I thank both the reviewers and the authors for their responses. There are a few of matters to consider to allow this paper to be published in HESS and to be completed in a major revision. Some points on which the reviews have covered and one or two I feel must be better addressed in the manuscript. My comments on this discussion phase are as follows:
1) Novelty – Reviewer 1 brings up the issue of a lack of perceived novelty about the paper. As I stated in my initial assessment I feel the paper is ‘novel’ in the sense there are very few papers that have cascaded results through from the meteorology. However the reporting is quite basic and in lack lacks scientific rigour in a number of respects. Not all have been fully teased out by the review process including:
a) There really is a complete lack of detail how the models have each been run/calibrated. This must be improved significantly including the rational behind using certain parameter/model structure/etc. limits and ranges and techniques to evaluate them
b) The authors seem to have no clear grasp of how the resultant uncertainties/sensitivity to final propagated outputs is reliant on their experimental design. In that they clearly choose a set of WRF outputs that have highly varying skill to observations but include no weighting function (or even rejection) of these outputs. Then there is such limited information on the hydrological model it’s difficult to determine what was done (and similar comments could be made for the inundation extent). The issue here is their core conclusion is based on the fact that ‘uncertainty does not increase’. Well that might be the case but without a sensible and rationale approach to the limits for each modelling component in the cascade this is not an outcome that can be justified. The paper is very weak in this regard and must be significantly improved. For example it isn’t rationale to me to have an ensemble of input uncertainties but then seemingly use a completely deterministic hydrological model and then suggest the uncertainties do not increase, this is not exploring an uncertain cascade and the conclusions cannot be justified from it (and again in the flooding part of the modelling this is deterministic). Finally just to drive this point home it seems difficult that the authors in their conclusions then recognise that models are an imperfect representation (so how can single structure/parameter combinations be justified?). I would also argue the ‘epistemic’ nature of the WRF outputs, given observed knowledge of the system was available, was not evaluated sensibly in a probabilistic/uncertainty approach such as others have done…..
c) The authors fail to discuss properly the dynamics of the inundated area and how much this is a valley filling event (so perhaps a lack of sensitivity to inundation area changes would be expected). So are other dynamics in the model changing (like simulated depths?). I also feel there needs to be much better processing of the SPOT images to show the ‘observed’ extent of the flooding and to effectively relate that to the simulated extent. This is not adequately done at the moment

I do agree with most, if not all, of the points raised by the reviewers. I think the response of the authors seems to be welcome (improving links to a wider range of research papers and improving readability and an understanding of what techniques have been deployed). If the authors cover these points well in a revised manuscript and include further work to address the points above then the revised paper will be considered for publication, but this might result in another round of discussion first as the changes are quite considerable.

Dear Authors,

I appreciate the time that you have taken to revise your manuscript. The initial reviewers were in disagreement with your paper, one suggested rejection and the other accept with changes. I tried to reflect most of the concerns from the most critical reviewer because I believed they provided the more comprehensive review.
But I cannot see how a manuscript can go forwards, with regards to cascading meteorology through to flood inundation extent, without considering the quality of the ensemble of meteorological inputs. At the moment the paper shows some WRF models have very poor skill and some are good compared to the observed data. This must be used in some way to weight (or at least reject) certain runs before cascading these inputs. It doesn't seem at all consistent with the general methodology if all these runs then drive the rest of the cascade. Please can this be addressed properly else I will have no excuse but to send this out to full review again (and may need to in any case). I hope you understand the point I am trying to make. Thanks again for all your time in making the changes you have done thus far, best wishes, Jim

The paper is now getting close to be acceptable for publication. However the 2nd reviewer has identified some omissions and also some inconsistencies which need to be addressed before this happens. These should not add too much to the paper length but the justification for differences in model skill has to be scientifically justified and the overall aims of the uncertainty cascade and what it does and doesn't do. So if the authors can address these points, respond and complete the modifications necessary we should be able to move to publication, best wishes, Jim

Whilst the author have done a good job of completing most of the authors concerns I still think they need to highlight a little more in the methods and conclusions the limitations of their lack of uncertainty evaluation of the hydrodynamic modelling.

Thats not to say this has to be included but it still has potentially some implications of the resultant uncertainties. So there does need to be a stronger and more thoughtful discussion of this in core areas of the manuscript to be clear to the reader and clear the authors recognise what they have and haven't done in the discussion and conclusion section, best wishes, Jim