Predicting Suitable Habitats of Camptotheca acuminata Considering Both Climatic and Soil Variables
Round 1
Reviewer 1 Report
I have read through the manuscript “ Predicting suitable habitats of Camptotheca acuminata considering both climatic and soil variables ".
The authors developed SDMs considering both climate and soil variables. The idea of the paper is very important and quite relevant for developing SDMs that can inform policy and management of medicinal plants such as Camptotecha acuminata. However there are some major flaws due to which I cannot recommend the paper to be published in its current form. I will recommend a major revision.
General issues
- Soil and climate are confounding variables. By building two different SDMS with two sets of variables which are essentially correlated to a certain degree you are "unconfounding" the effects of climate and soil which will lead to statistical bias.( see Lederer et al 2019)
- In this case, climate affects soil and plant occurrence, and only by having both in the model (or structural equation model(Chakraborty et al. 2019), if you prefer) you get the best possible estimation.
- The other important point the one-step model ( climate+ soil together) gets around is the correct quantification of uncertainty.
- Although climate and soil will be correlated, I would advice to live with such a correlation to a certain limit rather than having two different models and mechanically combining them later on. See (Dormann et al. 2013)
- Another option would be to combine a model with both climate and soil, and climate and soil separately and then access the model performance, transferability etc. (This is for authors to decide)
Chakraborty D, Jandl R, Kapeller S, Schueler S (2019) Disentangling the role of climate and soil on tree growth and its interaction with seed origin. Sci Total Environ 654:393–401. doi: 10.1016/j.scitotenv.2018.11.093
Dormann CF, Elith J, Bacher S, et al (2013) Collinearity: A review of methods to deal with it and a simulation study evaluating their performance. Ecography 36:027–046. doi: 10.1111/j.1600-0587.2012.07348.x
Lederer D J, Bell SC, Branson RD, Chalmers J D, Marshall R , Maslove D. M, … & Stewart P W. (2019) Control of confounding and reporting of results in causal inference studies. Guidance for authors from editors of respiratory, sleep, and critical care journals. Annals of the American Thoracic Society, 16(1), 22-28.
Author Response
Response to Reviewer 1 Comments
Point 1: The authors developed SDMs considering both climate and soil variables. The idea of the paper is very important and quite relevant for developing SDMs that can inform policy and management of medicinal plants such as Camptotecha acuminata. However there are some major flaws due to which I cannot recommend the paper to be published in its current form. I will recommend a major revision.
Response: Thanks for the positive comments. However, we do not see the major flaws as you mentioned in our approach. We have explained it below.
Point 2: Soil and climate are confounding variables. By building two different SDMS with two sets of variables which are essentially correlated to a certain degree you are "unconfounding" the effects of climate and soil which will lead to statistical bias.( see Lederer et al 2019).
Response: Lederer et al 2019 reviewed common mistakes in data analysis that lead to statistical flaws. However, we do not think that the independent consideration of climate and soil variables in our case falls into this category. We have explained it in answering the following questions.
Point 3: In this case, climate affects soil and plant occurrence, and only by having both in the model (or structural equation model(Chakraborty et al. 2019), if you prefer) you get the best possible estimation.
Response: Using only climate variables is the majority in building a SDM. A small number of studies (including Chakraborty et al. 2019, not a niche model though) incorporated soil variables in the model, which is a different approach from considering only climate variables. A major problem of this approach is that some variance explained by soil variables in the model could have been explained by climate variables if soil variables were not included, and vise versa. As there are no future projections for soil, the explanatory power of soil variables in the model is not used in projecting the future, resulting a overestimate of the model accuracy. On the other hand, if the opposite happens, the soil effect would be masked by climate variables. However, to our knowledge, this issue has not been addressed yet. Our two-step approach is one step forward of the climate-only approach without confounding between climate and soil variables. We first predict the total areas with suitable climate conditions (the climatic niche) for a species, within which suitable soil conditions may not be available in some areas. Our second step is to exclude such areas where soil conditions are not suitable within the climatic niche. We believe our two-step approach can better reflect the true nature of climate and soil variables in determining the distribution of the species than that including only climate variables or including both climate and soil variables into the same model. This is a novelty of the paper. We believe that it is an important contribution to this field.
We have revised the text in Lines 64-77 and rewritten Lines 283-298 to explain our approach better.
Point 4: The other important point the one-step model ( climate+ soil together) gets around is the correct quantification of uncertainty.
Response: We disagree with this point, as we explained above.
Point 5: Although climate and soil will be correlated, I would advice to live with such a correlation to a certain limit rather than having two different models and mechanically combining them later on. See (Dormann et al. 2013).
Response: Since we do not see any major problem with our two-step approach, we do not have to bear the problem of compromising the power of the model predictions for the future.
Point 6: Another option would be to combine a model with both climate and soil, and climate and soil separately and then access the model performance, transferability etc. (This is for authors to decide).
Response: explained above.
Chakraborty D, Jandl R, Kapeller S, Schueler S (2019) Disentangling the role of climate and soil on tree growth and its interaction with seed origin. Sci Total Environ 654:393–401. doi: 10.1016/j.scitotenv.2018.11.093.
Thanks for the reference, but it is not directly related to niche-based models.
Dormann CF, Elith J, Bacher S, et al (2013) Collinearity: A review of methods to deal with it and a simulation study evaluating their performance. Ecography 36:027–046. doi: 10.1111/j.1600-0587.2012.07348.x
Thanks. Collinearity is a concern between climate or soil variables. However, we believe that including both climate and soil variables into the same model is beyond a collinearity issue.
Lederer D J, Bell SC, Branson RD, Chalmers J D, Marshall R , Maslove D. M, … & Stewart P W. (2019) Control of confounding and reporting of results in causal inference studies. Guidance for authors from editors of respiratory, sleep, and critical care journals. Annals of the American Thoracic Society, 16(1), 22-28.
Reviewer 2 Report
General
Both climate and soil are important to habitat suitability; building models that incorporate both in the projections of future habitat suitability (and distributions) is important. This paper provides a concise description of the use of such models to map potential future habitat suitability for a medically important plant species. For the most part, the writing is clear but I do note some areas for improvement:
Introduction
The authors pose that their two-step approach to resolving climate and soils in habitat suitability modeling is novel, but this could be better demonstrated. Specifically, It would be helpful to include description of other ways in which researchers have considered both climate and soils in projections of future species distributions toward better explaining/demonstrating the novelty of their approach.
Methods
Figure 1. Is there any explanation for the single point in the top right part of the map?
Were all of the distribution data reflective of naturally occurring populations? Or were some of these data from cultivated occurrences?
Explanation/justification for the two future climate change scenarios used should be provided.
More explanation of why 6 and 12 climate and soil variables, respectively, were used in the climate niche model (rather than the 16 and 30 possibilities previously described) would clarify confusion about these numbers in the Methods.
Discussion
The authors explain that their approach is better than approaches that include both climate and soil factors in the same model, but the reason for this is not clear and could benefit from additional explanation.
More discussion of potential links between climate and edaphic factors (which are not always completely stable) should be provided.
Discussion of non-climate, non-soil factors (such as biotic interactions, seed dispersal, etc.) that also could influence distribution of the studied species should be included.
The subsection on genetic conservation would benefit from discussion of any knowledge of the genetic diversity, genetic structure, etc. of the species.
Author Response
Response to Reviewer 2 Comments
Point 1: Both climate and soil are important to habitat suitability; building models that incorporate both in the projections of future habitat suitability (and distributions) is important. This paper provides a concise description of the use of such models to map potential future habitat suitability for a medically important plant species. For the most part, the writing is clear but I do note some areas for improvement:
Response: Thanks for the positive comment.
Point 2: The authors pose that their two-step approach to resolving climate and soils in habitat suitability modeling is novel, but this could be better demonstrated. Specifically, it would be helpful to include description of other ways in which researchers have considered both climate and soils in projections of future species distributions toward better explaining/demonstrating the novelty of their approach.
Response: Thanks. We have revised the text in Lines 64-77 and rewritten Lines 283-298 to explain our approach better.
Point 3: Figure 1. Is there any explanation for the single point in the top right part of the map?
Response: Thanks for pointing this out. We have checked the original data again and found that the point in the top-right part of Figure 1 was a location for an introduced plantation on the campus of Beijing Forestry University. That data point was removed from our training dataset but forgot to delete it from the map. Figure 1 has been updated.
Point 4: Were all of the distribution data reflective of naturally occurring populations? Or were some of these data from cultivated occurrences?
Response: Yes, all of the distribution data reflect naturally occurring populations. Data points reflecting artificially introduced plantations were removed during the dataset cleaning process. This has been clarified in Line 101.
Point 5: Explanation/justification for the two future climate change scenarios used should be provided.
Response: These two scenarios are the most widely used ones and represent intermediate and business-as-usual scenarios, respectively. We have added the explanation in Lines 117-118.
Point 6: More explanation of why 6 and 12 climate and soil variables, respectively, were used in the climate niche model (rather than the 16 and 30 possibilities previously described) would clarify confusion about these numbers in the Methods.
Response: Thanks. We have rewritten relevant sections in MM (Lines 132-133, Table 1 and 2) and Results to make it clear.
Point 7: The authors explain that their approach is better than approaches that include both climate and soil factors in the same model, but the reason for this is not clear and could benefit from additional explanation.
Response: Thanks. We have revised the text in Lines 64-77 and rewritten Lines 283-298 to explain our approach better.
Point 8: More discussion of potential links between climate and edaphic factors (which are not always completely stable) should be provided.
Response: Thanks. We have addressed your suggestion in Lines 311-315.
Point 9: Discussion of non-climate, non-soil factors (such as biotic interactions, seed dispersal, etc.) that also could influence distribution of the studied species should be included.
Response: Another good point. Thanks. This point has been addressed in Lines 337-340.
Point 10: The subsection on genetic conservation would benefit from discussion of any knowledge of the genetic diversity, genetic structure, etc. of the species.
Response: Thanks for the good suggestion. We have added some relevant information and discussion in Lines 358-361.
Round 2
Reviewer 1 Report
The Authors have tried to respond to the issues I found in the manuscript, However I am not satisfied with the reasoning's they provided Especially with following
- The issue of Soil and climate are confounding variables. By building two different SDMS with two sets of variables which are essentially correlated to a certain degree you are "unconfounding" the effects of climate and soil which will lead to statistical bias.( see Lederer et al 2019)
The Authors don't provide a convincing answer as to how they deal with the biased estimates in the models by taking away confounding variables.
I agree that soil and climate can be collinear, and sometimes soil effects can mask climate effects and vice versa. If we see this also from the data quality point of view climate data at a 30 arc sec resolution is good enough but not soil. Soil changes in a matter of few meters and such rasterized soil dataset used by the authors ((http://www.iiasa.ac.at/web/home/research/researchPrograms/water/HWSD.html) don't capture the real soil dynamics.
If we consider a scenario that climate is conducive for the species to occur and soil limitations predicts that the species should not occur (ie low probability of occurrence). Here we cannot conclusively say that the soil limitations are real since the soil data is not as good as climate data (to put it simply)
Therefore a joint model would do better justice by taking into consideration this uncertainty and confounding issue.
Author Response
Response to Reviewer #1 Comments
- The issue of Soil and climate are confounding variables. By building two different SDMS with two sets of variables which are essentially correlated to a certain degree you are "unconfounding" the effects of climate and soil which will lead to statistical bias.( see Lederer et al 2019)
The Authors don't provide a convincing answer as to how they deal with the biased estimates in the models by taking away confounding variables.
Response: Thanks for the suggestion. We have added the combined model into our study to quantify the confounding effects (changes have been made in MM, Results and Discussion). We found that climate variables (91.6% contribution to the model) clearly masked the contribution from soil variables (8.4%) in the combined model. However, the contribution of each climate variable was also reduced to some extent relative to that in the climatic model. This results suggest that the combined model basically does not represent soil effect, while the climatic effect is also slightly compromised.
I agree that soil and climate can be collinear, and sometimes soil effects can mask climate effects and vice versa. If we see this also from the data quality point of view climate data at a 30 arc sec resolution is good enough but not soil. Soil changes in a matter of few meters and such rasterized soil dataset used by the authors ((http://www.iiasa.ac.at/web/home/research/researchPrograms/water/HWSD.html) don't capture the real soil dynamics.
If we consider a scenario that climate is conducive for the species to occur and soil limitations predicts that the species should not occur (ie low probability of occurrence). Here we cannot conclusively say that the soil limitations are real since the soil data is not as good as climate data (to put it simply)
Therefore a joint model would do better justice by taking into consideration this uncertainty and confounding issue.
Response: We do not agree on this. Both climatic and soil variables are far from 100% accurate. Soil variables may need finer resolution, but we can only use the best available data. Plus, this problem is not specific to our approach; it applies to all kinds of models, separately or combined.
Nice to see that you agree that “sometimes soil effects can mask climate effects and vice versa.” We believe that it always happen, just a matter of magnitude. If soil masks climate, the model would comprise its capacity for predicting the future as there is no soil data for the future. If climate masks soil, soil effects would not fully be reflected. Either way is a serious matter. These issues can be avoided by our two-step approach. Plus, the two-step approach can better reflect the true nature – climate set the basic niche, while soil put additional restraint on the top of the climate niche. We found it simple and straightforward. Thus, we are not convinced to give up on our new approach and to go back to the combining approach.
