Integrating CMIP6 and Remote Sensing Datasets for Current and Future Flood Susceptibility Projections Using Machine Learning Under Climate Change Scenarios in Demak District for Future Sustainable Planning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Decision: Major Revisions

Dear Authors,

Thank you for submitting your manuscript titled "Integrating Coupled Model Intercomparison Project Phase 6 Global Climate Model and Remote Sensing Datasets for Current and Future Flood Susceptibility Projections Using Machine Learning Under Climate Change Scenarios in Demak District."

The topic of your study is highly relevant and contributes meaningfully to the growing body of literature on climate-induced flood risk assessment using modern data-driven techniques. Your integration of CMIP6 climate projections, remote sensing datasets, and machine learning models is commendable and holds significant promise for practical applications in disaster risk management and planning.

However, before the manuscript can be considered for publication, substantial improvements are necessary. Key aspects related to the methodological clarity, data integration process, model validation, and interpretation of results require further elaboration and refinement. The manuscript would also benefit from enhancements in structure, flow, and overall presentation to meet the journal’s standards.

We therefore invite you to revise your manuscript thoroughly and address all concerns raised, with detailed explanations and improvements where appropriate.

We look forward to receiving your revised submission.

Kind regards,

Sincerely,

Reviewer

 

 

My Comments

 

 

1. The number of keywords (eight) is excessive and should be reduced to five or six for clarity and compliance with common journal standards. Some keywords are overlapping or can be merged (e.g., flood susceptibility model and flooding), and others can be simplified (e.g., CMIP6 GCM can be shortened to just CMIP6).
2. The introduction is too lengthy and could be more concise. Some content is repetitive, especially regarding flood impacts and modeling approaches, which can be streamlined for better readability.
3. The novelty of the study is not clearly stated. While the integration of CMIP6, remote sensing, and machine learning is modern, the introduction should explicitly highlight what is new or original compared to prior studies.
4. Overuse of clustered references (e.g., [2,3], [16–18]) suggests surface-level citation. Consider discussing key references more meaningfully to demonstrate critical engagement.
5. Formatting issues such as the placeholder text “Academic Editor: Firstname Lastname” and journal metadata (lines 23–30) should be removed or placed appropriately, as they disrupt the flow of the introduction.
6. The description of the study area is clear, but it would be helpful to provide a map or coordinate reference to better locate Demak District for readers unfamiliar with the region.
7. The use of Sentinel-1 SAR imagery and the Otsu method for flood inventory mapping is appropriate. However, the validation method relies heavily on UNOSAT data without field survey data, which should be clearly stated as a limitation.
8.  More explanation is needed about the stratified random sampling method for flood and non-flood points — specify how samples were distributed spatially and temporally.
9. The selection of 14 flood conditioning factors is comprehensive. Still, a discussion on the potential multicollinearity and the rationale for choosing each factor based on prior studies should be emphasized here (also linked to the multicollinearity section).
10. The detailed description of topographic and hydrological factors (elevation, slope, aspect, curvature) is good, but the explanation can be shortened or moved to supplementary materials for better flow.
11. The discussion of the indices (TWI, SPI, HAND, NDVI, NDSI) is thorough. However, equations (1)-(4) should be numbered consistently and explained with example values or ranges relevant to the study area.
12. The CMIP6 GCM data selection and validation are appropriate. However, the spatial resolution of precipitation data (925 meters to 5 km) varies widely — discuss how this impacts model consistency.
13. Validation of GCM outputs using gauge stations is good, but the limitation of only three stations for a 995 km² area should be mentioned.
14. The multicollinearity analysis (VIF, tolerance) is well included but should be explicitly linked to the final selection of conditioning factors — mention which variables were excluded or retained.
15. The explanation of machine learning models (MLP-NN, RF, SVM, XGBoost) is clear but too detailed for the main text — consider summarizing and referring to references for detailed descriptions. 
16. The evaluation metrics are comprehensive and appropriate. Please clarify if all metrics were used for model selection or only AUC and accuracy. 
17. The symmetrical uncertainty (SU) and rating metric (RM) methods for GCM performance assessment are well explained. It would be useful to clarify why models below 0.70 RM were excluded.
18. The multi-model ensemble approach and weighting are good practices. Provide more details about how uncertainty was quantified and how ensemble results were validated. 
19. The statistical evaluation (Pearson r, NSE, RMSE, md) is appropriate. Include brief interpretation of acceptable threshold values for these metrics in this context.
20. Multicollinearity Analysis (Section 3.1):Although VIF values are below 5, consider discussing the potential impact of variables with VIF closer to 3.5 (e.g., SPI) on model stability. Suggest including a brief explanation or threshold justification for readers unfamiliar with VIF/tolerance criteria. 
21. Feature Importance (Section 3.2):The variation in feature importance across models is noted, but it would help to discuss why certain variables (e.g., NDVI, LULC) perform differently across algorithms. Suggest adding insights on how these differences could affect practical flood management decisions.
22. Model Validation (Section 3.3):Cross-validation method (5-fold) is mentioned, but it would improve clarity to describe the data splitting approach (e.g., stratified sampling) and if hyperparameter tuning was applied. Suggest adding confidence intervals or statistical significance testing for key metrics to better support claims of model superiority. 
23. Flood Susceptibility Maps (Section 3.4):Maps comparison is clear, but the description could benefit from quantitative spatial metrics (e.g., spatial autocorrelation or Kappa spatial statistics) to support visual observations. Suggest highlighting limitations in areas where models disagree and possible reasons (data quality, resolution, or local environmental factors). 
24. Future Flood Projection (Section 3.5):The use of SSP scenarios and multi-model ensembles is appropriate, but the section could better clarify how uncertainties in climate models propagate through to flood susceptibility predictions. Suggest discussing limitations of using only precipitation projections, considering other factors like land use change or river management in future flood risks. 
25. Presentation and Readability:Tables and figures are comprehensive, but some parts could be improved with clearer captions and referencing in the text. Suggest ensuring consistency in abbreviations and defining all terms at first use for broader readership accessibility.
26. Strengthen the explanation of XGBoost's superiority: The paragraph mentions XGBoost's advantages briefly. Consider expanding on why it performs better compared to other models in this specific flood susceptibility context, possibly with references to handling feature interactions or imbalanced data. Suggest clarifying if hyperparameter tuning was done specifically for XGBoost.
27. Clarify the linkage between environmental factors and model results:You mention strong correlations of elevation, precipitation, and land use with flood susceptibility. Consider adding brief quantitative support (e.g., correlation coefficients or feature importance ranks) from your results to reinforce this. Suggest discussing any surprising or counterintuitive findings regarding these variables if present.
28. Expand on SSP scenario impacts with clearer structure:The explanation of SSP scenarios and temperature/precipitation effects is a bit dense and interrupted by table snippets. Recommend restructuring for clarity, e.g., summarize key points before introducing detailed numbers or tables. Explain the practical implications of rising flood susceptibility percentages for local stakeholders or planners.
29. Address assumptions and limitations clearly:You note the assumption that land use and infrastructure remain unchanged, which is important. Suggest explicitly discussing how this assumption might limit the projections’ realism. Also, mention any other limitations, such as uncertainties in GCM precipitation projections or spatial resolution limits.
30. Improve the discussion on SAR data use:The value of Sentinel-1 SAR is well noted, but consider elaborating on any challenges (e.g., temporal gaps, mixed pixel issues) and possible integration with other data sources (optical satellites, drones). Suggest pointing out how frequently SAR imagery was available for this study and how this affected model training.
31. Future research directions:The suggestion to integrate socioeconomic data and LULC prediction models (CA-MARKOV, CA-ANN) is excellent. Consider specifying which socioeconomic factors (population density, infrastructure vulnerability, etc.) would be most critical for Demak. You may also highlight the potential of coupling flood susceptibility models with early warning systems.
32. General language and flow:Some sentences could be simplified for better readability; avoid overly long sentences. Make sure all acronyms (e.g., SAR, LULC) are defined at first mention (though likely done earlier in the paper). Watch for minor grammatical slips (e.g., “their utility remain dependent” → “their utility remains dependent”).
33. The conclusion effectively summarizes the main findings but could start with a stronger, more explicit statement of the study’s overall contribution or novelty.For example: “This study presents a novel integration of Sentinel-1 SAR data, CMIP6 climate projections, and advanced machine learning models to robustly predict current and future flood susceptibility in Demak District.”
34. Emphasize how the improved flood susceptibility maps and projections can support local policymakers, urban planners, and disaster management agencies. You could add a sentence on how this work can guide land use planning, infrastructure development, or early warning systems.
35. Mention limitations briefly:Adding a brief acknowledgment of limitations (e.g., assumption of static land use, uncertainties in climate projections) will make the conclusion balanced.
36. Suggest future directions: You can briefly recommend future research, such as incorporating dynamic land use changes or socioeconomic vulnerability into models, or improving temporal resolution of flood monitoring.
37. Language polishing:Replace “providing a clear spatial representation of flood susceptibility” with a stronger phrase like “offering detailed, spatially explicit flood risk assessments.” Instead of “emphasizing the increasing impact,” use “highlighting the escalating risks posed by climate change.” Last sentence could be rephrased for more impact: “These findings underscore the urgent need for adaptive flood management and climate resilience planning in Demak District.”
38. Some entries (e.g., reference 22 and 68; reference 65 and 66) appear to be duplicates or near-duplicates. Make sure each entry is unique and not repeated unintentionally.
39. Formatting Consistency: A few inconsistencies exist in journal names (some have full names, others have abbreviations), volume/issue details, and DOI formatting. For example :Remote Sens (Basel) vs. Remote Sensing. Some references like #88 include journal name and issue but lack a proper DOI format.
40. Recommendation:Use a consistent citation style (e.g., APA, Chicago, or Elsevier/Harvard style) and ensure uniform formatting across all references.

 

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The topic is relevant. The article's problematic is consistent, bringing significant contributions to the field of study regarding actions outside of urban environmental planning to prevent flooding, especially in light of the impacts of global climate change. The methodology is consistent, and the statistics are sound. The results are robust, but they need to be compared with those from similar studies conducted in other cities, as the current discussion is merely descriptive.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This study develops flood susceptibility models for the Demak District in Indonesia by integrating remote sensing data, machine learning techniques, and CMIP6 Global Climate Model (GCM) data. It aims to map current flood susceptibility using Sentinel-1 SAR data and predict future flood susceptibility under three Shared Socioeconomic Pathway (SSP) scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) from 2021 to 2100. The XGBoost model shows the best performance in both current and future projections, indicating a significant increase in flood susceptibility, especially under higher emission scenarios.

1. The study uses remote sensing data with varying spatial resolutions (e.g., 30 m for SRTM DEM and 10 m for LULC). Consider incorporating higher-resolution datasets, where possible, especially for critical variables such as precipitation and land use, to enhance model accuracy. For instance, the 5000 m resolution of CHIRPS precipitation data could be replaced with finer-resolution data if available.
2. While the study uses Sentinel-1 SAR data for flood inventory mapping, integrating real-time satellite data could enhance the dynamic analysis of flood events. This would allow for more up-to-date flood susceptibility assessments.
3. The study selects 14 flood-conditioning factors. Including additional factors such as soil moisture content or detailed land cover changes could further improve model accuracy.
4. The study focuses on spatial flood susceptibility. Adding a temporal analysis component to assess how flood susceptibility changes within shorter time intervals (e.g., seasonal variations) could provide more comprehensive insights.
5. The machine learning models are trained using a dataset with an 80-20 split for training and testing. Consider using cross-validation techniques beyond the 5-fold cross-validation mentioned in the methodology to ensure robust model calibration.
6. While a multi-model ensemble approach is used to reduce uncertainties in future projections, quantifying the uncertainties associated with the current flood susceptibility model would strengthen the study's reliability.
7. The study validates the flood inventory map using UNOSAT data for the March 2024 flood. Expanding this validation to include more historical flood events would enhance the credibility of the flood susceptibility model.
8. The study projects future flood susceptibility under three SSP scenarios. A sensitivity analysis to determine how variations within each SSP scenario (e.g., different assumptions about population growth or technological development) affect flood susceptibility could provide deeper insights.
9. Combining machine-learning models with local expert knowledge could improve the interpretation of flood conditioning factors and the validation of flood susceptibility maps.
10. The study highlights the increasing flood susceptibility in the Demak District. Including specific adaptation strategies or recommendations based on the flood susceptibility maps would make the study more actionable for local policymakers.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for your submission to Sustainability. Your manuscript presents a scientifically sound and methodologically innovative approach by integrating CMIP6 global climate models, remote sensing (Sentinel-1 SAR), and machine learning techniques to assess current and future flood susceptibility in Demak District, Indonesia. The topic is highly relevant, especially in the context of increasing climate-related flood risk, and the study has strong potential for practical application in regional planning and disaster risk reduction.

Strengths of your Manuscript

• The multi-model ensemble of CMIP6 GCMs combined with ground station validation (high R and NSE values) provides confidence in the projected precipitation data.
• The use of Sentinel-1 SAR to derive flood inventories is appropriate and validated against UNOSAT maps, with high accuracy (>89%).
• The comparison of machine learning models and the selection of XGBoost as the optimal classifier is justified by performance metrics (AUC = 0.9291, Accuracy = 87.8%, Kappa = 0.752).
• The spatial visualization of flood susceptibility across SSP scenarios is clear and informative.

Here are some suggestions to improve your manuscript:

1. Discussion and Practical Implications:
   While the results are clearly presented, the discussion section could be strengthened by interpreting the broader implications of your findings. Do consider discussing:
• How flood susceptibility projections under climate change can inform local land-use planning, infrastructure design, and adaptation policy.
• How can the methodology be applied in data-scarce regions or integrated into early warning systems?
• How does this study compare with similar regional flood modeling studies in Southeast Asia or globally?
2. Conclusion and Future Directions:
   The conclusion could be enhanced by:
• Summarizing not only the findings but also specific policy recommendations or operational uses of your model.
• Suggesting avenues for future research, such as integration with socioeconomic vulnerability data or coupling with hydrodynamic models for real-time flood forecasting.
3. Limitations:
   Please explicitly acknowledge the limitations of the study, including:
• Potential misclassification or temporal mismatch in remote sensing data.
• Limitations of using NDVI or SAR (This remotely sensed data may be limited to lower resolutions and certain timeframes, provided free to the public) alone for flood mapping in urban or vegetated zones.
• Uncertainty is associated with climate projections and the downscaling approach used.
4. Discrepancy Between NDVI and LULC Maps (Figure 4):
   I have observed a noticeable inconsistency between Figure 4(h) (NDVI) and Figure 4(i) (Land Use and Land Cover). Specifically, some areas classified as built-up in the LULC map show unexpectedly high NDVI values, while some cropland or forested areas display lower NDVI values. This could stem from differences in spatial resolution, temporal coverage, or classification methodology between Landsat-8 and ESRI datasets. We recommend clarifying:
• Whether both datasets represent the same season and year.
• How the layers were harmonized and spatially aligned.
• Adding a color legend and value scale to both maps for better interpretability.
5. Figures and Visual Clarity:
• Ensure consistency in color symbology and scale bars across maps in Figures 9–14.
• Consider including a summary table of changes in flood susceptibility by SSP scenario to aid non-technical readers.

My final recommendations:

Your study makes a valuable contribution to climate adaptation research and flood risk assessment. With these minor but important improvements, particularly in expanding the discussion, clarifying visual discrepancies, and stating limitations, your paper will be well-positioned for publication and impact within the sustainability science community.

We look forward to seeing your revised manuscript.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors After reviewing the updated version, I am satisfied that the authors have addressed the major comments from the second round, and the manuscript meets the journal’s publication standards.   In my opinion, the paper can be accepted in its current form.

 

 

 

 

 

 

My Comments

 

Model Validation – Please clarify whether any validation or cross-validation procedure was applied beyond the 70/30 train–test split. If not, provide a rationale and discuss any implications for model robustness.

Overfitting Mitigation – While you note avoiding deeper architectures, it would be useful to explain whether other overfitting mitigation methods (e.g., dropout, L2 regularization, early stopping) were considered or tested.

Scalability and Generalization – Expand on how the models could be scaled or adapted for larger and more diverse datasets. Please consider discussing potential use of transfer learning or cross-industry datasets.

Practical Applications – Include a brief example or case scenario illustrating how the models could be applied in a real-world construction project during execution.

Reference Formatting and Accessibility – Ensure all citations follow a consistent format, add missing DOIs where available, and provide English translations for non-English titles (e.g., Croatian references).

In the abstract, consider adding the cost overrun model’s MAPE value for completeness.

In the metric interpretation section, a short “before vs. after tuning” example would make the discussion more concrete.

Some references still have inconsistent author name formats; please standardize them.

Overall, the paper is now much stronger and only requires these final adjustments to be ready for publication.

Recommendation: Minor Revision

 

Sincerely,

Reviewer

 

Author Response

Dear Reviewer 1,

We would like to sincerely thank you for your valuable time and constructive comments, which have greatly contributed to improving the quality and clarity of our manuscript. We have carefully reviewed and addressed each of your suggestions and have made corresponding revisions in the manuscript. The followings are point-by-point responses to comments. Our replies are given in bold following each comment.

Comments and Responses,

1. Model Validation – Please clarify whether any validation or cross-validation procedure was applied beyond the 70/30 train–test split. If not, provide a rationale and discuss any implications for model robustness.

>>> Thank you for your insightful question regarding our model validation procedure. We would like to clarify that in addition to the train-test split, we implemented a 5-fold cross-validation strategy. This approach was specifically used to enhance the model's robustness and generalizability, as stated in our Methodology section. This process was also an integral part of our hyperparameter optimization using GridSearchCV for each algorithm. Thank you again for your constructive feedback.

2. Overfitting Mitigation – While you note avoiding deeper architectures, it would be useful to explain whether other overfitting mitigation methods (e.g., dropout, L2 regularization, early stopping) were considered or tested.

>>> Thank you for your important question regarding overfitting mitigation. In our study, the primary methods for mitigating overfitting were a robust 5-fold cross-validation and careful hyperparameter optimization using GridSearchCV. This approach ensured we selected a model with balanced complexity. Furthermore, for models like XGBoost, we leveraged their effective built-in regularization methods. For the MLP-NN model, given its relatively shallow architecture, we found that the combination of cross-validation and hyperparameter tuning was sufficient for good generalization, as supported by the consistent model performance on both the training and test datasets (Table 9).

3. Scalability and Generalization – Expand on how the models could be scaled or adapted for larger and more diverse datasets. Please consider discussing potential use of transfer learning or cross-industry datasets.

>>> Thank you for this insightful suggestion to expand on our model's scalability and generalization.

Scalability: Our chosen top-performing model, XGBoost, is inherently capable of efficiently processing large datasets. This allows it to be scaled to larger geographical regions or higher-resolution datasets with minimal adjustments.

Generalization for Diverse Data: Our framework is flexible. As we mentioned in the discussion, the model can be extended to integrate more diverse data types, such as socioeconomic data (e.g., population density) and future Land Use Land Cover (LULC) predictions for more dynamic and contextual mapping.

4. Practical Applications – Include a brief example or case scenario illustrating how the models could be applied in a real-world construction project during execution.

 

>>> Thank you for the excellent question on practical applications. Here the example case scenario:

Scenario: A road construction project in Demak.

Our flood susceptibility map would serve as a dynamic risk management tool. When integrated with short-term precipitation forecasts, the system could issue an alert for a specific high-risk construction zone. This would enable the site manager to proactively:

• Move critical equipment and materials to safer, low-risk zones.
• Reschedule weather-sensitive tasks like concrete pouring.
• Install temporary flood barriers.

This transforms our model from a strategic planning tool into an operational decision-support system, helping to prevent project delays and financial losses during execution. Thank you for this valuable suggestion.

 

5. Reference Formatting and Accessibility – Ensure all citations follow a consistent format, add missing DOIs where available, and provide English translations for non-English titles (e.g., Croatian references).

 

>>> Thank you for your attention to the formatting details of our references.

We have reviewed our entire reference list to ensure the following:

• Consistent Format: All citations have been consistently formatted according to the journal's guidelines.
• DOIs: We have verified that DOIs are included for all available references.
• Non-English Titles: We would like to clarify that all sources cited in our manuscript are already in English; therefore, no titles require translation.

Thank you again for your thoroughness

Reviewer 2 Report

Comments and Suggestions for Authors

All recommendations were met by the authors.

Author Response

Dear Reviewer 2,

Thank you very much for your former comments in improving the quality of our manuscript, and the supporting of the revision.

Your sincerely,

Corresponding author