|Review of Najibi and Devineni (2018) for ESD|
I have re-reviewed the paper by Najibi and Devineni (2018) for ESD. My original major concern was the use of the DFO dataset as a single source of information on flood frequency and duration in which the conclusions of the paper are founded. I acknowledge that the authors have answered most of my original comments to a satisfactory degree, and indeed made many positive changes to the manuscript. This includes addressing issues around assumptions for statistical testing and including more information about the limitations of the DFO dataset.
However, after reading the revised manuscript the same concerns about the sole use of the DFO dataset still remain. Here is the author response to this concern: “Regarding the DFO dataset as a single source of information employed here, we should emphasize that to our knowledge, this is the first analysis of “global flood events” that focuses exclusively on the variability of “flood duration” derived from DFO dataset over the last three decades. The database has recorded flood inundation events using satellite sources and media verification since 1985, and currently has over 4200 entries with approximate location of the center of the area flooded, the dates and duration of flooding and notes as to societal impacts. This is the only global data set of this kind and we believe that an analysis of the trends provides value. Much of the prior studies either focused on rainfall-based datasets or model-based river flow data. In this regard, our study adds a new dimension to the flood literature (especially the understanding of the floods that last for longer time) at a global scale”.
My view is that just because this study is the first to analyse an event based dataset at the global scale is not the only criteria for scientific publication. The conclusions made are profound and could be misleading if they are not supported by other event-based and/or physical flood datasets. For example, there are five key conclusions outlined on Pg. 16 in the revised manuscript: “The frequency of flood events has increased”; “there is a statistically significant trend in the moments of flood duration at the global scale”; “The yearly number of moderate and long duration flood occurrences increased”; “there was no monotonic trend observed in the frequencies of short duration floods”, and “the increase in frequency of long duration floods during recent years can be related to the persistent patterns in the low-frequency climate induces”. There is a chance that these conclusions could be an artefact of the known changing quality and/or changing sources of information ingested into the DFO dataset since 1985, and not to mention the relatively short period of record used for a trend analysis (without trying to place shorter term patterns into their longer term context).
Of particular concern is how the ‘flood duration’ variable in the DFO dataset relates to physical flood inundation extent/duration of inundation, or just an artefact of news reporting. According to the DFO website (http://floodobservatory.colorado.edu/Archives/ArchiveNotes.html), flood duration is established from the reported flood start and end dates, and from this website: “Ocassionally there is no specific beginning date mentioned in news reports, only a month; in that case the DFO date will be the middle of that month. Ending dates are often harder to determine - sometimes the news will note when the floods start to recede. We make an estimate based on a qualitative judgement concerning the flood event”. To what degree of certainty does the flood duration variable used in the DFO, and hence a key aspect of the paper, relate to physical flood duration? Is this qualitative uncertainty mentioned on the DFO website only for a small number of floods in some lower income countries, or is the use of such qualitative estimates of flood duration common within the DFO dataset as a whole?
I acknowledge that the authors corroborate flood frequency from DFO with the EM-DAT dataset in their response to my original review, why did this not feature in the main analysis in the revised manuscript/or at least supplementary information? Are both datasets using similar sources of news reporting? Does the temporal pattern of flood frequency match physical flood frequency as observed with river flow gauges over the same time period, notwithstanding the known issue of low density gauges in many regions?
Overall, I do not feel confident accepting these conclusions based only on the DFO dataset. This is unfortunate as I think the authors have performed a nice analysis with very interesting and extremely worthwhile hypotheses, but at this time, in my opinion, the underlying dataset by itself does not provide a strong enough foundation for the application of these hypotheses and hence conclusions.
A possible suggestion for moving this study forward would be to perform a validation of reported flood frequency and duration based on in-situ observations, at least for regions with available overlapping data (as suggested in my original review). I understand this will be more straightforward for flood frequency than for flood duration (as flood inundation is not directly measured at a flow gauging station), but the only other option would be for a consistent satellite product to be available across the full 1985-2015 study period, which I do not believe exists(?). Nonetheless, some effort towards increased corroboration and/or validation even for a range of case study regions/basins would provide a path towards supporting the conclusions. I hope the authors understand my concern and can find the time to provide more validation.
Pg 3, Line 6: There has been several authors (e.g. Merz et al., 2012) who highlight the lack of scientific rigour in attribution studies in hydrology. The terms “attribute/attributed” should be avoided if not performing a rigorous formal attribution framework.
Pg 4, Line 2 and 3: Need to mention that ‘flood duration’ within DFO is simply calculated from the flood beginning and end date, and need to expand on to what degree the start and end date is from news reporting and/or from satellite images, and over which time periods (i.e. MODIS only started in 1999)?
Pg 7, Lines 26-29 & pg 8 Lines 1-7: This is methodological detail about trend tests and would be best placed in section 2.
Pg 10, Lines 2-3: Delete/rephrase sentence on “found an abrupt shift” as no longer assessing change points.
Pg 11, Line 24: Should “30 and 40%” not be “20 and 30%” if my interpretation of Fig. 7 is correct?
Merz, B., Vorogushyn, S., Uhlemann, S., Delgado, J., and Hundecha, Y.: HESS Opinions "More efforts and scientific rigour are needed to attribute trends in flood time series", Hydrol. Earth Syst. Sci., 16, 1379-1387, https://doi.org/10.5194/hess-16-1379-2012, 2012.