|The paper „Characteristics of rainfall events in RCM simulations for the Czech Republic“ by Svoboda et al., has been reviewed by two referees and revised by the authors. The revisions have (a) focussed the message, (b) sharpened the complementarity to the rich literature on RCM simulations of rainfall, and (c) highlighted the general value of the results for an audience beyond the Czech Republic. These were in my opinion the three main concerns and suggestions of the referees. However I have some comments/opinions for discussion, and some issues that may still need to be addressed/clarified before final publication.|
(1) It is true that many RCM studies on precipitation extremes exist and the authors cite some of this work. It is true that many of these analyses are also event-based, so there is limited novelty in that alone (Referee 1). However, I agree with the authors that most published RCM studies did not focus on sub-daily (e.g. hourly) rainfall, simply because the required RCM runs are rather new. I also agree with the authors that there is much to be said for event-based analysis from high-resolution rainfall because it is only with this kind of data/observations that more subtle and important details of the precipitation regime (duration and number of events, event intensity fluctuations, extremes) can be gleaned. I find the novelty of the analysis more clearly stated in the revised manuscript, being the analysis of RCM hourly-scale simulated event characteristics, and the comparison of RCM and point data, i.e. the problem of spatial averaging.
(2) In the introduction the authors state that “Although growing attention has been given to studies at sub-daily time scales in recent years, the complexity of physical processes related to sub-daily extremes and their simplification within climate model parameterizations might discourage researchers from assessment of simulated sub-daily precipitation, particularly since its validation is impaired by the lack of long and high-quality observed rainfall data series at hourly or sub-hourly time scales.” This statement seems to suggest that because RCMs cannot simulating hourly rainfall accurately and we have few hourly observation datasets, scientists are discouraged to compare RCMs with observations. I don’t think this reasoning is true. First of all there are enough long (20-30 yr) hourly records world-wide to do such comparisons. Secondly, I doubt climate modellers would agree that they are discouraged to compare their simulations with data at hourly resolutions, on the contrary. Please consider revising this or explain better what you mean. Also, the statement that only a few studies have dealt with characteristics of individual rainfall events (page 3, line 17) is not true, the authors themselves contradict this by the extensive literature that exists on defining idependent storms from rainfall data (Section 3.1).
(3) In the RCM simulations chapter additional clarification si needed, e.g. (a) write explicitely what is the time resolution of the individual model outputs (Table 1), I understand only very few RCMs in ENSEMBLES generated sub-daily rainfall, was it always hourly rainfall?; (b) explain what is the sensitivity to external forcing in HadCM3, this is not clear to non-modellers; (c) state explicitly what are the emission scenarios studied.
(4) Defining rainfall events based on MIT and independence is pretty straightforward theoretically -- the work of Restrepo-Posada and Eagleson (J. Hydrol., 1982) is a key reference for this. The ad-hoc basis following USLE mentioned on page 5, line 28, is not clear to me. I do not find the reference to Wischmeier and Smith (1978) at all relevant to the identification of independent strom events. Please explain or remove this citation. The authors then analyze only the 15% largest events, and this frequency was determined from observations exceeding 12.7 mm. Is this what gives Nse? I believe this is not explicitely said anywhere. This fraction was also extracted from the RCM simulations as well. In principle I think this is a good choice, because of the bias in event number in RCM simulations in general, however that this analysis refers only to the 15% largest events should be stated more prominently in the abstract, introduction and conclusions.
(5) The method to estimate and compare the areal-average rainfall of observations and RCM grids is in my opinion clearly explained in Section 3.3 (Referee 2). The key Fig 3 shows the ratios of RCM simulations to station-averages for three sampling area sizes. The conclusion is that on the average for the 15% largest events, RCMs produce more and longer events, with lower mean depths, and therefore much lower mean intensities than spatially average rain from stations. This applies also to the extremes on the average. Interestingly the very largest quantiles of rain intensity R are grossly underestimated in RCM simulations (Fig 4) and this likely affects the shorter convective events. This leads me to the underlying question behind this paper – RCM simulations of (heavy) rain are not ready to be used for extreme event analysis without some correction. The authors in the discussion claim that bias correction is not sufficient to solve this problem, and refer to other approaches such as the delta change approach by Sorup et al. (2016). Please expand this opinion with more substantiated arguments or statements why exactly the delta change approach is better.
(6) The altitude relationships shown in Fig 6 are rather confusing to interpret, i.e. are the apparently stronger altitude relations in RCMs actually related to the spatial distribution of simulated rain. Is it possible that the same elevations in different parts of the country have completely different climatologies? What do we learn from this analysis about the spatial distribution of rain is not clear to me. Also the authors stress that the biggest problem in RCM simulated data with elevation is in the frequency of heavy rain events and the seaosonal totals (bottom of page 13), yet these poor results are not shown, even though they are mentioned in the conclusions. Altogether this part seems detached from the rest of the work.
(7) In summary, the paper presents a rather pessimistic picture of the use of RCMs without correction for rainfall. I personally would like to read also a more optimistic opinion of the authors on where do we go for here and what are the ways in which RCM rainfall is usable and/or can be improved in the future.