Enhancing Flood Simulation in Data-Limited Glacial River Basins through Hybrid Modeling and Multi-Source Remote Sensing Data

Round 1

Reviewer 1 Report

The paper requires major modifications:

1-Authors have applied various ML models without mentioning setting parameters of Random Forest (RF), Gradient Boosting (GDBT), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), and Transformer (TF). After development of ML models, authors should provide setting parameters along with discussions and reasonable descriptions on selection of these parameters.

2-Literature review was poorly written. Authors should have mentioned a wide range of flood monitoring-related-investigations in terms of remote sensing techniques, artificial intelligence models, and typical satellite images. Authors can increase related literature to 30 papers.  For example, the papers  entitled "Flood monitoring by integration of Remote Sensing technique and Multi-Criteria Decision Making method" and "Flood risk mapping by remote sensing data and random forest technique" are well-matched examples for improving literature review.   

3-Nash coefficients need to be removed from the paper. Authors can use Scatter index, BIAS, and Discrepancy Ratio (DR) can be merged to the results and all tables as seen in Receiving More Accurate Predictions for Longitudinal Dispersion Coefficients in Water Pipelines: Training Group Method of Data Handling Using Extreme Learning Machine Conceptions.

4-Mapping zone of flood monitoring can be added to the paper.

5-Vulnerability of flood monitoring can be added to the results and discussion. Additionally, risk levels of flood monitoring should be done. After that, Kappa Coefficients (KCs) and ROC-AUC should be merged to the results and discussions as seen in the Evaluation of River Water Quality Index Using Remote Sensing and Artificial Intelligence Models

6-Add Taylor diagrams, Violin diagrams, and heat maps to the performance of ML models for both training and testing stages. 

7-Conclusion section should be significantly revised.

Author Response

1. Authors have applied various ML models without mentioning setting parameters of Random Forest (RF), Gradient Boosting (GDBT), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), and Transformer (TF). After development of ML models, authors should provide setting parameters along with discussions and reasonable descriptions on selection of these parameters.

Authors’ Response: Thank you for your thoughtful review and valuable feedback about setting parameters for the machine learning (ML) models used in our study. We agree with your suggestion and have listed optimal hyperparameter combinations of Random Forest (RF), Gradient Boosting (GDBT), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), Support Vector Machine (SVM) and Transformer (TF) in Table 1 in the revised manuscript.

We also agree with your suggestion about setting parameters along with discussions and reasonable descriptions on selecting these parameters. Nonetheless, the machine learning models employed in this manuscript remain black box models, lacking physical interpretability. Hence, it presents a challenge for us to analyze the impact of model-related parameters on the model outcomes or the underlying physical processes. Furthermore, the GridSearchCV technique, which is widely used in searching optimal parameters for machine learning (such as Ahmad et al., 2022; Alhakeem et al., 2022), has been employed in this manuscript to meticulously optimize model parameters for the selected ML models. In this way, the simulation results are considered to be satisfactory and reliable. The detailed revision in the text is: The GridSearchCV technique [36] was utilized in this study to optimize the hyperparameters. GridSearchCV meticulously navigates the ML model's performance landscape by exhaustively scanning the designated hyperparameter combinations, leading to the identification of the optimal hyperparameter configuration. Consequently, the optimal hyperparameter combinations for the six selected ML models have been obtained and are presented in Table 1 (Lines 250 - 254)

References:

Ahmad G N, Fatima H, Ullah S, et al. Efficient medical diagnosis of human heart diseases using machine learning techniques with and without GridSearchCV. IEEE Access 2022, 10: 80151-80173.

Alhakeem Z M, Jebur Y M, Henedy S N, et al. Prediction of ecofriendly concrete compressive strength using gradient boosting regression tree combined with GridSearchCV hyperparameter-optimization techniques. Materials 2022, 15(21): 7432.

 

2. Literature review was poorly written. Authors should have mentioned a wide range of flood monitoring-related-investigations in terms of remote sensing techniques, artificial intelligence models, and typical satellite images. Authors can increase related literature to 30 papers. For example, the papers entitled "Flood monitoring by integration of Remote Sensing technique and Multi-Criteria Decision Making method" and "Flood risk mapping by remote sensing data and random forest technique" are well-matched examples for improving literature review.

Authors’ Response: We appreciate your valuable feedback on our research. We agree that a wide range of flood monitoring-related-investigations in terms of remote sensing techniques, artificial intelligence models, and typical satellite images is an interesting and meaningful topic, however, it may not be directly pertinent to the primary focus of our study. The core research objective of this manuscript is to apply hybrid models in flood prediction in data-scarce glacial river basins. Hence, we confirm that the literature review should primarily center on the context of hybrid models in flood prediction, rather than providing an extensive literature review. Furthermore, current flood prediction research can be broadly categorized into two primary domains. The first domain focuses on simulating the velocity or discharge of floods at hydrological stations, while the second on concentrates on modeling the spatial distribution of floods. Both aspects are crucial for effective flood disaster forecasting. However, we acknowledge that attempting to encompass both of these complex aspects within a single manuscript would be challenging. Given that the core aim of our manuscript is to enhance the simulation of flood events specifically in data-scarce mountainous regions, we find it challenging to delve extensively into the topic of flood risk mapping. We hope that this clarification reaffirms the focus and scope of our research in the context of flood prediction in glacial river basins.

3. Nash coefficients need to be removed from the paper. Authors can use Scatter index, BIAS, and Discrepancy Ratio (DR) can be merged to the results and all tables as seen in Receiving More Accurate Predictions for Longitudinal Dispersion Coefficients in Water Pipelines: Training Group Method of Data Handling Using Extreme Learning Machine Conceptions.

Authors’ Response: Thank you for your feedback and suggestions regarding using Nash coefficients in our paper. However, there may be a slight misunderstanding in this regard. We firmly believe that Nash-Sutcliffe efficiency coefficient () holds significant relevance and is widely accepted as a standard metric for evaluating model performance in hydrological research, including flood prediction. While we acknowledge the utility of metrics such as Scatter Index, BIAS, and Discrepancy Ratio (DR), it's important to note that each metric serves distinct purposes. From the perspective of comprehensive understanding, they offer valuable insights but may not entirely replace . If you have any further suggestions or can provide specific reasons for the exclusion of  and the inclusion of the proposed metrics, please feel free to elaborate, and we will carefully consider your suggestions to improve the quality of our research.

4. Mapping zone of flood monitoring can be added to the paper.

Authors’ Response: We sincerely appreciate the reviewer’s valuable suggestions. The significance of mapping flood monitoring zones cannot be overstated, as it plays a crucial role in both flood monitoring and disaster prevention. However, it is indeed a complex and multifaceted subject, encompassing various aspects that demand thorough consideration. As outlined in point 2, the primary focus of our study centers on the application of hybrid models for flood prediction in data-scarce glacial river basins, rather than studying flood mapping. Expanding the manuscript to incorporate flood monitoring could indeed risk diluting the research's central theme. We highly value your recommendation, and we intend to explore the topic of flood monitoring in our following studies, as your suggestion. This will allow us to delve more deeply into this critical aspect of flood research while maintaining the current manuscript's focus intact.

5. Vulnerability of flood monitoring can be added to the results and discussion. Additionally, risk levels of flood monitoring should be done. After that, Kappa Coefficients (KCs) and ROC-AUC should be merged to the results and discussions as seen in the Evaluation of River Water Quality Index Using Remote Sensing and Artificial Intelligence Models

Authors’ Response: We appreciate the consideration of including the vulnerability of flood monitoring as a component for flood risk mapping. However, it's worth noting that effectively incorporating this aspect into flood simulations can be challenging. The applicability of KC and ROC-AUC metrics also poses difficulties in flood simulations. In our understanding, potential readers of this manuscript may want to know more about the performance and prospects of the hybrid model. Hence, we focus on these two aspects in the discussion session to provide potential readers with a more comprehensive understanding of the performance and potential of hybrid models. Meanwhile, we add limitations of the current study to present more insight into the combination of physically based models and ML for potential readers.

6. Add Taylor diagrams, Violin diagrams, and heat maps to the performance of ML models for both training and testing stages.

Authors’ Response: We appreciate your valuable suggestion. In our manuscript, we utilize machine learning models to simulate streamflow by employing the outputs and weather forcing data from the physically based SPHY model. We firmly believe that the Nash-Sutcliffe Efficiency (NSE) metric serves as a robust and adequate measure to quantify the performance of our simulation results, allowing us to assess the quality of our machine learning methods effectively. Hence, we are of the opinion that incorporating Taylor diagrams, Violin diagrams, and heat maps would be unnecessary in this study. We believe that NSE provides sufficient insight into the performance evaluation, aligning with the scope and objectives of our research. Hope to get your support!

7. Conclusion section should be significantly revised.

Authors’ Response:: Thank you for your feedback concerning the Conclusion section of our paper. We certainly consider your suggestion to improve this section as follows:

To enhance flood simulation in data-scarce glacial river basins, we present a novel hybrid modeling approach that leverages multi-source remote sensing data, a physically based hydrological model (SPHY), and machine learning (ML) techniques. Within this hybrid model, remote sensing data, including MODIS snow cover data and glacier mass balance data, are effectively employed to calibrate the SPHY model. The SPHY model generates crucial components like baseflow, rain runoff, snowmelt runoff, and glacier melt runoff in the high mountainous regions, which act as new inputs for the subsequent ML components. This newly developed hybrid model undergoes rigorous training and validation assessments. Subsequently, the best-performing hybrid model is selected through comprehensive comparisons, followed by an uncertainty analysis.

Through a case study within the Manas River basin in Central Asia, our study reveals several significant insights. First and foremost, the hybrid model (SPHY-RF) markedly enhances flood simulation accuracy compared to the standalone physical-based hydrological model (SPHY) which plays an important role in flood forecasting. Remarkably, SPHY-RF outperforms five other hybrid models (SPHY-GDBT, SPHY-LSTM, SPHY-DNN, SPHY-TF, SPHY-SVM) regarding both streamflow and flood simulation in the testing period. It may be due to the lack of training data, resulting in the performance of hybrid models based on deep learning algorithms is not as good as that of RF. Additionally, the integration of multi-objective optimization demonstrates its potential to improve streamflow simulation for the SPHY model, subsequently enhancing the streamflow and flood simulation performance of the SPHY-RF model. By utilizing bootstrap sampling, the 95% uncertainty interval for SPHY-RF is established, effectively covering 87.65% of flood events.

In conclusion, our findings highlight the substantial potential of the hybrid modeling approach for simulating floods in data-scarce glacial river basins. This approach not only establishes a robust framework for flood simulation and forecasting in complex environments but also lays the groundwork for investigating extreme hydrological events under the influence of a warming climate in alpine regions worldwide.

Above all of these can be found in lines 721-749 of the revised manuscript.

 

Author Response File: Author Response.docx

Reviewer 2 Report

In the introduction, the authors are encouraged to expand the section discussing the application of ML. For example, the authors have not included the improved ML models where integrated with hybrid optimizations. The authors can maybe refer to : Application of augmented bat algorithm with artificial neural network in forecasting river inflow in Malaysia, Investigation of Meta-heuristics Algorithms in ANN Streamflow Forecasting and Rainfall-runoff modelling using improved machine learning methods: Harris hawks optimizer vs. particle swarm optimization. Maybe by including these works the authors can capture the development that have been proposed to improve ML models. Still the authors can show the limitations of the hybrid models and then propose his method which is well presented in the paper.

Why the authors only use these Performance measures, maybe can justify.

“Table 2 Statistical evaluations of annual glacier melt, monthly snow cover fraction and daily streamflow simulated by the SPHY model.” Why there are negative values, please justify in the paper.

“Table 2 Statistical evaluations of annual glacier melt, monthly snow cover fraction and daily 389 streamflow simulated by the SPHY model.” I think the authors if possible mention the R2 in the scatter diagram and improve the figure.

 

What are the limitations of the current study, maybe should be included. 

Author Response

Reviewer 2：

In the introduction, the authors are encouraged to expand the section discussing the application of ML. For example, the authors have not included the improved ML models where integrated with hybrid optimizations. The authors can maybe refer to: Application of augmented bat algorithm with artificial neural network in forecasting river inflow in Malaysia, Investigation of Meta-heuristics Algorithms in ANN Streamflow Forecasting and Rainfall-runoff modelling using improved machine learning methods: Harris hawks optimizer vs. particle swarm optimization. Maybe by including these works the authors can capture the development that have been proposed to improve ML models. Still the authors can show the limitations of the hybrid models and then propose his method which is well presented in the paper.

Authors’ Response: Thank you for your valuable feedback regarding the introduction section of our manuscript. We appreciate your suggestion to expand the discussion on applying machine learning and hybrid optimization techniques. However, the primary objective of this study is to enhance flood prediction capabilities in data-scarce mountainous regions. Therefore, the logical progression in the introduction section entails reviewing existing methods potentially applicable to flood forecasting in such areas. By highlighting the identified issues, we introduce the hybrid modeling approach, which aligns seamlessly with the overarching goal of this study. While parameter optimization holds significance, it does not constitute the primary focus of this research. Delving into extensive discussions on parameter optimization might sidetrack readers from the essence of our study. In reality, we hold a strong interest in the matter of parameter optimization and plan to investigate the impact of various parameter optimization methods on the hybrid model in future research endeavors. We sincerely appreciate your constructive feedback and express our gratitude for your valuable contributions to our study. As for the limitations of the hybrid models, we emphasize this specifically in section 5.4 of the discussion to present more insights for potential readers.

1. Why the authors only use these Performance measures, maybe can justify.

Authors’ Response: We would like to express our gratitude for your questions regarding the performance metrics used in our study. Your feedback is invaluable and helps us clarify why we selected these specific performance metrics. We chose R, NSE, PBIAS, and RMSE as our performance metrics for several reasons, which are widely recognized and used in the fields of hydrology and water resources management: (1) Versatility: These metrics apply to various types of hydrological models and diverse application scenarios, enhancing the universality and comparability of our research findings. (2) Comprehensive Assessment: R, NSE, PBIAS, and RMSE cover different aspects of model performance. For instance, R and NSE consider linear correlation and consistency between model simulations and observations, PBIAS accounts for model bias, while RMSE measures the root mean square error of the model. Using these metrics simultaneously, we can comprehensively evaluate model performance and ensure that the model performs well across various dimensions. (3) Ease of Interpretation: These metrics are typically easy to interpret and communicate to non-experts, including policymakers and water resource managers.

Therefore, we have revised the corresponding part of the manuscript as follows:

Given the versatility and ease of interpretation, the following four performance measures were comprehensively used to qualitatively evaluate the performance of the developed models.

The relevant revisions can be found on lines 363-364 of the revised manuscript.

2. “Table 2 Statistical evaluations of annual glacier melt, monthly snow cover fraction and daily streamflow simulated by the SPHY model.” Why there are negative values, please justify in the paper.

Authors’ Response: Certainly, we appreciate your attention to Table 2. We've explained it, which can be found on lines 431-445 of the revised manuscript. We also show it as follows:

The observed performance dip during the testing period is primarily attributed to the suboptimal accuracy displayed in 2009 and 2010, manifesting as substantial overestimations and underestimations. Despite NSE values of 0.21 during training and -12.9 during testing, indicating subpar performance, the simulation remains relatively satisfactory. This is primarily due to the inherent discontinuities in both the temporal and spatial dimensions of the glacier mass balance data sourced from Hugonnet et al [1], necessitating interpolation. Consequently, the temporal alterations in glacier mass balance data are comparatively modest. In actuality, the melt of the Manas River glacier is significantly influenced by temperature and precipitation. A glance at Figure 3 reveals that in 2009 and 2010, there was a conspicuous reduction in snow cover area in contrast to other years, resulting in a significantly elevated glacier melt volume in 2010. Conversely, the diminished snow cover in 2009 contributed to a reduced glacier melt volume, primarily attributable to lower temperatures. This, in turn, results in noticeably diminished runoff for the same year which can be found in Figure 5a.

3. “Table 2 Statistical evaluations of annual glacier melt, monthly snow cover fraction and daily 389 streamflow simulated by the SPHY model.” I think the authors if possible mention the R2 in the scatter diagram and improve the figure.

Authors’ Response: Thank you for your valuable suggestion concerning Table 2 and the scatter diagram used for statistical evaluations. We recognize the importance of providing a comprehensive view of the model's performance, and your suggestion to include the coefficient of determination (R2) is noted. However, we'd like to explain that we have already utilized the correlation coefficient (CC), which is equivalent to R. While we acknowledge that R2 and R serve slightly different purposes, introducing R2 in addition to R might potentially lead to reader confusion. Additionally, it's worth noting that Nash-Sutcliffe Efficiency (NSE) and R2 are similar in assessing model performance. As a result, we have decided not to make this specific modification to the figure, as we believe that it may not add significant clarity but could potentially introduce complexity. We hope you understand our rationale for this decision. We genuinely value your feedback, and if you have any further recommendations or specific details you would like us to consider while improving the figure or any other aspect of the manuscript, please feel free to share them.

 

4. What are the limitations of the current study, maybe should be included.

Authors’ Response: Thank you for your valuable feedback. We appreciate your suggestion to include the limitations of our current study. Add discussion to the manuscripts as follows:

The results of this study suggest that integrating physical models and ML methods to predict floods in high-altitude mountainous areas is a promising approach but comes with several limitations. Firstly, this hybrid model lacks model interpretability. ML methods often manifest as black-box models, making it challenging to interpret their decision-making processes. Although the hybrid model's inputs in this study have physical properties, there are no physical constraints in the process of being used for ML, which makes it still hard to understand the contributing factors to a flood event. Secondly, the issue of data imbalance is noteworthy. Flood events are typically infrequent, resulting in datasets predominantly composed of non-flood events and a sparse representation of flood events. Extra measures may be necessary to balance the data to prevent the model from leaning excessively toward non-flood events. Thirdly, the risk of overfitting still exists. When integrating complex models with numerous parameters, there is a risk of overfitting, particularly when data is limited. For example, the best-performing model in this study, SPHY-RF, performed better in the training period than in the testing period, so it is necessary to use more data to train the model in practical applications. Furthermore, there are concerns regarding the applicability of this approach under changing climate conditions. Predicting floods under changing climate conditions is challenging. Changes in precipitation and temperature can lead to fluctuations in the outputs of models like SPHY. When incorporated into hybrid models, these variations can compound, making it difficult to apply the model to predict floods under changing climate conditions. Despite these limitations, integrating physical models and machine learning methods still promises to improve flood prediction accuracy in high-altitude mountainous regions, especially when data is scarce. Overcoming these challenges requires further research and methodological enhancements to fully harness the potential of this approach.

The relevant revisions can be found on lines 695-720 of the revised manuscript.

Author Response File: Author Response.docx