A New Algorithm Using Support Vector Machines to Detect and Monitor Bloom-Forming Pseudo-nitzschia from OLCI Data
Round 1
Reviewer 1 Report (Previous Reviewer 1)
Comments and Suggestions for Authors1. (Line 176) So every sample collection data corresponds to a date with qualified Sentinel-3 measurements? e.g., no missing date for satellite data? all satellite data have passed quality control?
2. (Line 250) What are the optimal values found in the first phase?
3. (Section 4) Based on your title, Section 4.1 and 4.2 should not be main results of this manuscript. Please update your manuscript title, or move the major contents of these two sections into other sections (or even appendix).
4. (Line 413) What is the r value for your proposed nonlinear model?
5. (Line 503) Where is Table 6?
6. (Figure 6(a)) To present the predicted ability of the model, it is necessary to provide a plot to show the errors between these two-time series.
Comments on the Quality of English LanguageNone
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report (Previous Reviewer 2)
Comments and Suggestions for AuthorsDear authors,
from my side, the paper can be published, but please check the abbreviation for chlorophyll a, generally the abbreviation is written with the capital letter "C" and with "a" in italics.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report (Previous Reviewer 3)
Comments and Suggestions for AuthorsThe revised version allows supplementary details.
We propose to accept the paper.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report (Previous Reviewer 4)
Comments and Suggestions for AuthorsIt has been presented in the manuscript that a SVM method was used to develop a remote sensing model to monitor Pseudo-nitzschia spp. blooms based on Sentinel-3 OLCI images, which provided complementary information to in-situ monitoring program. Validation results showed a good accuracy of the model of PONI. Although the manuscript holds a merit to be published, some efforts are needed to dig deeper as I have some difficulty in understanding the manuscript in current content. The main problem is that the structure of the article needs to be adjusted. I will address myself in detail as follows.
1. How was the detection threshold determined? What is the theoretical basis?
2. Is the data scaling method selected based on the characteristics of the data? How is it different from log transformation?
3. Have band combinations been considered as input parameters for machine learning models? Many studies have pointed out that band combinations reflect the better spectral characteristics of algae.
4. There are too many descriptions of methods in the results section, which can be stated in the methods or discussion section. The results section only shows the experimental results.
5. Why the author wants to compare the linear model with the SVM model instead of other machine learning methods?
6. All the content in Section 4.3 can be put into the discussion section.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report (Previous Reviewer 1)
Comments and Suggestions for AuthorsVia Fig. S4, the predictive ability of your model is not as good as expected.
Comments on the Quality of English Languageneed minor polishing.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1. How to match Sentinel-3 data with INTECMAR data in terms of space and time.
2. (Line 259-265) The procedure of this approach is not clear enough. Please provide an algorithm and figure/table results to show how to achieve both optimal values.
3. (Section 4) Half of this section is used for analyzing the original data but only half is used for the model and results. Why is the first half so important?
4. (Line 438-439) r=0.42 is a very bad value between observations and predictions.
5. (Table 5) For Bloom detection, most "Prec." values are low.
6. (Fig. 4) Need a figure with higher resolution. For all scatter plots, their correlations are all obvious weak.
7. (Line 542) Where is "Table 6"?
8. (Fig. 6(a)) Need additional new plot to show the errors between these two time series.
Comments on the Quality of English Language
Quality is high.
Author Response
see attachment please
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Authors,
I found the MS very interesting and well-written. However, I have some minor comments and suggestions to improve the MS to make it fully suitable for publication.
Introduction
L50 Please write the "a" in the word Chl a “ in italics.
L70 Please write the author's name and then the number of references as the sentence does not make sense like this.
L101 Again, the same as in L70.
L173 I only want to check whether it is 25x or 250x, 40x or 400x magnification?
Methods
I also want to check if the graphs are done by another program or by the models and software mentioned in MS, especially for Figure 4?
L219 Check your references, I think it should be 36 first, and then the higher number 47.
Results
Figure 4: Please increase the visibility of the graphical representation and also the legend text of the x and y axes.
Discussion
L498 has the same comment as L70.
The discussion goes more in the direction of describing your results than in the discussion part, can you also give more references in the discussion part, comparability with some studies that are comparable to yours?
Author Response
see attachment please
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper addresses a machine learning method to identify harmful algal blooms from satellite imagery over the Galician coast. The paper is very well written and pleasant to follow. The research presented follows the best standard. We recommend accepting the paper. Here follows several comments.
The references could be completed by some monitoring actions such as eurohab.com
Many plumes over the sea surface could be present: turbidity, run-off, river plume and theses ones could be in conjunction with HAB on a satellite image.
The prediction of the conditions favouring bloom start could be more detailed, lines 603-606 are useful for this point.
“prevalence” could be detailed, line 276.
The method and the parameters are well described.
“Seven stations” is “six stations” line 302.
Check 2017 line 331.
In figure 2, one year is different from the others, and it is not detailed in lines 342-349.
104 is 10^4 line 388.
End of table 5 is unclear because several lines are present under the linear case and the 5 sign + are not detailed. The lines 487-492 are difficult to link with table 5.
The dates and the captions FINAL and INITIAL in fig 5 are unclear.
“that” may be “than” line 630.
Check years 2016 and 2018, caption fig 7.
“,.” line 850, years in bold policy lines 859, 864.
Author Response
see attachment please
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsIt has been presented in the manuscript that a SVM method was used to develop a remote sensing model to monitor Pseudo-nitzschia spp. Blooms based on Sentinel-3 OLCI images, which provided complementary information to in-situ monitoring program. Validation results showed a good accuracy of the model of PONI. Although the manuscript holds a merit to be published, many efforts are needed to dig deeper as I have some difficulty in understanding the manuscript in current content. The main problem is that the method part lacked lacks sufficient explanation and some discussions were so lengthy, which made me confused for the improved part of result and discussion. I will address myself in detail as follows.
1. Part of Line 342-371 is recommended to be stated in Section 3.
2. Why were only the effects of the single band considered in sections 4.2 and 4.3?
3. It is recommended to use satellite images to show the characteristics of different types of samples. The current content method is not intuitive.
4. I don’t understand why the linear model and the SVM model are used to compare? Why not compare with other machine learning models or GLM models?
5. Image quality needs to be improved
6. The results showed that the distributions of different types had an impact on the accuracy of the model. Is it possible to discuss the impact on model accuracy due to the uneven distribution of algal bloom samples and non-algal bloom samples?
7. Section 4.4 should be stated in the discussion section.
8. It is recommended to simplify the description in section 5.3
9. Part 5.4 is recommended to be stated in the results section.
Author Response
see attachment please
Author Response File: Author Response.pdf