Quantitative Assessment of Climatic and Reservoir-Induced Effects on River Water Temperature Using Bayesian Network-Based Approach

Round 1

Reviewer 1 Report

The manuscript submitted for review presents the process of building a Bayesian network-based model, and then its use to simulate water temperature and its flow in an intact environment (by building a dam). The comparison of real and simulated values made it possible to determine the influence of climate change and independently the influence of a huge hydrotechnical structure on changes in temperature and water flow in the river. I consider the presented research results to be extremely interesting and certainly helpful in other studies aimed at determining the consequences of the construction of this water reservoir on the biocenoses of the Yangtze River. Therefore, I recommend accepting the manuscript for publication, suggesting a number of corrections indicated below:

 

Line 14: More important than changing the "regulations" (and therefore changing the law) is changing the water flow itself in the river?

Line 22: The name of the tank (Three Gorges Reservoir) must be followed by the abbreviation (TGR) used afterwards.

Line 97: Instead of "from 70–139m by October 2003" one should write: from 70 m in ...... to 139 m in October 2003. By the way, I propose to standardize the information provided - according to the data in lines 308-314, "the water level rising from 70 to 143m during 2003-2005 ”.

Lines 115-117: It is necessary to specify in what time units the time delay is defined, i.e. what do the numbers from 1 to 30 mean?

Lines 181: Why were two different markings of the studied periods introduced - post-TGD and pre-TGD (Lines 261-262) and (more commonly used) pre-TGR and post-TGR?

Line 182: What is RWT?

Lines 187-189: Since Eq. 2 determines Δ_TOT ?, move the sentence in lines 195-196 to this paragraph. ("ΔTOT is the total change of WT before and 195 after dam construction.")

Lines 197-198: Two paragraphs should be removed.

Lines 219-220: What are lag-1, lag-2, ..., lag-30? Until now, such markings have not appeared in the text.

Lines 228: Instead of "16.89 ° C" it should read "16.89 ° C". Same for "17.50 ◦C" (no spaces).

Line 237: What should mean "*" in the title of table 1, and what "•" after 1.41 in September for ANFIS in the post-TGR period?

Line 249: "As listed in Table 2," - is not needed.

Line 250-252: Inequality, which is the criterion for verifying the usefulness of the model, should be described in detail after the paragraph in lines 198-202. Besides, the criterion defined in this way is met only by the deviations of the ε models in spring and winter,  e.g. |ε_Spring| < |1,1e-4*ΔTGR_Spring.

Lines 260: Instead of "... which was the least influenced season", I propose "which indicates that the construction of the TGR had the least impact on changes in water temperature during this period."

Lines 281-289: And what does the abbreviation "M_bias" mean?

Lines 291-292: Figure 3 (b) shows only the variations of Q. The variations of WT are shown in Figure 3 (c).

Lines 307: The assumptions of this analysis should also be mentioned in the "Material and methods" section.

Lines 308-314: Transfer the information in this paragraph to the Material and methods section.

Lines 313-314: Please correct phase numbers - now all three periods are in phase 1.

Line 320: And how to comment on the situation from the turn of 2003 to 2004, when the observed WT was much higher than the simulated WT? This situation repeated itself in the years 2006/2007 and 2009/2010.

Lines 347-348: These values do not appear until Conclusions.

Author Response

Response to Comments from Reviewer 1

 

The manuscript submitted for review presents the process of building a Bayesian network-based model, and then its use to simulate water temperature and its flow in an intact environment (by building a dam). The comparison of real and simulated values made it possible to determine the influence of climate change and independently the influence of a huge hydrotechnical structure on changes in temperature and water flow in the river. I consider the presented research results to be extremely interesting and certainly helpful in other studies aimed at determining the consequences of the construction of this water reservoir on the biocenoses of the Yangtze River. Therefore, I recommend accepting the manuscript for publication, suggesting a number of corrections indicated below:

Response: firstly, it is our great appreciations for these constructed suggestions. We have made revisions to the original manuscript point by point.

 

Line 14: More important than changing the "regulations" (and therefore changing the law) is changing the water flow itself in the river?

Response: we used this sentence to demonstrate the meaning that the water flow itself has been changed by human-induced interventions (such as the dam construction) or climate change (such as air temperature variations).

 

Line 22: The name of the tank (Three Gorges Reservoir) must be followed by the abbreviation (TGR) used afterwards.

Response: we have made this revision.

 

Line 97: Instead of "from 70–139m by October 2003" one should write: from 70 m in ...... to 139 m in October 2003. By the way, I propose to standardize the information provided - according to the data in lines 308-314, "the water level rising from 70 to 143m during 2003-2005 ”.

Response: We have revised as “The TGR which is located along the main stream of Yangtze River between Chongqing and Yichang, started to impound water with the water level rising from 70m in 2003 to 143m in 2005.” And we also standardize the information provided.

 

Lines 115-117: It is necessary to specify in what time units the time delay is defined, i.e. what do the numbers from 1 to 30 mean?

Response: We have revised as “If we use  denote the target variable, time lags ranging from 1 to 30 days of air temperature which can be represented by  in this study are considered as possible input variables.” The time units of the time delay is day, so 1 to 30 means 1 to 30 days delay.

 

 

Lines 181: Why were two different markings of the studied periods introduced - post-TGD and pre-TGD (Lines 261-262) and (more commonly used) pre-TGR and post-TGR?

Response: in the Section 2 (study area and data), we have defined these two periods: “Here the pre-TGR (1984-2002) is defined as the period before the construction of reservoir while the post-TGR (2003-2013) is defined as the period with the operating TGR.” And in Section 3.3, we have showed “Due to the effect of Q over thermal inertia of the river, WT of Yichang station would be influenced by the operation of TGR. As a result, the BNM model is applied to reconstruct WT in the absence of TGR to make a quantitively assessment of the influence of TGR on WT. In order to realize the above research goal, data on Yichang were spilt into two parts: (1) a pre-TGR calibration period (1984–2002) and (2) a post-TGR period (2003–2013). The data during pre-TGR period was used for calibration. The calibrated BNM-based model was then run over the post-TGD period to predict the expected WT at Yichang in the absence of the TGR by replacing the observed value of Q to TGR with daily inflow values of Q, which represents natural discharge series getting rid of human-induced interventions (i.e. operation of TGR).” Totally speaking, the potential conditional dependence between AT, Q (natural discharge) and WT during pre-TGR period was used to predict (reconstruct) the natural WT series based on daily inflow values of Q of TGR during post-TGR period.

 

Line 182: What is RWT?

Response: It was a mistake, which has been revised as “WT”

 

Lines 187-189: Since Eq. 2 determines Δ_TOT ?, move the sentence in lines 195-196 to this paragraph. ("ΔTOT is the total change of WT before and 195 after dam construction.")

Response: we have adopted this suggestion.

 

Lines 197-198: Two paragraphs should be removed.

Response: we made this revision.

 

Lines 219-220: What are lag-1, lag-2, ..., lag-30? Until now, such markings have not appeared in the text.

Response: we made this revision as follows:“Typical network structure obtained for January. Notations for the variables are as follows: Water temperature (WT), discharge (Q). AT_1, AT_2,…, AT_30 represent the time lags ranging from 1 to 30 days of air temperature (AT) series.”

 

 

Lines 228: Instead of "16.89 ° C" it should read "16.89 ° C". Same for "17.50 ◦C" (no spaces).

Response: we have unified the unit as “The average annual AT during pre-TGR period (1984–2002) was 16.89 ℃, and increased to 17.50 ℃ during the post-TGR period (2003–2013)”

 

Line 237: What should mean "*" in the title of table 1, and what "•" after 1.41 in September for ANFIS in the post-TGR period?

Response: we have deleted these two clerical errors.

 

 

Line 249: "As listed in Table 2," - is not needed.

Response: we have made this revision.

 

 

Line 250-252: Inequality, which is the criterion for verifying the usefulness of the model, should be described in detail after the paragraph in lines 198-202. Besides, the criterion defined in this way is met only by the deviations of the ε models in spring and winter,  e.g. |ε_Spring| < |1,1e-4*ΔTGR_Spring.

Response: Since the model bias meets the condition that , the proposed BNM model can be credible to make a quantitative assessment of the influence of the TGR on WT. We have revised this criterion as |ε| < |1.1e-3*ΔTGR|) reflects that the proposed model is reasonable for quantifying the influence of the TGR on WT. And all seasonal models are capable of quantifying the influence of the TGR on WT.

 

 

Lines 260: Instead of "... which was the least influenced season", I propose "which indicates that the construction of the TGR had the least impact on changes in water temperature during this period."

Response: we have adopted this suggestion and corresponding revisions are made to the manuscript:“The summer season experienced a cooling of 0.57℃ which indicates that the construction of the TGR had the least impact on changes in water temperature during this period.”

 

 

Lines 281-289: And what does the abbreviation "M_bias" mean?

Response: M_bias denotes model bias, which has been added to title of Figure 2.

 

Lines 291-292: Figure 3 (b) shows only the variations of Q. The variations of WT are shown in Figure 3 (c).

Response: Figure 3(a) showed the variations of response relation between WT and AT from pre-TGR to post-TGR period while Figure 3(b) showed the variations of response relation between WT and Q. And Figure 2(b) and (c) show the temporal variations of WT and Q.

 

Lines 307: The assumptions of this analysis should also be mentioned in the "Material and methods" section.

Lines 308-314: Transfer the information in this paragraph to the Material and methods section.

Response: we have added these assumptions to methodology:“Distinguish the thermal effect of different operation phases at TGR: Based on the reconstructed WT time series during post-TGR, the temporal variability at seasonal scale of WT at Yichang during different operation stages of the TGR can be addressed. Since the operation of TGR experienced three phases: (1) TGR started impound water with the water level from 70m in 2003 to 143m in 2005; (2) TGR experienced a transition phase with water level rising from 145m in 2006 to 156m in 2008; (3) TGR experienced the final standard normal phase with water level ring from 145m in 2009 to 175m in 2013. ”

 

 

Lines 313-314: Please correct phase numbers - now all three periods are in phase 1.

Response: we have corrected phase numbers.

 

Line 320: And how to comment on the situation from the turn of 2003 to 2004, when the observed WT was much higher than the simulated WT? This situation repeated itself in the years 2006/2007 and 2009/2010.

Response: Since TGR acted as a heating source from October to January of next year, the simulated WT with natural discharge as input variables (Simulated WT here can be regarded as natural WT series in the absence of TGR) should be smaller than observed WT which is heated by TGR.

 

Lines 347-348: These values do not appear until Conclusions.

Response: We have added the statistics to section 4.1 “According to three performance metrics, the proposed BNM model has reduced RMSE by 16.4%, and raised the KGE and NSE by 15.5% in average compared to other two models, which recommends the BNM had a superior training effect compared to ANFIS and SVR models.”

Author Response File: Author Response.docx

Reviewer 2 Report

The authors present a study of the impact of the Three Gorges Dam on the downstream Yangtze River, and seek to separate the effects of the impoundment from climate-related impacts. One detail that would be useful to add would be a statement of from where (what level) in the reservoir the water is released, as this would have implications for water temperature given the thermal profile of reservoirs and degree of stratification observed. Figures, tables and references are all appropriate. There were a few typographical errors: line 152, date should be data; line 215, hep possibly should be helps.  Line 237 has an extraneous * at the end of the caption; otherwise a footnote is missing. Line 299, recommending probably should be indicating.  Line 347, model appears twice in the same sentence--delete the second model. Figure 3b shows a linear trend line but the graphic is clearly not linear so fitting a linear regression may not be the best statistic to use. Figure 3a is not so seriously impacted although the observations also are nonlinear. More appropriate statistics probably should be used. This, together with the level of the lake discharge, could significantly affect the conclusions regarding climatic effects versus reservoir impacts.

Author Response

Response to Comment from Reviewer 2

The authors present a study of the impact of the Three Gorges Dam on the downstream Yangtze River, and seek to separate the effects of the impoundment from climate-related impacts. One detail that would be useful to add would be a statement of from where (what level) in the reservoir the water is released, as this would have implications for water temperature given the thermal profile of reservoirs and degree of stratification observed. Figures, tables and references are all appropriate. There were a few typographical errors: line 152, date should be data; line 215, hep possibly should be helps.  Line 237 has an extraneous * at the end of the caption; otherwise a footnote is missing. Line 299, recommending probably should be indicating. Line 347, model appears twice in the same sentence--delete the second model. Figure 3b shows a linear trend line but the graphic is clearly not linear so fitting a linear regression may not be the best statistic to use. Figure 3a is not so seriously impacted although the observations also are nonlinear. More appropriate statistics probably should be used. This, together with the level of the lake discharge, could significantly affect the conclusions regarding climatic effects versus reservoir impacts.

 

Response: line 152, “date” has been revised as “data”; line 215, “help” has been corrected as “helps”;

 Line 237 has an extraneous *, we have deleted it; Line 299, “recommending” has been revised as “indicating”; Line 347, we have deleted the second “model”.

In this study, we adopted linear regression method to quantify the variations of response relation between WT and AT from pre-TGR to post-TGR period (Figure 3(a)) and the variations of response relation between WT and Q (Figure 3(b)). Former literatures about the thermal classification introduced by Kelleher et al (2012) and further developed by Piccolroaz et al (2016), the slope of the linear regression between RWT and AT can be used to distinguish two thermal patterns. So the linear regression method can quantify the difference of slope to show that the presence of the reservoir exerts as strong influence on the river thermal response.

Round 2

Reviewer 2 Report

The authors have revised their manuscript and have addressed the issues raised by the reviewer. The paper is recommended for acceptance.