Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Due to steep gradients, short channel lengths, and limited storage capacity, small streams exhibit very rapid and nonlinear flood responses, making real-time prediction difficult. To overcome these limitations, this study proposes a measurement-based real-time flood forecasting framework that integrates high-frequency observation data from SMMS and AWS, short-term rainfall forecasts from MAPLE, and rainfall-discharge and water depth-discharge relationships based on 4-parameter logistic (4PL) regression. Furthermore, to extend the prediction range beyond gauging stations to ungauged sections, this study utilized both HEC-RAS scenario-based depth-discharge curves and a Manning’s formula-based approach, achieving an average discharge prediction accuracy of approximately 89% and a depth prediction accuracy of approximately 90% across 12 small streams. However, the analysis of the factors causing differences in prediction performance among streams is somewhat limited, and the criteria for applying the Manning equation and the HEC-RAS-based method to ungauged reaches should be more clearly presented. In addition, further discussion is needed regarding the treatment of tide-affected reaches and the applicability of the framework to other regions.

1. This study holds significant academic and practical value in that it derives 4PL-based rainfall-discharge nomograms and water depth-discharge calibration curves from long-term observational data, thereby reducing the high computational burden and difficulties associated with real-time application inherent in existing numerical hydrological models. In particular, the proposal of a simple yet rapid, observation-based forecasting system for basins with short flood arrival times demonstrates high practical applicability for the development of early warning systems. Furthermore, the platform-based forecasting structure integrating SMMS and AWS is evaluated as a study that effectively balances field applicability and operational feasibility.

2. The validation results for the 12 small streams showed an average flow prediction accuracy of 89.08% and an average water depth prediction accuracy of 89.93%, with the RMSE also generally at a satisfactory level. However, performance varies by stream, and at some locations, accuracy remains in the 70% range. Therefore, it is believed that the explanatory power of the research results would be further enhanced if a more detailed analysis were conducted to determine which factors, such as watershed area and riverbed slope, caused these differences in prediction performance.

3. The decision to use both the HEC-RAS scenario-based rating curve and the Manning equation-based method in parallel for the extension of ungauged sections is a practical approach. However, as the Manning equation calculates each section independently, it does not fully account for upstream-downstream interactions, drainage effects, the influence of hydraulic structures, or turbulent flow. Consequently, the paper acknowledges the possibility that water depths in steep-gradient small streams may be underestimated or estimated discontinuously. Accordingly, it would be beneficial to establish clearer criteria regarding which method (the Manning equation or the HEC-RAS-based method) should be prioritized under what conditions when predicting ungauged sections.

4. The study found that water depths calculated using HEC-RAS were slightly lower than those obtained using the Manning formula in the Songnam and Jungdong rivers, which are located near the coast. One of the reasons cited for this discrepancy is the application of steady-flow point-line flow analysis rather than unsteady-flow analysis to sections affected by tides. Therefore, future research should explore the possibility of incorporating unsteady-flow analysis for sections subject to tidal influences.

5. This study is highly valuable in that it developed a model using long-term observational data from 12 small streams in Korea and demonstrated scalability by considering both monitored and unmonitored sites. However, since the proposed framework fundamentally relies on sufficient observational data and pre-established relationships, further validation is needed to determine whether it can achieve the same performance when applied to newly established sites, small streams with short data accumulation periods, or overseas regions with different topographic and climatic characteristics. It is anticipated that the study’s scalability will be further enhanced once it is supplemented through such additional validation.

I've reviewed the paper and found some corrections, so I think the paper needs minor revisions.

 

Author Response

Response to Reviewer 1 Comments

 

We sincerely thank the reviewer for the constructive and thoughtful comments, which have significantly improved the quality of our manuscript. All comments have been carefully considered, and the manuscript has been revised accordingly. Our detailed responses are provided below.

 

Comment1: Limited analysis of performance differences among streams

 

Response 1: The discussion has been expanded to better interpret variability in prediction performance among streams.

 

Location: Section 5.1 and Section 6 (Conclusions)

Revision: Added discussion indicating that lower prediction performance may be associated with steeper channel slopes, irregular channel geometry, and uncertainty in rainfall and hydraulic parameters. The revised manuscript now provides a more cautious interpretation of performance variability among streams, while a more detailed quantitative analysis of these relationships is suggested for future work.

 

Comment2: Criteria for applying Manning vs HEC–RAS methods

 

Response 2: Selection criteria have been clarified to improve practical applicability.

 

Location: Section 4.2, Section 5.2

Revision: Defined that the Manning-based method is suitable for rapid preliminary estimation in reaches with relatively uniform geometry and low hydraulic complexity. The HEC–RAS-based approach is recommended for reaches with complex hydraulic conditions, such as abrupt slope changes or irregular channel sections. Clarified that the Manning-based method is appropriate for real-time screening, whereas the HEC–RAS-based approach provides more reliable depth estimation in critical reaches.

 

Comment3: Treatment of tide-affected reaches

 

Response 3: Tidal effects and associated limitations have been addressed.

 

Location: Section 5.2

Revision: Added discussion on coastal streams (e.g., Songnam and Jungdong), where discrepancies between Manning-based and HEC–RAS-based predictions were observed. Clarified that these differences are likely due to steady-flow assumptions in HEC–RAS, while the reaches are influenced by tidal boundary conditions. Stated that steady-flow analysis may underestimate water depth in such environments and suggested incorporating unsteady-flow simulations in future studies.

 

 

Comment4: Applicability to other regions

 

Response 4: Applicability limitations have been clarified.

 

Location: Section 6 (Conclusions)

Revision: Added statement that, although the framework shows strong performance in twelve Korean small streams, its applicability to other regions requires further validation. Clarified that the framework depends on long-term observational data and may have limited performance in newly monitored or data-scarce basins. Emphasized the need for site-specific calibration using local observational data.

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Editor, I have reviewed the manuscript entitled "Measurement-Based Framework for Real-Time Flood Prediction in Small Streams Using Rainfall–Discharge Nomographs and Depth–Discharge Rating Curves". The manuscript is quite well-structured and, in terms of its scope, sample size, and methods, it is a valuable contribution to the literature. However, I believe it could be accepted with some major corrections in the methods and application sections, as well as minor stylistic adjustments. My suggestions are:

1- It is recommended that the abstract include an explanation of "conventional approaches" or details of which approaches are being used. As it stands, it is too general and weakens the originality of the work.

2- The weakest part of the manuscript is the literature review and the lack of reference to previous studies. For example, the introduction section should be improved by providing a detailed explanation of modeling tools such as HEC-RAS and Manning values, which are important parameters. It would be crucial to explain the advantages of the methods and where they are applied, thus informing readers about why these methods are used. The following studies could be reviewed for this purpose. 1- https://doi.org/10.3390/su16031226 2- DOI: https://doi.org/10.15233/gfz.2020.37.9

3- Do the numerical expressions in lines 94-104 rely on a source? Providing supporting evidence or citations for this section is recommended.

4- How was the S value obtained in Table 1? Was it arithmetic, harmonic, or with a GIS system? Was this slope value used in the modeling?

5- Figure 1 is quite successful, but providing a figure showing the country from a broader perspective (where is Wonju?) after the study area, or adding a small figure integrated into this figure, would help in understanding the study area. Also, the explanations in lines 127 and 128 are already available and can be deleted.

 6- In the Method heading, regarding the HEC-RAS modeling approach, were single values ​​entered for the model limits, such as the Manning value (although the equation text is included in Figure 2, it is recommended to provide this equation) for the left, right, and midsections? Currently, it appears that the modeling is done in one dimension, but details are not provided. It is suggested that the values ​​used for modeling and the workflow be added.

7- While the term "estimates" is used in Line 239, "Prediction" is used in the title. It is recommended to use a single term in the article.

8- How was the Manning value determined in Line 281? What kind of modeling was done in HEC-RAS? Were buildings and bridges defined? What was the base flow used? What is the resolution and year?

9- Can the feature expressed in Line 288 be applied to other rivers? It is recommended to address this in the discussion section.

10- Did the authors perform or recommend a calibration of their modeling, or is there a citation? I particularly recommend that future studies, even if not this one, utilize FLO2D, etc. The calibration of the results with a second model using the tools is recommended.

11- It is suggested that lines 297-303 be moved to the methods section.

12- Is there a visual or methodology for "field survey" in line 327? The Manning value is one of the most important parameters of modeling, and it is recommended that the authors examine this section in detail and provide it to the reader. The articles suggested in other sections can be reviewed to improve this section.

13- Consider providing Figure 5 on a single page and in high resolution, especially since the text is unreadable in the output format.

14- By adding a heading or paragraph titled "Evaluation" at the end of the methods section, the R² equation and the other error term equation RMSE can be included, thus making the findings understandable from the perspective of the error term, which might otherwise be misleading (Table 4).

15- Was the LM optimization performed in an ANN environment?

16- It is recommended that information about "flow simulations" be provided in line 520.

17- In line 529, as far as I understand, were the manning values ​​defined for HEC-RAS in 1D modeling taken as a single value? Are the right and left slopes the same as the river?

18- It is suggested that line 596 be rewritten and a different conjunction be used instead of the word "Overall". Also, if an AI writing tool like GPT is used, it is suggested to add this to the acknowledgments section. 

BEst

Author Response

Response to Reviewer 2 Comments

 

We sincerely thank the reviewer for the detailed and constructive comments, which have been carefully addressed through corresponding revisions to the manuscript, as detailed below.

 

Comment1: Abstract clarification

 

Response 1: The abstract has been revised to provide a clearer description of conventional approaches, including physics-based hydrodynamic models and data-driven methods. Their limitations in real-time applications are now more explicitly stated to better highlight the motivation and contribution of the proposed framework.

Location: Abstract

Revision: Clarified the limitations of conventional approaches and emphasized the integration of observation-based regression relationships with real-time monitoring data.

 

Comment2: Literature review improvement

 

Response 2: The Introduction has been expanded to include a more detailed discussion of hydraulic modeling approaches, including HEC–RAS and the Manning equation, together with their advantages and limitations. This revision improves the clarity of the methodological background and the motivation for the proposed framework.

Location: Section 1 (Introduction)

Revision: Added explanation of hydraulic modeling tools and their limitations in small-stream environments.

 

Comment3: Source for numerical expressions

 

Response 3: Supporting references have been added to clarify the source of the numerical expressions.

 

Location: Section 2.1 (relevant lines)

Revision: Added citation to National Disaster Management Research Institute (2025) to clarify the source of numerical expressions.

 

Comment4: Slope (S) calculation

 

Response 4: The derivation of channel slope has been clarified.

 

Location: Section 2.1 (Table 1 description)

Revision: Specified that channel slope was derived from surveyed coordinate data in the Comprehensive Small Stream Maintenance Plan reports and used as input in both Manning-based calculations and HEC–RAS simulations.

 

Comment5: Figure 1 improvement

 

Response 5: Figure 1 has been improved.

 

Location: Figure 1

Revision: Added a legend and improved spatial interpretation. Redundant explanatory text has been removed.

 

Comment6: HEC–RAS modelling details

 

Response 6: Additional explanation has been provided regarding the HEC–RAS-based scenario simulations.

Location: Section 4.2

Revision: Clarified the use of cross-sectional data, roughness coefficients, and upstream/downstream boundary conditions in the scenario-based simulations and explained that this approach was adopted to support efficient prediction in small streams.

 

Comment7: Terminology consistency

 

Response 7: Terminology has been revised to improve consistency.

 

Location: Entire manuscript

Revision: The use of “prediction” has been emphasized and inconsistencies have been reduced throughout the manuscript.

 

Comment8: Manning coefficient and modelling details

 

Response 8: Additional explanation has been included.

 

Location: Section 4.2 and Section 5.2

Revision: Clarified that roughness coefficients were determined based on field survey data and that simplified hydraulic representations were adopted in the modelling framework.

 

Comment9: Applicability to other rivers (streams)

 

Response 9: The applicability of the framework has been further clarified.

 

Location: Section 6 (Conclusions)

Revision: Stated that the framework is primarily applicable to small streams with similar characteristics and requires site-specific calibration for other regions.

 

Comment10: Calibration

 

Response 10: Calibration methodology has been clarified.

 

Location: Section 4.1

Revision: Stated that the nonlinear regression process using the Levenberg–Marquardt (LM) algorithm functions as a calibration procedure based on observed data.

 

Comment11: Section reorganization

 

Response 11: Content has been reorganized.

 

Location: Section 4 (Methods)

Revision: The methodological description has been reorganized to improve logical flow.

 

Comment12: Field survey clarification

 

Response 12: Field survey procedures have been expanded.

 

Location: Section 2.1

Revision: Added description of survey data, including cross-sectional measurements, channel geometry, and roughness estimation.

 

Comment13: Figure 5 readability

 

Response 13: Figure quality has been improved.

 

Location: Figure 5

Revision: Improved layout and resolution for better readability.

 

Comment14: Evaluation section

 

Response 14: We thank the reviewer for this valuable suggestion. In the revised manuscript, the performance evaluation has been clarified by explicitly presenting the coefficient of determination (R²) for the 4PL regression relationships and the RMSE for discharge and flood depth prediction. Although a separate subsection titled “Model Evaluation” was not added, the relevant evaluation criteria and their interpretation were clarified in the main text to improve the transparency of model assessment.

 

Location: Section 4.1 and Section 5.1

Revision:

• Clarified the goodness-of-fit of the rainfall–discharge nomographs and depth–discharge rating curves using R² in Section 4.1 and Table 3.
• Clarified prediction performance using RMSE and discrepancy-based accuracy in Section 5.1 and Table 5.
• Added explanation that the discrepancy-based accuracy metric differs from standard hydrological indicators and should be interpreted accordingly.

 

Comment15: LM optimization

 

Response 15: Methodological clarification has been added.

 

Location: Section 4.1

Revision: Clarified that the LM method was applied as a nonlinear least-squares optimization approach for parameter estimation.

 

Comment16: Flow simulation explanation

 

Response 16: Additional explanation has been included.

 

Location: Section 4.2

Revision: Clarified that scenario-based flow simulations were conducted to derive depth–discharge relationships under varying discharge conditions.

 

Comment17: Manning coefficient application

 

Response 17: The treatment of roughness coefficients has been clarified.

 

Location: Section 5.2

Revision: Clarified that representative roughness coefficients were used and that simplified assumptions were applied in the modelling.

 

Comment18: Stylistic revision and AI usage

 

Response 18: The manuscript has been revised for clarity and readability only.

 

Location: Section 6 (Conclusions)

Revision: Improved sentence structure and removed ambiguous expressions.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript entitled “Measurement-Based Framework for Real-Time Flood Prediction in Small Streams Using Rainfall–Discharge Nomographs and Depth–Discharge Rating Curves” presents a measurement-driven framework integrating real-time monitoring, rainfall nowcasting, and nonlinear regression models for flood prediction in small-stream environments. The study addresses an important and timely problem, particularly given the increasing vulnerability of small basins to flash floods. The manuscript is generally well-structured and demonstrates practical relevance; however, several methodological, conceptual, and presentation-related issues require substantial clarification and improvement before the manuscript can be considered for publication. Major revisions are required. The detailed comments are provided below.

1. The manuscript claims novelty in integrating measurement-based approaches with regression-based hydraulic relationships. However, similar frameworks combining rating curves, rainfall-runoff relationships, and real-time monitoring have been widely reported. The authors should clearly differentiate their contribution from existing studies, especially recent AI/data-driven flood prediction frameworks. At present, the novelty appears incremental rather than substantial.
2. The use of the four-parameter logistic (4PL) regression model is a central component of the proposed framework; however, its selection is not adequately justified. The manuscript does not explain why the 4PL model was preferred over other commonly used nonlinear formulations, such as power-law, exponential relationships, or more advanced machine learning approaches. Moreover, no comparative analysis with alternative models is presented. Without such benchmarking, it is difficult to assess whether the 4PL model provides superior performance or is simply one of many possible fitting options. Therefore, the claimed effectiveness of the model remains insufficiently supported.
3. The reported coefficients of determination (approximately 0.99 for all sites; see Table 3, page 11) are exceptionally high and raise concerns regarding potential overfitting, particularly given that the models are calibrated using only peak values. Such near-perfect fits are uncommon in hydrological applications and may indicate that the model is overly tailored to the calibration dataset rather than capturing generalizable relationships. Furthermore, the manuscript does not include any form of cross-validation, independent testing, or robustness analysis to assess model reliability. To ensure the credibility of the results, it is essential that the authors validate the model using unseen data or apply techniques such as k-fold cross-validation.
4. The methodology relies exclusively on peak rainfall–discharge and discharge–depth pairs, which represents a significant simplification of the underlying hydrological processes. By focusing only on peak values, the approach neglects the full hydrograph dynamics, including the rising and falling limbs that are critical for understanding flood evolution and timing. As a result, the temporal behavior of flood events is not explicitly modeled, limiting the framework’s ability to capture continuous flow variations. This simplification may reduce the applicability of the proposed method for real-time, continuous forecasting and operational decision-making.
5. The assumption that discharge remains constant along stream reaches (page 10) is unrealistic in many hydrological contexts, as it neglects important processes such as tributary inflows, lateral runoff contributions, and potential transmission losses. This simplification may introduce significant bias in the estimation of flood depth, particularly in heterogeneous or larger catchments. The authors should provide a clear justification for this assumption based on watershed characteristics or support it with a sensitivity analysis to quantify its impact on the results.
6. The framework is calibrated using data from only 12 small streams in Korea, which raises concerns about its broader applicability. The study does not sufficiently address the variability in climatic, geomorphological, and hydrological conditions that may exist in other regions. Furthermore, no transferability or external validation test is provided. The authors should more critically discuss the limitations of applying this framework beyond the study area and, if possible, include validation in different hydrological settings.
7. The performance of the proposed model is primarily assessed using accuracy (based on a narrow discrepancy interval) and RMSE, which are not sufficient for comprehensive hydrological model evaluation. Standard and widely accepted metrics such as Nash–Sutcliffe Efficiency (NSE), Kling–Gupta Efficiency (KGE), and Mean Absolute Error (MAE) should be included to provide a more robust and comparable assessment of model performance.
8. The definition of accuracy as the proportion of predictions within a very narrow error band (±0.01) (page 17) is unconventional and potentially misleading. This metric may artificially inflate the perceived performance of the model and is not directly comparable with metrics used in other studies. The authors are encouraged to adopt standard evaluation criteria commonly used in hydrology to ensure transparency and comparability.
9. Although the manuscript discusses errors in rainfall forecasts (Figure 7, page 15), it does not quantify how these uncertainties propagate into discharge and flood depth predictions. This omission is critical, as uncertainty in rainfall input is a major source of error in flood forecasting systems. The authors should include an uncertainty propagation or sensitivity analysis to evaluate the impact of rainfall forecast errors on model outputs.
10. The manuscript does not include any comparison between the proposed framework and existing approaches, such as physically based hydrodynamic models, machine learning models, or hybrid techniques. Without such benchmarking, it is difficult to assess the relative performance and advantages of the proposed method. A comparative analysis is necessary to demonstrate its effectiveness and practical value.
11. The study mentions an “AI-assisted flood prediction module” (page 6), but does not provide sufficient details regarding its structure, algorithms, or implementation. As presented, this component appears conceptual rather than operational. The authors should either provide a clear and detailed description of the AI methodology or reconsider including this claim if it is not a substantive part of the framework.
12. The manuscript states that the datasets used in the study are not publicly available, which limits the reproducibility and transparency of the research. To improve scientific rigor, the authors should consider providing at least partial datasets, synthetic data, or detailed guidelines that would allow other researchers to replicate the study.
13. Several figures (e.g., Figures 8–12) suffer from poor readability due to small size, unclear legends, and insufficient labeling of axes and units. This reduces the clarity and interpretability of the results. The authors should improve the quality and presentation of all figures to ensure that they are self-explanatory and accessible to readers.
14. While the manuscript is generally understandable, it contains repetitive statements, overly long and complex sentences, and minor grammatical inconsistencies. These issues affect readability and clarity. A thorough language revision by a professional editor is recommended.
15. The discussion section does not provide sufficient depth or critical analysis. It lacks meaningful comparison with previous studies, does not adequately explain discrepancies among different streams, and only superficially addresses the limitations of the proposed approach. A more comprehensive and critical discussion is needed to strengthen the manuscript.
16. Although the framework is presented as operational, key aspects such as computational efficiency, real-time implementation, and system integration are not clearly demonstrated. The authors should provide quantitative information on computational time and practical deployment to support claims of real-time applicability.
17. The manuscript does not include a sensitivity analysis of key parameters, such as the roughness coefficient or the 4PL model parameters. This is an important omission, as parameter uncertainty can significantly affect model reliability. Including a sensitivity analysis would strengthen confidence in the results.
18. Certain terms, such as “nomograph,” are used without sufficient explanation, which may limit accessibility for an international audience. The authors should provide clear definitions and context for specialized terminology.
19. Although some limitations are briefly mentioned, they are not critically analyzed. A more transparent and detailed discussion of uncertainties and methodological limitations is necessary to provide a balanced assessment of the framework.
20. The conclusions present the framework as highly effective and broadly applicable; however, these claims are not fully supported by the analyses conducted. The authors should moderate their conclusions to better reflect the limitations and scope of the study.

Author Response

Response to Reviewer 3 Comments

 

We sincerely thank the reviewer for the thorough and insightful evaluation of our manuscript and for recognizing its practical relevance and importance; all comments have been carefully addressed, and the manuscript has been substantially revised to improve clarity, robustness, and scientific rigor, as detailed below.

 

Comment1: Novelty is unclear

 

Response 1: We agree that individual components such as rating curves and rainfall–runoff relationships have been widely studied. In the revised manuscript, we have clarified that the main contribution of this study lies in the integration of high-frequency monitoring data, rainfall nowcasting, and nonlinear regression-based hydraulic relationships into a unified operational framework specifically designed for small streams with rapid hydrological response.

 

Location: Section 1 (Introduction) and Section 5 (Discussion)

Revision: Clarified the distinction between the proposed framework and conventional hydrodynamic and data-driven approaches, emphasizing the integrated and operational nature of the system.

 

Comment2: Justification of 4PL model

 

Response 2: We agree that additional justification was necessary. The revised manuscript includes an expanded explanation of the suitability of the 4PL model for representing nonlinear asymptotic relationships. We also included a comparison with alternative formulations (e.g., power-law and exponential relationships), indicating that the 4PL model provides stable parameter behaviour and consistently high goodness-of-fit for the observed datasets.

 

Location: Section 4.1

Revision: Added explanation of model characteristics and comparison with alternative nonlinear regression forms.

 

Comment 3–4: Overfitting and lack of validation

 

Response 3: We acknowledge the reviewer’s concern regarding potential overfitting. In the revised manuscript, the dataset was divided into calibration and validation subsets, and model performance was evaluated using independent events. The results indicate that the regression relationships maintain stable predictive performance across the analysed datasets.

 

Location: Section 4.1

Revision: Added description of calibration and validation procedure using independent events.

 

Comment 5: Peak-only limitation

 

Response 5: We agree that the use of peak values is a simplification. This limitation has been explicitly stated in the revised manuscript. The framework focuses on peak relationships and does not represent the full hydrograph dynamics, which may limit its applicability for continuous flow simulation.

 

Location: Section 4.1 and Section 6 (Conclusions)

Revision: Added explicit statement on the limitation of peak-based relationships.

 

Comment 6: Constant discharge assumption

 

Response 6: We agree that this assumption represents a simplification. The revised manuscript clarifies that this assumption may be reasonable for small streams with limited tributary inflow but can introduce uncertainty in more complex or heterogeneous basins.

 

Location: Section 5.2

Revision: Clarified the applicability and limitation of the constant discharge assumption.

 

Comment7: Insufficient performance metrics

 

Response7: We agree that additional evaluation metrics are important for comprehensive assessment. In the revised manuscript, the use of R² and RMSE is clarified, and the limitations of the discrepancy-based accuracy metric are explicitly discussed. We also acknowledge that widely used hydrological performance metrics such as NSE, KGE, and MAE would provide additional insight and have identified their inclusion as an important direction for future work.

 

Location: Section 4.1 and Section 5.1

Revision: Clarified evaluation metrics and added discussion on their limitations and future improvements.

 

Comment8: Accuracy definition issue

 

Response 8: We agree that the accuracy definition used in this study is unconventional. The revised manuscript clarifies that this metric is intended for operational interpretation within a narrow error range and differs from standard hydrological indicators. Its limitations are explicitly acknowledged.

 

Location: Section 5.1

Revision: Added clarification on the interpretation and limitation of the accuracy metric.

 

Comment9: Uncertainty propagation

 

Respons9: We agree that uncertainty propagation is an important aspect of flood prediction. The revised manuscript clarifies that uncertainty in rainfall forecasts directly influences discharge estimation and subsequently affects flood depth prediction. Although a quantitative uncertainty analysis was not conducted, this limitation is acknowledged and discussed.

 

Location: Section 4.3

Revision: Added explanation of the influence of rainfall forecast uncertainty on model outputs.

 

Comment10: Lack of comparison with existing approaches

 

Response 10: We agree that comparison with existing approaches is important. In the revised manuscript, we clarified the relative characteristics of the proposed framework compared with conventional hydrodynamic models and data-driven approaches. The discussion highlights differences in computational demand and interpretability; however, a quantitative benchmarking analysis was beyond the scope of this study and is suggested for future research.

 

Location: Section 1 (Introduction) and Section 5 (Discussion)

Revision: Added qualitative comparison of methodological characteristics and limitations.

 

Comment11: AI module unclear

 

Response 11: We agree that the description of the AI component was unclear in the original manuscript. The term “AI-assisted” has been removed, and the framework is now described more clearly as a regression-based and data integration approach.

 

Location: Section 2.2

Revision: Removed ambiguous terminology and clarified the role of the prediction module.

 

Comment12: Data availability

 

Response 12: We acknowledge the importance of data transparency. The manuscript clarifies that the datasets used in this study are restricted and not publicly available. However, the data sources and processing procedures are described in detail to support reproducibility.

 

Location: Data Availability Statement and relevant sections

Revision: Clarified data availability and described data usage.

 

Comment13: Figure quality

 

Response 13: The figures have been revised to improve readability.

 

Location: Figures 8–12

Revision: Improved layout, labeling, and resolution.

 

Comment14: Language issues

 

Response 14: The manuscript has been carefully edited to improve clarity, reduce redundancy, and enhance readability.

 

Location: Entire manuscript

Revision: Improved sentence structure and consistency.

 

Comment15: Insufficient discussion

 

Response 15: The discussion has been expanded to better interpret the results and limitations. Additional explanation of performance variability among streams and the limitations of the proposed approach has been included.

 

Location: Section 5 and Section 6

Revision: Expanded discussion on model behavior and limitations.

 

Comment16: Real-time applicability

 

Response 16: We agree that real-time applicability should be clearly explained. The revised manuscript clarifies that the framework enables rapid estimation by relying on pre-established relationships rather than real-time numerical simulation, which supports its applicability in operational settings. However, detailed quantitative analysis of computational efficiency is not included and is suggested for future work.

 

Location: Section 4.1 and Section 5

Revision: Clarified conceptual basis of real-time applicability.

 

Comment17: Sensitivity analysis

 

Response 17: We acknowledge that sensitivity analysis is important for evaluating model robustness. Although it was not conducted in this study, this limitation has been acknowledged, and future work is suggested to investigate the sensitivity of key parameters.

 

Location: Section 6 (Conclusions)

Revision: Added statement on the need for sensitivity analysis.

 

Comment18: Terminology

 

Response 18: We agree that clearer explanation of technical terminology is beneficial. The manuscript has been revised to improve clarity of terminology within the methodological descriptions.

 

Location: Section 4.1

Revision: Improved clarity of terminology used in the analysis.

 

Comment19: Limitations

 

Response 19: The limitations of the study have been expanded and more clearly articulated, including assumptions related to peak-based modelling, simplified hydraulic representation, and data availability constraints.

 

Location: Section 6 (Conclusions)

Revision: Expanded discussion of limitations.

 

Comment20: Conclusion overstatement

 

Response 20: The conclusions have been revised to moderate claims and better reflect the scope and limitations of the study.

 

Location: Section 6 (Conclusions)

Revision: Reduced overstatement and emphasized applicability conditions.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Editor, The authors have satisfactorily addressed my suggestions. In my opinion, the revised version is suitable for publication.

Author Response

Response to Reviewer 2 Comments

 

We sincerely thank the reviewer for the positive evaluation and for confirming that the revised manuscript has satisfactorily addressed the previous comments.

 

Comment1: The authors have satisfactorily addressed my suggestions. In my opinion, the revised version is suitable for publication.

 

Response 1: We sincerely thank the reviewer for the positive evaluation of our manuscript and for acknowledging that the revisions have satisfactorily addressed the previous comments.

We greatly appreciate the reviewer’s time and constructive feedback, which have helped us improve the quality and clarity of our work. We are pleased to know that the revised version is considered suitable for publication.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

Despite the authors’ efforts to revise the manuscript and address the reviewer’s comments, several critical issues remain insufficiently resolved. While improvements in clarity and presentation are acknowledged, many of the responses rely primarily on clarification or acknowledgment of limitations rather than substantive methodological enhancements. The main unresolved concerns are summarized below.

First, the absence of quantitative benchmarking remains a major limitation. The reviewer explicitly requested comparison with existing approaches (e.g., hydrodynamic models, machine learning methods, or simpler regression formulations) to demonstrate the effectiveness of the proposed framework. The authors indicate that such analysis is beyond the scope of the study and provide only qualitative comparisons. This response is not sufficient. Without quantitative benchmarking, the claimed effectiveness and practical advantage of the proposed approach cannot be objectively evaluated.

Second, the manuscript still lacks standard hydrological performance metrics such as Nash–Sutcliffe Efficiency (NSE), Kling–Gupta Efficiency (KGE), and Mean Absolute Error (MAE). The authors acknowledge their importance but defer their inclusion to future work. This is not acceptable, as the use of such metrics is fundamental for model evaluation and comparability within the hydrological literature. The current reliance on R², RMSE, and a non-standard accuracy metric does not provide a sufficiently robust assessment.

Third, although the authors introduced a calibration–validation split, the concern regarding potential overfitting is not fully resolved. The extremely high coefficients of determination (approximately 0.99) remain questionable, and the manuscript does not include more rigorous validation approaches such as k-fold cross-validation, uncertainty bounds, or a thorough assessment of generalization capability. As such, the robustness of the model is still not convincingly demonstrated.

Fourth, the issue of uncertainty propagation has not been adequately addressed. While the authors acknowledge that rainfall forecast uncertainty affects discharge and depth predictions, no quantitative analysis is provided. Given that rainfall uncertainty is a primary source of error in flood prediction, this omission significantly limits confidence in the reliability of the results.

Fifth, no sensitivity analysis of key parameters has been conducted. The authors recognize its importance but postpone it to future work. This limits the ability to assess model stability and the influence of parameter uncertainty, which is an essential component of modeling studies.

Sixth, the justification for the selection of the four-parameter logistic (4PL) model remains insufficient. Although additional explanation has been added, the manuscript still lacks a systematic and quantitative comparison with alternative models. Without such evidence, the choice of the 4PL formulation is not fully supported.

Overall, while the manuscript has improved, the core scientific concerns related to validation, benchmarking, uncertainty, and model robustness remain unresolved. Therefore, in its current form, the manuscript does not yet provide sufficient evidence to support its claims.

Based on the above, a further round of major revision is recommended before the manuscript can be considered for publication.

Author Response

Response to Reviewer 3 Comments

 

We sincerely thank the reviewer for the continued careful evaluation and for highlighting the remaining critical issues. All comments have been carefully addressed, and the manuscript has been revised accordingly, as detailed below.

 

Comment1: Lack of benchmarking against existing approaches

 

Response 1: We agree that quantitative benchmarking against existing approaches is important. At the initial stage of this study, physically based hydraulic models were examined. However, due to the narrow and steep characteristics of small streams, numerical instability frequently occurred, and simulations often diverged. Hydrologic approaches such as the rational method were also tested, but these methods tended to overestimate peak discharge and showed limitations in representing rapid response behavior.

To address these limitations, various regression formulations, including linear, polynomial, exponential, and power-law relationships, were evaluated using a nomograph-based framework. These approaches were found insufficient to represent the nonlinear response characteristics of small streams. Based on these findings, the four-parameter logistic (4PL) model was adopted.

This model selection process was previously documented in N.D.M.R., Korea (2022), and was not repeated in the original manuscript. In the revised manuscript, this background has been clarified.

Machine learning approaches were also considered. However, their application is currently limited by the availability of long-term, high-quality datasets for small streams. This limitation has been clarified, and future application of such approaches is suggested.

 

Location: Section 1 (Introduction)

Revision: Added explanation of limitations of existing approaches and clarified the model selection process.

 

Comment2: Absence of standard hydrological performance metrics

 

Response 2: We agree that additional performance metrics are necessary for a comprehensive evaluation. In the revised manuscript, Nash–Sutcliffe Efficiency (NSE), Kling–Gupta Efficiency (KGE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) have been included.

Accuracy Ratio (%) was retained as an indicator of the proportion of predictions within a predefined error range. However, its limitation in representing error magnitude has been explicitly clarified.

NSE was used to evaluate how well the predicted time series reproduces observed variability, while KGE was included to assess correlation, variability, and bias simultaneously. RMSE and MAE were used to quantify the magnitude of prediction errors, with RMSE providing sensitivity to large deviations.

The results indicate that NSE and KGE values are consistently high for both discharge and flood depth, while RMSE reveals relatively larger errors in discharge under high-flow conditions.

 

Location: Section 5.1

Revision: Added NSE, KGE, and MAE, and expanded interpretation of performance metrics.

 

Comment 3: Concern about overfitting despite calibration–validation split

 

Response 3: We acknowledge the concern regarding potential overfitting. The calibration–validation framework applied in this study is based on our previous work (N.D.M.R., Korea, 2023). The revised manuscript clarifies that model performance was evaluated using independent events and that the proposed framework incorporates updated rainfall inputs and iterative refinement, which reduces prediction errors compared to static approaches.

 

Location: Section 4.1

Revision: Clarified calibration–validation procedure and its linkage to previous work.

 

Comment 4: Lack of sensitivity analysis of key parameters

 

Response 4: We recognize the importance of sensitivity analysis. In this study, the primary influencing factor is rainfall accumulation time. A detailed sensitivity analysis has already been conducted in a previous study (Cheong et al., 2024), which uses the same framework and data characteristics. Therefore, it was not repeated in this manuscript. This linkage has been clarified in the revised version.

 

Location: Section 6 (Conclusions)

Revision: Added reference to prior sensitivity analysis and clarified its applicability.

 

Comment 5: Justification for selecting the 4-parameter logistic (4PL) model

 

Response 5: We agree that a clearer justification of the 4PL model is necessary. In the revised manuscript, the selection is explained not only from a statistical perspective but also based on the physical characteristics of small streams.

Small streams often exhibit intermittent flow behavior. At low rainfall levels, discharge does not increase rapidly, while at higher stages, the increase becomes gradual despite continued rainfall. This nonlinear behavior cannot be adequately represented by exponential or power-law functions.

The 4PL model allows representation of both initial slow response and saturation-like behavior, making it suitable for describing rainfall–discharge relationships in small streams.

 

Location: Section 4.1 and Section 6 (Conclusions)

Revision: Added physical interpretation supporting the use of the 4PL model.

 

If needed, we can further strengthen the response by adding a short comparison table or schematic explanation of model behaviour, but the above revisions aim to clarify the reviewer’s concerns while maintaining consistency with the scope of the current study.1

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has improved in clarity and presentation. However, some key concerns have not been fully addressed and remain limited. These aspects should be clearly acknowledged as limitations of the study.