Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study applied the CA-MLP-ANN algorithm to predict LULC changes, aiming to clarify the current status of built-up areas and forecast the urban development trajectory in the X22J catchment. However, the manuscript requires substantial revisions.

The abstract needs significant modification. The background information is excessive, while the description of research methods is insufficient. The emphasized points in the conclusion lack direct relevance to the findings of this study.

The focus is not sufficiently narrowed to the core issues of the current research. The first and third paragraphs are overly lengthy and should be condensed and integrated. The second paragraph would be more appropriately placed as the penultimate paragraph. The research gap stated in the final paragraph is unclear, and the last two sentences fail to effectively highlight the existing problems and the innovation of this study.

There are several issues with Table 1. The first column lacks a heading. The headings and content of the third, fourth, and fifth columns are wrong. Furthermore, the sources of various data (e.g., institutions, URLs) are not specified. The uncertainty associated with the various data is also not addressed, which could significantly impact subsequent analyses. pLease refer to:

“Examining the reliable trend of global urban land use efficiency from 1985 to 2020 using robust indicators and analysis tools” and “How well do we really know the world? Uncertainty in GIScience.”

The description of GEE is excessive and unnecessary in section 2.3.

The content of Table 2 constitutes common knowledge and is unnecessary.

This section of 2.4 is missing.

The section of 2.5.1 is titled "Spatial Variable Data," but the content primarily discusses MOLUSCE.

The unit for slope in the legend of Figure 3 is meters (m), whereas it is typically expressed in degrees. Please clarify this.

The sample sizes used for accuracy validation in Tables 3-5  are too small.

Sections 3.1.1 and 3.1.3: These sections share the same title. Furthermore, Section 3.1.2 appears to be missing.

Please explain the practical significance of a prediction Kappa coefficient of 0.52 and an overall accuracy score of 0.75 for guiding future work.

Some paragraphs in the Discussion section are not closely related to the research presented in this paper.

The latter part of the Conclusion section actually contains content more suitable for the Discussion, rather than stating the research conclusions of this study.

Author Response

Comment: This study applied the CA-MLP-ANN algorithm to predict LULC changes, aiming to clarify the current status of built-up areas and forecast the urban development trajectory in the X22J catchment. However, the manuscript requires substantial revisions.

Response: The authors are grateful to the reviewer for taking the time to review the manuscript, and they hope they have thoroughly addressed all the comments and concerns raised.

Comment: The abstract needs significant modification. The background information is excessive, while the description of research methods is insufficient. The emphasized points in the conclusion lack direct relevance to the findings of this study.

Response: Thank you for pointing this out. The authors have revised the abstract by reducing the background information and adding a bit more details of the methodology, results and the major conclusion in terms of urban growth

Comment: The focus is not sufficiently narrowed to the core issues of the current research. The first and third paragraphs are overly lengthy and should be condensed and integrated. The second paragraph would be more appropriately placed as the penultimate paragraph. The research gap stated in the final paragraph is unclear, and the last two sentences fail to effectively highlight the existing problems and the innovation of this study.

Response: The entire introduction has been revised as per the reviewer’s comment; it should be however noted that the length of the section was drafted to provide sufficient background and rationale to the identified problem.

Comment: There are several issues with Table 1. The first column lacks a heading. The headings and content of the third, fourth, and fifth columns are wrong. Furthermore, the sources of various data (e.g., institutions, URLs) are not specified.

Response: The authors agree with the comment, and the table has been corrected with added URLs and Institutions

Comment: The uncertainty associated with the various data is also not addressed, which could significantly impact subsequent analyses. Please refer to:

• • “Examining the reliable trend of global urban land use efficiency from 1985 to 2020 using robust indicators and analysis tools” and “How well do we really know the world? Uncertainty in GIScience.”

Response: Agreed. The authors have attempted to address the comment in lines 153 - 155 and Table 2. The authors further added another spatial variable, i.e, (Distance to City), and a recommendation has been added in Section 5 for future research to focus on spatial variable data and its quality assurance using more advanced methods and techniques.

Comment: The description of GEE is excessive and unnecessary in section 2.3.

Response:  Thank you for pointing this out. The description has been reduced, leaving the reason for selecting GEE as the analysis software in this study

Comment: The content of Table 2 constitutes common knowledge and is unnecessary.

Response: The author acknowledges the reviewer’s comment and therefore the table has been removed as per the reviewer’s comment, and the land use classes of interest in the study have been infused in the last line of the last paragraph of section 2.3. Please see line 201-203

Comment: This section of 2.4 is missing.
Response: Agreed. The numbering has been revised throughout the entire manuscript.

Comment: The section of 2.5.1 is titled "Spatial Variable Data," but the content primarily discusses MOLUSCE.
Response:  The authors agree with this comment, and so this has been addressed with some information being moved to 2.4.2.

Comment: The unit for slope in the legend of Figure 3 is meters (m), whereas it is typically expressed in degrees. Please clarify this.
Response: Initially, the authors reprojected the slope layer and used the units indicated under "information" in the layer property; however, this has now been corrected to degrees as per the reviewer’s indication

Comment: The sample sizes used for accuracy validation in Tables 3-5 are too small.
Response: The authors agree with the comment and therefore, the sample numbers have been increased to 112, 122 and 207  for 1990, 2007 and 2024, respectively

Comment: Sections 3.1.1 and 3.1.3: These sections share the same title. Furthermore, Section 3.1.2 appears to be missing.
Response:  Agreed. This has been addressed throughout the document

Comment: Please explain the practical significance of a prediction Kappa coefficient of 0.52 and an overall accuracy score of 0.75 for guiding future work.
Response: The prediction kappa and overall accuracy mean that there is room for improvement; therefore, future studies could explore additional covariates or advanced modelling techniques that would yield higher accuracy results. This is also indicated under limitations in section 5

Comment: Some paragraphs in the Discussion section are not closely related to the research presented in this paper.
Response: The authors have removed some paragraphs that they believe the reviewer may be alluding to. However, it is worth noting that the aim was to present the potential impacts of unplanned urbanisation. Should the removed paragraphs not be the ones the reviewer was referring to, the authors request that the reviewer perhaps highlight this part, and they are more than willing to revise it.

Comment: The latter part of the Conclusion section actually contains content more suitable for the Discussion, rather than stating the research conclusions of this study.
Response: The authors have attempted to rewrite the conclusion with more emphasis on the research conclusions

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,

The manuscript you provided is well-written and has the potential to offer valuable insights for urban and environmental planners, contributing to the existing literature in this field. Here are specific comments:

Abstract vs. Results: In the abstract, you report a Kappa coefficient of 0.52 and an overall correctness of 74 %, while later in the results, you mention “an overall score of 0.75”. Please clarify which values correspond to the Kappa coefficient and which to the overall accuracy (or another metric), and make sure the terminology and numbers are consistent throughout the manuscript.

Line 185 – It is not clear why in the workflow you mention Predicted 2020 map. Is it a mistake? Shouldn't it be 2024?

Lines L192–L198: Please revise “Landsat 4-5-TM and Sentinel-1-2” to clearly indicate which sensors were actually used for optical classification

Line 205-208 The statement “it doesn’t require pre-processing because everything can be done within the same platform” is too strong. If possible, specify which pre-processing steps were actually applied in GEE (cloud masking, median composites, resampling, etc.) rather than implying that no pre-processing was necessary. It is probably done in GEE.

Input variables and resolutions (Table 1): Please provide the full name “OpenStreetMap” and explain how the “Distance to road” and “Population density” maps at 30 m resolution were derived (source, resampling, projection).

Parameters in MOLUSCE: Please justify the choice of the key model parameters — 1,000 iterations, five hidden layers, learning rate 0.01, and momentum 0.05 — and also specify the value and type of the kernel (n×n) used in the neighbourhood analysis. In addition, please report the variable-importance or usefulness metric produced in the “Evaluating Correlation” step so that readers can understand which drivers most influenced the simulation results.

Line 221 – “Only images with cloud cover less than 0 were selected” is physically impossible. Please correct the threshold (e.g., < 10 % cloud cover) or clarify the actual criterion used for image selection.

Figure 5 – 1990&2024 not 2023

Line 422 – 1999???

Author Response

Comment: Dear authors,
The manuscript you provided is well-written and has the potential to offer valuable insights for urban and environmental planners, contributing to the existing literature in this field. Here are specific comments:

Response: Thank you for taking the time to review the manuscript and for the vote of confidence in the research. It means a lot to us. We have hope that we have thoroughly addressed all your comments and concerns. We really appreciate your input.

Comments: Abstract vs. Results: In the abstract, you report a Kappa coefficient of 0.52 and an overall correctness of 74 %, while later in the results, you mention “an overall score of 0.75”. Please clarify which values correspond to the Kappa coefficient and which to the overall accuracy (or another metric), and make sure the terminology and numbers are consistent throughout the manuscript.
Response: Duly noted, and we have corrected the errors as well as proofread the entire manuscript for any inconsistencies.

Comment: Line 185 – It is not clear why, in the workflow, you mention the Predicted 2020 map. Is it a mistake? Shouldn't it be 2024?
Response:  Thank you for pointing this out. That was a simple mistake, and it has been corrected

Comment: Lines L192–L198: Please revise “Landsat 4-5-TM and Sentinel-1-2” to clearly indicate which sensors were actually used for optical classification
Response: Agreed. This has been corrected as per the reviewer’s suggestion

Comment: Line 205-208 The statement “it doesn’t require pre-processing because everything can be done within the same platform” is too strong. If possible, specify which pre-processing steps were actually applied in GEE (cloud masking, median composites, resampling, etc.) rather than implying that no pre-processing was necessary. It is probably done in GEE.
Response: The authors agree with the comment; therefore, the statement has been replaced with some of the preprocessing, processing and post-processing tasks carried out during the classification.

Comment: Input variables and resolutions (Table 1): Please provide the full name “OpenStreetMap” and explain how the “Distance to road” and “Population density” maps at 30 m resolution were derived (source, resampling, projection).
Response: Agreed. This comment has been addressed in Table 1, and lines 220 – 223 further explain how the 30 m resolution was derived and the projection applied.

Comment: Parameters in MOLUSCE: Please justify the choice of the key model parameters — 1,000 iterations, five hidden layers, learning rate 0.01, and momentum 0.05 — and also specify the value and type of the kernel (n×n) used in the neighbourhood analysis. In addition, please report the variable-importance or usefulness metric produced in the “Evaluating Correlation” step so that readers can understand which drivers most influenced the simulation results.
Response: The reason for the key model has been stated, the kernel value has been specified in line 254, and Table 2 presents the produced "Evaluating correlations"

Comment: Line 221 – “Only images with cloud cover less than 0 were selected” is physically impossible. Please correct the threshold (e.g., < 10 % cloud cover) or clarify the actual criterion used for image selection.
Response: Agreed. This has been corrected to <1 because that is the threshold used in the model

Comment: Figure 5 – 1990&2024 not 2023
Response: Agreed. It is 2024, and it has been corrected accordingly

Comment: Line 422 – 1999???
Response:  Thank you for pointing this out. The error has been addressed, and the correct value is 1990.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The uncertainty associated with the various datasets remains inadequately addressed. We strongly recommend consulting the suggested references. 

Author Response

Comment: The uncertainty associated with the various datasets remains inadequately addressed. We strongly recommend consulting the suggested references

Response: Thank you for the comment, and we agree. We have tried to address this in lines 207 - 211

For land use prediction, Uncertainty in MOLUSCE is addressed by producing certainty maps, incorporating multiple spatial drivers, using robust sampling and multiple modelling algorithms, running iterative simulations, and validating predictions with independent data. This also addresses overfitting in the simulation.

For classifying using random forest in GEE setting, the output mode of the Random Forest classifier was set to 'multiprobability' to allow it to generate a probability map for each class, reflecting the likelihood of each pixel belonging to that class. This probabilistic output captures the classifier's confidence, providing a continuous measure rather than just hard labels. This probabilistic output allows identification of ambiguous or uncertain areas with low maximum class probabilities, which can be quantified as uncertainty layers, e.g., 1−max(class probabilities). This method overcomes the limitations of traditional discrete classifications, which ignore confidence levels, and helps identify ambiguous areas that require further investigation. Coupling these uncertainty measures with standard validation metrics improves overall map reliability and addresses uncertainties.

We hope the given methodological description addresses the reviewer’s concerns.

Reviewer 2 Report

Comments and Suggestions for Authors

No other comments.

Author Response

Thank you so much for the review and contribution towards the quality of the manuscript.