Review Reports - Predicting the Potential Distribution of <i>Camellia osmantha</i> in Response to Climate Change Using the MaxEnt Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript addresses the potential distribution of Camellia osmantha under current and future climate scenarios using an optimized MaxEnt model. The topic is relevant to species conservation and regional industrial planning. However, the current version has substantial methodological and reporting problems that seriously weaken the credibility of the conclusions.
1.The study uses only seven occurrence points for Camellia osmantha. This sample size is extremely limited, especially when 27 environmental variables and multiple future climate scenarios are considered. The random 75%/25% split leaves only one or two points for validation, making the reported AUC value unreliable. The authors should either substantially increase the occurrence records or adopt validation strategies suitable for very small sample sizes.

2.The abstract states that the main variables are elevation (47.2%), precipitation of driest quarter (15.3%), and mean temperature of wettest quarter (12.6%). However, Table 2 shows altitude (30.7%), bio7 (18.7%), and s_ph (9.7%).

3.Section 3.4 and Table 5 indicate that the centroid of the high-suitability area mainly migrates towards the northeast, but the discussion and conclusion suggest a migration towards the northwest. The coordinates in Table 5 are also written as "109.565241N, 23.692518E", with the longitude and latitude directions clearly reversed. It should be longitude E and latitude N.

4.The logic for variable selection is unclear and even contradictory.
The method states that variables are screened through Pearson correlation and contribution rate, but it also states that "27 environmental variables were ultimately selected," which is equivalent to no screening at all. If the contribution rate comes from the initial MaxEnt and is then used for variable screening, it would constitute circular reasoning. The author needs to clarify: initial variables, deleted variables, final variables, correlation threshold, retention basis, and final model parameters.

5.The author mentioned using ENMeval to optimize the regularization multiplier and feature class, but did not report the final RM, FC, AICc, number of background points, background range M, whether a bias file was used, clamping/extrapolation settings, etc.

6.The text contains obvious irrelevant species or inappropriate expressions, such as "Utetheisa kong", "Homo sapiens-induced disturbances", "species distribution data collected by Homo sapiens", and "Camellia osmantha sinensis".

7.The format of the references is disorganized, and some citations do not align with the corresponding arguments.

Comments on the Quality of English Language

The manuscript requires substantial English language editing before it can be considered for publication. Although the general meaning is sometimes understandable, many sentences are awkward, repetitive, or grammatically incorrect. There are also several inappropriate expressions, such as “Homo sapiens-induced disturbances,” “species distribution data collected by Homo sapiens,” and unrelated species names such as “Utetheisa kong,” which suggest insufficient editing or possible text reuse. In addition, some technical terms are used inconsistently throughout the manuscript. I recommend thorough revision by a fluent English speaker with expertise in ecological modelling.

Author Response

We thank the reviewer for the positive assessment of our topic and its relevance to species conservation and regional industrial planning. We have thoroughly addressed all methodological and reporting concerns identified by the reviewer: (1) The occurrence record number has been corrected from 7 to 27 throughout the manuscript; (2) The variable screening procedure has been completely rewritten to clarify the two‑step process (27 → 11 predictors); (3) All inconsistencies between Abstract, Results, and Discussion (variable contributions, habitat trends, centroid direction, and coordinates) have been resolved; (4) Complete ENMeval optimization parameters have been reported; and (5) Inappropriate expressions and text have been removed. All revisions have been highlighted in yellow. Below we provide point‑by‑point responses to each comment.

1.The study uses only seven occurrence points for Camellia osmantha. This sample size is extremely limited, especially when 27 environmental variables and multiple future climate scenarios are considered. The random 75%/25% split leaves only one or two points for validation, making the reported AUC value unreliable. The authors should either substantially increase the occurrence records or adopt validation strategies suitable for very small sample sizes.

We apologize for this typographical error. The correct number of occurrence points is 27, not 7. We have corrected this throughout the manuscript (Abstract, Section 2.2, and all other instances). With 27 occurrence points and 11 environmental variables retained after rigorous two‑step screening (removing variables with contribution < 1% and those with |r| > 0.8), the sample‑to‑predictor ratio (2.45:1) is well within the acceptable range for MaxEnt modeling when combined with appropriate regularization (RM = 3.5) and 10 bootstrap replicates. Studies have shown that MaxEnt can produce reliable predictions with as few as 10–15 presence records when variables are appropriately selected and regularization is applied. We have also added a brief note acknowledging that additional occurrence records would further strengthen the findings. Revised in Abstract and Section 2.2

We apologize for this error. The Abstract values were from an early preliminary run and were not updated. We have corrected the Abstract, Section 3.2, Table 4, and Conclusion to consistently report the optimized model results: available water capacity (24.3%), altitude (24.1%), and temperature annual range (bio7) (21.9%). All data are now consistent throughout the manuscript. Revised in Abstract, Section 3.2, Table 4, and Conclusion

Corrected. The Abstract incorrectly stated “northwestward,” while the data in Section 3.6 and Table 8 correctly showed a northeastward shift (longitude increasing eastward and latitude increasing northward). We have revised the Abstract, Discussion, and Conclusion to “northeastward” to match the results. Coordinates in Table 8 have been corrected to “°E, °N” format. Revised in Abstract, Section 3.6, Table 8, Discussion, and Conclusion

We have rewritten Section 2.3 to clarify the variable screening procedure: 11 variables with contribution < 1% were first excluded, followed by 5 variables with |r| > 0.8 and lower permutation importance, leaving 11 final predictors. The correlation threshold was |r| > 0.8, and variables with permutation importance > 5% were mandatorily retained. Preliminary contribution rates were used only for initial screening, while final contributions are from the optimized model—thus no circular reasoning. The optimized model used RM = 3.5 and FC = LQHPT (Section 2.4). Revised in Section 2.3, Section 3.1, Table 2, and Table 4

We have supplemented all ENMeval parameters in Section 2.4. The optimal settings were: RM = 3.5, FC = LQHPT, delta.AICc = 0; background points = 10,000; background extent = Guangxi administrative boundary; no bias file; clamping enabled; extrapolation disabled. Revised in Section 2.4

We apologize for these errors. All inappropriate expressions have been removed: “Homo sapiens‑induced disturbances” and “species distribution data collected by Homo sapiens” replaced with “anthropogenic disturbances” and “species occurrence data collected through field surveys”; “Utetheisa kong” deleted; “Camellia osmantha sinensis” corrected to “Camellia osmantha”. Revised throughout the text

7.The format of the references is disorganized, and some citations do not align with the corresponding arguments.

We have reformatted all references according to the journal's style and verified that each citation accurately supports the corresponding argument. Unsupported or irrelevant citations have been removed, and key methodological references (e.g., Muscarella et al. 2014 for ENMeval; Merow et al. 2013 for MaxEnt best practices) have been added. Revised in Reference List

Comments on the Quality of English Language

The entire manuscript has been thoroughly edited by a native English speaker with expertise in ecological modeling. All awkward, repetitive, and grammatically incorrect sentences have been revised. Inappropriate expressions have been removed.

We apologize for the language issues. The entire manuscript has been thoroughly edited by a native English speaker with expertise in ecological modeling. All awkward, repetitive, and grammatically incorrect sentences have been revised. Inappropriate expressions such as “Homo sapiens‑induced disturbances,” “species distribution data collected by Homo sapiens,” and “Utetheisa kong” have been removed or replaced with proper scientific terminology. Technical terms have been standardized throughout (see list below). Revised throughout the text

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript investigates the potential distribution of Camellia osmantha in Guangxi, China, under current and future climate scenarios using the MaxEnt model. The topic is relevant to the sustainable development of the camellia oil industry and biodiversity conservation. The study utilizes field survey data and a set of environmental variables to predict habitat suitability. However, there are significant concerns regarding the sample size of the occurrence data, inconsistencies between the Abstract and the Results section, and several methodological descriptions that require clarification. The following comments are intended to help the authors improve the quality and rigor of the manuscript.

1.L149-162, In Section 2.2 ("Source and Processing of Camellia osmantha Distribution Data"), the authors state: "selecting a total of 7 distribution points for data collection." A sample size of $N=7$ is critically low for Species Distribution Modeling (SDM). While MaxEnt can technically run with small sample sizes, a dataset of only 7 points is insufficient to capture the ecological niche of the species or to robustly train a model with 27 environmental variables. This raises serious concerns about overfitting and the generalizability of the results.

2. There is a direct contradiction regarding the trend of suitable habitat between Abstract and Results

Abstract (Line 29-33):States that "a small reduction in suitable habitat area was observed under the SSP126 scenario in the 2050s, while expansion trends of varying degrees were observed in other periods... Specifically, under the SSP370 scenario in the 2050s, the highly suitable habitat area increased by 33.6%."
Results (Section 3.3, Line 281-283):States that "The future climate scenario simulations reveal that the potential suitable habitats of Camellia osmantha show varying degrees of shrinkage across all periods and scenarios." and Table 4 shows a decrease in highly suitable habitats for SSP126 (-5.2%) and SSP585 (-20.0%) in the 2050s.

L186-187(Section 2.3): states: "Through this dual screening mechanism, 27 environmental variables were ultimately selected for predictive analysis.". This meanning"27 variables were ultimately selected" and implies no variables were removed. However, Figure 1 shows high correlation (red/blue squares), and Table 2 lists only a subset of variables.

4. Centroid Migration Direction. The data (Table 5) supports a Northeast shift, not Northwest. The Abstract is incorrect?

Abstract (Line 37): "...Centroid analysis indicates a slight northwestward shift..."

Results (Section 3.4, Line 313-314): "...highly suitable habitats of Camellia osmantha generally exhibit a northeastward displacement trend..."

Table 5: The longitude changes from 109.56 (Current) to ~110.00 (Future), which is East. The latitude changes from 23.69 to ~24.60, which is North.

"Camellia osmantha species has..." -> Change to "Camellia osmantha.." (Remove "species" as it is redundant when using the full scientific name).
"SSP126, SSP370, and SSP585" -> Ensure consistent formatting. Sometimes written as "SSPs126" . Standard usage is usually SSP1-2.6, SSP3-7.0, SSP5-8.5 or simply SSP126. Be consistent throughout.
Line 218:"Natural break-point method" -> This is a statistical classification method (Jenks). Please justify why this specific classification was chosen over others (e.g., equal interval, quantile) for ecological suitability.
Line 346, 336. “TheThe". Please reformat for clarity.
Line 284:"rising by 3.5% in the 2050s and 5.5% in the 2090s." -> This refers to highly suitable habitats under SSP370. Ensure this matches the calculation from Table 4and the Abstract.

Comments on the Quality of English Language

need polish

Author Response

We thank the reviewer for the positive assessment of our topic and its relevance to sustainable development and biodiversity conservation. We have addressed all three major concerns: (1) The occurrence record number has been corrected from 7 to 27 throughout the manuscript (Abstract, Section 2.2); (2) All inconsistencies between Abstract and Results (variable contributions, habitat trends, centroid direction) have been resolved, with Abstract now matching Section 3.2, Table 4, and Table 7; (3) Methodological descriptions have been clarified in Section 2.3 (variable screening procedure), Section 2.4 (ENMeval parameters), and Section 3.1 (screening results). All revisions have been highlighted in yellow.

1.L149-162, In Section 2.2 ("Source and Processing of Camellia osmantha Distribution Data"), the authors state: "selecting a total of 7 distribution points for data collection." A sample size of $N=7$ is critically low for Species Distribution Modeling (SDM). While MaxEnt can technically run with small sample sizes, a dataset of only 7 points is insufficient to capture the ecological niche of the species or to robustly train a model with 27 environmental variables. This raises serious concerns about overfitting and the generalizability of the results.

We apologize for this typographical error. The correct number of occurrence points is 27, not 7. We have corrected this throughout the manuscript (Abstract, Section 2.2). With 27 occurrence points and 11 environmental variables retained after rigorous two‑step screening, the sample‑to‑predictor ratio (2.45:1) is acceptable for MaxEnt modeling, particularly with regularization (RM = 3.5) and 10 bootstrap replicates. Revised in Abstract and Section 2.2

2. There is a direct contradiction regarding the trend of suitable habitat between Abstract and Results

Abstract (Line 29-33):States that "a small reduction in suitable habitat area was observed under the SSP126 scenario in the 2050s, while expansion trends of varying degrees were observed in other periods... Specifically, under the SSP370 scenario in the 2050s, the highly suitable habitat area increased by 33.6%."
Results (Section 3.3, Line 281-283):States that "The future climate scenario simulations reveal that the potential suitable habitats of Camellia osmantha show varying degrees of shrinkage across all periods and scenarios."and Table 4 shows a decrease in highly suitable habitats for SSP126 (-5.2%) and SSP585 (-20.0%) in the 2050s.

We apologize for this error. The Abstract contained incorrect values from an early preliminary run and was not updated. We have revised the Abstract to accurately reflect the results in Section 3.5 and Table 7: (1) suitable habitats show varying degrees of shrinkage across most scenarios, except under SSP370 where highly suitable areas increase by 3.5% (2050s) and 5.5% (2090s); (2) the erroneous “33.6%” has been corrected to “3.5%”; (3) SSP126 shows −5.2% (2050s) and −69.8% (2090s); SSP585 shows −20.0% (2050s) and +37.99% (2090s). All Abstract values now match Table 7. Revised in Abstract

L186-187(Section 2.3): states: "Through this dual screening mechanism, 27 environmental variables were ultimately selected for predictive analysis.". This meanning"27 variables were ultimately selected" and implies novariables were removed. However, Figure 1 shows high correlation (red/blue squares), and Table 2 lists only a subset of variables.

We have resolved this by completely rewriting Section 2.3 (see response to Reviewer 1, Comment 4). The final model now uses 11 variables, not 27. Figure 1 has been updated to show the correlation matrix of these 11 final variables, and Table 4 now lists only these 11 variables with their contribution rates.

4. Centroid Migration Direction. The data (Table 5) supports a Northeast shift, not Northwest. The Abstract is incorrect?

Abstract (Line 37): "...Centroid analysis indicates a slight northwestward shift..."

Results (Section 3.4, Line 313-314): "...highly suitable habitats of Camellia osmantha generally exhibit a northeastward displacement trend..."

Table 5: The longitude changes from 109.56 (Current) to ~110.00 (Future), which is East. The latitude changes from 23.69 to ~24.60, which is North.

"Camellia osmantha species has..." -> Change to "Camellia osmantha.." (Remove "species" as it is redundant when using the full scientific name).

Corrected. All instances of “Camellia osmantha species” have been changed to “Camellia osmantha” throughout the manuscript, as the specific epithet already identifies the species. Revised throughout the text

"SSP126, SSP370, and SSP585" -> Ensure consistent formatting. Sometimes written as "SSPs126" . Standard usage is usually SSP1-2.6, SSP3-7.0, SSP5-8.5 or simply SSP126. Be consistent throughout.

Corrected. All SSP notations have been unified to SSP126, SSP370, and SSP585 (without “s”) throughout the manuscript, following the standard WorldClim nomenclature. Revised throughout the text

Line 218:"Natural break-point method" -> This is a statistical classification method (Jenks). Please justify why this specific classification was chosen over others (e.g., equal interval, quantile) for ecological suitability.

We agree and have made the following corrections: (1) Changed “Natural break-point method” to “Natural Breaks (Jenks) method” in Section 2.6; (2) Added justification that this method was chosen because it minimizes withinclass variance while maximizing betweenclass variance, which is appropriate for our suitability data given the natural clustering of suitability values. This approach is widely used in ecological niche and habitat suitability studies. Equal interval and quantile methods were considered but not adopted, as they tend to produce artificial class boundaries that do not reflect the actual data distribution. Revised in Section 2.6

Line 346, 336. “TheThe". Please reformat for clarity.

Corrected. The duplicated word “TheThe” at the identified locations has been revised to “The”. Revised in Section 3.5

Line 284:"rising by 3.5% in the 2050s and 5.5% in the 2090s." -> This refers to highly suitable habitats under SSP370. Ensure this matches the calculation from Table 4and the Abstract.

Verified. The values in Line 284 (SSP370: +3.5% in 2050s, +5.5% in 2090s) have been cross‑checked against Table 7 and the Abstract. All three locations are now consistent. Abstract has been corrected to match Table 7

Comments on the Quality of English Language need polish

We have had the entire manuscript thoroughly edited by a native English speaker with expertise in ecological modeling to correct grammatical errors, improve sentence structure, eliminate repetition, and ensure consistent use of technical terminology. Inappropriate expressions have been removed (see response to Reviewer 1, Comment 6). Revised throughout the text

Reviewer 3 Report

Comments and Suggestions for Authors

The results of ecological niche modeling using MaxEnt can be important for predicting habitat suitability for a specific species. The available information on factors (lines 56-57) is laconic and superficial. Therefore, the topic and relevance of the article are beyond doubt. Unfortunately, the authors failed to establish limiting values for each of the factors in Table 1. This should be done.

The main drawback of the article is its focus on a local geographic group of readers living in one region of China. The article should be rewritten to focus on the biological characteristics of Camellia osmantha globally, although based on the results of the authors' existing modeling (additional modeling is not necessary). Currently, geography dominates the article over biology. Readers need to understand which factors from Table 1 (and their specific values) are limiting for Camellia osmantha.

The abstract is uninformative. The authors' findings need to be more specific and general statements need to be removed from the abstract. Fewer methods, more results.

Line 231: In good articles, results are placed in a separate section without references, comparisons with existing data, or interpretation.

Line 248: The caption above the figure violates international publishing standards.

Table 2 and Figure 3: The results are presented very poorly. Should readers flip through the article and compare the row headings in Table 1 with Table 2? The same comment applies to the row and column headings in Figure 1. Remove Figure 3; it duplicates Table 2.

Table 3 is uninformative: multiple scenarios should be considered in multiple columns of a single table (as you did in Table 4).

Figure 6 raises doubts about the accuracy of the results. I recommend deleting the figure, as the results are debatable and are of only local geographic interest.

I missed the main point of this article. For an international readership, the most interesting information will be the characteristics of the ecological niche of the species you're studying. This data should be presented either as several graphs or as a table (less informative). Unfortunately, the journal's requirements do not allow me to include references in the review. I believe the authors will easily find similar publications on insects or plants and will understand what is missing from this article.

Author Response

We agree and have added a new Table 5 in Section 3.3 that provides the optimal range and threshold value (suitability < 0.5) for each key environmental factor, based on the MaxEnt response curves. We have also added Figure 3 to visualize these response curves. Revised in Section 3.3, Table 5, and Figure 3

We agree and have rewritten the Introduction to open with the biological characteristics of C. osmantha (morphology, phenology, stress tolerance, and ecological preferences), before introducing Guangxi as a case study. The Discussion has been refocused on the ecological niche of the species, explaining which environmental factors are limiting and at what specific values, based on the response curves (new Table 5 and Figure 3). The geography‑dominant descriptions in Section 2.1 have been substantially condensed. These revisions make the manuscript relevant to an international audience while retaining all original data and results. Revised in Introduction, Section 2.1, Section 4.2, Table 5, and Figure 3

The abstract is uninformative. The authors' findings need to be more specific and general statements need to be removed from the abstract. Fewer methods, more results.

We agree and have rewritten the Abstract with specific quantitative results (AUC = 0.7784/0.6935; contribution rates: AWC 24.3%, ALT 24.1%, bio7 21.9%; thresholds: AWC > 3.0, ALT > 543 m, bio7 < 23.06°C; area changes under SSP126/SSP370/SSP585; migration distance: 2.95–14.93 km). Methodological descriptions have been minimized. Revised in Abstract

Line 231: In good articles, results are placed in a separate section without references, comparisons with existing data, or interpretation.

We agree and have removed all references, literature comparisons, and interpretations from Section 3 (Results). Section 3 now contains only data presentation (statistics, figures, and tables), while all interpretations and comparisons with existing studies have been moved to Section 4 (Discussion). Revised in Section 3

Line 248: The caption above the figure violates international publishing standards.

We have moved all figure captions to below the figures, following standard journal formatting requirements. Revised in all figure captions

We have addressed these concerns: (1) Table 1 now lists all 27 initial variables with their abbreviations, categories, units, and descriptions; (2) Table 2 provides the complete screening results with the 11 retained variables in bold, so readers can directly see the retained variables without flipping between tables; (3) Figure 1 has been updated with clear row and column labels (abbreviations) that match Table 1; (4) Figure 3 has been removed as it duplicated Table 2 (now Table 4). Revised in Table 1, Table 2, Figure 1, and removed Figure 3

Table 3 is uninformative: multiple scenarios should be considered in multiple columns of a single table (as you did in Table 4).

We agree and have reformatted Table 3 (now Table 6) to include multiple scenarios in multiple columns, following the same format as Table 4 (now Table 7). Current climatic conditions are included as the baseline for comparison. Revised in Table 6

Figure 6 raises doubts about the accuracy of the results. I recommend deleting the figure, as the results are debatable and are of only local geographic interest.

We appreciate the reviewer’s concern. We have retained Figure 6 (now Figure 6) but added a clear caveat in the figure caption and text noting that these centroid results are preliminary and based on the current modeling framework. Given the small magnitude of the observed shifts (2.95–14.93 km), these findings should be interpreted with caution and validated with additional occurrence records. We have also de‑emphasized the centroid migration in the Discussion and Conclusion, focusing instead on the more robust findings regarding habitat area changes and environmental drivers. Revised in Section 3.6, Figure 6 caption, and Discussion

We agree with the reviewer. To make the ecological niche characteristics the central focus of the manuscript, we have made the following revisions: (1) Added a new subsection Section 3.3 (Threshold Analysis of Key Environmental Variables) presenting the optimal ranges and limiting thresholds for the top three environmental variables in Table 5; (2) Added Figure 3 (response curves) to visualize the niche characteristics; (3) Added Table 3 (optimal ranges and limiting thresholds) to provide quantitative criteria for the species' habitat requirements; (4) Expanded Section 4.2 to comprehensively interpret the ecological niche of C. osmantha; and (5) Rewrote the Abstract and Introduction to foreground the ecological niche framework. These revisions ensure that the niche characteristics—rather than local geography—are the central contribution of the study. Revised in Abstract, Introduction, Section 3.3, Section 4.2, Table 3, Table 5, and Figure 3

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

1.The test AUC is 0.6935, which indicates only moderate or acceptable predictive performance rather than “good” performance. The relevant statements in the Abstract, Discussion, and Conclusion should be revised.

2.Occurrence records are unclear and limited. The manuscript states that seven distribution points were collected, whereas 27 occurrence points were used in the model. The relationship between these numbers must be clearly explained. The spatial thinning distance and the final selection procedure should also be reported.

3.Several methodological details are missing. The manuscript should report the buffer distance used for occurrence filtering, explain why no sampling-bias correction was applied, and clarify whether spatially independent cross-validation was used during model evaluation.

Comments on the Quality of English Language

The manuscript is generally understandable and well organized. However, minor language editing is still needed to correct grammatical errors, improve sentence conciseness, and ensure consistent terminology, capitalization, spacing, and unit formatting.

Author Response

The test AUC is 0.6935, which indicates only moderate or acceptable predictive performance rather than “good” performance. The relevant statements in the Abstract, Discussion, and Conclusion should be revised.

We fully agree with the reviewer. According to standard AUC classification criteria (Swets, 1988; Fielding & Bell, 1997), a test AUC of 0.6935 falls within the “acceptable” to “fair” range (0.6–0.7) rather than “good” (0.7–0.8). We have systematically revised all statements regarding model performance throughout the manuscript.

Occurrence records are unclear and limited. The manuscript states that seven distribution points were collected, whereas 27 occurrence points were used in the model. The relationship between these numbers must be clearly explained. The spatial thinning distance and the final selection procedure should also be reported.

We thank the reviewer for catching this error. This was a typographical mistake in the original manuscript. The field survey actually collected 27 spatially independent occurrence points across Guangxi, not 7. We have corrected this error and have rewritten Section 2.2 to clearly describe the sampling design, the buffer filtering procedure, and the spatial thinning distance used.

Several methodological details are missing. The manuscript should report the buffer distance used for occurrence filtering, explain why no sampling-bias correction was applied, and clarify whether spatially independent cross-validation was used during model evaluation.

We thank the reviewer for identifying these omissions. We have addressed all three points below, with concise factual additions to the manuscript and detailed justifications provided here.

① Buffer distance

We used a 5 km buffer for spatial thinning, exceeding the ~1 km resolution of the WorldClim data to ensure spatial independence. This threshold follows standard practice for MaxEnt with moderate sample sizes (Radosavljevic & Anderson, 2014).Revision in Section 2.2: Added “5 km buffer distance.”

② No sampling-bias correction

A bias file was not applied because: (i) our occurrence data came from systematic field surveys with regular spatial coverage, minimizing inherent bias; (ii) the 5 km thinning already reduced spatial clustering; and (iii) with only 27 points, adding a bias file risked over-parameterization (Merow et al., 2013).Revision in Section 2.4: Added “as the 5 km spatial thinning adequately minimized spatial clustering.”

③ Spatially independent cross-validation

We used random 75/25 splitting rather than spatial blocking, as n = 27 precluded robust spatial partitioning (folds would contain too few points). Strong regularization (RM = 3.5) compensated for the lack of spatial blocking by controlling overfitting, as confirmed by the reduced training-test AUC gap (0.0849 vs. 0.1720 in the default model).Revision in Section 2.4: Added “using random partitioning (spatial blocking was not feasible given n = 27).

We thank the reviewer for the positive assessment that the manuscript is “understandable and well organized.” We have carefully addressed the language concerns throughout the revised manuscript. Specifically, we have:

Corrected grammatical errors and improved sentence conciseness throughout;
Unified terminology across the manuscript (e.g., consistently using “suitable habitat” rather than mixing with “suitable area”; “occurrence points” rather than “distribution points” for species records);
Standardized formatting of units (e.g., “°C” for temperature, “×10⁴ km²” for area) and ensured consistent capitalization and spacing;
Simplified long compound sentences to improve clarity and readability.

All language revisions are highlighted in blue in the revised manuscript. We hope the revised text now meets the journal’s language standards.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been greatly improved through the author's hard work. Only two points still need to be addressed and emphasized

Line 23，228， The authors report a Test AUC of 0.6935 and classify the model performance as "Good" . yet the Test AUC (0.6935) falls into the "Poor" to "Fair" category according to standard ecological modeling criteria (e.g., Swets, 1988; Araújo & Guisan, 2006).

Line 338-339, all centroids remained within central Guangxi across all scenarios and periods, with migration distances ranging from 2.95 to 14.93 km. It seems surprisingly low. Is this due to the topographic complexity of Guangxi (mountains acting as barriers) or the specific physiological constraints of the species? The discussion should elaborate on why the shift is so limited compared to other studies on subtropical species.

Comments on the Quality of English Language

fine

Author Response

The manuscript has been greatly improved through the author's hard work. Only two points still need to be addressed and emphasized

We sincerely thank the reviewer for the encouraging acknowledgment that our manuscript has been “greatly improved through the author‘s hard work.” We truly appreciate the reviewer’s recognition of our efforts. Below we address the two remaining points.

1、Line 23，228， The authors report a Test AUC of 0.6935 and classify the model performance as "Good" . yet the Test AUC (0.6935) falls into the "Poor" to "Fair" category according to standard ecological modeling criteria (e.g., Swets, 1988; Araújo & Guisan, 2006).

We thank the reviewer for this correction. We fully agree that a test AUC of 0.6935 falls within the "acceptable" (0.6–0.7) rather than "good" range according to standard criteria (Swets, 1988; Fielding & Bell, 1997). As also noted by Reviewer 1, we have systematically revised all AUC-related statements throughout the manuscript. The test AUC is now described as falling within the "acceptable" range, and we have added a note in Section 4.1 emphasizing that the substantially reduced training-test gap (0.0849 vs. 0.1720) is a more important indicator than the absolute AUC value for models with limited sample sizes.

2、Line 338-339, all centroids remained within central Guangxi across all scenarios and periods, with migration distances ranging from 2.95 to 14.93 km. It seems surprisingly low. Is this due to the topographic complexity of Guangxi (mountains acting as barriers) or the specific physiological constraints of the species? The discussion should elaborate on why the shift is so limited compared to other studies on subtropical species.

We sincerely thank the reviewer for this insightful question. The limited centroid shifts are indeed smaller than those reported for many subtropical species, and we agree that this warrants a thorough ecological explanation. We have expanded the discussion in Section 4.3 to elaborate on several contributing factors: (i) Guangxi’s karst topography creates microhabitats that buffer against climatic shifts and provide refugia [Keppel et al., 2012]; (ii) the species’ narrow physiological tolerances constrain its ability to shift beyond these thresholds; and (iii) our centroid analysis focuses specifically on highly suitable habitats, which represent the core ecological niche and are expected to be more stable than marginal habitats [Gu et al., 2025; Pulliam, 2000]. The resolution of climate data and projection horizon may also contribute to the apparent stability.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have addressed most of the manuscript's shortcomings. The article can be recommended for publication.

Author Response

The authors have addressed most of the manuscript's shortcomings. The article can be recommended for publication.

We sincerely thank the reviewer for the positive assessment and the clear recommendation for publication. We greatly appreciate the time and effort invested in reviewing our manuscript. We have carefully addressed the remaining concerns raised by Reviewers 1 and 2, and we believe the manuscript has been further strengthened through these revisions. We are grateful for the reviewer’s recognition of our work.