Monitoring Forest Dynamics and Conducting Restoration Assessment Using Multi-Source Earth Observation Data in Northern Andes, Colombia

Round 1

Reviewer 1 Report

Dear Authors,

Please find my comment in the attached file.

Regards

Comments for author File: Comments.pdf

Author Response

Dear reviewer,

Thanks for your time to review our proposed manuscript.

The following list describes how we integrate all your contributions. All your requested  modificatios, these were succesfullly integrated on the new version and hope that succesfully improve the manuscrip quality. We are open to any request and proposed improvement,

1. Please clarify the research problem. R: 14-16 Abstract was modified to introduce to the researcho problem.
2. The development of this methodological approach could be introduced in introduction by using appropriate references. R: Citations and justification of the methodological approach was included in the introduction focus on the application of EO data for restoration monitoring.
   
3. Please check this reference error. R: Corrected- FIgure 8
4. Discussion requires comparison of your findigs with the similar studies or methods used . In your discussion part I cant find any single reference cited. Please introduce the appropriate chapter with analysis of corresponding studies. R:We included in discussion comparision of our findings with those reported in the bibliography regarding accuracy, limitation of monitoring tropical and mountain regions.
   
5. Please make abroad conclusion if your method could be used in other countries as well. R: We integrated some recmendations of advantages and limitations to apply the proposed approach regarding mountain and tropical areas.
   

   This research emphasizes that the application of these methods can be extended to tropical regions, and certain techniques can mitigate spatiotemporal gaps to produce annual mosaics. It is anticipated that integrating Landsat-7 may result in decreased accuracy due to its infrequent revisits and SLC-off malfunction. However, the reported lower accuracy estimates do not pose a significant issue when generating yearly mosaics inclusive of dry seasons. Yet, this limitation becomes pertinent when imagery from only the rainy season is integrated, necessitating consideration of seasonality in future analyses.

   
   

Author Response File: Author Response.pdf

Reviewer 2 Report

This article describes work on deforestation and forest regeneration in Colombia. The authors used Landsat and Sentinel-2 multispectral images from 1996-2021 to monitor deforestation and Sentinel-1 radar images to monitor restoration. The subject matter of the study is relevant and the authors have collected a lot of data which have been properly processed. In general, the study is well described, but I have questions to the part of the description of forest restoration and some minor edits to the text.

1. When assessing forest restoration, you had the majority of plots in the Intermediate state, and the difference with the other states is significant.  In Supplementary Table 1 and Figure 1, you can see that the number of plots in the Intermediate state increases and decreases without causing an increase in the number of plots in the Advanced state. Also, a decrease in the Advanced plots does not cause an increase in the Completed plots. Why does this happen? Doesn't this mean that the states are not properly estimated? Add these explanations to the results description or to the Discussion section.

2. Subsection 3.2.1 states that the overall classification accuracy for the period 1996-2000 was 89 %. But in the description of the method you write that you repeated the classification procedure with the improvement of the dataset until the accuracy of 95% was achieved. Why was this not done for the first period 1996-2000?

3. Reference to several consecutive paragraphs is usually in the format [21-24] rather than [21, 22, 23, 24].

4. Figure 2 is too small, text is not readable.

5. Figure 7 is too small, and too close shades of red are chosen for different periods of deforestation - it is impossible to distinguish them on the map.

6. Figure 10 - the colors of the polygons in the image do not match what is written in the caption to this figure. This may be due to differences in monitor settings or perception, but I see yellow, red and blue-green (mint) polygons in the image, not blue and orange as indicated in the caption. Check this out.

7. Figure 10 - there should be a typo in the figure caption, as the R32-2016 plot description is more appropriate for R18-2018

Author Response

Dear reviewer,

Thanks for your time to review our proposed manuscript.

The following list describes how we integrate all your contributions. All your requested  modificatios, these were succesfullly integrated on the new version and hope that succesfully improve the manuscrip quality. We are open to any request and proposed improvement,

1. We included a sentence in Disucssion where apparently no patterns were detected when analyzing the estimated state of restoration with the year of implementation. A limitation of this study is that more detailed information regarding the type of implementation (active or passive) and the species used in restoration efforts was not available. th discussion section we included that no aparanet
2. We clarify in the methodology that iterations were undertaken to achieve accuracy levels that would prevent overestimation or underestimation of deforestation events, for which a threshold of 90% was applied for this category; iterations were performed until accuracy levels for this category reached or exceeded 90%. The period 1996-2000 presents lower accuracy levels due to reasons explained in the discussion, including the presence of clouds caused by the low revisit frequency and the quality of Landsat inputs for this period. Threhgold is associated to deforestation accuracy not overall accuracy of the map. Lines 226-228
3. Format was modified to be presented in the correct way. 21-24] rather than [21, 22, 23, 24]
4. Figure text size was increased, and corrections were made based on methods clarifications. High resolution is attach to ensure the reader can increase figure zoom to make it more readible
5. Figure wwas modified to contrast deforestationocurring in different periods. And text size incresed.
6. Polygon colors were modified to be easily associated to color distributions
7. Caption tex was corrected, including the plot descriptions

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors attempted to investigate the spatiotemporal dynamics of forests in the northern Andes using multiple earth observation data. The topic is important in the context of sustainability; however, the current structure of the manuscript is not suitable and requires modifications. Moreover, the implemented methodologies require elaboration and justifications. I provided specific comments below.

1- Line 19 and Line 23; Please correct the dates, as they should be 2021.

2- Line 24; Please include statistical values regarding the obtained accuracies of the generated maps.

3- Lines 88-89; The authors stated the existence of limitations in previous methodologies and protocols for forest monitoring; however, they did not explain what these limitations are. Based on the current state of the Introduction, the contribution of the manuscript is not justified. The authors should clearly mention relevant and recent studies, along with their limitation(s), to justify the necessity of their work. The authors should also highlight the contribution of the manuscript in the context of forest monitoring and methodology.

4- Section 2; This section requires a thorough revision. Although the explanations are written in good English, the structure is confusing. The authors are advised to introduce the sources of remote sensing data, the reference samples, and then the pure methodology separately. Currently, these three sections are combined without proper division.

5- How did you handle the spatial inconsistency between datasets, i.e., Landsat archive at 30m and Sentinel at 10m? Please provide further explanations in detail.

6- Please add the number of satellite images for each year in a relevant section related to datasets.

7-Line 187; Why were both median and last pixel values computed?

8- Line 193-195; How did you tune the hyperparameters of the RF classifier? Please state the hyperparameters of the RF. The explanations regarding the reference samples are not complete. The authors should add a new subsection and explicitly explain the reference data collection procedure. The number of reference samples both for training and validation (in the later stage) is not enough for such a large area.

9- Lines 197-198; Why was only a visual interpretation used for initial evaluations? How did you understand the accuracy is over 95%? I did not find these explanations convincing, so the authors should consider some statistical values to make the results more robust and reliable.

10- Line 207; Please explicitly elaborate on reference data preparation for all studied years. Why was no automatic framework considered for classification using migrated training samples of the baseline year to other years? The manual selection of reference samples is a limitation of the study and should be discussed in a relevant section.

11- Line 233; Please correct the "the [43] was adopted" for clarification.

12- Line 238; the number of validation samples is too low for such a large area. More validation samples are required for better investigations.

13- Line 238; I dont understand the objective of the first sentence.

14- Figure 5; Based on the first subplot (upper left), it is clear that the proposed data pre-processing workflow was not capable of generating a suitable image mosaic. There are many clouds and cloud shadows visible in the image. However, the authors mentioned cloud masking and removal in the previous stage (e.g., Line 155). Please clarify the existence of many clouds and cloud shadows. These two parameters would definitely make the results unreliable and would introduce uncertainty in the time series analysis. Further enhancement and pre-processing steps are required to generate a good-quality image for the initial year. Moreover, it is clear that the Landsat-7 data includes stripping. Why was no specific pre-processing used to de-stripe the Landsat-7 images? This data also required more enhanced preparation.

15- Line 393: It is reported that the map accuracy is 89%, while in the methodology section, the authors mentioned iterative revision using visual interpretation to ensure 95% accuracy. This contrasts the aims of the implemented method and the provided explanations.

In general, the writing style is good and it is easy to follow the explanations, though minor grammatical errors and typos exist in several parts. I recommend authors re-read the paper carefully and apply the required modifications to enhance the English quality of the manuscript.

Author Response

Dear reviewer,

Thanks for your time to review onour proposed manuscript.

The following list describes how we integrate all your contributions. All your requested  modificatios, these were succesfullly integrated on the new version and hope that succesfully improve the manuscrip quality. We are open to any request and proposed improvement.

1. Corrected
2. Corrected. Quality control accuracy was clarified. Text was incosistent.
3. The contribution of this manuscript was explained and some relevant and recent studies are mentioned
4. Section 2 was restructured based on reviewers recomendations.
5. Pan-sharpen methods were included on the manuscript section 2
   
6. We are generating a Supplmentary material with the list of each single image used on the analysis. This will be included on the final version of the manuscript
7. Median was computed from all pixels fro the same year without cloud,and integrated on PCA and random forest analysis. Last pixel was integrated as support information for quality control and training data set.
8. A sentence of RF classifier parameters for classifications was integrated to the section 2.
9. Validation procedures regarding accuracy were clarified on the manuscript. Olofsson approach were applied based on omission and comission statistical estimators.
10. The text on the manuscrit was corrected and the method clarified. Instead, automation was performed using the baseline of 2021, based on the final forest/non-forest product of 2021. Subsequent years employed a semi-automated approach integrating principal component analysis and random forest to estimate stable forest/non-forest areas (highly correlated by imagery throughout the years) and deforestation areas (low correlation by imagery throughout the years). 253-274
11. Corrected.
12. Initially proposed validation point sugested by the corresponding mapped area proportions (defined by Olofsson) was balanced to increase the validation points. Invcreasing to 100 points fora ch class. This was reviwed on the text of the manuscript
13. The sentence was modified to make it more comprehensive
14. A sentence was included to describe the missing description on how to Fill LANDSAT-7 generated by SLC-off. In Discussion section the limitations of the procedures are included.
15. We clarify in the methodology that accuracy threshold was applied just to the deforestation class to reduce the omission and comission error regrding this class. Iterations were undertaken to achieve accuracy levels that would prevent overestimation or underestimation of deforestation events, for which a threshold of 90% was applied for this category; iterations were performed until accuracy levels for this category reached or exceeded 90%. The period 1996-2000 presents lower accuracy levels due to reasons explained in the discussion, including the presence of clouds caused by the low revisit frequency and the quality of Landsat inputs for this period. Threshold is associated to deforestation accuracy not overall accuracy of the map. Lines 226-228

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The authors addressed the raised comments, the responses were rather convincing, and the manuscript has improved accordingly. The first comment of the previous round has not been applied, and the dates are not correct. Please correct the study period in lines 20 and 24. 

The manuscript requires further English enhancement. The authors should review the entire manuscript and correct grammatical issues.

Author Response

Dear reviewer,

Thanks for your time to review our proposed manuscript.

The following list describes how we integrate all your contributions for the second round review. All your requested  modifications were succesfullly integrated on the new version and hope that succesfully improve the manuscrip quality. We are open to any request and proposed improvement.

1. The first comment of the previous round has not been applied, and the dates are not correct. Please correct the study period in lines 20 and 24. R:/ The dates associated to the study period were corrected. Line 20 describes the 1996-2021 perdios that is the time frame for the complete study. Line 23-24 mentions the 1996-2000 period were the highest deforestation rate were estimated based on the multitemporal analysis.
2. A complete review of the entire manuscrit was implemented for english enhancement. All the modifications are highlighted on yellow.

Best regards

Carlos Pedraza