Predicting the Potential Global Geographical Distribution of Two Icerya Species under Climate Change

Round 1

Reviewer 1 Report

Establishment of alien insects depends on the presence of natural enemies as well as climatic conditions. Therefore, authors should state the importance of other factors such as enemy free spaces in the Discussion section.

Author Response

Comments and Suggestions for Authors：Establishment of alien insects depends on the presence of natural enemies as well as climatic conditions. Therefore, authors should state the importance of other factors such as enemy free spaces in the Discussion section.

Answer: We added the effect of natural enemies on the invasion of two scale insects in the discussion.

Reviewer 2 Report

General comments

In my opinion, the MS currently does not meet the requirements for publication. Beside the poor quality of English, I point here important issues in the modelling methods. I will here only focus on my major concerns. For minor points, I will wait for the second revision round (because models outputs will change and then the main conclusions and the text will likely be modified).

Major comments

1. I have a major concern about the quality of the English language. In my opinion, the text does not meet the minimum required standards for publication. All the MS will obligatory need a revision by a native English speaker.
2. I fully disagree with the environmental variables selection approach. First, I advise authors to use only bioclim variables (bio1àbio19) and remove the use of monthly precipitation and temperature descriptors. The main reason is that bioclim are variables that have the same ecological meaning everywhere in the world and then are the most suited for such continental-scale SDM. For example, in January it is winter in northern hemisphere while it is summer in the southern hemisphere. Thus, monthly descriptors are not “comparable” among locations from an ecological perspective and should be avoided in continental-scale SDM.
3. L135: Authors say they remove correlated variables (correlation index <0.8). Please give details on which index you uses. Is this the Pearson’s index?
4. The environmental variables should ideally have an ecological meaning and an putative impact on the biology of the species under study. So I would advise to use proxies of maximum and minimum temperatures or precipitations (such as bio10, bio11 , etc.). Also, I advise authors to use few descriptors in order to avoid model overfitting. This is particularly true for invasion risk assessment. It is well known, and I can confirm it from my own experience, that models fitted with few climate variables (for example something between 2 and 5 descriptors) display often a better transferability in new time or in new regions than complex models with many climate descriptors. For more details, please read Elith et al. 2010 or Jimenez-Valverde et al. 2011.

Elith, J., Kearney, M., & Phillips, S. (2010). The art of modelling range‐shifting species. Methods in ecology and evolution, 1(4), 330-342.

Jiménez-Valverde, A., Peterson, A. T., Soberón, J., Overton, J. M., Aragón, P., & Lobo, J. M. (2011). Use of niche models in invasive species risk assessments. Biological invasions, 13(12), 2785-2797.

5. To select the best climate descriptors, I would advise the authors of (i) fitting different MaxEnt models using different climate datasets and using only the presence records available in the native range of these species (2) evaluate the predictive power of these models using only records available in the invaded range. This approach of evaluating model accuracy with independent data is highly recommended and avoid pitfalls associated to an artificial inflation of accuracy metrics due to spatial autocorrelation in training and evaluation datasets. This approach will allow you to select climate descriptors enhancing model transferability. Once you have found the “best” climate dataset, you can fit your full models using all available occurrences.
6. The authors do not explain where they generated background points in MaxEnt. Please be aware that the selection of the background area has an high impact on models outputs. Please, try to select a background region accessible for the species under study. For example, if a species is only distributed in Asia, it is recommended to generate background points in Asia…if the species is only distributed in America, generate background points in America, etc.
7. L196 : Authors mention the existence of presence records in greenhouses. Please, these records should be discarded from model calibration dataset since greenhouses display particular conditions different from outdoor conditions.
8. Please be very careful with LPT threshold: from my own experience, this threshold may lead to high false positive rate. I understand LPT is a very conservative approach and it is suited for invasion risk. However, the value of LPT is very low and so, models could overestimate the potential range of these species. In Appendix, you could show results using a threshold maximizing the sum of sensitivity and specificity so that the reader might compare results. Alternatively, you could also show in the main text the probability of occurrence predicted by Maxent.
9. I do not think that calculation of the extent of suitable areas in every continent is necessary. I would advise authors to present only the predictions maps and comment these maps in the discussion.
10. L368 -370 Authors say : “Highly suitable areas are more likely to be successfully invaded than moderately suitable areas and low suitable areas. The trend of diffusion is gradually from high suitable areas to moderately and low suitable areas” --> this is an obvious statement: I would remove it from the MS.

Author Response

General comments

In my opinion, the MS currently does not meet the requirements for publication. Beside the poor quality of English, I point here important issues in the modelling methods. I will here only focus on my major concerns. For minor points, I will wait for the second revision round (because models outputs will change and then the main conclusions and the text will likely be modified).

Answer: Thank you very much for your comments, which are very important for our research. We edit and proofread this article again by JJ Scientific Consultant Ltd, UK. At the same time, we rebuilt the model.

Major comments

1 I have a major concern about the quality of the English language. In my opinion, the text does not meet the minimum required standards for publication. All the MS will obligatory need a revision by a native English speaker.

Answer：This MS has been revised by JJ Scientific Consultant Ltd, UK.

2. I fully disagree with the environmental variables selection approach. First, I advise authors to use only bioclim variables (bio1àbio19) and remove the use of monthly precipitation and temperature descriptors. The main reason is that bioclim are variables that have the same ecological meaning everywhere in the world and then are the most suited for such continental-scale SDM. For example, in January it is winter in northern hemisphere while it is summer in the southern hemisphere. Thus, monthly descriptors are not “comparable” among locations from an ecological perspective and should be avoided in continental-scale SDM.

Answer：Your comments are very correct. Our previous studies have been modeling at a small scale. The last modeling did not consider the variables are not “comparable” among locations from an ecological perspective at a large scale. In this model, we only select the climate variables of bio1-bio19.

3. L135: Authors say they remove correlated variables (correlation index <0.8). Please give details on which index you uses. Is this the Pearson’s index?

Answers:

The correlation matrix shows the values of the correlation coefficients that depict the relationship between two datasets. In the case of a set of raster layers, the correlation matrix presents the cell values from one raster layer as they relate to the cell values of another layer. The correlation between two layers is a measure of dependency between the layers. It is the ratio of the covariance between the two layers divided by the product of their standard deviations. Because it is a ratio, it is a unitless number

Correlation ranges from +1 to -1. A positive correlation indicates a direct relationship between two layers, such as when the cell values of one layer increase, the cell values of another layer are also likely to increase. A negative correlation means that one variable changes inversely to the other. A correlation of zero means that two layers are independent of one another.

We provide the correlation matrix in the Supplementary Materials.

4. The environmental variables should ideally have an ecological meaning and an putative impact on the biology of the species under study. So I would advise to use proxies of maximum and minimum temperatures or precipitations (such as bio10, bio11 , etc.). Also, I advise authors to use few descriptors in order to avoid model overfitting. This is particularly true for invasion risk assessment. It is well known, and I can confirm it from my own experience, that models fitted with few climate variables (for example something between 2 and 5 descriptors) display often a better transferability in new time or in new regions than complex models with many climate descriptors. For more details, please read Elith et al. 2010 or Jimenez-Valverde et al. 2011.

Elith, J., Kearney, M., & Phillips, S. (2010). The art of modelling range‐shifting species. Methods in ecology and evolution, 1(4), 330-342.

Jiménez-Valverde, A., Peterson, A. T., Soberón, J., Overton, J. M., Aragón, P., & Lobo, J. M. (2011). Use of niche models in invasive species risk assessments. Biological invasions, 13(12), 2785-2797.

Answer: In this model, only five climate variables are selected to avoid model overfitting.

5. To select the best climate descriptors, I would advise the authors of (i) fitting different MaxEnt models using different climate datasets and using only the presence records available in the native range of these species (2) evaluate the predictive power of these models using only records available in the invaded range. This approach of evaluating model accuracy with independent data is highly recommended and avoid pitfalls associated to an artificial inflation of accuracy metrics due to spatial autocorrelation in training and evaluation datasets. This approach will allow you to select climate descriptors enhancing model transferability. Once you have found the “best” climate dataset, you can fit your full models using all available occurrences.

Answer: The distribution point data used in this article excludes the greenhouse and indoor occurrence data, and only retains the natural occurrence data. In order to avoid spatial autocorrelation, we adopted the approach of evaluating model accuracy with independent data. We screened out a quarter of the occurrence data as the test sample, and the remaining three quarters as the training sample. After selecting the best climate dataset, we selected all occurrence data to fit the complete model.

6. The authors do not explain where they generated background points in MaxEnt. Please be aware that the selection of the background area has an high impact on models outputs. Please, try to select a background region accessible for the species under study. For example, if a species is only distributed in Asia, it is recommended to generate background points in Asia…if the species is only distributed in America, generate background points in America, etc.

Answer: Because I. aegyptiaca is naturally distributed in Asia and Africa, we set MaxEnt to generate background points for I. aegyptiaca in Asia and Africa, and set the projection layers to global (excluding Antarctica). I. purchasi is distributed in six continents, so the background points generation and the projection layers were set to global (excluding Antarctica).

7. L196 : Authors mention the existence of presence records in greenhouses. Please, these records should be discarded from model calibration dataset since greenhouses display particular conditions different from outdoor conditions.

Answer：We exclude greenhouse and indoor occurrence data and only retain naturally occurring data.

8. Please be very careful with LPT threshold: from my own experience, this threshold may lead to high false positive rate. I understand LPT is a very conservative approach and it is suited for invasion risk. However, the value of LPT is very low and so, models could overestimate the potential range of these species. In Appendix, you could show results using a threshold maximizing the sum of sensitivity and specificity so that the reader might compare results. Alternatively, you could also show in the main text the probability of occurrence predicted by Maxent.

Answer：In the Supplementary Materials, we provide maps of the suitable areas of the two scale insects under “Maximum training sensitivity plus specificity Cloglog threshold”.

9. I do not think that calculation of the extent of suitable areas in every continent is necessary. I would advise authors to present only the predictions maps and comment these maps in the discussion.

Answer：Changed as your suggestion.

10. L368 -370 Authors say : “Highly suitable areas are more likely to be successfully invaded than moderately suitable areas and low suitable areas. The trend of diffusion is gradually from high suitable areas to moderately and low suitable areas” --> this is an obvious statement: I would remove it from the MS.

Answer：We re-discussed this on the time scale.