Species-Level Classification and Mapping of a Mangrove Forest Using Random Forest—Utilisation of AVIRIS-NG and Sentinel Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

This article used AVIRIS-NG and Sentinel data to classify mangrove forests at species-level. The authors found that the accuracy of species-level classification result derived from Sentinel-2 data with broad bands is similar to that extracted from AVIRIS-NG with narrow bands, indicating that Sentinel-2 data have potential to apply in mangrove classification at species-level. Additionally, the authors also found that the Sentinel-1 data have little contribution on classify mangroves at species-level. Overall, this work is on a topic of significant interest and falls with the scope of the journal. However, the article is not well-organized. The writing and presentation should be further improved.

 

Abstract:

1. Page1, line 15-17: I think that the first sentence of abstract should emphasize the importance and main contribution of this article. I would recommend revising this sentence.
2. Page 1, line 22-24: I think the meaning of this sentence is that Sentinel-2 data have potential to discriminate various mangrove species. I would recommend improving this sentence to present more clearly.

Introduction:

This section found there is lack of the research on using multi-sensor remotely sensed datasets to map species-level mangroves through reviewing previous studies. Therefore, this article focused on this topic. However, the topic in this section differs from that shown in results. I would recommend improving this section further.

 

Materials and Methods

1. Accuracy assessment is very important. I would recommend introducing this part in details in one subsection.
2. In the subsections 2.2 and 2.3, the acquisition time and corresponding tidal height of satellite, airborne and field data should be added. The spectral features of multispectral data is highly influenced by tide. The results derived from data with different tidal height are not comparable.

 

Results

1. There are two maps derived by from AVIRIS-NG and Sentinel-2 data separately. In section 3.1, you should show that the accuracy assessment is conducted on which map.
2. Page 7-8, Line 242-254: this paragraph should be placed in the section 2.
3. In the section 3.2, you should show more details about the species-level mangrove map in figures.

Author Response

Thank you for your valuable comments and suggestions.

Please find the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The submission by Dev Behera is well written and was easy to follow and understand thanks to clearly defined objectives and quality presentation of results and figures. The application of a random forest model along with variable importance metrics fit this data set and problem definition very well. I recommend this paper for publication with only minor suggestions as provided below.

 

Minor changes:

p.2, line 48 Change to something like, “…data are commonly used to distinguish….”

 

General comments:

Figure 3 is an excellent figure as is shows similar patterns predicted from models using two different data sets. We can also use this figure to get a sense for where uncertainties might exist, and which regions represent a challenge for identifying the dominant species.

Figure 4 (p.9) Displaying variable importance from an RF model is also an excellent idea, and Figure 4 is a great start toward the goal of understanding variable importance. While this figure is sufficient for getting the main idea across to readers, an R package such as Boruta (https://cran.r-project.org/web/packages/Boruta/Boruta.pdf) provides more detail and insight into variable performance using random forests and resampling techniques. Boruta also provides insight into which variables are not important and such an insight could help other researchers know which variables they might want to eliminate from their predictive models. I am not suggesting that Boruta must be used in your paper. However, I do suggest using it to validate and better understand and provide additional context into the results presented in Figure 4.

With regard to Table 3 (p. 7), this table does a good job of giving the reader context into the challenges of predicting the dominant species, particularly when the dominant species is only marginally the majority (say 50% to 70%). Could you include a table (or text) to indicate model performance as a function of percentage dominance (perhaps for the different coarse bands of dominance percentage)? I would conjecture that the RF model performs with a higher accuracy when species dominance is higher.

Author Response

Thank you for your valuable comments and suggestions.

Please find the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

After revision, the logic and writing of this paper still need substantial improvements.

 

Major comments:

 

Abstract:

This section was not well organized. The limitation of previous studies and the unique contribution of this paper was not well summarized and emphasized. You can reference other published articles.

 

Materials and Methods

In section 2.5, you should introduce the process of training sample selection in details.

 

Results

In section 3.2, the comparison of species-level classification accuracies derived from AVIRIS-NG, Sentinel-2, Sentinel 1 & 2 can be presented in a table.

 

you should show details about the species-level mangrove map in more figures.

 

Author Response

Thanks for your valuable suggestions.

Please see the attachment.

Author Response File: Author Response.docx