Identifying Suitable Restoration and Conservation Areas for Dracaena cinnabari Balf.f. in Socotra, Yemen

Round 1

Reviewer 1 Report

Tittle. The title conveys the main message of the paper — the issues addressed and the relationships among the issues.

Abstract. The abstract is concise, provides a clear overview, includes essential facts for the paper, and concludes with a final point that places the work described in a broader context.

Keywords. These are enough for the topic.

Introduction. The introduction includes background to provide an appreciation for the context of the work presented but it is not enough. It’s necessary to include a brief review of previous research about restoration and conservation areas for Dracaena cinnabari, or indicate that there isn’t any one.

Material and methods.  In this section, the authors describe the correct steps that followed during conducting their study.

Line 143 ― It is to correct the word errs by errors.

Results and discussion. This section was well written and shows all data with good descriptions. The results say about the objective that motivates the research.

Conclusion. I suggest including the major conclusions but briefly written.

Figures.

Moving Figures 1-5  to Supplementary material.

Author Response

We appreciate the comments and see how it benefits the overall flow and coherence of the document.

An additional paragraph has been added to the introduction to clarify the points raised by the reviewer.

Line 143 typo has been corrected.

Introduction has been included.

Figures have been moved to Supplementary material.

Reviewer 2 Report

Dear Authors,

I was glad to read your manuscript, which is an interesting piece of work. The manuscript seeks the most promising ways to preserve the populations of Dracaena cinnabarina on Socotra island by combining distribution and accessibility models in order to find suitable places for reforestation and protection. The paper combines spatial and statistical methods in a smart way. In my opinion the main point of the paper is the combination of accessibility models with distribution models. Due to the combination it reaches two goals in one step since, this combination makes possible to find the most easily accessible sites, which are the most suitable ones, too.

However, I support this work I think that the manuscript requires some further improvement in order to meet the publication standards.

My major concerns are the followings:

- In spite of the fact that you present your objectives and motivations in the abstract, there is nothing about these objectives in the introduction. The introduction should contain the facts what happened so far in this topic and what gaps you are intend to fill. Regarding the abstract, in my opinion, it should contain the objectives, the description of the data and methods, and the major findings. The keywords must be refined to fit better with the manuscript. For example, I would skip Google Earth Engine as it plays only the role of a data, and tool provider. I would not put the species name twice in the keywords.

The section for materials and methods is mixed up with the section of results. The description of an implemented method is followed by the results that you obtained with it. This should be separated into different chapters.

In Tables 1-2 it is not clarified what the Scale is for? It is also needs clarification why the VIF value of 10 was used as threshold in variable selection (Table 3).

Based on the inner variable structure you select only 7 variables out of the 24. How can you be sure that this seven contain those that correspond the best with the distribution of target species? If you have only 7 predictor variables what is the point in using random forest?

Soil variables require a more detailed description. What is the source of this data? How were they obtained? What is known about the spatial uncertainties of this data? It is very interesting that the pH value of the upper 5 cm plays the most important role in the distribution prediction of a tree species. It would be nice to read something about this in the discussion.

Minor concerns:

The exponents are missing from the equation in line 187.

I would put an additional map showing the best places for reforestation and the best places for conservation.

Author Response

We appreciate the comments and see how it benefit the overall flow and coherence of the document.

An additional paragraph has been added to the introduction to clarify the points raised by the reviewer.

The paper has been structured to facilitate replication of the results for different contexts. We believe it is easier to follow when intermediate steps are clear in the Materials and Methods section. The core results are presented and discussed in the Results section, and only intermediate layers are part of the Material and Methods.

As per suggestion of the other Reviewer, we have move the Figures to Supplementary material, which also contributes to the separation of Results and Materials and Methods.

VIF threshold was defined based on the work developed by Trevor A. Craney and James G. Surles on Model-Dependent Variance Inflation Factor Cutoff Values. https://www.tandfonline.com/doi/abs/10.1081/QEN-120001878

it was identified a high level of collinearity between certain variables amongst the 24 selected. A few of these variables were redundant, for example, "bio08" (Mean temperature of driest quarter) is identical to "bio10" (Mean temperature of coldest quarter). This happens due to the driest quarte coinciding with the coldest quarter. The same occurs for "bio07" (Mean temperature of wettest quarter) and "bio09" (Mean temperature of warmest quarter). Other variables were also highly correlated. The decision to select one variable over another was also based on what type of data is currently available (or can be easily collected) at landscape level by experts replicating the study in other areas.  An additional summary of the collinearity test can be found here: https://docs.google.com/document/d/1Z0xtANVqKqq35OlXORfUIsxOw5tpL_GXisi55J5EA28/edit?usp=sharing

Scale is ancillary information on how the data values are stored in the Google Earth Engine layers. It serves as a reminder for the reader to be attentive to this peculiarity.

We have used RandomForest due to its versatility and prompt availability in Google Earth Engine. It allowed the team to simulate the distribution using different combination of variables before arriving to the final 7 selected.

pH was consistently a strong predictor is all simulations. Further studies will need to be conducted to investigate it further. We believe pH might be associated with other soil nutrient of importance to the species, which can be explain strong influence.   

Soil layers were acquired from ISRIC — World Soil Information (International Soil Reference and Information Centre) and are generated from a data base with soil class and property information from more than 190.000 soil profiles around the world.

Exponents added to the equation.

Considering the energy cost component, the best place will highly depend on the local characteristics and resources available to the restoration/conservation effort. Investments in accessibility, the use of vehicles, and seasons, can all impact the optimum allocation. Therefore, we leave the best area allocation to be define by the users, in conjunction with tailored ancillary data.