Reconstructing Three Decades of Land Use and Land Cover Changes in Brazilian Biomes with Landsat Archive and Earth Engine

Round 1

Reviewer 1 Report

Comments on "Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine".

 

The paper assessed and evaluated land use and land cover changes over Brazil for 1980 to 2017 using Landsat series, which is an important topic to addressed. The paper reads well, and methods are well presented. Overall, it deserves publication by Remote Sensing, but before that, I want the authors to clarify a few things.

(i) Forest definition: I like to see more detailed descriptions about land cover type definitions (more than shown in Table 2), especially for forests. Do they comply with the FAO forest definition (at least 0.5 ha size with canopy, density greater than 10%, tree height greater than 5 m) or something original?

 (ii) Comparison with the existing global land cover data: How do interannual patterns of the new LULC differ from the global data, such as Land Use Harmonization (LUH)? It is important to know the degree of consistency and inconsistency between the global and region-specific data.

 

Author Response

Reviewer 1

Open Review

(x) I would not like to sign my review report 
( ) I would like to sign my review report 

English language and style

( ) Extensive editing of English language and style required 
( ) Moderate English changes required 
( ) English language and style are fine/minor spell check required 
(x) I don't feel qualified to judge about the English language and style 

 

	Yes	Can be improved	Must be improved	Not applicable
Does the introduction provide sufficient background and include all relevant references?	(x)	( )	( )	( )
Is the research design appropriate?	(x)	( )	( )	( )
Are the methods adequately described?	(x)	( )	( )	( )
Are the results clearly presented?	(x)	( )	( )	( )
Are the conclusions supported by the results?	(x)	( )	( )	( )

Comments and Suggestions for Authors

Comments on "Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine".

The paper assessed and evaluated land use and land cover changes over Brazil for 1980 to 2017 using Landsat series, which is an important topic to addressed. The paper reads well, and methods are well presented. Overall, it deserves publication by Remote Sensing, but before that, I want the authors to clarify a few things.

(i) Forest definition: I like to see more detailed descriptions about land cover type definitions (more than shown in Table 2), especially for forests. Do they comply with the FAO forest definition (at least 0.5 ha size with canopy, density greater than 10%, tree height greater than 5 m) or something original?

Answer: thanks for pointing that out. We have included all the LULC class descriptions in the S2 table in the supplementary material indicating also how the MapBiomas LULC classification scheme translates to FAO, IBGE and other classification scheme. We stated in the manuscript that the Forest class includes old growth mature forest (i.e., > 30-year-old), early stage (i.e., 5-15-year-old) and advanced secondary growth (i.e., 15-30-year old) forests, pristine forests that have not undergone anthropogenic conversion, savanna woodlands, mangroves and forest plantation. The minimum mapping unit of MapBiomas methodology is 0.5 ha, now clearly stated in the Spatial filter section.

 (ii) Comparison with the existing global land cover data: How do interannual patterns of the new LULC differ from the global data, such as Land Use Harmonization (LUH)? It is important to know the degree of consistency and inconsistency between the global and region-specific data.

Answer: No attempt has been made to compare the MapBiomas LULC results with other LULC regional and global products in this manuscript. This is an ongoing research question for our group. However, we made clear in the discussion that we are pursuing this path, and added the results of a study that compared out LULC map of 2015 from MapBiomas with their LULC map produced with PROBA-V multispectral images for all Brazilian biomes at 100 m pixel size. The level of agreement among the main LULC of these maps found in this study was 69%.

 

Author Response File: Author Response.docx

Reviewer 2 Report

Review report: remotesensing-886105

Manuscript Number: remotesensing-886105

Title: Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine

General remarks

This study provided methods, results, and analysis are potentially useful for science and society. Thus, I recommend this manuscript for acceptance for publication in the journal of Remote Sensing after Minor revision from the authors.

Detailed comments

Title

The title of the manuscript is fine and it cover overall content of the manuscript.

Abstract

The abstract of the manuscript written finely.

1. Introduction

Line 76: [3,5,6]. Please strictly avoid "Reference Lumping". Instead, please locate them where they exactly belong.

 

Line 85:  [13-16], Strictly avoid such "Reference Lumping". Instead, please locate them where they exactly belong.

 

Line 100:  [29-31], Avoid such "Reference Lumping". Instead, please locate them where they exactly belong.

 

2. Materials and Methods

Line 155: Figure 1. Need to improve this figure as:

[1] Show Brazil’s map in the right side up part framework, and highlight the study area.

[2] Below of the right side framework, show province map and highlight the study area. Further, add grid line with latitude and longitude.

[3] Left side in main map, show the details of the study area with Brazilian biomes and if possible the color DEM map with elevation variation.

In this way, the readers easily get the idea about geographical features of the study area.

The other overall write-up of this section is fine.

3. Results

The analysis of the results section provided in details. It helpful for further researchers to know the details findings of such study.

 

4. Discussion

The discussion provided in this section concise and analysis in a good way.

 

5. Conclusions

The conclusions of the manuscript is concise and fine.

Author Response

Reviewer 2

Open Review

(x) I would not like to sign my review report 
( ) I would like to sign my review report 

English language and style

( ) Extensive editing of English language and style required 
( ) Moderate English changes required 
(x) English language and style are fine/minor spell check required 
( ) I don't feel qualified to judge about the English language and style 

 

	Yes	Can be improved	Must be improved	Not applicable
Does the introduction provide sufficient background and include all relevant references?	(x)	( )	( )	( )
Is the research design appropriate?	(x)	( )	( )	( )
Are the methods adequately described?	(x)	( )	( )	( )
Are the results clearly presented?	(x)	( )	( )	( )
Are the conclusions supported by the results?	(x)	( )	( )	( )

Comments and Suggestions for Authors

Review report: remotesensing-886105

Manuscript Number: remotesensing-886105

Title: Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine

General remarks

This study provided methods, results, and analysis are potentially useful for science and society. Thus, I recommend this manuscript for acceptance for publication in the journal of Remote Sensing after Minor revision from the authors.

Detailed comments

Title

The title of the manuscript is fine and it cover overall content of the manuscript.

Answer: Thanks!

Abstract

The abstract of the manuscript written finely.

Answer: Thanks!

1. Introduction

Line 76: [3,5,6]. Please strictly avoid "Reference Lumping". Instead, please locate them where they exactly belong.

Answer: Thanks for pointing that out. In fact, we noticed that our references were mostly wrongly cited because the citation management software used lost the links with the references. We corrected the citation problems. We also made sure that the references were not lumped.

Line 85:  [13-16], Strictly avoid such "Reference Lumping". Instead, please locate them where they exactly belong.

Answer: Fixed. Unnecessary references were removed. Included only [13].

Line 100:  [29-31], Avoid such "Reference Lumping". Instead, please locate them where they exactly belong.

Answer: Fixed, now [27–29], and these references are not lumping. They sustain the fact that conservation policies for the Pampa biome is being neglected.

27. Overbeck, G.E.; Müller, S.C.; Fidelis, A.; Pfadenhauer, J.; Pillar, V.D.; Blanco, C.C.; Boldrini, I.I.; Both, R.; Forneck, E.D. Brazil’s neglected biome: The South Brazilian Campos. Perspect. Plant Ecol. Evol. Syst. 2007, 9, 101–116.
28. Carlucci, M.B.; Luza, A.L.; Hartz, S.M.; Duarte, L.D.S. Forests, shrublands and grasslands in southern Brazil are neglected and have specific needs for their conservation. Reply to Overbeck et al. Nat. e Conserv. 2016, 14, 155–157.
29. Overbeck, G.E.; Vélez-Martin, E.; Scarano, F.R.; Lewinsohn, T.M.; Fonseca, C.R.; Meyer, S.T.; Müller, S.C.; Ceotto, P.; Dadalt, L.; Durigan, G.; et al. Conservation in Brazil needs to include non-forest ecosystems. Divers. Distrib. 2015, 21, 1455–1460.

 

1. Materials and Methods

Line 155: Figure 1. Need to improve this figure as:

[1] Show Brazil’s map in the right side up part framework, and highlight the study area.

[2] Below of the right side framework, show province map and highlight the study area. Further, add grid line with latitude and longitude.

[3] Left side in main map, show the details of the study area with Brazilian biomes and if possible the color DEM map with elevation variation.

In this way, the readers easily get the idea about geographical features of the study area.

The other overall write-up of this section is fine.

Answer: We have improved Figure 1. Thanks for your suggestion. Now, Brazil’s map is on the right side, the country boundary is highlighted to indicate the study area, including the biome and state (provinces) boundaries in the legend (on the right bottom side). We provided also more geographic context of South America and included a subtle shaded DEM background.

1. Results

The analysis of the results section provided in details. It helpful for further researchers to know the details findings of such study.

 Answer: Thanks!

1. Discussion

The discussion provided in this section concise and analysis in a good way.

 Answer: Thanks! However, we’ve made minor changes in the Conclusions as requested by another reviewer.

1. Conclusions

The conclusions of the manuscript is concise and fine.

 Answer: Thanks! However, we’ve made changes in the Conclusions as requested by another reviewer.

 

Reviewer 3 Report

„Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine” is an interesting article. Authors set three research goals to present the method of reconstructing the annual time-series of LULC maps for all the Brazilian biomes during 1985 and 2017, to assess the extent, rates and main drivers of LULC change in the Brazilian biomes between 1985 and 2017 and to present the MapBiomas image processing and classification protocol which map main land cover classes separately for each biome and common cross-cutting land use themes.

The applied methodology is complex (composed of 7 stages) but adequate and presented understandably. The results are elaborated in enough detail and in line with the aim of the paper. I find the paper acceptable for publication after minor revision.

 

Some remarks & questions:

 

Page 8–9 rows 258–273, whether the use of a different type of samples for training the random forest classifier for Amazon biome and other biomes did not affect the obtained classification results

Page 15 row  461 Figure 5. Does not show “Land cover and land use transformation between 1985 and 2017 in Brazil”, but the state LULC status in 1985 and 2017.

Page 18 row 528–529 Figure 7 no legible marking of parts A and B. There are polygons on the map with white borders, is it an administrative division? the division into biomes is missing in part of the figure. Unnecessary “).”

Presented drawings 5 and 7  difficult to read. I think the article would gain if LULC changes would be presented on the maps for each biome.

 

I suggest complementing the description of “LULC Spatial and Temporal Trends in Brazil” with possible factors that influenced the described changes in every biome. For example, "Amazon region has lost much more forest area than the other biomes in Brazil" which is a well-known phenomenon and its reasons are the subject of many studies, but "The Pampa and the Atlantic Forest showed an increase in forest cover in this period ", which is interesting considering that “the Forest class diminished (...) at country level”. It is very interesting what factors influenced these changes. I also suggest paying more attention to the analysis of spatial distribution of LULC changes in biomes.

Author Response

Reviewer 3

Open Review

( ) I would not like to sign my review report 
(x) I would like to sign my review report 

English language and style

( ) Extensive editing of English language and style required 
( ) Moderate English changes required 
( ) English language and style are fine/minor spell check required 
(x) I don't feel qualified to judge about the English language and style 

 

	Yes	Can be improved	Must be improved	Not applicable
Does the introduction provide sufficient background and include all relevant references?	(x)	( )	( )	( )
Is the research design appropriate?	(x)	( )	( )	( )
Are the methods adequately described?	(x)	( )	( )	( )
Are the results clearly presented?	( )	(x)	( )	( )
Are the conclusions supported by the results?	( )	(x)	( )	( )

Comments and Suggestions for Authors

„Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine” is an interesting article. Authors set three research goals to present the method of reconstructing the annual time-series of LULC maps for all the Brazilian biomes during 1985 and 2017, to assess the extent, rates and main drivers of LULC change in the Brazilian biomes between 1985 and 2017 and to present the MapBiomas image processing and classification protocol which map main land cover classes separately for each biome and common cross-cutting land use themes.

The applied methodology is complex (composed of 7 stages) but adequate and presented understandably. The results are elaborated in enough detail and in line with the aim of the paper. I find the paper acceptable for publication after minor revision.

 Some remarks & questions:

Page 8–9 rows 258–273, whether the use of a different type of samples for training the random forest classifier for Amazon biome and other biomes did not affect the obtained classification results

Answer: The proposed LULC mapping methodology gives autonomy for the biomes to adapt the parameters of the mapping of the protocol. One of the parameters is how to sample to train data for the random forest classifier. The classification results are also evaluated independently. The mapping integration requires each biome to generate LULC that are spatially coherent specially along the biome transitions. This is the requirement to accept the independent classification results of each biome. We have added this explanation to the discussion to clarify this issue. Thanks for bringing that up.Please, see Lines 709-713 in the Discussion for that:

“Besides the iterative mapping collections, we implemented a flexible LULC mapping protocol which allows each biome to define the feature space and samples for training the random forest classifier (S1 Appendix), as long as the biome maps can follow the map integration protocol to guarantee spatial and temporal coherence along the their transitional ecotone zones.”

 

 

Page 15 row  461 Figure 5. Does not show “Land cover and land use transformation between 1985 and 2017 in Brazil”, but the state LULC status in 1985 and 2017.

Answer: Thanks for pointing that out. We’ve fixed that by deleting the word ‘transformation’ and highlighting that each map depicts LULC status in each year. Figure 5 was also graphically improved, and now highlights only the biome boundaries, as requested in your comment below.

Page 18 row 528–529 Figure 7 no legible marking of parts A and B. There are polygons on the map with white borders, is it an administrative division? the division into biomes is missing in part of the figure. Unnecessary “).”

Answer: Figure 7 was improved for legibility. Now, we are indicating ‘right’ and ‘left’ for parts A and B, respectively. We also deleted the “)” typo error.

Presented drawings 5 and 7 difficult to read. I think the article would gain if LULC changes would be presented on the maps for each biome.

Answer: The administrative boundaries of Figures 5 and 7 were removed, and we kept only the biome boundaries (which are indicated in the map legends).

I suggest complementing the description of “LULC Spatial and Temporal Trends in Brazil” with possible factors that influenced the described changes in every biome. For example, "Amazon region has lost much more forest area than the other biomes in Brazil" which is a well-known phenomenon and its reasons are the subject of many studies, but "The Pampa and the Atlantic Forest showed an increase in forest cover in this period ", which is interesting considering that “the Forest class diminished (...) at country level”. It is very interesting what factors influenced these changes. I also suggest paying more attention to the analysis of spatial distribution of LULC changes in biomes.

Answer: Thanks for highlighting that. The section “LULC Spatial and Temporal Trends in Brazil” focuses on LULC changes at the country level. We present the results of LULC in the “3.4 LULC Biome Transitions” section (please, refer to Figure 7 and Tables 7 and 8). We added ‘Biome’ to this section title to make it clearer.

 

Reviewer 4 Report

Review Summary:

I enjoyed reading this paper entitled " Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine" with a focus on LULC classification and change analysis on Brazilian biomes. Authors laborious attempt to generate LULC for the entire country of Brazil for almost the entire period of Landsat collections on an annual basis is a praiseworthy work. I see how many sectors can utilize these data ranging from making action plans by policy makers to studying associated processes on land surface, sub-surface, and atmosphere by researchers. I have no doubt that the level of data analysis work required by this project is huge. I think the authors have an opportunity to strengthen the article by meticulously documenting their methodology and interpreting the results. I also have a view that authors need a solid point on how one would still be able to use the results, such as a decrease of forest by 10% etc., while this percentage falls within their error margin.

I have outlined my comments below in two sections: Major Comments and Specific Comments

Major comments:

1. The authors need to very clearly state in their methodology section about how they come up with one single annual product from the series of Landsat data. I see some wet period/ dry period or June to October mentioned for different biomes in their supplemental files. Crux of this description deserves to be stated on the main article and then it can be supplemented by supplementary materials.  

2. Authors also need to state what modifications were needed in computations while using Landsat 5 vs. 7 vs. OLI.

3. Authors should also report and discuss the classification accuracy for each year.

4. With a sound interpretation incorporating local knowledge, authors need to discuss enough on why it was expected that different biomes have different rate of change for each LULC type.

5. The authors mentioned in their introduction section that there were many previous LULC prepared by other agencies, hence a discussion on how the accuracy/application of this dataset is superior than previous ones would be helpful.

6. Conclusion: conclusion section needs to be rewritten. The points made in lines 650-653 are good. Authors need to sum up what did they learn from: the use of more than 3-decades of Landsat data, cloud-computing platform of GEE to carry out the entire workflow, use of one reference data versus other, classifying in one biomes or themes  vs other, classification accuracy in wet year vs. dry year, accuracy variation while using image from June vs. image from October. These highlights in conclusion section will properly conclude your work and will be helpful for those willing to expand/reproduce this work to their region. 

Specific Comments:

Line 83: use of and (English)

Line 88, 89 and line 93 give contradicting meaning: which biome is the most altered one? Clarify it.

Line 95: pressure of its natural vegetation – unclear mostly because of the preposition “of”, clarify

Line 104 and 105: explain why some biomes were studied detail and some were not.

Line 107: IBGE- spell it out, keep spaces consistent between years

Line 118: closed square bracket without an open square bracket

Line 121 and 122: A sentence on how cloud computing allows to select cloud-free pixels would be helpful.

126-127: clarify what do you mean by step-wise process based on LULC collection

Line 127: This study focuses on? (word In is not required)

Line 131: which maps (English)

Line 149: the Biomes? (English)

Line 175: sentence fragment is used after but, no complete sentence (English)

Table 2: In writing the class name “non vegetated area” make word case consistent with prior writing

Table 2: The description of classes/sub-classes have used qualitative terms such as predominance. For instance, predominance of herbaceous is used in grassland and predominance of tree species is used in forest. There are several other instances where the qualitative terms are used. The quantitative description would be helpful if available.

Lines 180-187: the subclass listing is not consistent with the table 2. For example, Non-vegetated area has rocky outcrop and mining as separate level 2 classes on the table, but not in the description.

Line 190: Spell out what does S2 mean?

Line 201: Needs reference for orthorectification part

Line 205: use lower level subheadings e.g. 2.3.1 Preprocessing

Line 213-216: consider rewriting the sentence to make it more clear.

Line 218: “combined to produce annual temporal mosaics targeting specific period of the year” this is the key step of your LULC generation and need more detail on what specific period for which biome and why?

Line 229-232: Parenthesis opened, but not closed.

Line 242-243: Explain why the temporal mosaic procedure is not optimal for certain LULC

Line 248: varied over time (?) (English)

249: sentence unclear because of use of “and” in too many places without putting other punctuation marks

Line 266: “that did not changed” (English)

Line 269: assessed? (English)

Line 270: Rewrite to clarify – reclassify to the correct on if so? (English)

Line 295: ti (write i in subscript)

Line 309: rather than using “similar to described above”, first label each sub-section with their subheading no. (see comment given for line 205)

Line 312-314: rewrite and clarify

Line 315: Capitalize each word

325-329: the text is matching with what they provide on S3 Appendix.

Line 363-368: I could not become clear on what I was reading on this section.

Figure 3: I don’t see a specific reason why authors choose this kind of graph instead of regular line or bar graphs.

Figure 4: Why the non-vegetated areas show no signal of absolute change in LULC area in any biome?

Line 416: they used the word circa and then in the parenthesis they noted figure number etc.

Line 450: are shown? (English)

Line 452: Call it class level 2 to be consistent, not legend level 2

Line 457: lay? (English)

607-609: I would rather prefer to state what specifics could make the results better rather than implying that progresses are being made and will be published in the future

Line 644: In this study, we present an? (English)

Authors have included a sentence of acknowledgement for their data and analysis platform Earth Engine on their conclusion section. Take it out from the conclusion.

 

Best Wishes,

 

Author Response

Reviewer 4

Open Review

(x) I would not like to sign my review report 
( ) I would like to sign my review report 

English language and style

( ) Extensive editing of English language and style required 
(x) Moderate English changes required 
( ) English language and style are fine/minor spell check required 
( ) I don't feel qualified to judge about the English language and style 

 

	Yes	Can be improved	Must be improved	Not applicable
Does the introduction provide sufficient background and include all relevant references?	( )	(x)	( )	( )
Is the research design appropriate?	(x)	( )	( )	( )
Are the methods adequately described?	( )	( )	(x)	( )
Are the results clearly presented?	( )	(x)	( )	( )
Are the conclusions supported by the results?	( )	( )	(x)	( )

Comments and Suggestions for Authors

Review Summary:

I enjoyed reading this paper entitled " Reconstructing three decades of land use and land cover changes in Brazilian biomes with Landsat Archive and Earth Engine" with a focus on LULC classification and change analysis on Brazilian biomes. Authors laborious attempt to generate LULC for the entire country of Brazil for almost the entire period of Landsat collections on an annual basis is a praiseworthy work. I see how many sectors can utilize these data ranging from making action plans by policy makers to studying associated processes on land surface, sub-surface, and atmosphere by researchers. I have no doubt that the level of data analysis work required by this project is huge. I think the authors have an opportunity to strengthen the article by meticulously documenting their methodology and interpreting the results. I also have a view that authors need a solid point on how one would still be able to use the results, such as a decrease of forest by 10% etc., while this percentage falls within their error margin.

I have outlined my comments below in two sections: Major Comments and Specific Comments

Answer: Thanks for the detailed and constructive review of our manuscript. We answer the comments in the detail below.

Regarding your question in the Review Summary above about the ability to use our results to estimate changes, we reinforced that the change analysis focus on the +30-year period and that further investigation is necessary to understand the impact of LUCC classification error to estimate yearly change.

Major comments:

1. The authors need to very clearly state in their methodology section about how they come up with one single annual product from the series of Landsat data. I see some wet period/ dry period or June to October mentioned for different biomes in their supplemental files. Crux of this description deserves to be stated on the main article and then it can be supplemented by supplementary materials.

 Answer: Thanks for highlighting this point. We added an explanation that each biome has a reference period for building the annual mosaics which is kept fixed over time (lines 243-248). Therefore, we are able to estimate annual change at the biome level, but we don’t produce an annual mosaic at the country level.

Regarding the wet/dry periods, we used this information as a feature attribute input to the random forest classifier. This is clearer now in the method section.

1. Authors also need to state what modifications were needed in computations while using Landsat 5 vs. 7 vs. OLI.

Answer: we used TOA Landsat Tier 1 imagery data for Landsat 5, 7 and 8, which is already orthorectified based on ground control points and digital elevation model. We did not apply any normalization for these imagery data sets because the data is already in TOA reflectance. We improved the explanation for this in the 2.3 Remote Sensing Dataset section.

2. Authors should also report and discuss the classification accuracy for each year.

Answer: The classification accuracy for each year during 1985 and 2017 is now available in the S2 Figure and S2 Dataset. In Tables 4 and 5, we provide the average accuracy, standard error at LULC classes in Levels 1 and 2 in each biome and Brazil during 1985 to 2017. The variability of the accuracy for each year did not vary much (i.e., <1%) according to the standard error. We made this clear now in the results (section 3.1 LULC Map Accuracy).

4. With a sound interpretation incorporating local knowledge, authors need to discuss enough on why it was expected that different biomes have different rate of change for each LULC type.

Answer: The rate of change is not affected by the local knowledge incorporated in the LULC classification. The knowledge of local experts is key for building the annual mosaics and for training the random forest. The different rate of change of each LULC class is due to the different drivers of change of each biome. We provided a summary of the main land use and land cover drivers of each biome in the introduction and study area sections highlighting that.

5. The authors mentioned in their introduction section that there were many previous LULC prepared by other agencies, hence a discussion on how the accuracy/application of this dataset is superior than previous ones would be helpful.

Answer: we refrained from including the comparison of the LULC products because it requires using a common reference data for accuracy assessment and spatial agreement analysis. However, we added more information about this issue as requested also by Reviewer 1. Please, see the answer for Reviewer 1 below:

“No attempt has been made to compare the MapBiomas LULC results with other LULC regional and global products in this manuscript. This is an ongoing research question for our group. However, we made clear in the discussion that we are pursuing this path, and added the results of our study that compared out LULC map of 2015 with their LULC map produced with PROBA-V multispectral images for all Brazilian biomes at 100 m pixel size. The level of agreement among the main LULC of these maps found in this study was 69%.”

6. Conclusion: conclusion section needs to be rewritten. The points made in lines 650-653 are good. Authors need to sum up what did they learn from: the use of more than 3-decades of Landsat data, cloud-computing platform of GEE to carry out the entire workflow, use of one reference data versus other, classifying in one biomes or themes  vs other, classification accuracy in wet year vs. dry year, accuracy variation while using image from June vs. image from October. These highlights in conclusion section will properly conclude your work and will be helpful for those willing to expand/reproduce this work to their region. 

Answer: Thanks for pointing that out. We have included most of your suggestion in the Conclusion.

Specific Comments:

Line 83: use of and (English)

Answer: Rephrased.

Line 88, 89 and line 93 give contradicting meaning: which biome is the most altered one? Clarify it.

Answer: Clarified.

Line 95: pressure of its natural vegetation – unclear mostly because of the preposition “of”, clarify

Answer: Clarified.

Line 104 and 105: explain why some biomes were studied detail and some were not.

Answer: Explained.

Line 107: IBGE- spell it out, keep spaces consistent between years

Answer: Corrected.

Line 118: closed square bracket without an open square bracket

Answer: Corrected.

Line 121 and 122: A sentence on how cloud computing allows to select cloud-free pixels would be helpful.

Answer: Suggestion incorporated.

126-127: clarify what do you mean by step-wise process based on LULC collection

Answer: Step-wise term replaced by a clearer phrase.

Line 127: This study focuses on? (word In is not required)

Answer: Phrase deleted.

Line 131: which maps (English)

Answer: Thanks, we added it.

Line 149: the Biomes? (English)

Answer: Rephrased.

Line 175: sentence fragment is used after but, no complete sentence (English)

Answer: Sentence fragment adjusted.

Table 2: In writing the class name “non vegetated area” make word case consistent with prior writing

Answer: Adjusted.

Table 2: The description of classes/sub-classes have used qualitative terms such as predominance. For instance, predominance of herbaceous is used in grassland and predominance of tree species is used in forest. There are several other instances where the qualitative terms are used. The quantitative description would be helpful if available.

Answer: No quantitative description is available, as the intent was to make a general description of a class that includes a variety of phytophisiognomies.

Lines 180-187: the subclass listing is not consistent with the table 2. For example, Non-vegetated area has rocky outcrop and mining as separate level 2 classes on the table, but not in the description.

Answer: Missing subclasses included.

Line 190: Spell out what does S2 mean?

Answer: S2 Table refers to second table shown at the Supplementary Material section and nomenclature follows journal guideline.

Line 201: Needs reference for orthorectification part

Answer: Reference [47] accounts for that.

[47]Zhu, Z. Change detection using landsat time series: A review of frequencies, preprocessing, algorithms, and applications. ISPRS J. Photogramm. Remote Sens. 2017, 130, 370–384.

Line 205: use lower level subheadings e.g. 2.3.1 Preprocessing

Answer: Subheadings inserted.

Line 213-216: consider rewriting the sentence to make it more clear.

Answer: Fixed. Now it reads:

“The first pre-processing step was to build cloud-free annual tile Landsat mosaics yearly. Cloud and cloud shadow masks were applied to all Landsat scenes. For that, we used the Temporal Dark Outlier Mask (TDOM) [48] algorithm and the Band Quality Assessment (BQA) information available in the Landsat Collection.”

Line 218: “combined to produce annual temporal mosaics targeting specific period of the year” this is the key step of your LULC generation and need more detail on what specific period for which biome and why?

Answer: We included a new sentence detailing that, and added information about the specific annual period for mosaicking for each biome in the S3 Table (lines 251-256). Now it reads:

“Each biome had the flexibility to define the optimal period of the year to build the annual mosaics because the cloud condition and phenological behavior of the LULC varies across the biomes. Therefore, the annual image mosaics have different reference period for each biome. This characteristic of the mosaics does not affect LULC annual change estimates because the reference period of each biome is the same over time, allowing to build a single annual product at that level (S3 Table).”

Line 229-232: Parenthesis opened, but not closed.

Answer: “(see” was deleted.

Line 242-243: Explain why the temporal mosaic procedure is not optimal for certain LULC

Answer: This happens because not all LULC classes have the optimal spectral separability in a given optimal period. For example, natural grassland and pasture have different phenological behavior over the year.

Line 248: varied over time (?) (English)

Answer: “overtime” was replaced by “temporally”.

249: sentence unclear because of use of “and” in too many places without putting other punctuation marks

Answer: Adjusted.

Line 266: “that did not changed” (English)

Answer: Corrected.

Line 269: assessed? (English)

Answer: Corrected.

Line 270: Rewrite to clarify – reclassify to the correct on if so? (English)

Answer: Rewritten.

Line 295: ti (write i in subscript)

Answer: Corrected.

Line 309: rather than using “similar to described above”, first label each sub-section with their subheading no. (see comment given for line 205)

Answer: Corrected.

Line 312-314: rewrite and clarify

Answer: Done.

Line 315: Capitalize each word

Answer: Corrected.

325-329: the text is matching with what they provide on S3 Appendix.

Answer: This text was excluded from the S3 Appendix.

Line 363-368: I could not become clear on what I was reading on this section.

Answer: This phrase was rewritten.

Figure 3: I don’t see a specific reason why authors choose this kind of graph instead of regular line or bar graphs.

Answer: We used a panel of area graph (which is similar to a bar graph because of the high data density of the time-series) with the Loess smoothing function because we wanted to highlight the trend behavior of each class.

Figure 4: Why the non-vegetated areas show no signal of absolute change in LULC area in any biome?

Answer: The non-vegetated area has smaller coverage/extent in all biomes that cannot be depicted in the graph scale. In fact, this class has a change signal that is showed better in the relative change graph (panel B of Figure 4). The non-vegetated area class has a high (relative) variability, and also shows different behavior amongst the biomes.

Line 416: they used the word circa and then in the parenthesis they noted figure number etc.

Answer: Deleted “with a peak circa”.

Line 450: are shown? (English)

Answer: Corrected.

Line 452: Call it class level 2 to be consistent, not legend level 2

Answer: Corrected.

Line 457: lay? (English)

Answer: English reviewed.

607-609: I would rather prefer to state what specifics could make the results better rather than implying that progresses are being made and will be published in the future

Answer: The first paragraph of the discussion states the value of the results, and the second one demonstrates how the results are being used in science and societal applications. The statement below (lines 607-609 referred in this comment) explains further steps that our research is conducting to understand the LULC transitions and estimate its uncertainties.

“However, our research group of the MapBiomas Project is already exploring methods to understand the frequency t pixel changes its LULC class and the number of times it happens [53,65]”

Line 644: In this study, we present an? (English)

Answer: Corrected.

Authors have included a sentence of acknowledgment for their data and analysis platform Earth Engine on their conclusion section. Take it out from the conclusion.

Answer: This acknowledgment was deleted in the conclusion.

Best Wishes,