|The authors are to be commended for the improvements to their manuscript and the interesting results they present. A number of issues still remain outstanding, however, which require addressing before publication is to be considered:|
1. Language remains an issue in this manuscript. I appreciate the efforts of the authors, but the language as a whole remains difficult to follow and poor translation has led to some questionable commentary in some places. The copyeditors of the journal will be able to help with some of the more cosmetic errors, but the substantive meaning of what has been said needs to be checked by the authors. In particular, I'm not sure diagnostic/prognostic variables is a useful descriptor, and in the abstract it states that runoff is 80% of total uncertainties -- please make clear this is the total of considered uncertainties, given an analysis of very large uncertainties (e.g. terrain data errors, river bathymetry) is not considered.
2. I still do not think the section on the AIC analysis provides useful conclusions in understanding model skill. The floods being simulated are extrapolations from these distributions, and the degree to which they fit 35 years of data does not illustrate how useful they are for extrapolation. If I am wrong about this, then please add further discussion explaining why this is a relevant test of the model, as I am presently unconvinced.
3. Some interpretations of the results and comparison with the literature do appear quite tokenistic, rather than contributing to a rich discussion. E.g. P18 L3-13 loosely says routing is and is not important. P18 L14-24 is a little rushed in its review of these papers and it is not clear how this relates to the study in question. P11 L5-10 suggests the spatial patterns are not consistent with Schellekens et al., but I disagree. Although the proportion of the contribution to the uncertainty by runoff forcing is fairly consistent in space (Fig 4e), the magnitude of the uncertainty (Fig 4c; the relevant figure for the point being made about Schellekens' conclusion) is larger in mountainous and arid regions.
4. Section 3.3 is still a fairly long-winded way of saying that runoff is most important. I would suggest condensing 3.3.1 and 3.3.2 without using so much text, as the conclusion is quite simple -- consider what the reader truly needs to know. A GRDC gauge is mentioned, but then no analysis is done using this, which is confusing. Section 3.3.3 is interesting, but the authors should be much more explicit about the uncertainties in the exposure data they are using: how accurate is 1km resolution population data? If you analysed sensitivities with the 90m flood model with higher resolution population data, I expect conclusions would be very different. There is lots of literature on the accuracy of these global population datasets and the impact of resolution that should be discussed.
5. Some figures require improvement. The x axes on Figs 6+7 make no sense. Fig 5 is too small also (where is the yellow cross?).