Trend Detection in Annual Streamflow Extremes in Brazil
Round 1
Reviewer 1 Report
This manuscript presents a comprehensive trend detection analysis of eight streamflow extreme indices using daily data from 1106 gauging stations in 12 hydrological regions of Brazil. This manuscript applied a Modified Trend-Free Pre-Whitening (MTFPW) method to control autocorrelation errors at the regional level and used False Discovery Rate (FDR) to avoid trend identification errors cause by multiplicity of hypothesis tests in statistics. It has the potential to be published in Water. However, several aspects should be paid more attention and need further improvement. Specific comments and suggestions are as follows:
- The abstract of the manuscript does not describe the methods used by the authors to address data autocorrelation and multiplicity, which should be included to provide a more specific understanding of the manuscript.
- In line 37, sentence like "[3] challenged..." should be changed to "Milly et al. challenged..." , it is recommended to specify the author of the references, rather than use the serial number directly as the subject or object.
- In line 107, does "Type 1 Error" refer to the “autocorrelation problem” or the “multiplicity problem” ?
- The introductory section does not seem to present the interrelationship between reservoir scheduling and extreme flows in streams. And section 2.4 also states that attribution is not the objective of this study, thus the section on evaluating reservoirs seems like a bit of a rudimentary patchwork, which would be more appropriate in a study of changing factors analysis.
- In lines 259-264, it is recommended to give more detail to explain the relationship between and .
- Why do you choose the eight Streamflow Indices for this study?
- In lines 275-302, what are the advantages of MTFPW over other methods?
- One of the purposes of the study is to compare three different methods of trend analysis. However, in section 3.2, each method is presented in a different way, so that the comparison is not intuitive.
- Please correct the index error in lines 548-549 and 556, where a warning is displayed “Erro! Fonte de referência não encontrada”.
- Why are gauges with at least 60 years of data mainly located in southeastern part of Brazil?
- How were daily streamflow data measured at 1106 gauges obtained?
- How to assess the role of reservoirs in the observed significant trends? Please explain in detail.
- In figure9, what does ‘Q7 – DoR < 0.02’ mean？
Author Response
May 20th 2022
Response to Reviewer1
Dear Reviewer 1,
We thank you for your care in reviewing this manuscript and for your constructive suggestions. Below we describe the changes we have made to address your comments. We were pleased that for the most part the manuscript was well received.
Reviewer 1:
(1) The abstract of the manuscript does not describe the methods used by the authors to address data autocorrelation and multiplicity, which should be included to provide a more specific understanding of the manuscript.
Response:
We have included the methods in the abstract. We had to change part of the original text to comply with the 200-word limit.
(2) In line 37, sentence like "[3] challenged..." should be changed to "Milly et al. challenged..." , it is recommended to specify the author of the references, rather than use the serial number directly as the subject or object.
Response:
We thank you for the suggestion. We are not used to the numbering citation configuration. We agree that including the name of the author(s) as the subject of the sentence improves readability. We have made this change in different parts of the manuscript.
(3) In line 107, does "Type 1 Error" refer to the “autocorrelation problem” or the “multiplicity problem” ?
Response:
Both autocorrelation and multiplicity can distort the probability of Type 1 Error. That is the reason we used the Modified Trend-Free Prewhitening to deal with the former, and the False Discovery Rate to account for the latter.
So, to make it clearer in the document, we have changed the sentence in Line 107 as follows,
“The effects of serial correlation in the data and the multiplicity of test in regional analysis are discussed and properly addressed in the methodology, avoiding the excessive number of gauges with false positive trends.”
(4) The introductory section does not seem to present the interrelationship between reservoir scheduling and extreme flows in streams. And section 2.4 also states that attribution is not the objective of this study, thus the section on evaluating reservoirs seems like a bit of a rudimentary patchwork, which would be more appropriate in a study of changing factors analysis.
Response:
We have decided to change the analysis regarding the reservoirs in the second part of the results. As we don’t know the dates of construction nor the operation rules of many reservoirs in our dataset, we decided to focus only on those gauges at which we are very confident reservoirs have no effect. In this second part, we show the results only for gauges that have DoR less than or equal to 0.02.
Results for this second part show that the spatial pattern of changes when the whole dataset is used is basically the same when only unaffected gauges are used. We see this as evidence that reservoirs is not the only explanation for the increase of Q7 in the southern Brazil not for the decrease of QX1D for the northern part.
(5) In lines 259-264, it is recommended to give more detail to explain the relationship between and .
Response:
Unfortunately, the comments were truncated. We believe the reviewer is referring to the Greek variable rho used in eq. (8), the variable r_1 used in eq. (9), and perhaps even the variable r^*_1 used in eq. (10).
The Greek letter rho is used to represent the true (population) and unknown value of the correlation. The variable r_1 is the estimator that uses the observations in the sample to estimate the sample correlation. And finally, the variable r^*_1 is an unbiased estimator of the correlation, which means is a different version of r_1.
We have changed the manuscript, so these differences are clear.
(6) Why do you choose the eight Streamflow Indices for this study?
Response:
We have attempted to estimate changes in different aspects of the streamflow series, ranging from low to high flows, so water managers can have a more comprehensive description of the current situation. Using these different indices allow for further and more specific impact assessments, such as changes in flood damages, which may be affected by changes in QX1d. Impacts on structures due to changes in volume, such as reservoirs, can be assessed by changes in both QX5d and QX30d. We could say the same for potential impacts on water availability and that is why we have decided to use indices, such as Q7, which is often used on water quality studies around the world and is employed by some states in Brazil in the water permit issuing process. Q30 and the other two seasonal low flow indices are often used to characterize some aspects of drougt.
These indices and similar ones are used elsewhere, such as in Chagas and Chaffe (2018), Gudmundsson et al. (2019), Zhang et al. (2016), and Bard et al. (2015).
References:
Chagas, V. B. P., & Chaffe, P. L. B. (2018). The role of land cover in the
propagation of rainfall into streamflow trends. Water Resources Research, 54,
5986–6004. https://doi.org/10.1029/ 2018WR022947
Gudmundsson, L., Leonard, M.,Do, H. X., Westra, S., &Seneviratne, S. I.
(2019). Observed trends in global indicators of mean and extreme streamflow. Geophysical Research Letters, 46, 756–766. https://doi.org/ 10.1029/2018GL079725
Zhang, X. S., Amirthanathan, G. E., Bari, M. A., Laugesen, R. M., Shin, D., Kent, D. M., MacDonald, A. M., Turner, M. E., and Tuteja, N. K.: How streamflow has changed across Australia since the 1950s: evidence from the network of hydrologic reference stations, Hydrol. Earth Syst. Sci., 20, 3947–3965, https://doi.org/10.5194/hess-20-3947-2016, 2016.
Antoine Bard, Benjamin Renard, Michel Lang, Ignazio Giuntoli, Jane Korck, Gernot Koboltschnig, Mitja Janža, Michele d’Amico, David Volken. Trends in the . hydrologic regime of Alpine rivers, Journal of Hydrology, Volume 529, Part 3, 2015, Pages 1823-1837, https://doi.org/10.1016/j.jhydrol.2015.07.052.
(7) In lines 275-302, what are the advantages of MTFPW over other methods?
Response:
We have included the following text in the manuscript,
“The MTFPW provides a good tradeoff between Type 1 and Type 2 Errors. The TFPW has a much larger statistical power that the PW method, but it fails to control the Type 1 Error. The MTFPW, based on a relatively small change in the TFPW method, provides a better control of the Type 1 Error without losing much statistical power.”
(8) One of the purposes of the study is to compare three different methods of trend analysis. However, in section 3.2, each method is presented in a different way, so that the comparison is not intuitive.
Response:
In fact, we want to show the importance of dealing with serial correlation and the multiplicity of tests. We use section 3.1 to show that both issues play a role in the results. We do this by comparing the results obtained by three different approaches, say MK, MK-MTFPW, and MK-MTFPW-FDR. The results show that ignoring these points can distort the interpretation of reality.
In section 3.2, we don’t make comparisons. All the results are based on only one approach, the MK-MTFPW-FDR, which is the one that is capable of correctly dealing with correlation and multiplicity.
The first paragraph of section 3.2 states the following, “The previous section showed the impacts that both serial correlation and multiplicity have on the results of trend detection. From this point onwards, all the results presented here were obtained by the MK-MTFPW-FDR approach, which accounts for both issues. Moreover, the control of the Type 1 Error at the regional level using the FDR procedure was applied carried separately for each hydrographic region.”
(9) Please correct the index error in lines 548-549 and 556, where a warning is displayed “Erro! Fonte de referência não encontrada”.
Response:
We apologize for these errors. The original word document of the manuscript is free of error. We think the error was made during the generation of the pdf document during the submission process. We should have checked the pdf more carefully before finishing up the submission. We will make sure this won’t happen again.
(10) Why are gauges with at least 60 years of data mainly located in southeastern part of Brazil?
Response:
In the past, the Brazilian Government heavily invested in streamflow monitoring at sites that had high potential for hydropower generation and were close to areas with large energy demand, usually the more populated regions. A great part of these sites is in the southeastern Brazil, a very populated and economically developed region then and now. This explains the fact that most of sites with at least 60 years are located in this region.
(11) How were daily streamflow data measured at 1106 gauges obtained?
Response:
We obtained the daily streamflow from the webservice provided by the National Water Agency of Brazil (https://telemetriaws1.ana.gov.br/ServiceANA.asmx). We have included this information in the Data Availability Statement of the paper, as follows,
“Data Availability Statement: The daily streamflow data can be obtained from the webservice maintained by the National Water Agency of Brazil (https://telemetriaws1.ana.gov.br/ServiceANA.asmx). The annual time series of the streamflow indices for the twelve hydrographic regions used in the paper are available in a GitHub repository (doi: 10.5281/zenodo.6551909). The scripts used in this study can also be accessed in another GitHub repository (doi: 10.5281/zenodo.6555934).”
(12) How to assess the role of reservoirs in the observed significant trends? Please explain in detail.
Response:
As we mentioned before, we have changed the second part of the analysis of our results. We now look only at those gauges that we are very confident reservoirs play no role in observed changes.
(13) In figure9, what does ‘Q7 – DoR < 0.02’ mean？
Response:
It means the gauges shown in that specific map, which displays the results for Q7, have DoR < 0.02. A value of DoR = 0.02 means that the total storage capacity, considering all the reservoirs located upstream of that gauge, is equal to 2% of the estimated long-term annual mean flow at that gauge. We assume that gauges with DoR < 0.02 are unaffeced by reservoirs.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Dear authors,
I have attached below my suggestions for your article.
All the best,
Comments for author File: Comments.pdf
Author Response
May 20th 2022
Response to Reviewer 2
Dear Reviewer 2,
We thank you for your care in reviewing this manuscript. Below we describe the changes we have made to address your comments. We were pleased that for the most part the manuscript was well received.
Reviewer 2:
(1) Major points in the article which needs clarification, refinement, reanalysis, rewrites and/or additional information and suggestions for what could be done to improve the article: 1. Presentation of figures.
Response:
This issue was also raised by another reviewer. We have updated our figures. They are all in very high quality now. Besides, during the submission process, some Figures were not correctly generated from the word document to the pdf version of the manuscript. We have corrected this as well.
(2) Minor points like figures/tables not being mentioned in the text, a missing reference, typos, and other inconsistencies: 1. Corrections to the wording or presentation of figures or tables in the text
Response:
We have made some changes where we think they were needed.
(3) Attention Line 548 and 556 error entering reference.
Response:
During the submission process, more specifically during the generation of the pdf version of the manuscript, unfortunately, cross-references to Figure 8 were lost and we haven’t noticed the problem then. We have made the correction and the cross-references are fine now.
(4) I recommend that figure 1 and figure 2 be with the scale, coordinates and cardinal points.
Response:
We have made the suggested changes to these two figures.
(5) Perhaps in figure 6 and 7 note with a, b, c, d and the panels shown (even if you have passed the method used). Pay attention to figure 9, to redo its presentation. Attention please to the presentation of the titles related to the tables.
Response:
We have included letters (a), (b), (c), and (d) in Figures 6 and 7.
(6) The analysis is an interesting one, especially due to the large area researched. The size of the area is also the one that can distort the results obtained. It is interesting to conduct new hydrological regionalization studies of this space. Could not verify the origin of the data used, the link presented at line 623 is not good.
Response:
We are glad the reviewer finds the analysis interesting. The reason for suggesting a hydrological regionalization is not totally clear to us. We understand that a more in-depth discussion on the reasons of why we obtained such a spatial pattern of trends is an important one but is beyond the scope of this paper.
We have prepared a dataset with the data used in this study and made it available in a public GitHub repository with a specific DOI number. Besides, we have also included the address of the webservice where the reader can access the daily streamflow values that were used in this study.
We have provided this information in the data availability statement at the end of the paper,
“Data Availability Statement: The daily streamflow data can be obtained from the webservice maintained by the National Water Agency of Brazil (https://telemetriaws1.ana.gov.br/ServiceANA.asmx). The annual time series of the streamflow indices for the twelve hydrographic regions used in the paper are available in a GitHub repository (doi: 10.5281/zenodo.6551909). The scripts used in this study can also be accessed in another GitHub repository (doi: 10.5281/zenodo.6555934).”
(7) Spatial analysis of long-term data can provide important information on the evolution of the hydrological regime. The use of some sets of indices is beneficial but also recommend the analysis of geographical factors that can define certain hydrological phenomena.
Response:
This is a very interesting topic related to the understanding of the temporal changes in hydrological variables, but we believe that this new analysis is beyond the scope of this paper. We understand that identifying and estimating the magnitude of trends using sound statistical methods to control Type 1 Error employing more than 1100 gauges for a country of continental size is worth publishing. And we hope the reviewer agrees with us.
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear authors, the manuscript is well conceived and informative but it has some inadequacies. I shall highlight them and the authors might improve the quality and readability of this research paper accordingly.
1. The cross-cutting explanation of the methodology should be described by a schematic and conceptual flowchart. The methodology should be described clarifying the definition of the variables and their modelling.
2. Please include all the Software codes that have been implemented in this particular research work in the appendix section for independent simulation, testing, validation and integration.
3. The introduction section does not provide a succinct theoretical basis for the study. I would like to ask the authors to expand and highlight the advantages of their approaches that bring benefits in the solved issues.
4. If possible, it will be good if the authors could add a graphical representation summarizing their results which compares controls, simulation results, all the parameters and variables directly related to streamflow extreme and serial correlation.
5. Please, briefly add future perspectives and further applied applications of this specific research work in the discussion section.
6. The techniques and/or models presented and mentioned in the manuscript require sufficient details (including calibration, sensitivity analysis and validation) to allow other researchers to develop and test the applications later on. Please include the parameters that I have mentioned here. More simulations and comparisons that show the advantages and the drawbacks of the proposed schema are needed.
7. The most relevant data-results should be summarized and demonstrated by a graph and a corresponding table.
8. Please, highlight the outliers in all the tables and graphs, where relevant.
9. Please include a geo-referenced map of the study areas shown within the map of Brazil and Brazil within a world map.
10. All the geo-referenced maps and images should be included in high resolution, so that the details and the legends in each figure could be read properly.
11. Please kindly include all the texts in the manuscript in English language. Otherwise, it gets difficult to follow and understand the entire text, properly.
12. Figure 9.b is not legible. Please make it appropriate in the revised manuscript.
13. All the maps and associated figures should be geo-referenced.
14. Please improve the quality of English Language in the manuscript.I
Author Response
May 20th 2022
Response to Reviewer 3
Dear Reviewer 3,
We thank you for your care in reviewing this manuscript and for your constructive suggestions. Below we describe the changes we have made to address your comments. We were pleased that for the most part the manuscript was well received.
Reviewer 3:
(1) The cross-cutting explanation of the methodology should be described by a schematic and conceptual flowchart. The methodology should be described clarifying the definition of the variables and their modelling.
Response:
We have included a flowchart that depicts the methodology used in this study. It includes all aspects of our study, from the estimation of the flow indices based on daily values of streamflow, to the methods used to deal with both serial correlation and multiplicity.
We have also made some small changes in the description of the methodology and clarified the use of the three variables used in the section that describes the serial correlation.
(2) Please include all the Software codes that have been implemented in this particular research work in the appendix section for independent simulation, testing, validation and integration.
Response:
We have organized the code used in this study and made it available in a GitHub repository. We have also created a DOI for this repository and included it in the “Data Availability Statement” of the paper, as presented below,
Data Availability Statement: The daily streamflow data can be obtained from the webservice maintained by the National Water Agency of Brazil (https://telemetriaws1.ana.gov.br/ServiceANA.asmx). The annual time series of the streamflow indices for the twelve hydrographic regions used in the paper are available in a GitHub repository (doi: 10.5281/zenodo.6551909). The scripts used in this study can also be accessed in another GitHub repository (doi: 10.5281/zenodo.6555934)..
(3) The introduction section does not provide a succinct theoretical basis for the study. I would like to ask the authors to expand and highlight the advantages of their approaches that bring benefits in the solved issues.
Response:
We have included a discussion about the effects of serial correlation and multiplicity on the results of trend detection, providing a general description of methods that have been devised to deal with such issues. We have also mentioned the benefits of the methods we have used.
(4) If possible, it will be good if the authors could add a graphical representation summarizing their results which compares controls, simulation results, all the parameters and variables directly related to streamflow extreme and serial correlation.
Response:
We haven’t used any simulation to compare results based on different approaches. One of the goals of our paper is to show the importance of using methods that are able to take serial correlation and multiplicity in trend detection studies. We know, based on other studies that extensively used Monte Carlo simulation, that if these issues are neglected, results can be distorted because the probability of making a Type 1 Error gets larger than the significance level. Even though these results are available in the literature, we still observe studies being published that don’t take these two issues into account in the analysis. We want to make that point clear and show the results for Brazil as a whole.
So, in our first part of the results, section 3.1, we compare the trend detection results for Brazil using three different approaches: MK, MK-MTFPW, and MK-MTFPW-FDR. The first approach is very naïve because it neglects both issues. The second approach deals with serial correlation, but it is blind to the multiplicity problem. Only the MK-MTFPW can account for both issues in the analysis. So, our way to show the effects of both serial correlation and multiplicity was to compare the results obtained by these three approaches.
Only then, we pursue our second goal that is to provide a comprehensive trend detection analysis for several streamflow extreme indices using sound statistical methods.
Having said that, we believe our results are properly presented. I think the inclusion of new figures or tables would not provide any new insight to what have already been shown.
(5) Please, briefly add future perspectives and further applied applications of this specific research work in the discussion section.
Response:
We agree with the reviewer. We have included the following paragraph in the Conclusions,
“Further research is needed to investigate the role of potential natural and hu-man-induced factors in the estimated changes in streamflow extremes. Climate variability and change, as well as human activities such as agricultural expansion, urbanization, and construction of reservoirs are all factors that must be examined in an attribution analysis that must be carried out in a more refined spatial scale.”
(6) The techniques and/or models presented and mentioned in the manuscript require sufficient details (including calibration, sensitivity analysis and validation) to allow other researchers to develop and test the applications later on. Please include the parameters that I have mentioned here. More simulations and comparisons that show the advantages and the drawbacks of the proposed schema are needed.
Response:
We don’t really understand what Reviewer 3 means with this comment. As we mentioned before, we haven’t made any simulation in our study. We performed a trend detection analysis following, in general, what most of other studies published in the specialized literature have done. In section 3.2 we simply used the MK-MTFPW-FDR approach to obtain the trend results for the whole country. And we tried to show in Section 3.1 that using the MK-MTFPW-FDR was the right thing to do.
Our study differs from some of those mentioned in the Introduction (not all of them) in the sense that we carefully considered the issues of serial correlation and multiplicity, which we and others think are very important. Actually, we don’t just think that dealing with these two issues is important, it was shown by many others in the past, based on Monte Carlo simulation, that these issues should be properly addressed in trend detection studies.
A complete Monte Carlo study to show the behavior of different methods to deal with both serial correlation and multiplicity is a very interesting step but is out of the scope of this paper. In fact, the authors of this manuscript carried out an extensive Monte Carlo study that showed the benefits of using the MTFPW to deal with serial correlation when compared to other methods, but this is the topic of another very long paper that is not published yet.
(7) The most relevant data-results should be summarized and demonstrated by a graph and a corresponding table.
Response:
Our study generated an immense amount of results. What is presented in Tables and Figures are already a summary of what has been obtained. We thought very carefully about what to present and what should be left out. We think the current Figures and Tables provides a good support for our analysis and for the good understanding of the readers.
We respectfully disagree with reviewer in this point. We think that including new Figures or Tables could harm the understanding of the paper instead of helping the reader.
We hope Reviewer 3 understands our view in this issue and accept our decision.
(8) Please, highlight the outliers in all the tables and graphs, where relevant.
Response:
Comment #8 is related to comment #7. We have thought of highlighting some number presented in Table 2. In fact, we have indeed made some changes in this direction, but the visual aspect of the final Table was not good. Table 2 is already a large table with lots of information. Trying to highlight some of the results made it harder to read.
Our Figures also provide lotos of information. Trying to highlight some points made them worse than their original versions.
So, we kindly ask the reviewer to agree with keeping Figures and Tables the way they are. We sincerely believe we should let the tables and figures the way they are presented in the current version of the manuscript, after the improvements made during the revision process.
(9) Please include a geo-referenced map of the study areas shown within the map of Brazil and Brazil within a world map.
Response:
A new map was included in the text.
(10) All the geo-referenced maps and images should be included in high resolution, so that the details and the legends in each figure could be read properly.
Response:
We have made changes in all Figures. They were all made in high resolution.
(11) Please kindly include all the texts in the manuscript in English language. Otherwise, it gets difficult to follow and understand the entire text, properly.
Response:
We believe the reviewer is referring to some errors in the pdf version of the document that appeared during the submission process. We believe these errors were made during the generation of the pdf file during the submission process because the word version of the manuscript was free of error. We apologize for the inconvenience. We will make sure the generation of the pdf file is free of error.
(12) Figure 9.b is not legible. Please make it appropriate in the revised manuscript.
Response:
We apologize for the inconvenience of finding these sorts of errors during the revision process. The word version of our manuscript had no errors at all. We believe the problem has been created during the generation of the pdf version when we submitted the paper in the journal system. We failed to catch the error and we apologize for that. We will make sure this will not happen again.
(13) All the maps and associated figures should be geo-referenced.
Response:
We have made the changes.
(14) Please improve the quality of English Language in the manuscript.I
Response:
We have sent the manuscript to two experienced researchers whose first language is English. They have made some small suggestions in the text. All suggestions were finally made in the manuscript.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have made proper improvement in the revised version. I suggest a thorough check of the whole manuscript.
Reviewer 3 Report
Dear Authors, I am okay with the changes made. Thanks.
