Review Reports - First Characterization of Megafire Refugia in a South American Subtropical Mountain Forest

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper investigates the mechanisms influencing fire refugia formation in typical mountainous regions of Argentina following megafires. The research addresses a novel theme and explores several interesting scientific questions. I have identified minor issues that should be addressed before the manuscript can be considered for publication.

Keywords: The terms "ecosystem recovery," "forest regeneration," "regrowth," and "sprouting" are semantically overlapping and all describe forest restoration processes. I recommend removing one or two of these keywords and incorporating terminology specifically related to "fire refugia."

Lines 210-228: The figure layout format requires correction.

Lines 328-351: The figure layout format requires correction.

Conclusions: This section currently reads more as a discussion of topics rather than a synthesis of findings. Please revise to include a concise summary of the specific research results and main conclusions derived from this study.

Author Response

Reviewer 1

R. This paper investigates the mechanisms influencing fire refugia formation in typical mountainous regions of Argentina following megafires. The research addresses a novel theme and explores several interesting scientific questions. I have identified minor issues that should be addressed before the manuscript can be considered for publication.

R. Keywords: The terms "ecosystem recovery," "forest regeneration," "regrowth," and "sprouting" are semantically overlapping and all describe forest restoration processes. I recommend removing one or two of these keywords and incorporating terminology specifically related to "fire refugia."

Authors: Thank you, the key words were changed.

R. Lines 210-228: The figure layout format requires correction.

R. Lines 328-351: The figure layout format requires correction.

Authors: The format was corrected.

R. Conclusions: This section currently reads more as a discussion of topics rather than a synthesis of findings. Please revise to include a concise summary of the specific research results and main conclusions derived from this study.

Authors: We incorporated a synthesis of finding at the beginning of the conclusions. We left, but reduced to a shorter version, the text with management implications because we considered it pertinent in the context of the manuscript being part of a Special Issue on Forest Fire Detection, Prevention and Management.

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,
Your study makes a significant contribution to understanding fire refugia in subtropical mountain forests and provides practical insights for post-fire restoration strategies. The unexpected finding of high-quality refugia in upper topographic positions under megafire conditions presents an important advancement in the field. Your robust methodology using 208 field plots and the clear management implications offer meaningful guidance for both researchers and practitioners. I believe this research will serve as an important reference for future studies in fire ecology and forest management.

Author Response

Reviewer 2

R. Dear Authors,
Your study makes a significant contribution to understanding fire refugia in subtropical mountain forests and provides practical insights for post-fire restoration strategies. The unexpected finding of high-quality refugia in upper topographic positions under megafire conditions presents an important advancement in the field. Your robust methodology using 208 field plots and the clear management implications offer meaningful guidance for both researchers and practitioners. I believe this research will serve as an important reference for future studies in fire ecology and forest management.

Authors: Thank you very much, we hope so!

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

This article is written on a very relevant topic and devoted to megafire refugia characterization for South American subtropical mountain forests. To improve the paper before promotion, I suggest some notes (accordingly the sections).

Section “Abstract” is quite informative, since it contains the explanation of research of post-fire vegetation responses two years after the megafires using five criteria. However, the main research methods, objective and main findings of the research also should be given (presumably after line 31). Also the presentation of the text should be in the third person, not like “we ised…” (line 26, etc.).

Section 1. Introduction: in this section, research hypothesis should be given (presumably after line 90). But before it, it is necessary to outline clearly the gap in megafire refugia research that your study aims to fill.

Section 2. Materials and Methods:

Fire cases should be clearly described, with their characteristics (burning period, ignition sources, damaged area, flora and fauna species, etc.) indicated, possibly in tabular form.

Furthermore, Equation (1) – line 164 – was introduced without validation. Therefore, it is unclear how the "Topographic position" parameter is represented by the Authors as a linear dimension, although the map assumes dimetric dimensions (X and Y axes).

The research methods presented in Table 1 are based on a simple analysis of primary data and are not supported by correlation analysis. Therefore, the inclusion of random values in the sample cannot be ruled out.

The figure on page 7 is superimposed on the text.

The Authors' logic in using the number of criteria and the quality of fire refugia (Figure on page 7) is unclear; this issue requires further substantiation.

The Akaike Information Criterion mentioned in line 248 requires justification in terms of the appropriateness of the Authors' choice.

I suggest adding tables from attached Supplementary file to the Article, after proper analysis.

Section 3. Results: The Authors provided good quantitative analyses, since the figure on page 10 is rather informative. However, the Authors omitted a qualitative analysis of the fire refuge's condition.

Section 4. Discussion: This section appears as a detailed description of the research results (Section 3). The discussion section should contain the main breakthrough results, a description of the obstacles and limitations of the research, and future ways to overcome them.

Section 5. Conclusions: No notes to this section.

Common note: the manuscript must be formatted accordingly MDPI template, including References.

Good luck!

Author Response

Reviewer 3

R. Dear Authors,

R. Section “Abstract” is quite informative, since it contains the explanation of research of post-fire vegetation responses two years after the megafires using five criteria. However, the main research methods, objective and main findings of the research also should be given (presumably after line 31). Also the presentation of the text should be in the third person, not like “we used…” (line 26, etc.).

Authors: Thank you, we included the objective, methods and main findings. We have retained the first-person style in the abstract and main text. This choice aligns with the journal’s conventions, and we believe that first-person writing enhances the flow and reader engagement. Additionally, it highlights the researchers’ role in design decisions and interpretations, promoting scientific transparency. However, we ensured that the tone remains formal and the focus remains on the research rather than the authors.

R. Section 1. Introduction: in this section, research hypothesis should be given (presumably after line 90). But before it, it is necessary to outline clearly the gap in megafire refugia research that your study aims to fill.

Authors: You will find these in the last two paragraphs of the introduction.

R. Section 2. Materials and Methods:

R. Fire cases should be clearly described, with their characteristics (burning period, ignition sources, damaged area, flora and fauna species, etc.) indicated, possibly in tabular form.

Authors: Thanks for the suggestion, fire characteristics were complemented within the section: “2.2 Study fires.”

R. Furthermore, Equation (1) – line 164 – was introduced without validation. Therefore, it is unclear how the "Topographic position" parameter is represented by the Authors as a linear dimension, although the map assumes dimetric dimensions (X and Y axes).

Authors: The equation used to calculate topographic position represents a normalized measure of each plot’s relative position within the local elevation range, defined by the highest and lowest elevations within a 315 m radius. This index was based on Cingolani et al 2008 (Predicting cover types in a mountain range with long evolutionary grazing history: a GIS approach. Journal of Biogeography, 35:3, 538-55) and was designed to express where a site lies along the local topographic gradient, from valley bottoms (values close to 0) to ridge tops (values close to 1). This metric does not describe a horizontal (X–Y) spatial dimension, but rather a relative altitudinal position within the local relief. This index was explained in the main text of the manuscript.

R. The research methods presented in Table 1 are based on a simple analysis of primary data and are not supported by correlation analysis. Therefore, the inclusion of random values in the sample cannot be ruled out.

Authors: Table 1 was intended to provide a descriptive summary of the variables used in the analyses rather than to present their statistical relationships. You can now find the correlations among variables in supplementary materials.

R. The figure on page 7 is superimposed on the text.

Authors: Thanks, the format was corrected.

R. The Authors' logic in using the number of criteria and the quality of fire refugia (Figure on page 7) is unclear; this issue requires further substantiation.

Authors: In the present version, we better explain the criteria and quality of refugia construction, see section 2.5

R. The Akaike Information Criterion mentioned in line 248 requires justification in terms of the appropriateness of the Authors' choice.

Authors: We have added a justification for the use of the Akaike Information Criterion (AIC) in the revised manuscript.

R. I suggest adding tables from attached Supplementary file to the Article, after proper analysis.

Authors: We incorporated the information of table S1 of the previous version into the main text of the manuscript, in the "results" section. Accordingly, we deleted that table from in the supplementary material (now, Table S1 corresponds to the previous Table S2 and so on). We would be happy to add some or all the four tables from supplementary material as tables in the text if the editor considers it appropriate, but it would make the article very long, we are not so sure about this.

R. Section 3. Results: The Authors provided good quantitative analyses, since the figure on page 10 is rather informative. However, the Authors omitted a qualitative analysis of the fire refuge's condition.

Authors: We added a little more text with qualitative descriptions of the sites with different refuge quality scores .

R. Section 4. Discussion: This section appears as a detailed description of the research results (Section 3). The discussion section should contain the main breakthrough results, a description of the obstacles and limitations of the research, and future ways to overcome them.

*Authors: We greatly reduced the detailed descriptions of the results, and emphasized the main and novel results, and the obstacles and limitations of the research, together with suggestions of future ways to continue the research.

R. Section 5. Conclusions: No notes to this section.

R. Common note: the manuscript must be formatted accordingly MDPI template, including References.

Authors: We formatted accordingly MDPI template, including References.

Good luck!

Authors: Thank you!

Reviewer 4 Report

Comments and Suggestions for Authors

General comment:

The manuscript addresses an important ecological question and provides valuable field data from a poorly studied region. Statistical treatment and methodological rigor require substantial improvement before publication. In particular, the mis-specification of the dependent variable, lack of spatial analysis, and minor but numerous typographic errors detract from the otherwise solid forest area.

Specific comments:

In Table 2 coefficients and z-values are presented clearly, but the table title should specify that the model used a Generalized Linear Mixed Model (GLMM), not simply a “generalized linear model,” to maintain methodological consistency.
The study design (208 plots, 1 ha each) is robust and well distributed, but the spacing criterion (“at least 60 m apart”) is insufficient to ensure spatial independence of 1-ha plots. Spatial autocorrelation could bias model results, particularly given the clustered transect design.
Pre-fire conditions were partially reconstructed post hoc from field data (“estimated pre-fire values when the fire could change these values”), which introduces potential recall or observer bias. No validation method (pre-fire imagery or field calibration) was described for these estimates.
The authors state to have used “field and GIS-derived variables,” but the integration process of grid matching between 30 m DEM and 1-ha plot boundaries is not explained, which may affect variable precision.
The refugia quality index was created by summing binary variables. This method assumes equal weighting of criteria, which may not be ecologically justified. A weighted scoring or multivariate approach would be statistically sounder.
The conversion of this index (0–5) into a binomial structure within GLMMs is questionable, since it is ordinal rather than binary. An ordinal logistic regression would be more appropriate.
The “manual forward stepwise” model selection using AIC is outdated and prone to overfitting. More robust alternatives include information-theoretic model averaging or regularization approaches. I suggest to use LASSO model to do this.
The model does not appear to have been validated. No mention of residual diagnostics, cross-validation, or overdispersion checks.
Collinearity screening at |r| > 0.50 is unusually conservative. I think that common practice is to exclude variables above |r| > 0.7 or 0.8.

Constructive feedback:

The stated objective to identify pre-fire conditions that determine the presence and quality of fire refugia is clear but somewhat narrow. It omits mention of post-fire ecological implications, although these are discussed later. The hypothesis section is well articulated but could be strengthened by defining “quality” in ecological terms before introducing quantitative criteria.

Authors used of binomial GLMM on ordinal data. Please replace with ordinal logistic regression or cumulative link mixed model. Treating the fire refugia quality as a binomial variable undermines the ordinal nature of the index. This may distort p-values and effect estimates. The interpretation of “non-significant” variables (p < 0.11) as “informative” based solely on AIC is debatable. Including variables with p > 0.05 but <0.11 without cross-validation can introduce Type I error. The authors did not report model fit statistics (pseudo-R², dispersion parameter, or random effect variance). Without these, the explanatory strength of the model is indeterminate. Although the data are spatial, no spatial autocorrelation test (maybe Moran’s I and if necessary, include spatial random effects) was applied to residuals, leaving open the risk of pseudoreplication.

The phrase “pos-fire proportion of resprouted trees” (line 199) contains a typographical error (“pos-fire” instead of “post-fire”). The Abbreviations section includes “Geographic Information Sistem” instead of “System.”

Reference style is inconsistent with MDPI formatting (lowercase journal names “fire ecol 2025” and proper capitalization “Fire Ecology”).

In the Conclusions section, URLs are embedded directly within the text, which deviates from standard academic formatting. It is recommended to relocate these citations to the Discussion section, where they can be more appropriately contextualized. Furthermore, the Conclusions should be rewritten to ensure clarity, coherence, and alignment with academic conventions. The conclusions summarize findings but largely restate results with no sufficient synthesis. There is minimal discussion of mechanistic processes or predictive implications for future fires.

Summary:

The final section includes policy recommendations (restoration prioritization) but lacks explicit linkage to the statistical results supporting those claims. The authors mention that “extreme weather conditions weaken the effects of other factors” (line 455–457), which contradicts their analytical results showing significant topographic effects. This inconsistency weakens the logical coherence of the conclusion.

Author Response

Reviewer 4

R. General comment:

R. The manuscript addresses an important ecological question and provides valuable field data from a poorly studied region. Statistical treatment and methodological rigor require substantial improvement before publication. In particular, the mis-specification of the dependent variable, lack of spatial analysis, and minor but numerous typographic errors detract from the otherwise solid forest area.

Authors: We improved statistical treatment and methodological rigor according to suggestions (see below), many thanks. We also checked typographic errors ourselves and English was revised by an expert.

R. Specific comments:

R. In Table 2 coefficients and z-values are presented clearly, but the table title should specify that the model used a Generalized Linear Mixed Model (GLMM), not simply a “generalized linear model,” to maintain methodological consistency.

Authors: We have specifically clarified the model used, which was changed to an ordinal logistic regression model in the present version following the suggestions of another reviewer (see below).

R. The study design (208 plots, 1 ha each) is robust and well distributed, but the spacing criterion (“at least 60 m apart”) is insufficient to ensure spatial independence of 1-ha plots. Spatial autocorrelation could bias model results, particularly given the clustered transect design.

Authors: This is an important point. To control this, we tested spatial autocorrelation of residuals by a semivariogram and a Mantel test, and no spatial autocorrelation was detected. We incorporated the corresponding information in the “statistical analysis” subsection of methods, and in “Topographic and land cover effects” subsection of results.

R Pre-fire conditions were partially reconstructed post hoc from field data (“estimated pre-fire values when the fire could change these values”), which introduces potential recall or observer bias. No validation method (pre-fire imagery or field calibration) was described for these estimates.

*Authors: Yes, this is a drawback to our study, even when we of course made efforts to avoid potential recall or observer bias. We did not perform a formal field calibration with data to report, but during the first three field trips we systematically performed evaluations simultaneously by two or more researchers which we then discussed to standardize, and we stopped this procedure when estimations became very similar between observers. After calibration, all estimations were performed by a single person (Ricardo Suarez). We explain in the manuscript that dominant woody plants conserve their charred trunks, therefore their pre-fire cover and height could be estimated. In the present version we added that herbaceous cover were estimated from regrowth. Our dominant herbaceous vegetation consist of tussock grasses, which shortly after the fires began to regrow, reaching in two years covers similar to those previous to the fire (Cingolani AM, Poca M, Whitworth-Hulse JI, Giorgis MA, Vaieretti MV, Herrero L, Navarro Ramos S & Renison D. 2020. Fire reduces dry season low flows in a subtropical highland of central Argentina. Journal of Hydrology 590:125538.). Additonally, note that three of the eight resulting explanatory variables were not expected by us, so at least those should presumably be free of bias. Regarding pre-fire measurements using imagery, we incorporated this suggestion into the discussion as suggestions for future studies.

R. The authors state to have used “field and GIS-derived variables,” but the integration process of grid matching between 30 m DEM and 1-ha plot boundaries is not explained, which may affect variable precision.

Authors: Elevation, topographic position and roughness values were obtained from the 30 m DEM. In table 1 was detailed the matching relation between pixels and the 1-ha plots. Elevation was obtained from the pixel of the center of the plot. Topographic position index was obtained by applying the described formula on the 315m radius circle around the center of the plot. Roughness was obtained averaging the 9 pixels surrounding the center of the plot.

R. The refugia quality index was created by summing binary variables. This method assumes equal weighting of criteria, which may not be ecologically justified. A weighted scoring or multivariate approach would be statistically sounder.

Authors: We feel there is still not enough empirical or theoretical basis to determine the relative ecological importance of each attribute in our system, thus assigning differential weights would only introduce further subjectivity. Our goal was to develop a score that integrates multiple, relevant indicators of refugia quality, ensuring a broad and balanced representation of site conditions without prioritizing any single variable.

R. The conversion of this index (0–5) into a binomial structure within GLMMs is questionable, since it is ordinal rather than binary. An ordinal logistic regression would be more appropriate.

Authors: Yes, correct. We have changed the previous GLM by an ordinal regression analysis to statistically evaluate how environmental predictors relate to the fire refugia quality score. Results were similar than in the previous version, except that one additional variable was included now.

R. The “manual forward stepwise” model selection using AIC is outdated and prone to overfitting. More robust alternatives include information-theoretic model averaging or regularization approaches. I suggest to use LASSO model to do this.

Authors: We updated the method using a stepwise forward selection based on AIC ( function stepAIC() from MASS package), restricting the inclusion of variables that were highly or moderately correlated with those already selected.

R. The model does not appear to have been validated. No mention of residual diagnostics, cross-validation, or overdispersion checks.

Authors: After running the new model (i.e. the ordinal logistic regression) we performed thorough residual diagnostics (now described in Methods. No overdispersion was detected and all the model assumptions were met as we now state in the results section.

R. Collinearity screening at |r| > 0.50 is unusually conservative. I think that common practice is to exclude variables above |r| > 0.7 or 0.8.

Authors: We adopted a conservative collinearity threshold (|r| > 0.50) to minimize redundancy among explanatory variables and ensure that each retained variable represented a relatively independent ecological factor. As correlations among predictors increase, the interpretation of model results becomes more complex and potentially misleading. Moreover, visualizing model predictions also becomes less straightforward, as fixing all non-plotted predictors at their mean values may not reflect realistic conditions.

Constructive feedback:

R. The stated objective to identify pre-fire conditions that determine the presence and quality of fire refugia is clear but somewhat narrow. It omits mention of post-fire ecological implications, although these are discussed later. The hypothesis section is well articulated but could be strengthened by defining “quality” in ecological terms before introducing quantitative criteria.

*Authors: We agree this was missing. We now added explanation highlighted in yellow before stating the hypothesis.

R. Authors used of binomial GLMM on ordinal data. Please replace with ordinal logistic regression or cumulative link mixed model. Treating the fire refugia quality as a binomial variable undermines the ordinal nature of the index. This may distort p-values and effect estimates. The interpretation of “non-significant” variables (p < 0.11) as “informative” based solely on AIC is debatable. Including variables with p > 0.05 but <0.11 without cross-validation can introduce Type I error. The authors did not report model fit statistics (pseudo-R², dispersion parameter, or random effect variance). Without these, the explanatory strength of the model is indeterminate. Although the data are spatial, no spatial autocorrelation test (maybe Moran’s I and if necessary, include spatial random effects) was applied to residuals, leaving open the risk of pseudoreplication.

*Authors: We remade all analyses using logistic ordinal regression, all results were updated in the new version. We maintained the interpretation of selected variables as potentially relevant even when they were not statistically significant, as long as they were identified as informative by AIC. However, such interpretations were made with caution. This decision was based on their relatively low p-values (< 0.15 in our study) and their support from the Akaike Information Criterion. In biological research, it is common to consider p-values up to 0.2 as potentially biologically meaningful. This approach allows for a more nuanced, less binary interpretation and prevents to neglect variables that may be relevant. In Fig. 3, we now display only variables with p < 0.1. As a statistic of model fit we now reported McFadden pseudo-R².

We tested model residuals and conducted standard diagnostic checks, including assessment of the residual distribution (expected to be uniform as we used DHARMa-like residuals), overdispersion, and potential relationships with fitted values, and with each predictor variable. No significant biases or violations of model assumptions were detected. Spatial autocorrelation was tested through semivariograms and using the Mantel approach to evaluate whether model residuals were spatially structured and no spatial structure was detected.

R. The phrase “pos-fire proportion of resprouted trees” (line 199) contains a typographical error (“pos-fire” instead of “post-fire”). The Abbreviations section includes “Geographic Information Sistem” instead of “System.”

Authors: Those errors were corrected, thanks.

R. Reference style is inconsistent with MDPI formatting (lowercase journal names “fire ecol 2025” and proper capitalization “Fire Ecology”).

Authors: The errors were corrected appropriately.

R. In the Conclusions section, URLs are embedded directly within the text, which deviates from standard academic formatting. It is recommended to relocate these citations to the Discussion section, where they can be more appropriately contextualized. Furthermore, the Conclusions should be rewritten to ensure clarity, coherence, and alignment with academic conventions. The conclusions summarize findings but largely restate results with no sufficient synthesis. There is minimal discussion of mechanistic processes or predictive implications for future fires.

*Authors: We better discuss possible mechanism and now have a new section of the discussion where we suggest that the mechanisms should be better studied and more replicates of this study with other megafires are needed to increase restrictiveness. We relocated to methods the URLs of restoration projects, see last paragraph of “Study area”.

Summary:

*Authors: Yes, that was confusing. We changed to a more logical first paragraph to the discussion.

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Good luck!

Author Response

Comments 1: "Good luck". Response 1: "thanks!"

Reviewer 4 Report

Comments and Suggestions for Authors

Authors replied to my comments correctly. Some of the things they couldn't improve, but it was explained why. The general paper now is better than the previous revised paper.

Author Response

Comments 1: Authors replied to my comments correctly. Some of the things they couldn't improve, but it was explained why. The general paper now is better than the previous revised paper.

Response 1: We are glad you found the paper better, many thanks for your time