Round 1

Reviewer 1 Report

Please check the English grammar and spelling. Otherwise I think that the authors have revised the original version of the manuscript according to the suggestions I pointed out last time.

Author Response

Dear Reviewer,

Thank you very much for reviewing our manuscript and final positive evaluation of our work.

Reviewer 2 Report

Review comments: Manuscript ID:- remotesensing-1190211

The manuscript (ID: remotesensing-1190211) entitled with “ Spatiotemporal Modeling of Coniferous Forests Dynamics  Along The Southern Edge of Their Range In The Central Russian Plain” has gone through a better  revision as compared to the earlier version (submitted with manuscript ID: remotesensing-1131805). Major issues from my side were already taken into account. However, all major issues from my side were not taken into account.

From my previous comments and suggestions, the following not yet fully addressed:

1. “ L12-29:abstract to be in the order of brief introduction, objectives, methods, findings and conclusion, which is not the case in the current version.” Though the authors reposed as they already made a “complete redone,” it still needs to consider my suggestion seriously. The major part is about introduction and methodological description, see L12-28.
2. “L235-236: Include explanation for the choice of the reference years (2006-2019). Including the choice of the reference years are important for checking the relevance of the current study.
3. (L237-238: “…..Field data were collected using standard methodology [73] in forest sample plots of 400–625 m2 area….,”, for which year dataset/reference year the field data set collected? How the field data for the historical one collected?). This is also not addressed at all, see L241-42.

 

4. (L324-325: “…Classification quality was assessed using the confusion matrix and the overall accuracy…,” how testing samples collected? Was it collected together with training samples? If so, which proportion used between training and testing samples?) This is not also addressed fully, see L400-401. Was the testing sample collected together with training samples?
5. (L504: Table 2, there is a big deviation between Kappa and Overall accuracy in all different models. Why? Such kind of deviation shows problem related to the accuracy assessment strategies. In addition, the accuracy values are too small. How it is possible to trust the model outcomes with lower values, and to use them as input for any other analysis? I suggest the authors to check these again.). This is also partially addressed. The provided reference about the use of low Kappa value (0.47) is too old (1997). I suggest referring the recent one. I also suggest the authors to comment about the deviation between Kappa and Overall accuracy. I think it arises due to their accuracy strategy. I suggest running them again.

L711-745: Conclusion is more of summary. I suggest including major conclusions from this study.

Author Response

Dear Reviewer,

Thanks a lot for reviewing our manuscript and evaluating our work. We have tried to answer all your questions and hope that our comments will be sufficient.

“ L12-29:abstract to be in the order of brief introduction, objectives, methods, findings and conclusion, which is not the case in the current version.” Though the authors reposed as they already made a “complete redone,” it still needs to consider my suggestion seriously. The major part is about introduction and methodological description, see L12-28. Response 1: Thank you very much, we corrected the annotation again

2. “L235-236: Include explanation for the choice of the reference years (2006-2019). Including the choice of the reference years are important for checking the relevance of the current study.  Response 2: Field data of the last 14 years were used in the study. Earlier relevés were excluded because they did not reflect the current state of the ontogenetic phase of tree species in populations. Narrowing the data set collected period to 5 or 10 years would reduce the number of relevés and the representativeness of the resalts. Significant changes in the tree layer, according to our and literature data (Abaturov, AV, Melankholin, PN Estestvennaya dinamika lesa na postoyannykh probnykh ploshchadyakh v Podmoskov'e [Natural dynamics of the forest on the permanent sample plots in the Moscow region]. Tula, Grif i K., 2005. 336 p.; Korotkov, V.N.; Logofet, D.O.; Loreau, M. Succession in Mixed Boreal Forest of Russia: Markov Models and Non-Markov Effects. Ecological Modelling 2001, 142, 25–38), observed for wood cover in the hemiboreal zone for 20-40 years.

3. (L237-238: “…..Field data were collected using standard methodology [73] in forest sample plots of 400–625 m2 area….,”, for which year dataset/reference year the field data set collected? How the field data for the historical one collected?). This is also not addressed at all, see L241-42. Response 3: The answer to the question is close to explanation for the choice of the reference years. Historical assessment of the spatial distribution of forests was carried out on the basis of remote sensing data, collecting the field data for the historical period of years was not possible.

4. (L324-325: “…Classification quality was assessed using the confusion matrix and the overall accuracy…,” how testing samples collected? Was it collected together with training samples? If so, which proportion used between training and testing samples?) This is not also addressed fully, see L400-401. Was the testing sample collected together with training samples?  Response 4: Classification quality is assessed using the confusion matrix and the overall accuracy [89]. The following strategy was used to select test samples. We selected test sample from the initial database of relevés and we did not use them for modeling only for testing. Since the amount of relevés is limited and non-equal between different formations we tried to make the proportion of test sample about 10-15% from total quantity of relevés in each formation. As a result, we obtained test sample proportions 8-19% and amounts 10-100 relevés (Table 1).

5. (L504: Table 2, there is a big deviation between Kappa and Overall accuracy in all different models. Why? Such kind of deviation shows problem related to the accuracy assessment strategies. In addition, the accuracy values are too small. How it is possible to trust the model outcomes with lower values, and to use them as input for any other analysis? I suggest the authors to check these again.). This is also partially addressed. The provided reference about the use of low Kappa value (0.47) is too old (1997). I suggest referring the recent one. I also suggest the authors to comment about the deviation between Kappa and Overall accuracy. I think it arises due to their accuracy strategy. I suggest running them again. Response 5: The deviation between Kappa and Overall accuracy is caused by two factors which influence Kappa sensitivity: i) different size of samples between different formations, and ii) non-equal level of agreement for different formations. Both factors are especially typical for pine-spruce formation. This formation is very rare, it occupies about 3% of the forest area and it has the smallest amount of relevés and, consequently, small percentage of test sample. The way to improve the deviation is to increase the amount of relevés in pine-spruce formation. We did not find relevant recent studies which classify the Kappa levels. The more recent studies are medical ones which looks not suitable for Remote Sensing. Moreover, all of them cite the Landis and Koch study.

L711-745: Conclusion is more of summary. I suggest including major conclusions from this study. Response 6: Thank you very much. Text updated

Reviewer 3 Report

The authors made a lot of affords to improve the quality of the manuscript. However, I have still some comments on the current version of the paper.

ln16 – check punctuation

ln20, ln22-23, ln30 – you used construction “made it possible” three times in the short text of the abstract – please revise

ln134 – “dynamic model ... of forest dynamics” sounds a bit awkward

section 2.2 – I still do not feel like the information on the historical background of forest formation is worth inclusion in this paper. I agree that historical aspects predefined the current state of the forest structure in the region, however, your work must be focused more on remote sensing. The periods you described here start from the 10^th century, your study covers the period since 1980^th.

ln321 – use Google Earth Engine instead of “earth engine”: your repository is not publicly available, please share it

ln375 – your explanation of Random Forest classifier in this sentence is very simple.  Given the next paragraph, you can remove this sentence from the manuscript. In addition, you should cite Breiman L. (2001)

section 2.3.3 – why you did not provide an explanation of the KNN algorithm?

section 2.3.4 – please, explain how you got updated for different periods GFC forest masks

ln404 – check punctuations

Table 2 – without a description of the KNN algorithm I cannot understand how you used 32 nearest neighbors to predict categorical variables (classes). Generally, k-NN (see for example McRoberts & Tomppo, 2008) is an imputation method that is extensively used in forest inventory. Maybe, you reported here another approach, please explain

You made great work and I believe that is a very significant contribution to understanding trends of forest dynamics of your study area. Nevertheless, you focused too much on field data and geobotanical issues which is reducing the significance of your study for science and application of remote sensing technology.

 

Author Response

Dear Reviewer,

Thanks a lot for reviewing our manuscript and appreciating our work. We have tried to correct the text in accordance with all your comments with the exception of section 2.2

Please find my detailed replies to your comments appended below.

ln16 – check punctuation	Thanks. Text updated
ln20, ln22-23, ln30 – you used construction “made it possible” three times in the short text of the abstract – please revise	Thanks. Text updated
ln134 – “dynamic model ... of forest dynamics” sounds a bit awkward	Thanks. Text updated
section 2.2 – I still do not feel like the information on the historical background of forest formation is worth inclusion in this paper. I agree that historical aspects predefined the current state of the forest structure in the region; however, your work must be focused more on remote sensing. The periods you described here start from the 10th century, your study covers the period since 1980th.	We largely agree with the opinion of the reviewer, however, we consider it expedient to leave this text. First, the previous history of intensive land use does indeed determine the state of forest cover. At the same time, despite the strong anthropogenic pressure and a high proportion of coniferous plantations, the state of the forest cover, as the result of successional dynamics, approaches the composition of zonal forests. This conclusion is in good agreement with our models. Secondly, we decided that the peculiarities of the main trends of forest management in the European part of Russia would be of interest to foreign specialists in the context of the history of the formation of coniferous forests (since this information is presented mainly in Russian-language literature sources). In addition, the mention of early history (10th -13th centuries) is only two sentences and takes up little space. Finally, there is a fourth reason for the preservation of the historical aspect - this is the absence of objections to subsection 2.2. From other reviewers. We will be ready to shorten this section if the editorial board's opinion coincides with you.
ln321 – use Google Earth Engine instead of “earth engine”: your repository is not publicly available, please share it	We updated the link – the script is available now
ln375 – your explanation of Random Forest classifier in this sentence is very simple.  Given the next paragraph, you can remove this sentence from the manuscript. In addition, you should cite Breiman L. (2001)	Text updated according to the comment.
section 2.3.3 – why you did not provide an explanation of the KNN algorithm?	We updated the text and added the k-NN description
section 2.3.4 – please, explain how you got updated for different periods GFC forest masks	Text updated Additionally for benchmarking our modeling results we use the Global Forest Watch (GFW) data on forest cover using the layer “percent forest cover” for two periods: 2010 and 2020 by calculation of forest cover in 2000 minus forest loss (cumulative for 2010 and 2020) and plus forest gain (cumulative for 2012).
ln404 – check punctuations	Text updated
Table 2 – without a description of the KNN algorithm I cannot understand how you used 32 nearest neighbors to predict categorical variables (classes). Generally, k-NN (see for example McRoberts & Tomppo, 2008) is an imputation method that is extensively used in forest inventory. Maybe, you reported here another approach, please explain	We started from assumption “best number of neighbours = SQRT (number of samples)” which equals 36 and then tested to increase/decrease number of neighbors. The best result was with 32 neighbors.

Round 2

Reviewer 2 Report

Review comments: Manuscript ID:- remotesensing-1190211

The manuscript (remotesensing-1190211) entitled with “Spatiotemporal Modeling of Coniferous Forests Dynamics Along the Southern Edge of Their Range in the Central Russian Plain” has gone through a better revision as compared to the earlier version. Major issues from my side were already taken into account. Thus, I recommend considering this manuscript for publication.

Reviewer 3 Report

The authors kindly addressed all my comments or provided an appropriate explanation, so I have no further suggestions.

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

A brief summary 

The paper aims to investigate a spatial distribution of pine and spruce forests within Moscow Region using field observation and Landsat data. The subject of the study is interesting, however I think that authors should have focused more on appropriate use of remote sensing techniques. Given the structure of the manuscript and authors contribution to remote sensing methodology, I would recommend considering another journal for publication of this research. In current form this study is out of scope of MDPI Remote sensing.

 

Broad comments

The paper has some major drawbacks that will not allow me to recommend it for publication in MDPI Remote Sensing:

1. The manuscript must be completely rewritten to meet the aims of the journal. In the result section authors report mostly on their findings that are based on field survey data analysis. This information may be useful for ecologists and botanists but not for specialists of remote sensing. There was few information extracted from satellite images processing. I believe that sections 3.1 and 3.2 out of the scope of MDPI Remote Sensing.
2. I have found some wrong citations which are used to refer to some methods or datasets, however these publications do not provide their description (see my comments below).
3. Description of data and methods is unclear. The text flow is mixed and description of classification approach is not enough.
4. On my opinion, authors applied incorrect approach for remote sensing pre-processing. They used 7-years period to create median Landsat mosaic while different forest disturbances could uccur within this time frame. In addition, for training they used spatial features extracted from images for 2010 and applied them to other periods. I belive that such simple approach will not work. There is also no information on forest mask accuracy which was produced from GFW data. Nevertheless, authors discuss the areal changes of different forest formations within forest masks. Ultimately, there is no information how land cover was mapped.
5. Authors used seven classifiers and did not provided detailed analisys of their performance.

 

Specific comments

Title

line 3 – remove uppercase letter in “CentraL”

 

Abstract

lines 19-29 – use another sentence construction, e.g. “we mapped ... using supervised classification”

line 28 – “MR” has not been explained before

line 29 – change “care” to “treatment’

line 31 – you used “random forest” with lowercase letters in the abstract

 

Introduction

This section is mostly focused on current problems of forests in MR, however I think that the introduction section should include more analysis regarding application of remote sensing to detect trends in forest cover change. Given the aims and scopes of the journal, the introduction must be re-written. There are also a lot of sentences with not-English words flow.

line 52 – lightnings is more common for “atmospheric electrical discharges”

line 53 - the statement that “pine is highly resistant to fires” is questionable; there is number of examples inthe literature according to which pine stands belongs to fire-prone forests

line 57 – what do you mean by “silviculture forests”?

line 64 – do not use “silviculture” in terms of planting

line 70 – please explain “MR”; use “disturbance” instead of “violations”

line 75 – start sentence from uppercase letter

line 92 – correct “km2”

line 102-103 – I did not understand what you tried to explain in this sentence – how statistical processing facilitates improved forest cover assessment

line 111-115 – these couple of sentences fit more to methods section

line 111-120 – I hardly can understand why you decided to include this information into introduction section

line 126 – you write “identify … vegetation mapping” as an aim of your paper

 

Material and Methods

Authors should improve the structure of this section. Methods include field and remote sensing data description. I believe there must be two sub-sections for field observation and remote sensing data.

line 167-173 – this information looks like discussion, nevertheless I did not understand the main idea of this portion of text

lines 174-233 – I don’t feel like “2.2. History of coniferous forests formation”  is a subject of Material and Methods section

2.3. Methods

This section mixes data and methods descriptions. Authors should disaggregate from here field survey data, remote sensing data. Then describe methods for satellite imagery pre-processing and classification. Authors use “modelling” instead of “classification”, in case of kNN one should use “imputation”.

line 283 – indicate that Juice 7.0 is software

line 298 – the sentence is unclear

line 300 - provide level of image processing used in this study

line 304 – use “Google Earth Engine”

line 305 – the repository pointed by reference #85 does not exist

line 306 – you combined images for 7 years, so what the year is reported in this paper? what's about forest disturbances and other factors of forest area dynamics – I think that you should have applied classifiers for yearly mosaics; how did you harmonize spectral reflectance of TM and OLI sensors?

line 311 - I cannot find description of all indices in reference #86

line 312 – what do you mean be “modelling”? I believe there must be “classification”

line 314 – so, what is the ratio you referenced at? I think that such information can be provided in this paper without pointing to your previous work

line 317 - why did you decided to use only 2010 mosaic to train model while data were collected during 2006-2019 (line 236)

line 318 – “train” is applied to classifier, not mosaic; on my opinion, this approach is not applicable without more sophisticated temporal normalization of spectral reflectance over years; I would recommend using some temporal segmentation approach for time series processing (for example LandTrendr)

line 319-320 – what type of data you used to detect forest disturbances

line 323-324 – LibSVM and KNN were not explained before

line 325 – there are no logic in putting together confusion matrix (table) and overall accuracy (measure) as means of the classification performance assessmnet – overall accuracy is calculated based on confusion matrix; check punctuation

line 328 – how did you extracted yearly forest cover from the GFW data – this map includes treecover only for the base layer? in addition, this sentence should go into previous paragraph because your data and methods description is a bit confusing

line 324 ­– the reference #87 is wrong – this paper does not present specified methods

 

Result

The section 3.1 is just analysis of ground sampling information and is a subject of data section. In addition, there is no need to provide so detailed description in the manuscript: you describe different forest formation however it was not resulted from remote sensing data processing. Inclusion of such information makes your contribution to methods of satellite data processing negligible.

Table 2 – how you justified the number of NN (32)?; given the size of the reference data set, your results will be smoothed and majority of predicted values will be close to the mean value

Table 3 – which methods does this table refer to?

line 508 – you cannot train mosaic

Figure 9 – correct the title – it should not start from panels descriptions; how did you disaggregate unforested areas? the data description does not provide any explanation on sample distribution to get land cover classification

 

Discussion

The information that authors discuss in lines 538-579 comes from other than remote sensing sources of data

lines 561-565 – I found firstly here your intention to use this type of model; how it can be applied to Landsat imagery remains unclear

lines 596-598 – based on your statement, you cannot correctly compare results from different periods: I absolutely agree with you and you should improve the methods

lines 606-608 – I do not think so; I would recommend you reading some papers regarding spectral reflectance harmonization for different Landsat sensors (e.g. Roy, D.P.; Kovalskyy, V.; Zhang, H.K.; Vermote, E.F.; Yan, L.; Kumar, S.S.; Egorov, A. Characterization of Landsat-7 to Landsat-8 Reflective Wavelength and Normalized Difference Vegetation Index Continuity. Remote Sensing of Environment 2016, 185, 57–70, doi:10.1016/j.rse.2015.12.024)

 

Conclusion

This section is too long

 

Reference

There are a lot of reference written in Russian (e.g, 48, 50-64 etc.) which complicates understanding what you want to say: I think, it has no sense for majority of international readers

Reviewer 2 Report

Review comments: Manuscript ID:-remotesensing-1131805

General Comment

This is an interesting manuscript about assessing modern composition and spatial distribution of forests with Norway spruce (Picea abies (L.) Karst.) and Scots pine (Pinus sylvestris L.) along the southern edge of their  range in the central Russian plain within the Moscow Region. However, there are aspects that require improvement and clear presentation before being considering for publication. Specific comments and suggestions are included below.

Specific comments and suggestions

L12-29: I suggest the abstract in the order of brief introduction, objectives, methods, findings and conclusion, which is not the case in the current version.

L34-37: add source(s)

L109-112: “LULCC surveys are very effective when they implement Landsat time-series, spectral-temporal modeling and modeling based on spectral signatures of land/vegetation cover types [42–44]. We use this approach in remote sensing and modeling section of methodology of the current study…..”, how this can be possible for mapping and modelling for the dynamics of a specific types of forests (Norway spruce (Picea abies (L.) Karst.) and Scots pine (Pinus sylvestris L.))?

L235-236: Include explanation for the choice of the reference years (2006-2019).

L236-237: “…The total number of relevés is 1608, of which 906 are in  coniferous forests (Table 1)…,” what is the base for deciding such number of plots?

L237-238: “…..Field data were collected using standard methodology [73] in  forest sample plots of 400–625 m2 area….,”, for which year dataset/reference year the field data set collected? How the field data for the historical one collected?

L290-292: why analysis of the dynamics and the current distribution performed based on Landsat 5 images (1990 and 2010)?

L324-325: “…Classification quality was  assessed using the confusion matrix and the overall accuracy…,” how testing samples collected? Was it collected together with training samples? If so, which proportion used between training and testing samples?

L1-5: Abstract is more of a summary about the approach, which is not self-contained. I suggest including key findings of the paper and conclusions.

L386-400: All presented are methods not results. I suggest moving them into appropriate place of the method section.

L695: How ground trough data collected? Which sampling strategy used for collecting the ground truth data of accuracy assessment?

L504: Table 2, there is a big deviation between Kappa and Overall accuracy in all different models. Why? Such kind of deviation shows problem related to the accuracy assessment strategies. In addition, the accuracy values are too small. How it is possible to trust the model outcomes with lower values, and to use them as input for any other analysis? I suggest the authors to check these again.

Reviewer 3 Report

See attached comments.

Comments for author File: Comments.pdf

Reviewer 4 Report

Review report: Spatiotemporal Modeling of Coniferous Forests Dynamics Along The Southern Edge of Their Range In The Central Russian Plain

 

In general, the paper is well established and the experimental design supports the research objectives and enables thorough analyses on the investigated phenomenon. The introduction section is well established to provide sufficient background information for the proposed study. However, I would recommend the authors to have a more global rather than local perspective here, and present the research background in a broader context. I mean, Section 2.2 already contains information about the historical aspects of the study area, so there might be some overlap now. Also, I would have liked to see more methodological discussion, i.e., some grounds for dynamic and retrospective modelling. This is a matter of opinion, of course, but now it remains somewhat unclear what is the methodological novelty of this study. I like the way the authors clearly present the objectives of the study at the end of introduction (and structurize the results accordingly).

 

The methods and materials section contains a very thorough description of the characteristics and history of the forests within the study area, which helps the reader understand the playground and the key features of it. The study methods have also explained very well, being both detail and easy-to-understand. However, for readability, I would recommend the authors to think about the structure of Section 2.3 and consider whether it would be feasible to structurize the text with some subheadings. It will help the reader follow the text; when it is about doing something with the field data, when it is about doing something with the remote sensing data, and when it is about combining information from both data sources, modelling and so on. Also, I came across an unfamiliar terminology - a “relevé”. I do not know if it is just me, but I think it would be beneficial to explain with a few words what that actually means without the need for a non-geobotanist to search for its meaning from the Internet.

 

The results section is structured according to the main research objectives presented in the introduction, so there is no fear of confusion and the reader finds what he/she is looking for. The most important results are presented with nice tables and figures to support the text. In the discussion section, the results of the proposed study are reviewed with respect to the previous findings, and casualties are searched for the findings. This section provides thorough discussion on the topic, and is in general well established. The conclusions really well and concisely binds together the most important findings of the study.

 

I would encourage the authors to look at the introduction as well as the materials and methods sections as suggested above and check the correct spelling and grammar throughout the manuscript.

 

Minor comments:

L28, 69 Please describe the acronym ‘MR’. I guess it means the Moscow region...

L143 Figure 1. Please add a scale bar and a north arrow into the maps

L345 Table 1. Please check the layout of the table as it seems somewhat confusing at least in the pdf-version of the manuscript.