Predicting Habitat Suitability and Conserving Juniperus spp. Habitat Using SVM and Maximum Entropy Machine Learning Techniques

Round 1

Reviewer 1 Report

Predicting habitat suitability and conserving Juniperus sp. habitat using SVM and Maximum Entropy machine learning techniques

Abdolrahman Rahimian Boogar, Hassan Salehi, Hamid Reza Pourghasemi and Thomas Blaschke

The article presents a study that compares two habitat suitability models for predicting the distribution of Juniperus habitat in the Sepidan Area in the Fars province of Iran.  The article needs to better establish the aim and objectives of the paper. The use of two models is not justified in the introduction within the context of the study. For example, is the paper a critical analysis of the two modelling approaches or is it suggesting that by applying both models the predicted maps provide a more rigorous predicted dataset to inform conservation planning? The context for the study is also problematic as it indicates drought and climate change have driven a change in Juniperus distribution, but potential changes (either observed or predicted) are not factored into the rest of the article. The reader is also left confused on whether it is a conservation study or a study that underpins expansion of cultivation.

The methods used in the paper are on the whole sound, with the process of model development and validation underpinned by sound methodology and datasets. The analysis and subsequent presentation of the results, however, do not adequately interrogate the key differences in the SVM and MaxEnt model predictions.  Independent validation indicates both methods are produce valid results, but there is limited critical analysis of the differences between the two predictions. Alternatively, integration of the results to demonstrate sensitivity or uncertainty in the resultant map could be useful to support decision making.The importance of effective factors provides some interesting insights into the key variables driving the models, but this is presented as a single analysis and not for independently for each model? Likewise, the discussion does not explore the significance of the two differing but independently validated model results or highlight some key implications such as the control of temperature on Juniperus distribution and the likely impact of climate change.

The figures used in the article are clear, but the article needs a through edit and review of the language used in the article.

The foundation and modelling are undertaken in this paper are sound, and with a major revision to better present and interpret this work I believe it will be suitable for publication.

Some specific comments are below.

Abstract

L16 – delete or reword:  ‘and nearly to its development in this area’

Introduction

The introduction needs a thorough edit to address the long and complicated sentence structure and clarify the language used.

The introduction needs to provide more context and justification for why the SVM and MaxENT methods should both be used – what are their relative strengths and weaknesses and why is this important to test? The current wording suggests both methods are valid, and does not explain the value of testing both approaches in the context of this study. This should also be captured in the framing of the aims and objectives of the study.

Methods

The methods are reasonabliy clearly written, but still warrant some editing. Some description of the ecology of Juniperus would be worthwhile to add to provide a context for the habitat suitability analyses.

L75 Change 72.44 to 73 degrees

Figure 1d unnecessary?

L88 – 700ha

L113 –  why was the IDW method used for soil interpolation and was the accuracy of this validated?

L115 – reword sentence

L121 – break sentence at ‘…maxent/).’ And reword second half of the sentence.

152-157 – reword

Results

L172 – Table 2 Elevation close to threshold-given relationship to derived parameters should be excluded.

L178-181 – Reword sentence and remove decimal places from percentage values.

L182 – As well as total area, the predicted pattern should be compared. Undertake a cross tabulation of the two predicted maps to identify the specifically which classes contain the most confusion.

L186 – was the model run removing with the 3 most significant factors only given they appear to account for most variation?  

Discussion

See overall comments.

Author Response

Hi

Please see in attached file.

Best

Author Response File: Author Response.docx

Reviewer 2 Report

Overall, this appears to be a very good and focused study, with useful information that can be applied on the agriculture side.  Below are my specific comments:

Abstract:
line 16: what does "nearly to its development in this area." mean? Have drought/climate issues impacted the growth of Juniperus in the region? Please clarify.

Intro:
lines 36-39: second sentence of intro is poorly worded, must reword.
line 41: 'habitat suitability assessment' is not a 'tool' but rather an 'approach' or 'method'
line 44: add 'an' in front of 'ecological'
lines 65-66: delete 'a noteworthy tool'

Materials/Methods:
lines 69-70: delete second half of sentence, no need to include coordinates since they are in the map.
line 75: spell out digital elevation model (DEM) at first mention
Figure 1: it would be better in (a) if you give the surrounding countries a light gray (10%) shade so that it is clear what the boundaries of Iran are. Change the fonts to italics for the two bodies of water. Delete 'Value' in (c) and delete underscores in 'Validation_Samples' and 'Training_Samples'. Adjust your scale bars to more logical intervals (e.g., 2000 instead of 1850) and I would suggest writing out 'Meters" in both (c) and (d). Coordinates are not needed on all 4 sides of (c) and (d) - I would suggest eliminating the top and left side coordinates. The figure is fuzzy, but that could be related to the peer review upload process - make sure it's at least 300dpi for final submission. Change (c) description in the caption to read "Juniperus spp. locations with elevation in the background". After looking at this, there's really no need for figure 1d - I recommend eliminating and enlarging (c) to fit the area.
line 88: 700 ha?
lines 88-89: put "app version 17.6" in paretheses ()
line 91: to my knowledge, Hawth's Tools does not work in any version of ArcGIS that is 10 or higher; it changed to Geospatial Modeling Environment (http://www.spatialecology.com/gme/index.htm). Did you use GME? Or did you just use the Geostatistical Analyst to perform the subset split?
line 107: delete extra parentheses after F
lines 108-110: what was the resolution of the climate data? Or was this state data that had to be interpolated? If so, what interpolation method did you use and did you match the DEM resolution with the output?
line 115: it is unclear what you are referring to as being excluded - please clarify
line 132: add "and" after "commentary"
lines 136-138: reword - "...system, which is widely used in the field of remote sensing, and has the ability to handle small samples sizes."
line 146: do you mean "mathematical approach"?
lines 152-154: "The ROC..." sentence is a bit confusing and unnecessary - I recommend deleting.
lines 155-157: sentence is unclear and needs to be reworded
Figure 2: no need to have the coordinates for any of these maps, you've already established the location with Figure 1. These just make figure 2 way too busy and a bit difficult to read. Delete "Value" in all legends. More than 2 decimals is unnecessary (a, c, d, f, g, h, i, j, k, l, m, and n). Since the scale and orientation are the same across all 6 maps, only 1 north arrow and 1 scale bar are needed. Adjust the scale to 2000 meters. Since these maps span 3 pages, I don't think you'll be able to publish these as a single figure (although you need to consult the publisher). I recommend developing 3 figures with 5 maps each.Table 1. Recommend adding Resolution as a category in the table.

Results:

lines 175-176: since natural breaks classification is only based on binning the raw values (and not directly related to model output), I would recommend examining other break value methods that are based on threshold values output by the models themselves. For example, this article (https://www.mdpi.com/2071-1050/10/4/1081/htm) uses the minimum training presence logistic and the maximum test sensitivity plus specificity thresholds that were output by MaxEnt. This is a more model-based method of interpreting the results, whereas natural breaks is a bit more arbitrary.
lines 187-188: do you mean the jackknife test?

Discussion:
lines 221-223: starting with "but..." - this part of the sentence is confusing and needs to be reworded
line 225: do you mean 'statistically'?
line 228: change "displayed correct estimation" to "were better than random"
line 239: delete "determined as"
line 241: delete "climate"

Additional info needed for Results/Discussion sections: It may be more difficult for SVM results, but MaxEnt produces response curves. You've indicated which variables seem to influence the models the most (e.g., max/min temp and rainfall), but what range of those variables represents the optimal suitability? Does the species prefer lower temperatures and higher rainfall? This is not clear and would add useful info to the paper.

Author Response

Hi

Please see in attached file.

Best

Author Response File: Author Response.docx

Reviewer 3 Report

This study tested two machine learning models to assess the habitat suitability of Juniperus sp., suggesting that the abilities of SVM and MaxEnt are similar for this purpose, based on species occurrence data and ecological factors. Results indicated the importance of temperature and rainfall, which is relevant also for future managing and conservation actions.

In my view, the manuscript is clear and well written, and is of potential interest to scientists dealing with botany, conservation ecology and landscape management.

Author Response

Hi

Please see in attached file.

Best

Author Response File: Author Response.docx

Reviewer 4 Report

The article is essentially correct. Unfortunately, it does not carry much scientific value.

In the introduction, the authors devoted too little space to the description of the content of subsequent chapters of the article. This should be corrected.

In figure 1, the section marked (d) is not needed. The research area is shown in part (c).

(91) The authors should describe in more detail the operation of algorithms in Hawth's Tools in ArcGIS 10.6.1.

The article would have a much greater scientific value if it showed a true prediction of Juniperus sp. based on prediction of changing climatic factors. The studies could also be supplemented with the use of past data if possible.

The biggest objection to the article is the dubious connection with the subject of the jurnal.

Author Response

Hi

Please see in attached file.

Best

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Congratulations on your revised paper-this has addressed the issues noted in the first version and is ready for publication.

Author Response

Dear Reviewer

Thank you so much for your positive comment.

Reviewer 4 Report

Figure 1 is for improvement. Please make it completely new in ArcMap (the authors wrote that they have access to it), use spatial data from the same source for three data frames and the same projection.

Author Response

Dear Reviewer

Thank you so much for your positive comment.

Please see revised version of Fig. 1

Best

Author Response File: Author Response.docx