Application of Parallel Factor Analysis (PARAFAC) to the Regional Characterisation of Vineyard Blocks Using Remote Sensing Time Series

Round 1

Reviewer 1 Report

Dear authors, 

Please find my comments in the pdf attached.

Best wishes,

AA

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Although there are many technical frameworks used for information mining of spatio-temporal-spectral remote sensing data, such as continuous wavelet decomposition, empirical mode decomposition, and principal component analysis, as the authors said, there are few solutions to multidimensional decomposition. In light of this, the present work is very interesting and important. This work introduces the PARAFAC methodology in the field of chemometrics into the field of remote sensing applications. PARAFAC is an unsupervised decomposition method. The authors tried to correspond its components with the concept of end element in the field of remote sensing and made a lot of assumptions and inferences. However, these assertions seem to lack sufficient data support, and the expert assessment-based approach cannot provide enough evidence to support the core conclusions of the article. Therefore, I cannot recommend proceeding to the next step with the current version. major concerns: 1, The authors inferred that the first two components of PARAFAC correspond to soil and vegetation information, but the current data analysis is not enough to support this assertion. It is strongly recommended to add necessary experiments to prove, for example, (a) a simulation data experiment, such as using endmembers mixing; (b) the retrieval of agronomic parameters, such as grape yield and leaf area index. 2, Although industry experts have given detailed assessments and the authors have also provided detailed discussions, this does not form direct evidence for the above assumptions. 3, Figure 4-5. were these figures corresponding to only one of the 4978 vineyard blocks? if yes, why not show all samples or show their Mean, median, and standard deviation? minor concerns: 1, L53. Unnamed? Aerial Vehicles 2, L93-96. How to understand "the simplest and most restricted"? Besides, this sentence may need to be rewritten because it is confusing. 3, L97-98. what is "pure spectra"? did you mean raw or original spectra?

4, L193-194, 203, Figure 2, Table 1. There are some non-English usages in the text, tables, and figures. Please correct them.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

 

In comparison to many contemporary papers that are trying to design different decision support systems based on the existing knowledge about the environmental conditions and state of the crops in the field, the presented study brings some fresh air into this research area by proposing an exploratory approach for understanding of the underlying environmental factors that are driving certain processes in the agricultural production. Presented results confirm that it has strong potential to be used as a first step in better understanding of the variabilities related to the crop production in some agricultural region, which are usually hindered due to large spatial and temporal extent of the processes. Moreover, it enables certain insights without ground-truth data collection campaigns and could be a very useful tool for design of different cost efficient data sampling strategies. It is based on the assumption that there could be certain hidden variables or factors that could explain the structure of the acquired measurements collected in the corresponding data cubes extending over spatial, spectral and temporal domain. Such unsupervised identification of the certain factors that could be given interpretation by the domain experts is often neglected in many novel remote sensing analyses, and therefore existence of such methods should be better emphasized in the literature. Therefore, I would strongly suggest to consider presented results for publication and provide the readers with opportunity to see the results of such exploratory methods in practice.

Paper is well written and organized. I would suggest to maybe further discuss what would be some alternative methods for tensor decomposition that could be applied in similar fashion, and what you would see as the main obstacle(s) for wider adoption of this type of process identification approaches, besides the necessity to have certain domain knowledge or information about the area characteristics that could ease the interpretation of the score maps and underlying factors by the experts. Also, in the current study there were only two significant factors, which served to deconstruct the measurements into information revealing maps, in that sense, please try to comment on how many factors should be expected in different applications and what would be the best strategies for determining number of factors and lowering their mutual overlapping (in ideal case we would like to have no overlapping influences and to be able to have crisp picture of the main influences that could be controlled by the field management or some adopted agricultural practice). Such discussion would further facilitate adoption of this kind of exploratory analyses.

 

Nice work, good luck.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Dear authors,

Happy with the changes. 

Best wishes,

AA

Reviewer 2 Report

I think the authors are negative in revising the present paper. I haven't seen enough improvement in the revised version, and I still have to reject it.