Mapping Forest Types in China with 10 m Resolution Based on Spectral–Spatial–Temporal Features

Round 1

Reviewer 1 Report

This paper was very well written and the work is very sound. I am impressed with the descriptions of the methods and the results. This work represents an enormous amount of data and you have done an excellent job of not only interpreting the data but also translating the results for the reader. In the attached, I have a few very minor technical edits for you to consider. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The novelty/originality is not justified to highlight that the manuscript contains sufficient contributions to the new body of knowledge. The knowledge gap needs to be further addressed.

1. The introduction is very weak.
2. The research results are very poor. Findings are duplicates. What is the innovation of this research?
3. Writing is weak.
4. There is really no new result.
5. This manuscript writing style is not enough for considering as a academic writing. But the biggest problem is less novelty.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper explores the use of spectral, spatial, and temporal (SST) features, supported by earth observation big data, for large-scale, for-forest-type discrimination. The results provide a general and reliable forest-type classification method, and reveal relative SST feature contributions to different geographical areas. The paper is supported by the remote sensors data and current technology available in the literature. It is described in a clear manner that makes it easy to understand.

Figure 4a – Notheast (NE)  ---  Northeast

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

This manuscript addresses very important topic of forest type mapping at a large scale using middle and high resolution satellite data from Sentinel-2 and Landsat imagery. Additionally, the study covers different types of landscape covering various forest types and climate regions. This is an important aspect of the presented study. The combination of spectral, spatial and temporal features seems to be a good approach that help solving different issues related to forest type mapping. However, not all aspects of this research seems to be clear and they needs to be clarified before it could be published. They are either some mistakes in the research work or explanations are not clear enough. The most important issues are pointed out below.

 

More specific issues to the manuscript are provided below:

- please re-read carefully the manuscript and correct small but numerous typos.

- lines 46-62 – the authors write about some examples of forest mapping but there is no clear conclusions coming from this examples. A clear research gap needs to be pointed out. The only sentence (lines 59-62, which is by the way difficult to read and needs to be rewritten more clearly) addressing the literature states that the provided examples describes products derived from spectral features. However, the provided examples shows that both spectral, temporal and spatial features were used. Therefore, there is contradiction in the author statement and the provided examples do not confirm the author’s conclusions. Moreover, mapping forest and forest types is a very broad topic in the field of remote sensing and should be presented here more deeply. Forest was one of the first land cover type mapped globally!

- lines 74-76 – this sentence also need to be rewritten so that it can be read easily.

- Figure 1 – please increase the size of symbols in the legend of the main map. They are very small

- line 110 – the first satellite of Sentinel-2 was launched in 2015!!! Correct in the whole manuscript

- line 110 – author mentioned availability of data from Sentinel-2 covering period 2013(incorrect)- 2019 but in line 167 a period covering years 2017-2019 is mentioned. Which one is correct?

- lines 110-112 – what the authors mean by this sentence? Sentinel-2 data organised into composite including temporal features and red-edge bands? All S2 images are with red-edge bands! And what about the temporal features? Are there ready to use features available?

- line 113 – what does it mean ‘between the growing and deciduous seasons’? I think if the growing season is combined with leaf-off periods, it covers the whole year? Is not it right? So, which data was finally used? And, is the term ‘deciduous season’ correct? It is the first time I see it. I would rather use the leaf-off periods expression. Does it mean something different?

- line 117 – why such a long period was used (1985-2019) if the year 2018 was analysed?

- section 2.2.2. – do all the reference sources provide the same kind of information? Are the second and third one up-to-date? The different data sets should be probably listed by bullets.

- section 2.3 – use bullets or points to list the methodological steps

- Figure 2 – the image composites were prepared separately for the growing and leaf-of seasons?

- section 2.3.1. is very confusing, especially points 1 and 2. This should be explained more clearly.

- line 162 – again, why so long period?

- lines 167-169 – why data from three years were used? Why not only 2018?

- 178-179 – what kind of quality assessment was performed? Even if some details are provided in ref. 23, the authors should provide some insight into this method

- line 197 – 199 – these features are usually called textural features instead of spatial features.

- line 190 – what was the reason to not use all S2 bands?

- line 200-206 – so finally which kind of image data was classified? Which year was analysed? 2018? 2017-2019? Or 1985 – 2019? Which image composites were classified? Annual? Monthly? Bi-annual (growing season and leaf-off season)?

- Table 2 – Spectral features -  one or more NIR bands?; spatial features – S2 annual composition? Is it another product?

- lines 227-239 – why the features were compared pair-wise? It is not much useful as all type of forests were mapped together.

In the RF classification first feature selection was performed and this allowed to select the most important features. I assume that in this case all features were used together to get the most accurate forest type map. So why in the next step features were compared in pairs? This process of course may indicate some more or less important features but they are used all together anyway in the final classification. Therefore, analysis which type of features are more useful is a bit less important. The most important is to indicate which one are used in the most accurate classification. This study shows, however, that there are no dominant features. Each classification, in every region used different set of ‘optimal’ feature sets so no clear recommendations can be indicated. This has to be explained clearly in the discussion.

- line 243 – 244 – where the number of execution of RF (20) comes from?

- line 245, 253, 266, 274 – which type of accuracy is it??

- Figure 3 – where it the list of features as indicated in the caption?

- line 299 - what is called inland areas? all forested areas are inland

- Figure 4 caption – again, is it for sure 2018?

- section 3.3. – are accuracy measures presented here different than those from section 3.1? why?

- line 403 – were the compositions made of Landsat or S2 too?

- line 418 – ‘These pixels were included in the NIR image composite, as 418 it showed the least noise over forested areas’ – what is this? Does the image composite is composed only of NIR band?

- line 423 -  what are it the reference images? Where they mentioned before?

- Figure 7 – the caption is not self-explanatory, and graphs should have labels.

- section 4.3 – I cannot understand which features type was found to be more important: temporal or spectral. The first two paragraphs of this section are opposite.

- line 518 – map for 2018? It has been already mentioned that this date is confusing; please write resolution (10 m) without the hyphen as in the previous sections.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Manuscript is improved. But introduction section still is poor and it should be improved.

Please provide more literature review. And improve introduction in term of quality and quantity.

Author Response

Please see the attachment.

Author Response File: Author Response.docx