Analysing Process and Probability of Built-Up Expansion Using Machine Learning and Fuzzy Logic in English Bazar, West Bengal

Round 1

Reviewer 1 Report

The manuscript investigates the use and application of SVM in land use retrieval within English Bazar Block of West Bengal, India, and applies the brand new landscape fragmentation technique and frequency approach into statistical analysis, for identifying potential areas or spaces for future urban expansion. The topic is important and interesting, especially for developing cities. The research gap mentioned in Line 85 is indeed a scientific advancement. I think the manuscript should be published after the following points and typos / grammatical changes are properly addressed:

Major Corrections

Lines 63-76: Since this study is more focused on developing cities / countries, it will be better to include some relevant studies (for example how Support Vector Machine (SVM), Random Forest (RF), or other machine learning approaches were recently applied in other developing countries like Pakistan, South Africa etc., for conducting land use retrieval. Further, References [37] and [38] are not too relevant to the study, and should be removed from this manuscript, and replaced by studies as aforementioned, with focus on machine learning and statistical retrieval algorithms, especially the application of SVM (the technique adopted in current study) in retrieval of LUCC maps and urban dynamics in Pakistan, India and so on.

Table 1: How did the authors quantify the cloud cover on those dates, or periods? If not, how can we ensure that there is no cloud coverage in those retrieval period / dates? It would be better to outline the ideas behind.

Lines 144-146: The authors should also highlight the use of hyperplane within SVM. Please refer to recent studies of applying Support Vector Machine into LUCC retrieval, which should consist of more details about SVM. It will be better to provide more details of SVM in this section.

Line 250: Please define the parameters used in post classification validation.

Figure 3: Why are built up area mostly concentrated in North Eastern / Eastern area? Any spatial reasoning?

Table 3: Need some descriptive paragraph to compare and contrast the values obtained in Table 3, otherwise all numbers seem good, and we are not sure which is better than which? Moreover, what are the number of cells / grids considered in each of these 3 years?

Figure 4: What do these numerical values represent? It will be better to give an indicator, for example, above what number will the change be obvious and significant?

Figure 5: What are the reasons of converting from built up to non-built up area?

Figure 9: How do you define "Stability"? Is it using the LUCC maps within the 3 years?

Lines 400-425, 426-440, 441-448 seem to be more suitable to be put in Conclusion, rather than Discussion, as Discussion should focus on some extensions based on the research results or findings.

Lines 488-490: What are the criteria of these "high probability future growth hubs" and "moderate probability future growth centres"? Any mathematical formulation / justified parameters / scoring indices?

Lines 527-530:  Is 30 m really considered as "coarse" resolution? If that is the case, what kind of datasets of even finer resolutions should researchers and scientists use?

Also, what kind of deep learning models / machine learning techniques can be used to overcome the shortcoming? How can it be implemented? SVM or RF were used very often till now, and they are actually not conventional algorithms.

Therefore, these limitations may not be essential or obvious. Could the authors suggest other limitations, say from the perspectives of data handling and retrieval, cloud cover etc.?

Minor Corrections

Line 48: What kind of urban problem is addressed in Reference [17]?

Line 112: For "pollution", do the authors refer to air pollution or sound pollution, or other kinds of pollution?

Line 183: why 3 times?

Lines 238-245: need some more numerical values within this paragraph.

Lines 309-310: Any citation on this, about the expansion of transportation network?

Label of Figure 7: should include a layer for 0, labeled in white

Line 361: Please label "NH-34" on one of the sub-plots of each figure, so that it will be easier for readers to follow.

Lines 364, 368 and 371-372, 393-398: These lines and contexts have mentioned a lot of places and towns, please label them accordingly in the spatial plots.

Lines 494-497: Can perhaps add some environmental considerations here, because it concerns about introducing more green space.

Line 516: what do you mean "the built-up area block"?

Grammatical Changes / Errors

Lines 45-46: Add comma - "[11-14], which has resulted in unplanned expansion of these towns, as well as"

Line 64: [5, 27-29]

Line 138: presented in Table 2

Line 146: Since the training data are important for...

Line 164: was examined in Figure 2

Line 170: Percentage Change, also, the product sign should be replaced by the mathematical notation

Lines 197-198: dominant, diversify and connective model

Line 204: , and 1 indicating that

Line 211: fulfill

Lines 403-404: transformation from agricultural land into built-up areas.

Line 433: patch [58], where core indicates

Lines 460-461: Therefore, the new ....migrants.

Line 500: Bold "Figure 11"

Lines 501-502: where's the NH34? Also, there is no close bracket after 2021.

Line 505: below --> above

Lines 526-527: restricting urban sprawl and nurturing compact green cities,

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

It is appreciated that the manuscript is easy to follow and not too long. With interested I read the manuscript, the message is clear and of interest to the community. Authors utilized multi-temporal Landsat satellites images to investigate built-up expansion using machine learning and fuzzy logic methods. I would suggest a major revision before the manuscript could be accepted. Please allow me to clarify.

1. Cannot emphasis more on correcting the language in terms of maintaining consistency (past/ present tense, active/passive etc.) and many other grammar errors. I would suggest choosing one style and stick to it throughout the manuscript.

2. Elaborate more on how SVM was trained: was it one v/s all, or one v/s one training, description of hyperparameters and their selection strategy etc. (these are important information to assess the authenticity of machine learning modeling).

3. There is some information provided for validation in section 3.2, however, there is no information provided for training/test: How many samples were used? What was training and test ratio? Was there any sampling method involved? etc. Additionally, provide detail on sample distribution.

4. Landsat 5 and 8 has considerable difference in their band passes, and they have different number of bands too, how this difference was reconciled? was there any particular subset of bands chosen for this study, if yes, what was the rational? which level data products were considered from Landsat 5 and 8? If Level-1 was considered, was there any radiometric calibration performed (EXTREAMLY IMPORTANT for multi-temporal analysis)? What physical unit was used (TOA reflectance/ radiance/ raw pixel values etc.)?

5. Please replace Figure 6 with a better-quality figure.

6. In your discussion section could you please also include how this methodology could be adapted for other satellite images. Were there any efforts put on increasing temporal resolution of the dataset? How this would affect the results/insights? Lastly, elaborate more on implication of the methodology in general.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The article “Analyzing process and probability of built-up expansion using 2 machine learning and fuzzy logic in English Bazar, West Bengal” is well written and structured – except for the conclusion. The conclusion must be rewritten. The presentation of the results meets international standards.

However, the authors do not present anything new, that they are not known, as well on the plans of data, methodology, algorithms, and results on the urban fabric, the urban dynamics, etc. The study uses the usual approaches and processing methods (SVM, fuzzy logic, fragmentation) - including satellite images from the Landsat 5 TM to Landsat 8 OLI series - with results that are quite typical in the important literature on the issues of urbanization in India.

The explanation of the choice of construction of fuzzy logic indices is interesting. It would have had the merit of being more developed and reframed in the general context of the scientific literature in this specific field. This is also one of the shortcomings of this article: the lack of bibliographic references referring to the approaches, algorithms and results that would make it possible to better identify the contributions.

The results in terms of knowledge production are interesting but remain also typical. It would have been interesting to better integrate them with the knowledge and associated scientific references which are numerous. Moreover, a comparison with demographic and economic changes would make sense in this study. They are vaguely mentioned (line 520).

 The figures must be improved in terms of semantics, and quality, and meet the basic rules of cartography. The summary is too vague, and the elements must be specified: line 19 (satellite data?) line 22 type of landscape fragmentation methods?)

Smart missing:

Line 44: missing a “,” Line 64: [ ] Line 89: space

Lines 206-2018: extend this part. This is the “innovative” part of this publication.

Fig.2: Graphic semiology to be resumed Fig. 3, 5, 6, 7, 8, 9: serious problems of semiology and semantics in cartography.

Fig. 11: A big, unforgivable mistake. What is a Google Earth image? What are you talking about! These are colourful compositions of commonly used commercial satellites!

The conclusion must be completely rewritten. It is uninteresting.

The question of the scale of analysis is not addressed. Why limit us for this type of study to only Landsat images (archives - “historical depth aside”-).

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The revised version of this manuscript looks much better than before, except the following points, which the authors should address as well before publication:

Lines 79-95 (in revised manuscript): The authors only cite sources from different mega-cities of India, but not as what we have suggested in previous review report - in Pakistan, South Africa as well. The following references should be added:

https://www.mdpi.com/2072-4292/13/16/3337 

https://www.sciencedirect.com/science/article/pii/S0924271621001635

https://www.ajol.info/index.php/sajg/article/view/125790/115329

Further references have to be added too.

Further, "we first identified the land use features, including built-up area, as an essential input for further research using SVM for 2001-2021", the approach of using SVM is not new, as shown in the above several references. Therefore, the authors should present the new breakthrough of this study in a more clear manner.

Table 1: Regarding cloud cover, the situation is now clearer, and the authors should cite relevant source in their manuscript as well.

Lines 208-219 (in revised manuscript) - please kindly refer to the following two references for more details of the statistical parameters and statistical principles of SVM used:

https://www.mdpi.com/2072-4292/13/16/3337  

https://www.thaiscience.info/journals/Article/WJST/10958633.pdf

Lines 381-385 (in revised manuscript): The factor and reason mentioned are quite reasonable and well explained. Are there any local government sources? Please kindly cite some official documents from local government.

Line 385: authority (spelling)

Line 425: The "2" of km2 should be upper-scripted.

Section 4 (Discussion): There are several paragraphs and main focus of these available paragraphs added. Can we separate them into different sub-sections, like Section 4.1, 4.2...?

Other than that, I think the manuscript has certain scientific impact, and the study is meaningful. After addressing the above points and citing more relevant references, the manuscript is good to be published.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Thank you for making efforts to address all the comments/concerns. I recommend publication of this manuscript in the present form. 

Author Response

We are very thankful to you for accepting our work.

Reviewer 3 Report

The authors have done a great effort of work to improve the submitted manuscript by taking into account all the remarks of the reviewers. Four remarks:

-lines 142-150/ 165-172: the choice of the number of spectral bands restricted to the visible range is not relevant. The authors do not use the spectral capacities of the TM and OLI sensors. It would have been more appropriate to use the spectral bands of the SPOT series (visible and near-infrared) which have a much better spatial resolution on the ground, more suited to the study problem. Why not pan-sharpened the multispectral images with panchromatic band 8?

-Line 498-439: The built-up expansion probability model with its parameters should be detailed. What is the error level of the simulation model?

-I still don't know what is the high-resolution image (false colour composition) under Google Earth presented in this article?

-The quality of the figures needs to be improved.

Author Response

Please see the attachment

Author Response File: Author Response.pdf