The manuscript addresses an important problem in hydrology and water resources management. The integration of ensemble rainfall forecasts with the Switch Prediction Method (SPM) and machine learning techniques is relevant, and the ShihMen Reservoir case study provides a solid application context. The methodological framework is clearly presented, and the paper is generally well organized. These are important strengths of the study.

Nevertheless, there are several areas where improvements would help to increase the clarity, novelty, and broader impact of the work:

1. Research novelty not sufficiently highlighted

Although the Abstract and Introduction provide a comprehensive description of the study background and methodology, the novelty of the proposed approach compared with existing methods is not clearly articulated. I recommend that the authors strengthen the concluding part of both sections by explicitly stating the uniqueness of this study and clarifying how the proposed framework improves upon conventional approaches. This will better highlight the scientific contribution.

2. Outdated references

The list of references added by the authors appears to rely heavily on older studies. A scientific article should also reflect the state-of-the-art. It is recommended to include more recent literature and to provide a review of previous research on similar issues to enhance the timeliness and relevance of the study.

3. Insufficient technical details of SPM

The Switch Prediction Method (SPM) is only briefly described, making it difficult for readers to fully understand its operational mechanism. More details are needed. This will also help ensure reproducibility.

4. Limited discussion on practical implications

The Discussion and Conclusion are focused mainly on the technical aspects of the model. However, the practical significance for reservoir operation remains underdeveloped. It is suggested to elaborate on how 72-hour inflow forecasts—even with certain errors—can still provide critical lead time for reservoir managers to adjust release strategies and enhance operational safety. This would substantially strengthen the applied value of the study.

5. Future research directions too narrowly technical

The future work outlined in the Conclusion mainly emphasizes model improvement. To broaden the impact, the authors should consider including wider perspectives such as cross-basin applicability, feasibility under climate change conditions, or hybridization with physically-based models. This would enhance the forward-looking dimension of the paper and its relevance to a broader research community.

Overall comments

The manuscript presents an interesting and practical framework for long-term reservoir inflow forecasting by integrating the Switch Prediction Method (SPM) with machine learning models. The topic is relevant, especially for typhoon-prone regions such as Taiwan, and the study shows potential contribution to operational hydrology.

However, several issues related to clarity, methodological consistency, and presentation should be addressed before publication.

Abstract

The abstract is informative; however, it would benefit from a more concise and consistent presentation of results. It is recommended to report only the best-performing model with representative numerical values to improve clarity.

Introduction

The objective of the study is generally clear; however, the novelty is not sufficiently emphasized. Please revise the objective section to clearly state:

-what is new in this study

-how it differs from existing ensemble or ML-based approaches

It may be helpful to explicitly highlight the contribution of integrating SPM with MSF and machine learning models.

Methodology

The manuscript would benefit from improved reproducibility. It is recommended to include a pseudo-code or algorithmic flow for the proposed framework (SPM + MSF + ML models).

Results and Discussion

The results are well presented; however, the discussion lacks comparison with existing studies. It is recommended to include comparison statements with previous research to better position the contribution of this work.

The manuscript does not explicitly discuss the advantages and limitations of the proposed framework.

It is recommended to add a dedicated discussion including:

-strengths (e.g., adaptability, uncertainty reduction)

-limitations (e.g., data dependency, lack of extrapolation capability)

Conclusion and Future Work

The conclusion summarizes the findings well; however, future research directions are only briefly mentioned.

Please expand the future scope section, including:

-application to other watersheds

-real-time forecasting

Figures and Tables

Please ensure that all figures are properly cited and clearly indicate whether they are original or adapted from other sources.

References

Please carefully review all references to ensure they are relevant and closely related to the study.

The use of non-peer-reviewed sources (e.g., websites) is not recommended and should be avoided.

Comments:

This manuscript proposes an integrated framework combining the Switch Prediction Method (SPM), ensemble rainfall forecasts, and machine learning models (SVM and LSTM) for 72-hour reservoir inflow forecasting during typhoon events. The topic is relevant to reservoir operation and flood management, and the manuscript is generally well organized.

However, the reviewer finds that the inclusion of only two machine learning models does not provide a sufficiently comprehensive evaluation of the proposed framework. Given the significant developments in deep learning and transformer-based forecasting methods in recent years, the authors should better position their work within the current state of the art and clarify the advantages, limitations, and potential competitiveness of the proposed framework relative to more advanced architectures.

Line 121: The study was conducted using only 18 instances of typhoons from one reservoir catchment. Although the forecasts generated were impressive, it is challenging to determine how effective the developed framework would be if applied in a different hydrologic environment. It is essential for the authors to highlight the shortcomings of the available data and present more supportive evidence if possible.

Lines 263-267: Describe the use of grid search for parameter optimization. However, the manuscript does not provide sufficient details regarding the search ranges, validation procedure, or total number of parameter combinations evaluated. Without such information, the study is difficult to reproduce. The reviewer recommends expanding the methodological description.