Review Reports

Round 1

Reviewer 1 Report

This is an interesting paper on a rare endemic species and it's definitely worth the time and effort. The author's leverage SDMs, specifically MaxENT models, to predict the past, current, and future, distribution of the endemic L. oblata. However, as I outline in details below, there are a few fundamental conceptual drawbacks that can severely limit the value of this paper's findings. As it is right now, the paper merely highlights a shift in the climatic niche of L.oblata across large time periods, and it would be a stretch to discuss anything more than just that.

1. One, the assumption that only climate variables change, while topographic, soil and land cover/ land use types stay the same over time. Especially, given the large periods of time, from past last glacial maximum (22 thousand years ago) to 2070, involved in this study. The authors need to justify such a strong assumption behind the model and discuss its consequences in the eventual results. 
2. The study area used for MaxENT modeling is not quite clear. Assuming the study area used for modeling is as shown in the maps of Figures 3 and 5, it is not surprising the authors found very high AUC values. For an endemic with such a narrow distribution range, any modeling without true absence data is likely to result in models with high discrimination capacity but of little ecological value. I strongly suggest authors find a more biologically relevant and meaningful way to define and restrict their pseudoabsence distribution so that AUC values are not overly generous. Check this recent paper, might help develop more nuanced models with restricted pseudoabsences. It can be argued that the observed shift in centroid is merely an artifact of changing climatic conditions across the study area during the large swaths of time periods.
3. Also, not clear from the methods used as to why altitude was not explicitly used as a predictor in addition to slope, especially when it's evident that L. oblate is restricted to high elevations and mountain tops. Including altitude could also help readers to better understand the "nowhere to go" hypothesis and inference.
4. The use of the term 'climate refugia' is quite misleading in the context of this study and its findings. Climate refugia are areas that remain relatively buffered from contemporary climate change over time and enable the persistence of valued physical, ecological, and socio-cultural resources. In that sense, I don't see how predicted future changes in climatic conditions that are within L. oblata's climatic niche be considered refugia? In other words, without showing how microclimatic factors buffer or minimize changes from current climate conditions of a L.oblata location, climate refugia are hard to infer. In all likelihood, the authors have found only climatically suitable conditions in the future for multiple models of climate change scenarios. Here is a recent paper that captures climate refugia as a function of microclimate factors

Author Response

Please see the attachment.

 

Author Response File: Author Response.docx

Reviewer 2 Report

The study is trying to investigate the effect of the climate change on potential distribution of an endangered shrub species in China suing an SDM model.

The manuscript is well-written and easy to read as it has logical flow. The figures are also well presented, and the structure of the manuscript is standard. 

I think that the title of the research needs some attention as it should contain something about back-casting or prediction of historical times.  

My main question / concern is about the approach that authors have adopted from the Zhao et al. (2020). I am not sure why different MaxEnt models have been developed for variables used while one model could have been done for variables? As the authors may know, MaxEnt is able to take the categorial variables as well as continuous variables? So, it is not clear why they have fragmented their modelling process into 3. I understand that some have tried different regularization values for models included categorical variables, even so the authors mention the defaults settings have been used in their model .The other issue is the way that the values are calculated for so called CHS model. How this is performed? How the intersection is done for three components? I rereferred to the original paper by Zhao et al. (2020) but I couldn’t find more explanation. In addition, assuming such process has been done in a sound way, what are the criteria for making suggested threshold in CHF model i. e. suitable, highly suitable etc….Is this arbitrary, or the rei s a rationale behind it. Such rationale can potentially fundamentally change the output or the interpretations.

Some other detailed comments/questions can be found in the “pdf” file attached.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

 

 

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

In my opinion, the way that CHS is constructed is not sound. The authors are biased regarding the role of some environmental factors and by separating them into different model there is the risk of biasing the outputs. The MaxEnt mode is capable of analysing continuous and categorial data and I find no such need to separate the variables. The have cited previous studies where the idea seems to have been promoted by a limited group of authors and I surprised why this has not been questioned in those publications.

I have gone through the response provided by the authors. Thanks for calrifying most of the question being asked. Howere, I have still concern about some rationale and reasoning explanied in the response. Here are my comments in regards to some of the authors' responses:

 

Point 2: my main question / concern is about the approach that authors have adopted from the Zhao et al. (2020). I am not sure why different MaxEnt models have been developed for variables used while one model could have been done for variables? As the authors may know, MaxEnt is able to take the categorial variables as well as continuous variables? So, it is not clear why they have fragmented their modelling process into 3. I understand that some have tried different regularization values for models included categorical variables, even so the authors mention the defaults settings have been used in their model. The other issue is the way that the values are calculated for so called CHS model. How this is performed? How the intersection is done for three components? I rereferred to the original paper by Zhao et al. (2020) but I couldn’t find more explanation. In addition, assuming such process has been done in a sound way, what are the criteria for making suggested threshold in CHS model i. e. suitable, highly suitable etc….Is this arbitrary, or the rei s a rationale behind it. Such rationale can potentially fundamentally change the output or the interpretations.

Response 2: the comprehensive habitat suitability (CHS) model was first used to predict a forest ectomycorrhizal fungus (Tricholoma matsutake) by Guo et al. (2017). T. matsutake has unique growth requirements. It does not grow in the absence of suitable vegetation and soil conditions, even if climate and topography are favourable. Thus, soil type and vegetation type variables were used as the important limiting factors during the modeling process.

 

Comment: But this works the other way around as well. If the soil type is right but climate is not favourite the species won’t grow. I still can not accept the rationale behind separating the variables. It seems that the authors had this predisposition to assume the role of soil and vegetation is more important than other variables such as climate. Even if we accept that then when building CHS model, how this has been taken into account? should more weight be given to such output? As I understand the authors mention that the results were multiplied. While it is reasonable to accept some species have specific soil or vegetation conditions, but if its that obvious, the model should pick up such relationship, if we trust the model is not able to do so why we should trust the model for other variables? Are the evidence that if we use all variable in one model (MaxEnt here), the model is not able to make reasonable predictions? If so, the authors should cite such studies that have practiced it not just assumed it.

I still didn’t get a clear reply for this question as well: The other issue is the way that the values are calculated for so called CHS model. How this is performed? I am curios to see how exactly the model is constructed.

The authors haven’t responded to the following question as well:

In addition, assuming such process has been done in a sound way, what are the criteria for making suggested threshold in CHS model i. e. suitable, highly suitable etc….Is this arbitrary, or the rei s a rationale behind it. Such rationale can potentially fundamentally change the output or the interpretations.

 

 

Point 12: Discussion 4.1 Line 386: But which group of factors are more dominant?

Response 12: we separately constructed three models, the key factors determining the distribution of L. oblata were also separately discussed. Hence, it is hard to compare which group of factors are more dominant.

Comment: But isn’t this true that by separating these you have indirectly assumed and mentioned soil type and vegetation are more important? If not then why not running the model with all variables especially that the model is capable of doing so.

Author Response

Please see the attachment.

Author Response File: Author Response.docx