Combining Color Indices and Textures of UAV-Based Digital Imagery for Rice LAI Estimation

Round 1

Reviewer 1 Report

The content of the paper is good I think. However there are some improvements that could be made to the paper. Firstly, the English needs some work with the help of a native speaker and proof-reader. The detail is good. However, more information could be provided about the site and the crop with a more detailed map and also some ground photographs as well as some more information about the UAV used and the sensors - some photographs would be helpful. More could also be said about the platform and perhaps some more context at the end concerning the viabilityof such low-cost systems in relation to the commercial applications of the hardware and apps now becoming available e,g, DroneAg's Skippy Scout App. This would help to contextualise the research and the application of this technology in a practical sense for the farmer etc.  Some pictures of the imagery aslo would be useful. I am a little curious also as to why the imagery is quite so old and how this relates to the current platforms and technology that have advanced considerably since then. Maybe some comment on this should be included.

Author Response

Response to Review Comments

Detailed Responses to Review Comments:

n  The content of the paper is good I think. However, there are some improvements that could be made to the paper. Firstly, the English needs some work with the help of a native speaker and proof-reader.

Response:

Thank you for your comments. The revised manuscript has been proofread by a native speaker. 

n  The detail is good. However, more information could be provided about the site and the crop with a more detailed map and also some ground photographs as well as some more information about the UAV used and the sensors - some photographs would be helpful. More could also be said about the platform and perhaps some more context at the end concerning the viability of such low-cost systems in relation to the commercial applications of the hardware and apps now becoming available e,g, DroneAg's Skippy Scout App. This would help to contextualise the research and the application of this technology in a practical sense for the farmer etc. 

Response:

More information about the locations, treatments, sensing device, and ground photographs has been added in the section of Materials and Methods (mainly in Figure 1 and Figure 2).

Certainly, as you mentioned, the rapid development of hardware and software of consumer-grade UAVs has partially reshaped the industry of drone-based precision agriculture. Accordingly, we have added more discussion in section 4.3 on new technical developments in this revised manuscript.

n  Some pictures of the imagery also would be useful. I am a little curious also as to why the imagery is quite so old and how this relates to the current platforms and technology that have advanced considerably since then. Maybe some comment on this should be included.

Response:

The images were collected by a widely used UAV platform in two experiments conducted during 2016 and 2017. As our objectives of this study were to detect the predictive capabilities of color indices and texture features, as well as to evaluate various regression methods for better LAI estimation, rather than real-time rice LAI monitoring, we feel it is fine to use these images for our purposes even though they are a couple of years old. 

Reviewer 2 Report

The article entitled „Combining color indices and textures of UAV-based 2 digital imagery for rice LAI estimation” describe a method to LAI estimation using color indices and texture obtained from RGB images taken from UAV. The developement of alternative methods for the robust and inexpensive evaluation of one of the most important plant parameter is undoubtedly very important for precision agriculture. This method can be useful, for example, in determining the nutritional needs of plants and the development of fertilizer plans for agricultural crops.

The article is well constructed (easy to follow), and analysis was well performed with the most advanced statistical methods. The only remark concerns Fig. 11, which in my opinion is unnecessary because it duplicate data Thus I reccommend to publish this article in the present form if the language requirements are met.

Author Response

Response to Review Comments

Detailed Responses to Review Comments:

n  The article entitled „Combining color indices and textures of UAV-based 2 digital imagery for rice LAI estimation” describe a method to LAI estimation using color indices and texture obtained from RGB images taken from UAV. The development of alternative methods for the robust and inexpensive evaluation of one of the most important plant parameter is undoubtedly very important for precision agriculture. This method can be useful, for example, in determining the nutritional needs of plants and the development of fertilizer plans for agricultural crops.

The article is well constructed (easy to follow), and analysis was well performed with the most advanced statistical methods. The only remark concerns Fig. 11, which in my opinion is unnecessary because it duplicate data Thus I recommend to publish this article in the present form if the language requirements are met.

Response:

Thank you for your comments. Although Figure 12 (in the revised manuscript) includes some information provided in Table 7 and Figure 11, it mainly highlights the performance of the best LR models and multivariate regression models. Moreover, it directly shows comparisons between the index-based and multivariate regression models, and among different input treatments, which facilitates the model evaluation and discussion in later sections. Therefore, we tend to retain this figure in the manuscript for a better reading experience.

Reviewer 3 Report

In their study the authors present different methods to estimate leaf area index of rice with the help of color indices and texture features and their combination in UAV-based high spatial-resolution RGB images. The focus was on a low cost system with commercially available RGB cameras. The methods were described very accurately so that it is simple for the reader to comprehend it. The paper is well written. All needed issues are discussed in the introduction and the methods section. All results are well described.

L301 : For texture-based models your focus is only across all growth stages. But each single growth stage has a rather good performance you described in the text (see L308). Why is the estimation across all stages so bad? It seems like the texture-based models estimate the variability within each growth stage not bad but the increasing absolute values of LAI across growth stages is not reflect. It could improve the models if you involve the growth stage as a factor in the models.

L306 : “..varied significantly…” How have you tested the significance?

Figure4 : The both green colored bars (PI and BT) are difficult to distinguish.  Another color would improve the figure.

Table 5 – I guess the results are across all four stages. It should be an additional information in the heading of table 5

Table 7 – I think this are results of the test dataset and not the training dataset. Is the heading description right?

Author Response

Response to Review Comments

Detailed Responses to Review Comments:

n  In their study the authors present different methods to estimate leaf area index of rice with the help of color indices and texture features and their combination in UAV-based high spatial-resolution RGB images. The focus was on a low cost system with commercially available RGB cameras. The methods were described very accurately so that it is simple for the reader to comprehend it. The paper is well written. All needed issues are discussed in the introduction and the methods section. All results are well described.

Response:

Thank you for your comments.

n  L301: For texture-based models your focus is only across all growth stages. But each single growth stage has a rather good performance you described in the text (see L308). Why is the estimation across all stages so bad? It seems like the texture-based models estimate the variability within each growth stage not bad but the increasing absolute values of LAI across growth stages is not reflect. It could improve the models if you involve the growth stage as a factor in the models.

Response:

We acknowledge your comments. As revealed in Figure 6, most NDTI performed well for LAI estimation in the single stage but showed poor associations across all stages. This may demonstrate that the increasing absolute values of LAI could not be estimated well by most NDTIs based on LR models, and the factor of growth stages should be considered.

However, the judgment of growth stages is partly subjective and rough, which may not be used as an accurate variable for quantitative estimation. In addition, different climatic factors such as temperature may lead to diverse growth process that affects the stage selection and practical estimation of single-stage based LR models.

In this study, we put forward a promising and robust predictive method for LAI estimation across various growth stages. However, more quantifiable variables such as accumulated degree days (including temperature information) and sunshine duration should be analyzed to indicate the growth process, and to be utilized for predictive modeling in future studies. Discussion of this part has been added in section 4.3.

n  L306: “..varied significantly…” How have you tested the significance?

Response:

A significance test was not conducted. Therefore, “significantly” has been revised to “notably” to avoid potential misunderstanding.

n  Figure 4: The both green colored bars (PI and BT) are difficult to distinguish.  Another color would improve the figure.

Response:

The color schemes in Figures 4 and 5 have been revised in the new manuscript.

n  Table 5 – I guess the results are across all four stages. It should be an additional information in the heading of table 5

Response:

Accepted. “across all four stages” has been added to the headings of Tables 5 and 6.

n  Table 7 – I think this are results of the test dataset and not the training dataset. Is the heading description right?

Response:

Thanks for pointing this out. The heading of Table 7 has been corrected.