Modeling Spongy Moth Forest Mortality in Rhode Island Temperate Deciduous Forest
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe manuscript uses a large amount of data to predict tree mortality, but there are several issues with the paper.
Major Issues:
1. The title does not seem to align with the content of the paper.
Although both the title and the introduction mention the spongy moth outbreak, most of the variables seem unrelated to the outbreak. Only line 88 mentions, "Forests near urban areas may have higher exposure to invasive plants or pests." If the variables (Tables 1 - 3) are related to the spongy moth outbreak, this should have been clarified in the introduction.
2. The analysis model is overly simplistic.
Why does this paper use random forest instead of other machine learning models such as gradient boosting decision trees or support vector machines? Why not use a variety of machine learning models and select the best-performing one for subsequent analysis? This could make the results more convincing.
Minor Issues:
3. The basis for the classification in the 2-class and 3-class models mentioned in lines 306 to 309 is unclear.
4. When using the random forest model, the authors seem to have used 5-fold cross-validation (mentioned in line 310). Is the data used in the SHAP analysis from one of the model training results?
5. The SHAP-related analysis is too simple, relying only on beeswarm (Figures 4 and 5). Please add dependence scatter plots.
6. There are numerous formatting issues, such as the font size in Sections 2.2 and 2.3 not being consistent with other sections. The format of Equation (1) is inconsistent with other equations. Additionally, the formatting of the tables and citations seems to differ significantly from other published papers. The authors should download the formatting template from the journal's official website for correction.
Author Response
Thank you for your comments and suggestions. Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis study examines the impact of the Spongy Moth outbreak on tree mortality in temperate deciduous forests in Rhode Island, addressing a meaningful and practical research question. The manuscript uses a Random Forest model combined with satellite remote sensing and geospatial data to predict forest mortality rates. The overall research methodology is reasonable, the results are clear, and the findings have valuable implications for forest management and ecological conservation. I recommend minor revisions before acceptance for publication.
1.The introduction is currently lengthy and repetitive, with extensive citations of related studies but insufficient synthesis of key insights. It would be better to streamline the background and literature review sections, emphasizing the research objectives and innovative aspects of the study.
2.The current interpretation of the model's variables relies only on basic SHAP analysis, without detailed exploration of variable interactions or their ecological implications. I suggest using tools like partial dependence plots to enrich the discussion on the ecological significance of key variables and to improve the presentation of the data.
3.Finally, the conclusions are limited to temperate deciduous forests in Rhode Island. The authors should consider discussing the potential applicability of the model to other regions or clearly defining the limitations of the study's scope.
Author Response
Thank you for your comments and suggestions. Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe manuscript "Modeling Spongy Moth Forest Mortality in Rhode Island Temperate Deciduous Forest" provides relevant insights into the use of Random Forest modeling to predict forest mortality. Although the study is novel and relevant, several aspects need further clarification, refinement, or enhancement to improve the overall quality of the manuscript. Specific comments are provided below, organized by section and line numbers, to help guide revisions.
General Comments
1. Clarity of Objectives: The objectives of the study are well-defined, although some methodological choices, such as weighting in the defoliation index, are not well justified.
2. Although comprehensive, the methodological depth has left many gray areas open to explanation, especially on data processing and validation.
3. Results and Interpretation: Results were clearly presented, but discussion, especially in terms of the ecological implications of urban proximity and partial defoliation, could have been better integrated.
4. References: Although appropriate, the references are outdated or scant in some important areas, such as soil and defoliation metrics. The addition of recent studies would enhance the manuscript.
Line-by-Line Comments
- Lines 7–13: Classifying forest mortality as low, medium, and high is done without any ecological rationale for the thresholds used. For example, >5 or >11 trees/ha. The rationale or references should be discussed.
- Lines 22–29: Though the impact of invasive pests is very well highlighted, the background on Spongy moth-specific impacts on mortality draws heavily from older references, such as Davidson et al. (1999). Update with more recent studies reflecting advancement in pest ecology.
- Lines 47–54: While the introduction does cite examples of pest-induced mortality in other forest types, it would have greater contextual relevance to include a brief statement on why deciduous forests differ in mortality factors, for example, species diversity.
- Lines 211–229 (Defoliation Index): The formula of defoliation index is simple: Defoliation Index=fD1+2×fD2+3×fD3, but weights are not explained. Justification of such a choice or sensitivity analysis should be performed to prove that this is the right choice.
- Lines 233–239 (Canopy Cover - FRCI): Clearly define the height threshold (>3 m) that will be used for "FirstCanopy". Explain how noise, i.e., understory vegetation or small shrubs will be filtered to measure the actual canopy cover.
- Lines 243–253: Drought Index is summer-averaged, which could mask peak drought stress. Consider testing finer temporal resolutions or interaction effects (e.g., drought × defoliation).
- Lines 267–283: Soil Metrics exhibited low importance within models; this was probably a consequence of low resolution-1 ha minimum mapping unit size. Combining soil characteristics into composite indices-for example, water retention potential-could enhance their predictive utility.
- Lines 289–298: The training dataset is dominated by high mortality tiles (>20 trees/ha), and the model could be biased. Consider stratified sampling or weighting techniques.
- Lines 342–348: The finding that 23% of Rhode Island's forest experienced significant defoliation is really important. However, partial defoliation events (e.g., < 75% crown defoliation) are not accounted for. These may lead to longer-term mortality. In such cases, try the integration of partial defoliation data by mapping gradients of defoliation using finer resolution imagery, such as Sentinel-2 or aerial data.
- Lines 394–404 (Variable Importance): The top predictors are the defoliation index, coastal proximity, and canopy cover, while less impactful variables, such as soil and slope metrics, have not been discussed as well. Discuss possible reasons why these predictors are less ecologically relevant or underused.
- Lines 447–453: Interpretation of the result of SHAP analysis. Discuss for example why low canopy should be related to higher mortality, independently of any heating and evaporation effects. May disturbance preconditioning, like large storms have occurred?
- Lines 460–466 (Urban Proximity): The finding of lower mortality associated with urban proximity was unexpected. Consider active urban forest management, microclimatic effects, or pest control efforts as possible confounding variables.
- Defoliation metrics: Include those works that involve advanced remote sensing techniques, such as Sentinel-2 or LiDAR-based severity mapping (e.g., Meng et al., 2022).
- Urban Forestry: Include studies on the management and pest control of urban trees, such as Nowak & Crane 2002.
- Topographic and Soil Impacts: More recent works on soil moisture and nutrient variability in mortality modeling can give better context.
Author Response
Thank you for your comments and suggestions. Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for Authors1.The abstract should be supplemented with additional background information and ecological significance, highlighting the harmful impacts of the Spongy moth and the representativeness of the study area.
2.In lines 23–129, the introduction section should strengthen the comparison between domestic and international studies to highlight the innovation of this study. Additionally, the citation format is incorrect and does not conform to the Forests journal standard.
3.Consider elaborating on why Rhode Island's temperate deciduous forests were chosen as the study area and whether they possess any specific representativeness or uniqueness.
4.Although the study mentions using Landsat-derived defoliation maps and geospatial data, it does not emphasize the innovation of the methods or results. It is recommended to clarify how this study goes beyond or supplements previous research, for example, whether it is the first to combine specific variables or achieve breakthroughs on a regional scale.
5.A brief explanation of model performance could be added. For instance, why did the 3-class classification model (65% accuracy) perform worse than the 2-class classification model (82% accuracy)? Is this related to data distribution or classification difficulty?
6.The discussion section should include a comparative analysis with existing studies to highlight the contributions of this research. It is recommended to clearly state the contributions, limitations, and future research directions of this study. For example, can this method be applied to other regions? Would higher-resolution data be needed to optimize the model?
Author Response
Thank you for your comments and suggestions. Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have made revisions based on my recommendations. I agree to accept it.
Reviewer 4 Report
Comments and Suggestions for AuthorsThe quality of the paper has been improved after revisions, and I agree to its publication.