We would like to thank Olga Petrucci for taking the time to analyze our paper and data, as well as drawing our attention to multiple issues related to the data. Unfortunately, we have not been able to reproduce the statistics quoted by the author, or find the event-specific problems mentioned. We have all individually checked the data, paper and the website; as we explain below, the author probably has gotten the wrong impression of the data by analyzing only a small portion of the dataset. We would be very grateful for information which file(s) were used as basis of the posted comment, so we can potentially change the formatting of the data to avoid repetition of the problem. Below we reply to each paragraph of the comment:

1. “Your DB, for a period of 128 years, in 42 European countries, reported 704 fatalities”. The dataset covers a period of 151 years, of which in 150 separate years there is at least one event, and at least fatality. The total number of fatalities of the 2521 events is 19,322. We don’t know what subset of HANZE was used by the author to arrive at 128 years and 704 fatalities. We should also note that though the other dataset (FFEM) has 2875 fatalities (as opposed to 3455 in HANZE), almost half of it is in Asian countries not covered by HANZE (Turkey and Israel).
2. “for Italy we compare fatalities in the same time span reported in HANZE and in Papagiannaki et al, 2022, we found 102 vs 425 fatalities”. For Italy, 1980-2020, HANZE includes 625 fatalities, therefore we again think that the author only looked at some subset of HANZE.
3. We use “maintain” in the sense that the cited databases are currently accessible online, as most of them aren’t actually updated (at least not regularly). We did realize that event-specific data in AVI end in 2001 and some summary data in 2003; we used the environmental yearbooks published by the Italian government afterwards in lieu of AVI. If we get a chance to revise our manuscript in ESSD, we will modify the wording in the text to increase its precision.
4. “for my region (ITF6) … I only found 74 fatalities”. Events in HANZE that have at least partially occurred in Calabria have a total of 429 fatalities, not 74. The author writes further that “no trace of catastrophic events killing hundreds of people in autumn 1951 and 1953, or 6 people in 1996 in Crotone or 12 in 2000 (Soverato) and 10 victims in Raganello flash flood in 2018”. None of this is correct. The 1951, 1953, 1996 and 2000 events are all in HANZE, as can be quickly checked through the website: https://naturalhazards.eu/details,16134, https://naturalhazards.eu/details,16164, https://naturalhazards.eu/details,16512, https://naturalhazards.eu/details,16527. As for the 2018 event, we specifically excluded it. This is mentioned directly in the paper (L508-L511): “certain flood-related cases of fatalities were excluded here: … Fatalities related to sports activities in remote areas, such as canoeing or hiking, in connection to sudden localized flash floods, as they cannot be reliably modelled both in terms of hazard and exposure, e.g. … Italy 2018 (10 fatalities, FloodList, 2023).”
5. As evidenced above, we don’t imply that the region is “risk-free”. Even a look at Fig. 4, or our online map https://naturalhazards.eu/map, gives the exact opposite impression. We mention several times the limitations of the data, and as the study doesn’t say anything about future impacts of climate change in particular, it can’t be used directly for any planning purposes by any reasonable decision-maker.
6. It is not true that we show no fatalities for the 1998 Sarno disaster. The event was indeed mostly a landslide, but the AVI catalogue attributes a very small share of impacts (1 fatality, 35 affected) in Caserta province to flood. Therefore, we included those impacts as such, while describing the full impact of the landslide in the “Notes” field for context: https://naturalhazards.eu/details,16517 . Also, we mention several times the problem of co-occurring hazards and the fact that they cannot always be separated from floods (e.g. L131-140, L374-384, L565-582). In the data, we indicate co-occurring landslides or mudslides for 227 events (in “Notes” field).
7. The comment’s final paragraph is, in conclusion, erroneous as it is based only on a fraction of our data. Further, we didn’t claim that our work is complete, but the opposite: we indicate the issue of incompleteness several times, including just before the results (L238-240), extensively in the discussion, and in the conclusions.

We encourage the author to have another look at our dataset, especially through our website. We would be very interested in any further feedback that can help us improve the study. Kind regards on behalf of the co-authors,

Dominik Paprotny

We thank the reviewer for taking the time to analyze our paper. Below we list the reviewer’s comments with our responses afterwards.

Comment 1: Item 2.1: It is not clear what these criteria are. What is specifically meant by the "specific meaning of the term flood"? I understand you specify the criteria later; however, it reads very confusingly to have it here without going into detail. I would suggest modifying this paragraph.

Response 1: the paragraph is an introduction to the section, where individual criteria are explained in the following subsections. We will modify the paragraph to make clear where the criteria are described in detail.

Comment 2: Line 103: I see it as a limitation that you only consider floods with high consequences. EM-DAT and existing datasets also focus on these high-damage floods. However, smaller-scale events can have high damages when considering their cumulative consequences. I would suggest writing about this in the limitations section.

Response 2: we would disagree that we only consider floods with high consequences. We can note that EM-DAT has a minimum threshold of 10 fatalities, a “call for international assistance or an emergency declaration”, or 100 affected. While the first two indicate only very major floods, the third is theoretically lower than ours (200 affected), but in practice rarely applied. Also, the other database cited by the reviewer, FFEM, only considers events with fatalities. By contrast, 907 events (36%) in HANZE caused no fatalities, which amounts to almost twice the number of all EM-DAT events. We also include events with as few as 10 affected persons due to relatively low economic threshold in particular (1 million euro in 2020 prices), which enables capturing floods that impacted commercial assets or infrastructure, but spared residential zones. We will make this point clear in the discussion.

Comment 3: Line 112: I would mention already here that only one criterion needs to be fulfilled.

Response 3: We will move the sentence before the list of criteria in the revision.

Comment 4: Item 2.2: This item needs better explanation. How did you find the correct sources? How can this be reproduced by other researchers? Which keywords were used to search for articles or papers in different languages? What are these local sources? Is it a local newspaper article? How were ambiguities solved? Was this done manually or with natural language processing? How many people verified this information? What is the intercoder reliability? There is a vast literature on document analysis (the method you used here, I assume) that should be referenced at least once.

Response 4: we will extend the information on the processing of the data. In general, all materials were manually processed by the main author. Even if some materials were found and preprocessed by a research assistant, it was still only finalized and entered into the database by one researcher. Further, we consider the open availability of the data, focus on public online resources, and the citation of sources per each individual record (the latter not available in EM-DAT or FFEM), are the best means of reproducibility and verification of the data.

Comment 5: Figure 1: Why is there so little information for 2020? Did you consider the entire year?

Response 5: 2020 includes 30 events, which is below the average for previous decades (about 40 per year), but more than e.g. 2011, 2012 or 2017. It stands out in the graphs simply because it is the only year in the 2020s decade, whereas previous decades include all years.

Comment 6: Figure 3: This figure is not appropriate for color-blind people. Also, there is a lot of information overload. I suggest having two separate figures. The pie charts should be removed as they sometimes cover the entire country. It would be interesting to have information about how reliable/complete the information for each country is.

Response 6: we will check the color scheme and split the figure into two. We will add a table to show the completeness of impact data, and a qualitative indication of the perceived completeness of the records.

Comment 7: Discussion: A comparison with the database on fatalities by Pagagiannaki (2022) for the countries covered in both databases should be made.

Response 7: we will add such a comparison. As FFEM was considered a source in HANZE, there are no major differences. However, for some events FFEM was the only source that could be found, and therefore they could not be included as they didn’t provide information on the type of event. In some cases, the fatalities in FFEM were found based on other sources to be falling outside the scope of the study (hiking or boating accidents) or in a few instances, not flood-related. In the same study area, HANZE has 1796 fatalities since 1980 compared to 1524 in FFEM.

Comment 8: Line 282: What does 1305 mean here? Is it the number of events in Hanze before 1980?

Response 8: We will rewrite the sentence to make it clear that it is the number of events in HANZE before 1990.

Comment 9: Figure 4: It would be nice if you could compare these results with water gauge levels over the years, not only with other databases. Can you do this for at least some of the countries? This would improve reliability.

Response 9: it is unclear to us how such comparison should look like. Water levels translate only partially into impacts, due to large variation in flood protection levels and exposure. Additionally, the availability of gauge data is highly heterogenous in Europe. Therefore, such a comparison would not be informative in our opinion. We are currently merging a hydrological reanalysis with gauge data and documentary records of flood impacts and non-impacts, but due to very extensive modelling work involved it is outside the scope of this publication.

Comment 10: Figure 5: Please add Hanze V1 here too.

Response 10: We will include HANZE v1 in the revision.

Comment 11: The direct comparison with EM-DAT is not done in a fair way since the criteria for inclusion in EM-DAT are different. If you consider the same criteria as EM-DAT, how many more events does your database have? Please add two sentences with this comparison.

Response 11: The thresholds in EM-DAT are both higher (fatalities) and lower (persons affected) than ours, and additionally include the troublesome criterion of “call for international assistance or an emergency declaration”. Consequently, it is not possible to fully homogenize the criteria. When filtering events from both datasets using only the first two criteria, HANZE still includes three times more events than EM-DAT. As we mention already in section 4.1, we only found seven floods (out of 528) in EM-DAT that are not included in HANZE due to falling outside our criteria.

Comment 12: 4.2: The revision of the entries in Hanze 2 highlights how variable the methodology is and how much it depends on personal interpretation and different sources. This limitation should be highlighted.

Response 12: we would disagree with such an interpretation of the revisions made to the database. The main cause of revisions is the incorporation of a vastly greater number of sources (almost three times more). That an individual record was “revised” often doesn’t mean that previous information was changed, but added where it was previously missing. A revision was indicated when changes were made to any field, even to non-mandatory supplementary information (such as precipitation amount or notes). Availability of sources changes constantly, whether it’s a new research paper on flood events in Malta or a newly digitized 1960s post-flood government report from Austria. Each adds new level of detail and precision that changes the previous assessment that was sometimes based on very limited or outrightly incorrect secondary information, frequently those included in international databases.

Comment 13: Line 355: It would be important to have an assessment of the completeness and reliability for each country, even if it's qualitative. This information is crucial for people using this dataset and taking it for granted.

Response 13: assessing the “true” number of floods per country is not possible. As noted in response 9, we are trying to partially address this by employing extensive model reconstruction of past floods, which falls outside the scope of this paper. We will add a table to show the completeness of impact data, and a qualitative indication of the perceived completeness of flood records, per country.

Comment 14: Line 382: The references provided are not that useful, in my view. For example, how do I know which paper is (UNDRR 2020)? It would be great if you provided full references in the Hanze_events.csv dataset (not separate as it is now).

Response 14: HANZE database follows the standard academic citation format. As the list of references is alphabetical, it is not difficult to find UNDRR (2022) there, as in the bibliography of this research paper.

We thank the reviewer for taking the time to analyze our paper. Below we list the reviewer’s comments with our responses afterwards.

Comment 1:  How is the flood extent validated? Is it done considering the DFO database? I highly recommend to validate the flood extent from your database with the ones reported in the Global Flood Database by Tellman et al. (2021)

Response 1: The flood extent is based directly on the documentary sources that are cited per each flood. We disagree with the reviewer that HANZE should be validated with the Global Flood Database. This is because satellite-derived footprints are not direct observations, but results of algorithmic detection of temporary water coverage. We are currently undertaking a reanalysis of past floods with use of hydrodynamic modelling, and employed the Global Flood Database for verification of modelled flood extent. In the process, we have found very large discrepancies between impacts based on satellite footprints and reported impacts. For the UK, the Global Flood Database indicates an order of magnitude higher number of persons affected than detailed post-flood government surveys. In other countries, the extent of flooding is vastly underestimated, showing even two orders of magnitude lower number of persons affected. We plan to include the comparison in our follow-up study, while for now we consider that our data should be used to validate satellite-derived footprints, rather then the other way around. In this context, we didn’t use spatial footprints from DFO in the database, only quantitative and textual information contained in it.

Comment 2:  How can you differentiate between riverine and compound floods in delta areas? Are you comparing coastal and riverine floods that occurred in the same area at the same time?

Response 2: Yes, the events in deltas were considered compound, if there was a storm surge coinciding with a high river flow. 

Comment 3: Is there a reason why 2020 has a limited number of reported flood impacts?

Response 3: 2020 includes 30 events, which is below the average for previous decades (about 40 per year), but more than e.g. 2011, 2012 or 2017. It stands out in the graphs simply because it is the only year in the 2020s decade, whereas previous decades include all years.

Comment 4: Figure 3 is quite dense in information. Is there a way to remove the pie charts and make them as a separate figure?

Response 4: we will split the figure into two for better clarity of information.

Comment 5: Is Figure 4 referring to the old or new version of the HANZE database? If yes, refer to HANZE v2.1 as the new version and HANZE v1.0 as the old one within the paper. It isn't easy to understand if you are referring to the new or old version

Response 5: we will clarify in the figures that they refer to the new version. In the paper, results of v1 are only shown in section 4.2.

Comment 6: In section 2.1.1, are you also excluding flooding due to levee failure in addition to dam failure? If yes, how can you verify that those flood impacts were not due to levee failure?

Response 6: levee (dike) failures are a typical component of flood events, therefore they are included. Dam breaks can similarly be part of flood events. What we explicitly exclude, are “floods caused entirely by dam failures unrelated with a severe meteorological event”, as we write in section 2.1.1. This distinction means that we include dam breaks with a hydrological cause (reservoir overflow due to swollen river), but exclude purely geotechnical failures (like the notorious Stava valley tailing dam failure in 1985 that is often included in catalogues major floods). One can also imagine dike failures unrelated to high water levels, as happened in the Netherlands due to the 2003 drought and heat wave. Such unusual occurrences without a hydrological driver are not included.

Comment 7: Can you explain the differences between HANZE 2.1 and 1.0 obtained in Portugal?

Response 7: we already explain the difference in section 4.2: “originally most footprints were based on Zêzere et al. (2014), who indicated them only by upper-level administrative districts. In HANZE v2, the footprints were revised completely using detailed geocoded impacts by Pereira (2017).” It is therefore entirely a result of incorporating better information compared to what was previously available.