Projected Climatic and Hydrologic Changes to Lake Victoria Basin Rivers under Three RCP Emission Scenarios for 2015–2100 and Impacts on the Water Sector
Round 1
Reviewer 1 Report
Comments and suggestions are in the attached file
Comments for author File: 
Comments.pdf
Author Response
REVIEWER 1
The paper analyses precipitation, temperature and annual discharge behaviour for Lake Victoria basins under three RCP scenarios considering a specific modelling configuration. The rainfall-runoff transformation process is simulated using Vector Autoregressive Moving Average Process (VARMAX) calibrated on observed precipitation (CHIRPS) and temperature.
Major comments
1) The paper is long and not easy to read, especially the descriptions of the climatic projection (line between 285 and 392) should be summarized and presented more effectively.
We have summarized the climatic projections and re-written the text in lines 285 to 392.
2) All the analyses are based on one climatological model configuration (a specific RCM nested in a specific GCM). It is well known that the projections provided by climatic model are subject to uncertainties. The uncertainties are most challenging for extremes, for projections at local scales, and for the influence of natural climate variability. The uncertainty assessment is usually based on the results of an ensemble multiple models. If such an ensemble is not available this should be properly stated and addressed, it is the main, serious, limitation for the analysis. If ensemble is available, or its analysis has been published, the position of the chosen model configuration, with respect to the average trends of the ensemble, must be evaluated.
We agree that uncertainties are a generic feature of all statistical models. In the field of climate projections ensemble multiple models are often used to address this. We did not use ensemble multiple models but provide 95% pointwise confidence bands and projection standard errors to quantify uncertainty in the projections. Similarly, we provide standard errors of the model parameter estimates, which also measure uncertainty.
The VARMAX model can also be implemented using the Bayesian Monte Carlo Iterations. This allows simulation of many possible future discharge trajectories and quantification of uncertainties in parameter estimates and future projections in terms of 95% credible limits. The greater flexibility of the Bayesian implementation of the VARMAX model is made possible through repeated sampling from the posterior distribution. But, as with Bayesian techniques in general, this can be computationally very expensive.
3) The Technical Report Climate Scenarios for Switzerland https:Unaturalsciences.ch/service/publications/107868-ch2018---climate-scenarios-for switzerland offers numerous ideas to improve the representation of results and discussions about uncertainty evaluation and bias correction of climatological variable.
4) Is the VARMAX model appropriate to represent the rainfall-runoff behaviour of a basin? References about the use of this model in hydrology are missing.
We cannot answer this question in general terms but our specific application suggests that the representation of discharge is intuitively appealing and reasonable.
5) The VARMAX model configuration, used to simulate the annual discharge under climate change is not clearly presented. The terms in 510 table are not described (e.g what is XLO_l_l? el? log_discharge(t-1) among variable does it means that discharge is used as variable to predict discharge?). Which is the configuration of the model used? Does it vary for each basin?
We have now provided additional descriptions of the parameters in the tables in question. The symbol referred simply means that discharge at time t is dependent on or is serially correlated with discharge up to p time steps earlier, at time t-p, as specified by the order (p) of the autoregressive model. Additionally, the residuals in discharge at time t is similarly correlated with the residuals up to q time steps earlier, t-q, as specified by the moving average order q. The model also includes lagged values of covariates such as rainfall up to lag s.
6) Tables are hardly understandable -proper descriptions are not provided - explanation about the name of the variable and the contents of the columns are missing (e.g.table4 what is Mavrain2, AICc, etc.)
We have now provide more detailed descriptions of the tables. Mavrain2 is the
moving average of annual rainfall computed over the current and the preceding year.
AICc is the corrected Akaike Information Criterion, etc.
7) Have been the model calibrated using CHIRPS rainfall?
The models were calibrated using the CHIRPS rainfall enriched with additional station records
by US Geological Society, Famine Early Warning System (FEWNET).
8) Have been precipitation and temperature predicted by the RCM properly bias corrected? Which technique have been used? It is well known that climatic models have biases in the representation of temperature and precipitation that must be corrected before their quantitative use.
We received help from the Intergovernmental Centre for Climate Prediction with
the preparation of the predicted precipitation and temperature data as well as bias
correction under each of the RCM scenarios
9) Paragraph 3.1- Why is presented this analysis? Was the linear model usedto simulate annual discharge under climate change?
Historic discharge was first related to historic rainfall and temperature using linear, nonlinear and semi-parametric models involving penalized cubic basis splines. But the linear models had the greatest strength of support in the data based on AICc and was therefore used for projection.
10) If the model VARMAX reproduces annual discharge why in line 427 "mean monthly discharge" is mentioned as projected variable? In the captures of figures 5, 6, 7 mean monthly discharge is mentioned - please clarify.
This was an error. Mean monthly discharge has now been replaced throughout with mean
annual discharge. We thank the reviewer for catching this error.
11) Figure 2, 3, 4 - boxplots (for specific areas) instead of maps will be probably more informative. The plotted variable is not indicated in the capture of the figure.
The figures are intended to show the spatial distribution of rainfall corresponding to the 23 catchments which the box plots will not capture.
12) lssanga river: results in figure 6 seems unrealistic, increases in 2071-2100 are of the order of 10"3 respect to the historical and 2015-2030?
We have revised and re-run the model for Issanga. All the other results are unchanged.
13) Comments and analyses about changes in seasonality of rainfall and temperature are extensively reported in all the paper but the hydrological model selected (VARMAX) does not allow to analyse the impact of these behaviour and trend on discharge. This should be properly stated in the paper. A more appropriate hydrological analysis could have been selected.
We related changes in historic discharge with annual and seasonal rainfall and temperature components. However, only the annual rainfall components had the strongest support in the data based on information criteria. Moreover, bivariate correlation analyses (reported in the manuscript) reinforce the conclusions arrived at based on the information criteria. Since we expected the apriori that variation in the annual and seasonal rainfall and temperature components affect variation in discharge we present all the results for completeness and transparency. As well, because the seasonal rainfall and temperature components may vary in opposing directions or to contrasting extents, as some indeed do, analysing their anticipated future variations under the three climate change scenarios yield important additional insights.
14) This paper analyses the total annual discharge, the impact of climate change on flood trend and magnitude cannot be scientifically addressed by this work. Paragraphs 4.1,4.2 and 4.3 should be reformulated and incorporate in the introduction because they are mainly a review of literature results, the impacts of climate change illustrated in those paragraphs are not result directly linked to the results of the paper.
We do not claim to model the impact of climate change on flood trend and magnitude. Instead, we infer what the spikes and troughs in anticipated discharge trajectories imply for floods and droughts based on known and long-standing historical knowledge for this region.
Minor comments
1) Provide the area of the basins.
The catchment area has been added in table 1
2) Line 190 - clarify that LN is the logarithm
We have now clarified that LN is the natural logarithm in the revision.
3) Line 397 - what do you mean for historic base line?
We mean the historic reference period. We have now made this clearer in the revision.
4) Figure 5, 6,7 -the lower bar of the variance is missing
We have replaced the old figures with new ones with both lower and upper standard error bars.
5) Paragraph 3.2.2 - too long
We agree and have re-written the paragraph and split it into smaller paragraphs in the revision.
6) Table at page 12 - characters are not readable
We have updated this table and the characters are now all readable.
Reviewer 2 Report
comments are given in attachment
Comments for author File: 
Comments.docx
Author Response
Reviewer 2
Review comments:
Overall the method introduced is interesting – replacing hydrological modelling with derived statistics. It is especially relevant because the study covers quite a large area and a variety of river basins.
More attention should be given to the discussion how this method compares to hydrological modelling and its validity under changing climate conditions.
Abstract:
- I would consequently use the word projections instead of predictions, the latter suggests more certainty which is not realistic in the case of climate change.
We agree and have replaced prediction with projection throughout the text.
- Please explain VARMAX the first time it is used.
We have done this.
- The text refers to 23 river basins and later to the Western part of the catchment. I suggest always using the word basin when referring to the whole Lake Victoria basin. And use catchments for the individual rivers. As it is now it is confusing.
We agree and have done this.
Introduction:-
The introduction is a bit sloppy. There is repetition and not for all information the relevance to this study is clear.
We have removed the repetitions and synthesized the paragraphs in a better way.
- The first two sentences bring exactly the same message.
We have revised this,
- Rephrase the sentence on line 39
We have revised this
- Line 41 mentions many countries : is this worldwide or in Africa.
We have revised this to read ‘globally’
- Line 50 what is the difference between agricultural output and productive actvitiy
For clarity, we have revised the sentence and removed the productive activity
- The first time LVB is used it should be fully spelled out.
We have done this.
- line 65 I would use relation instead of dependence
Ammended
- Line 89 ‘works for the situation with long term data coverage’ what is meant here and is this the case for precipitation and discharge in the area?
We have revised this to show precipitation and discharge data
- Line 94 to 97 these sentences are repetition as well.
The repetition has been removed
Materials and methods:
- At line 120 Lake Victoria Basin is finally spelled out fully – but not needed anymore.
Done
- The CHIRPS product is used, but because this is a blended product its performance and possibly its estimates vary over time – did this influence the results of this assessments
- Various studies have been conducted to review the performances of CHIRPS and rainfall data from metrological centres by regions and countries (Funk et al., 2015; Dinku 2014; Katsanos, 2015). The CHIRPS data used in this work were already evaluated in parts in numerous studies across East Africa (e.g., Funk et al., 2015; Husak et al., 2013; Mukhopadhyay et al., 2019; ).
Climate projections:
- line 152 explanation of RCPs is already done in relatively high detail in the Introduction is this needed at two locations?
We have removed the details from the introduction
2.4 Statistical analysis:
- the first line refers to forecasting – should this be climate projections?
Ammended to read projections
- Is the analysis done on both annual and monthly time-scales?
The analysis was done on the annual time scale because only annual averages of discharge data were available. We have revised the text to remove any references to month. However, changes in projected rainfall, minimum and maximum temperatures were analyzed at the monthly level and aggregated to the seasonal level (long wet season-March-April-May (MAM), short wet season—October-November-December (OND) and long dry season-June-July-August-September (JJAS)).
- Is there a good reference for the VARMAX method – please add
Yes.
SAS Institute Inc. 2018. SAS/ETS® 15.1 User’s Guide. Cary, NC: SAS Institute Inc.
- Line 205 mentions some river characteristics – were discharge extremes also considered? Yes
- Line 257: The model is only presented for annual values would this be the same for other characteristics is it possible to extent with formulas for other characteristics?
The model is presented for annual values in the manuscript because we only had access to annual averages of discharge. The model is very general and can accommodate monthly or daily values, if such data are available. As explained in the text, the model can accommodate various components of variation in such fine-resolution data, such for instance as seasonality, etc.
- The model is based on historic data – how certain are you that it also applies under future climate change conditions? Are the relations still valid when extremes are reached that were not part of the historical period?
The model is based on historical data and uses evolution equations to project future changes in discharge like other systems for time series forecasting with covariates. The uncertainly in the projected future values are quantified by confidence limits around the projected values. The model can be implemented in either frequentist or Bayesian mode using Monte Carlo Iterations. The Vector Autoregressive Moving Average Models with covariates (VARMAX) assume short-range dependence but data sets displaying long-range dependence can be readily modelled in the same framework using the Vector Autoregressive Fractionally Integrated Moving Average Models with Covariates (VARFIMAX). Out-of-sample one step ad head and multistep-ahead forecasts are supported. These allow reliable computation of forecast error and model selection criterion.
The forecasts are conditional on the projected future rainfall and temperature and the three scenarios. It is also possible to incorporate the uncertainty of parameters in the forecasts in the Bayesian context and evaluate forecasts using tests of bias, efficiency, and accuracy. Thus, while it is possible that extreme conditions may emerge in future that widely deviate from those observed in the past, this was not the case for any of the 23 rivers under any of the three emission scenarios.
If such extreme values lead to nonlinear transitions in the trajectory of change, then nonlinear structural time series (state space) models may be preferable. Such models use splines to capture nonlinearity. As explained in the text, we considered such models but they had little support in the data.
Results
- The results are presented with a lot of detail but read hard because no details on the basin are given. For example line 280 mentions a river with deviating results is there a physical reason for this?
Seem like there is a physical reason. Katonga River is a unique river among the rivers of the LVB, it connects lake George in the west and Lake victoria in the east. It generally flows eastwards however during flood conditions, part of it flows west, this could lead to a weak to negative correlation between rainfall and discharge
- Description of figure 2 to 4 last line: ‘each row corresponds to’ à ‘the rows corresponds to’
Each row corresponds to spatial patterns for the annual and seasonal components of rainfall or temperature.
- Line 340 – ‘greater warming in the southwestern part’ is there an explanation for this?
We speculate that this is linked to local processes, from previous studies the explanation that seems plausible is related to soil moisture which is a key variable in the climate system, a negative soil moisture anomaly can lead to an increase in surface temperature through a negative anomaly of evapotranspiration. This is has been projected for the JAS season in East Africa (Koster et al 2006, Dosio et al 2016)
- Line 393: remove ‘the three’
Done
- Line 396 : ‘we use the deviation of the projected river flow rate’ this should be better explained. How are the seasonal changes established and how is the change in daily variability incorporated.
By this we mean that we projected discharge and compared (subtracted) averaged values of the projected discharge for the 2030s, 2050s and 2070s with the average for the baseline or reference period. Seasonal changes were computed only for the projected rainfall, minimum and maximum temperatures and not for river discharge. To account for seasonal and daily variability in discharge, the VARMAX or VARFIMAX models would require the monthly or daily data. These were unfortunately not available to us.
- Line 427 and overall: explain how the results compare to the historical period
Done
- Line 444: give a physical explanation
Done
- Line 446 – 448: rephrase sentence
Done
- line 478: Is there a biophysical explanation
It is not very clear if there is one because a number of mechanisms such as soil moisture, topography and local processes have been proposed by previous workers. We have discussed these in the discussion.
- line 540 – 541: Rephrase second part of the sentence
Done
- line 588: ‘we infer that groundwater recharge will be more pronounced’ unclear what is mend and unclear how this comes.
We have revised to show that the impact of high rainfall is increased in gw recharge
- line 633 – 637: Split into multiple sentences
Done
- Line 646: discussion on impact for hydropower – how does this relate to the projected changes?
We have revised this section to clearly state that HEP production relies on the rivers in the LVB
- line 669: VARMAX is suitable for data poor regions – but in this situation there were years of data available. How can the statistical relations be established with little data?
The statistical relationships require rich data to reliably establish. By data poor we mean that we need much more limited types of data to make the projections.
Reviewer 3 Report
Olaka et al. provide a timely and sound manuscript that explores the future changes in climate and its associated consequences on hydrologic regime in Lake Victoria Basin. Authors provide an excellent background and description of methods and did an extensive job exploring these impacts on 23 rivers which is very impressive. They performed numerous statistical tests to back their findings and applied these findings to come up with suggestions for future mitigation and management for this basin. The flow of manuscript is smooth and writing is strong. Although there are some areas of Discussion (especially the first few paragraphs) which are more rehashing of the Results section and therefore redundant. Overall, this is useful and important contribution and I recommend publishing it after they address changes/comments below.
Authors used 3 RCPs while only using 1 Earth system model (MPI-ESM-LR). There are so many studies which emphasize on using multiple AOGCMs to capture a range of plausible projections and reduce uncertainty. I am not implying to redo all analysis using multiple AOGCMs but at least they should justify why they only used one and why did they chose this specific model. Following on that, there is no mention of uncertainty in this study! Every modeling study should accompany with uncertainty analysis but the way that authors wrote this MS, reader gets the impression that they are certain about all these projections. They need to address uncertainties associated with their studies and shortcomings in their research.
Authors only used a statistical model which at the end links all changes in discharge to precipitation, how about future changes in the region (e.g. changes in vegetation, land-use)? Any comments on these?
My other concern is about using two different datasets with different grid size; one which is CHIRPS with 0.05 degree and the other one is 0.44 degree. How do these do work together? Also what is the temporal scale of these datasets (daily, weekly)?
Last but not least, inter-annual variability is important and in graphs which show forecasting of discharge, they only look at annual time series, how about monthly and seasonal? In my opinion and according to their findings, monthly variations is more important that annual average. Also in the supplementary materials, I only got supplementary tables and no figures which they referenced in the text!
Below are some minor comments:
Line 18: add “period” after 2015-2100 and please do that for all the text.
L20-21: make the seasons similar to text; e.g. MAM
L40: add urbanization.
L48: reference?
L62: change “on” to “and”.
L65: define LVB before using it.
L68: add , after relatively.
L83: change it to spatiotemporal
L84: add “of” after “period”
L110: the under-construction dam does not contribute to 800 MW of power.
L118: add “for” after “supply”
L119: add “,” after “water stressed”.
L130: what is the resolution of time series?
L164: how about the period that covers 2100 since the last one ends on 2085.
L172: why January and February are missing in all the text?
L181: why six moving average? Justification? Why lag of 0-4 years?
L183: six lags or 5 (0-4?
L202-206: replace period with “,” when you list.
L234-241: ditto.
L244: is r2 of 0.35 significant??
So many statistical test but I did not find the results of all them in supplementary tables.
L267-272: define low, moderate, high correlation, what are the thresholds?
L400: Last column of the table should be 2071-2100, fix please.
L432: define LN (title of Y-axis)
L433: it seems to me that pattern is the same and the driver is only RCP across all rivers, discuss please.
L433: tile of graph has a typo; change “n” to in.
L435: What is the light green bar graph imposed on top of historical values? Nothing found in legend.
L450-475: no mention of seasonal variability.
L489: delet “(“ before “87%”
L524: delete “r” before “considered”.
L526: change ‘ to .
L598: change “whose” to “which its”.
L626-660: consolidate and remove redundancy with results, mostly is a rehash of the “Results”.
Author Response
Response to Reviewer
Comments and Suggestions for Authors
Olaka et al. provide a timely and sound manuscript that explores the future changes in climate and its associated consequences on hydrologic regime in Lake Victoria Basin. Authors provide an excellent background and description of methods and did an extensive job exploring these impacts on 23 rivers which is very impressive. They performed numerous statistical tests to back their findings and applied these findings to come up with suggestions for future mitigation and management for this basin. The flow of manuscript is smooth and writing is strong. Although there are some areas of Discussion (especially the first few paragraphs) which are more rehashing of the Results section and therefore redundant. Overall, this is useful and important contribution and I recommend publishing it after they address changes/comments below.
Authors used 3 RCPs while only using 1 Earth system model (MPI-ESM-LR). There are so many studies which emphasize on using multiple AOGCMs to capture a range of plausible projections and reduce uncertainty. I am not implying to redo all analysis using multiple AOGCMs but at least they should justify why they only used one and why did they chose this specific model.
The projected climate data used in this study was generated from the Rossby Center Regional Atmospheric Model (RCA4). This model was selected based on a study by Endris et al., (2015) who evaluated a number of ensemble models. This model provides one of the possible climate scenarios for the region and it important to access other possible scenario.
We agree that it is important to capture a range of plausible projection scenarios to reduce uncertainty. We do this by considering the best case (RCP 2.6), worst case (RCP 8.5) and one intermediate (RCP 4.5) scenario. For each scenario, we provide 95% pointwise projection confidence bands that further quantify the uncertainty associated with each projection scenario. Many other scenarios can potentially be considered between the two boundary scenarios but their projections can be expected a priori to fall between the two boundary scenarios. We also assess the performance of our model performance against the historical time series as the benchmark. In all the cases, the agreement between model projections and historic observations is reasonable. Finally, since we are dealing with 23 rivers and performing extensive model selection for each, considering a large ensemble of potential future projection scenarios would make the model output too extensive to condense into a single manuscript.
Following on that, there is no mention of uncertainty in this study! Every modeling study should accompany with uncertainty analysis but the way that authors wrote this MS, reader gets the impression that they are certain about all these projections. They need to address uncertainties associated with their studies and shortcomings in their research.
We explicitly consider model selection uncertainty in the relationships between discharge and rainfall and use information theoretics to choose between contending models of functional forms relating discharge to rainfall and the specific component of rainfall (lags or cumulative moving averages of rainfall). In particular, we use the corrected Akaike Information Criterion to choose models and rainfall components. Secondly, we provide standard errors of the model parameter estimates, which quantify parameter estimation uncertainty. Thirdly, we provide 95% pointwise projection confidence bands that quantify the uncertainty associated with the projections for each scenario. These are included in each graphical plot of the historic and projected future discharge trajectories for each river. Lastly, we consider uncertainty in the projected future scenarios by considering three RCP scenarios. Although we consider and quantify several sources of uncertainty, we recognize that there are many other sources of uncertainty in practice and that it is exceedingly difficult to consider all of them comprehensively in any single study.
Authors only used a statistical model which at the end links all changes in discharge to precipitation, how about future changes in the region (e.g. changes in vegetation, land-use)? Any comments on these?
We considered rainfall, minimum and maximum temperatures but model selection identified the contribution of temperature as being unimportant after accounting for rainfall. It would indeed be desirable to consider the potential contributions of other factors such as changes in vegetation, land use and so on. But we did not have good historical time series on these other factors. But changes in these other factors almost certainly affect river discharge largely by influencing rainfall, surface run-off of rain water and retention of rain water in soils.
My other concern is about using two different datasets with different grid size; one which is CHIRPS with 0.05 degree and the other one is 0.44 degree. How do these do work together? Also what is the temporal scale of these datasets (daily, weekly)?
The 0.44 degree projected dataset that is the best data resolution that is currently available for East Africa (Endris et al. 2015). The CHIRPS data sets were based on daily and aggregated to monthly totals. The same was with the projected data. In these analysis we were interested in the temporal changes and data were aggregated for the various components.
Reference
Endris Hussen Seid; Lennard Christopher; Hewitson Bruce; Dosio Alessandro; Nikulin Grigory; Panitz Hans-Juergen (2015) Teleconnection responses in multi GCM driven CORDEX RCMs over Eastern Africa. Springer. doi: 10.1007/s00382-015-2734-7
Last but not least, inter-annual variability is important and in graphs which show forecasting of discharge, they only look at annual time series, how about monthly and seasonal? In my opinion and according to their findings, monthly variations is more important that annual average.
We agree that there is substantial (monthly) seasonal variation in rainfall and hence in discharge. It would thus have been interesting to analyze interannual variation in seasonality in discharge. However, we had access only to annual discharge data but to monthly rainfall and temperature data. This permitted interannual forecasting of discharge but precluded consideration of potential changes in seasonality in discharge. But in our other recent analysis of temporal variation in rainfall in the region, we have demonstrated that rainfall seasonality is stably modulated by the movements of the Intertropical Convergence Zone (ITCZ) and as result has varied little at least since 1914 (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0202814).
Also in the supplementary materials, I only got supplementary tables and no figures which they referenced in the text!
Below are some minor comments:
Line 18: add “period” after 2015-2100 and please do that for all the text. DONE
L20-21: make the seasons similar to text; e.g. MAM DONE
L40: add urbanization. DONE
L48: reference? – MOVED (ref 15) and inserted
L62: change “on” to “and”. _DONE
L65: define LVB before using it. - DONE
L68: add, after relatively.- NOT Correct grammatically
L83: change it to spatiotemporal –Done line 83
L84: add “of” after “period” – Not adopted to keep correct grammar
L110: the under-construction dam does not contribute to 800 MW of power.- correct 800MW is from 5 stations, as stated in the sentence
L118: add “for” after “supply” –.. supply infrastructure is one compound term, we have added freshwater before supply to maintain meaning of sentence
L119: add “,” after “water stressed”. Done (line 123)
L130: what is the resolution of time series?
The river discharge time series is annual. The rainfall and temperature time series have a monthly resolution. The projected rainfall and temperature series also have a monthly resolution.
L164: how about the period that covers 2100 since the last one ends on 2085. The spatial climate projection are analysed upto 2085 this was to show the spatial distribution for rainfall and temperature while the river discharge has been projected upto 2100 using the VARMAX model,
L172: why January and February are missing in all the text?
Our explanation is that we focused on the main rainy seasons (MAM and OND) and the main dry season (JJAS).
L181: why six moving average? Justification? Why lag of 0-5 years?
We computed cumulative moving averages of prior annual rainfall spanning 0 to 5 years because annual rainfall shows irregular cylces with cycle periods of about 5 years (see e.g.
Ogutu, J. O., Piepho, H. P., Dublin, H. T., Bhola, N., & Reid, R. S. (2008). El Niño‐southern oscillation, rainfall, temperature and normalized difference vegetation index fluctuations in the Mara‐Serengeti ecosystem. African Journal of Ecology, 46(2), 132-143.
Bartzke, G. S., Ogutu, J. O., Mukhopadhyay, S., Mtui, D., Dublin, H. T., & Piepho, H. P. (2018). Rainfall trends and variation in the Maasai Mara ecosystem and their implications for animal population and biodiversity dynamics. PloS one, 13(9), e0202814.)
We then selected the moving average best supported by the data using information theoretics.
L183: six lags or 6 (0-5)?
L202-206: replace period with “,” when you list. Done 2006-2015
L234-241: ditto. Done- 262-272
L244: is r2 of 0.35 significant?? The p-value is 0.0463, which is less than the threshold p-value of 0.05 and hence this significant as are all the other r2 values.
So many statistical test but I did not find the results of all them in supplementary tables. Results of most of the statistical tests are provided in supplementary tables S1, S4-S11. Results of some of the statistical procedures used to select the VARMAX model, such as selecting the orders of the autogressive (AR) and moving average (MA) components of the VARMAX model were omitted for brevity.
L267-272: define low, moderate, high correlation, what are the thresholds?
We define high correlation as (0.50-0.83) and low to moderate correlation as 0.31-0.49.
L400: Last column of the table should be 2071-2100, fix please. DONE
L432: define LN (title of Y-axis)
LN is the Natural logarithm
L433: it seems to me that pattern is the same and the driver is only RCP across all rivers, discuss please.
There are some differences across rivers but are somewhat masked and harder to visualize because of
differences in the scale of discharge. But it is also largely true that most of the variation is across RCPs.
L433: tile of graph has a typo; change “n” to in.
Corrected.
L435: What is the light green bar graph imposed on top of historical values? Nothing found in legend.
Both bars are for historical values. The chartreuse (bright green) bar is the average of the observed historic series. The wider medium slate blue bar is the value for the historic series predicted by the fitted model. This is meant to help evaluate how well the model does with historic data.
L450-475: no mention of seasonal variability.
Projected seasonal variability was analyzed only for rainfall and temperature. For river discharge we cab only infer what projected future fluctuations in rainfall imply because our model uses annual resolution due to lack of monthly data on discharge.
L489: delet “(“ before “87%” done on Line 526
L524: delete “r” before “considered”. DONE L562
L526: change ‘ to . Done
L598: change “whose” to “which its”. Grammar maintained
L626-660: consolidate and remove redundancy with results, mostly is a rehash of the “Results”. Done
Author Response File: Author Response.docx
