Analysis of Land Use/Cover Changes and Driving Forces in a Typical Subtropical Region of South Africa

Round 1

Reviewer 1 Report

The article presents an analysis of land use/cover (LUCC) change in KwaZulu‐Natal Province and Mpumalanga Province, Eastern South Africa, spanning from 1995 to 2020, using three classification methods. The driving factors behind LUCC was also analysis. While the article is comprehensive in content and well-structured, it lacks innovation in the field of remote sensing science. The methods employed are traditional, and the analysis does not yield significant new findings. It appears that the article primarily utilizes remote sensing data as a foundation for analysis. It is recommended that the article reduces its emphasis on conventional data processing and delves deeper into the analysis of LUCC driving factors. Then, it is suggested that the article be submitted to other relevant journals.

Comments for author File: Comments.docx

Moderate editing of English language required

Author Response

Dear Mr./Ms. Reviewer,

Thanks for your valuable suggestions and comments on our manuscript, and we highly appreciated your professional level. We have revised the manuscript following your guide and now uploaded a response letter as an attachment. Hope to hear your feedback soon.

Best wishes to you!

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Reviewer 2 Report

Please see the attachment.

Comments for author File: Comments.pdf

Author Response

Dear Mr./Ms. Reviewer,

Thanks for your valuable suggestions and comments on our manuscript, and we highly appreciated your professional level. We have revised the manuscript following your guide and now uploaded a response letter as an attachment. Hope to hear your feedback soon.

Best wishes to you!

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript produced LULC maps of KwaZulu‐Natal Province and Mpumalanga Province in eastern South Africa every five years from 1995 to 2020 using different classification schemes, and explored the driving factors of LUCC of this region by using the optimal parameter‐based geodetector model by using the LULC maps with the highest classification accuracy.

This manuscript is not innovative enough, but considering its potential contribution to new understanding of human-land relations in the region, it is recommended to be accepted after major revision.

Major problems:

1.The overall structure of this manuscript is oriented to method, but the title is oriented to remote sensing regional application, there is a mismatch. The structure of application-oriented articles is mostly ‘Introduction, Research Area and Data, Methods, Conclusions’, while the structure of method-oriented articles is mostly ‘Introduction, Materials and Methods, Experimental Results’.

2.The transition between the introduction paragraphs is abrupt and gives a sense of patchwork. It is recommended to first address the contradiction between human and land in the region, then introduce LUCC and LULC, then introduce technical issues, and finally compare and evaluate different technologies based on the needs of this region.

3.The title still needs to be considered, and some titles do not completely match the content. For example, the primary heading of section II is materials and methods, but the secondary heading is study areas and data. The pixel-based classification in section 2.3.3 mainly introduces classifiers, while the object-based classification in section 2.3.4 not only introduces the concept and parameters of object-based classification, but also includes classifiers.

4.The existing classification schemes cannot explain the performance difference of different classification methods in classification accuracy. Due to different features used, it cannot be explained that the improvement in classification accuracy is brought by object-oriented strategy, resulting in insufficient evidence to prove the relevant arguments in section 4.1.

Minor problems:

1. Data category partitioning is not accurate enough. The MOD21A1D LST data is also a kind of remote sensing data.

2. Some sentences are too absolute and not accurate enough. For example, the Driving Force Analysis in the first sentence of the second paragraph should be a method that can be used for ‘exploring the intrinsic mechanisms …’, not just for this purpose as you described.

3.The abbreviation of OPGD on line 233 should be given in the first time it appears on line 114.

4.The few examples listed to illustrate the diversity of classification systems are not representative, and more authoritative classification systems such as IGBP and FAO are recommended.

5.In addition to the number of samples, it is best to give the spatial distribution of all samples.

6. It is not enough to give only OA and KAPPA in the accuracy evaluation section. It is suggested to give the confusion matrix of each classification in the form of attachment.

The writing level of the article is good, making it both easy to read and understand.

Author Response

Dear Mr./Ms. Reviewer,

Thanks for your valuable suggestions and comments on our manuscript, and we highly appreciated your professional level. We have revised the manuscript following your guide and now uploaded a response letter as an attachment. Hope to hear your feedback soon.

Best wishes to you!

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Reviewer 4 Report

This study aims at making an effective land use and land cover (LULC) classification for the case area. Besides, it explores the trends in time for Land use/cover change (LULC), and analyze the driving factors of Land use/cover change (LUCC).

Three different classification methods were employed where the object-based classifier outperformed the other methods.  The LULC produced by the most effective method (OBIA) was taken as the basic data, and the optimal parameter‐based geodetector (OPGD) model was used to analyze the driving factors of LUCC in the region. Land use and land cover (LULC) data every five years including six series between 25-years period (1995-2020) were explored through the trends of change of LULC.

Major Comments

About classification method:

One of the classification methods being compared for its effectiveness in the present study is an unsupervised method. It is very obvious that this method will produce low accuracy output compared to that of supervised methods that employ a training procedure. Therefore, I find it unnecessary and useless to include K-means clustering into the classification for comparisons. Authors if they need to make a sound comparison of classification methods, they should be making it amongst the supervised classifiers that employ the same training set or they should shift to unsupervised classifiers and make comparisons among unsupervised classifiers. Here it is important for the authors what they want to emphasize. Do they compare the object based vs state of the art pixel-based classifiers? A comparison between most widely used classification methods for LULC in the literature might be a reasonable means to shift towards.

Authors claim that they explore driving factors of LUCC. However, in the exploration part, what they did was exploring driving factors and LULC. There is no relationship demonstrated between driving factors and “change” in particular.

Sampling:

Sampling method for supervised classification is obscured. More information should be provided about the sampling method.

R168: “The number of samples for different LULC types was determined by their area ratio to the total area of the study area [29].

This is called a “stratified sampling” and authors should denote this. Authors should also address whether this is a stratified random sampling or not. If the sampling was not random, than they should be explaining and verifying their sample selection method. They should also address if this is a pixel by pixel collection or a collection as group of pixels using a polygon selection tool.

There is also a testing. Selection of test samples is as important as the selection of samples. Did the authors collect all the samples (a total of 7000) and then based on a ratio 5000/2000 divided 2000 of them randomly out of 7000 for testing? How did the authors select testing samples? A biased collection would change the accuracy results. These are very unclear.

Sampling for object-based classification is obscured as well.

R226: How do the authors collect a total of 5000 objects as training samples? The object-based classification comes after image segmentation procedure. Then do the authors still use pixel-based training samples and how?

How do they involve additional features (the features that they did not use in Random forest classification such as GLCM, shape, size) as training samples into classification?

Data and materials:

Features used in classification:

Principal component analysis (PCA) was employed for the eight texture features and first two principal components that contained more than 90% of the texture information were added as input feature to classification.

I suggest that authors provide PCA parameters (any rotation? etc.) and interpret the components based on their correlated factors. Authors can provide a “component matrix” for this.

Dependent independent variables

R280: In this study, land use was divided into six classes. Drawing on past research results, the grading indices for farmland, forest, grassland, water, constructed land, and unused land were 3, 2, 2, 2, 4, and 1, respectively [38].

Reviewer: The grading index seemingly has a great influence on the outcome of the “Land use intensity”. Therefore, it seems quite important. What does this grade refer to? What is its meaning?

Authors should verify the grades based on the similar studies in the literature or a sound method explained in the text. A single study by the authors of the present manuscript is not very convincing.

R287: In line with previous research, the coefficients of impact intensity were obtained by averaging the values from the Lohani checklist method, the Leopold matrix method, and the Delphi method.

Reviewer: Authors provide no explanation of any of these methods, their pros/cons. Why do authors not employ one of the methods with advantages, but take the average of their outputs as a whole?

Driving factors relationships:

R 403: Notably, in 2015, the interaction between nighttime light intensity and elevation showed the strongest explanatory power, reaching 0.743.”

Reviewer: These kind of statements needs more interpretation. Authors should refer to the nature or the mechanisms that produce this relationship. For instance, is this the case? “As the elevation rises, the urbanization tends to decrease, so that the LULC relevant with human interference?”

Reviewer: Doesn’t the OPGD give the direction of association, i.e. positively or negatively. E.g is elevation negatively related with nighttime light intensity, or pop density as an indicator of urbanization. It may be useful to mention the direction of the relationship.

Minor comments:

R161: “Regional LULC can be divided into various classification systems depending on the application objectives.”

Reviewer: It is not the regional LULC that is divided into classification systems. Instead: “Regional LULC can be divided based on various …”  better explains the statement.

R92 “This model focuses on the spatial perspective,”

Reviewer: Does the OPGD focusses on spatial perspective? How? I do not see a clue based on the equations 1, and 2.

R180: “Four spectral index features, namely, NDVI, MNDWI, NDBI, and BSI, were generated based on the imagery of each year.”

Reviewer: the extended names for the abbreviations and the equations of indices should be provided.

R183:

Reviewer:  It is understood that all the features above (spectral and topographic) formed the feature combination used for pixel‐based classification. Object-based classification uses further features, eg. geometric, and size and shape. But authors give this information in text which is not easy to capture. Please provide a table that give the data sets used for each classification method.

R221: “Object‐oriented classification uses image segmentation technology to divide all pixels in remote sensing images into globally covered and closely connected objects based on feature similarity and then conducts sample collection and classification for each object.

Reviewer: This sentence is not explanatory and confusing.

“group” all pixels rather than “divide all pixels”

What do the authors mean by “globally”?

“Spectral/spatial similarity” rather than “feature similarity”

“… then conducts sample collection and classification”? authors should describe this in detail in sample selection section.

R312 “As shown in Figure 3, farmlands were distributed mainly in the southeastern coastal areas and the central‐northern plateau regions, grasslands were located primarily in Kruger National Park…

Reviewer: I do not see any spatial references in Figure 3 as mentioned in the text.

R338: … optimal grouping methods and breakpoint numbers for different independent variables and the same independent variable in different years, revealing the spatial and temporal differentiation of the independent variables.

Reviewer: Especially differing breakpoint numbers for the same independent variable in different year. This should be making the discretization, and hence OPGD “q-value” for year (5-year) not compatible. How do you explain this?

R512 which met the requirements of subsequent research.

Reviewer: Authors mention “subsequent research” throughout the text several times. Subsequent has a meaning like following. Is it what they mean? Or do they mean relevant literature? Then they should refer to that literature.

R531:

Reviewer: When analyzing the factors affecting land use intensity, some factors, such as road network density, road area, and distance from roads, are missing.

Reviewer: Road data is a quite available; particularly in open data sources recently. Why do the authors mention this as a lack, and how do they verify this?

Can be improved.

Author Response

Dear Mr./Ms. Reviewer,

Thanks for your valuable suggestions and comments on our manuscript, and we highly appreciated your professional level. We have revised the manuscript following your guide and now uploaded a response letter as an attachment. Hope to hear your feedback soon.

Best wishes to you!

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The whole structure of this revision is complete and the analysis is profound. However, the remote sensing methods used are traditional and do not provide academic value in the field of remote sensing. It is advisable to reduce the focus on remote sensing image interpretation and instead concentrate on analyzing the driving factors of land use change. Suggest submissions to “Sustainability ”or “Land”.

 Minor editing of English language required

Author Response

Dear reviewer, 

Thank you for your valuable feedback, and the response letter is uploaded as attachment.

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript is sufficiently improved. 

Author Response

Dear reviewer, 

Thank you for your valuable feedback, and the response letter is uploaded as attachment.

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Reviewer 3 Report

All problems raised in the first round of review have been appropriately addressed in the revision version. Therefore, it is considered that the revised version meets the standards for publication.

Author Response

Dear reviewer, 

Thank you for your valuable feedback, and the response letter is uploaded as attachment.

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf

Reviewer 4 Report

Authors have revised the manuscript fairly enough in line with my comments and suggestions.

Author Response

Dear reviewer, 

Thank you for your valuable feedback, and the response letter is uploaded as attachment.

Sincerely,

Sikai Wang

Author Response File: Author Response.pdf