Projected Shifts in Colombian Sweet Potato Germplasm Under Climate Change

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The introduction is overly lengthy and emphasizes only the limited number of studies on the impact of climate change on sweet potato distribution, which is insufficient to highlight the study's novelty. It is recommended to consolidate and streamline this section to improve logical coherence. Moreover, the current shortcomings of existing distribution models or conservation strategies in Colombia should be clearly stated, as well as the unique contribution and perspective of this study.

Please clarify which SSP scenario the MPI-ESM1-2-HR model corresponds to (e.g., SSP2-4.5, SSP5-8.5), both in the methods and results sections.

The manuscript mentions 5,000 pseudo-absence points but does not specify how these were generated (e.g., background points, random sampling, environmental filtering). This information is crucial as it affects model fairness and result robustness.

The manuscript lacks details regarding species occurrence data processing: how were duplicate records handled? What strategy was used for filtering points at the spatial resolution employed? Was spatial clustering bias addressed (e.g., through spatial thinning)?

The machine learning models lack interpretability mechanisms. In addition to AUC, it is recommended to assess feature importance or model uncertainty to enhance the explanatory power of the models.

For the MaxEnt model, please provide details on parameter settings (e.g., whether automatic tuning was applied), and clarify whether sample bias correction was considered.

Regarding the projected distribution changes, were key metrics such as percentage change in the distribution area, range shift, or niche overlap calculated? These indicators are essential for assessing the impact of climate change.

Lines 256–262 present Welch test results for temperature and precipitation differences, but lack ecological interpretation. Please elaborate on how these climate differences influence the adaptive strategies or potential migration capacity of wild vs. cultivated populations.

Line 46, “global average temperature is notably 3°C higher than the estimated range of 14 ± 0.5°C,” could be misleading. Please clarify the reference baseline period (e.g., 1850–1900).

The definition of “presence threshold >90%” is unclear. Please explain how this threshold was determined.

The manuscript briefly mentions “future pre-breeding” at the end; however, the scientific implications for breeding applications should be strengthened. Please specify which phenotypic traits (e.g., drought tolerance, heat resistance, altitude adaptation) should be prioritized, and whether wild genetic resources could be integrated to improve cultivated varieties.

Line 466: “[53] affirmed…” should be revised for consistency in citation style and author identification.

Citation formatting is inconsistent—some appear as “[1,2]”, others as “[17–23]”, and some use direct URLs. Please unify the referencing format throughout.

While using VIF < 10 is a common standard, it is advisable to cite a foundational reference or briefly justify the selection of this threshold.

Figure 6, the text is too small to be legible and should be enlarged for clarity.

Line 70, “Afterall” is a spelling error

Author Response

Comments 1: The introduction is overly lengthy and emphasizes only the limited number of studies on the impact of climate change on sweet potato distribution, which is insufficient to highlight the study's novelty. It is recommended to consolidate and streamline this section to improve logical coherence. Moreover, the current shortcomings of existing distribution models or conservation strategies in Colombia should be clearly stated, as well as the unique contribution and perspective of this study.

Response 1: Thank you for this suggestion. We have revised the Introduction to enhance its logical coherence and streamline redundant content. We consolidated the background on sweet potato distribution and climate impact, and included a clearer articulation of the knowledge gaps in existing distribution models and conservation strategies for Colombia. Our study’s unique contribution—namely, the integration of seven machine learning models and a high-resolution assessment of wild vs. landrace sweet potato under CMIP6 projections—is now emphasized more clearly in the final paragraph of the Introduction (lines 85–99).

Comments 2: Please clarify which SSP scenario the MPI-ESM1-2-HR model corresponds to (e.g., SSP2-4.5, SSP5-8.5), both in the methods and results sections.

Response 2: Thank you for pointing this out. We have clarified that the MPI-ESM1-2-HR model corresponds to SSP5–8.5, a high-emissions scenario consistent with continued fossil fuel development. This information was added to the Methods section in lines 193–195 of the revised manuscript.

Comments 3: The manuscript mentions 5,000 pseudo-absence points but does not specify how these were generated (e.g., background points, random sampling, environmental filtering). This information is crucial as it affects model fairness and result robustness.

Response 3: We appreciate this observation. In the revised Methods section (lines 200–206), we now clarify that for the MaxEnt model, 5,000 pseudo-absence points were randomly generated using the randomPoints function from the dismo R package, excluding known presence locations to avoid spatial overlap. For the supervised machine learning models, we generated pseudo-absence points in equal number to the presence records for each dataset, ensuring a balanced dataset suitable for classification algorithms.

Comments 4: The manuscript lacks details regarding species occurrence data processing: how were duplicate records handled? What strategy was used for filtering points at the spatial resolution employed? Was spatial clustering bias addressed (e.g., through spatial thinning)?

Response 4: Thank you for this valuable comment. While we previously mentioned in Section 2.1 that non-redundant records were used, we have now clarified the data cleaning process in greater detail (lines 178–185 in the revised version). Specifically, we removed exact duplicates with identical geographic coordinates and retained only one occurrence per 1 km² grid cell, consistent with the spatial resolution of the environmental variables. These steps were taken to reduce spatial redundancy and minimize the risk of model overfitting.

Comments 5: The machine learning models lack interpretability mechanisms. In addition to AUC, it is recommended to assess feature importance or model uncertainty to enhance the explanatory power of the models.

Response 5: Thank you for this valuable suggestion. To improve model transparency and comparability, we used AUC as a common metric across both the MaxEnt and supervised machine learning models. Each algorithm was optimized using a custom script and the caret package, applying grid search strategies for model-specific hyperparameter tuning.

We have now expanded the Results section (lines 320–333) to detail the optimal configurations for each dataset. For landraces, the best-performing Random Forest model used mtry = 7, min.node.size = 3, and 700 trees; the optimal KNN model used k = 2; and the GBM model used a shrinkage of 0.1, interaction depth of 2, and 1000 trees. For wild populations, Random Forest used mtry = 3, min.node.size = 3, and 700 trees; KNN used k = 2; and GBM used the same shrinkage and depth with 500 trees. Naïve Bayes, logistic regression, and LDA were implemented using default or non-tunable parameters.

Although some supervised models showed high AUCs in training, they exhibited greater variance between training and testing performance, suggesting overfitting risks. In contrast, MaxEnt yielded more stable behavior across datasets and was thus prioritized in downstream projections.

Comments 6: For the MaxEnt model, please provide details on parameter settings (e.g., whether automatic tuning was applied), and clarify whether sample bias correction was considered.

Response 6: Thank you for your suggestion. We have now provided detailed information on the MaxEnt model configuration in the Methods section (lines 225–231). Models were trained using the ENMevaluate function from the ENMeval R package, which wraps the maxnet algorithm and allows systematic tuning of model settings. Specifically, we evaluated combinations of feature classes (fc = c("L", "LQ", "LQH", "H")) and regularization multipliers (rm = 1:3), and used block partitioning (partitions = "block") to ensure spatially independent model evaluation. The final configuration for each plant genetic resource was selected based on model performance metrics. A total of 5,000 background points were randomly sampled using the randomPoints function from the dismo package, excluding known presences to avoid spatial overlap.

Although we did not apply a bias file, we minimized sampling bias by removing duplicate records at identical geographic coordinates and within the same 1 km² grid cell. This step helped reduce spatial redundancy and potential overfitting in the models.

Comments 7: Regarding the projected distribution changes, were key metrics such as percentage change in the distribution area, range shift, or niche overlap calculated? These indicators are essential for assessing the impact of climate change.

Comments 8: The definition of “presence threshold >90%” is unclear. Please explain how this threshold was determined.

Response 7 and 8: Thank you for highlighting this important point. We apologize for the inconsistency in terminology—throughout the revised manuscript, we have corrected all instances to reflect the accurate threshold, which was 95%, not 90%. This threshold was applied to the continuous suitability outputs of the MaxEnt and machine learning models in order to binarize them into presence–absence predictions. As now described in the Methods section (lines 249–255), this 95% probability threshold was selected as a conservative criterion to identify only those areas with the highest environmental suitability, reducing false positives and emphasizing core distribution zones.

This approach aligns with existing literature on MaxEnt modeling, which recommends high-threshold criteria (e.g., percentiles or upper quantiles) when prioritizing areas of greater predictive confidence, especially in conservation-focused studies (Merow et al., 2013; Liu et al., 2013). We have now incorporated these references into the manuscript to clarify the rationale for this choice.

Comments 9: Lines 256–262 present Welch test results for temperature and precipitation differences, but lack ecological interpretation. Please elaborate on how these climate differences influence the adaptive strategies or potential migration capacity of wild vs. cultivated populations.

Response 9: Thank you for this thoughtful observation. We have now included a paragraph immediately following the Welch test results (lines 284–294 in the revised manuscript) to provide ecological interpretation. Specifically, we emphasize that the significantly cooler and wetter conditions associated with wild accessions suggest narrower ecological tolerances and potential vulnerability to future climate change. In contrast, landraces—occurring in warmer and drier environments—may reflect broader adaptive capacity, either due to physiological plasticity or long-term human-mediated selection. These findings imply that wild populations may face greater barriers to altitudinal migration or in situ adaptation, highlighting the urgency of targeted conservation actions.

Comments 10: Line 46, “global average temperature is notably 3°C higher than the estimated range of 14 ± 0.5°C,” could be misleading. Please clarify the reference baseline period (e.g., 1850–1900).

Response 10: Thank you for your observation. We have revised the sentence to reflect the current scientifically accepted estimate of global warming relative to the pre-industrial baseline (1850–1900). Specifically, we clarified that the global average temperature has increased by approximately 1.1–1.3 °C compared to the pre-industrial average, as reported by the IPCC (2023). The previous figure of 3 °C, which refers to projected future warming under high-emissions scenarios, has been removed to avoid confusion. This correction is now reflected in line 44 of the revised version.

Comments 11: The manuscript briefly mentions “future pre-breeding” at the end; however, the scientific implications for breeding applications should be strengthened. Please specify which phenotypic traits (e.g., drought tolerance, heat resistance, altitude adaptation) should be prioritized, and whether wild genetic resources could be integrated to improve cultivated varieties.

Response 11: Thank you for your insightful suggestion. In response, we have expanded the final paragraph of the Discussion section (now located in lines 583–596) to better highlight the scientific implications for breeding applications. The revised paragraph emphasizes the prioritization of phenotypic traits such as drought tolerance, heat resistance, and altitude adaptation in the context of climate change. It also discusses how genome–environment association (GEA) analyses can help identify adaptive loci, referencing successful applications in crops like common bean and forest species such as Fagus sylvatica. Furthermore, we highlight the potential integration of wild genetic resources through marker-assisted selection and genome editing tools like CRISPR/Cas to support future pre-breeding strategies in sweet potato.

Comments 12: Line 466: “[53] affirmed…” should be revised for consistency in citation style and author identification.

Response 12: Donde

Comments 13: Citation formatting is inconsistent—some appear as “[1,2]”, others as “[17–23]”, and some use direct URLs. Please unify the referencing format throughout.

Response 13: Thank you for your observation. We confirm that the citation formatting follows a consistent style throughout the manuscript: references are separated by commas when they are non-consecutive (e.g., [1,3,7]), and grouped using a hyphen when they are consecutive (e.g., [17–23]).

Comments 14: While using VIF < 10 is a common standard, it is advisable to cite a foundational reference or briefly justify the selection of this threshold.

Response 14: Thank you for your observation. To support the use of the VIF < 10 threshold in ecological niche modeling, we now cite two foundational references: Dormann et al. (2013) and Guisan et al. (2017). These have been added to the revised Methods section.

Comments 15: Figure 6, the text is too small to be legible and should be enlarged for clarity.

Response 15: Thank you for the observation. We have enlarged the text in Figure 6 to improve legibility in the revised version of the manuscript. Additionally, we have included the high-resolution version of the figure in the attached PDF to ensure optimal quality for publication.

Comments 16: Line 70, “Afterall” is a spelling error

Response 16: Done

Reviewer 2 Report

Comments and Suggestions for Authors

Main comment

In this study, the authors evaluated the impact of climate change on wild and landrace sweet potato (Ipomoea batatas) in Colombia, using an ecological niche modeling (Maxent approach + 6 machine learning methods) based on climatic variables. The results predict a 50% range contraction for wild accessions by 2081, accompanied by an upward altitudinal shift, while landraces show a 36% reduction in suitable areas, highlighting the need for targeted conservation strategies to mitigate genetic erosion and ensure food security. These findings could be highly relevant given the great importance of sweet potato as food crop, and the recognized importance of Colombia as a key agrobiodiversity hotspot. Although the authors' effort in implementing multiple ecological niche modeling approaches is noteworthy, the manuscript suffers from severe flaws, which can be summarized in the following list of points. The manuscript, as it is, is by no means scientifically sound in terms of conceptualization, reporting of results and discussion.

(1) lack of proper contextualization and theoretical introduction of ENM.

Despite being a study primarily focused on ecological niche modeling (ENM), in the Introduction there is few mention of ENM and how this approach is increasingly used in predicting cropland suitability under future climate scenarios. At lines 68-82, the topic of "determining the effect of climate change on the distribution of species" is - very generally – introduced, but the paragraph is confused and scientifically misleading (e.g., confounding distributional shifts with niche shifts and adaptation; see next point). The use of maximum entropy approach for determining the potential future distribution of species is mentioned at lines 140-143; however, Maxent is not the only available algorithm for ENM (indeed the authors used other approaches, too) and, most importantly, the basic theoretical foundations of niche modeling are not discussed. For example, which definition of niche did the authors consider in their analysis? Which kind of data did they use (e.g., presence-only vs presence/absence)? Which kind of variables did they consider? How does this reflect in the ecological meaning of their results? On the other hand, the authors talk at length about concepts which are not strictly related to their study: for example, they spent > 15 lines speaking about ecoregions, but the authors did not stratify their ENM analysis by ecoregion, nor discussed ecoregion-specific conservation strategies. This lack of proper contextualization weakens the paper’s theoretical foundation and is linked to critical misinterpretations of the models results made by the authors (see point 2).

(2) misinterpretation of ecological niche theory and of ENMs results.

In this manuscript, the authors speak multiple time about "adaptation", in relation to changes in species distributions and ENMs; below some examples.

Lines 99-102: "recognizing the distribution of sweet potato based on its current ability to adapt...";

Lines 368-369: "Given the projected temperature increase for this period, sweet potato landraces are expected to adapt to…";

Lines 389-390: "However, a new concentration point appears in the southern Pacific region, where the area's characteristics likely support the adaptation of...";

Lines 468-471: "This behavior suggests that spatial distribution not only involves adaptation but also…";

Lines 521-522: "overall, this modelling effort has enable to predict range contraction in wild populations migrating to highlands, and adaptation of landraces...";

Lines 532-536: "In contrast, landraces exhibited greater stability and adaptability in lowland regions..."; "This divergent behavior suggests differences in adaptive capacity".

This leads me to assume that the authors do not have a strong grasp of the fundamental theoretical concepts of ecological niche theory and ENMs. In ENM, you model the niche of a species based on species occurrences and environmental variable and, as a by-product, you can also get a spatially explicit habitat suitability map. Then, supposing that the niche does not change in time (i.e., niche conservatism), you can reproject the distribution of the species according to future environmental conditions. Under this approach, you are by no means evaluating/modeling species adaptation, nor adaptation potential; on the contrary, the basic assumption is of niche conservatism (see e.g., Peterson et al. 2011. Ecological niches and geographic distributions. Princeton University Press). Niche conservatism assumes that species' ecological preferences remain unchanged over time (i.e., niches are conserved), i.e., no evolutionary change (null hypothesis of no adaptation). As a consequence, ENM (e.g., MaxEnt) predicts habitat suitability, not adaptation. Statements like those mentioned above are unsupported by ENM: they would require evidence of genetic adaptation (e.g., adaptation genomics studies, common-garden experiments) or, at least, specific tests of niche divergence (e.g., Schoner’s D – but the latter are not conclusive for adaptation and, if niche conservatism is violated, future projections are biased), not just correlative niche modeling. Same when the authors discuss their results claiming about"resilient landraces" and "adaptability": those are all misinterpretations. All these statements suggest unclear conceptual framing by the authors. Indeed, at lines 210-212 (section 2.3. Potential Niche Distribution Modeling), they wrote: "This analysis was based on the assumption of niche conservatism, suggesting that range shifts would happen through selection, extinction, and migration". Claiming niche conservatism allows for "selection" is a serious theoretical mistake, which among other things contradict the core assumption of ENM (that niches are stable over time). This isn’t just semantics – it’s foundational to interpreting ENM results. Therefore, the authors should reinterpret all their results in light of a correct understanding of the theory underlying ENMs. Alternatively, if the authors intend to test for niche evolution (e.g., via genotype-environment associations), they must explicitly model niche shifts (e.g., between wild and landrace sweet potato populations, but I guess this would be complicated in a cultivated plant).

Note: among other things, also the the authors' statement that "the Maximum Entropy algorithm... applies maximum entropy and Bayesian inference techniques to estimate probability distributions of occurrences" (line 214) is partially incorrect: although some extensions of the Maxent algorithm integrate Bayesian modeling (Bayesian Maximum Entropy), to my knowledge the classic MaxEnt model (as implemented in the function MaxNet of R package ENMeval, which the authors cite) does not use Bayesian inference.

 

(3) Limitations deriving from the exclusion of potentially relevant ecological variables.

Modeling the potential distribution of a species based of only climatic variables can sometimes be acceptable (depending on the aims of the study), but it’s a very reductive approach. The authors should clearly specify that they are modeling only the climatic niche (environmental tolerances to temperature/precipitation), and discuss their results in the light of all the limitations of this approach. However, in the case of crops such as potato or sweet potato, factors other that climate and, in particular, edaphic variables, can be crucial limiting factors for its distribution and cultivation (see e.g., Raymundo et al. 2014. Potato, sweet potato, and yam models for climate change: A review. Field Crops Research). For example, soil type was found to be one of the most important variables in a Maxent model of potato potential distribution in Pakistan (Khalil et al. 2021. Climate change and potential distribution of potato (Solanum tuberosum) crop cultivation in Pakistan using Maxent. AIMS Agriculture and Food). Therefore, the authors should incorporate some soil variables in their models, extracting data from SoilGrids, the FAO soil portal, etc., or, at least reinterpret their results clearly discussing the associated limitations.

Minor comments

Lines 51-52: please replace “and intensify global warming” with something like “representing the primary driver of ongoing global warming” (AR6 is clear about that).

Line 70: given the object of your study, please add “and cropland suitability” after “the distribution of species”.

Lines 74-75: correct “the high plasticity of sweet potato crop” with something like “whose high plasticity”.

Line 94: add “to” before “some extent”.

Lines 230-233: I really don’t understand here. Weren’t the future scenarios tested in your study four? Why you first speak about 2020-2040, and then you list the others? Moreover, haow can be data for 2020-2040 be “historical data”? Please clarify.

Lines 238-240: what do you mean here for “minor changes” and “reduced/enhanced distribution”? Which variation threshold did you used?

In section "3.5 Altitudinal Migration to Counter Climate Change", the authors reported about a shift in the future altitudinal distribution of sweet potato. However, in the Methods there is no mention on how they calculate it, nor which source (e.g., DEM) they used for the elevation data (also note: there are two different sections called 3.5; please correct).

Line 210: please correct "sceneries" into "scenarios".

Line 524: please correct "porgrams" into "programs".

Author Response

Comments 1: Lack of proper contextualization and theoretical introduction of ENM
Response 1: We thank the reviewer for this valuable observation. We fully agree on the importance of strengthening the theoretical foundation of Ecological Niche Modeling (ENM), given its central role in our study. In response, we have substantially revised the second half of the Introduction, particularly lines 66–84 in the revised version, to explicitly address:

• The conceptual basis of ENMs, including the assumption of niche conservatism (as previously mentioned in line 211 of the original manuscript, but now more appropriately introduced in the Introduction with the corresponding citation);
• The distinction between the fundamental and realized niche, and the interpretation of ENM outputs as projections of environmental suitability, rather than direct indicators of species distribution or adaptive capacity.

These additions are now supported by key references, including Peterson et al. (2011), to ensure conceptual rigor and alignment with current ecological modeling frameworks.

Furthermore, we have streamlined and repositioned content that was not directly relevant—such as the extended discussion on ecoregions—by linking it more explicitly to the conservation rationale that motivates our modeling objectives, and reducing its prominence in the revised manuscript.

 

Comments 2: Misinterpretation of ecological niche theory and of ENM results

Response 2: We greatly appreciate this clarification. Following your observation, we carefully reviewed all sections where terminology such as “adaptation”, “adaptive capacity”, or “resilience” was used inappropriately in the context of correlative ENMs. These have now been revised to reflect that:

• Our models assume niche conservatism, i.e., no evolutionary adaptation;
• We infer potential shifts in habitat suitability, not adaptive responses;

Any statements suggesting adaptation or resilience have been rephrased as spatial or climatic suitability, unless explicitly supported by prior physiological or genomic evidence.

We are grateful for this observation, which has substantially improved the scientific accuracy of our interpretation.

Comments 3: Limitations deriving from the exclusion of potentially relevant ecological variables

Response 3: We thank the reviewer for this thoughtful and important comment regarding the exclusion of potentially relevant ecological variables, particularly edaphic factors. We fully agree that soil characteristics—such as texture, fertility, and drainage—can be key determinants of the realized distribution and cultivation potential of crops like sweet potato, as clearly demonstrated in studies such as Raymundo et al. (2014) and Khalil et al. (2021).

As rightly pointed out, integrating edaphic variables from sources such as SoilGrids or the FAO Soil Portal could improve the spatial resolution of present-day suitability estimates. However, we would like to clarify that the primary objective of our study is to conduct a temporal comparative modeling of the species' climatic niche under various climate change scenarios, using projections derived from CMIP6.

In this context, including edaphic variables—while potentially useful for improving the current spatial accuracy of our models—presents notable limitations for projecting future distributions, because:

1. Global soil datasets such as SoilGrids and GSOCseq are static and do not include future projections under climate scenarios (e.g., SSP2-4.5, SSP5-8.5).
2. Assuming the constancy of edaphic conditions over long timeframes (e.g., up to 2100) could be misleading, particularly given anticipated shifts in land use, degradation, and climate–soil interactions.
3. This could introduce temporal inconsistencies or noise into the model outputs, thereby reducing the robustness of long-term projections.

To address this important point, we have added a dedicated clarification in the revised version of the manuscript (lines 525–530), where we acknowledge that edaphic variables could improve present-day model accuracy but would constrain the reliability of future projections due to their static nature. We also emphasize the potential value of developing dynamic, climate-linked soil datasets that could be integrated into future modeling frameworks. Furthermore, we explicitly cited the valuable references suggested—Raymundo et al. (2014) and Khalil et al. (2021)—to strengthen the scientific basis of this discussion.

 

Minor Comments

Comments 4: Lines 51-52: please replace “and intensify global warming” with something like “representing the primary driver of ongoing global warming” (AR6 is clear about that).

Response 4: Lines 51–52: Reworded as suggested: “...representing the primary driver of ongoing global warming.”

Comments 5: Line 70: given the object of your study, please add “and cropland suitability” after “the distribution of species”.

Response 5: Line 70: Added “and cropland suitability”.

Comments 6: Lines 74-75: correct “the high plasticity of sweet potato crop” with something like “whose high plasticity”.

Response 6: Line 74–75: Rephrased

Comments 7: Line 94: add “to” before “some extent”.

Response 7: Line 94: Corrected

Comments 8: Lines 230-233: I really don’t understand here. Weren’t the future scenarios tested in your study four? Why you first speak about 2020-2040, and then you list the others? Moreover, haow can be data for 2020-2040 be “historical data”? Please clarify.

Response 8: Lines 230-233: Thank you for pointing this out. We realized that throughout the text we mistakenly referred to the first future projection period as 2020–2040, when it should be 2021–2040. We used nearly 30 years of historical data from WorldClim to calibrate the models, and then projected them using CMIP6-based future scenarios.

Thank you for this observation. We apologize for the inconsistency in terminology—throughout the revised manuscript, we have corrected all instances to reflect the accurate threshold, which was 95%, not 90%. We have now clarified the methodology used to estimate altitudinal shifts in section 2.3 Specifically, we added the following explanation in lines 240–246 of the revised Methods: “A 95% presence probability threshold was applied to binarize presence–absence predictions. This conservative threshold was selected to delineate areas of high environmental suitability, thereby minimizing false positives and identifying core habitat zones, as recommended in previous studies using MaxEnt for conservation-oriented projections [43,44]. For cells identified as presence, elevation values were extracted using the elevation layer from WorldClim 2.1, which is derived from SRTM (Shuttle Radar Topography Mission) elevation data, to calculate the altitudinal distribution of predicted occurrences. Maps for each climate change scenario were compared per plant genetic resource to assess their climate sensitivity.” We also corrected the section numbering to avoid duplication of "3.5".

Comments 9: Line 210: please correct "sceneries" into "scenarios".

Response 9: Line 210: Done

Comments 10: Line 524: please correct "porgrams" into "programs".

Response 10: Line 524: Done

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have carefully addressed all of my comments, implementing the suggested changes to the manuscript or providing detailed and convincing responses.
The Introduction has been properly rewritten to offer a more accurate contextualization and theoretical background on ENM, with less relevant information removed.
The authors acknowledged their previous misinterpretation of the ENM results in terms of adaptation and have carefully revised the relevant sections.
The exclusion of edaphic variables from the ENM models was convincingly justified by the authors, who now also recognize the limitations associated with this choice.
Inaccurate or unclear points have been resolved.
Overall, I strongly believe that the manuscript has been significantly improved.

Author Response

We sincerely thank the reviewer for their thorough evaluation and positive feedback. Your constructive comments have greatly contributed to strengthening the manuscript.