Comparative Analysis of Machine-Learning Models for Soil Moisture Estimation Using High-Resolution Remote-Sensing Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors attempt to analyze various machine-learning models for soil moisture retrieval by combining Sentinel-1 SAR, Sentinel-2 multispectral data, as well as NASA DEM and geographical code and temporal code features, which is very interesting and valuable for soil moisture retrieval based on SAR data. However, more modifications for the manuscript are needed. Some comments are listed as follow.

 

Major comments

1. The introduction has to be further improved to clearly explain why the authors chose the machine learning method rather than the empirical or semi-empirical models, or the interferometry technique.

2. From Figures 7, 8, 9, it is hardly said the VV has high correlation with SM. The time series of SM has obvious difference with SM. Please compute the correlation coefficient (R) between the VV and SM, and then address this question.

3. In Figure 9, the LAI has significant difference between 2019 and 2020. please further check the processed LAI data.

4. There are obvious stripes in Figure 10 and Figure 12 except the Figure 12(b). Please further check the input variables, especially for the Sentinel-1 data.

5. Please specify the innovative of this article in the sections of introduction, abstract, and conclusion…

6. Do the machine learning methods use the same training and test datasets?

7. Please give the cross-correlation (R) of the feature parameters. It is necessary to keep the input feature datasets have minimum redundancy.

 

Minor comments

1. Line 40: please revise as “…utilizes the scattering of microwaves to…”

2. Please further modify the Figure 1, such as: the text on the figure is blurry, the figure title, …

3. What’s the spatial resolution for the distribution of soil moisture in Figures 10 and 12.

4. What’s the spatial resolution for the processed Sentinel-1 SAR data?

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

In general, the manuscript is well-written and easy to understand. The authors compared various machine learning techniques and assessed their accuracies in estimating soil moisture in the manuscript. As the research interest in machine learning grows, this study could help researchers further understand some benefits and limitations of machine learning for applications related to climate change studies.

Some suggestions have been provided to help improve the manuscript, with detailed comments shown below:

Lines 101 to 107: How were different machine learning models selected for this study? Were there any considerations or assumptions made in choosing the ensemble learning and deep learning models to be used?

Lines 202 to 210: Although these machine learning models have been widely used, it may be useful to include a summary of the main characteristics, advantages and disadvantages of these techniques to provide readers with an overview and enhance their understanding of this topic.

Figure 4: What does the color contour represent in the figure? Is it the accuracy between measured and predicted values (i.e., blue refers to low match whereas red means exact match)?

Lines 301-305: Please provide references to substantiate the statements.

Lines 478-480: How will the findings from this study be applied to other studies if the model cannot be generalized due to limited site data? Are there any specific issues that future researchers should consider if the machine learning models developed in this study are to be used?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

In this manuscript, machine learning methods are used to estimate soil moisture from synthetic aperture radar (SAR) and optical remote sensing data.  The authors compare using various machine learning methods alone with various combinations.  They find that the combinations are more skillful than the lone methods, especially if the combinations contain similar types of machine learning methods.

 

This is a well-written and timely paper.  Still, I have some minor comments that should be addressed before publication:

 

Line 119:  “The ShanDian River, emerging at the Hebei-Inner Mongolia border, encompassing a basin of roughly 12,700 km².”  This is not a complete sentence, as there is no verb and subject.  You should connect this to the next sentence.

 

Line 335-337:  “[R]adar backscatter signals effectively reflect soil moisture conditions rather than vegetation information.”  This contradicts with the sentence that follows in Lines 341-343:  “Previous research [48] has shown that vegetation coverage can influence the estimation of soil moisture using radar backscatter signals…”  One or both sentences need to be revised.  One possibility:  Replace the word “influence” with “degrade” which is more accurate.

 

Line 365:  The authors mention a “crop area,” but the general reader won’t know where that is unless they know the area.  Therefore, the authors need to show land cover type, say, in Figure 1.  MODIS provides such a data set, for instance.  

 

Also, it’s hard to see the text in the figures.  They need to be bigger, especially in Figures 10 and 11.

Comments on the Quality of English Language

The English is mostly acceptable.  There is only one comment I have regarding an error in English at Line 119 which is given in my main comments.

Author Response

Please see the attachment.

Author Response File: Author Response.docx