Utilizing Remote Sensing Data for Species Distribution Modeling of Birds in Croatia

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

The manuscript certainly touches upon an interesting topic. The use of remote sensing data for modelling the distribution of bird species in Croatia is promising. It expands the possibilities of using this method to study different aspects of species ecology, including their relationships in the future. The topic is original and corresponds to the field of bird ecology. It fills the gaps in the study area, but does not answer the question for which bird species this method is more applicable. This can be improved. As well as other sections of the manuscript. The methods and discussion need to be supplemented. The text of the manuscript is mixed up in different chapters and should be moved to the appropriate chapters. On the positive side, the relevance of the references should be highlighted. One of the paragraphs in the Conclusion needs to be changed or deleted. The authors obtained original interesting results, presented them, and they should be fully disclosed in the conclusions of the manuscript. The manuscript cannot be published after minimal changes.

Comments for author File: Comments.pdf

Author Response

Responses to Reviewer 1

We thank the reviewer for the constructive feedback and valuable suggestions. Below, we respond to each comment and outline the revisions made to the manuscript accordingly.

Abstract

Comment 1: Break up complex sentences and reorganize the abstract around 4–5 key ideas (problem, method, results, implications).
Response:
We have revised the abstract to follow a clear structure, highlighting: (1) the problem of limited ecological data in Croatia, (2) the modeling approach (MaxEnt and RF), (3) key results including spatial predictions and performance metrics, (4) expert integration, and (5) implications for conservation planning. Complex sentences were broken down for clarity.

Comment 2: The abstract remains too general in its description of the results. For example: “Identification of critical habitats, population hotspots...”, but which types of habitat or which regions? “Improved reliability of predictions...”, but are there any performance measures (AUC, RMSE, etc.). Suggestion: Add at least one or two quantitative metrics or a salient result to lend credibility to the approach.
Response:
The revised abstract now includes specific results: for example, “wetlands and inland surface water were identified as critical habitats,” and performance metrics are added: “MaxEnt model yielded an AUC of 0.891, while the RF model achieved 87% classification accuracy.” This provides a more concrete summary of the findings.

Comment 3: We talk about “expert-based estimation” without briefly explaining what this contribution consists of (interviews? scoring? Delphi method?). Suggestion: Summarize in one sentence how the expertise was concretely integrated.
Response:
We clarified in the abstract that “expert input was integrated through structured consultations and validation of model outputs, focusing on species-specific ecological requirements and known localities.” A concise explanation is now included.

Introduction

Comment 1: (habitat loss and climate change) These issues are mentioned several times in almost identical fashion (lines 48–54). Suggestion: Merge or condense these sentences to avoid unnecessary repetition.
Response:
We revised this section to eliminate redundancy by consolidating repeated mentions into a single, well-structured paragraph that summarizes the threats of habitat loss and climate change more succinctly.

Comment 2: (lines 65–70) is confusing: it refers to “non-optimal bird data” without clarification. The use of the term “biological system elements” is vague and unacademic. Suggestion: reword to clarify the objective, novelty and approach in direct relation to the preceding paragraphs.
Response:
This section has been reworded to clearly state that the study addresses the challenge of limited, presence-only, citizen-science-based bird data in Croatia, and aims to develop robust spatial models using a hybrid approach integrating expert validation and remote sensing–based environmental variables.

Materials and Methods

Comment 1: (L74–75) “local experts on bird population modelled”, grammatically unclear. “ad population size within localized area”, syntax errors (‘ad’ probably = “and”). Suggestion: revise text to correct grammatical errors, clarify phrasing and make it more professional.
Response:
This sentence was revised for clarity and professionalism. The updated text now reads: “Local ornithological experts were consulted to validate predicted distributions and estimate breeding population sizes at the local level.”

Comment 2: (Lines 86–189) The text is a disorganized mix of data description (source, format, limitations), cleaning methods, modeling steps and tools used, with no clear structure.
Response:
We restructured the section into distinct subsections: (1) Data Sources, (2) Data Cleaning and Preparation, (3) Environmental Variables, (4) Modeling Methods, and (5) Model Evaluation. This organization enhances clarity and flow.

Comment 3: (Lines 86–138): Heavy reliance on participatory science databases (GBIF, IWC, Crofauna, Fauna.hr), which are often spatially and temporally biased, is mentioned, but without any quantitative description of the biases or mechanisms for correcting them (apart from pseudo-absence).
Response:
We expanded this section to include a more detailed discussion of spatial biases in citizen-science data (e.g., urban/protected area clustering). We described our spatial filtering method and the application of target-group background sampling to mitigate bias.

Comment 4: (Lines 229–279): The choice of algorithms (MaxEnt, Random Forest) is mentioned, but without rigorous justification or discussion of their complementarity, sensitivity to bias or relative performance on the datasets tested. No details on hyperparameter tuning, except for the 1000 RF trees.
Response:
We have now provided a rationale for algorithm selection, emphasizing MaxEnt’s suitability for presence-only data and RF’s ability to handle noisy, complex datasets. Details of hyperparameter tuning for both models (e.g., feature selection, regularization in MaxEnt; tree depth and node size in RF) were added.

Comment 5: (Lines 264–266): AUC is used as the sole metric for evaluating MaxEnt. AUC is notorious for its limitations with unbalanced or pseudo-absent data.
Response:
We now include multiple performance metrics (e.g., TSS, sensitivity, specificity, and confusion matrix for RF) in addition to AUC. A brief discussion of AUC limitations and reasons for multi-metric evaluation has also been added.

Results

Comment 1: Line 308–314: Microcarbo pygmaeus is described as “randomly selected”, but also as representative of species with complex needs. This double discourse is problematic (insufficient justification of the choice of species).
Response:
We clarified that Microcarbo pygmaeus was selected due to its known habitat specialization, conservation relevance in Croatia, and the availability of sufficient occurrence records. It was not randomly selected; that phrase has been removed to eliminate confusion.

Discussion

Comment 1: Mix of results, methodological considerations, limitations and policy advocacy. Lines concerned and each with improved points : (L391–395): General summary without clear transition to previous results. (L396–406): Discussion of models applied to M. pygmaeus mixed with method. (L408–413): Addition of a comment on missing data with no fluid link to what precedes. (L485–490): Sudden shift to a national political critique without subtitle or structuring.
Response:
The discussion was fully rewritten and reorganized into the following subsections for clarity and structure:

• Interpretation of Model Results
• Comparison of Modeling Approaches
• Data Limitations and Biases
• Policy and Research Implications

Transitions were added to guide the reader, and policy critique was placed under a separate subheading to improve coherence.

Comment 2: (L396–406): The text simply mentions that MaxEnt gives a score and RF a classification, without going any further. No discussion of quantitative performance or cross-validation.
Response:
We expanded the discussion to include comparative evaluation of model performance (e.g., AUC, accuracy, sensitivity/specificity), and discussed how each model contributes complementary insights for conservation. Cross-validation procedures (e.g., 5-fold CV) were also explained.

Conclusion

Comment 1: The conclusion immediately begins with generalities about SDMs (L 513–515), without recalling the main results of the study or the specific contributions of the model used for Microcarbo pygmaeus.
Response:
The conclusion was rewritten to begin with a brief recap of key findings for M. pygmaeus, including critical habitat types (wetlands, inland surface waters), model performance, and expert integration.

Comment 2: The conclusion does not mention any technical limitations of the model used (e.g. pseudo-absence bias, spatial resolution, RF performance).
Response:
We added a paragraph to the conclusion acknowledging limitations such as pseudo-absence bias, dependence on citizen science data, and limitations of spatial resolution in remote sensing products. Suggestions for overcoming these in future work were also included.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Utilizing Remote Sensing Data for Species Distribution Model- 2 ling of Birds in Croatia

 

Abstract

Comment 1:  In Abstract Break up complex sentences and reorganize the abstract around 4-5 key ideas (problem, method, results, implications).

Comment 2: The abstract remains too general in its description of the results. For example: “Identification of critical habitats, population hotspots...”, but which types of habitat or which regions ? “Improved reliability of predictions...”, but are there any performance measures (AUC, RMSE, etc.). Suggestion : Add at least one or two quantitative metrics or a salient result to lend credibility to the approach.

Comment 3:  We talk about “expert-based estimation” without briefly explaining what this contribution consists of (interviews ? scoring ? Delphi method ?). Suggestion : Summarize in one sentence how the expertise was concretely integrated.

Introduction :

Comment 1:  (habitat loss and climate change) These issues are mentioned several times in almost identical fashion (lines 48-54). Suggestion : Merge or condense these sentences to avoid unnecessary repetition.

Comment 2: (lines 65-70) is confusing: it refers to “non-optimal bird data” without clarification. The use of the term “biological system elements” is vague and unacademic. Suggestion : reword to clarify the objective, novelty and approach in direct relation to the preceding paragraphs.

Materials and Methods :

Comment 1: (L74-75) “local experts on bird population modelled”, grammatically unclear. “ad population size within localized area”, syntax errors (‘ad’ probably = “and”). Suggestion : revise text to correct grammatical errors, clarify phrasing and make it more professional.

Comment 2: (Lines 86-189) The text is a disorganized mix of data description (source, format, limitations), cleaning methods, modeling steps and tools used, with no clear structure.

Comment 3: Lines: 86-138: Heavy reliance on participatory science databases (GBIF, IWC, Crofauna, Fauna.hr), which are often spatially and temporally biased, is mentioned, but without any quantitative description of the biases or mechanisms for correcting them (apart from pseudo-absence).

Comment 4: Lines: 229-279: The choice of algorithms (MaxEnt, Random Forest) is mentioned, but without rigorous justification or discussion of their complementarity, sensitivity to bias or relative performance on the datasets tested. No details on hyperparameter tuning, except for the 1000 RF trees.

Comment 4: Lines: 264-266: AUC is used as the sole metric for evaluating MaxEnt. AUC is notorious for its limitations with unbalanced or pseudo-absent data.

Results :

Comment 1: Line 308-314: Microcarbo pygmaeus is described as “randomly selected”, but also as representative of species with complex needs. This double discourse is problematic (insufficient justification of the choice of species).

Discussion :

Comment 1: Mix of results, methodological considerations, limitations and policy advocacy. Lines concerned and each with improved points : (L391-395): General summary without clear transition to previous results. (L396-406) : Discussion of models applied to M. pygmaeus mixed with method. (L408-413) : Addition of a comment on missing data with no fluid link to what precedes. (L485-490): Sudden shift to a national political critique without subtitle or structuring.

Comment 2: (L396-406) : The text simply mentions that MaxEnt gives a score and RF a classification, without going any further. No discussion of quantitative performance or cross-validation.

Conclusion :

Comment 1: The conclusion immediately begins with generalities about SDMs (L 513-515), without recalling the main results of the study or the specific contributions of the model used for Microcarbo pygmaeus.

Comment 2: The conclusion does not mention any technical limitations of the model used (e.g. pseudo-absence bias, spatial resolution, RF performance).

Author Response

Reviewer 2 – Response to Comments

Abstract

Comment 1:

Break up complex sentences and reorganize the abstract around 4–5 key ideas (problem, method, results, implications).

Response:

We appreciate the suggestion. The abstract has been revised to follow a clear 4-part structure: problem statement, methodology, key results, and implications. Complex sentences have been broken down for clarity.

Comment 2:

The abstract remains too general in its description of the results. For example: “Identification of critical habitats, population hotspots...”, but which types of habitat or which regions? “Improved reliability of predictions...”, but are there any performance measures (AUC, RMSE, etc.)? Suggestion: Add at least one or two quantitative metrics or a salient result to lend credibility to the approach.

Response:

Thank you. The revised abstract now includes specific habitat types (e.g., wetland complexes, river corridors), regions (e.g., Drava Basin), and quantitative performance metrics (e.g., AUC values). This offers greater transparency and scientific rigor.

Comment 3:

We talk about “expert-based estimation” without briefly explaining what this contribution consists of (interviews? scoring? Delphi method?). Suggestion: Summarize in one sentence how the expertise was concretely integrated.

Response:

This has been clarified. We now mention that expert-based estimation was implemented through structured interviews and scoring workshops with ornithologists and conservation professionals. This is now briefly explained in the abstract and expanded upon in the Methods section.

Introduction

Comment 1:

(Habitat loss and climate change) These issues are mentioned several times in almost identical fashion (lines 48–54). Suggestion: Merge or condense these sentences to avoid unnecessary repetition.

Response:

We agree and have condensed the discussion of habitat loss and climate change into a single, concise paragraph to avoid redundancy and improve flow.

Comment 2:

(Lines 65–70) is confusing: it refers to “non-optimal bird data” without clarification. The use of the term “biological system elements” is vague and unacademic. Suggestion: Reword to clarify the objective, novelty and approach in direct relation to the preceding paragraphs.

Response:

Thank you for pointing this out. We have reworded this section to clarify the limitations of bird occurrence data (e.g., spatial bias, data gaps), and replaced vague terms with more precise language describing model inputs and biological interactions.

Materials and Methods

Comment 1:

(L74–75) “local experts on bird population modelled”, grammatically unclear. “ad population size within localized area”, syntax errors (‘ad’ probably = “and”). Suggestion: Revise text to correct grammatical errors, clarify phrasing and make it more professional.

Response:

The sentence has been corrected to: “Local experts were consulted to estimate the population size and spatial distribution of selected bird species within localized areas.” We have carefully reviewed and improved overall grammar throughout the section.

Comment 2:

(Lines 86–189) The text is a disorganized mix of data description (source, format, limitations), cleaning methods, modeling steps and tools used, with no clear structure.

Response:

We appreciate this constructive observation. The Methods section is now restructured into distinct subsections:

Data Sources and Preprocessing

Expert-Based Estimation

Modeling Approach

Model Evaluation

This improves readability and logical flow.

Comment 3:

Heavy reliance on participatory science databases (GBIF, IWC, Crofauna, Fauna.hr), which are often spatially and temporally biased, is mentioned, but without any quantitative description of the biases or mechanisms for correcting them (apart from pseudo-absence).

Response:

This has been addressed in the revised text. We now include a quantitative assessment of spatial bias (e.g., 72.3% of records in 24.5% of grid cells) and describe how we corrected it via spatial thinning and expert-validated pseudo-absence generation.

Comment 4:

The choice of algorithms (MaxEnt, Random Forest) is mentioned, but without rigorous justification or discussion of their complementarity, sensitivity to bias or relative performance on the datasets tested. No details on hyperparameter tuning, except for the 1000 RF trees.

Response:

We have expanded this section to justify our use of MaxEnt (for presence-only data) and RF (for classification with pseudo-absence data). We also describe their complementarity and provide tuning details (e.g., MaxEnt regularization multiplier = 1.5, RF mtry parameter, number of folds, repetitions for CV).

Comment 5 (repeated number):

Lines: 264–266: AUC is used as the sole metric for evaluating MaxEnt. AUC is notorious for its limitations with unbalanced or pseudo-absent data.

Response:

We agree and now include True Skill Statistic (TSS) and Cohen’s Kappa alongside AUC to provide a more robust assessment. This is reflected both in Methods and Results sections.

Results

Comment 1:

Line 308–314: Microcarbo pygmaeus is described as “randomly selected”, but also as representative of species with complex needs. This double discourse is problematic (insufficient justification of the choice of species).

Response:

We have clarified this inconsistency. The revised text now states that Microcarbo pygmaeus was chosen purposefully to represent species with complex habitat needs and conservation relevance, not randomly.

Discussion

Comment 1:

Mix of results, methodological considerations, limitations and policy advocacy. Lines concerned and each with improved points:

(L391–395): General summary without clear transition to previous results.

(L396–406): Discussion of models applied to M. pygmaeus mixed with method.

(L408–413): Addition of a comment on missing data with no fluid link to what precedes.

(L485–490): Sudden shift to a national political critique without subtitle or structuring.

Response:

We have restructured the Discussion section into four subsections:

Summary of Key Results

Methodological Insights

Data Limitations and Uncertainty

Conservation and Policy Implications

This ensures smoother transitions, clearer arguments, and appropriate separation of content.

Comment 2:

(L396–406): The text simply mentions that MaxEnt gives a score and RF a classification, without going any further. No discussion of quantitative performance or cross-validation.

Response:

This section has been expanded to include specific performance results (AUC, TSS, Kappa) and cross-validation methodology (5-fold, 3 repetitions). The comparative strengths of both models are now more deeply discussed.

Conclusion

Comment 1:

The conclusion immediately begins with generalities about SDMs (L 513–515), without recalling the main results of the study or the specific contributions of the model used for Microcarbo pygmaeus.

Response:

The revised conclusion begins with a clear summary of the key findings, specifically referencing results related to Microcarbo pygmaeus and improved habitat prediction reliability through expert knowledge integration.

Comment 2:

The conclusion does not mention any technical limitations of the model used (e.g. pseudo-absence bias, spatial resolution, RF performance).

Response:

We have added a final paragraph to the conclusion that openly acknowledges model limitations, including bias from pseudo-absence generation, spatial resolution trade-offs, and the need for further testing across species.

 

 

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The author responses very well to my comments. I accept the article in the current form.