Data Gap Filling Using Cloud-Based Distributed Markov Chain Cellular Automata Framework for Land Use and Land Cover Change Analysis: Inner Mongolia as a Case Study

Round 1

Reviewer 1 Report

This manuscript used the cloud-based distributed Markov Chain Cellular-Automata framework for simulation to fill data gap for land use/land cover (LULC) change analysis.  The research topic is of importance. Nevertheless, the paper has the following problems that should be addressed.

(1) This study used many data processing and programming models and platforms such as Apache Spark, Apache Giraph and MapReduce, which are not conventional to people who do not deal with big data. The author(s) should provide a brief introduction to them the first time they are used in the paper. Similarly, what does the abbreviation DFS (in L355) mean? Depth First Search?

(2) Although the author(s) provided the link to supplementary materials at the end of the text, not every reader will have the time to look into the supplementary tables and figures. Some important information needs to be retained in the paper rather than in the supplementary materials. For example, the definitions of constraint factors and restraint factor and the mechanism of how they affect LULC changes are not clearly explained in Section 3.4.3 although the details are available in Table S1.

(3) The overall accuracy and Kappa Statistics of the four models in Section 4.2 can be better compared with each other in a table rather than in a lengthy paragraph (L400-L426). Please consider compiling the data from Tables S3-S6 in the supplementary materials to create a new table in the text.

(4) The two maps of LULC types in Figure 5 are too small. It is difficult to see the difference between the two maps at such a map scale.

(5) In Table 1, four subclasses of forest(f), sand(s), road (r), and human land use and all other non-grassland(l) are reclassified into one class Non-grassland (NG).  “Especially in the Xilingol League area, the simulated NG (non-grassland, i.e., human land use) (Figure 6c) was significantly less than the observed NG in this area (Figure 6d).” Why did non-grassland only indicate human land use here?

(6) This paper needs improvement in terms of language. A few sentences are very wordy. There are also minor grammatical errors in the paper. Please consider using professional proofread service to improve the quality of writing.

Author Response

Please see the attached word file. Many thanks for your insightful comments.

Author Response File: Author Response.docx

Reviewer 2 Report

Thank you very much for the opportunity to read such an interesting material! The article is devoted to a very urgent problem. And the suggested solution seems to be very effective. The methodology is described in detail. This should ensure reproducible results. Used modern and effective methods of statistical analysis. Attention is paid to both software and hardware aspects. 

I think the material can be published in its current form. Only minor corrections can be made to the text:

1) Line 125. It seems this is not a header, but a fragment of a template. Maybe it should be written "Study area" and not "Subsection"?

2) There are a lot of small details in Figure 5. Therefore, it is advisable to make Figure 5 larger. The page width and page margins allow you to increase it.

3) In figure S1 in the appendix, you can mark the boundaries of the three fragments from figure 6. This will help to understand where in figures 6A, 6C, 6E the simulated fragments of LULC are located in the gaps areas. 

 

Author Response

Please see the attached word file. Many thanks for your insightful comments.

Author Response File: Author Response.docx

Reviewer 3 Report

This manuscript focuses on a topic of sure interest for the readership of the Remote Sensing journal. However, it has some unclear issues. Please see the following list of comments:

 

- You should mention the case study in the title.

- What are the innovative contributions of your manuscript to science? (since several studies have similar methodological approaches in other case studies?)

- The methodology needs more explanation of issues and alternative approaches to overcome. More discussion of issues and alternative approaches could be helpful.

- please place a scale bar on the map in figure 1.

- line 136: why did you use ten classes? this has to be justified.

- The use of data showed in section 2.2. it has to be justified and supported with bibliographic references. Furthermore, the original scale of each of these variables/factors should be indicated.

- Figure 2 has poor readability.

- The section you designated as Discussion is actually the results. Therefore, you should change what you designated as results for discussion.

- section 4.1. Experiment environment should be in the data and methods section not the discussion/results section.

- on the map of figures 5 and 6, please put a scale, a frame of the study area, and spell out the meaning of the legend (instead of representing acronyms).

- A real discussion is missing. The authors need substantial revision to include an in-depth discussion on those important results including comparing your results with previous studies and explaining better why your results are similar or different from previous findings.

- Please address the flaws and limitations of your approach.

Author Response

Please see the attached word file. Many thanks for your insightful comments.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Thank you and your colleagues for the changes that you have made to this manuscript and how well you have answered the comments and suggestions.

Author Response

“The authors have successfully addressed almost all the suggestions from the reviewers. I would suggest the authors to further improve their manuscript, taking into account the following comment from the 3rd reviewer (not fully addressed): ‘The authors need substantial revision to include an in-depth discussion on those important results including comparing your results with previous studies and explaining better why your results are similar or different from previous findings.’”

We carefully revisited this comment and substantially enhanced the discussion section. We added two new paragraphs in this revision, discussing the differences of our approaches and results compared with previous studies concerning missing data gap filling. In the first paragraph, we contrasted our approach from the size of the study area with the current literature. We explained that our cloud-based LULC data gap filling framework was designed to fill missing data gaps over a vast study area, where the spatial disparity of LULC features is manifest. According to our experiments in Inner Mongolia, the spatial disparity affected the final gap-filling results and, thus, cannot be ignored. Hence, we proposed sub-simulation-space (SSS) strategies to address the influences of spatial disparity on missing data gap filling. We discussed this difference by referencing several previous findings.

In the newly added second paragraph, we pointed out that the classic Markov Chain CA models use the past available data to train and obtain CA transition rules and apply these rules to simulate future states with references. Our new gap-filling approach simultaneously advocated spatial and temporal data mining to fine-tune CA transition rules. For the temporal data mining, we used the available data in 2016 to train transition rules to fill the missing data gaps in the same year, 2016. We trained and derived two transition matrices, 2000-2010 and 2010-2016. We then weighted the two transitions embedded in our distributed solution to improve the accuracy of the results significantly. We provided the accuracy test results and several references to support our statement. Therefore, our data filling algorithm is not simply simulating future states from past available data but treating “the known future” as part of data mining. The known future here indicated the available information in the simulation year. Thus, our new algorithm is genuinely filling the missing data gaps.

We used these two newly added paragraphs to reorganize the discussion section. The revised discussion is more in-depth and coherent. Please read the first three paragraphs in the Discussion Section for details.

Finally, we want to use this opportunity to thank the academic editor and the anonymous third reviewer for this insightful comment. We feel confident that we have significantly strengthened the discussion section when addressing this comment in this second round of revision.