Rapid Determination of Soil Horizons and Suborders Based on VIS-NIR-SWIR Spectroscopy and Machine Learning Models

Round 1

Reviewer 1 Report

The specific comments can be seen in the attached pdf-file.

• The discussion section is not well focused on the research results.

There are some number errors.

Comments for author File: Comments.pdf

The English of this manuscript need to be improved.

Author Response

Dear reviewer, please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,

Here are some general comments I made about your article.

Below are some general comments and specific points following the peer-reviewed version attached because because there is no line numbering in the version made available for review.

General comments:

 

The article under review is adequate in terms of form and content. It brings a background on soil spectroscopy and its applications. The idea is not necessarily new but advances in the testing of models, seeking to point out what has the best results for predicting horizons and classifying subclasses. This research aims to leverage modern technology, including machine learning and spectral data analysis, to enhance the classification of soil horizons and suborders, contributing to a more efficient and accurate understanding of soil characteristics.

The introduction contains the mandatory elements such as theoretical framework, justification (although it uses weak arguments), research problem, hypothesis and objective(s). Despite being Latinized, the writing in English is intelligible to the reader and demonstrates well-founded research. However, it needs to improve in terms of arguments and examples in paragraphs, in addition to the excessive use of the passive voice.

 

The study presents a method to classify Soil Horizon and Suborder Using VIS-NIR-SWIR Spectroscopy and Machine Learning Models. The authors argue that it is a way to improve the efficiency of the traditional methods to categorize soils using advanced analytical and computational techniques, particularly machine learning models.

Background: The authors provide sufficient research references, but many are outdated. They highlight the need to develop more efficient and cost-effective methods for classifying soils regarding their horizons and suborders. The justification is that traditional methods have limitations in terms of cost and time, and therefore, there's an increasing interest in integrating these conventional methods with machine learning and spectral reflectance data analysis.

 

Objectives: The authors distributed the goals in subtopics. I do not consider this form because the paper is not a list of procedures performed in methodology (opinion of mine). The aim must be short, clear and direct in only one period.

Methodology: The researchers collected soil samples from seven different locations in SC state and classified them in the Brazilian and American soil classification systems. They used a spectroradiometer (ASD Fieldspec) to obtain spectral reflectance data along every soil monolith, resulting in 800 samples per monolith for horizon classification and 5600 for soil classification in suborder level.

The design of the study is suitable and detailed enough. However, the date of sampling is missing.

 

Data Analysis:

The authors used the Principal Component Analysis (PCA) to analyze the spectral fingerprints of the soil samples. Thus, they divided the spectral data into training (70%) and test (30%) sets for model training and evaluation, respectively.

Five ML algorithms modeled the datasets using adjustment parameters to optimize their performance.

Performance Metrics of the study were accuracy and F-Score, which assessed the ML models' task in classifying soil horizons and suborders.

 

There were the following key findings:

Particle size and soil organic carbon have significantly influenced the soils' spectral response.

The PCA analysis showed that topsoil horizons tended to cluster together in most monoliths, while subsoil horizons had more overlap in their data.

Among the ML models, the Artificial Neural Network achieved the highest accuracy in classifying soil horizons, while the Support Vector Machine had the lowest performance. The classification of soil suborders reached high accuracies.

Conclusion: The study's results suggest that using spectral data and ML models can be highly effective in discriminating and classifying soil horizons and suborders. This approach enhances traditional soil classification methods and improves soil analysis's predictive capabilities. It remains to consider the perspectives its results may represent in future studies.

Comments for author File: Comments.pdf

I know that appositions, ellipses, deictic elements and passive voice are frequently used as a writing style in Latin languages. However, English uses a direct, short and objective language to avoid possible ambiguities. For example, see the last period of the first Introduction paragraph:

"Soil classification improves the analysis of its formation processes and its respective physical and chemical properties, which is crucial for optimizing land use, enhancing the preservation of this natural resource."

Words like "which" can became difficult to retrieve the referenced information. Are you referring to classification or analysis? Be clearer when referencing information.

Another serious lack is the argumentation that each paragraph must have. It will be necessary to explain the reason for certain statements, or at least to exemplify.

Author Response

Dear reviewer, please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

1、The length of profile is 160cm and the spot size of contact probe is 1cm. Whether 5 spectroscopy were collected every 1 cm?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors, I have saw that you followed my suggestions.

I believe it substantially improved the quality of your paper (which was already good).

I have some minor adjustments and suggestions that still remain to be observed, as follows:

1 -In future manuscripts, number the lines. The MDPI's template already has them. So follow it to make reviewing and editing easier.

2 - According to which/how many researchers? It is mandatory to write numbers for reference. However, you can also cite 'First Author et al. (ANO) [38]' for clarity and coherence of the text, as mentioned. Please consider doing this or removing the marked part for clarity and consistency.

3 -  'Conclusion' is the name of the topic, regardless of the number of insights you found.

 

Without further ado, I wish you success in your future research.

 

Best regards

Comments for author File: Comments.pdf

The Quality of English Language has been improved. No serious issues to be reported here.

Author Response

Dear reviewer, please see the attachment.

Author Response File: Author Response.pdf