Spatial and Temporal Variations of Forest Cover in Developing Countries
Round 1
Reviewer 1 Report
The paper discusses the temporal variation of forest cover for those countries, which are usually summarized as „developing countries“ with a share of forest coverage of more than 3%. In addition, using a binary logistic regression model, an attempt is done to determine some driving factors of forest transition.
The definition “developing countries” is not very clear, even in the source mentioned (WESP report) the term is not used. The map in Figure 2 is not consistent with the definition. A proper definition for the regions or countries used in the study should be given.
The number of “developing countries” is reduced by those countries, which have less than 3% forest cover. But the list of countries used finally for the study is not give, neither as a list or a map. Thus the user has no idea, which countries are included. Recommendation: Table A1 should only those countries included in the study.
Related to this, it is not clear how the countries in figure 6 have been selected. What criteria has been applied? Countries like Brazil, which show a big change in the forest area are not included, whereas other countries with a very small forest area like Israel, are included. At least, some justification about the selection should be given.
Also the term “forest” is not clearly defined. This is important, if different sources are compared as done in the discussion., se e.g. https://news.mongabay.com/2017/03/in-defining-plantations-as-forest-fao-attracts-criticism/
The factors used in the regression model are not justified. Why these variables have been used? In the result and discussion there is the attempt to set up some chain of causalities. But this is not really proved by the results. The regression model just shows some correlation, but gives no idea about cause and effects. Thus her a better justification about the selection of the variables should be given and the discussion about the results related to the regression model should be improved.
ll156: sever à severe
ll 199: radio à ratio
ll 86 is duplicated: the forest area is not “other data”, but has been already mentioned in 2.1.2
Author Response
Many thanks for your patient comments! We learned a lot from them. There are our responses to your comments.
Point 1: The definition “developing countries” is not very clear, even in the source mentioned (WESP report) the term is not used. The map in Figure 2 is not consistent with the definition. A proper definition for the regions or countries used in the study should be given.
Response 1: Many thanks!
We have added the definition of developing countries,
”The developing countries referred to the developing economies from the classification results of the countries in the WESP report.”
Then we modified the study area in Figure 2.
Point 2: The number of “developing countries” is reduced by those countries, which have less than 3% forest cover. But the list of countries used finally for the study is not give, neither as a list or a map. Thus the user has no idea, which countries are included. Recommendation: Table A1 should only those countries included in the study.
Response 2: Many thanks for your comment!
In our original manuscript, in the part of detecting changes in forest cover at the global, continental, and country scales, all 98 developing countries were included; and in the part of detecting forest transition, we excluded countries with forest coverage below 3% to rule out accidental factors. According to your suggestions, in the revised manuscript, only developing countries with forest coverage of more than 3% are included, and we presented the list in Table A1. Meanwhile, because of this revision, some values in the 3.1 part have been re-calculated.
Point 3: Related to this, it is not clear how the countries in figure 6 have been selected. What criteria has been applied? Countries like Brazil, which show a big change in the forest area are not included, whereas other countries with a very small forest area like Israel, are included. At least, some justification about the selection should be given.
Response 3: Many thanks!
Figure 6 showed the countries that the turning point was during 1992 and 2015, just as the figure’s name “DFT countries (countries during forest transition) and their turning point”. Brazil have not experienced forest transition until 2015, which belonged to BFT countries (countries before forest transition), and Israel is a DFT countries. In the revised manuscript, we also presented the change of forest coverage of BFT countries and PFT countries in the supplementary materials (Figure A1 and A2).
Point 4: Also the term “forest” is not clearly defined. This is important, if different sources are compared as done in the discussion., se e.g. https://news.mongabay.com/2017/03/in-defining-plantations-as-forest-fao-attracts-criticism/
Response 4: Many thanks! Actually, the term “forest” is very important! The forest definition was further explained in the Supplementary Table A2, and in this study, the plantations are contained in forests, just like the definition of FAO. The Table A2 was listed as follows:
Table A2. The vegetation types of the forests
Types | Qualification |
Tree cover, broadleaved, evergreen | Closed to open (>15%) |
Tree cover, broadleaved, deciduous | Closed to open (> 15%) |
Tree cover, needleleaved, evergreen | Closed to open (> 15%) |
Tree cover, needleleaved, deciduous | Closed to open (> 15%) |
Tree cover, mixed leaf type (broadleaved and needleleaved) | - |
Mosaic tree and shrub / herbaceous cover | (>50%) / (< 50%) |
Tree cover, flooded, fresh or brakish water | - |
Tree cover, flooded, saline water | - |
Point 5: The factors used in the regression model are not justified. Why these variables have been used? In the result and discussion there is the attempt to set up some chain of causalities. But this is not really proved by the results. The regression model just shows some correlation, but gives no idea about cause and effects. Thus her a better justification about the selection of the variables should be given and the discussion about the results related to the regression model should be improved.
Response 5: Thanks! We have added the description of variable selection, including reasons, theoretical basis and corresponding proxy variables. Meanwhile, we also added relevant references, and the revised description was as the following:
“According to previous research, forest scarcity and the economic development are regarded as two main pathways of FT. In the former pathway, in some countries with little import ability and stable demand for forest products, the prices of the forest products rise under the spur of scarce forest resources, and promote the land conversion to woodland. Therefore, the forest area was regarded as one important potential driving factors for FT, and we used forest coverage in the framework of interpretation to reveal this characteristic. In the latter pathway, during the rapid urbanization process, the labors shift from marginal agriculture to industrial production, and the abandoned farmlands revert to forest. This pathway related to economic development and urbanization, therefore, the GDP per capita and urbanization level were put into the model. In later studies… Globalization could shift the pressure on forest resources to other countries to reduce deforestation. As a result, the ratio of export value to import value of forest products that could reveal the trade conditions of the country was designed as an independent variable. Overall, considering the availability and comparability of global data, four independent variables (Forest_r, GDP_per, Urbanization_r, Trade_r) were listed as the potential FT driving factors in this study at the global scale and continental scale.”
Thus, in our manuscript, the cause and effects was reflected in the variables selection, the variables were potential driving factors, according to previous studies. In addition, in the discussion section, we improved some statements and added some relative references.
We hope the revised version can better explain the relationships.
Point 6: ll156: sever à severe
Response 6: Many thanks! We have corrected this word.
Point 7: ll 199: radio à ratio
Response 7: Many thanks! We have corrected this word.
Point 8: ll 86 is duplicated: the forest area is not “other data”, but has been already mentioned in 2.1.2
Response 8: Many thanks! We have deleted the description of forest data in the section of “other data”.
Author Response File: 
Author Response.docx
Reviewer 2 Report
@page { margin: 2cm } p { margin-bottom: 0.25cm; line-height: 115% }
The authors a very interesting manuscript where they analyse the changes on forest extension in several developing countries over time, using of course remote sensing data and techniques. This information is essential for many conservation studies. The manuscript is very well written and clear, the methods are sound, and the conclusions fits the results. I really enjoyed reading the manuscript. Sorry, I have very few comments to improve the manuscript:
1. Figure 4 must be presented at global scale. I think that one of the most important results is to plot at global scale the trends on forest changes.
2. The logistic regression can be improved. The data must be split in training and test to validate the model. You can use a validation metric such as AUC (Area under the curve of the ROC plot) or TSS (True Skill Statistics). Please, provide the value of a pseudo R2 to indicate the amount of variance explained by the model. Which is the prevalence of the logistic regression, the proportion of presences and absences? In logistic regression, prevalence must be close to 50% (a few more absences than presences). The variables introduced in the model do not need to be normal, but residuals yes. Please, check if residuals are normal.
Other comments:
Figure 1 – How are the limits of three types of FT defined?
P 4- Authors stated that the logistic regression was composed by 1795 records. I think this value may be wrong. For the logistic regression, there must be a record per country and year – 75 countries and 23 years = 1725.
Table 1 – Are this the only variables that can drive forest increase?
L142 – Please provide the SD for total forest area as well.
Figure 3 – It will help to put the names over the lines at the right side of the plot. I had difficulties to distinguish the colours.
Figure 6 – I would like to see these plots for all countries. You can put this information as Supplementary Materials.
Author Response
Many thanks for your patient comments! We learned a lot from it. There are our responses to your comments.
Point 1: Figure 4 must be presented at global scale. I think that one of the most important results is to plot at global scale the trends on forest changes.
Response 1: Many thanks for your comments, we have presented the changes at global scale, the modified figure was followed in the word file.
Point 2: The logistic regression can be improved. The data must be split in training and test to validate the model. You can use a validation metric such as AUC (Area under the curve of the ROC plot) or TSS (True Skill Statistics). Please, provide the value of a pseudo R2 to indicate the amount of variance explained by the model. Which is the prevalence of the logistic regression, the proportion of presences and absences? In logistic regression, prevalence must be close to 50% (a few more absences than presences). The variables introduced in the model do not need to be normal, but residuals yes. Please, check if residuals are normal.
Response 2: Many thanks for your professional guidance! The comments remind us a lot of knowledge we have missed. We searched for the relevant materials carefully according to your opinions. First, the purpose of the regression is to explore the driving factors of the forest transition, not prediction. Therefore, we did not do the training and test. Second, xtlogit is quite different from the logit model, according to the some materials (https://www.statalist.org/forums/forum/general-stata-discussion/general/14 26166-roc-curve-after-xtlogit), it cannot do the ROC plot, “-lroc- (instructions to do ROC plot) is written to run only after -logit-, -logistic-, or -probit-, not -xtlogit-”. Third, the resulting report of the xtlogit-random effect model did not provide the pseudo R2, as the Stata manual said, ”the Stata return -e(chi2)- for the random-effect specification”. Therefore, we presented the Wald chi2 and Log likelihood to reveal the model effect. Fourth, our logistic model is the panel-logistic model, in time series of 1992–2015, some countries always presented (presence); some countries always presented (absence); and some presented (presence) and (absence). The proportion of presences is 54%, which is close to 50%, but the presences are a bit more, for our explanation aim, this is acceptable.
Finally, according to the some materials (https://stackoverflow.com/questions/ 12146914/what-is-the-difference-between-linear-regression-and-logistic-regression), “In logistic regression, residuals need to be independent but not normally distributed.”
I do not know if there is anything wrong with my understanding, if there is, please correct me. Thank you very much for your patient comments again!
Other comments:
Point 3: Figure 1 – How are the limits of three types of FT defined?
Response 3：Thanks for your comments! As we mentioned in the manuscript, during the period of 1992–2015, countries with the declining forest coverage are defined as before forest transition (BFT); countries with continuously increasing forest coverage are regarded as post forest transition (PFT); and countries whose forest coverage has changed from reduction to expansion during the period are defined as during forest transition (DFT). The limit points in this figure was 1992 and 2015. When the country experienced forest transition at 1992 or 2015, it was defined as the “DFT countries” (experienced forest transition during 1992 and 2015).
Point 4: Authors stated that the logistic regression was composed by 1795 records. I think this value may be wrong. For the logistic regression, there must be a record per country and year – 75 countries and 23 years = 1725.
Response 4：Many thanks for pointing out this mistake for us! It has been corrected.
Point 5: Table 1 – Are this the only variables that can drive forest increase?
Response 5：Many thanks for your comment! We selected the variables according to the previous studies and data accessibility; some other potential factors such as government policies were difficult to acquire global statistic data. Besides, there are potential relationships between some variables, such as forest coverage and agriculture land area. In order to make the results as clear as possible, we only selected the major variables. In the revised manuscript, we tried to have a better justification about the selection of the variables.
Point 6: L142 – Please provide the SD for total forest area as well.
Response 6：Many thanks for your suggestion! We have added the SD value after the first sentence of the paragraph,
” and its standard deviation was 24.6%”.
Point 7: Figure 3 – It will help to put the names over the lines at the right side of the plot. I had difficulties to distinguish the colours.
Response 7：This suggestion is excellent! We have modified this figure as follows.
Point 8: Figure 6 – I would like to see these plots for all countries. You can put this information as Supplementary Materials.
Response 8：Many thanks for your suggestion! We have added the plots that showed the forest coverage change of BFT countries and PFT countries, in Appendix A-Figure A1 and A2. The figures were followed in the word file:
Figure A1. Forest coverage change of BFT countries
Figure A2. Forest coverage change of PFT countries
Besides, according to other reviewer’s suggestions, in the revised manuscript, only developing countries with forest coverage of more than 3% are included, and we presented the list of these countries in Table A1. Because of this revision, some values in the 3.1 part have been re-calculated.
Author Response File: Author Response.docx
Reviewer 3 Report
Spatial and Temporal Variation of Forest Cover in Developing Countries.
This is a stimulating paper on spatial and temporal variation of forest cover in some Developing Countries, from 1992-2015. The paper works, and the relevance of the main ecological functions of forests. The paper starts with a discussion of some of the most important ecological questions of forests, in developing countries.
Some specific points of discussion in the paper are:
1. Pag. 1. The description of forests products and their ecological functions, perhaps could be introduced with a bit more of details and discussions. Description for such an important point is too weak.
2. Pág 3. Explanation about the binary logistic regression model should be a bit more clear. Which is the meaning of “a macro perspective”?
3. Some stronger theoretical foundations to justify the explaining variables in Table 1 is needed.
4. Presentation of explaining variables is a bit weak.
5. Table 2. If panel data are used, could you justify why a different constant term is not used in each regression? There is only one constant term in each regression?
Author Response
Many thanks for your patient comments! We learned a lot from it. Here are our responses to your comments.
Point 1: Pag. 1. The description of forests products and their ecological functions, perhaps could be introduced with a bit more of details and discussions. Description for such an important point is too weak.
Response 1: Many thanks for your comment! After reading some relevant literatures, we have expanded the description of the importance of forests, just as the following:
“The global forest change is always concerned by the public [1-6], because human has heavy reliance on forests. On the one hand, forests could provide forest products [7], not only for food, but also for medicine, fodder for livestock, fuel and shelter [8]. It was estimated that the livelihoods of 1.6 billion rural people are dependent on forests [9]. On the other hand, the ecological services provided by forests can solve some environmental problems [10,11], such as reducing soil erosion and improving water quality by regulating hydrological cycle [12,13], ameliorating climate change through increasing carbon sequestration [14] and protecting biodiversity [15]”.
Point 2：Pág 3. Explanation about the binary logistic regression model should be a bit more clear. Which is the meaning of “a macro perspective”?
Response 2: Many thanks! We have expanded the description of the binary logistic regression model, just as the following:
“In this study, we established one regression model for all developing countries and every continents. Forest transition status was considered as the binary dependent variable. If the country experienced forest transition, we set the dependent variable=1; otherwise, it was 0. In addition, four potential driving forces (Forest_r, GDP_per, Urbanization_r, Trade_r) were regarded as independent variables.”
And we also added interpretation of logit model formulas,
“where b0 is a constant, and the parameters bi (i=1, 2, 3, …, k) can reveal the impact of each independent variable for the outcome.”
As for the “a macro perspective”, I meant an overall analysis based on country and continent scales, but we removed this term in the revised manuscript, because of the confusion.
Point 3：Some stronger theoretical foundations to justify the explaining variables in Table 1 is needed.
Response 3: Many thanks for your comment! We have added the description of variable selection, including reasons, theoretical basis and corresponding proxy variables, meanwhile, we also added relevant references, and the revised description was as the following:
“According to previous research, forest scarcity and the economic development are regarded as two main pathways of FT. In the former pathway, in some countries with little import ability and stable demand for forest products, the prices of the forest products rise under the spur of scarce forest resources, and promote the land conversion to woodland. Therefore, the forest area was regarded as one important potential driving factors for FT, and we used forest coverage in the framework of interpretation to reveal this characteristic. In the latter pathway, during the rapid urbanization process, the labors shift from marginal agriculture to industrial production, and the abandoned farmlands revert to forest. This pathway related to economic development and urbanization, therefore, the GDP per capita and urbanization level were put into the model. In later studies… Globalization could shift the pressure on forest resources to other countries to reduce deforestation. As a result, the ratio of export value to import value of forest products that could reveal the trade conditions of the country was designed as an independent variable. Overall, considering the availability and comparability of global data, four independent variables (Forest_r, GDP_per, Urbanization_r, Trade_r) were listed as the potential FT driving factors in this study at the global scale and continental scale.”
Point 4：Presentation of explaining variables is a bit weak.
Response 4: Many thanks! According to the comment 3, we have modified this paragraph, including the reason why we chose these variables, and their descriptions. We hope the revised version can clearly explain these variables.
Point 5：Table 2. If panel data are used, could you justify why a different constant term is not used in each regression? There is only one constant term in each regression?
Response 5: Many thanks! Actually, there is not only one constant term in each panel regression. The only one constant reported in the Stata is the average constant, according to the Stata manual. It is difficult to provide all the constant terms in the limited space. In the revised manuscript, we added the description of the constant in the table foot,
“The constant is the average constant of each regression.”
However, if my understanding is wrong, please correct me, many thanks for your patience!
Besides, according to other reviewer’s suggestions, in the revised manuscript, only developing countries with forest coverage of more than 3% are included, and we presented the list of these countries in Table A1. Because of this revision, some values in the 3.1 part have been re-calculated.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
In the new version all suggestions have been considered.
Reviewer 3 Report
In my opinion, the paper can be perfectly accepted and published in Sustainability.
