Runoff Evolution Characteristics and Predictive Analysis of Chushandian Reservoir
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsPage 1: Title and Abstract
The abstract is too dense and technical for general readers. Terms like “quasi-2.32-year periodicity” and “cross-wavelet transform” need simplification or brief clarification. Consider simplifying and highlighting the novelty of the study in lay terms. The impact of the findings (e.g., on reservoir operation or regional planning) is not emphasized. No numerical performance metrics for the models are presented. Add a short sentence at the end of the abstract highlighting practical significance. Mention model accuracy (e.g., NSE scores) briefly.
Page 2: Introduction
The novelty of the work is not clearly identified. Several general references are used; it needs more recent or specific citations for runoff modeling. A literature gap is implied but not directly articulated. Specify what previous studies lack—e.g., “Few studies integrate ESMD and CNN-LSTM to explore reservoir runoff dynamics…” State clearly what gap the paper addresses (e.g., "Previous studies did not integrate ESMD with CNN-LSTM to study long-term trends in reservoir runoff"). Repetition about human-climate interaction should be avoided.
Page 3: Study Area and ESMD Method
No information about data quality control or potential gaps in the 58-year record is given. The description of ESMD is overly detailed and technical for a general hydrology audience. The author can add a table or figure summarizing the data (e.g., missing years, mean values). Also, dense equations and algorithmic steps in the supplementary material can be shifted.
Page 4–5: Heuristic Segmentation and Cross-Wavelet Analysis
No discussion of model assumptions (e.g., stationarity, noise level). The choice of teleconnection indices is not well justified. Explain why these indices (e.g., AO, NAO, DMI) were selected. Clarify how the significance threshold (0.65) in mutation detection was determined.
Page 6–7: Prediction Models (GM-BP, CNN-LSTM)
Missing hyperparameter details: number of layers, neurons, activation functions, training epochs, data normalization methods. No mention of how overfitting was controlled or whether a validation set was used. A table listing model configurations. Include performance metrics on both training and test datasets.
While the reservoir’s role is explained, more information on the data's spatial resolution and quality control measures is needed. Key equations should be better formatted or referenced to supplementary materials. Lacks hyperparameter details (e.g., number of layers, epochs, activation functions). Absent from mentioning overfitting controls (e.g., dropout, validation loss).
Page 8–9: ESMD Results and Periodicity
The correlation coefficients in Table 1 are relatively weak beyond IMF2; the implications of this are not discussed. The trend in Figure 6 is interpreted without a statistical test for trend significance.
Page 10–11: Mutation and Cross-Wavelet Results
Figures are too dense; interpretive clarity is lost. No linkage to land-use or management changes around 2010 when the mutation occurred. Add a table summarizing key resonance periods across all indices. Provide possible physical reasons for the 2010 abrupt change.
Page 12: Prediction Results
CNN-LSTM performs worse than GM-BP, contrary to expectations. This is not critically analyzed. Figure 10 could be enhanced with clearer legend and confidence bands. Explore reasons for the underperformance of CNN-LSTM (e.g., model tuning, data scarcity). Discuss applicability to real-time forecasting scenarios.
Page 13–14: Discussion
The discussion is descriptive rather than analytical. There is little engagement with uncertainties or limitations. Over-reliance on earlier studies for comparison.The manuscript needs deeper exploration of why CNN-LSTM underperformed? Also, more critical evaluation of climate-reservoir causality is needed. Discuss land-use changes, upstream developments, or specific management practices that might explain the mutation point in 2010.
Page 15: Conclusion
No mention of model limitations or recommendations for operational water managers. Add a sentence or two about model transferability, future improvements, or the need for real-time forecasting data. No practical recommendations or future directions provided. References lack consistency in formatting, and some are outdated (e.g., ref. [14] from 2002).
Comments on the Quality of English LanguageEnsure consistency in variable notation (e.g., X vs x(t); IMF1 vs IMF-1).
Reference formatting is inconsistent (some authors use full names, others initials).
Author Response
Please find the comments as attached.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript deal with the problem of runoff characteristics and predictive analysis. It separately compares 3 methodologies, 2 deep learning models, and evaluates the possible impact of 9 teleconnection factors on runoff. The manuscript has a standard structure with relevant references.
General comments
- Clearly present abbreviations and their meaning from the beginning of the introduction section.
- The predictive analysis is mentioned in the title of the manuscript. I expected some prediction into the future; however, proposed only test of the model was done without any clear prediction to future.
Specific comments
- data description – source of data is missing
- statistical analysis – missing information in the material and methods section, which and how the tools were used
- results – Periodic and trend characteristics, also in table 1 and figure 6 – which coefficient is R
- information in the conclusion doubt the abstract; some further work needs should be presented in the conclusion
- some international references would be nice to add – it looks like only Chinese authors are referenced
- References 10, 35, 43 and 45 are the same. Also, 38 and 39
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsAbstract
The abstract, while informative, is dense and lacks a clear breakdown of methods vs. results. It would benefit from clearer structure (e.g., objectives, methods, results, implications).
Introduction
Overuse of citations within long paragraphs—makes it hard to follow. The research gap is not distinctly separated from literature review. Limited discussion of data limitations or the spatial heterogeneity of the basin.
Methodology
Excessive algorithmic details may overwhelm readers not focused on computational hydrology. The rationale for selecting certain parameters (e.g., threshold in segmentation, epochs in training) is not sufficiently justified beyond empirical choices.
Results
The results section tends to repeat content from the methods and discussion. Figures (e.g., wavelet coherence maps) are not visually discussed in detail. Weak explanation of why 2010 marked a runoff mutation year beyond linking to drought.
Predictive method
The training data span (1960–2017) may include nonstationary trends—this issue is acknowledged but not addressed through methods like data detrending or ensemble modeling. No cross-validation or uncertainty quantification.
Conclusion
Some points reiterate previous sections without adding strategic conclusions for practitioners.
To enhance the manuscript:
Clarify the research gap and practical contributions. Simplify and reorganize sections for readability. Address model uncertainty more robustly. Enrich discussion with implications for reservoir management and climate adaptation.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsI appreciate the work the authors did during the review of the paper. Most of the comments have been incorporated and I therefore recommend the paper for publication.
Author Response
I sincerely appreciate your time and effort in reviewing this manuscript. Your rigorous and insightful comments, as well as your commitment to academic excellence, have been invaluable in improving this work. I are deeply grateful for your dedication and the thoughtful feedback you have provided. Thank you for your support and contributions.
