General Comments

The overall concept of this paper is neatly done: a 12-member ensemble of baseline and future climate (12 km resolution) is input to a grid-based hydrological model (1 km resolution) to characterise the impact of climate change on flood events. The strength of the paper is in its focus on areal flood events, where the joint interaction between the factors that cause floods over a range of temporal and spatial scales is implicitly accommodated by the use of a gridded daily continuous simulation model. All inferences about changes to flood risk are made using 30-year sequences of daily floods, as derived from the 12-member ensemble of climate projections. Differentiating impacts by the areal extent and duration of floods of varying severity is novel, as is the exploration of possible changes in their spatial dependency.

There are, however, some aspects to this work which are potentially problematic, and these need to be addressed by further explanation and/or revision.

Specific Comments

The key issues that I am struggling with are as follows:

• It is difficult for a dynamically downscaled rainfall products to reproduce rainfall quantiles over the temporal and spatial (meso-) scales relevant to catchment flooding, and I was surprised to read (lines 62-63) that “due to the focus on … extremes rather than the whole regime in general” that no bias correction was applied. Bias-correcting projected extremes is as important, if not more important, than a central tendency measure. The rainfall-based simulation of floods is critically dependent on the correct representation of the frequency distribution of areal rainfalls, and I think it important to provide evidence that the frequency of areal rainfall extremes derived from the UKCP18 data compare reasonably well with observations. To this end, providing evidence that distributions fitted to n-day maxima extracted from UKCP18 (preferably for a range of areal extents relevant to the adopted spatial limits) are reasonably consistent with those based on observational data. I searched for any such evaluations in the Met Office documents (the citations provided for these need to be improved and corrected in the manuscript) but I could not find anything specifically relevant to the rainfall behaviour of most interest.
• On the basis of the information provided it is difficult to be comfortable with the reported probabilities of exceedance (PoE). In concept the approach of adopting a merged CDF on the basis of empirical and fitted distributions is fine, my difficulty is with the inferred annual PoEs. I suspect that there is a problem with the way that the Poisson approximation is applied, and I suggest that the authors compare (or replace) their analysis with the more straightforward approach based on fitting the GPA distribution to the POT2 series, where the annual quantiles are obtained by the simple expedient of factoring the exceedance probabilities by N/M, where N is the number of years in the record and M is the number of maxima extracted. The key reason for my discomfort with the PoEs reported is the severity of the identified events. For example, in Figure 2 it appears that 3 (possibly 4?) events with return periods of 1000 years have been observed in a single 30 year sequence. I appreciate the need to consider the influence of spatial dependency and the trading space for time issues here, but still, this number of extreme events is higher than expected (and higher than I suspect would be extrapolated by Tawn et al, 2019). A crude estimate of the likelihood of this could be obtained by estimating the notional number of largely independent catchments across the UK. If we adopt a spatial dependence limit of 120km (from line 220 in the paper) then the notional upper limit of the spatial extent of an event might be around 45000 km², which yields around 5 or so independent catchments (or “trials”) in each year. Given that the likelihood of a 1 in 1000 event occurring in a 30-year period is 0.029 (from the Binomial distribution), then there is about a 13% chance you would see a single 1000-year event in one of the five independent catchments somewhere across the UK in a 30-year period. However, we would actually need to have around 50 independent catchments in the UK to see three 1000-year events occurring in a 30-year period with any likelihood, and this corresponds to an asymptotic dependence limit of only around 40km, which is very low given the information presented in Figure 7. The number of exceedances shown in Figure 5 is larger again, but this may be due to how the ensemble members are combined (discussed in the next point).
• If my understanding is correct (lines 175-177), the 12-member ensemble from UKCP18 has been lumped together and used in the preparation of the results as summarised in Figures 3 to 7. I think this approach confounds the absolute interpretation of the reported frequencies and return periods, and I suggest that it would be more useful to treat each ensemble member as a source of aleatory uncertainty over a 30-year period. Thus, rather than reporting, say, that there are 17 events larger than 1000-year event in DJF (Fig 5) under baseline conditions, it would be more useful to report on the average (or median) frequency/quantile across the 12-member ensemble, where the highest and lowest ensemble member provides an indication of the upper and lower bounds of the sampling uncertainty in each 30-year period.
• Lastly, no discussion is provided on how the asymptotic independence metric varies with distance (lower panel, Figure 7). I think the metrics used by Coles to explore asymptotic behaviour would benefit from additional explanation here as they are not intuitively obvious; specifically, the way in which the independence metric is defined is easily misinterpreted and without explanation it appears odd that the degree of independence is decreasing with increasing distance, which is exactly the opposite of what one would expect (and as shown in the dependency metric in the upper two panels of Figure 7, which is consistent with intuition).

Rory Nathan

Department of Infrastructure Engineering

University of Melbourne

Widespread flooding dynamics changing under climate change: characterising floods using UKCP18

The authors present a study on flood changes under climate change in the UK. The focus lies specifically on changes in modelled flood return periods based on an ensemble climate projection. The main point of analysis is the changes in widespread flooding. The authors find that there is more widespread flooding in winter and less in summer in the future projected climate. Further analysis included changes in return period, area covered and duration of events between current and future climate.

Overall, I like that the article focuses specifically on simultaneously occurring flood events under climate change. The analysis and results presented here show thorough, good work. I would have wished for a bit more focus on how the uncertainty of the climate ensemble translates into the results and more discussion of the results regarding potential drivers of change. While I have lots of comments and open questions, all of them are minor and should be quick to address.

Introduction:

Since a large part of your results section talks about spatial dependence, can you motivate this analysis in the introduction? Especially since several people have already written about flood coherence/synchrony (Brunner et al. 2020) including results for the UK (Berghuijs et al. 2019).  

Methodology:

Can you elaborate on why you chose the Grid-to-Grid model for this analysis and how well it performs in streamflow/flood prediction under the current climate? This would allow some estimate how reliable future projections might be.  

I like that you thoroughly elaborate on your choice of thresholds regarding POT and inundation extend. Can you supplement this with a sentence along the lines of “Widespread events are defined as…”.

Can you elaborate more on the method chosen for asymptotic dependence and, more importantly, elaborate on what that means? I have not encountered this method before, nor did I understand by the end of the paper, what it actually tells me. If you are interested in using an established method, I can refer you again to the papers by Brunner et al, (2020) or Berghuijs et al. (2019). Their results should also be discussed in line 254 since it relates to your proposed further work.

Results:

There seems to be a mix between results and discussion in the results section (e.g. lines 184-190 are discussion, not results). You could either call the results section “Results and Discussion” or move any discussion from the results section to “Discussion and Conclusion”. Generally, the discussion could be more elaborate (see below).

You quite often talk about an “increase in the range”, “little change”, “less asymptotic dependence”, “extend slightly”. Can you support these statements with numbers?

Line 205: “On the right of some panels (future winter and autumn) is a set of events with a peak return period of at least 1000 years.” From what I see, all panels have events up and over a return period of 1000 years.

Even though you use a climate ensemble as input data for the hydrological model, the presented results mostly do not give an overview of the uncertainty the different climate projections introduce. Can you please give an indication of how the ensemble spread demonstrates uncertainty in the results? Especially since you state in the abstract: “Results were consistent across ensemble members, with none showing significant difference in distribution.” Since the two main conclusions are about the seasonal shift and spatial dependence, the results in Figure 3 are not enough to support this statement across all findings.

Figure 2: Can you include in the caption what the percent inundation refers to? Is this percent grid cells or percent land area?

Figure 3: I would prefer if you would present a summary figure for the different model ensembles. After all, since the ensemble runs represent uncertainty, only presenting, comparing and analysing individual ensemble members does not make sense. 

Figure 4: Since you are using ensemble results, can you include uncertainty bars into the event count? Secondly, the caption says that you take the sum of all ensemble results. I would think that the mean or median (and potentially even the range) is the more appropriate measure. This is the case for Figures 5 and 6 as well.

Figure 5+6: Is there a specific reason why you have return period once on the x-axis and once on the y-axis? If not, I would recommend choosing one or the other, not both.

Discussion:

Although the analysis itself does not focus on drivers of change, there have been several published articles on how hydrology and specifically floods are changing in the UK. I think the discussion would benefit from discussing the results of this study in the context of previous findings. For example, there is a projected increase in winter atmospheric rivers in the UK which are likely to bring widespread flooding (Lavers et al, 2013). Furthermore, floods in the UK are strongly associated with soil moisture timing (Blöschl et al, 2017). Do changes in the soil moisture influence in the increase/decrease of widespread flooding in the UK?

General comments:

There seems to be an issue with your referencing system. I found at least three references cited in the text to be missing in the reference list (Coles, 2001; Jiminéz Cisneros et al, 2014; and Paz et al, 2006). I did not check all of them, so there could be more. Furthermore, the reference list is not always sorted alphabetically (e.g. Robson et al and Rudd et al should be before Sayers et al) and some references do not start on a new line (e.g. Chen et al. ).

Data availability: What is EIDC?

There are missing spaces in lines 201, 223, 225, and 227, and an “s” missing in asymptotic in line 218.

Line 181: There seems to be a word missing after “widespread”.

References

Berghuijs, W. R., Allen, S. T., Harrigan, S., & Kirchner, J. W. (2019). Growing spatial scales of synchronous river flooding in Europe. Geophysical Research Letters, 46(3), 1423-1428.

Blöschl, G., Hall, J., Parajka, J., Perdigão, R. A., Merz, B., Arheimer, B., ... & Živković, N. (2017). Changing climate shifts timing of European floods. Science, 357(6351), 588-590.

Brunner, M. I., Gilleland, E., Wood, A., Swain, D. L., & Clark, M. (2020). Spatial dependence of floods shaped by spatiotemporal variations in meteorological and land‐surface processes. Geophysical Research Letters, 47(13), e2020GL088000.

Lavers, D. A., Allan, R. P., Villarini, G., Lloyd-Hughes, B., Brayshaw, D. J., & Wade, A. J. (2013). Future changes in atmospheric rivers and their implications for winter flooding in Britain. Environmental Research Letters, 8(3), 034010.