Review Reports

Round 1

Reviewer 1 Report (Previous Reviewer 1)

This article examines the deforestation rate in the KPK region of Pakistan over the past four decades. A Random Forest classifier was used to improve the accuracy of land-use classification using remotely sensed images. The paper is well written and the results are presented in detail while addressing their implications for land use policies and deforestation. My suggestions for minor revisions are as follows.

The paper is currently submitted to Sustainability instead of Remote Sensing or other journals. The discussion on remote sensing techniques would be more concise. And the authors may want to strengthen the discussion on policy issues or the contribution of this case study to the literature focusing on sustainability or the issue of deforestation.

In addition, the authors may want to address the limitation of this study in the end, in terms of both the methods and the case. 

Author Response

Point 1:  The paper is currently submitted to Sustainability instead of Remote Sensing or other journals. The discussion on remote sensing techniques would be more concise. And the authors may want to strengthen the discussion on policy issues or the contribution of this case study to the literature focusing on sustainability or the issue of deforestation.

Response 1:

Effective policy is key to addressing the issue of deforestation and promoting sustainable forest management. Some policy measures that can contribute to this goal include: Developing and enforcing clear regulations and laws prohibiting illegal logging and other activities contributing to deforestation. This can include measures such as strengthening penalties for violators and creating mechanisms for monitoring and enforcing compliance.  Promoting land use planning and zoning that considers the environmental and social impacts of different land uses. This can help ensure that forests are protected and that development sustainably takes place and respects the rights of local communities. Encourage sustainable forestry practices, such as selective logging and reforestation, through incentives and education programs. This can help ensure that forests are managed in a way that preserves their ecological value and supports local communities livelihoods.

This study can contribute to the literature on sustainability and deforestation issues by providing a detailed analysis of the forest and other vegetation cover assessments using satellite remote sensing to implement a specific context and their impact on forest management. By examining the successes and challenges of these measures, the study can provide valuable insights and lessons for policymakers and other stakeholders working to address these issues. Additionally, the analysis can highlight the importance of considering the perspectives and needs of local communities in the development and implementation of policy, as well as the role that stakeholder engagement can play in promoting sustainable forest management.

 

Point 2:  In addition, the authors may want to address the limitation of this study in the end, in terms of both the methods and the case.

Response 2:

Limitations of the study include that satellite data may not accurately map forests, especially in areas with complex topography or dense canopy cover. In these cases, ground-based data may be needed to supplement the satellite data and improve the accuracy of the forest map. Second, satellite data may be subject to spatial resolution limitations, which can affect the accuracy of the forest map. For example, data with low spatial resolution may not provide sufficient detail to map small-scale features such as individual trees accurately. Third, the accuracy of the forest map may be affected by the quality and availability of ancillary data, such as digital elevation models and land cover maps, which are often used to improve the accuracy of the forest map. Finally, the analysis results may be limited by the sample size and the representativeness of the case study area. A larger sample size and a more diverse range of case study areas may provide more robust and generalizable results. The Random forest method is a relatively complex algorithm that can be difficult to understand and interpret for some users. This can make it challenging to explain the results of a random forest model to non-technical stakeholders. The model's output may be unreliable if the input data is incomplete, noisy, or biased. Overfitting is another limitation of this study. The random forest can be prone to overfitting, which occurs when a model is too closely tailored to the training data and doesn't generalize well to new data. This can lead to poor performance on unseen data.

Reviewer 2 Report (Previous Reviewer 2)

Authors have well addressed all my concerns. This manuscript could be accepted for publication.

Author Response

Point 1: The authors have well addressed all my concerns. This manuscript could be accepted for publication.

Response 1:

Thank you for your significant reviews and suggestions and for recommending the paper for publication.

Reviewer 3 Report (New Reviewer)

The effort of mapping LULC dynamic and analyzing forest cover using Landsat time-series and RF is interesting and would bring a significant contribution in this field.

Besides that, manuscript needs serious improvement.

L259. I guess figure 3 is missing from text

L316. Figure 4 is missing too.

L360. Formula for specificity is missing.

Table 2. To avoid class imbalance it is a good idea to resample dataset to equal parts. RF has an ability for overfitting with class imbalance

Table 5. Please add to the table precision and recall metrics

Table 5(L497). Please specify, percentage in each timestep calculated from 1980 or from previous?

Please clarify, why using stack of vegetation indices has better accuracy than Landsat bands? In my opinion processing of bands is faster. Increase in accuracy shouldn’t be as much as expected.

I wish that my comment would be helpful in improving the quality of this research.

Thank you.

Author Response

Point 1:  L259. I guess figure 3 is missing from text.

Response 1:

Figure numbers were incorrect. Corrections were made according to the instructions.

 

Point 2:  L316. Figure 4 is missing too.

Response 2:

Figure numbers were incorrect. Corrections were made according to the instructions.

 

Point 3:  L360. The formula for specificity is missing.

Response 3:

The formula of specificity was added to the manuscript.

Point 4:  Table 2. To avoid class im balance, it is a good idea to resample the dataset to equal parts. RF has an ability to overfit with class imbalance.

Response 4:

It is generally a good idea to balance the classes in a dataset when training a machine learning model, especially when the classes are imbalanced. One way to balance the classes in a dataset is to resample the dataset by either oversampling the minority class or undersampling the majority class. In particular, if the majority class is much larger than the minority class, the RF model may overfit the majority class and have poor performance on the minority class. To address this issue, we tried balancing the classes in the dataset before training the RF model. It is also possible to modify the RF training process to address the class imbalance. For example, we used class weights or set the class_weight parameter in the RF model to give more importance to the minority class. We also use the min_samples_leaf parameter to control the minimum number of samples required at a leaf node, which can help prevent overfitting to the majority class. The detail of the classes and features selection for the study are explained in the result section, and changes are made to improve the methodology section.

 

Point 5:  Table 5. Please add to the table precision and recall metrics

Response 5:

Precision and recall metrics were added to the manuscript in the table.

 

Point 6:  Table 5(L497). Please specify, percentage in each timestep calculated from 1980 or from previous?

Response 6:

Table 5 indicates the increase and decrease in the area of forest, woodland and other land cover changes from 1980 to 2020 with five years of time spin in percentage. The values for 1980 were not present in the table as we started our estimation in 1980.

 

Point 7:  Please clarify, why using stack of vegetation indices has better accuracy than Landsat bands? In my opinion processing of bands is faster. Increase in accuracy shouldn't be as much as expected.

Response 7:

Processing individual bands may be faster than calculating multiple indices, but the increased accuracy of using indices can often outweigh the additional processing time. Additionally, modern computing hardware and software can make it possible to process large volumes of data quickly. Using a stack of vegetation indices, rather than individual Landsat bands, can often improve the accuracy of the land cover classification and other types of analysis because the indices are specifically designed to highlight the unique reflectance characteristics of vegetation. This can make distinguishing between different vegetation types and other land cover classes easier. The choice of which bands or indices to use will depend on the specific goals and requirements of the analysis.

Round 2

Reviewer 3 Report (New Reviewer)

The manuscript has significantly improved from the last revision and can be published in present form

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This paper presents an empirical analysis of land use and land cover change with a focus on forest land areas in Pakistan's northwestern province of Malam Jabba. Results indicate a very high loss rate of even green forest areas in the region. I find this article is more focused on the technical part of the story instead of its implications for sustainable development. I suggest the authors make extensive revisions before it can be accepted for publication. 

First, besides the methodological contribution by using random forest procedure, what would be the most interesting finding as compared to other areas that also experience deforestation? For example, the paper concludes that overgrazing, urbanization, road construction, firewood collection are major driving forces behind the loss of forest areas. I would suggest the authors further elaborate these dynamics.

Second, I would suggest the authors add a section titled discussion in this article. The fourth section includes lots of plain discussion on the results. A more in-depth analysis of these dynamics I mentioned above is needed.

Third, besides classification techniques, would it be possible if a regression analysis would be conducted to better quantify the so-called dynamics of deforestation? 

Author Response

Environmental problems like overgrazing, urbanization, road construction, and firewood collection are major driving forces behind the loss of forest areas throughout the district of Swat. The above phenomenon is elaborated using the literature available on the area of interest. However, Malam Jabba is the small hill station which contains forests on a significant portion.  A section with a title discussion is added to compare different studies conducted on the study area and district Swat. A regression analysis needs more variables to better quantify the so-called dynamics of deforestation in the region, but lack of historical data, this study only focuses on using the RF method to address the challenges in the remote region. In the discussion section, the results are compared with the results of deforestation and LULC in different developed and underdeveloped countries, comparing the scenario concerning the current changes in land features.

Reviewer 2 Report

Authors used landsat time-series imagery and random forest classification method to map the lulc dynamic in Malam Jabba region. Generally, lulc study belonged to a typical area. In this manuscript, authors took a small region as the study area. Took several vegetation indexes to extaract forest, wood land and other land. Authors had done many analysis. Generally, this paper's novelty is high, but the structure is poor.    This manuscript used Landsat images to do the LULC classification in Malam Jabba region. Authors used images to calculate several vegetation indexes to extract the forest, wood land and other land. Random forest method was applied to do this progress.      LULC especially forest dynamic change analysis and detection is important for sustaniable development goal. Especially in the background of global warming. Authors used several images to detect the forest, woodland extraction accuarcy. Authough the study area is small, it address a specific gap of forest accuarate extraction. Existing framfork provide a light for this field.     This subject area should be the environment sustainable development, vegetation detection and forest extraction is important for regional sustainable development. Authors provide a framework to extract forest information more accuractely.     Methodology should be more detailed. Including the formula you used and the description. The sturuture should be well organized.Especially in the discussion part, authors should compare there results with other similar studies including methods application or classification accuracy comparsion.     Yes, authors have provide consistent conclusions, however, in the abstract part, as i addressed, the description about the study area and background is too vague. Suggestu authors provide concise information about this manuscript's meaning.   all references were appropaiate. Suggest authors provide more newest relavant papers in this field. As LULC or forest detection studies had already achieved fruitful results.     Tables and figures should be more concise, especially for figures, please provide a sharper dpi images.

 

Abstract: In the begining, don't talk too much about your study area. These content should be in study area part.

Author Response

The structure is organized by arranging some paragraphs and sections in the articles. In the discussion part, different correlative comparisons were made by using similar studies, including methods application or classification accuracy comparison.  The scenario regarding the current changes in land features is compared with the outcomes of deforestation and LULC in various developed and developing countries.  The error in the abstract is addressed, and the study area description was discussed in more detail according to the instruction. More papers related to the latest studies on the exact domains and application methods were added to the article. Tables and figures are made more concise, mainly providing a map was enhanced to make it more understandable.

Reviewer 3 Report

Mapping LULC dynamic and its potential implication on forest cover in Malam Jabba Region with Landsat time-series imagery and Random Forest classification

Within this manuscript, time-series of Landsat imagery were used for mapping forest cover in Malam Jabba Region using Random Forest classifier. The results show that Malam Jabba's total forest land area in 1980 was about 236 km2 and shrank to 152 km2 by 2020. The Authors claim that the information presented in this research will be a great help in the future for guiding future efforts to sustainably manage the forest ecosystem..

Although the topic of the manuscript is of wide interest in the scientific community and for this journal, I am seeing several serious methodological and scientific writing flaws in this study.

Review summary

·         First of all, English writing and grammar needs to be improved in the manuscript.

·         Secondly, an abbreviation is mentioned in the title (LULC) which is not explained in whole manuscript. I know that it is wide known what LULC means, nevertheless, it has to be explained

·         Please, indicate the scientific contributions of the manuscript in the new paragraph in Introduction. In a current form, the only scientific contribution of the study is mentioned in LN 142-145, whereas other part of the paragraph already explains e.g., training and validation points, etc. Furthermore, LN 140 – 142 mention „(..) to predict and assess forest fire risk in this region.“, and after reading the manuscript, I did not see any research about forest fires?!?

·         The Authors mention feature selection (FS) in LN 114, and in section Results the selected features were mentioned, but we do not know what method for FS was used (e.g., MDA, MDG, Boruta, VSURF)

·         LN 236: VCP abbreviation is not introduced; LN 251: 'Baran land' ??; LN 252: DBSI values ??

·         When Authors describe Random Forest algorithm in Section 2, they already reference to a Figure 4 and 6 which are placed several pages later, that is really hard to follow

·         LN 365: pan-sharpening should be described in section Materials and methods, not Results

·         Table 4 and LN370: Altitude lower than 1700m – this is wrong

·         Figure 5 and Table 6, Figure 7 present the same information, they are excessively redundant

·         Figure 8, 11 and 13 are very hard to read/decode, again, they present information which were already presented before

·         Figure 15 is definitely an error matrix, it presents information which were derived from the error matrix

At the same time, the attempted methodology is very actual and interesting, but the most important issue is that a lack of novelty and too many redundant information. My final opinion is that this research has a solid potential and it is very interesting, but in a current presented form, it does not fulfill the Journal criterion.

 

Author Response

English writing and grammar mistakes were addressed to enhance the quality of the manuscript.  The abbreviation LULC and its implication were explained in many sections of the manuscript according to the instruction. A new paragraph was included in the Introduction section, which indicates the scientific contributions and importance of the manuscript. LN 140 – 142 mentions related to predicting and assessing forest fire risk in this region were removed as it doesn’t have a direct relation with this study, but in the past, many events related to forest fire have been recorded in the region. VCP abbreviation of VCP, LULC DEM and DBSI was introduced in different sections of the article. Figures 4 and 6 were arranged related to the random forest algorithm. The methodology related to the feature selection (FS) in LN 114, and in the section and the results, the selected features were explained according to the reviewer’s comments. Pan-sharpening were rearranged and described in the section Materials and methods. Correction related to the Altitude was done in LN 370. Excessively redundant figures 5, 7, 8, 11, and 13 were addressed and modified. Novelty and the scientific importance of this article are inserted and explained in the different paragraphs of the article as per directed by the reviewer.