Simulating the Impacts of Climate Change on the Hydrology of Doğancı Dam in Bursa, Turkey, Using Feed-Forward Neural Networks

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents a well-structured and timely study applying feed-forward artificial neural networks (ANNs) to assess the hydrological impacts of climate change on the Doğancı Dam in Turkey. The paper is methodologically sound, with a clearly defined modeling framework, comprehensive dataset, and thorough performance evaluation of different ANN training algorithms. The use of multiple meteorological input combinations adds analytical depth, and the study offers practical implications for water resource planning under changing climatic conditions. Revision Suggestions as following:

1.Strengthen the Introduction with Broader Context and Justification.The introduction provides useful background but could benefit from a deeper discussion on the global importance of dam hydrology under climate change. Consider emphasizing the broader relevance of the study by citing more recent global or regional works addressing dam vulnerability, ANN-based forecasting, or climate resilience in water systems.

2. Add More Supporting References. In several parts of the Introduction and Method sections, the discussion of ANN advantages or applications would be strengthened by referencing additional recent studies, particularly from the last 5 years. This will enhance the academic depth and relevance of the literature review. Such as: Climate change impact assessment on a German lowland river using long short-term memory and conceptual hydrological models;

Assessing climate change and human impacts on runoff and hydrological droughts in the Yellow River Basin using a machine learning-enhanced hydrological modeling approach；

Specific impacts of climate change on the hydrological patterns and land use dynamics in the Arghandab River Basin, Kandahar, Afghanistan；Exploring the temporal and spatial effects of city size on regional economic integration: Evidence from the Yangtze River Economic Belt in China.

3. Improve Image Resolution and Clarity. Some figures (e.g., Figures 5, 6, and 7) appear to be of low resolution in the current PDF version. Enhancing the resolution and ensuring clarity of axis labels and legends would improve readability and professional presentation.
4. Enhance Comparative Discussion with Related Studies. The Results and Discussion section would benefit from a more explicit comparison with findings from similar studies (e.g., on other dams or using different modeling techniques). Discussing consistencies or discrepancies would help situate your results within the existing literature and demonstrate their novelty.

5.Briefly Mention Practical or Policy Implications in the discussion part. While the conclusion touches on the implications for water management, it would be helpful to signal the relevance for policymakers or practitioners earlier in the paper—possibly at the end of the Introduction or in the Abstract.

Author Response

Reviewer 1

This manuscript presents a well-structured and timely study applying feed-forward artificial neural networks (ANNs) to assess the hydrological impacts of climate change on the Doğancı Dam in Turkey. The paper is methodologically sound, with a clearly defined modelling framework, a comprehensive dataset, and a thorough performance evaluation of different ANN training algorithms. The use of multiple meteorological input combinations adds analytical depth, and the study offers practical implications for water resource planning under changing climatic conditions. Revision Suggestions as following:

1. Strengthen the Introduction with Broader Context and Justification. The introduction provides useful background but could benefit from a deeper discussion on the global importance of dam hydrology under climate change. Consider emphasizing the broader relevance of the study by citing more recent global or regional works addressing dam vulnerability, ANN-based forecasting, or climate resilience in water systems.

Answer: We thank the reviewer for this insightful suggestion. In response, we have expanded the Introduction section of the manuscript to better highlight the broader relevance of our study within the global context of climate resilience and dam vulnerability. A new paragraph has been added that references recent studies which demonstrate the application of ANN models in forecasting dam inflows and evaluating infrastructure risks under climate variability. The paragraph states:

“In recent years, the global scientific community has placed increasing emphasis on the resilience of water infrastructure, particularly dams and reservoirs, to resist the adverse impacts of climate change. The intensifying frequency of extreme weather events, including droughts and floods, has underscored the urgent need for accurate forecasting tools that can support climate-resilient water management strategies. Artificial neural networks have gained prominence in this regard. A related study modeled the impact of climate change on the water quality of a dam using artificial neural networks, demonstrating high predictive accuracy [27]. Recent studies, such as Kim, et al. [28] have demonstrated the effectiveness of ANN models in predicting monthly inflows to the reservoir using climate indices, enabling proactive adaptation to variable climatic conditions. Similarly, another research integrated ANN-based modeling with 30-year precipitation data to assess flood risk and operational vulnerability in Korean multipurpose dams, offering critical insights for infrastructure planning [29]. These advancements reflect a broader paradigm shift, as noted in the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report [30], which identifies machine learning models as key components of adaptive and anticipatory water governance.”

 

2. Add More Supporting References. In several parts of the Introduction and Method sections, the discussion of ANN advantages or applications would be strengthened by referencing additional recent studies, particularly from the last 5 years. This will enhance the academic depth and relevance of the literature review. Such as: Climate change impact assessment on a German lowland river using long short-term memory and conceptual hydrological models;

Assessing climate change and human impacts on runoff and hydrological droughts in the Yellow River Basin using a machine learning-enhanced hydrological modeling approach

Specific impacts of climate change on the hydrological patterns and land use dynamics in the Arghandab River Basin, Kandahar, Afghanistan:Exploring the temporal and spatial effects of city size on regional economic integration: Evidence from the Yangtze River Economic Belt in China.

Answer: We thank the reviewer for the constructive suggestion. In response, the following paragraph has been incorporated in the Introduction to address this recommendation:

“Current advances have expanded the capabilities of ANN-based models, incorporating deep learning architectures and hybrid approaches for better climate impact predictions. For example, enhanced machine learning hydrological models (such as soil and water assessment tools and random forest algorithms) have been applied to the Yellow River Basin to assess the combined effects of climate change and anthropogenic activity on monthly runoffs and droughts [31]. While ANN is preferred for hydrological forecasting, its applications have also extended to interdisciplinary areas like regional economic integration and spatial planning under climate variability, as demonstrated in the Yangtze River Economic Belt study by Chen, et al. [32]. These examples demonstrate the growing academic and practical relevance of ANN and related AI models in climate resilience and water systems.”

3. Improve Image Resolution and Clarity. Some figures (e.g., Figures 5, 6, and 7) appear to be of low resolution in the current PDF version. Enhancing the resolution and ensuring clarity of axis labels and legends would improve readability and professional presentation.

Answer: We thank the reviewer for the helpful recommendation. Hence, Figures 5, 6, and 7 have been revised to improve their resolution and enhance the clarity of labels.

 

4. Enhance Comparative Discussion with Related Studies. The Results and Discussion section would benefit from a more explicit comparison with findings from similar studies (e.g., on other dams or using different modeling techniques). Discussing consistencies or discrepancies would help situate your results within the existing literature and demonstrate their novelty.

Answer: We appreciate this suggestion. In response, we have revised the Results and Discussion section to include a more explicit comparison with findings from similar studies involving different dams and modeling techniques. The following paragraph has been added to address the comment:

“Other modeling approaches have also been applied for dam inflow prediction, yielding varying levels of performance. For instance, Awan and Bae [66] utilized an adaptive neuro-fuzzy inference system to forecast long-term monthly inflows to major dams in South Korea using diverse meteorological input variables. Their optimal model achieved a correlation coefficient of up to 0.97. In a separate study on Egypt’s Aswan Dam, a forecast-based adaptive reservoir operation framework was implemented to evaluate the dam’s long-term resilience under climate change scenarios. This approach, which employed monthly precipitation and temperature data, achieved a maximum correlation coefficient of 0.92 [67]. Another model, i.e., the variation analogue model, was modified by Amnatsan, et al. [68] for the standardized inflow management of Sirikit dam in Thailand under low-flow and high-flow periods. Their model produced a minimum root mean square error of 115.55 and a peak R value of 0.98. In comparison, the present research demonstrates competitive predictive performance, further validating the reliability and robustness of the proposed ANN-based modeling approach.”

5. Briefly Mention Practical or Policy Implications in the discussion part. While the conclusion touches on the implications for water management, it would be helpful to signal the relevance for policymakers or practitioners earlier in the paper—possibly at the end of the Introduction or in the Abstract.

Answer: We appreciate the reviewer’s suggestion to emphasize the practical implications of the study. In response, we have added a sentence to the end of the Introduction to highlight the relevance of our findings for policymakers and practitioners. These additions underscore the utility of the proposed modeling framework in guiding data-informed dam operation strategies and climate adaptation planning. The sentence added is:

“By enhancing the accuracy of hydrological forecasting under climate variability, the proposed modeling framework can support decision-makers in developing robust water management plans and adapting operational strategies to ensure long-term water security.”

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript "Mapping Water Yield Service Flows in the Transnational Area of Tumen River" utilized the artificial neural networks to determine the relation of water quantity, i.e., water level, water inflow, and outflow fluctuations in Doğancı dam in Turkey with the changing climatic parameters. The study tackles the widespread global issue of climate-related factors that pose a threat to water security worldwide. The results of the study offer practical value for governmental bodies, water resource planners, and climate adaptation policymakers. However, a few areas need improvement to enhance clarity, scientific rigor, and readability.

1. Line 18-38: The abstract does not clearly state the novelty or comparative advantages of this study relative to existing work. It is recommended to explicitly highlight the innovative aspects (e.g., methodological advancements, unique input variable combinations, or regional applicability) to strengthen the contribution and impact of the research.
2. Line 40: The graphical abstract is overall well-designed, but I think it could be slightly more refined.
3. Introduction: A more structured critical analysis of methodological advancements and key challenges in existing studies is needed.
4. Introduction: It is recommended to further supplement the recent climate change trends in Bursa, Turkey (such as specific data on temperature and precipitation changes) to enhance the urgency of the study.
5. Line 133: The map on the left in Figure 1 is too blurry, which seriously affects its readability.
6. Line 356-358: Enhancing the color palette of Figure 5 would not only improve the overall coherence of the article but also make the information it presents more intuitive.
7. Line 387-388: The font sizes in the left and right pictures of Figure 7 are inconsistent.

 

Author Response

Reviewer 2

The manuscript "Mapping Water Yield Service Flows in the Transnational Area of Tumen River" utilized the artificial neural networks to determine the relation of water quantity, i.e., water level, water inflow, and outflow fluctuations in Doğancı dam in Turkey with the changing climatic parameters. The study tackles the widespread global issue of climate-related factors that pose a threat to water security worldwide. The results of the study offer practical value for governmental bodies, water resource planners, and climate adaptation policymakers. However, a few areas need improvement to enhance clarity, scientific rigor, and readability.

1. Line 18-38: The abstract does not clearly state the novelty or comparative advantages of this study relative to existing work. It is recommended to explicitly highlight the innovative aspects (e.g., methodological advancements, unique input variable combinations, or regional applicability) to strengthen the contribution and impact of the research.

Answer: We appreciate the reviewer’s valuable suggestion. In response, we have revised the Abstract to highlight the novelty and comparative strengths of the study. The following sentence has been added to the Abstract to address this comment:

"The novelty of the study lies in the use of certain meteorological inputs, particularly snow depth, for dam inflow forecasting, which is seldom explored. Moreover, the study compares two widely used ANN training algorithms and applies the modeling framework to a region of strategic importance for Turkey’s water security."

This addition clarifies the methodological advancements and regional relevance of the research.

2. Line 40: The graphical abstract is overall well-designed, but I think it could be slightly more refined.

Answer: We acknowledge the reviewer for the positive feedback on the graphical abstract. In response to the suggestion, we have made slight improvements to enhance its clarity and visual appeal.

1. Introduction: A more structured critical analysis of methodological advancements and key challenges in existing studies is needed.

Answer: We thank the reviewer for this suggestion. In response, we have refined the Introduction to include a more critical discussion of methodological advancements and limitations in the context of hydrological modeling. Artificial neural networks (ANNs) are compared with statistical methods. Additionally, the advantages of ANN models in handling non-linear and multivariate relationships have been highlighted.

To address this comment, the following sentences have been added to the Introduction:

"The traditional statistical models struggle with capturing the non-linear and dynamic interactions that define hydrological processes. Some established models, such as the Penman-Monteith model, rely heavily on site-specific data. When generalized or non-local data are used, these models often produce overestimations and lead to significant uncertainty in estimation results [10]. In contrast, artificial intelligence or machine-learning techniques can learn complex patterns directly from data without assuming a functional relationship, making them highly suitable for non-linear systems [11, 12].”

"ANNs are capable of approximating complex functional relationships. This makes them particularly suited for modeling hydrological systems influenced by multiple interdependent variables [16]. Furthermore, recent developments have improved artificial neural networks to more sophisticated architectures like long short-term memory networks and hybrid models that integrate fuzzy logic, or ensemble learning to improve accuracy, better data interpretation and generalization [17].”

We have also ensured that the Results and Discussion section includes a comparison between our findings and those from previous similar studies (Table 3), demonstrating the efficiency and robustness of the proposed models in relation to established benchmarks.

2. Introduction: It is recommended to further supplement the recent climate change trends in Bursa, Turkey (such as specific data on temperature and precipitation changes) to enhance the urgency of the study

Answer: To address the comment regarding the need to supplement recent climate change trends in Bursa, Turkey, we have added the following sentences to the Introduction section for clarification:

“In Bursa, long-term meteorological data from 1984 to 2013 show that the highest and lowest average monthly temperatures recorded were 25.1 °C and 5.6 °C, with the mean annual temperature being between 14 and 16 °C. This shows a significant fluctuation in the air temperature over the years. Likewise, the simulations performed in a study by Katip [34] using the standardised precipitation index were indicative of meteorological and agricultural drought in Bursa. This implies the growing vulnerability to climate extremes in the region.”

5. Line 133: The map on the left in Figure 1 is too blurry, which seriously affects its readability.

Answer: We thank the reviewer for pointing this out. In response, the map on the left in Figure 1 has been replaced and revised to improve its visual clarity and readability.

3. Line 356-358: Enhancing the color palette of Figure 5 would not only improve the overall coherence of the article but also make the information it presents more intuitive.

Answer: We sincerely thank the reviewer for the valuable suggestion regarding the color palette of Figure 5. However, the figure was generated using MATLAB’s default plotting functions to ensure clarity and consistency. Therefore, adjusting the color palette within the existing framework might present some challenges.

7. Line 387-388: The font sizes in the left and right pictures of Figure 7 are inconsistent.

Answer: We appreciate the reviewer for noticing this mistake. The figure has been revised to ensure consistency in font sizes between the left and right images, improving overall clarity and uniformity.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

Reviewer Comments

General Aspects

The introduction is constructed with the presentation of related articles. However, the study's importance section does not clarify its contribution to science. The study has new applicability in the local context, but in terms of innovation, it does not present a clear differential compared to other existing studies.

The manuscript should specify the proposed model's advantages compared to established evaporation and water balance models in the literature and detail its improvements over physical models. For instance, do existing approaches, such as Penman's reservoir balance combined with a hydrological model, not achieve similar outcomes?

Methodology

The paper mentions multiple reservoirs within the basin. It should be clarified whether these reservoirs interfere with the flow rates at the Doğancı Dam and whether this factor was considered in the analysis.

Figure 1 can be improved. It should include the contributing watershed area and the locations of the meteorological stations from which data were obtained.

ANN optimisation using Evolutionary Algorithms (EAs) and their variations has shown promising results compared to Non-Linear Programming models such as Levenberg-Marquardt (LM), which can converge to local optima. The authors should state why LM was employed instead of EAs to train the ANN.

The description of the neural network model (potentially sections 2.1 and 2.3 to 2.5 in the manuscript) can be summarised. Given that MATLAB and its standard systems were used in the implementation, standard references should be indicated.

The methodology employs monthly data. The authors should explain why daily data were not used, as daily data might capture hydrological dynamics with more detail.

Figure 4, which illustrates the ANN architectures for Model 1 (a), Model 2 (b), and Model 3 (c), shows minimal structural differences beyond the number of input neurons (five for Model 1, two for Model 2, and three for Model 3). The authors should consider whether presenting all three diagrams is necessary or if one representative diagram, noting the variations in input layer size, would be sufficient.

 

 

 

 

Results and Discussion

The presentation of results is confusing in parts. Units are missing on the axes in figures (e.g., Figures 5, 6, and 7). The methodology states that monthly data were used; however, input parameters for Model 1 are described with "daily" terms (e.g., "total daily global solar radiations," "monthly average total daily solar intensity," etc.). This discrepancy regarding the data time scale requires clarification.

 

 

 

 

Evaluating the model's performance against time series is fundamental for analysing flow, level, and hydrological balance. For the best-performing model, it is strongly recommended to present time-series plots of observed and calculated output variables (dam level and flow rates).

If possible, the error metrics in Table 3 (comparison with previous studies) should be standardised to ensure a consistent comparison. Table 3 should be moved to the introduction or a dedicated literature review section, as its purpose is to contextualise the current study within existing research rather than to present direct results of this work.

 

 

Figures 5a and 5b, which depict results for Model 1, show data points tightly clustered along the Y = T line due to high correlation coefficients (R values of approximately 0.999). This tight clustering, particularly if values are scaled or originate near zero, might be misinterpreted (e.g., as a single point at (0,0) by the user). The authors should ensure the clarity of these plots to avoid potential misinterpretation of the data distribution shown.

Comments on the Quality of English Language

Sometimes sentences sound a bit literal or could be rearranged for a more idiomatic English flow. For example, it is reasonable to say "Climate change is one of the significant environmental issues at the present time"; but, "Climate change is currently one of the most significant environmental issues" would sound somewhat more natural. "To the author's knowledge, this form of study is novel in the local context, making it unique and equally important from a water management point of view," says another example. The term "equally important" begs the issue "equally important to what?". Maybe more direct would be "also important" or "particularly important". These are not major mistakes endangering knowledge; polishing would help.

Concision: The book might occasionally be just slightly more succinct. "The main step in beginning the analysis was collecting data," for instance, could be reduced to "The first step in the analysis was data collecting."
Minor and sporadic discrepancies in the use of articles (a/an/the) or prepositions are common difficulties for non-native speakers when writing in English. Still, none of the obvious mistakes that seriously hampered understanding surfaced during the reading.

Author Response

Reviewer 3

General Aspects

1. The introduction is constructed with the presentation of related articles. However, the study's importance section does not clarify its contribution to science. The study has new applicability in the local context, but in terms of innovation, it does not present a clear differential compared to other existing studies.

Answer: We appreciate the reviewer’s observation. In response, we have revised the Introduction to clarify the scientific contribution and novelty of our study. Specifically, we now emphasize that the study goes beyond local applicability by (i) integrating a diverse set of meteorological variables with hydrological outcomes, (ii) systematically comparing input parameter sets to identify the most effective climatic drivers, and (iii) evaluating the performance of two ANN training algorithms. These aspects demonstrate the methodological and applicative innovation of our approach in the context of ANN-based hydrological forecasting. The newly added paragraph is:

“Prior studies often relied on single-variable models such as precipitation or temperature alone. In contrast, this study systematically compares multiple combinations of meteorological inputs to identify the most predictive parameters using a robust feed-forward neural network architecture. The comparative evaluation of two ANN training algorithms (Levenberg-Marquardt and resilient backpropagation) provides a methodological contribution to improving hydrological forecasting performance. Thus, beyond its local applicability, the study contributes to the advancement of ANN-based modeling frameworks for climate-resilient water management. Thus, tackling the widespread global issue of climate-related factors that pose a threat to water security worldwide. By enhancing the accuracy of hydrological forecasting under climate variability, the proposed modeling framework can support decision-makers in developing robust water management plans and adapting operational strategies to ensure long-term water security.”

2. The manuscript should specify the proposed model's advantages compared to established evaporation and water balance models in the literature and detail its improvements over physical models. For instance, do existing approaches, such as Penman's reservoir balance combined with a hydrological model, not achieve similar outcomes?

Answer: We thank the reviewer for this important comment. A detailed literature review was conducted before selecting the proposed ANN-based modeling approach. The limitations of previously established models have been briefly mentioned in lines 66-72.

Evaporation or water balance models rely heavily on energy or mass balance principles and typically estimate evaporation or evapotranspiration from meteorological variables such as temperature, solar radiation, windspeed and/or humidity. While useful, these models primarily focus on the evaporative component of the hydrological cycle and often neglect other essential components such as direct inflows (e.g., tributaries) and controlled outflows (e.g., dam releases). Additionally, these models require extensive site-specific data, are sensitive to calibration, and often underperform in scenarios involving complex, nonlinear relationships.

In contrast, ANN model integrates both inflows and outflows alongside multiple meteorological inputs, offering a more holistic approach to hydrological modeling. This allows for a more accurate and adaptive prediction of water volume dynamics in reservoirs, particularly under changing climatic conditions. The ANN’s ability to handle nonlinearities, limited data, and multi-variable interactions makes it a more suitable and robust alternative for this study.

Methodology

3. The paper mentions multiple reservoirs within the basin. It should be clarified whether these reservoirs interfere with the flow rates at the Doğancı Dam and whether this factor was considered in the analysis.

Answer: We acknowledge the reviewer’s observation. The flowrate data used in this study represent the total inflow and outflow at the Doğancı Dam, as recorded by the local water authority. The data inherently reflect any upstream influences, including those from other reservoirs within the basin. However, the analysis focused solely on the net hydrological behaviour of the Doğancı Dam. Following sentence has been added to ‘Data Attainment’ section for more clarity:

“The inflow and outflow data used for Doğancı Dam were obtained as aggregated values, representing the total flow at the dam from the upstream reservoirs or other similar sources.”

4. Figure 1 can be improved. It should include the contributing watershed area and the locations of the meteorological stations from which data were obtained.

Answer: In response to this valuable comment, Figure 1 has been revised to improve its clarity and overall presentation. While we acknowledge the importance of indicating meteorological stations, the meteorological data used in this study was aggregated from all districts of the Bursa region, which are represented in Bursa’s map on the left.

5. ANN optimisation using Evolutionary Algorithms (EAs) and their variations has shown promising results compared to Non-Linear Programming models such as Levenberg-Marquardt (LM), which can converge to local optima. The authors should state why LM was employed instead of EAs to train the ANN.

Answer: We appreciate the reviewer’s thoughtful comment and agree that Evolutionary Algorithms (EAs) offer promising capabilities in optimizing neural network training. However, the Levenberg-Marquardt algorithm was selected in this study due to its computational efficiency, faster convergence, and proven performance in hydrological time series modeling. LM combines the advantages of both gradient descent and the Gauss-Newton method, making it well-suited for applications where training speed and accuracy are critical.

Moreover, this study also included Resilient Backpropagation as a secondary training algorithm to compare performance and mitigate potential convergence issues. The use of these two well-established algorithms allowed us to assess the influence of training mechanisms on prediction quality without significantly increasing model complexity or computational time.

Nonetheless, we recognize the value of EA-based optimization and will consider its application in future work. A sentence is added in conclusion to support this statement:

“Additionally, future studies may also explore some other established algorithms, such as evolutionary algorithms, for training and comparing ANN models.

6. The description of the neural network model (potentially sections 2.1 and 2.3 to 2.5 in the manuscript) can be summarised. Given that MATLAB and its standard systems were used in the implementation, standard references should be indicated.

Answer: We sincerely thank the reviewer for this helpful suggestion. In response, the content in Sections 2.1 and 2.3 to 2.5 has been condensed to retain only the most relevant methodological information regarding the ANN model structure, input combinations, training process, and evaluation metrics. Additionally, a standard reference for MATLAB has been included in the revised manuscript to acknowledge the software and tools used for model development and implementation.

7. The methodology employs monthly data. The authors should explain why daily data were not used, as daily data might capture hydrological dynamics with more detail.

Answer: We appreciate this important observation. The study relied on monthly data, as this was the form in which consistent and complete hydrological and meteorological records were made available by the relevant water authorities. Moreover, the objective of this research was to conduct a long-term trend analysis of hydrological parameters in relation to climate variables, rather than capturing short-term or event-based dynamics. Using monthly data provides a more stable and smoothed dataset, reducing the effects of noise, outliers, or extreme short-term fluctuations that could compromise model generalization. We have added a clarification to the manuscript accordingly as follows in the data attainment section:

“As mentioned, monthly data of the parameters was used in this study due to the completeness of long-term meteorological and hydrological records in this format. Additionally, the use of monthly data supports the long-term trend analysis presented in the study and helps to smooth short-term variability, making it suitable for water resource planning.”

 

8. Figure 4, which illustrates the ANN architectures for Model 1 (a), Model 2 (b), and Model 3 (c), shows minimal structural differences beyond the number of input neurons (five for Model 1, two for Model 2, and three for Model 3). The authors should consider whether presenting all three diagrams is necessary or if one representative diagram, noting the variations in input layer size, would be sufficient.

Answer: In response to the reviewer’s kind recommendation, Figure 4 has been revised to present a single representative neural network architecture. The updated figure indicates the variations in input layer size for Models 1, 2, and 3.

Results and Discussion

9. The presentation of results is confusing in parts. Units are missing on the axes in figures (e.g., Figures 5, 6, and 7). The methodology states that monthly data were used; however, input parameters for Model 1 are described with "daily" terms (e.g., "total daily global solar radiations," "monthly average total daily solar intensity," etc.). This discrepancy regarding the data time scale requires clarification.

Answer: We acknowledge this thoughtful observation. To clarify, the monthly average total daily global solar radiation refers to the average of the total daily radiation values recorded within a given month. This metric reflects the solar energy received during the month on average. All other meteorological and hydrological parameters used in the study are also based on monthly average values. This approach ensures consistency in the temporal resolution across all model inputs while preserving the seasonal variability essential for long-term hydrological trend analysis.

Additionally, the figures (5, 6, and 7) have been revised to include proper axis labels. Regarding the gradient, mean squared error, and correlation coefficient, these metrics are dimensionless (as the data was normalized), which is why no units were defined for the axes.  We trust these revisions improve the overall readability and presentation of the results.

10. Evaluating the model's performance against time series is fundamental for analysing flow, level, and hydrological balance. For the best-performing model, it is strongly recommended to present time-series plots of observed and calculated output variables (dam level and flow rates).

Answer: We appreciate the insightful suggestion regarding the inclusion of time-series plots for observed and predicted output variables. While we agree that time-series analysis is important for understanding temporal hydrological dynamics, the primary objective of this study was to evaluate the modeling performance using statistical indicators, specifically, error metrics and correlation coefficients, which provide a comprehensive measure of predictive accuracy across the dataset.

Including multiple time-series plots for each output variable may potentially overwhelm or distract the reader from the main statistical focus. Therefore, to maintain clarity and avoid visual overload, we have chosen to highlight aggregated performance measures instead.

11. If possible, the error metrics in Table 3 (comparison with previous studies) should be standardised to ensure a consistent comparison. Table 3 should be moved to the introduction or a dedicated literature review section, as its purpose is to contextualise the current study within existing research rather than to present direct results of this work.

Answer: We thank the reviewer for this valuable comment. We acknowledge the suggestion to reposition Table 3; however, we believe its current placement, following the results and model evaluation sections, is more appropriate in the context of this study. The purpose of Table 3 is to offer a comparative perspective on the performance of the proposed ANN models once their predictive outcomes have been fully presented. By situating the comparison after the modeling and evaluation results, the reader can more clearly assess the relative performance of our approach against established studies.

Furthermore, standardizing the error metrics across the studies presented in Table 3 is challenging, as each study involves different predicted outputs. Therefore, a generic overview has been provided to offer a broad yet informative context for the readers. The values cited for previous studies were reported directly from the respective publications and represent the performance across the whole datasets.

12. Figures 5a and 5b, which depict results for Model 1, show data points tightly clustered along the Y = T line due to high correlation coefficients (R values of approximately 0.999). This tight clustering, particularly if values are scaled or originate near zero, might be misinterpreted (e.g., as a single point at (0,0) by the user). The authors should ensure the clarity of these plots to avoid potential misinterpretation of the data distribution shown.

Answer: The comment is well-appreciated. We acknowledge that the visual clustering of points tightly along the Y = T line may raise concerns about the clarity and interpretability of the plot, particularly when the dataset is normalized and model performance is high.

To clarify, the scatter plots are based on normalized data which is a standard practice in training artificial neural networks. This step is essential for ensuring convergence and stability during training. As a result, the observed and predicted values are normally compressed into a narrow range. Given the high correlation coefficient and the minimal error achieved by Model 1, the predicted values closely mirrored the observed values. This leads to points being plotted directly on top of one another along the Y = T line.

The visual effect may give the appearance of a condensed point cloud or reduced data spread, particularly near the origin, despite the underlying dataset containing a full range of values. This should not be interpreted as data sparsity, lack of variability, or convergence to a singularity such as (0,0). Rather, it is a visual consequence of both normalization and the high predictive performance of the ANN model. A data normalization step has been added in ‘ANN Application’ section for more clarity as follows:

“Prior to model training, the data were normalized using the following equation:

  Equation 5

Where, D represents a dimensionless normalized value. In contrast, Di is the normalized value for the ith measurement in the data, and Dmax and Dmin are the maximum and minimum normalized scores of all the training and testing data taken.”

Comments on the Quality of English Language

Sometimes sentences sound a bit literal or could be rearranged for a more idiomatic English flow. For example, it is reasonable to say "Climate change is one of the significant environmental issues at the present time"; but, "Climate change is currently one of the most significant environmental issues" would sound somewhat more natural. "To the author's knowledge, this form of study is novel in the local context, making it unique and equally important from a water management point of view," says another example. The term "equally important" begs the issue "equally important to what?". Maybe more direct would be "also important" or "particularly important". These are not major mistakes endangering knowledge; polishing would help.

Concision: The book might occasionally be just slightly more succinct. "The main step in beginning the analysis was collecting data," for instance, could be reduced to "The first step in the analysis was data collecting."

Minor and sporadic discrepancies in the use of articles (a/an/the) or prepositions are common difficulties for non-native speakers when writing in English. Still, none of the obvious mistakes that seriously hampered understanding surfaced during the reading.

Answer: We sincerely appreciate the reviewer’s constructive feedback regarding the language and phrasing used in the manuscript. As non-native English speakers, we acknowledge that certain expressions may have sounded overly literal or less idiomatic. We re-read the manuscript and have revised a few phrases and sentence structures to the best of our understanding.

 

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

All the comments have been well solved, and I have no further comments