Biophysical Variable Retrieval of Silage Maize with Gaussian Process Regression and Hyperparameter Optimization Algorithms

Round 1

Reviewer 1 Report

The present study proposes a kernel-based machine learning algorithm capable of performing adaptive and nonlinear data fitting so as to generate a suitable, accurate, and robust algorithm for spatio-temporal estimation of leaf area index, fractional vegetation cover, and biomass. factors using Sentinel-2 images.

The authors propose an interesting methodology and observations trying to robust the estimation and manage the calculation time. However, the experimental part of the manuscript is not comprehensive enough to highlight the superiority of the proposed method. It is hoped that the authors will make further improvements to the manuscript. Here are my specific comments：

 

1. Eq (2) using “Math type” or “Equation” of word office to generate a proper form of fractional expression instead of using “/”
2. Ln:215, please use Equation 2 instead of “Eq. 2” in the article. Also, fix “Eq3” in ln:318 accordingly.
3. Line up the Eq 1, 2, and 3 numbers.  
4. The numbering of figure captions is in “Figure 1.” format. The reference of figures in the context should be consistent, ie.  using “Figure 1.” instead of “Fig.1.”. Please revise the article.
5. Please specify the definition of “length-scale”. It seems to relate to the spectrum wavelength, but it’s not well defined in the article.
6. Fig 7 title, “wavelength” instead of “wevelength”.
7. Fig 8, please specify the unit of the intensity color bars in the context.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper proposed an optimized GPR model to estimate LAI, FCOVER, and biomass with S2 data. Four optimizing algorithms were performed and the best one was selected according the validation with field data. The algorithm was also compared with ANN and RF to demonstrated the robustness. The paper is well organized. The introduction covers main research in this field and gives the required details and information. The result and conclusion are clear and consistent. This work is of certain significance in improving LAI, FCOVER, and biomass estimation with remote sensing data.

Only some minor comments were demonstrated:

1.      The paper tried four optimizing strategies to improve the GPR algorithm, and compared them with ANN and RF. But what and how SVR, KRR, RVM, GPR, and VH-GPR used were not clearly written. In 363-374, the runtime of SVR, KPP etc. were demonstrated, but the reason why they were demonstrated was confused.

2.      Too many algorithms were referred in this paper. Their roles in the paper were not well introduced. The authors should clarify it in the instruction and method section.

3.      The Fig. 5 flowchart is quite blurry and some unnecessary red wavy lines are drawn below some words.

4.      Line 383, the meaning of length-scale and its formula should be added.

5.      Line 114, leads to -> lead to.

The English language is mostly proper. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This paper presents a Gaussian Processes Regression and Hyperparameters Optimization Algorithm, which was shown to outperform kernel-based machine learning regression algorithms. The innovation is attractive. The paper is well-organized, and the experiments are sufficient. I think the paper can be recommended for publication with following minor revisions:

 

 

1. In Abstract, the first sentence ‘Among the key factors across hydrological, agricultural, and irrigation management studies include quantification of vegetation biophysical variables such as leaf area index (LAI), fractional 19 vegetation cover (fCover), and biomass.’ is faulty wording or formulation.

 

2. In Introduction, authors should add developments on classical model-based methods, such as the classical least squares or sparse regression methods, such as

 

[1] Jian Li and P. Stoica, “An adaptive filtering approach to spectral estimation and sar imaging,” IEEE Transactions on Signal Processing, vol. 44, no. 6, pp. 1469– 1484, June 1996.

 

[2] Zhang Y ,  Luo J ,  Li J , et al. Fast Inverse-Scattering Reconstruction for Airborne High-Squint Radar Imagery Based on Doppler Centroid Compensation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, PP(99):1-17.

 

Meanwhile, the advantage and disadvantage should be stated.

 

3. The resolution of figures is quite poor.

4. In equation (2), symbols will be better.

5. When comparing RMSE, the Cramer-Rao bound should be compared.

6. The format and captions of subfigures should be improved. There are also improper usage of subfigure number.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

The authors failed to provide what is the conventional method for predicting vegetation biophysical variables in order to justify the algorithms in the manuscript. 

The pictures in Fig. 1d, 2, and 3 in the manuscript are absolutely not necessary. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 5 Report

 Authors proposes a kernel-based machine learning algorithm capable of performing adaptive and nonlinear data fitting to generate a suitable, accurate, and robust algorithm for spatio-temporal estimation of the three mentioned variables using Sentinel-2 images. Gaussian process regression (GPR) - particle swarm optimization (PSO), GPR - genetic algorithm (GA), GPR - tabu search (TS), and GPR - simulated annealing (SA) hyperparameters optimized algorithms were developed and compared against kernel-based machine learning regression algorithms, artificial neural network (ANN), and random forest (RF) algorithms. Experiments show that GPR-PSO algorithm outperformed other algorithms under study in terms of robustness, and accuracy.

The quality of presentation and scientific soundness are both high. I recommend this paper to be accepted in this renowned journal.

no

Author Response

Please see the attachment.

thank you very much. We improved the manuscript. 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thank you for the improvement. The article is qualified to be published in my opinion.

Reviewer 4 Report

The original data is not clear.

Some present tenses should be in past tenses.