Historical Changes in Land Use and Suitability for Future Agriculture Expansion in Western Bahia, Brazil
, Allan Turini Speroto 1, Marcos Heil Costa 2 and Emily Ane Dionizio 2Round 1
Reviewer 1 Report
The manuscript is very interesting and suitable for publication in the journal. Its is well structures. Methods and data analysis are well presented. The manuscript is almost ready for publication.
I would like to ask the authors to comment on the intermediate trend analysis (per decade) for better clarity of the spatio-temporal changes (see Figure 3). For example, it is obvious that during 2000 significant changes occurred.
Author Response
Response to Reviewer 1 Comments
The manuscript is very interesting and suitable for publication in the journal. Its is well structures. Methods and data analysis are well presented. The manuscript is almost ready for publication.
Response: We would like to thank the reviewer for his/her time on this review. We are glad the reviewer considered our case study interesting and suitable for publication in this journal.
I would like to ask the authors to comment on the intermediate trend analysis (per decade) for better clarity of the spatio-temporal changes (see Figure 3). For example, it is obvious that during 2000 significant changes occurred.
Response: We improved the text by adding a new paragraph analysing the trends in land use per decade (1990-2000, 2000-2010 and 2010-2020). See the first paragraph after Table 3 in section 3.1 Land Use and Land Cover Classification. The text now reads:
At the decadal time scale, the total agriculture areas increased by 1.13 Mha in 1990-2000, 1.07 Mha in 2000-2010, and 0.970 Mha in 2010-2020. The average linear expansion rate has decreased slightly in the period (0.112, 0.107 and 0.097 Mha yr−1, respectively), but the decadal growth rate was approximately twice as high in 1990-2000 (57%) as the growth rate in 2000-2010 (35%) and 2010-2020 (23%). Rainfed crops increased by 0.833, 0.958, and 0.606 Mha from 1990-2000, 2000-2010, and 2010-2020, respectively, with a higher linear expansion rate in the first two decades (0.0833, 0.0958, and 0.0606 Mha yr−1, respectively) and a decadal growth rate of 102%, 58%, and 23%, respectively. Irrigated crops started from 24,200 ha in 1990 and increased by 49,700 and 34,600 ha from 1990-2000 and 2000-2010, respectively, but accelerated to a change of 109,000 ha during 2010-2020, with growth rates of 205%, 47% and 101%, respectively. Pasturelands had the highest growth in 1990-2000, increasing 0.274 Mha (27,400 ha yr−1, or 26% in 10 years), over five times higher than in 2000-2010 (4,980 ha yr−1, or 4% in 10 years). After that, pasturelands decreased at the rate of 2,110 ha yr−1 during 2010-2020.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Original Manuscript ID: remotesensing-1112105
Original Article Title: “Historical Changes in Land Use and Suitability for Future Agriculture Expansion in Western Bahia, Brazil“
This study is interesting, but the authors should consider these comments:
- The introduction part should be more informative, including the main objectives of your proposed study.
- The authors can add the description of the study area to the materials and methods part.
- It would be useful to provide some recent references regarding the Changes in Land Use and Suitability for Future Agriculture Expansion in the introduction and discussion parts.
- In the legend of figure 1, the symbols used to characterize the ground truth site not clear; please improve it.
- Line 105: why all images were filtered for the dry season of the Cerrado biome (from April to August)? Please explain the reason why you select this season.
- Line 108-111: need rewrite for ease read.
- Please simplify the Flowchart in Figure 2.
- The Authors should add the description and sources of the datasets used in this study.
- The Authors should add some references to support the results, and compare the results with previous work.
Sincerely Yours.
Comments for author File: Comments.docx
Author Response
Response to Reviewer 2 Comments
Original Article Title: “Historical Changes in Land Use and Suitability for Future Agriculture Expansion in Western Bahia, Brazil“
This study is interesting, but the authors should consider these comments:
Response: First, we would like to thank the reviewer for his/her time on this review. We are glad the reviewer considered our case study interesting for publication in this journal.
Point 1: The introduction part should be more informative, including the main objectives of your proposed study.
Response 1: We improved the text to make the objective more explicit. See the third paragraph in the Introduction.
Point 2: The authors can add the description of the study area to the materials and methods part.
Response 2: We added the section “2.1 Study Area” into Section 2. Materials and Methods. We included the description of the area and Figure 1 in this section.
Point 3: It would be useful to provide some recent references regarding the Changes in Land Use and Suitability for Future Agriculture Expansion in the introduction and discussion parts.
Response 3: We added several recent references related to changes in land use to the introduction and discussion. Following a request by Reviewer 3, we also added a new paragraph discussing the environmental impacts of changes in land use. A total of 13 new references were added to the manuscript. However, we could not find studies that mix the agriculture expansion with land suitability, which highlights the novelty of our study.
Point 4: In the legend of figure 1, the symbols used to characterize the ground truth site not clear; please improve it.
Response 4: We increase the stroke width of the symbols in Figure 1 to make them more visible. The Figure 1 was also moved to Section 2.1.
Point 5: Line 105: why all images were filtered for the dry season of the Cerrado biome (from April to August)? Please explain the reason why you select this season.
Response 5: All images were filtered for the dry season of the Cerrado biome (from April to August) to minimize the commission errors in the classification of forest formations and to avoid loss of information due cloud shadows. In this season, there are fewer clouds on the atmosphere than in the wet season which improves the classification algorithm. We included this information in the text to make it more informative. Please, check the fourth paragraph of section 2.2 Land Use and Land Cover Classification.
Point 6: Line 108-111: need rewrite for ease read.
Response 6: We have rewritten the text to clarify the text. See the fourth paragraph of section 2.2 Land Use and Land Cover Classification.
Point 7: Please simplify the Flowchart in Figure 2.
Response 7: We believe that this detailed flowchart helps the reader understand more clearly our methodology. Some researchers that read the manuscript before submission (see Acknowledgements) commented that this flowchart helped them understand the image processing. Thus, we preferred to keep the figure as is in order to maintain the clarity and detailing of our methods.
Point 8: The Authors should add the description and sources of the datasets used in this study.
Response 8: True enough. A new paragraph was added summarizing and referencing the sources of the datasets used in the study. (3rd paragraph in the new section 2.2 Land Use and Land Cover Classification).
Point 9: The Authors should add some references to support the results, and compare the results with previous work.
Response 9: This has been included. Please see response to point 3.
Author Response File: Author Response.docx
Reviewer 3 Report
The article “Historical Changes in Land Use and Suitability for Future Agriculture Expansion in Western Bahia, Brazil” is contributing to a better knowledge of land use dynamics at place in Brazil. The manuscript is more specifically analyzing land cover and land use change during the period 1990 to 2020 in the Western Bahia region, known for its agricultural production increase. There are two part in the manuscript, one about a remote sensing analysis of the land cover and land use change from 1990 to 2020, and a second part using a spatial multi-criteria decision analysis to identify land suitability for agriculture.
Broad comments
The manuscript is a good piece of work, it is well written and the analysis are clearly explained. My comments are mainly addressing the first part of the analysis about the land cover and land use with satellite imageries as it is my main area of expertise and focus of the journal where the authors submitted. I have a few minor comments and suggestions about the remote sensing analysis.
Yearly Landsat composites: how do you select the “best observed pixels”? I understand you have only one composite per year of analysis for the dry season, not a time series. How is the time series reduced to a single value?
Training and validation dataset: Appreciable that you have access to field data to guide the drawing of the polygons. However, I would like to have more information about the training data you used.
- Did you use a specific sampling scheme for the creation of the training and validation dataset (stratified based on another map? Random?)
- What is the logic behind sampling points in the polygons for training the RF algorithms and not using all the pixels inside the polygons?
- Can you clarify whether you use only training data from 2017 to classify the different maps? And if so, how did you deal with that?
- Can you share the number of polygons and pixels for each class of your classification in the training and validation datset?
Results:
About the accuracy assessment of the LULC maps, I do not find reference to the good practices usually used in the field of estimated areas derived from remote sensing.
Reporting the confusion matrix, per class producer and user accuracy are generally favoured to the Kappa coefficient
Foody, Giles M. 2020. “Explaining the Unsuitability of the Kappa Coefficient in the Assessment and Comparison of the Accuracy of Thematic Maps Obtained by Image Classification.” Remote Sensing of Environment 239 (August 2019): 111630. https://doi.org/10.1016/j.rse.2019.111630.
Pontius, Robert Gilmore, and Marco Millones. 2011. “Death to Kappa: Birth of Quantity Disagreement and Allocation Disagreement for Accuracy Assessment.” International Journal of Remote Sensing 32 (15): 4407–29. https://doi.org/10.1080/01431161.2011.552923.
The accuracy assessment of the change of LULC classes is a challenging task. The comparison to the existing products is interesting but the transition from one land use should be more specifically addressed or discussed in the manuscript.
Arevalo (2018) address the topic of land use change assessement over years in the context of REDD+
Arévalo, Paulo, Pontus Olofsson, and Curtis E. Woodcock. 2020. “Continuous Monitoring of Land Change Activities and Post-Disturbance Dynamics from Landsat Time Series: A Test Methodology for REDD+ Reporting.” Remote Sensing of Environment 238 (January 2018): 111051. https://doi.org/10.1016/j.rse.2019.01.013.
I am missing an information about the features importance from the 40 indexes used in the classification. Have you analyzed that?
On the land suitability part:
I have mainly a comment about the importance of the natural vegetation classes. Do I understand correctly that in the suitability analysis for agriculture and pasture land any natural vegetation can be converted if the forest/savanna/grassland is not protected by the Forest Code of 2012?
Regarding global issue of deforestation, biodiversity losses and climate change, it would be expected to have a discussion about the environmental impact of a statement such as “agricultural area could nearly double in the region”
Specific comments
I spotted an extensive use of the “we” in the text, that according to me is not always justified
Figure 1: the classes of ground truth sites are not easily identified in the map.
l.162 corrected pixel is a strange formulation
l.319 why approximately 30 m?
Figure 6 : adjust the scale for b) and c) and add the year for each data point if possible
l-149-154 describe briefly the dataset and years available
l 228-231 refer to table 2 sooner in the text
l.318 Is the final product yearly or 5-years?
Figure 9 the blue color for highly and moderately suitable is difficult to distinguish
Figure 10 I don’t find it easy to extract the relevant information from that figure. How did the Forest code of 2012 impacted the area of restricted areas?
Author Response
Response to Reviewer 3 Comments
The article “Historical Changes in Land Use and Suitability for Future Agriculture Expansion in Western Bahia, Brazil” is contributing to a better knowledge of land use dynamics at place in Brazil. The manuscript is more specifically analyzing land cover and land use change during the period 1990 to 2020 in the Western Bahia region, known for its agricultural production increase. There are two part in the manuscript, one about a remote sensing analysis of the land cover and land use change from 1990 to 2020, and a second part using a spatial multi-criteria decision analysis to identify land suitability for agriculture.
Broad comments
The manuscript is a good piece of work, it is well written and the analysis are clearly explained. My comments are mainly addressing the first part of the analysis about the land cover and land use with satellite imageries as it is my main area of expertise and focus of the journal where the authors submitted. I have a few minor comments and suggestions about the remote sensing analysis.
Response: First, we would like to thank the reviewer for his/her time on this review. We are glad the reviewer considered our case study important to contribute to the knowledge of the dynamics of land use at place in Brazil, especially in Western Bahia.
Yearly Landsat composites: how do you select the “best observed pixels”? I understand you have only one composite per year of analysis for the dry season, not a time series. How is the time series reduced to a single value?
Response: We generated thirty-one annual mosaics from 1990 to 2020 using the median values of the pixels uncontaminated by clouds and cloud shadows filtered from the images time series.
This paragraph was re-written to clarify this issue. Please, check the fourth paragraph of the (renumbered) Section 2.2 Land Use and Land Cover Classification.
Training and validation dataset: Appreciable that you have access to field data to guide the drawing of the polygons. However, I would like to have more information about the training data you used.
- Did you use a specific sampling scheme for the creation of the training and validation dataset (stratified based on another map? Random?)
- What is the logic behind sampling points in the polygons for training the RF algorithms and not using all the pixels inside the polygons?
Response to 1 and 2: This part of manuscript was re-written to clarify this issue. Please, check the sixth paragraph of section 2.2 Land Use and Land Cover Classification. We include this information:
“These field samples were then used to assist in manually drawing 2,707 similar polygons in neighbor regions of the mosaic generated for that year (112 polygons in forest formations, 121 in savanna formations, 121 in grasslands, 35 in mosaics of crop or pastures, 294 in rainfed crops, 1411 in irrigated crops, 309 in pasturelands, 204 in water bodies and 100 in urban areas/farm buildings). Random pixels were sampled inside the drawn polygons (stratified sample approach) to expand the sampling points from 120 to about 28,500 points. These training data were sufficient to train the classifier, since the use of all pixels within the polygons made the computational processing time much longer”.
- Can you clarify whether you use only training data from 2017 to classify the different maps? And if so, how did you deal with that?
Response to 3: Yes, we only used 2017 training samples to classify the entire series. We train the algorithm to learn all time series based on 2017 samples. We used space-time metrics to minimize inconsistencies in the series classification. This was described in the original version of the manuscript. Please, check the third and fourth paragraphs after Figure 2 in section 2.2 Land Use and Land Cover Classification.
- Can you share the number of polygons and pixels for each class of your classification in the training and validation datset?
Response to 4: Yes, they are included in the response to comments 1 and 2 above.
Results:
- About the accuracy assessment of the LULC maps, I do not find reference to the good practices usually used in the field of estimated areas derived from remote sensing.
Reporting the confusion matrix, per class producer and user accuracy are generally favoured to the Kappa coefficient
Foody, Giles M. 2020. “Explaining the Unsuitability of the Kappa Coefficient in the Assessment and Comparison of the Accuracy of Thematic Maps Obtained by Image Classification.” Remote Sensing of Environment 239 (August 2019): 111630. https://doi.org/10.1016/j.rse.2019.111630.
Pontius, Robert Gilmore, and Marco Millones. 2011. “Death to Kappa: Birth of Quantity Disagreement and Allocation Disagreement for Accuracy Assessment.” International Journal of Remote Sensing 32 (15): 4407–29. https://doi.org/10.1080/01431161.2011.552923.
Response to 1: We have initially used the Kappa index because it is a classic metric. We have also defined the cross-validation of data as being a more robust indicator of accuracy (See Figures 5 and 6 and associated discussion). Considering this comment, we removed the index and re-wrote the text including the references suggested. Please, check the first paragraph after Figure 4 in Section 3.1. The main validation is associated to Figures 5 and 6.
- The accuracy assessment of the change of LULC classes is a challenging task. The comparison to the existing products is interesting but the transition from one land use should be more specifically addressed or discussed in the manuscript.
Arevalo (2018) address the topic of land use change assessement over years in the context of REDD+
Arévalo, Paulo, Pontus Olofsson, and Curtis E. Woodcock. 2020. “Continuous Monitoring of Land Change Activities and Post-Disturbance Dynamics from Landsat Time Series: A Test Methodology for REDD+ Reporting.” Remote Sensing of Environment 238 (January 2018): 111051. https://doi.org/10.1016/j.rse.2019.01.013.
Response to 2: Reviewer 1 has also requested this. The manuscript has been re-written and we included over a page of discussion for the transition between LULC. First, we described a summarized discussion per decade, followed by a detailed discussion of the transition between 1990 and 2020. A temporal evaluation of the land use transition, as described by Arevalo et al., (2018) would be an article by itself, and is beyond our goals here. However, we provide the entire database available for download for anyone that wants to do this analysis. Here, we show the year-by-year changes in Figure 4, and the transitions of each class of land use between 1990 and 2020 in Table 3.
- I am missing an information about the features importance from the 40 indexes used in the classification. Have you analyzed that?
Response to 3: In the 8th paragraph, before Figure 2, we explained that were used correlations between bands and a PCA analysis, to select the 14 indexes for the final classification. No additional analysis was made to select them.
On the land suitability part:
- I have mainly a comment about the importance of the natural vegetation classes. Do I understand correctly that in the suitability analysis for agriculture and pasture land any natural vegetation can be converted if the forest/savanna/grassland is not protected by the Forest Code of 2012?
Response to 1: The reviewer is partially correct. Any area of natural vegetation unprotected by the Forest Code of 2012 or by APAs can be converted to cropland or pasture. But there are also other limitations related to relief and climate, that may render the land unsuitable for agriculture.
- Regarding global issue of deforestation, biodiversity losses and climate change, it would be expected to have a discussion about the environmental impact of a statement such as “agricultural area could nearly double in the region”
Response to 2: We agree with the reviewer. We included the discussion of the environmental impacts to Section 4, with a new paragraph dedicated to this. This information has been also added to the conclusions.
Specific comments
I spotted an extensive use of the “we” in the text, that according to me is not always justified
Response: That is a matter of style. We revised the manuscript to avoid the use of first person in some cases to please the reviewer, but mostly we follow the manual of Style – Ten Lessons in Clarity and Grace, by Joseph M. Williams. In Chapter 3, for the sake of clarity, the manual recommends to clearly explicit characters and actions in a sentence. To avoid any doubts about whether a procedure was done by the authors or taken from someone else, for every procedure done by the authors, we explicit say that “we did it”.
Figure 1: the classes of ground truth sites are not easily identified in the map.
Response: This was also requested by Reviewer 2. We improved the Figure 1 to make it clearly.
l.162 corrected pixel is a strange formulation
Response: We changed the word “corrected” to “replaced”.
l.319 why approximately 30 m?
Response: 30 meters is a nominal resolution. A more accurate term is 1 arc-second (approximately 30 meters). We included this information on manuscript.
Figure 6 : adjust the scale for b) and c) and add the year for each data point if possible
Response: The idea of maintaining the same scale in Figure 6 a, b and c is precisely to show the different proportions between the crops, pasturelands and mosaics of pasture and crops in relation to the 1:1 line. The largest errors are in the mosaics, but they represent a relatively small amount of area. The dots are very clustered, labeling the dots will make the image very confusing.
l-149-154 describe briefly the dataset and years available
Response: We improved the text to clarify this issue. Please, check the second paragraph after Figure 2 in section 2.2 Land Use and Land Cover Classification.
l 228-231 refer to table 2 sooner in the text
Response: We added the reference to Table 2 sooner in the text. Please, verify the second paragraph in section 2.3.2. Criteria Thresholds.
l.318 Is the final product yearly or 5-years?
Response: The final product of LULCC is yearly. We improved the Figure 3 caption to clarify this issue.
Figure 9 the blue color for highly and moderately suitable is difficult to distinguish
Response: Maybe it is a monitor or printer calibration issue. The shades of blue are quite distinct in the scale of Figure 8 and in Figure 10, for example. The shades are the same used in figures 8a, b, 9, and 10. Anyway, the suitability maps are available for download at htts://obahia.dea.ufv.br. Anyone interested in plotting these maps can download the files from there.
Figure 10 I don’t find it easy to extract the relevant information from that figure. How did the Forest code of 2012 impacted the area of restricted areas?
Response: This figure is a synthesis of the entire paper, visually showing the past expansion of agriculture and irrigation, the amount of preservation areas, and a mix of the potential for agriculture expansion across each suitability class. The impact of the Forest Code is clearly visible in Figure 10. There is twice as much area protected as remaining highly suitable areas for agriculture expansion (we included this sentence in the discussion text).
Author Response File: Author Response.docx
