LUCC Simulation Based on RF-CNN-LSTM-CA Model with High-Quality Seed Selection Iterative Algorithm

Round 1

Reviewer 1 Report

The abstract should be revised. At the beginning, a phrase must be introduced that puts the entire research in the international context to highlight the importance of the study.

The results obtained are directly presented in the abstract, some methodological points should also be introduced. A lot of abbreviations are also used in the abstract (it would be useful if each abbreviation were replaced with the full name, in the first instance).

The introduction of the graphic abstract is to be appreciated

It is worth noting that the authors include in the text a table summarizing the databases used. Figure 2 must be redone, in its current format it does not highlight the particularities of the database. Legends should be introduced to each database to highlight it and at the same time a color palette should be chosen in such a way that specific values can be differentiated.

The methodology followed is well highlighted, an added value is given by its schematization within a very suggestive workflow.

The conclusions part should be improved

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper compares several deep-learning and machine-learning inspired models for predicting land-use change in a major metropolitan area within the country of China. The authors conclude that hybrid methods that combine machine learning and deep learning approaches achieve better accuracy than the state-of-the-art approaches for predicting land use change.

One major issue with the paper is the way in which the model results are organized. There are several accuracy comparisons between pairs of models scattered throughout, but at no point are all the model results contained in once place. I finished reading not really knowing which of the author’s proposed models were truly best.

Further, the paper lacks sufficient detail about how exactly the models are being trained. Which years are used for training and which for testing? If each year except the first is used for testing, then this should be clarified and accuracy should be tracked across years.

It is also interesting to me that the increases in the kappa coefficient seem to be mostly trivial when compared to previous methods. Occasionally, the increases in the FOM are more substantial (10+%), which begs the question why these models only show appreciable improvements on one of the two metrics. A better explanation of the seemingly contradictory results should be provided in the text. 

Line/section/figure-specific comments:

Section 1: The addition of one or two paragraphs at the beginning of the introduction to better define LUCC terminology would be helpful. What exactly do you mean by “neighborhood” and “conversion rules”? I think I know but want to make sure we are precisely on the same page.

Redefine all acronyms in the text. Don’t rely on the acronym definitions in the abstract.

Line 29: “LUCC has become a powerful tool to study the process of LUCC” doesn’t make any sense because LUCC can’t be the tool to study itself. Please clarify. Similarly,
“LUCC suitability of LUCC” does not make sense on Line 214.

Line 106: Please define “GDP” acronym.

Equation (7) Note in the equation that the sum iterates on m, not k. That is evident from the text but should also be evident in the actual equation.

Line 237-240: It would be helpful if the authors would provide a more formal definition of the kappa coefficient in the text.

Line 240: It seems as though an “acceptable” kappa value is problem specific. Nowhere in the reference you cite [32] do they demonstrate why 0.7 is an acceptable cutoff. That cutoff needs better justification in the text or the claim needs to be omitted from the paper.

Line 249-253: Capitalize “B”, “C” and “D” in the text.

 Table 2: The authors not a change in the number of pixels in the study region but provide no explanation as to why. That explanation should be provided. It also calls into question how accuracy is being measured, since measure changes in a pixel state rely on those pixels being constant over time.

Figure 9: How is variable importance measured. Is Random Forest permutation variable importance? This needs to be clarified in the text and in the caption.

Figure 10/Table 6: It is not clear in the text how land use can continually expand in all categories. My assumption was the amount of land was fixed and its use was changing over time. This would result in a “zero-sum game” where gains in one land class must be balanced by losses in another. Please clarify somewhere in the text why the total number of land keeps changing throughout the paper.

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

It is a good piece of work with many interesting innovations. There are some typo's which make it difficult for the reader to access how much the innovations have improved results and I suggest that these are improved. The figure and table captions also need to be more informative. I feel strongly that the authors should make both the code and output data available through a dynamic link in text - I believe this is the way forward to help researchers from less well resourced locations and to improve transparency. This is particularly so for research that is state funded, as this article is. A more appropriate reference on the values of Kappa that represent good results are required - the current reference cited is either an error or just inappropriate as it does not discuss appropriate levels of Kappa or provide discussions and comparisons that are relevant to the matter.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

This manuscript requires major English language editing.

Specific comments to authors.

Line 29-30: The use of LUCC twice in this statement introduces confusion.

Line 32: Consider deleting, 'in the world....'

Line 33: What changes the cellular state? This is not clear

Line 55: Not proper to start with 'And..'

Line 59: Year of publication is missing

Line 61: Year of publication is missing

Line 78-87: Consider re-writing this part. It is not clear

Line 93: Consider changing 'China's inland' to 'Inland China'

Line 94-95: Consider deleting this sentence

Line 100: Add space between 'areas' and 'Figure'

Line 105: change to 'soil type..'

Line 120: Add the source of ASTER data

Line 124: This part of the sentence seems misplaced

Line 125: Consider changing the sentence to, 'Population data were extracted from

Line 133: What is the spatial resolution?

Line 134: Remove capital letter

Figure 1: Please include the legends for the raster data

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I appreciate the authors' well-crafted response to revision and thought that they generally did a decent job responding to my line-by-line comments. I agree that the new paper format is easier to read than the last. 

I originally included the comments in the editor-only response and I assumed that the editor would raise those editor-only issues with the authors: "it is not clear from the reported results that the increases in accuracy, at least for the kappa values, are worth adopting the new approach."

That issue remains unchanged and there is no way for the authors to address that without completely revamping the study, which is why I made it an "editor only" comment. By providing only a single year (2015 to 2020) of accuracy results in a single locale, there is no way to know if these seemingly small accuracy improvements are circumstantial or represent a worthwhile step forward in how we model land use change. Some comparison to traditional land use modeling approaches, rather than "degraded" versions of other machine learning model approaches would need to be included. This is not something I expect the authors to directly address in a review response, as it involves a serious change in their study design. Thus, I would recommend that this paper be "rejected" in its current form and/or you find another reviewer to provide an alternative view. 

To be clear, the writing is much improved and the response to review was good, but I still maintain that the papers conclusions are not substantial enough to warrant publication in this venue.  

Author Response

Please see the attachment.

Author Response File: Author Response.docx