Detecting Woody Plants in Southern Arizona Using Data from the National Ecological Observatory Network (NEON)

Round 1

Reviewer 1 Report

The topic is very interesting. Undoubtedly, this study provide a priori knowledge and scientific support for the development of  woody plant encroachment (WPE) mapping method. However, the content and structure of the article was confusing, especially the unreasonable introduction of introduction and result. On the one hand,  introduction is very long (It should be straightforward). The section of result are too technical, one the other hand. Also, i found that the only 1-2 references closely related to WPE mapping, i.e., 21 and 22.

Author Response

Point 1: The content and structure of the article was confusing, especially the unreasonable introduction of introduction and result. On the one hand, introduction is very long (It should be straightforward). The section of result are too technical, one the other hand

Response 1: 

We removed significant portions of the introduction to reduce complexity. 

We reordered and shortened the first paragraph of results and removed content in the in variable selection section to reduce complexity.

Point 2: I found that the only 1-2 references closely related to WPE mapping, i.e., 21 and 22.

Response 2:

Other references cited that are closely related to WPE and FWC mapping include [2], [16], [23], [24], [28], [32], [33], and [34].

 

 

 

Reviewer 2 Report

 

The paper presents an object-based method to estimate the fraction of woody plant in drylands using remote sensing imagery (LiDAR) and in situ data. Different data science / machine learning algorithm are compared.

 

The paper is very well written  and very well structured with an impressive and  informative introduction retracing all is needed to know to appreciate the approach undertaken. The methodological description is also very precise, facilitating potential comparisons with the literature.

 

The results are described within a vivid series of appropriate figures and tables. The only minor concern is that,  I feel that some paragraphs in the discussion could be better placed or summarised in the conclusion?

 

 

 

Author Response

Point 1: I feel that some paragraphs in the discussion could be better placed or summarised in the conclusion.

Response 1: 

We moved a few paragraphs from discussion to conclusion and added some content to discussion to address assumptions and future recommendations.

Thank you for the time and effort spent reviewing our manuscript, your feedback has been very helpful for improving our article for publication.

Reviewer 3 Report

Very nice paper!

Some minor comments for improvement:

Line 439: although the link to the supplementary materials is provided at the end of the paper it would be good to indicate that.

Line 575: please repeat here the Research Questions from p. 4 to ease reading.

As WPE is a dynamic process from my understanding monitoring its development is more important than a simple estimation of the current status. Of course it is important to establish a baseline (as good as possible) from which monitoring of the evolution needs to start. And the paper provides exactly a sound methodology for that. However, the paper would improve if you would address the multi-temporal monitoring aspect  in the discussion.

 

Author Response

Point 1: Line 439: although the link to the supplementary materials is provided at the end of the paper it would be good to indicate that.

Response 1: 

Indication of Supplementary Materials added at Lines 340-341.

Point 2: Line 575: please repeat here the Research Questions from p. 4 to ease reading.

Response 2: 

Research questions have been restated at their respective locations within the Discussion.

Point 3: The paper would improve if you would address the multi-temporal monitoring aspect  in the discussion.

Response 3:

Added content to discussion to address multi-temporal aspect of these methods/data.

Thank you for the time and effort spent reviewing our manuscript, your feedback has been very helpful for improving our article for publication.

Reviewer 4 Report

1. For estimating FWC, it is more important to accurately measure woody features, i.e., metrics or variables than selecting machine learning algorithms and variables. The methods for measuring woody features in this manuscript is too simple and lack evaluation.

2. There is too much background knowledge in the Introduction. Current methods for measuring woody features should be added.

3. The training data is too few to support the effectiveness of feature and machine learning algorithm selection.  In fact, when there are enough data with enough features, there is little difference between different machine learning algorithms. 

Author Response

Point 1: It is more important to accurately measure woody features, i.e., metrics or variables than selecting machine learning algorithms and variables. The methods for measuring woody features in this manuscript is too simple and lack evaluation.

Response 1:

As this is a relatively new methodology with novel data, it contains elements of uncertainty by nature. We understood this as we conducted our research and tested as many fundamental vegetative indicators (3 vegetation structure and 4 vegetation index variables) as feasible with the given size, scope, and budget of our work. Data quality is completely subject to the data source with our methods, but we deemed the in-situ vegetation measurements and corresponding remote sensing data collected by NEON to be adequate for evaluating our methodology and testing to see how well solely NEON data performs with this vegetation classification task. We agree that there are many other vegetation variables that need to be evaluated to improve our understanding of our methods and we encourage future studies to evaluate those variables in our discussion.

Point 2: There is too much background knowledge in the Introduction. Current methods for measuring woody features should be added.

Response 2:

We removed significant portions of the introduction to add simplicity. Current methods for measuring woody features via remote sensing are briefly described in the introduction, but not fully detailed for the sake of brevity. 

Point 3: The training data is too few to support the effectiveness of feature and machine learning algorithm selection.  

Response 3:

We agree that the training data size is too small to fully support the predictive power of the machine learning algorithms tested. However, one of the goals of this study was to lay groundwork methodology with currently available data that can be continuously improved as the next two decades of NEON data is collected. We added content to discussion to address this point.

Round 2

Reviewer 1 Report

• It has been revised according to relevant suggestions. Recommended for publication.

•  

Author Response

Thank you for the time and effort spent reviewing our manuscript, your feedback has been very helpful for improving our article for publication.

Reviewer 4 Report

The response is acceptable

Author Response

Thank you for the time and effort spent reviewing our manuscript, your feedback has been very helpful for improving our article for publication.