Interpretable Network Framework for Predicting the Spatial Distribution of Chromium in Soil

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article proposes a lightweight PHMS-Transformer model for predicting the spatial distribution of heavy metals (taking chromium as an example) in soil. This model combines multi-source environmental variables and adopts shallow encoders, dynamic pooling strategies and multi-head self-attention mechanisms. While alleviating the overfitting problem, it can effectively capture the complex nonlinear relationships among variables. The research was carried out in the border area between Chizhou and Xuancheng, Anhui Province, and 2,035 data points were collected. By comparing with various machine learning models, the PHMS-Transformer performs better in terms of prediction accuracy, generalization ability, etc. Furthermore, the SHAP model was used to interpret the prediction results, clarifying the key environmental factors affecting the spatial distribution of chromium, providing a theoretical basis and practical guidance for soil environmental protection and the prevention and control of heavy metal pollution.

My advice is to accept after minor revisions,The paper can in principle be accepted after revision based on the reviewer’s comments. I am given for minor revisions.Modification suggestions are as follows:

1. Introduction（p37）：the overall content is rather brief and fails to fully elaborate on the current situation, hazards of soil heavy metal pollution and the deficiencies of existing research, resulting in the research background and significance not being prominent enough.
2. 2.1. Soil Sampling and Chemical Analysis（p165）：only the chemical analysis of the collected soil samples was mentioned, but the verification process of the accuracy and reliability of the analysis methods was not mentioned. It is suggested to supplement relevant content, such as whether to conduct comparative analysis and verification with standard samples, to enhance the credibility of the experimental results.
3. 2.2.2. Feature Screening（p184）：Section 2.2.2 In the feature screening section, although the 22 significant features screened out were mentioned, the screening process and basis of these features were not elaborated in detail.
4. 2.3. PHMS-Transformer（p225）：Some expressions in the article are not easy to understand and may cause difficulties for readers in comprehension. For example, in the description of the structure of the PHMS-Transformer model in Section 2.3, a large number of professional terms and complex sentence structures are used. It is suggested to reexpress it in more concise and clear language, or add schematic diagrams to assist in the explanation to improve readability.
5. 3. Interpretation of the Model Prediction Results（p395）：The resolution of some charts in the article is relatively low, not clear and aesthetically pleasing enough, which affects the overall quality of the article. For example, in Figures 3.3 and 3.4, the annotations of the legends and coordinate axes are not clear and definite enough. It is suggested that the size of the pictures be uniform
6. Conclusions（p433）：The main conclusions of the PHMS-Transformer model were summarized, but the limitations of the research and the specific contents of future prospects were not mentioned. It is suggested to supplement.
7. Reference（p466）：The references are cited too few, and the format of some literature is not standardized.

Author Response

We replied to the reviewers' questions through the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Editor

After detailed readings in the manuscript, entitled: "An Interpretable Network Framework for Predicting the Spatial Distribution of Heavy Metals in Soil", It should be noted that research into the spatial distribution of heavy metals in soil is essential for understanding heavy metal soil pollution and accurately assessing the associated environmental risks. Objectively analyzing the Chizhou-Xuancheng border area in southern Anhui Province, machine learning models, including AdaBoost, GBDT, XGBoost, and MLP, were used to predict the heavy metal Cr in conjunction with environmental covariates. The results are of great importance to readers of this journal, as they confirm that the PHMS-Transformer model achieves superior performance, demonstrating high accuracy and generalization capacity, aimed at soil environmental protection and heavy metal pollution control. I suggest ACCEPTING the manuscript with major corrections:

1 – I suggest replacing the title: "An Interpretable Network Framework for Predicting the Spatial Distribution of Heavy Metals in Soil", with: "Interpretable Network Framework for Predicting the Spatial Distribution of Heavy Metals in Soil", I believe it would be clearer to the reader.

2 - At the end of the “Abstract,” you could highlight the global importance of this study, which would further pique the curiosity of readers of this journal.

3 - On page 2, lines 49 to 51, the following excerpt: “In-depth research on predicting the spatial distribution of soil heavy metal pollution is crucial for developing effective environmental protection policies and sustainable development strategies”, contains very strong statements, I suggest that it be supported with sources from the literature.

4 - At the end of the introduction, the importance of the study on a global scale should be better included.

5 - Throughout the manuscript, figures should be referenced in the text before they are displayed, to allow for better understanding by the reader.

6 - The figures are in excellent resolution, congratulations to the authors.

7 - In the methodology section, on page 9, lines 268 to 276, I suggest that this text be better substantiated with bibliographical references that support the arguments.

8 - In “Results and Discussion,” item “3.1. Evaluation Metrics” has no bibliographic citations. I suggest inserting these discussions.

9 - In “Results and Discussion,” item “3.2. Evaluation of the performance of different models.” I suggest inserting these discussions with sources from the literature.

10 - The conclusion is well founded, together with the good quality of the English used in the text, which is clear and understandable. Congratulations.

Author Response

We replied to the reviewers' questions through the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The paper addresses the prediction of the spatial distribution of heavy metals in soil, a critical issue for environmental management and public health.
The introduction provides a comprehensive overview of heavy metal pollution, including the natural and anthropogenic factors that influence it, as well as existing prediction methods (statistical, machine learning, and deep learning). It discusses several models (AdaBoost, GBDT, XGBoost, MLP, Transformer, and PHMS-Transformer) and the use of performance indicators (MAE, RMSE, R²).
The description of sampling and laboratory analysis (XRF) could be further developed, as could the selection of environmental variables and geographic and physicochemical data sources.
However, although the title and parts of the abstract and introduction refer broadly to "heavy metals in soil", the entire empirical analysis is focused exclusively on chromium. The data collection, modelling, SHAP interpretation, and spatial maps presented refer solely to the concentration of Cr in soil and not to other heavy metals such as Pb, Cd, Ni, or Zn. By using the general term "heavy metals," the reader is led to expect a multielement study, with analysis of distribution, interactions, or familiar sources of several metals. Conclusions about migration mechanisms, adsorption, influence of pH, distance to rivers, etc., are specific to Cr and cannot be automatically generalized to other metals without prior analysis.
In revising the article, authors should:
- Adjust the title and abstract to reflect that the study focuses specifically on chromium clearly.
- Reword the introduction to contextualize the general importance of heavy metal pollution, but make it clear from the outset that the focus of the study is solely on Cr due to its local relevance and data availability.
- Reserve generalizations about "heavy metals" for future discussion in case the model is intended to be applied to other elements.

Author Response

We replied to the reviewers' questions through the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Editor
The authors carefully addressed the requested revisions and implemented all recommendations. I suggest ACCEPTING the manuscript for publication.