Combining Machine Learning Models and Satellite Data of an Extreme Flood Event for Flood Susceptibility Mapping
, Ioannis Benekos 2, Theano Iliopoulou 1, Panayiotis Dimitriadis 1 and Demetris Koutsoyiannis 1,*
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe article bearing the title” Combining Machine Learning Models and Satellite Data of an Extreme Flood Event for Flood Susceptibility Mapping” is very interesting for the readers but is required major revisions as given below:
There is need to rewrite the abstract with quantitively way
What is novelty of the article, author just compare the different machine learning model
There is also needed to revise the introduction, 148- 154 lines are unnecessary
There is also needed to draw flow chart of the methodology
Materials and methods comprised on the study area instead of results
Results and discussions are given comparison, no quantitively description is given, please rewrite with quantitively description instead descriptive
Also suggest to incorporate the results in hydrological model and check its improvement in flood mapping as well as consider socio economic aspect
Author Response
Please see attached.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe present study focuses on combining machine learning models and satellite data to map flood susceptibility in the event of an extreme flood. The authors offer the following observations for consideration:
1) Add details about the sample dataset, including the flood and non-flood samples for the test and training sets, in a table.
2) Add the source of the non-flood samples.
3) Visualize the location of the flood and non-flood samples on a map.
4) Enhance the visualization of Figure 7.
5) Enhance the quality of figures such as Figure 8. The font sizes are very small.
6) Add a coordinate system using latitude and longitude.
7) Use “Altitude” or “Elevation” rather than “DEM”.
Minor Comments:
I recommend a few additional articles that may strengthen your introduction. If you find them relevant, feel free to incorporate them; otherwise, you may disregard them.
doi.org/10.1016/j.jhydrol.2025.133553
Author Response
Please see attached.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis article presents a critical review of the application of artificial intelligence (AI) in urban flood prediction over the past two decades. The review highlights the limitations of traditional physically-based hydrodynamic models—such as high computational demands and the need for high-resolution data—and explores how AI techniques have been increasingly used to enhance forecasting accuracy, speed, and scalability.
The article is well-written and clear. However, the section that requires more detail is the determination of the flood susceptibility map for a 1000-year return period. How was the classification of flood susceptibility into low, medium, and high categories performed?
Author Response
Please see attached.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThis study clearly highlights the practical value of applying machine learning techniques to flood susceptibility mapping in the Thessaly region, an area historically prone to severe flood events. By leveraging environmental, topographic, and hydrological data-particularly satellite information related to Storm Daniel-the research has developed predictive models that have achieved particularly high levels of accuracy.
While acknowledging the potential of the proposed approach, it is useful to raise some critical considerations that can help better interpret the results and guide future developments.
Despite the promising performance of the Random Forest and XGBoost models, which achieved AUC values close to 0.97it is important to question some methodological issues that may have influenced, and in part inflated, these parameters. This reflection is essential to fully understand the true potential of the approach and outline improvement strategies.
A first aspect concerns the construction of the dataset: randomly selecting 3 950 points while maintaining a perfect balance between flooded and non-flooded areas certainly simplified the classification task, improving the performance of the models. However, this ideal balance does not reflect the real spatial distribution of the territory, where the non-flooded areas are far more numerous. As a result, the model runs the risk of being overtrained on a “perfect” case study, returning overly optimistic estimates that may not be confirmed in operational applications characterized by a strong imbalance between classes.
A second point of criticism concerns the relatively small sample size, modest by the standards required by machine learning models for good generalization. In addition, the choice to focus the analysis on a single extreme event (Daniel storm) could lead the model to capture patterns specific to that event, rather than general regularities of flood dynamics. To strengthen the robustness and adaptability of the model, it would be appropriate to include data from multiple historical events, even with different intensities and spatial distributions.
An additional element concerns the interpretability of the results. Why was it used? What is the value added? Why were those feature measures used? Why were not others used? The potential of such techniques is well known, but it should be contextualized by mentioning what has been done in the literature on the subject matter (Early Warning System). Finally, to ensure transparency and reproducibility, it would be desirable to make available the dataset used. This would not only allow validation of the results obtained, but also facilitate comparisons and future development by the scientific community.
In summary, the study is an important contribution, but it needs to be further enhanced by improving the construction of the dataset, expanding the number and variety of events considered, deepening the critical analysis of interpretability, and promoting data sharing for reproducibility.
Author Response
Please see attached.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe research article seems better version as compared to older version but still need to improve with minor revison.
Always data collection and study area before the methods, section 3 can be keep before section 2.
there is no need to line number 658 to 660, it is interdispilinary work.
the discussion section need to compare different flood modelling instead of general discussions, author can also compare his own work with previous results with similar environment.
it is requested to improve with minor revison.
Comments on the Quality of English Languagei am not native in english, it's better to improve the english native
Author Response
Please see attached file
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Authors,
Thank you for addressing the reviewer comments. The paper can be accepted in its current form.
Regards,
Reviewer
Author Response
Please see attached file
Author Response File: Author Response.pdf
