General comments

The authors present an analysis of drought characteristics in a Chinese river basin regulated by a large reservoir. They are concerned with drought characterization in terms of severity and duration, the delay between meteorological and hydrological drought and the role of a large reservoir in mitigating drought effects. Their work is based on statistical modelling of drought characteristics using the available data for their selected case study. Most of their work is based on already tested methods, so the originality is limited. The main contribution is the use of a Bayesian networks to model the drought propagation threshold. However, I am afraid the conclusions extracted from the discussion of results are only applicable to the specific case study under analysis.

The research topic is relevant for the readership of Natural Hazards and Earth System Sciences. However, the paper needs more elaboration before it can be finally published. The manuscript is poorly written, and many details require additional polish. It lacks a clear strategy for motivation of the work and the methodology is not entirely clear. Some figures need additional explanation and discussion.

The presentation of the work needs improvement. The background information provided in the introduction section is too general. It is a general discussion of droughts, with references to many papers, but it is not focused on the specific topics analysed in the paper or the methods that are applied. In its current drafting, it does not help the reader to understand the motivation of the work and identify where the authors go beyond the state of the art with their work. I suggest rewriting this section, stating the main research questions addressed in the work and referencing the authors that have tackled these questions before, summarizing their results and identifying the gaps to be filled by this work. The scope of the work presented in some of the references cited in the introduction (for instance, Wang et al, 2019; Wang et al., 2020) is very similar to that of this paper, and it should be discussed in greater detail.

I found the presentation of results a bit confusing. The methodology section presents the methods used, but it is not very helpful to understand the rationale for the work carried out. As it is, the paper is an exercise of fitting statistical models to data, but it lacks structure. The authors should clearly state their objectives at the beginning and devote the methodological section to present their strategy, indicating how they are going to tackle the research questions addressed in their work and why they chose the methods they are proposing. For instance, they analyse drought “resistance” without a proper definition of this term. I would have liked to see an orderly presentation of the objectives, the approach followed in their analyses, and the expected results, together with a clear definition of the metrics used to evaluate the performances of the models. It would have been very helpful to follow the presentation of results.

Apart from the lack of structure, the manuscript is not properly finished. There are too many grammar and typographical errors that indicate that the manuscript is still at an early stage of revision. For instance, the authors refer to the “tyson” polygon method (line 26), but I guess they mean “Thiessen”. The authors should have checked their manuscript for such evident errors before submission. You may find below several of such errors, but the list is far from complete.

Specific comments

The methods section is missing SRI and SPI. They are well-known drought indices, but their application is not straightforward because it requires normalization of the probability distributions. What transformations were applied and how good were the results?

The definition of drought “severity” is not clear. On line 1 of page 5, it is stated that drought severity is the “run length”, but run length is related to duration, not severity.

I have some trouble with the interpretation of the drought propagation time. It is stated that drought propagation time is the SPI time scale with the best correlation with the 1-month SRI. However, in the results section, the analyses are presented separately by months (figure 7). I interpret these results as the correlation coefficient obtained between n-month SPI and 1-month SRI for the values of each month in successive years. However, the values of the correlation coefficients change from month to month. If we focus on relatively high correlation coefficients (over 0.5), the band is larger in some months than in other. How can it be that the drought propagation time in June is 2-6 months, in July and August increases to 3-12 months and in September it drops to 2 months? Am I missing anything? I would expect some discussion of this in terms of droughts, physical processes, and reservoir operation, not only in terms of numerical results from the statistical analysis.

I am also troubled by the discussion of the propagation time. The sentence in line 5 of page 16 states: “Considering the lack of data before the construction of the DHF reservoir...” How can you make an analysis of the effect of a reservoir if there are no control data without the reservoir? The authors are attributing the differences in the results obtained in the different stations to the reservoir. How can they be sure of this if there are no data prior to the construction of the reservoir? In fact, they even state that the reservoir “weakened the drought resistance upstream the DHF reservoir”. I cannot imagine a way in which a reservoir may influence the threshold of rainfall deficit required to generate a drought of certain level in an upstream basin. This should be discussed. Could it be just variability in the data?

Finally, I was entirely lost with the presentation and discussion of the Bayesian network model to link cumulative precipitation deficit to hydrological droughts. What is the purpose of this analysis and why is a Bayesian model required? What is the utility of the conditional probability distributions shown in Figure 9? Are these distributions derived from the fitted models? What is their relationship with actual data and their usefulness? Please clarify all these questions.

Some of the figures need additional work or explanations. In Figure 1, I was not able to locate the stream gauge upstream of the reservoir (BKQ). The definition of severity is not clear in Figure 2. What are the bar charts that appear to the right of the plots in Figure 7?. Are these the PTMH?

General comment

The manuscript investigates the effects of large reservoirs on drought occurrence in the Hun River basin (China), by implementing a procedure accounting for meteorological and hydrological drought occurrence, through drought propagation within the system, included reservoirs. The manuscript is generally well written and interesting to read. Notwithstanding this, there are some minor to moderate issues that should be addressed by the Authors before the manuscript can be considered for publication. In summary, the Authors should improve the presentation of their work, especially the methodology section, in order to highlight the innovative contribution to the literature provide by their work, help readers understand results and possibly replicate the analysis for other case studies. My concerns are listed in the following; I hope that they will be helpful for manuscript improvement.

Specific comments

Abstract

The description of the work in the abstract should be improved; in fact, while reading the abstract, the work appears to be only a case study application. On the contrary, I believe that the Authors are proposing a methodological framework to evaluate the impact of meteorological droughts in regulated river systems.

Methodology

• (5) should be better explained, e.g. by adding a sketch. If this equation was previously used for the same scope, a reference should be added. I would also add a sketch describing depicting the general framework adopted here. A better description of the framework could help the reader understand the rationale behind the specific statistical tools implemented here.
• The CPD is used as condition to explain the occurrence of severity and duration of hydrological droughts; why using CPD and not meteorological drought characteristics estimated based on SPI (e.g. intensity, severity and durations of meteorological droughts)? This point is not very clear to me, and the same could be for a potential reader. Further, why using the 0.95 CPD quantile as a threshold for hydrological drought? Does it result for data analysis or from the hydrological response of the catchment?

Results

• What about drought frequency? It is mentioned at line 17 of page 9, yet not defined or investigated.
• Section 4.2. What do you mean by “periodicity”? Do you mean frequency or probability?
• At SY and XJWP stations, the correlation is very high for a large variety of SPI time scales; this makes the identification of PTMH values highly uncertain. This affects the conclusions drawn from figure 8; to overcome this issue I suggest to include in figure 8 the PTMH uncertainty bounds. This is the only major issue that should be addressed by the Authors.
• For clarity, in Figure 9 I suggest to use the same x-axis range for all D or S panels.

There are some typos in the text; I warmly suggest to revise the English language.

The manuscript ‘Comprehensive evaluation of hydrological drought and the effects of large reservoir on drought resistance in the Hun River basin, NE China‘ concerns the interesting subject of  hydrological drought and the effects of large reservoir on drought resistance.

The authors use methodologies that are adequate for the abovementioned subject.

In order for this manuscript to be accepted for publication in the Journal of ‘Natural Hazards and Earth System Sciences‘ the authors should take into consideration the following remarks:

• Improvement of structure of the manuscript, good definition of the objectives and clear presentation of the methodologies used. The above presentation should be so detailed as to help the reader.
• Linguistic improvement of the text
• Correction of errors that should be avoided: (3a) BKQ station is not presented in Figure 1. (3b) page 3, lines 17-18, change ‘agricultural irrigation’ to “irrigation”. (3c) page 3, line 27, change ‘tyson polygon’ to ‘Thiessen polygon’. (3d) Although indexes SPI and SRI are well known, they should be more detailed (refer to normal standardization). (3e) page 5, line 4, ‘…, otherwise it is defined as no drought event (c).’ According to table 1 when SRI is -1.0<SRI<=-0.5 there is Mild drought. (3f) Figure 4, XWP change in XJWP. (3g) page 7, line 24, present in a figure the linear slope of SRI changed from -0.089/10a to 0.469/10a. (3h) page 9, line 6 and table 2 use the same value -0.83 or -0.84. (3j) page 9, line 23, ‘the districts in the wast (XJWP)’ change in ‘the districts in the west (XJWP)’ (3i) page 11, lines 7-9, the values presented are out of the range of axes in the figure 6. (3k) page 12, line 12-13, ‘…PTMH ranged from 3 to 14 months, while the correlation coefficients were lower from late winter to early summer with PTMH ranged from 4 to 23 months at BKQ and DHF’ the values are out of range in figure 7. (3l) page 14, lines 3 and 4, ‘Higher temperature in summer …. low PTMH of spring and winter’ change to  ‘Higher temperature in summer …. long PTMH of spring and winter’
• Figure 9 was presented (page 16, lines 2 and 3) but it was not discussed.
• MAIN COMMENT. Page 16, lines 7-8: ‘…strengthened the drought resistance in the lower reaches of DHF reservoir while weakened the drought resistance in the upper reaches of DHF reservoir.’

It is necessary to explain how the construction of the reservoir affected the upper reaches of DHF reservoir.

• MAIN COMMENT. Page 16, lines 11-13: ‘With the increase of hydrological drought level, ….at severe hydrological drought level’. It is not enough to make an observation; authors should try to explain why this happens.

The paper focuses on the study of drought duration and severity using precipitation and runoff data. The main objective is to estimate drought propagation from precipitation to runoff deficit. Moreover, the propagation is studied using runoff data with a system including a large reservoir. Methodology includes copula estimation of the bivariate process of drought duration and severity.  I m not convinced by two points which are the test of adequacy for the copula. There is no test of the copula model. Moreover, the discussion of propagation is lacking coherency due to the fact that authors do not properly takes into account the physical system, The latter is composed by 4 hydrological stations one upstream; the reservoir two downstream the reservoir and one at the outlet of the reservoir.   

Page 2 Line 3 Hydrological drought refers to the condition when 3 the water level of a river or aquifer is lower than normal. This is not true

Page 2 line 7 the concept of drought resistance is not explained, it is not clear for me. No references are given. Any way I would think that it is not appropriate to compare the SRI severity for basins with different sizes. Severity should be scaled according to basin size.

Page 2 Line 8 Run 8 Run theory, a time series analysis method. Here Yevjevitch 1980 should be cited as first reference

Page 3 line 9 Authors need to mention the climate of the 3 studied sub regions DHF, SY, XJWP.

Pqge 3 line 27 what is tison polygon method?

Page 5 line 6 the combination of droughts is important to mention in the abstract

Page 5 line 11 why did authors selected these pdf?

Page 5 line 19 it is not clear whether the dependency between drought duration and severity is for the same variable rainfall or runoff

Page 5 Line 21 RMSE and AIC do not seem adequate to measure the adequacy of  Copula. Authors should use Kolmogorov Smirnov test or other tests based on Monte Carlo simulations and Cramer Von Mises statistics to check the copula model adequacy ( See for example the paper of Genest et al 2011 https://arxiv.org/pdf/1102.2078.pdf   )  

Page 5 line 24 how did authors computed F (d,s)?

Page 6 line 10 Did authors compute the Pearson correlation coef between SRI 1 month and SPI k months from k equal 1 to k equal 24?

Page 6 Line 13 it is not clear how authors used Bayesian network BN. There is no graphical model. The causal influences among variables are not described. May authors add a graphical model? I understood that authors contribute to identify or fit the conditional probability which is known as parameter learning component in learning BN but the structure learning is missing. A reference to Pearl papers should be included for example https://ftp.cs.ucla.edu/pub/stat_ser/R246.pdf

Page 6 Line 21 Authors do not use the conventional words within BN which are nodes and parents, response variable y here hydrological drought with two components duration and severity and the feature variable x that characterizes y and which is here is rainfall deficit

Page 7 line 21 what do authors mean by seasonal scale SRI? Is it SRI 3 months beginning in January; April etc…or is it SRI3 of 3 ;onth period beginning in January, then in February m then in March etc… ?

Page 9 line 1 figure 4 what does mean heat map?

Page 7 to 9 trend analysis should be first presented in methodology section and then in results section

Page 9 line 3 it is not clear whether the trend analysis is about SRI. It is not clear how are obtained the results of Table 2 . What are the series that are analyzed using M K test?

Page 9 line 18 why do authors analyze 1 month SRI?

Page 10 line 9 alpha equal 0.01 is problematic. Usually, we use 0.05. Other tests should be used. Also authors need to test the normality of the SPIs and SRIs series

Page 10 line 11 it is not clear whether Table 3 is about precipitation data, this should be indicated in the title

Page 10 discussion of the number of droughts. How did authors took into account the reservoir impacts on the runoff downstream ? Downstream stations SY and XJWP are affected by the reservoir operation . The comparison of the three hydrological stations does not hold because drought conditions in Downstream stations SY and XJWP are not due to natural variability but to both natural variability and storage decisions. This explains the seasonal characteristics discussed in Page 12 Line 8 and the results in Page 13 Line 7 and 8. However the comparison between DHF and BKQ holds. So in Fig. 7 c and d are not worth for comparison

Page 11 Line 14 authors choose 0.5 as reflecting moderate drought, They should explain why 0,5 corresponds to moderate drought. (the mean is greater than the median corresponding to 0.5).

Page 11 in Figure 6 and Table 5 why not representing results of station BKQ station?

Page 12 line 6 What is Poisson correlation coef?

Page 14 Line 1 to 3. I would not agree with authors conclusions, the result of 4 months as propagation time in relation with an interdistance of 40 km  is due to the fact that the reservoir exists. Otherwise I guess that the propagation time would be one month or less for such a distance  

Page 14 the complexity of the situation (lines 7 to 17) argues in favor of  drawing a Bayesian network BN  scheme.