WSN-Driven Advances in Soil Moisture Estimation: A Machine Learning Approach

Round 1

Reviewer 1 Report

Dear Authors,

The article addresses an important issue for agricultural productivity. The paper explores the integration of Wireless Sensor Networks with machine learning and deep learning to try to optimize soil moisture estimation. Five machine learning models were evaluated and compared using information from regions in India with the dataset from NASA’s SMAP project, Google Earth Engine, and Power Access Climate Data. The article is well written, the references are in line with the topic, the results are partially well presented, and there are small details that should be reviewed; there are the following suggestions:

1) Change the keywords “WSN; Remote sensing; Random Forest; LSTM”. Some do not reflect what was done in the text, and the idea is to include the exact word, not its abbreviation; it can generate wrong interpretations.

2) In line 74, it is better to review the sentence because you did not integrate your proposal with ground-based sensors; you used a platform with some of these sensors. This gives the reader an erroneous interpretation.

3) In lines 130 and 131, the first sentence is missing a comma, check.

4) Explain subsection 3.3 in more detail. “Power Access Climate Date”, lines 169 to 181.

5) In lines 186, 187, and 188, review the sentence because, using the link in ref [27], it is not possible to confirm the “soil moisture” part.

6) Explain section 4.4 better. “Dataset Preparation”, lines 244 to 258.

7) Explain in more detail how the optimal values from Table 1 in section 5 were found.

8) Explain Figure 3 in more detail; as it stands, it is not possible to properly understand several details of the figure.

Kind regards.

Dear Authors,

 In lines 130 and 131, the first sentence is missing a comma, check.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Based on the provided manuscript, I suggest the following revisions for a major revision to enhance the clarity, depth, and scientific rigor of the study:

Introduction and Literature Review:

Expand on the importance of soil moisture estimation for specific agricultural practices and environmental management, citing recent studies that highlight its critical role.

The literature review should more critically analyze existing methodologies, highlighting their strengths but also emphasizing their limitations in more detail. This can include a discussion on the scalability of WSNs, data reliability issues, and the challenges of integrating heterogeneous data sources.

Methodology:

Clarify the selection criteria for the study areas. While the manuscript mentions geographical diversity, a more systematic explanation of how these areas represent different agricultural practices and climatic zones in India would add value.

Provide more details on the data collection process, especially concerning WSNs. Discuss the frequency of data collection, the types of sensors used, their calibration processes, and any data quality control measures implemented.

The methodology section would benefit from a more detailed explanation of the data preprocessing steps. Specifically, elaborate on the missing value imputation methods and the criteria for feature selection. This could include discussing the rationale behind choosing certain lag values for time series transformation.

Machine Learning Models:

Offer a more comprehensive discussion on the choice of evaluation metrics. Explain why these particular metrics (MAE, MSE, RMSE, MAPE) are suitable for this study and how they relate to the practical application of soil moisture estimation.

For each ML model, discuss the theoretical basis that makes it suitable for soil moisture estimation. Additionally, detail the process of hyperparameter tuning, including the range of values tested and the rationale behind selecting the optimal values.

Results and Discussion:

The results section should include a more nuanced discussion of the findings. For instance, explain why the LSTM model outperforms others in greater detail, possibly by discussing its ability to capture temporal dependencies in the soil moisture data.

Discuss any potential biases or limitations of the models used. This could include addressing overfitting, the representativeness of the training data, and the generalizability of the models to other regions or soil types.

Conclusion and Future Work:

The conclusion could be strengthened by summarizing the key findings more concisely and discussing their implications for agriculture and environmental management in more specific terms.

For future work, suggest specific areas of research that could further improve soil moisture estimation models. This could include integrating additional data sources, exploring other ML algorithms, or developing real-time monitoring systems.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper presents a comprehensive investigation into soil moisture estimation, emphasizing its significance in agriculture, water resource management, and environmental monitoring. By integrating machine learning (ML) and deep learning (DL) techniques with wireless sensor networks (WSNs) and remote sensing data, the study aims to address the limitations of traditional methods and improve accuracy in soil moisture estimation. The introduction provides a thorough overview of the importance of soil moisture and the shortcomings of existing estimation techniques, setting a strong foundation for the study. The authors propose a novel methodology employing ML and DL models trained on satellite data and WSNs, with a particular focus on the LSTM model's performance compared to four other models (linear regression, SVM, decision trees, and random forest).  The paper highlights the effectiveness of the proposed approach, particularly the robust performance of the LSTM model in estimating soil moisture levels in five regions of India. Additionally, future research directions are outlined, emphasizing the importance of real-time monitoring systems and extensive validation studies. Overall, the paper offers valuable insights into advancing soil moisture estimation techniques, with practical implications for agricultural and environmental management. However, further clarity on certain methodological aspects and validation procedures could enhance the paper's rigor and impact.

Some points need to be addressed in the final version : 

 -- What is NOAA referring to? (figure 1 page 5)

 -- Equation at the bottom of the page is repeated twice (page 9)

 -- Not all the equations are enumerated (section 4.4)

 -- References 22 and 24 are the same

 -- Authors did not mention or discuss the limits of their approach.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

This study investigates the integration of Wireless Sensor Networks (WSNs) with machine learning (ML) and deep learning (DL) techniques to enhance soil moisture estimation. The authors need to address several remarks: 

1. The paper organization should be highlighted in the last paragraph of the introduction section.

2. The authors should provide compelling arguments for the data "consolidation" process, demonstrating its impact on result accuracy.

3. The process of "consolidation" lacks clarity; more details are needed regarding its necessity, inputs, outputs, and methodology.

4. The section "3. Preliminaries" requires restructuring to better align with the overall organization.

5. Computational costs of ML and DL should be compared to enrich the comparative analysis.

6. The authors should compare their results with the current state of the art.

7. Improved presentation methods are needed for the results; Table 2 lacks sufficient comparative analysis.

n/a

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 5 Report

The paperwork describes mainly the application of an ML/DP-based linear regression for a specific application. I find the paper well written and easy to follow, however, it seems to be a bit out of the field of the Electronics journal even if it covers a general application of AI even if data are acquired via wireless sensors. However, I give some comments besides suggesting to providing more details about the electronics readout network.

As a general comment, the authors use too many acronyms. For example:

- line 162: IDE and API are never used as acronyms, you can remove them

- line 174: CSV is never used as an acronym, you can remove it

- Figure 1 refers to NOAA data that are not described

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

the authors need to include what they have mentioned in their response to comment 6 in the conclusion section when they said:

"It's important to note that our study does not claim to have provided the "best" model for soil moisture estimation."

they have to reference the studies that provide the better accuracy and mention their counter argument: 

"Instead, we aim to offer a comprehensive analysis of the performance of different ML and DL models under similar conditions."

n/a

Author Response

Please see the attachment.

Author Response File: Author Response.pdf