Deep Learning for Vegetation Health Forecasting: A Case Study in Kenya
Round 1
Reviewer 1 Report
This manuscript proposes a method for vegetation health forecasting in Kenya where a number of droughts has occurred. Reasons for input data and deep learning models selection are sound, and the methods and metrics used for model explanation are SOTA . Experimental results are well presented. In summary, I would like to recommend that this manuscript be published in Remote Sensing. I have only three minor comments.
- The LSTM methods may perform badly for outliers, of which I did not find an appraisal or a comment in this manuscript. It should be better to give a tip to readers.
- Fig. 2, a,b,c were not labeled.
- Fig. 3, the schematic on the right is not clear, at least in the pdf file I read.
Supplementary comments:
1. What is the main question addressed by the research?
The authors seek to test two LSTM architectures for vegetation health forecasting. The LSTMs are not originated by them, but the method and the interpreting they used here is interesting and solid. This is an original work for vegetation health forecasting.
2. Do you consider the topic original or relevant in the field, and if so, why?
The topic is original and relevant in the field. They used remote sensing data for vegetation health forecasting.
3. What does it add to the subject area compared with other published material?
The method used for models interpreting makes it a highly valuable work, especially when compared to those only so-called improved algorithms.
4. What specific improvements could the authors consider regarding the methodology?
I have no idea, currently. They are good.
5. Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed?
Yes.
6. Are the references appropriate?
Yes.
7. Please include any additional comments on the tables and figures.
No further comment. Text, figures and tables in the pdf file are all in the right. It is not easy to print and read.
Author Response
We thank Reviewer 1 for their specific comments and will update the manuscript as follows:
- We will add a comment outlining the behaviour on outliers. For negative outliers, this is what Figure 8 is outlining, since values of 1 refer to VCI scores of less than 10.
We have added the following: “This demonstrates the possibility of using the LSTM to monitor and predict negative outliers, even in those circumstances where the VCI score is less than 10 (see Table \ref{tab:vdi}). We can see that 74\% of the values were correctly classified, and 25\% of values were incorrectly classified on these negative anomalies.”
- We will add the labels to Figure 2.
- We will update Figure 3, since the png appears to be corrupted.
Reviewer 2 Report
This paper presents two LSTM architectures to forecast vegetation for drought monitoring. The methods are appropriately presented, and the results are adequately discussed. In addition, the article is exciting and very well written.
Author Response
We thank Anonymous Reviewer 2 for their time and insights while reading our manuscript. We are pleased that the information was well received and that the community is excited by this research.
Reviewer 3 Report
This paper is very interesting and well written. Figure captions are a little bit long otherwise it can be accepted as it is.
Author Response
We thank Anonymous Reviewer 3 for their time and insights while reading our manuscript. We are pleased that the information was well received and that the community is interested in this research. We understand the reviewers comments, however, we believe that the text is required to be able to read and understand what is being shown in the figure captions.
