Evaluating Machine Learning-Based Approaches in Land Subsidence Susceptibility Mapping

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I have carefully reviewed this manuscript, and overall, it holds significant practical relevance and offers considerable reference value for the prediction and protection against land subsidence. However, there are still some issues that need to be addressed. Here are my suggestions:

 

1. In Section 2.3, the author briefly introduces the characteristics, strengths, and weaknesses of seven machine learning algorithms. It would be more thematic and beneficial for interdisciplinary applications of machine learning algorithms if a targeted introduction to the advantages and disadvantages of each algorithm regarding their suitability for predicting land subsidence could be provided. This would aid fellow researchers in better understanding the application of these seven machine learning algorithms in the context of land subsidence prediction.

 

2. On line 155, the authors mentioned that the subsidence points were taken from sensitive areas, while non-sensitive areas are considered less likely to experience ground subsidence. The distinction between sensitive and non-sensitive areas is based on previous research. Please indicate the referenced literature or data sources here. Additionally, briefly explain how you used previous studies and literature to determine the division between sensitive and non-sensitive areas. It's important to note that the classification of areas into sensitive and non-sensitive can vary, as there are multiple opinions on this matter, and the results of such classifications can differ under different algorithmic conditions. Therefore, it's necessary for the author to clarify this point.

 

3. On line 432, it is noted that Table 6 is missing. Additionally, regarding the accuracy range, if the accuracy mentioned by the author is between 0.5% and 0.76%, then the credibility of the model's accuracy is questionable. Please verify if there has been an error.

 

4. Discussion on limitations of this research is missing.

 

5. For the Results section, two suggestions could make the paper more broadly relevant in practical terms:

(1) In presenting the prediction results of various machine learning algorithms, it is possible to include more detailed statistical analyses and performance metric comparisons, such as confusion matrices, Receiver Operating Characteristic (ROC) curves, and the Area Under the ROC Curve (AUC). These statistical tools and metrics can provide a more comprehensive performance evaluation, helping readers to more clearly understand the efficacy and limitations of different models in practical applications. For example, by comparing the precision, recall, and F1 scores of the models, one can more finely reveal each model's capability in predicting positive and negative class samples.

(2) Incorporating a discussion on the strengths and weaknesses of each model, along with their specific applicable scenarios, in the results analysis. This would not only add depth and breadth to the paper but also provide practical guidance for researchers looking to apply these models in similar fields in the future. For instance, although some models may perform excellently in terms of overall accuracy, they might require more computational resources or more complex parameter tuning. Providing such analysis can help decision-makers choose the most suitable model based on their resources and needs.

Author Response

We sincerely appreciate the valuable comments and suggestions from the reviewer. The thorough review helped immensely in shaping the manuscript. The suggestions and comments have been closely followed, and revisions have been made accordingly. The following are the comments by the reviewer, along with our summarized responses. For the reviewer's convenience, the revisions were implemented in the manuscript using the track change tool.

"Please see the attachment."

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

 This paper presents the comparison of several machine learning approaches to seek the cause or to predict the land subsidence from easily obtainable datasets. The importance of comparison among the machine learning methods is understandable. However, in my experience, the performance of machine learning is strongly dependent on the choice of the predictor variables. From this point of view, this paper lacks the explanation on how and why these predictor variables are selected for this study. Also, the change of performance when we remove some predictor variables would be very important information for us to choose a machine learning approach in other areas with less datasets. Without addressing these points, the value of comparison is unclear for me. Inversely, with these points, the novelty of this paper becomes clear.

 For the other point by point comments are as follows.

 

1. l.80-82. The extrapolation performance of the machine learning approaches is still questionable because of the possible overfitting in general. I’m not sure what kind of “climate change” is considered by the authors, but I think it is not easy to agree with this statement without evidence.

2. l.93. “reduction in GWL” may imply the drop of hydraulic head. I recommend the clear description.

3. Figure 2. The scale is hidden by the map in (a). The legend is overlapped on the map in (c ). The fonts of characters are not systematic.

4. Figure 3 and 4. The aspect ratio looks distorted. The fonts of characters are not systematic. I expect all the predictor variables.

5. Figure 7. I expect a cross plot of observation vs prediction to show them align on 1:1 line.

6. Figure 8 and 9. I don’t understand why this information is presented in this form. I recommend the table form.

Comments on the Quality of English Language

1. l.136. “The novel contributions of this work were” -> “The expected novel contributions of this work are”

2. l.142. “used comprised”?

Author Response

We sincerely appreciate the valuable comments and suggestions from the reviewer. The thorough review helped immensely in shaping the manuscript. The suggestions and comments have been closely followed, and revisions have been made accordingly. The following are the comments by the reviewer, along with our summarized responses. For the reviewer's convenience, the revisions were implemented in the manuscript using the track change tool.

"Please see the attachment."

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

 

“Evaluation of Seven Machine Learning-based Approaches in Land Subsidence Susceptibility Mapping with Huge Datasets” is an interesting article however, following points shall be incorporated for the improvement of the manuscript.

 

1-      Title: title is relatively long it mut be rephrased and simplified.

2-      Introduction: The significance of land subsidence shall be further elaborated by case histories particularly from the study area.

3-      Since MLR approach is an ongoing research topic in geotechnical and geological circles, one particular and well-studied MLR method can by adopted to discuss in depth aspects of LS. It is not clear why seven different approaches were adopted.

4-      Methodology can be explained in a few sentences figure 1 is not necessary at this level.

5-       “The observational data” used for groundwater levels shall be explained.

6-      If dewatering /abstraction is the main reason for LS what are the alternate water resources available for the local population.

7-      What are the recommendations for local government and community to limit groundwater utilization.

 

 

Comments on the Quality of English Language

A manuscript requires extensive proofreading, some of the statements were difficult to understand.

Too many brackets were used in the text, a bit of rephrasing can adjust this problem.

Author Response

We sincerely appreciate the valuable comments and suggestions from the reviewer. The thorough review helped immensely in shaping the manuscript. The suggestions and comments have been closely followed, and revisions have been made accordingly. The following are the comments by the reviewer, along with our summarized responses. For the reviewer's convenience, the revisions were implemented in the manuscript using the track change tool.

"Please see the attachment."

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

With possible effects on infrastructure, the environment, and the economy, land subsidence is a serious natural hazard in both the Semnan Plain and the Kashmar Plain. To investigate land subsidence in these areas, this study used six classification-based MLAs (BRT, RF, SVM, BLR, CART, and LogR) and MLR, taking into account several variables, including lithology, slope, aspect, TWI, distance from the river, distance from the fault, land use, NDVI, and groundwater level.

 

Both study areas showed that the BRT approach accurately predicted the risks of land subsidence, and both RF and SVM performed well. In contrast, the least effective method was the LogR technique. The study demonstrated that lithology, distance from the river, NDVI, land use, and groundwater decline—mainly due to over-abstraction—contribute to land subsidence.

 

The potential of deep learning techniques should be investigated in future research, even though this study concentrated on classification-based MLAs and MLRs. In addition, future prediction models should consider variables like soil type, curvature, altitude, sedimentation rate, and others that were overlooked in this study to produce more accurate results. Future research on management scenarios to improve area resilience and lessen the negative effects of land subsidence on the environment and economy can build on the findings.

To ensure transparency and accuracy in your research, it is crucial to discuss the limitations of your study. This section should explicitly address all potential sources of bias, assumptions made during the modeling process, data limitations, and any other constraints that may have affected your findings. By acknowledging these limitations, you can demonstrate the credibility of your research and help readers interpret your results with greater confidence.

Clarify how findings can be applied by policymakers, urban planners, or environmental managers, adding a practical dimension to your research.

It's important to provide clear and specific recommendations for researchers who want to build on or replicate your work. Be assertive in suggesting particular regions or conditions where your models are most applicable and areas where further validation is necessary.

 

If possible, please indicate the accessibility of your dataset or provide information on where interested researchers can access the data. This promotes transparency and encourages reproducibility in scientific research.

 

Comments on the Quality of English Language

Should be look over, especially in long paragraphs.

Author Response

We sincerely appreciate the valuable comments and suggestions from the reviewer. The thorough review helped immensely in shaping the manuscript. The suggestions and comments have been closely followed, and revisions have been made accordingly. The following are the comments by the reviewer, along with our summarized responses. For the reviewer's convenience, the revisions were implemented in the manuscript using the track change tool.

"Please see the attachment."

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All my comments have been adequately addressed.

Reviewer 2 Report

Comments and Suggestions for Authors

I think the manuscript was well revised.

Reviewer 3 Report

Comments and Suggestions for Authors

The Paper has been improved and can be accepted after a final proofread

Comments on the Quality of English Language

The manuscript needs a final proofread