General comments

Author presents an interesting paper very dense of meteorological and hydrological information. In a sense I would say too dense. Author describes the 12 and 13 September 2019 very accurately, providing analysis of precipitation, even with the help of spatial moments of catchment rainfall, and analysis of flood response including runoff coefficients and lag times. Then a reconstruction of the model through a spatial distributed hydrological model is provided, including model performance analysis. Then author introduces a scale dependency analysis framework to relate rainfall to discharge. Finally the framework is applied by using data from an ensemble prediction system.

With all these points presented, I cannot understand the real objective of the work. From the title I understand this paper should be an analysis of the flood event, but more than this is presented. Author himself states that there are two objectives (L 90-97). I understand that the final objective should be to evaluate methods to assess hydrological risk in semiarid basins, but I think results presented are at too early stage.

So, given that the topic of the paper is very interesting to NHESS readers, I suggest to simplify the paper by removing section 5, not because it is not relevant, but to give it the proper space in a dedicated paper, to be submitted when work will be more mature.

 The author would like to thank the highly constructive comments and suggestions made by the reviewer. The primary objective of the study is to examine the main hydrometeorological processes of the 12-13 September 2019 widespread flash flooding. As consequence of the sustained rains resulting in large precipitation amounts, most of the examined catchments featured a delayed hydrological response. Until runoff thresholds were exceeded, infiltration-excess runoff generation did not start. It is in this context that the study proposes an exploration of simple scaling theory between flood magnitude and total rainfall amount. Admittedly, the reviewer is right when he/she stands that the final objective of using simple scaling theory between rainfall amount and flood magnitude should be to evaluate in depth this approach so as to assess hydrological risk in semiarid basins. However, this goal is out of the present scope of the manuscript as it just presents an exploratory analysis and the potentialities of this approach for flood risk management.

Specific comments:

L 375 Has the precipitation-discharge relationship been applied to observed data of the event?

Figure 10 shows the specific peak discharges versus total rainfall amounts for the radar-derived precipitation amounts against observed discharges and how it compares with the quantitative rainfall forecasts. As a future research step, the envelope curves could be constructed from observed rainfall amounts and the resulting discharges.  

L 415 How can the power law relationship be used to perform analysis for different return periods? Maybe using synthetic rainfall coming from intensity duration frequency curves? Is the precipitation return period equal to peak discharge return period in a highly non linear response basin like the semiarid karst ones presented in the paper?

Ideally, the envelope curves could be built from the available observed rainfall amounts and discharges for a determined catchment. These rainfall amounts would comprise different return periods. The corresponding discharges would be derived from these empirical envelope curves. No, there is not a direct relationship between the precipitation and discharge return periods in semi-arid basins with karstic substrates. As stated in the conclusions, future research should analyse long-term observations, where available, so as to establish how this relationship varies for a larger sample of extreme episodes in terms of properties in rainfall (i.e., intensity and type) and initial soil moisture conditions (i.e., seasonality) for a given basin. All these questions still remain open at this research step of the power-law relationship. The author would like to highlight that this manuscript just illustrates the potentialities of this approach from an illustrative case study.

Technical corrections

L 102 “areat”

Done

The manuscript “Hydrometeorological analysis of the 12 and 13 September 2019 widespread flash flooding in eastern Spain” by Arnaul Amengual presents an analysis of precipitation and flood response in several catchments impacted by a long-lasting heavy precipitation episode in September 2019, and in addition explores a simple scaling theory between flood magnitude and total rainfall amount. The manuscript reads well, but its structure is rather unusual. The author, after the introduction and case study section divides the manuscript into three sections (precipitation analysis, analysis of flood response and scale dependency between flood magnitude and rainfall amount), in which of each the methodology is provided and results presented. I would suggest to re-organize the paper in a more traditional structure, such that all the methodology is presented as first, followed by a section devoted to results presentation and discussion and a conclusive section in the end. Moreover, results should be inserted in a broader framework, highlighting differences/similarities with studies focusing on similar topics, which is rather missing in the current version of the manuscript.

The author would like to thank the highly constructive comments and suggestions made by the reviewer. His/her concerns have contributed to improve the structure and contents of the revised manuscript.

Concerning the structure of the current version of the manuscript, it follows the same organization than some of the previous published monographs in flash-flooding (e.g., Smith el at. 1996 and 2000; Ogden et al. 2000; Gaume et al. 2003 and 2004; Delrieu et al. 2005; Borga et al. 2007, Zanon et al. 2010), in which it is customary to divide the research in different sections, including the methods. Obviously, the author has no problem to use a more traditional structure in which all the methods are first provided and next the results are presented if requested. However, it is possible that this more traditional structure this structure could cause the reader to have to go up and down through the different sections while reading the article in order to remember which methods were applied to derive the different results.

The author agrees with the reviewer that the work could be better contextualized in the framework of previous research in flash flooding.

Please find more detailed comment below.

L57: There is plenty of studies about the use of the Generalized Extreme Value distribution, maybe the author could cite here some of the most relevant. The work by Metzger et al. (2020) was conducted with a specific focus on arid areas and concludes that the GEV and the partial duration series approach through the Generalized Pareto distribution are not optimal in the analyzed area. In the discussion, instead, they suggest to investigate the use of an emerging approach, i.e., the Metastatistical Extreme Value distribution in its full or simplified form (Marani and Ignaccolo (2015); Zorzetto et al. (2016); Marra et al. (2019); Miniussi et al. (2020) and references therein), as it leverages the use of the information contained by the whole distribution of streamflow values (in contrast to the largest ones only), is flexible in the choice of the underlying “ordinary distribution”, thus allowing for a more accurate description of the tail of the distribution (e.g., Mushtaq et al., 2021) and was proved outperforming traditional methods especially in the case of short observational records, which are of interest for the present study.

The author agrees with the reviewer concern. References about additional studies employing the Generalized Extreme Value distribution could be added in the revised version of the manuscript. The author was not aware of the novel Metastatistical Extreme Value (MEV) approach when working in this case study. Undoubtedly, the MEV distribution and its simplified form have arisen as suitable tools for dealing with the problematics of catchments featuring very limited number of floods per year or limited data series, somewhat hampering the optimal application of the Generalized Extreme Value or Generalized Pareto distributions.

L102-103: it would be helpful to see a map displaying for example the precipitation and temperature normals, so that it is immediately clear how their spatial pattern is.

Well, the author does not have available the necessary data to create these maps, although he agrees that they would be helpful. As an intermediate compromise was adopted by including a reference to a technical document by the Spanish Agency of Meteorology (AEMET) that presents this and other relevant climatological information (Chazarra-Bernabé et al. 2018)

L133-138: this paragraph needs clarification: 1) some details about the “dynamical fitting” (which is not called like this in the Cole and Moore paper) would allow a deeper understanding on the procedure employed; 2) how are the 227 independent daily pluviometers selected? Are they a completely independent dataset or are they sub-sampled from the one including 369 gauges?

Cole and Moore (2008) distinguish between static and dynamic gauge-adjustments of radar estimates. The static methods apply a mean bias correction uniformly to all the radar estimates. The dynamical approaches use gridded multi-quadric surface fitting that vary in time and space. In the latter methods the gauge-adjustment factors are calculated at the available radar temporal resolution. Therefore, these factors can be considered dynamic as they change from spatial to spatial rainfall radar pattern estimates at the different time-steps. This issue could be further clarified in the manuscript

The 227 independent daily pluviometers belong to the daily precipitation network of the Spanish Agency of Meteorology (AEMET). These stations perform daily measurements and they are different of the automatic network of the same state body, which measures rainfall with a temporal resolution of 10 minutes. Yes, they are a completely independent dataset.

L179 (and L433): how is the p-value of the OLS model computed?

The estimates computed by the OLS regression are the mean of Gaussian random variables. Then, t-statistics are computed by using these estimates and their standard errors. Next, the p-value is obtained as the probability of achieving a t value as large as or larger than the observed t value is the null hypothesis was true. That is, it is computed the upper tail probability of achieving the t values that are obtained from a t distribution with degrees of freedom equal to the residual degrees of freedom of the model. In this case, the p-value represents the probability of achieving a t value greater than the absolute values of the observed ts.

L343: what does “LNP” stand for? And why do you choose to evaluate the model performance by NSE and LNP? I suggest to add some comments on what they differ in, or the advantages/limitations of each one, so that their choice is supported.

L_NP is a skill measure of how well a model reproduces the observed system behaviour. It linearly combines various types of goodness-of-fit indices: the NSE efficiency coefficient and the error of peak time and peak discharge. Roux et al. (2011) called as the “L_NP” criterion, without any special meaning of the acronyms used. It was specifically designed to aid in warning decisions in emphasizing peak flow characteristics. That is, the L_NP score attempts to conciliate real time flood forecasting requirements with a better understanding of the physical phenomena involved in flood event generation.

Figures

Figure 9: what do the dashed lines represent?

The horizontal dashed lines in Figure 9 have no special meaning. They have been introduced just for clarifying purposes when highlighting the different hydrological responses in terms of lag time. The vertical dash line in Figure 9(e) indicates whether rainfall distribution was located near the basin outlet or the headwaters.

Typos/corrections

Done