Review Reports - Influence of Avocado Plantations as Driver of Land Use and Land Cover Change in Chile’s Aconcagua Basin

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study presents an innovative and well-supported analysis of the expansion of avocado crops in the Aconcagua Basin in Chile. The analysis is conducted for three time periods, using satellite imagery for three periods, using Random Forest for land-use classification, and various metrics for statistical and timely analysis of changes and transitions. Additionally, an analysis of fragmentation and connectivity of avocado crop patches and natural vegetation is included. These analyses significantly refine data interpretation and highlight the negative effects of allowing the expansion of these crops. This analysis highlights the effects of these crops and other types of orchards on natural vegetation. The information presented serves as a baseline for guiding public policies that regulate the expansion of agricultural activities in the region, limiting environmental impacts and impacting the ecological balance of the basin. The text is clear and well-structured.

Below I describe some elements that may contribute to strengthen the understanding and scope of the article.

In table 1: in Fruit Farm Land description add “except Avocado land. Also, add source at the end of the table.

Line 183: Specify the filter data applied to the date

Line 207 add (RF) after Random Forest

Can the author add more details for validate the land use data?

Give data about the overall accuracy (OA) and a Kappa index for each period analized.

Lines 342 to 344 are information that should be included in the methodology, not in the results.

Also for these lines: “Scenario 1 (S1): “avocado decrease”. Replacement of the avocado land above 550 m by native vegetation. - Scenario 2 (S2): “avocado expansion”. A 200-meter buffer around existing avocado patches. - Scenario 3 (S3): “Ecological restoration”. Complete replacement of avocado plantations by native vegetation.”

Why 550 m in S1, 200 m in S2? Can citations or references be added to justify this parameter?

Author Response

Dear reviewer:

We appreciate your helpful comments and suggestions. As a result, our manuscript has improved significantly. Additionally, some figures were improved, and some important comments were included in the methodology and discussion. The new changes appear in the new manuscript (yellow color).

Point No1: In table 1: in Fruit Farm Land description add “except Avocado land. Also, add source at the end of the table.

Response 1: The changes were made in table 1, as requested.

Point No2: Line 183: Specify the filter data applied to the date.

Response 2: The changes were made to the text.

The images were filtered by date, the area of interest, and cloud cover percentage (< 10%). For the dates, images from the entire year were considered, with a focus on the dry season. From these images, a median composition was generated to obtain a representative image of the studied period.

Point No3: Line 207 add (RF) after Random Forest

Response 3: The changes were made, as requested.

Point No 4: Can the author add more details for validate the land use data?

Response 4: The entire training and validation process for land use classification was explained in greater detail as required.

Classification validation was performed by randomly splitting the sample into 80% for training and 20% for validation (using the random column generation function). The performance was evaluated using the confusion matrix, obtaining overall accuracy (OA) and Kappa coefficient values.

Mitigation of bias and improvement of model stability:
Stratified sampling was employed by combining various point collections (e.g., water, urban, avocado, scrub, etc.) to ensure proportional representation of all classes. The division into training (80%) and validation (20%) through a random column, along with internal validation (using the error matrix and OOB calculation), contributes to the robustness and stability of the Random Forest classifier.

Point No 5: Give data about the overall accuracy (OA) and a Kappa index for each period analized.

Response 5: Data on overall accuracy (OA) and a Kappa index for each period analyzed were provided in more detail.

LULC	Precisión	Kappa
2003	0.86	0.82
2013	0.74	0.69
2023	0.74	0.69

Point No 6: Lines 342 to 344 are information that should be included in the methodology, not in the results.

Response 6: The changes were made but it was decided to keep them in case of forgetfulness and for better clarity for readers.

Point No 7: Also for these lines: “Scenario 1 (S1): “avocado decrease”. Replacement of the avocado land above 550 m by native vegetation. - Scenario 2 (S2): “avocado expansion”. A 200-meter buffer around existing avocado patches. - Scenario 3 (S3): “Ecological restoration”. Complete replacement of avocado plantations by native vegetation.”

Response 7: The changes were made, as requested.

Point No 8: Why 550 m in S1, 200 m in S2? Can citations or references be added to justify this parameter?

Response 8: These parameters were chosen based on local expert criteria and evaluation using a digital elevation model. It was observed that 550 m was an acceptable indicator that could be applied homogeneously throughout the basin, while 200 m was selected because it could exceed the maximum elevation of hills planted with avocados.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

1. The description of data sources and preprocessing steps lacks clarity and structure. It would be beneficial to systematically present the different datasets used, including their sources, acquisition process, spatial resolution, and preprocessing steps. For example, the study mentioned that field validation was conducted in 2023, it is unclear whether similar efforts were made in 2003 and 2013. If not, please specify how ground truth data were obtained for these years and what steps were taken to ensure its reliability. Additionally, separating data-related discussions from classification methods into distinct sections would enhance readability and provide a more logical flow.

2. The manuscript mentions that the dataset was randomly split into training and testing subsets (line 202), but it does not discuss strategies to mitigate bias or improve model stability. Was stratified sampling, k-fold cross-validation, or another approach used to ensure a balanced representation of different land cover classes? Given the dataset size, relying solely on the Out-of-Bag (OOB) evaluation mechanism from Random Forest may be less stable. Please provide a discussion about it.

3. The rationale for choosing RF over other classification methods is not well elaborated. The study mentions that the land use classification is based on the nomenclature of Zhao et al. and the study by Duran-Llacer et al., with modifications (line 177). A more detailed explanation of how the classification system was adapted and what methodological improvements were introduced would help highlight the study’s contributions.

Author Response

Dear reviewer:

Date filter, validation, and accuracy metrics (OA and Kappa):

Specific date filters were applied for each period:

2023: Images taken between January 1 and December 31 were considered, and a filter was also applied to select scenes with less than 10% cloud cover. A median composite was obtained from these images to obtain a representative image of the study period.
2013: A range from April 2013 to March 2014 was used to consistently capture the dry season. A median composite was obtained from these images to obtain a representative image of the study period.
2003: Images from the entire year (January 1 to December 31) were selected. A median composite was obtained from these images to obtain a representative image of the study period.

Classification validation was performed by randomly dividing the sample into 80% for training and 20% for validation (using the random column generation function). Performance was evaluated using the confusion matrix, yielding overall accuracy (OA) and Kappa coefficient values.

Mitigating bias and improving model stability: Stratified sampling was used by combining different point collections (e.g., water, urban, avocado, shrubland, etc.) to ensure that all classes were proportionally represented. The division into training (80%) and validation (20%) using a random column, along with internal validation (via the error matrix and the calculation of out-of-bounds), contributes to the robustness and stability of the Random Forest classifier.

Data sources and preprocessing: Landsat 8 images were used for 2013 and 2023, and Landsat 5 images for 2003, all at a resolution of 30 m. Specific date filters were applied (for example, for 2023, images from January 1 to December 31 were used) and a cloud percentage filter (less than 10% or 20% depending on the collection). In addition, indices such as NDVI and MNDWI were calculated, and topographic variables (DEM, slope, latitude, and longitude) were incorporated.

Supervised classification and validation: Real field points were used for 2023, and a model was trained and applied for 2023 and 2013, as both periods had images from the same sensor (Landsat 8). Because of this, the accuracy and kappa of both classifications are the same. While for 2003 (Landsat-5), training points were generated through photointerpretation, and some ground-collected points were retained in 2023, with no changes in coverage. Thus, a specific model was trained for 2003, since the model for 2023 and 2013 could not be directly applied, as it was trained with Landsat-8 data.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

An excellent and interesting manuscript. A few suggestions: I learned a lot about avocado production of which I knew nothing. I would have liked to have seen a brief summary of the practice of avocado cultivation. Some of the practices you identified - fruit farm, other agriculture, exotic plantations, bare soil could use some clear definition. I have some comment on a few of your figures which are very good but I would like several of them to be larger. An excellent discussion section.

Comments for author File: Comments.pdf

Author Response

Dear reviewer:

Point No1: I learned a lot about avocado production of which I knew nothing. I would have liked to have seen a brief summary of the practice of avocado cultivation.

Response 1: The practices summary was included in discussion and other aspects.

Point No2: Some of the practices you identified - fruit farm, other agriculture, exotic plantations, bare soil could use some clear definition.

Response 2: Some definition are included in the text.

Point No3: I have some comments on a few of your figures which are very good but I would like several of them to be larger.

Response 3: We appreciate your comments on the figures. Several changes and improvements were made following your suggestions, as well as those provided by other reviewers. We agree that optimizing space is essential for better visualization. However, according to the current template of the Land journal, the space allocated for figures is the maximum allowed. While we understand that this may limit visual presentation, it is an editorial matter that might be improved during subsequent editing stages. The same situation occurs with tables.

Point No4: An excellent discussion section

Response 4: Thank you very much. We appreciate your comment. Other aspects were discussed that we felt would enrich the discussion.

Response 5: There are no other high-quality photographs available for inclusion in the manuscript. The study is based on LULC classes previously used in similar research, including bare soil categories as defined in the consulted literature. The buffer area may encompass various classes, modeling avocado expansion toward nearby covers. Dense vegetation includes all natural plant species, such as shrubs, riparian vegetation, and trees with high leaf density.

Bees and other pollinators are beneficial, but a proliferation of pest insects or invasive species can negatively impact the environment. The paper addresses the challenges faced by small farmers, including lack of funding, high costs, limited infrastructure, and water scarcity. In contrast, large companies dominate the region with advanced irrigation and harvesting technologies.

Avocado trees, which require substantial water throughout the year, consume less than seasonal crops but increase water demand as cultivation areas expand.

Wildfires in Chile remain a significant environmental issue, with large areas affected each year, and concerns arise about post-fire land use for agriculture.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Dear Authors,

The article is basically fine regarding scope and methodology (not new but well-established).

The changes in avocado plantations are huge but still, affect only <1% of the total land, while the changes in natural vegetation and increase in the area of bare lands are tremendous, regarding the percentage of the total land area.

My major concern is the presentation of the results.

I do not really feel that the article has a well-organized string that leads the readers through the maps and data.

I also have major concerns about the self-explanatory manner of the presented figures. Please clarify. I added some hints in some of the maps but I suggest checking all of the figures and tables, e.g.:

Figure 6 is useless in its present form (sorry to say).
Figure 9. We do not know what the numbers are telling us on the map! Furthermore, there is no need to give 2 captions, I suggest deleting the one on the map, you have the caption below the figure, no need to put a shorter version on the figure! Same in SM3 and SM4.
In case of Table 3, there is no indication what is S1, S2 and S3. Must be explained!
I suggest avoiding short, non sel-explanatory titles, such as "3.1.1 Net Change". Titles that are non-specific (can be any title of any articles' any subchapter, so to say) should be avoided.
Wherever it is obvious/visible from the figures, please add before or after the figure that, e.g. the changes in avocado land occurred from 2003 to 2013, like in the case of Figure SM1 or in Figure 4.

Good luck!

Regards, Reviewer X

Comments for author File: Comments.pdf

Author Response

Dear reviewer:

Point No1: My major concern is the presentation of the results. I do not really feel that the article has a well-organized string that leads the readers through the maps and data.

Response 1: Some improvements were made to the presentation of the results with figures and structure, as well as methodology.

Point No2: Figure 6 is useless in its present form (sorry to say).

Response 2: Figure 6 was removed from the text. It was added to supplementary material.

Point No3: Figure 9. We do not know what the numbers are telling us on the map! Furthermore, there is no need to give 2 captions, I suggest deleting the one on the map, you have the caption below the figure, no need to put a shorter version on the figure! Same in SM3 and SM4.

Response 3: The changes were made, a request.

Point No4: In case of Table 3, there is no indication what is S1, S2 and S3. Must be explained!

Response 4: The explanation appears in the text. However, a note has been added to the table.

Point No5: I suggest avoiding short, non sel-explanatory titles, such as "3.1.1 Net Change". Titles that are non-specific (can be any title of any articles' any subchapter, so to say) should be avoided.

Response 5: Thanks for the suggestions. The changes have been made.

Point No6: Wherever it is obvious/visible from the figures, please add before or after the figure that, e.g. the changes in avocado land occurred from 2003 to 2013, like in the case of Figure SM1 or in Figure 4.

Response 6: The figures were improved, and it was always emphasized that the main change occurred between 2003 and 2013.

Author Response File: Author Response.docx