Identifying Climate and Human Impact Trends in Streamflow: A Case Study in Uruguay
Round 1
Reviewer 1 Report
@page { margin: 0.79in } p { margin-bottom: 0.1in; line-height: 115% }Review on Identifying climate and human impact trends in streamflow: a case study in Uruguay by Rafael Navas et al..
Overall comments:
Manuscript is well written and easy to follow. In my opinion, this paper should be published with some minor revision. I have some comments/suggestion that need to address before publication.
General comments
In the Introduction section, the authors conducted a literature review of land-use changes and their impact on water resources in different regions and basins. However, the authors failed to clarify the difference(s) between their study and the similar studies of other scholars, and justify the motivations of their study. The authors should point out: what are the drawbacks in the previous studies in the region, what is lacking and what is the novelty in this research?
In the methods section, homogeneity and normality of the input data for simulation are missing. The homogeneity of data is important regarding the quality assurance and normality of the data is important for applying the parametric tests e.g. regression. It needs to give some more strict procedure to prove that the data is of such good quality. Quality of the model is determined by calculating NSE based on monthly average simulation, why daily simulated data are not used for calculating NSE.
Although the model is well cited and discharge simulation is not the main aim of this paper but residual from the simulation is used, so I think the governing equation for discharge simulation from the model should be clearly presented.
minor comments:
line 63: To be consistent use catchment instead of the basin.
Line 75-77: What are the remaining percentage is for?
Line 85/Figure 3: I would like to see the average temperature. Please justify Why PET is low during the periods (month: 5,6 7,8) with high runoff.
Author Response
Overall comments:
Manuscript is well written and easy to follow. In my opinion, this paper should be published with some minor revision. I have some comments/suggestion that need to address before publication.
General comments
In the Introduction section, the authors conducted a literature review of land-use changes and their impact on water resources in different regions and basins. However, the authors failed to clarify the difference(s) between their study and the similar studies of other scholars and justify the motivations of their study. The authors should point out: what are the drawbacks in the previous studies in the region, what is lacking and what is the novelty in this research?
Authors response: We are a bit surprised by this comment, as we feel we clarified the motivation, the difference with other studies and the novelty in lines 45 – 53 in the original manuscript. We are happy to highlight previous studies in the region and therefore have expanded the previous section to now read:
“The motivation for this work is twofold, the first is methodological. As the classic time trend analysis can be problematic [25], this study proposes a different approach, which combines rainfall runoff modelling with a GAMM regression modelling of the residuals to specifically identify the drivers of trends. The second motivation is regional, past studies in the region have only compared the runoff response to similar rainfall events, before and after land use change [10], and compared landuse effects using paired catchments [9]. The problem with these prior studies is that the approach is difficult to extrapolate to other catchments in the region, and it does not eliminate the effect of climate variability, which this study addresses.” (lines 52-59 of the new Version July 6, 2019)
In the methods section, homogeneity and normality of the input data for simulation are missing. The homogeneity of data is important regarding the quality assurance and normality of the data is important for applying the parametric tests e.g. regression. It needs to give some more strict procedure to prove that the data is of such good quality.
Author response: We think the reviewer means residuals of the model rather than input data, because even for basic regression there is no need for the input data to be homogeneous or normally distributed, rather these are the assumptions for the residuals of the regression model. The beauty of the Generalised Additive Modelling models (and GAMM, see Wood, 2017) is that there is no need for the input data, or the residuals of the model to be normally distributed or homogenous, this can be fitted by making alternative assumptions about the residual distribution in the model. However, in this case we have assumed that the residuals of the model were Gaussian. This can be easily demonstrated with the plot of the residual analysis of the S4 model (All the other models were similar):
Clearly there are only slight problems with the normality of the residuals (and note the log scale) in the QQ plot and Histogram. There is no evidence of an unequal variance in the two right hand plots.
So overall, we believe that there are no issues with the modelling in terms of meeting the assumptions of the regression model.
Wood, S.N. Generalized additive models: an introduction with R, second edition ed.; Chapman & Hall/CRC
texts in statistical science, CRC Press/Taylor & Francis Group: Boca Raton, 2017
Quality of the model is determined by calculating NSE based on monthly average simulation, why daily simulated data are not used for calculating NSE.
Author response: The results of the NSE at the daily scale is added to table 2 (for illustration purposes). The GAMM analysis was performed at monthly scale since we want to simplify the autocorrelation in the model to a basic 1st order autocorrelation (line 136-137, Version June 17, 2019). For that reason, we think that NSE at monthly scale should be kept in the table.
Although the model is well cited and discharge simulation is not the main aim of this paper but residual from the simulation is used, so I think the governing equation for discharge simulation from the model should be clearly presented.
Author response: We appreciate your comment. Nevertheless, given the complex system of equations, presented by Perrin et al (2003) [26], which summarizes 23 equations, and nonexistence of a unique governing equation: for further details of rainfall-runoff computation we prefer to refer the reader to the cited literature.
Perrin, C.; Michel, C.; Andréassian, V. Improvement of a parsimonious model for streamflow simulation. J
Hydrol 2003, 279, 275–289.
minor comments:
line 63: To be consistent use catchment instead of the basin.
Author response: Suggestion accepted, changed to catchment
Line 75-77: What are the remaining percentage is for?
Author response: The phrase has been completed as follows: In 2015 the main land use was grassland (54%), followed by small grains and row crop agriculture (34%), forestry (Eucalyptus, 9%), and urban areas (3%) (Figure 1 b).
Line 85/Figure 3: I would like to see the average temperature. Please justify Why PET is low during the periods (month: 5,6 7,8) with high runoff.
Author response: Monthly Temperature has been added to figure 3. The high runoff and low PET is logical as these months represent the winter in the southern hemisphere as we point out on line 84 and 85 in the original manuscript.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The manuscript titled “Identifying climate and human impact trends in streamflow: a case study in Uruguay” addressed an interesting topic and I endorse it for publication in the Water journal after appropriate responses to my queries. My comments to the authors are listed below:
Major comments:
The procedures used to conduct the study are not explained well. The authors can make it simple and can use a schematic presentation of the step taken in their study. This will help the audience to follow their study easily.
There are some drawbacks of using R2. Why not the authors sued some other statistical measures for model selection, for example, AIC etc. I believe that using more such statistical measures will suggest the best model for your data. This is my suggestion only.
P-5, line 126-127: You evaluated GR4J hydrologic model on monthly basis after obtaining the simulation on daily frequency. I would suggest evaluating the model on daily basis to assess the performance of the model. Evaluating the model on monthly basis ignores different type of variations and data become smoother.
P-5, line 134-135: My concern is again that why you are doing your analysis on monthly basis. It will be better to do your work on daily basis as there are different types of variations and on monthly basis these variations are not considered in the analysis.
Table 2: Please calculate the mentioned statistic on daily frequency and you will find different values (most probably lower than these values).
Minor comments:
P-2, line 60: Don’t need to write the D and C capital as it is the continuation of the above statement.
P-10, line 213-15: You stated that “Model L2 and S2 have higher adjusted r2 than models L1 and S1, indicating these models explain more of the variation in the data”. Models explain the variation in which data, response variable or other data?
P-11, Line 246-48: Please change the wording “which at this point we cannot explain” in the sentence “Looking at the behavior, s(S) has a smooth trend with a peak during the winter, indicating a decrease in the streamflow, which at this point we cannot explain”.
P-11, Line 252: I suggest it is better to use pattern instead of shape.
Author Response
Reviewer 2
The manuscript titled “Identifying climate and human impact trends in streamflow: a case study in Uruguay” addressed an interesting topic and I endorse it for publication in the Water journal after appropriate responses to my queries. My comments to the authors are listed below:
Major comments:
The procedures used to conduct the study are not explained well. The authors can make it simple and can use a schematic presentation of the step taken in their study. This will help the audience to follow their study easily.
Author response: a schematic representation of the steps has been included in section 3 (new Figure 4.)
There are some drawbacks of using R2. Why not the authors sued some other statistical measures for model selection, for example, AIC etc. I believe that using more such statistical measures will suggest the best model for your data. This is my suggestion only.
Author response: Thank you for your suggestion, and we agree that if we were focussed purely on creating the best model for a prediction and associated model selection, the AIC would be the best performance criteria. Here we have however focussed on the adjusted r2 because it can be directly translated into the amount of variance in the data that is explained by the model. In the context of identifying trends in the model this is a more useful parameter.
P-5, line 126-127: You evaluated GR4J hydrologic model on monthly basis after obtaining the simulation on daily frequency. I would suggest evaluating the model on daily basis to assess the performance of the model. Evaluating the model on monthly basis ignores different type of variations and data become smoother.
Author response: Thank you for your comment, we added the skill of the GR4J model at daily scale to table 2.
P-5, line 134-135: My concern is again that why you are doing your analysis on monthly basis. It will be better to do your work on daily basis as there are different types of variations and on monthly basis these variations are not considered in the analysis.
Author response: We currently focus on the monthly basis. Of course, as you mention in your comment: “Evaluating the model on monthly basis ignores different type of variations and data become smoother”. However, as our data to explain trends (Land cover and water licences) is only at a monthly scale, we don´t believe that we can analyse this correctly at a daily scale. The question is also how much of the daily variation is actually related to the land use and water licence variation. We have, however, in the rainfall runoff modelling removed the climate variation at the daily scale, and climate is the main driver of the daily variation. Finally, run GAMM at monthly basis gives the next two advantages: “The aggregation to the monthly time step is firstly to simplify the management of autocorrelation in the model to a basic first order autocorrelation and secondly to speed-up the GAMM computations” (lines 135-137, Version June 17, 2019). Those are the main reasons why we upscale to monthly time steps.
Table 2: Please calculate the mentioned statistic on daily frequency and you will find different values (most probably lower than these values).
Author response: Suggestion Accepted goodness of fit on daily frequency has been added to table 2
Minor comments:
P-2, line 60: Don’t need to write the D and C capital as it is the continuation of the above statement.
Author response: Suggestion accepted.
P-10, line 213-15: You stated that “Model L2 and S2 have higher adjusted r2 than models L1 and S1, indicating these models explain more of the variation in the data”. Models explain the variation in which data, response variable or other data?
Author response: Changed to “Models explain more of the variation of TRR”
P-11, Line 246-48: Please change the wording “which at this point we cannot explain” in the sentence “Looking at the behavior, s(S) has a smooth trend with a peak during the winter, indicating a decrease in the streamflow, which at this point we cannot explain”.
Author response: Suggestion accepted, the phrase has been dropped since some hypothesis to explain this result are already given in the discussion section.
P-11, Line 252: I suggest it is better to use pattern instead of shape.
Author response: Suggestion accepted
