Next Article in Journal
Assessment of Urban Subsidence in the Lisbon Metropolitan Area (Central-West of Portugal) Applying Sentinel-1 SAR Dataset and Active Deformation Areas Procedure
Next Article in Special Issue
Rice Crop Height Inversion from TanDEM-X PolInSAR Data Using the RVoG Model Combined with the Logistic Growth Equation
Previous Article in Journal
Inter-Comparison of Diverse Heatwave Definitions in the Analysis of Spatiotemporally Contiguous Heatwave Events over China
Previous Article in Special Issue
Finite-Region Approximation of EM Fields in Layered Biaxial Anisotropic Media
 
 
Article
Peer-Review Record

A Comprehensive Comparison of Machine Learning and Feature Selection Methods for Maize Biomass Estimation Using Sentinel-1 SAR, Sentinel-2 Vegetation Indices, and Biophysical Variables

Remote Sens. 2022, 14(16), 4083; https://doi.org/10.3390/rs14164083
by Chi Xu 1, Yanling Ding 1,*, Xingming Zheng 2, Yeqiao Wang 3, Rui Zhang 4, Hongyan Zhang 1, Zewen Dai 1 and Qiaoyun Xie 5
Reviewer 1: Anonymous
Reviewer 2:
Reviewer 3:
Reviewer 4:
Reviewer 5:
Remote Sens. 2022, 14(16), 4083; https://doi.org/10.3390/rs14164083
Submission received: 25 June 2022 / Revised: 16 August 2022 / Accepted: 18 August 2022 / Published: 20 August 2022

Round 1

Reviewer 1 Report

The manuscript aims at exploring the potential of both S1 and S2 in estimating corn total dry aboveground biomass. Two machine learning methods were used in the assessments. The topic of the manuscript is of interest to the journal. However, all the assessments on the model calibration and validation were only conducted based on the three statistical indices, and their physical backgrounds on the model development and results were not well present and discussed. For instance, (1) why different models using different inputs of S1 or S2 (Figure 2) was not well present.  (2) Details on why the Jun_(VV+VH), Jun_VH, and Aug_(VH-VV) from S1 were best for the estimations were not discussed, and similar requests other selected indictors from S1 or S2 are suggested in the Discussion section. (3) Figure 7/8 shows that the biomass , especially the high biomass, the was poorly estimated. Saturation issues can be found for biomass> 0.2 kg/m2.  The authors may need to re-consider their model development on using both satellite data.

Author Response

Thank you for comments concerning our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. Those comments are valuable and very helpful. We have provided a point-by-point response to the comments. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this study, SAR polarization indices from Sentinel-1, vegetation indices from Sentinel-2, and biophysical variables (LAI, FCOVER, FAPAR, CWC and CAB) were used for maize biomass estimation. The RF-based model possessed the highest accuracies. The paper is on an interesting topic and the results are useful for the RS community. I list here below some comments to improve the presentation, but in general I would say that the manuscript requires minor revisions only.

 

2.2. Satellite data pre-processing and derived variables

How did you measure the biophysical variables? Do you have any information about the accuracies? More details are required.

 

2.3. Maize biomass modeling and Feature Selection

How did you optimize the hyperparameters of ML algorithms?

 

Figure 5.

You should have explained 'IncNodePurity' in the method section.

Author Response

Thank you for comments concerning our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. Those comments are valuable and very helpful. We have provided a point-by-point response to the comments. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper compares different combinations of polarization indices, vegetation indices, and biophysical variables derived from the sensor observations onboard Sentinel satellites.

The description for the methodology needs to be improved. More information on ground truth data collection should be provided in either the method section or the result section. The description for training data should be able to answer question like - How many data are collected for the biomass ground truth data? How are the biomass data collected?

Author Response

Thank you for comments concerning our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. Those comments are valuable and very helpful. We have provided a point-by-point response to the comments. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Overall Assessment

In this paper, gaussian Process Regression and random Forest algorithm were used to realize biomass inversion of maize based on Sentinel-1 and Sentinel-2 image data. The influence of characteristic variables under different image data on the estimation accuracy of maize biomass in different months was compared and analyzed. This study is innovative to some extent, but there are still some problems. The details are as follows:

Introduction

Point 1: Page 2, line 49~119 The overall introduction is confusing and not clear enough. It is suggested that the author should reconsider the composition from the perspectives of maize biomass estimation, image data, feature selection and the importance of optimal vegetation index.

Study areas and data

Point 2: Page 3, Figure 1. Please ask the author to verify whether the measured data in June shown in Figure 1 are too concentrated, resulting in poor representation of the estimated accuracy of the experiment for the whole research area.

Point 3: Page 4, line 140~141 The field measurement process of 10×10m sample plot is described in detail.

Point 4: Page 4, Please explain how the sample data measured in the field of 10*10m in corresponding to the remote sensing image?

 

Point 5: Formula (4) Whether "RPD" has been used as the accuracy evaluation index of vegetation or crop estimation, please list the literature.

Methodology

Point 6: Page 5, Whether the sample data in the paper are preprocessed, whether there are outliers and removed. 

Results

Point 7: It is suggested that the author analyze the differences between the estimated and measured biomass of different sample points according to Figure 3

Point 8: Please ask the author to confirm whether there are outliers in Figure 3 (A and C) that need to be removed. There are obviously large differences between the estimated values of some points and the actual values.

Point 9: Please ask the author to confirm whether there are outliers in Figure 7 and 8 that need to be removed. There is obviously a big difference between the estimated values of some points and the actual values. Therefore, it is suggested that the author remove outliers, which will significantly improve the estimation accuracy.

 

Author Response

Thank you for comments concerning our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. Those comments are valuable and very helpful. We have provided a point-by-point response to the comments. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 5 Report

Review

The paper intends to compare machine learning and feature selection methods to estimate maize biomass using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables. The research significance is worth encouraging. However, several issues and concerns in terms of structure, methodology, and meaning expression need to be addressed prior to the recommendation of paper acceptance.  

Major

1.      Why do you construct different machine learning models in three months separately? From the result, although you found that different machine learning methods or radar indices/optical VIs perform differently, as they are ONLY performing well in a specific stagy of the crop growth cycle. I would suggest you develop machine learning methods that are based on the data of the whole growth season.

2.      The structure of 2.3.3 and 2.3.4 can be reorganized. Pearson correlation coefficient should be performance metric that should be moved 2.2.4 together with R2, RMSE etc. however, these contents do not fit the heading of 2.3.4 model calibration and validation.

3.      I’m not clear that what are the three predictors in 2.2.3.

4.      Figure 2 and related text should be appeared as an independent subsection rather than be embraced into 2.3.4.

5.      As the core objective of the paper, feature selection method was not clearly shown in methodology section.

6.      As the author demonstrated in Introduction that Vis are sensitive to LOW biomass (the VIs may be saturated when biomass grows high), but the research was conducted with maize in June to August where maize biomass is no longer LOW. So, you may extend a discussion on this issue which might be the source of error.

 

Minor

1.      Introduction part and other sections related, two important references are recommended to strengthen the study progress and enrich the discussion.    Ghasemloo N et al., 2022, Journal of Geovisualization and Spatial Analysis,6(2): 1-12ï¼›Du P, et al., 2020, Journal of Geovisualization and Spatial Analysis, 4(1): 1-25.

2.      Line 28, where did you measure the biomass? The research site can be added here.

3.      Line 29 vertically-transmit-vertically-receive?

4.      Line 30 Difference of VH an VV-polarization, backscatter here with linear unit or dB?

5.      Lines 33, 37, what do you mean total indices and total VIs? Are you meaning all indices or Vis together?

6.      Line 63 the most extensively-> a most extensively

7.      Lines 73,75 In-text citation format might be incorrect. Same issues appear in the following context.

8.      Line 130 Northeast-> Northeastern

9.      Line 140-145, you may need to show how you got the areal biomass from single plant.

10.    Line 165, you should say Resampling ONLY for those bands with 20 and 60 m resolution.

Author Response

Thank you for comments concerning our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. Those comments are valuable and very helpful. We have provided a point-by-point response to the comments. Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

Your paper has been revised with outstanding research highlights and a clear structure, so I think it has achieved the requirements of "Remote Sensing" for publication.

Author Response

We appreciate the reviewer’s approval of our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. We are extremely grateful to your suggestion. All suggestions are valuable and helpful for revising and improving our paper, as well as the important guiding significance to our research in the future.

Best wishes for you. Hope you everything goes well.

Reviewer 5 Report

The authors have well addressed the reviewer's concerns, no further modification is required, thus acceptance is recommended. 

Author Response

We appreciate the reviewer’s approval of our manuscript entitled “A comprehensive comparison of machine learning and feature selection methods for maize biomass estimation using Sentinel-1 SAR, Sentinel-2 vegetation indices and biophysical variables ”. We are extremely grateful to your suggestion. All suggestions are valuable and helpful for revising and improving our paper, as well as the important guiding significance to our researches in the future.

Best wishes for you. Hope you everything goes well.

Back to TopTop