Dear Authors,
Thank you for your reply to the referees’ comments in the interactive discussion stage of the manuscript.

You have already indicated in your reply to the referees that you have already incorporated some of the comments. Apart from these points, I think it would be important that you add the elaborations/clarifications that you provided to the referees into your manuscript, as these sections of the manuscript clearly seem to be in need of further explanation. This would strengthen both the focus and the clarity of the manuscript.


After taking into consideration the manuscript, the referee comments, your responses to the referees and my own editorial comments (for details please see below), the manuscript would benefit from a thorough revision to clarify certain aspects and to further strengthen the document, particularly with regard to the validity of data and methods used.

I look forward to receiving the revised version of your manuscript that should incorporate the reviewers’ comments and the editorial remarks below.


With best wishes,
Julia Hall




General Comments:
Overall, plit is essential that all the comments of the reviewers are incorporated into the manuscript. Particularly, if points were in need for further elaboration (as pointed out by one of the reviewers) please add the explanations directly into the manuscript (with similar detail as in the response to the reviewer question).

I) The core of the manuscript is the analysis of the Dartmouth flood Observatory global data set. However, as also pointed out by one of the reviewers, the data might not have the required quality for the trend analysis performed in the manuscript.
Therefore, when the data is introduced, the authors need to present the uncertainties and possible data quality issues associated with the dataset in detail. That is, the quality assessment needs to be done a priori or at the latest accompany the data analysis procedure itself, so that the reader knows of the limitations and possible influences on the results so that the outcomes are not misleading. Performing the analysis and only later discussing possible issues with the data quality in the discussion (as currently done in the manuscript) is thus not appropriate. Additionally, the authors should discuss in detail, why in the data quality is sufficient to perform the analysis on the data (similar to their response to the reviewer). This is of particular importance with regard to the analysis for the flood frequency, as for example reporting bias has a strong impact on the results.

II) All methods need to be moved into the methods section and not appear in the results section. Additionally, the GLM framework is an important part of the paper and should be further elaborated. This means not presenting it at the end of the paper but making it a fundamental part of the analysis, which also requires the methodological details to be moved to the methods section.

III) In parts of the analysis, the characteristics of the flood duration distribution (moments) are analysed for a change. However, each year has a different sample size. This results in the possibility that the changes observed could be artificially caused by the different sample sizes. This is of particular importance for the southern mid-latitudes, where only 59 floods are recorded for the entire period, which means that on average there are about 2 floods per year. However, the maximum for one year in this region is 7 floods, which will leave many years without a flood and in turn not really allow for a meaningful analysis of the changes in any of the analysed variables.
Therefore, the authors need to show that the changes in sample size do not impact the results, or find a way how to correct for the changes in the sample size.

IV) Both trend analysis and change point analysis are performed. Generally, the presence of a change point influences the interpretation of the obtained statistically significant trends. (Villarini et al 2009) Therefore, after a detection of a change point, the timeseries before and after the change point should also be analysed separately to put the overall trend into perspective.
Additionally, the data seems to not have a monotonic change but rather an oscillatory behaviour (e.g. Fig 2). Instead of ignoring this characteristic and simply reporting the results of trend analysis, the authors need to at least discuss this or use some method that can account for such characteristics in the time series.

V) As already pointed out by reviewer 2 (point 4): The MK test might not be the best for zero inflated count data. How is a zero count being handled in the analysis (i.e. as a zero value or as missing data)? As this influences, the output of the MK test. Please consider using trend tests that are meaningful to the data used or explicitly justify why the MK is considered appropriate. See also Table 5 where one obtains a statistically significant trend using the MK test but the Sen Slope is zero. What is the physical meaning of this?

VI) The data is aggregated using latitudinal belts. This reasoning needs to be further elaborated, as the current reasoning is not convincing. When moving from west to east in each of these latitudinal belt, the temperature and precipitation characteristics change considerably for example due to increasing continentality. Therefore, ‘binning’ floods from for example the west coast of Europe with floods in central Russia or even norther China seems to be not quite intuitive a requires a better explanation.

VII) I agree with the reviewer that given the uncertainties with the data quality (homogeneity) the results should be corroborated with another global dataset such as the GRDC. There has been a previous study that showed agreement between the DFO in Europe and the US, however this does not mean that this is the case for regions in which the data is less uncertain. This fact needs to be mentioned/discussed.

VIII) I would suggest that in section 4 a discussion of the results in the context of already existing (partly smaller scale) studies on flood frequency /duration is conducted, as section 4.1 should be moved into the data section.

IX) Spatial aggregation (globally or into latitudinal belts) of a GPH destroys crucial information needed to characterise atmospheric circulations, namely spatial contrasts. Therefore, any trends calculated over time series resulting from these aggregated fields do not provide physically consistent information to characterise changes in atmospheric circulations. In fact, the weather is not determined by the value of the GPH per se but rather by spatial contrasts (related to spatial pressure gradients). Therefore, the analysis has no physical meaning with regard to flood frequency or duration. Similarly, a spatial aggregation of PWC (globally or into latitudinal belts) and the associated trend does not provide any information whether it will rain or not as it will depend where the PW content is located. Therefore, the analysis of these variables does not provide any additional insights.
Consequently, there is no physical value in using these aggregated GPH and PWC as predictors in the GLM. It is important that the authors reconsider the variables that they use for the GLM.

X) When rewriting, please pay attention to spelling, grammar and sentence structure.


Specific Comments:

Throughout the document the phrase ‘at all spatial scales’ is used several times. However, it is not clear what this expression means. Do you want to refer to ‘both at global and in the latitudinal belts’? If so, the expression could be misunderstood. The same potential for a misunderstanding applies to ‘latitudinal scales’. Please replace these phrases with something less ambiguous.

Table 9 and 10 contains colour, which will not be possible in the final publication, please remove the colours and adjust the tables so that they will be readable/understandable without the colours.

L 35 ‘This research…’ Which one? Please specify.
L39-40. ‘Understanding these trends can help…’ How can the understanding of what aspect of the trends help? Please specify.

L65: ‘…most susceptible…’ In your analysis only the percentages of the three classes of flood duration are given and 4 countries are shown as an example. This has nothing to do with susceptibility or ‘vulnerability’ as mentioned in other parts of the manuscript. A clear definition of ‘susceptibility’/‘vulnerability’ as used in this paper needs to be provided.

L68: ‘We consider…’ is this a hypothesis? Please be more specific.

L70: ‘Understanding the temporal trends (regime like behaviour)…’ The meaning is not clear. Additionally, the understanding … will help to understand better… In the current manuscript, the trends are not ‘understood’, but rather characterised. (See also comment above P2L39-40)

L 71-74: ‘this will ultimately lead to…’ How? I think this sentence is out of scope of the current research.

L 92: The data might be ‘globally consistent’ (i.e. same methods) but what about temporal consistency? Please elaborate.
L101: please provide more detail on how the ‘unusually large’ are being quantitatively determined.

L112 ‘as suggested by Env (2016)’ ? Is this an incomplete reverence?

L113 ‘…more consistent…’ compared to what? Please specify.

L115 ‘four countries that have a high flood frequency’ It is not clear why these countries were selected as from Fig 7 there seem to be other countries such as Indonesia or Vietnam that have a higher number of floods then Thailand. Please elaborate.

L 190: the maximum number of floods is not occurring around 2005. It is only the LOESS that peaks in this period for most of the analysed regions. I recommend a more careful interpretation of the results.

L213: please provide a ‘physical’ meaning of MAD.

L 259: ‘The magnitude…’ This is can be easily confused with flood magnitude. Please use ‘The length…’ or similar.

L 308: Please specify whether it is possible to have more than one flood of different length per year. If so the statement ‘to have at least one flood per year’ should be changed to ‘to have on average one flood per year’. Additionally, please elaborate why it is considered of importance to have the 31 floods per country.

L315: The analysis is interesting, but please provide some explanation/interpretation of the results.

L321, L323, L 325, Please add % behind each number.

L 442: Please replace the term ‘attributed’ as no attribution within a formal attribution framework has been performed.

L 446: Please elaborate why no formal autocorrelation testing was performed.

L394-395: The quotation of Wang and Zhou is not correct, since the statement cannot be substantiated from their analysis. Moreover, the statement is physically inconsistent.

L 396-398: This is not clear. Please rewrite.

L 453-462: I am not sure that this section deserves a separate section in the discussion of the results. I suggest rather incorporating it into a general discussion section.

L 491-494: This was not really shown in the paper. Please focus only what has been found in the current manuscript.

P 29: For Figure 7a) I suggest using a ‘ternary plot’ (also called triangle plot, simplex plot, Gibbs triangle or de Finetti diagram) instead for ease of interpretation.


References:
Villarini et al 2009, On the stationarity of annual flood peaks in the continental United States during the 20th century, WATER RESOURCES RESEARCH, VOL. 45, W08417, doi:10.1029/2008WR007645.

Dear Authors,
Thank you for submitting a revised version of your manuscript, which incorporated most of the referees comments.

Following the latest reviews provided by the referees and the referee reports in the previous round of the discussion, it becomes clear that the data source used in this analysis is still a major concern.
Particularly the comments provided by the Anonymous Referee #2 raise important points in this regard. For example, if in the data no date is mentioned the middle of the month is used. This certainly affects the calculation of the flood duration and any subsequent analysis of the time series .
Both referees recommend that the uncertainties need to be better explained and highlighted and that results obtained should at least be partly validated with additional data sets such as the one from the GRDC.
Therefore, to improve the confidence in the analysis, the authors should validate their results using a different data set. Regions for validation should be selected with particular focus on where the data quality in the DFO is perceived to be low.

Moreover, please take into account all the other points highlighted by the two referees, as they both raise valid concerns.

Additionally, the authors mention several times the changes before and after ~ 2000. This is an interesting finding; however, it also raises again concerns with the data homogeneity/reliability as the authors mention themselves that only consistent information in the data exists particularly since 1999. This fact needs to be further investigated and discussed in detail in the discussion section.

The abstract strongly highlights the results about trend results. Given that the DFO data is used (which has known data quality issues and all the concerns raised by the referees) a statement regarding the uncertainties associated with this data needs to be included. Additionally, in the data section 2.1 a more detailed discussion on data quality is needed so that the reader is aware of the potential issues BEFORE the data is analyzed and the results are presented.

Section 4.2.1. More detail to the results need to be added by linking the statistical results obtained to the actual physical mechanisms and how these interconnections interact with the floods and why there is the difference in their effects to different indices analyzed.



Additional comments:

P1L14-15: Add reference to this statement.

P2L 22: Unclear. Please elaborate more how 'temporal trend' and 'regime like behavior' are related.

P3L6: Add sentence on the country scale analysis.

P4L15: Add already a sentence in this section why these 4 countries were selected.

Section 3.2: Please add some elaboration to each sub section what these changes mean for the floods in a physical way. E.g. what does is mean when 'the asymmetrical/symmetrical behavior of the distribution … changes from 5 to ~8.

Section 3.4: Much detail on the methods presented in this section should be moved to the methods section.

P11L19-20: Please provide more detail on this data on damages (e.g. how was it derived in the original data set)

Section 3.4: Please add a paragraph summarizing the most important findings from this section. Currently this is not clear.

Section 4.1. Please discuss the possible effects on the trends obtained and the uncertainties associated with the in-homogeneity of the data pre- and post- inclusion of the MODIS product.

Figure 8a) The figure is difficult to read due to the thin line width of the gray points. Please increase the quality of the figure.

Dear Authors,
Thank you for submitting a revised version of your manuscript

The manuscript improved during the discussion phase, particularly due to the additional analysis performed to identify/account for possible data shortcomings/inaccuracies and the increased attention to the physical interpretation of the results obtained.

After the revisions there is one point remaining that should be better accounted for/discussed and communicated to the reader:
From the responses and the figures provided by the authors to the reviewers, it becomes evident that there is a likely bias of the start date of floods that is reported in the DFO.
The authors highlight that only 11% of the entire number of flood events starts on the 1st (6.5%) and the 15th (4.5%) of a month. However, the flood counts on these dates are considerably higher (more and almost double respectively) compared to the ~2.8% of flood events starting on an ‘average day’.
This indicates the existence of a bias in the reporting of the start date in the DFO database.
Additionally, when examining the distribution of flood dates across the month, it becomes also apparent that the day 1, 5, 10, 15, 20 and 25 have also an increased number of flood counts (roughly ~4%) , which also suggest a reporting bias (likely rounding of the start date to a number that can be divided by 5).
Therefore, the authors need to discussed the possibility of a reporting bias in the date in the data section, together with the possible implications of such a bias for the results of the analysis in the discussion section (and provide the figure with the counts per day in the appendix).
Additionally, the authors also need to check and report on this for the end date of the floods.
The authors should also note in the data section (as they have already done so in their response to the referees) that there seems to be no spatial pattern for these apparent reporting biases)

When conducting the revision, please also incorporate the technical corrections suggested by Referee #2.


Additional comments:
P4 L9: Please provide the reference to the publication in which the DFO reports this statement.

P14 L1-5: Please describe in more details what ‘very little errors’ are (i.e quantify what ‘very little’ entails in number of days). Additionally, given the discussed shortcomings of the DFO data, please replace ‘evidently reliable’ with a term that better describes the outcome of the comparison

P18L1-2 Please specify in more detail what ‘fairly reliable’ entails.


Figure 8a: The grey color used for the point data is too light, which makes it very difficult to distinguish the shapes used to indicate the different countries. Maybe the grey color used for the points could be replaced by the colors used for the countries. Additionally, the two different shades of red are too similar. Please consider using different colors.