indicator C is ideal to assess the difference between the maps by comparing their intersection to their union. The closer the maps are, the closer the value of C will be to 100.  To avoid the confusion, some comments will be added in the manuscript.

1. There is no justification whatsoever regarding the chosen resolution (100 m) for the simulations. Further, there is very limited discussion on how resolution affects the results, particularly in terms of validation and data merging. This is a very important factor when evaluating the results and the authors certainly need to address this point throughout the manuscript.
   
   As mentioned in the method section 3.1, the choice of the flood model configuration was supported by a sensitivity analysis not presented in this manuscript. The test-cases were also modelled using 50m resolution grids and the comparison to observed data was also applied as presented for the 100m models. Considering the validation data, the 100m models globally performed slightly better, probably due to the smoothing of local barriers and protection and thus generating larger hazard maps which could compensate for a slight under-estimation of the forcing (see, Melet et al. 2021 ²). At the same time, the 50m resolution drastically increased the computational time. In consequence, without quantifiable improvements from the finer models, a 100m resolution was chosen to support a balance between quality and computational feasibility. It is also important to note that 7920 models were developed for the creation of the flood catalogue, and 100m grid resolution matches the state of art for European flood map catalogues generated using a dynamic method (Vousdoukas et al. 2016 ³). This information will be added in the manuscript.
2. What are the relative peak water levels of the applied return periods? These need to be clearly stated.
   
   A table gathering range of TWL peak values for different oceanographic regions and for each RL will be added in the methodology.
3. The hydrograph of the synthetic events is rather simple – although even simpler hydrographs have been used in the past, there are studies which have employed more realistic synthetic hydrographs (e.g. MacPherson, L., Arns, A., Dangendorf, S., Vafeidis, A., and Jensen, J. (2019). A stochastic storm surge model for the German Baltic Sea coast. J. Geophys. Res. Oceans124, 2054–2071). Some further information on the decisions regarding the chosen shape would be useful.
   
   It is indeed possible to develop more elaborated shapes as presented in the MacPherson et al. (2019) ⁴ study. They used a stochastic approach to define a temporal approximation of extreme events for 45 locations in the German Baltic Sea based on data varying between 14 and 66y. Their work shows differences even in locations less than 100 km apart and in the same oceanographic region. The application of such a sophisticated method at European scale would require dedicated work to identify and analyse the shape of available hydrographs. This is an interesting comment that can improve the representativeness of future versions of flood catalogues but it was out of the scope of the present study. Additionally, according to the fair results obtained in the comparison between synthetic and real events (section 4.2), the shape of the synthetic TWL is not the primary source of uncertainty, unlike the DEM or the flood model numerical solver. Therefore, it is mainly for sake of simplicity for an application at European level that the symmetrical triangular shape was chosen. 
4. Figure 4: Differences in flood extents hardly visible, I would suggest to pick specific areas where there are strong differences and show the flood extents.
   
   Some new figures will be created to enhance the visualisation, selecting interesting areas, however relative differences between the Maximum Flooded Areas are already displayed in Figure 5.
5. Comparison with Vousdoukas results : state which horizontal resolution they employ as I assume would also strongly affect the resulting flood extents.
   
   It is also a 100m model. The information was added in the text.
6. Line 312: I find it unusual that friction in wetlands is higher than in urban areas. Reporting the roughness values and the sources (see previous comment) would be useful.
   
   Wetlands are usually present on the first dry cell line, areas which are usually covered by vegetation. It is then assumed that they will have a greater roughness than concrete areas. A table of the Manning’s coefficient will be added in the appendix, see point 1. 
7. Although the paper does not contain many grammatical or syntax errors, it would greatly benefit (in terms of readability) by improving the use of English throughout. Proofreading by a native speaker is recommended.
   
   Following the guidelines from the journal, we will make the request to submit the revised manuscript to a proofreading service. 

¹ Kiesel, J., Lorenz, M., König, M., Gräwe, U. and Vafeidis, A.T., 2023. Regional assessment of extreme sea levels and associated coastal flooding along the German Baltic Sea coast. Natural Hazards and Earth System Sciences, 23(9), pp.2961-2985.

² Melet, A., Irazoqui, M., Fernandez-Montblanc, T., & Ciavola, P. , 2021. Report on the calibration and validation of hindcasts and forecasts of total water level along European coasts, Deliverable 4.1 - ECFAS project (GA 101004211), www.ecfas.eu (Version 2). Zenodo. https://doi.org/10.5281/zenodo.7488687

³ Vousdoukas, M.I., Voukouvalas, E., Mentaschi, L., Dottori, F., Giardino, A., Bouziotas, D., Bianchi, A., Salamon, P. and Feyen, L., 2016. Developments in large-scale coastal flood hazard mapping. Natural Hazards and Earth System Sciences, 16(8), pp.1841-1853.

⁴ MacPherson, L.R., Arns, A., Dangendorf, S., Vafeidis, A.T. and Jensen, J., 2019. A stochastic extreme sea level model for the German Baltic Sea coast. Journal of Geophysical Research: Oceans, 124(3), pp.2054-2071.