Submit to this Journal Review for this Journal Propose a Special Issue

Article Menu

Share Help Cite Discuss in SciProfiles

Open AccessArticle

Peer-Review Record

Synergistic Use of Sentinel-1 and Sentinel-2 to Map Natural Forest and Acacia Plantation and Stand Ages in North-Central Vietnam

Remote Sens. 2021, 13(2), 185; https://doi.org/10.3390/rs13020185

by Ben Spracklen^* and Dominick V. Spracklen

Reviewer 1: Anonymous

Reviewer 2:

Erkki Tomppo

Reviewer 3: Anonymous

Reviewer 4: Anonymous

Remote Sens. 2021, 13(2), 185; https://doi.org/10.3390/rs13020185

Submission received: 21 November 2020 / Revised: 27 December 2020 / Accepted: 4 January 2021 / Published: 7 January 2021

Round 1

Reviewer 1 Report

This manuscript describes an interesting study into the use of Sentinel imagery to classify natural forest and acacia plantation and to classify stand ages. This study should be of interest to remote sensing, and is well written I have a general comment that the plantation area is quite restricted to two zones in the study area, which seems to be at lower elevation, close to villages and shallower slopes. The authors note that these 3 variables alone can explain 84.4% of the variation, which does not feature in the abstract or conclusion at all. Remote sensing with Sentinel 2 has only around 7% higher accuracy. While this is without the use of ancillary data, it seems plausible that the slope, and elevation are one of the factors being indirectly assessed. So how far can the study findings be extrapolated beyond this target area? Some caution in the abstract and conclusions is warranted. Another key point is that the 6 age classes can be classified with an accuracy of 71%, but the reality is that this is skewed towards greater accuracy at detecting stands less than 1 year old. It would be good to clarify this in the abstract and conclusions. Also the title notes 'Synergistic use of S1 and S2' but my take-home message was that S1 didn't add a lot of value, so I am not too sure of the synergies here.

Some more editorial comments are:

Line 8 "Tropical forests are heavily impacted..." is an emotive statement that doesn't really belong in a scientific paper. Also, the impact is not clear. Best to delete this sentence and start with "Many remote sensing studies..."
Line 13 - is there intended to be a classification accuracy (%) given for the combined S1 and S2?
L187 - please give the number of shapefiles in the training and testing sets
Line 272 - classifying acacia leaves as "tiny, dense and needle-like" doesn't accord with my definition of acacia leaves, which are actually quite large and flat. The leaves you describe sound like casuarina phyllodes or pine needles.

The figures also need a bit of work:

Fig. 2 - has too many acronyms that need to be expanded

Fig. 3 - the text is too small to read

Fig. 4 - The confusion matrix could be explained in more detail in the caption, especially around how the 'user' and 'producer' accuracy are presented

Fig. 5 - the text is too small to read

Once these concerns are addressed, I feel that the manuscript will be acceptable for publication in Remote Sensing

Author Response

Reviewer 1: We want to thank the Reviewer for their useful comments. In the following, we give the Reviewer’s comments, followed by our response in italics.

Response: Thanks for raising these important issues. We have added a section to the Discussion noting the importance of ancillary information to classification:” In the study area, on most of the flatter, lower-elevation sites close to villages, the natural forest has been overwhelmingly replaced by acacia plantation, while natural forest still dominates the steeper, higher elevation and more remote land. This meant that the classification accuracy using only ancillary data was extremely high at 84.4%, though the addition of S-1 or S-2 features did raise accuracy significantly by 3% and 8% respectively. Further, we note that without the ancillary features S-2 classification accuracy was reduced only slightly to 91.7%. Despite this, we note the possibility that slope and elevation may be being indirectly measured in the image values, and the applicability of our results to other areas where natural and plantation forest are more intermingled, and lie on similar terrain, should be tested further.”

In the Conclusions we have “We found Random Forest classification was effective at distinguishing acacia plantation under 2 years old from older plantation, but struggled to distinguish between the older plantation stands.”

In the Abstract we add “, with young plantation consistently separated from older. However, accuracy was lower at distinguishing between the older age classes.”

In the Conclusions we add “Elevation and slope were consistently highly rated features, and on their own obtained classification accuracies of 84%, reflecting the predominance of acacia plantation on the low-lying, flat terrain.”

Some more editorial comments are:

Line 8 "Tropical forests are heavily impacted..." is an emotive statement that doesn't really belong in a scientific paper. Also, the impact is not clear. Best to delete this sentence and start with "Many remote sensing studies..."

Response: deleted this line as requested

Line 13 - is there intended to be a classification accuracy (%) given for the combined S1 and S2?

Response: Apologies for the unclear wording. We have rewritten as “and 92.5% and 92.3% for S-2 and for S-1 and S-2 combined respectively”

L187 - please give the number of shapefiles in the training and testing sets

Response: Section 2 (Random Forests) we have changed to “The training set was composed of 15868 shapefiles, or 70% of total shapefile number, with the testing carried out on the remaining 30% (5306 shapefiles), as recommended by [47]”

Line 272 - classifying acacia leaves as "tiny, dense and needle-like" doesn't accord with my definition of acacia leaves, which are actually quite large and flat. The leaves you describe sound like casuarina phyllodes or pine needles.

Response: deleted these lines. Instead we add the following paragraph to Discussion: “After harvest the cross-polarised signal (VH) increased within a year to rough parity with natural forest. The co-polarised signal (VV) also rapidly increases after harvest but to a backscatter intensity lower than natural forest. Given the thick canopy and minimal forest clearings in both adult acacia and natural forest, we expect the backscatter to be dominated by canopy return, with minimal surface scattering. Natural forest may possess a higher degree of structural variation at S-1 resolution scale, which would result in a higher backscatter than the smoother acacia plantation canopy. Alternatively, the large, oblong leaves (about 4cm x 20cm) [61] of acacia may result in greater attenuation of the signal compared with the more diverse leaf sizes of the natural forest (such as tree species from the families Litsea, Machilus, Lauraceae and Euphorbiaceae.) [31]”

The figures also need a bit of work:

Fig. 2 - has too many acronyms that need to be expanded

Response: Completely reformulated figure so that it is hopefully clearer: flowchart runs vertically down the page, legend added, flowchart shapes reorganized, number of acronyms reduced, remaining acronyms explained in figure caption.

Fig. 3 - the text is too small to read

Response: We have increased the fontsize of the axis labels in Figure 3 and 5.

Fig. 4 - The confusion matrix could be explained in more detail in the caption, especially around how the 'user' and 'producer' accuracy are presented

Response: We agree these confusion matrices were unclear. We have switched to table format, with rows and columns clearly labelled, as well as user and producer accuracy.

Fig. 5 - the text is too small to read

Response: We have increased the fontsize of the axis labels in Figure 3 and 5.

Once these concerns are addressed, I feel that the manuscript will be acceptable for publication in Remote Sensing

Reviewer 2 Report

General comments

The authors use Sentinel-1 SAR data and Sentinel-2 optical area data and the both data sets
"to explore the accuracies in distinguishing between natural and plantation forests throughout the plantation lifecycle" and to estimate the plantation age
using a test site in Vietnam. Further, the authors
analyze of the changes of SAR backscatter (both VV and VH polarizations) and optical area intensities separately in harvested and non-harvested forests,
as well as seasonal variation.

The article is quite easy to read and proceeds logically.

Analyses of the changes of SAR backscatter and optical area intensities as well as seasonal variation separately in harvested and non-harvested forests
are interesting.

The authors have used the Random Forest /RF) classifier and write:
"Random forest (RF) [29] is a popular and powerful machine learning technique that has been
widely used for forest classification studies [30–33]."
Is this the only reason?
I am wondering would other methods, in addition to Random Forest, for example,
support vector machine, multinomial logistic regeression or any version of k-NN could also be tested and would give even larger accuracies.
Please comment the choice.

Stand age is a continuous variable.
What is the reason to classify it for the analyses?

Some material in the Results would suit better to section Discussions.
The section Discussions is missing now.

Please consider to rename the subsections in the section Results, e.g., instead of "Random forest" you could use "Accuracies of plantation and natural forest
classifications", or something better.

It would be interesting to see also the total area estimates and error estimates based on
the poststratified estimators (what is called a good practice).
These data sets could offer a good possibility to compare the real error and poststrafied error
estimators.

Please define all abbreviations, e.g., RDFI is not defined.

Please define all figures in rows and columns in the confusions matrices, Figures 4 and 8,
although they can guessed.
Concise tables would be easier to read.

Please use statistical tests to assess the significance of the differences in the classification accuracies when using different input data sets, e.g., in section 3.2.2, and also in the other comparisons.

Number all equations

Detailed comments

L 29:
Why not to discuss also carbon sink of plantations and natural forests
in addition to carbon storage?

L 50-51:
There are many references concerning the penetration and SAR frequency (wavelength).
Please cite some.

L 67:
"Can plantation age be effectively classified?"
Would it be more precise to write
"Can plantation age be accurately and timely classified?" ?

L 88:
A very detailed question:
Ho do you define the distance to the nearest village, from the centre point of the forest patch
(shapefile) to the centre point of the village, or as the shortest distance.

L 105-111:
Please tell the details, window size, etc., used in calculating the Co-Occurrence Matrix
and the textural features.

L 153
A reference could be given for the equation of SRR.

L 182-183
"We used an object-based approach (as opposed o a pixel-based classification),
with the mean and standard deviation of the pixel bands and their ..."
Usually, sigma nought, 10*log_10(average) is used.
Did you use that?

L 187-188
How did you decide the shares of 70% and 30%?

L 227-228,
Make sure that the '-' is connected to 14.5,
that is -14.5.
There us a line-brake between '-' and '14.5'.

Figure 3c,
Please consider to change the scale of the y-axis so that the seasonal variation could also be
seen.

Figure 4.
Would it be better to present the confusion matrices as tables.
The current figures are very large, take space and are not user-friendly
to compare to each other.

L 276-280 and L 367-372
The material on these lines would suit better to section Discussions.

Author Response

REVIEWER 2: We want to thank the Reviewer for their careful work on our paper, and for suggestions that we think have significantly improved its quality. In the following, we give the Reviewer’s comments, followed by our response in italics.

Response: We considered the use of SVM and CNN, and in the end went with RF for the following reasons, now added to the text (Section 2.3.4):

“Advantages include the ability to handle large numbers of input features, to estimate the importance of these features and its insensitivity to noise. These qualities make it a good choice for multi-source input data, and its accuracy at land-use classification has generally been found to be roughly comparable [37–40] to competitors such as Support Vector Machine (SVM) [41] and Convolutional Neural Networks (CNN.) [42] Additionally RF is relatively insensitive to the values of its free parameters, as opposed to CNN and SVM which require considerable fine-tuning, therefore making RF quick and easy-to-use.”

Stand age is a continuous variable.
What is the reason to classify it for the analyses?

Response: We looked at using RF regression for this problem, and decided to use classification for the following reasons: 1) due to obtaining our plantation ages from unevenly distributed (in time) Landsat images, our plantation age variable was not truly continuous, but classed according to available imagery 2) classification helped solve the unbalanced dataset – there were far more young acacia areas than old, and so by grouping the older stands together (3-5 year, 6-9 year) we could balance the dataset 3) comparison with the only other acacia age paper we could find, which also used classification.

Some material in the Results would suit better to section Discussions.
The section Discussions is missing now.

Response: Created a Discussion section and moved material from the Results to this section. We moved Lines 247-258, Lines 276-280; Lines 319-322, Lines 335-341, Lines 353-361 and Lines 367-379

Please consider to rename the subsections in the section Results, e.g., instead of "Random forest" you could use "Accuracies of plantation and natural forest
classifications", or something better.

Response: Changed subsection titles to “Natural forest and plantation classification accuracies”

It would be interesting to see also the total area estimates and error estimates based on the poststratified estimators (what is called a good practice).
These data sets could offer a good possibility to compare the real error and poststrafied error estimators.

Response: Added to Section 3.2.2::

“S-2 imagery and ancillary data classified 41247.7 ha and 9231.1 ha as natural forest and plantation respectively, with the reference MARD data giving 41084.8 ha and 9394 ha as natural forest and plantation respectively.”

.

Please define all abbreviations, e.g., RDFI is not defined.

Response: Apologies, RFDI should have been NDI (Normalised Difference Index). We have changed this mistake. Reduced abbreviations, defined abbreviations in Fig 2 caption.

Please define all figures in rows and columns in the confusions matrices, Figures 4 and 8,although they can guessed.
Concise tables would be easier to read.

Response: switched to using tables as suggested. Rows and columns now clearly labelled.

Please use statistical tests to assess the significance of the differences in the classification accuracies when using different input data sets, e.g., in section 3.2.2, and also in the other comparisons.

Response: Thank you for this suggestion. For thoroughness we decided to use 2 tests described in the paper as follows (Section 2.3.4):

“ We use a 2-proportion Z-test [44] to compare the proportions of correctly classified shapefiles between the pairs of interest: for example, comparing classification results using S-1 imagery and classification results using S-2 imagery. This tests the null hypothesis of no difference between the proportion of correctly classified shapefiles of each pair. Furthermore, we use McNemar’s test [45] for marginal homogeneity of the contingency table – in other words if the disagreements in classification between the 2 cases match. For both tests we report the χ² value and associated p value, with a statistically significant difference defined at the 5% level.”

Statistical test results are added to Sections 3.2.1. and 3.2.2 for:

comparing S-1 with S-2, S-2 with S-1 and S-2 combined, S-2 winter image and S-2 summer image, S-2 summer image and S-2 winter and summer image combined, S-2 with and without ancillary data, S-1 with and without ancillary data.

And to Section 3.3 for comparing S-2 classification and S-1 and S-2 classification combined

Number all equations

Response: all equations numbered

Detailed comments

L 29:
Why not to discuss also carbon sink of plantations and natural forests
in addition to carbon storage?
Response: Added as suggested.

L 50-51:
There are many references concerning the penetration and SAR frequency (wavelength).
Please cite some.

Response: We have added 2 citations, both comparing Sentinel-1 (C-band) with longer wavelength SAR: 1) Omar, H.; Misman, M.A.; Kassim, A.R. Synergetic of PALSAR-2 and Sentinel-1A SAR polarimetry for retrieving aboveground biomass in dipterocarp forest of Malaysia. Applied Sciences 2017, 7, 675.

2) Huang, X.; Ziniti, B.; Torbick, N.; Ducey, M.J. Assessment of forest above ground biomass estimation using multi-temporal C-band sentinel-1 and polarimetric L-band PALSAR-2 data. Remote Sensing 2018, 10, 1424

L 67:
"Can plantation age be effectively classified?"
Would it be more precise to write
"Can plantation age be accurately and timely classified?" ?

Response: Changed to “Can plantation age be accurately classified?”

L 88:
A very detailed question:
How do you define the distance to the nearest village, from the centre point of the forest patch(shapefile) to the centre point of the village, or as the shortest distance.

Response: distance was defined as from the centre point of the village to the centre point of the shapefile, we have clarified this point in section 2.2.3 Ancillary Data as follows: “, with distance defined as from the centre of the village to the centre of the forest shapefile.”

L 105-111:
Please tell the details, window size, etc., used in calculating the Co-Occurrence Matrix and the textural features.

Response: added more details to this section (Section 2.3.2): “These features were calculated for all images and both polarisations (VV and VH) for all angles, with a window size of 9x9, 32 quantization levels and a pixel displacement of 1.”

L 153
A reference could be given for the equation of SRR.

Response: Citation added to section 2.3.3: Tonolli, S.; Dalponte, M.; Neteler, M.; Rodeghiero, M.; Vescovo, L.; Gianelle, D. Fusion of airborne LiDAR and satellite multispectral data for the estimation of timber volume in the Southern Alps. Remote Sensing of Environment 2011, 115, 2486–2498.

L 182-183
"We used an object-based approach (as opposed o a pixel-based classification),
with the mean and standard deviation of the pixel bands and their ..."
Usually, sigma nought, 10*log_10(average) is used.
Did you use that?

Response: Yes, we have added a line to Section 2.2.1 to clarify: “This results in 2 features: the VV and VH polarized backscatter values (in decibels, dB.)”

L 187-188
How did you decide the shares of 70% and 30%?

Response: We followed the recommendation of Adelabu, S.; Mutanga, O.; Adam, E. Testing the reliability and stability of the internal accuracy assessment of random forest for classifying tree defoliation levels using different validation methods. Geocarto International 2015, 30, 810–821.

We have added the citation to Section 2.3.4 in order to clarify this.

L 227-228,
Make sure that the '-' is connected to 14.5,that is -14.5.
There us a line-brake between '-' and '14.5'.

Response: Thanks for spotting. We have stopped Microsoft Word from breaking the line.

Figure 3c,
Please consider to change the scale of the y-axis so that the seasonal variation could also be seen.

Response: We have added a zoomed inset figure to Figure 3c so that the seasonal variation can be clearly seen.

Figure 4.
Would it be better to present the confusion matrices as tables.
The current figures are very large, take space and are not user-friendly to compare to each other.

Response: changed to tables

L 276-280 and L 367-372
The material on these lines would suit better to section Discussions.

Response: created a Discussion section and moved this material to this section. We moved Lines 247-258, Lines 276-280; Lines 319-322, Lines 335-341, Lines 353-361 and Lines 367-379

Reviewer 3 Report

Comments for Authors

This manuscript is based on an interesting topic about mapping and differentiating forest types, natural versus plantations, in other to appropriately account and monitor service provided by forest. The authors explore both optical (Landsat and Sentinel-2) and radar (Sentinel-1 SAR) images to evaluate and compare the spectral and backscatter intensity between acacia plantations and natural forests cover. They try to compare the potential of different data sources (optical, radar, and their combination) to classification of forest types. Then, used spectral information (Landsat) to classify plantation age categories for sample site of acacia plantations. The authors have attempted at contributing to application of remote sensing data, but in its present form, I have some strong reservation as to the presentation and content of the manuscript.

I think that the introduction is not sufficiently elaborated with regards the necessary background why this study was conducted, the potentials of remote sensing along with the analysis procedures that were implemented in the study. A large proportion of the methodology is presented without elaborating the rational, purpose, and detail procedure; I have highlighted those identified in the comments.

Besides the general need to extensively revise and improve the English grammar, I have identified the following points that may help improve the manuscript in order to review

L8-9: How do forest plantations impact natural forest? The author should provide relevant background (as well, in the introduction section of the manuscript) for the ready to understand the rationale of this study.

L12-14: the author presents two classification accuracies (%) for three groups of data. As it reads, it seems that the 92.5% accuracy was obtained for both S-2 image group and the group of S-1 and S-2 combined. There is some ambiguity in reporting, which the author may clarify.

L14 – 15: Why was it so? The author does not discuss the key results in context of the data used for analysis, the nature of the vegetation, and in comparison with other literature. A discussion of the results is vital to situate the study in relation to existing literature, and to understand importance and impact of such findings.

L15-16: which data set was used for age-group classification of acacia plantations?

L17 -18: I do not understand what the meaning of visual imagery. For instance, one may ask if a radar image not visual?

L18 – 19: Is this conclusion based on the results? This is a general rationale for use of radar images for monitoring vegetation in area with frequent cloud cover.

L5 – 53: I am unable to get the main message in this statement. Is S-1 dependent of vegetation’s physical properties?

L30 – 31: How would such distinction be of great value?

L44 – 46: It is not clear how the 2 satellites of Sentinel1 (ascending and descending orbits) give a repeat time of 12 days or even 6 days overall. Please revise the sentence to clarify the idea of 12 day repeat cycle for a single orbit (ascending or descending) and a revisit cycle of 6 days when considering both orbits.

L50 -51: I find this statement ambiguous: does S-1 microwave signal have a high frequency? Whether or not the signal is dependent of the wavelength, and signals (waves) with long wavelength have a low frequency. However, depending on the context and in comparison to other wavelengths, the frequency of a microwave signal could be considered high or low. The author should provide this distinction and clarify their point of view.

L52 -57: The use of the term “visual” is not entirely clear here. In the author referring to optical images, I wonder if the both terms (visual can be used interchangeably in both the context of type of imaging sensor (line 53) and wavelength range in the EM spectrum (line 54-55). However, I suggest that, were appropriate, the author stick to the terms as largely used in remote sensing literature.

L58; How does plantation age impact carbon storage and species richness? What kind of species?

L65: I find this sentence incomplete. e.g. sources of what?

L67: In my opinion, this question is quite vague. Following from lines 58 to 61, one would expect as question such as, can plantation age be estimated? The research questions are not clear. I suggest that the research questions be rephrased and aligned with the specific goal of this study.

L70 – 75: It is necessary to provide geographic coordinates of the study area either in the text of Figure 1.

L77-79: The map is not informative: without tick marks, the geographic coordinates could be interpreted to be anywhere on the map. I find the legend incomplete: for instance, what does the red line represent? Also, it would be educative to provide a legend for the Elevation and the source of the elevation data.

L80 – 89: I strongly suggest that the author provides the necessary supporting reference(s) for the information on vegetation distribution and topography as I presume these data were not collected by the author.

L94: In my opinion, the figure caption does not appropriately describe the figure. The steps highlight on the figure are way beyond mere image pre-processing. I suggest that the author provide a more elaborate and concise description of the figure. What do the colours signify? In the present format, with juxtaposition of arrow and boxes, the figure is not comprehensible. I suggest that the figure be revised to a format that enhance interpretation the methodology – possibly grouping the pre-processing step and the different stages and types of analyses.

L97 – 99: These statements are unclear. In the present format, my interpretation is that VV and VH are both co-palarised, which is not the case. In line 99, it is also not clear which resolution the author is making reference to and how they are sampled.

L100 -101: What criteria were used in selected the five ascending images? Please, elaborate on the rationale and choice of the different image processing steps and analysis used in this study and provide relevant references where necessary.

L102-104: I guess this a standard procedure from the SNAP platform. This should be clearly stated and cited.

L105 – 109: Why and how do these texture features improve classification accuracy? Moreover, I doubt if this applies to every context of image analysis, study area, and type of images. Please, provide necessary background and justification for using the GLCM textures.

L109-110: What was the rationale and justification for choosing a window size of 5 by 5 and pixel displacement? Also, the author does not provide details about the orientation of texture estimation.

L112 – 123: Please, clearly justify the need and purpose of computing four radar indices. I strongly suggest that references be provided were appropriate, to support your choice of indices.

L121-122: I do not get the logic and reasoning why the RSI is based on a time series. Please elaborate your idea.

L124 – 127: What was the purpose of a time-series analysis? Considering the difference is geometry between ascending and descending images, is it logical to include both in a time-series analysis? To able to use images collected by both ascending and descending orbits together in the analysis, it would be necessary to pre-process and reconcile their geometric differences. If not done, as it appears in the study, the observed differences in backscatter may be largely contribute by disparity in geometry than ground features, leading to erroneous interpretation of the results and conclusions.

L126: I guess there is a misplace period (.)

L129 -130: What type of variation? For instance, in tree density.

L130 – 133: Which time-series is the author referring to here: time-series of S-1 or S-2 images? It is not clear how the time-series shows shifts in appearance; were the time series displayed as images or graphs? Please provide the corresponding figure or table for referencing.

L133: I do not understand what type of granules is being referred to here. S-2 granules? If so, state this explicitly in a revised form of the sentence.

L134 – 142: The author mention 6 images were used for the Random Forest classification. Two images were collected on February 15th 2018. However, it is not clear how many images were respectively collected on 16th May and 30th June 2018. On what basis was the 16th May images used as ‘master’ images.

L146 – 147: The time-series is referenced as a succeeding sections. I find this illogical since, according to the author, the analysis in sections 2.2.1 and 2.2.2 are based on it. A logical presentation would necessitate that the time series be present prior to these aforementioned sections.

L150: Which life cycle is the author referring to here?

L154: What was the rationale behind computing the ratios for the 10 bands of summer and winter image?

L157: It is quite confusing to grasp how many time-series were done at this point. I suggest the description of time-series, as well as other analyses, be presented in a separate second. Please, clarify and justify the purpose of the time series analysis. And, as previously mentioned, It may be necessary to present the time series analysis before the analysis that are based on it. Please, harmonise the usage of “time-series”, “time series”, or “times-series” in the entire manuscript.

Up to this point, my understanding is that most of the other analyses, using the estimated indices were based on “the time series”, not clear what this means: time-series analysis or not. I am not convinced that the time series was a key results as presented in the manuscript. I think that the results of time series were only exploratory, and these may not constitute a reliably grounds of conclusions of any sort.

L160: Here, the author mentioned that other studies justify usage of ancillary data to improve classification accuracies, and just one reference is cited. Please provide other references.

L162 – 163: How was the raster of village location produced? Please, elaborate with details of interpolation procedure and software.

L164: I suggest that this section be located prior to the previous sections in the manuscript.

L166 – 167: ….how the age of the plantation effects radar backscatter, spectral bands, and classification accuracy. This statement is unclear; I presume that author meant “affect”.

L167 – 171: Was the time series of S-1 and S-2 imagery conducted for the entire study area? Why so? I understand that a time-series analysis for the sampled forest areas (natural and plantation), as shown in Figure 3. Also, as aforementioned, I think that without correcting for geometric distortion, the time series from combination of images from both ascending and descending passes would be unreliable for making inferences on backscatter intensity over sampled field areas. Please clarify how this potential source of error was addressed.

L172 – 174: It is clearly elaborated how the S-2 RGB composite image help in selecting plantation areas. Which image bands were used for creating the RGB imagery? How were the RGB imagery interpreted? Please, provide these details, which may guide to understand the methods and possibly replicate your approach.

L171 -172: I am not sure how the difference in backscatter were tested. Please clarify how this was done and possibly provide summary statistics of the results.

L176: The author has not clarified the meaning of “our study cycle”.

L177 -179: Which time series is being referred to here? Please provide information about the reference figures within the text as is widely done in literature. Again, up to this point, it has not been stated if the time series was conducted across the plantation life cycle: what was the duration of plantation life cycle?

L183: I do not understand the meaning of “pixel bands”, their associated indices and textural measures.... Please clarify.

L187 – 188: Considering the difference in the respective number of shapefiles natural and plantations forest (from MARD database), it would be good to clarify the sampling procedure for the training and test data sets.

L190 – 193: What was the rational behind using 5 S-1 images for RF classification? Is one S-1 not capable of producing a similar results? Is it possible to also provide the error or confidence interval of classification?

L200 – 206: I strongly suggest that these information and details be clarified by including the used Path/Row on Figure 1 with necessary legend and indication of location of plantation areas, reserves, etc. This is will enhance reading and comprehension of the data and selected sites – North, East of KNT reserves.

L212 – 214: Could you provide relevant reference(s) for application of NBR. However, I find this statement somewhat inconsistent with the preceding text in the paragraph.

L215: What are Adjacent pixels? I find this statement incomprehensible.

L217: …..classified into 6 groups - …… I suggest the use of colon to initiate a list.

In general, there is not mention about where (software) the data processing, Landsat image-preprocessing and the classification were conducted.

L224 – 226: This is part of the methodology, or at least, I perceive this only as an exploratory result.

L226 – 232. I can grasp the main ideas in these sentences; they are quite difficult to read and understand.

L233 – 244: It is confusing to find the figure sub-label (a, b, c, etc.) on the charts; these are expected at the bottom of the respective charts.

L246 – 258: Are these entirely observable in the results in, for instance, Figure 3? Are there any supporting references for the stated facts about the indices or ratios? If so, please provide them accordingly.

L291 – 292: The figure caption does not appropriately describe the figure. For example, why are the different values presented in the corresponding colours? What do the percentages represent? What is the purpose of the third row and column, each lacking a title. I strongly suggest the use of widely used representation of a confusion matrix – tables.

L282 – 287: Why were these ancillary data most essential in the classification results? What explanation for the observation that the GLCM textures from the VH band were least important as classification predictors? I would expect such key results be addressed in a discussion section

L302 – 324: How is it that the summer image had a higher classification accuracy than the winder? Please, provide discussion section to address and explain the key results while considering other literature in the subject and similar studies.

L326 – 328: I think this figure is not explanatory enough. For instance, there are no map coordinates, indication of magnetic or geographic north, and map scale. Moreover, the legend is quite confusion; from interpretation, my understand is that the maps show, for the study area, the different training and test polygons used for the classification exercise. Yet, including the category of wrongly classified polygons gives room for another interpretation - a thematic map from the classification result. Please, clarify and provide appropriate map items to assist the reading in interpreting the map.

L343: Please provide the geographic orientation and coordinates on the map. I suggest to specify the year of classification to serve as reference for other studies or experiments in the future.

L345 – 352: I recommend that the widely used approach of presenting classification confusion matrix – in a table format. The last row and column do not have labels. Please, see me comment on Figure 5 about the description of the different numbers, percentages, and significance of the different colours.

L385 -387: The presented results do not provide evidence about the SWIR being consistently darker in acacia plantation than in natural forests.

L381 – 396: In my opinion, these conclusions are not sufficiently backed by the presented analyses, presentation and explanation of key results, and I question the reliability of the results considering the some highlighted possible sources of error in the presented analyses.

Comments for author File: Comments.pdf

Author Response

REVIEWER 3: We want to thank the Reviewer for their comments. We understand the time and effort this must have required and we appreciate the suggested changes which we feel have improved the paper. In the following, we give the Reviewer’s comments, followed by our response in italics.

Besides the general need to extensively revise and improve the English grammar, I have identified the following points that may help improve the manuscript in order to review

Response: Changed to in Introduction: “Forest plantations can reduce logging pressure on natural forests through providing an alternative source of timber. However, conversion of natural forest to plantation is a major driver of forest loss [2].”

Response: changed to “and 92.5% and 92.3% for S-2 and for S-1 and S-2 combined respectively” to clarify

Response: We have added a Discussion section to the paper where we discuss our findings more fully. We reword line as “We found that the ratio of the Short-Wave Infrared Band to the Red Band proved most effective in distinguishing acacia from natural forest.” as we feel this is more supported by Results.

L15-16: which data set was used for age-group classification of acacia plantations?

Response: added “on S-2 imagery” to clarify

L17 -18: I do not understand what the meaning of visual imagery. For instance, one may ask if a radar image not visual?

Response: replaced “visual” with “optical” here and, where appropriate, throughout the text

L18 – 19: Is this conclusion based on the results? This is a general rationale for use of radar images for monitoring vegetation in area with frequent cloud cover.

Response: Deleted this line

L5 – 53: I am unable to get the main message in this statement. Is S-1 dependent of vegetation’s physical properties?

Response: Tried to simplify by rewording and rearranging to “The SAR backscatter varies with the wavelength, polarization and incidence angle of the SAR signal. In addition, the SAR imagery from forest areas is dependent on the electrical properties and internal and external moisture content of the vegetation. The forest’s 3-dimensional structure, such as the roughness, size and orientation of the leaves and branches also affects the SAR backscatter.”

L30 – 31: How would such distinction be of great value?

Response: Changed to: “An ability to distinguish plantation and natural forests using remote sensing would be of great value by allowing the accurate monitoring of natural forest loss and plantation expansion.”

Response: Altered to “In Vietnam ascending and descending orbits have a repeat time of 12 days, giving an overall 6 day repeat time…”

Response: added citations and changed to “In the dense humid tropical canopies of Vietnam, the 5.5cm wavelength signal of Sentinel-1 will have limited penetration of the canopy. [12,13]“

Response: Apologies for our careless use of terminology. Changed so that now we refer to S-2 as optical, and reserve visual only for cases where we refer specifically to the Red, Green and Blue bands.

L58; How does plantation age impact carbon storage and species richness? What kind of species?

Response: We have significantly increased the detail on biodiversity in this section of the paper:” In the tropics, acacia plantations are often managed on very short (less than 5 year) rotations. However, there has been a recent push by the Vietnamese government [28,29] and certification bodies to switch to longer-term rotations, in an attempt to convert acacia plantation from woodchip to sawlog markets. Plantation age has important impacts on carbon storage [24] and species richness [25]. For example, 2 studies in Malaysia [26,27] found the number of bird species significantly increasing with age for 2, 5 and 7 year old acacia plantations. The 2 year old plantations were dominated by open-habitat and scrubland species, but mature acacia contained about 50% of the primary forest species, albeit lacking the more specialized and uncommon taxa. Determination of plantation age can therefore be useful to conservationists, landowners and for land-use planning purposes by allowing for an effective accounting of resource availability.”

L65: I find this sentence incomplete. e.g. sources of what?

Response: reworded to “What is the classification accuracy of Sentinel-1 (SAR), Sentinel-2 (optical) and S-1 and S-2 combined) for distinguishing natural forest and plantation?”

Response: We have lengthened the section detailing the rationale behind our research (see Comment to Ln58 above.) We have slightly rephrased our research question as: “Can acacia plantation age be accurately classified?”

L70 – 75: It is necessary to provide geographic coordinates of the study area either in the text of Figure 1.

Response: tickmarks added to map co-ordinates in Fig.1, co-ordinates of study area added to text “(E 106^o 17‘ to 106^o 56‘ – N 16^o 43’ to 17^o 32’)”

Response: Added tick marks, added red-line to legend, added legend for elevation, added source of elevation data to Figure caption “(sourced from Shuttle Radar Topography Mission data)”

Response: Reference given of detailed report on the area’s remaining natural forest “A Birdlife Report [31] gives a detailed overview of the remaining natural forest.” Mahood, S.; Van Trần, H. The biodiversity of Bac Huong Hoa Nature Reserve, Quang Tri Province, Vietnam; BirdLife International Vietnam Programme, 2008;

Response: Completely reformulated figure so that it is hopefully clearer: flowchart runs vertically down the page, legend added, flowchart shapes reorganized.

Response: Agree wording is confusing, have removed “co-polarised”. We add reference for resolution and reword to “These GRD images consist of VV (vertical send and vertical receive) and VH (vertical send and horizontal receive) polarisations, both with a resolution of 10m.”

Response: due to other comments by reviewer have switched to classification based on 1 descending orbit image. This image was chosen for its proximity in date to a S-2 image used for classification (February 20^th.)Added the following: “This date was chosen to correspond as closely as possible with a clear Sentinel-2 image.”

L102-104: I guess this a standard procedure from the SNAP platform. This should be clearly stated and cited.

Response: We’re not sure there is a single standard preprocessing procedure for Sentinel-1. We decided on our preprocessing based on the resulting imagery: in particular we decided that both terrain flattening and terrain correction was needed.

Response: added to Section 2.3.2 ”SAR imagery differs not only in intensity but also in texture (spatial variation.) Texture is a quantative measurement of the relationships of pixels with neighbouring pixels, often used to improve the accuracy of land-use classification studies. We might expect the texture of a planted forest of regularly spaced trees of the same age to differ from that of natural forest. Accordingly we chose the most frequently useed of the texture measures: the Grey-Level Co-Occurance Matrix (GLCM), which describes the frequency with which different pixel intensity values occur in an image.”

L109-110: What was the rationale and justification for choosing a window size of 5 by 5 and pixel displacement? Also, the author does not provide details about the orientation of texture estimation.

Response: added more details to this section 2.3.2: “These features were calculated for all images and both polarisations (VV and VH) for all angles, with a window size of 9x9, 32 quantization levels and a pixel displacement of 1.”

On further investigation, we switched window size to 9x9 for classification. We added to the text: “Choice of window size can be important in producing the most useful texture measure. Two forest classification studies [37,38] found larger window sizes gave the best classification performance. Further, it has been suggested [39] that while small window sizes are better-suited to heterogeneous environments with high local variance, larger window sizes are appropriate for more homogenous areas. Given our study area consists largely of thick and uniform canopy cover we therefore decided a large (9x9) window size was the most appropriate choice.”

L112 – 123: Please, clearly justify the need and purpose of computing four radar indices. I strongly suggest that references be provided were appropriate, to support your choice of indices.

Response: Added line and supporting citations “These indices have been proved to be effective in earlier work classifying plantation and natural forest. [23,38,39].”

Hoang, T.T.; Truong, V.T.; Hayashi, M.; Tadono, T.; Nasahara, K.N. New JAXA High-Resolution Land Use/Land Cover Map for Vietnam Aiming for Natural Forest and Plantation Forest Monitoring. Remote Sensing 2020, 12, 2707.

De Alban, J.D.T.; Connette, G.M.; Oswald, P.; Webb, E.L. Combined Landsat and L-band SAR data improves land cover classification and change detection in dynamic tropical landscapes. Remote Sensing 2018, 10, 306.

Sarzynski, T.; Giam, X.; Carrasco, L.; Lee, J.S.H. Combining Radar and Optical Imagery to Map Oil Palm Plantations in Sumatra, Indonesia, Using the Google Earth Engine. Remote Sensing 2020, 12, 1220.

L121-122: I do not get the logic and reasoning why the RSI is based on a time series. Please elaborate your idea.

Response: We accept that our sections are not in a logical order. Correspondingly, we have moved the time-series section forward so that it now appears (as Section 2.3.1) before the S-1 and S-2 processing sections (Section 2.3.2 and 2.3.3)

Response: We add the following lines to the Introduction to help explain our approach: “Our overall method of approach was to initially produce S-1 and S-2 time-series for natural forest and plantation for the period mid-2015 to mid-2020. These time-series would then be used to make a more informed judgement about what features to include in our classification model.”

We accept the point about the use of both ascending and descending images. We have switched to only using descending images throughout the paper.

L126: I guess there is a misplace period (.)

Response: changed

L129 -130: What type of variation? For instance, in tree density.

Response: added “in the vegetation”” to clarify

Response: added “S-2” to line to clarify. Removed reference to time-series so now reads:” All the forest, both natural and plantation, in the study region is evergreen, and variation in the vegetation over the year should be minimal. Nonetheless, we decided to test if the use of multiple images could improve classification accuracy. Accordingly in our S-2 classification we tested accuracy in the winter period, shortly after the end of the wet season, and accuracy in the summer period, towards the end of the study area’s dry period.”

L133: I do not understand what type of granules is being referred to here. S-2 granules? If so, state this explicitly in a revised form of the sentence.

Response: added “S-2” to before granules to make clear. Now reads: “Our study area was covered by two S-2 granules (QXD and QXE).”

Response: Sorry for lack of clarity. We have rewritten so that we hope is clearer: ”Therefore for the Random Forest classification we downloaded a total of 6 images as Level-1C Top-of-Atmosphere reflectance products: the ‘winter’ QXD and QXE S2 image from February 15^th2018 (https://scihub.copernicus.eu/) : (“S2A_MSIL1C_20180215TO31821_NO206_R118_T448QXD_20180215TO83955.SAFE” and S2A_MSIL1C_20180215T031821_N0206_R118_T48QXE_20200228T115658.SAFE”), a QXD and QXE ‘summer’ image from 16^th May, and a QXD and QXE ‘summer’ image from 30^th June 2018. Persistent cloud cover over the study area in summer 2018 meant that we needed images from 2 different dates to create a single cloud-free image. The 16^th May images were largely cloud-free and therefore classed as the ‘master’ images, with the June images used only to fill in any areas that were cloud-covered in the May image.”

Response: Agreed, and we have moved the time-series section to before the S-1 and S-2 processing sections. Time series is now 2.3.1, S-1 and S-2 now 2.3.2 and 2.3.3

L150: Which life cycle is the author referring to here?

Response: added “of an acacia plantation” to clarify

L154: What was the rationale behind computing the ratios for the 10 bands of summer and winter image?

Response: This was based on the S-2 time-series data. We accept that our ordering of sections was extremely confusing. We have rearranged so that the time-series section now appears before the S-2 processing section.

Response: We accept that the time-series used are of an exploratory nature, used to help our classification which is the main part of our analysis. We have added a section to the Introduction to help explain our approach. In Timeseries section we add line “In total, therefore, we have 4 time-series: an S-1 time-series and its control, and an S-2 time-series and its control.” We have switched entirely to using “time-series” throughout the manuscript.

As suggested time-series moved forward in the report so it now appears before the S-1 and S-2 processing sections.

L160: Here, the author mentioned that other studies justify usage of ancillary data to improve classification accuracies, and just one reference is cited. Please provide other references.

Response: added other references: Rodriguez-Galiano, V.F.; Ghimire, B.; Rogan, J.; Chica-Olmo, M.; Rigol-Sanchez, J.P. An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS Journal of Photogrammetry and Remote Sensing 2012, 67, 93–104.

Zimmermann, N.E.; Edwards, T.C.; Moisen, G.G.; Frescino, T.S.; Blackard, J.A. Remote sensing-based predictors improve distribution models of rare, early successional and broadleaf tree species in Utah. Journal of applied ecology 2007, 44, 1057–1067.

L162 – 163: How was the raster of village location produced? Please, elaborate with details of interpolation procedure and software.

Response: distance was defined as from the centre point of the village to the centre point of the shapefile, we have clarified this point in the section 2.2.3 Ancillary Data as follows: “, with distance defined as from the centre of the village to the centre of the forest shapefile.” Added citation for software (QGIS) used for processing.

L164: I suggest that this section be located prior to the previous sections in the manuscript.

Response: as suggested time-series section now moved prior to the S-1 and S-2 processing sections, so now in order S-1 data source and pre-processing, S-2 data source and preprocessing, Time series, S-1 processing, S-2 processing

L166 – 167: ….how the age of the plantation effects radar backscatter, spectral bands, and classification accuracy. This statement is unclear; I presume that author meant “affect”.

Response: “effects” changed to “affects”

Response: We accept the use of ascending and descending images in a time-series is potentially problematic. We switch to entirely using descending images, removing the ascending images, and adding in 10 new descending images as replacements.

Response: added to Section 2.3.1 “(Red-Green-Blue, comprising S-2 bands B2, B3 and B4)” as explanation. Add line “Acacia harvesting is obvious and unmistakable in the visible wavelengths.”

L171 -172: I am not sure how the difference in backscatter were tested. Please clarify how this was done and possibly provide summary statistics of the results.

Response: We accept the use of ascending and descending images in a time-series is potentially problematic. We switch to entirely using descending images.

L176: The author has not clarified the meaning of “our study cycle”.

Response: changed to “our study period of March 2015-June 2020” in Section 2.3.1

Response: in Section 2.3.1 changed to “From these S-1 and S-2 time-series we can see how band values in the years change following harvesting, and which bands vary most between plantation and natural forest.”

L183: I do not understand the meaning of “pixel bands”, their associated indices and textural measures.... Please clarify.

Response: deleted “pixel”

Response: While shapefile number was not exactly equal between the classes (11809 and 9365) we still felt that random sampling was appropriate.

Response: In accordance with this and other comments by the Reviewer we have switched to using a single descending S-1 image for the classification.

Response: Path/rows added to inset figure of Figure 1. In section 2.3.5, we add reference to the inset map of Figure 7, which shows location of subset plantations: “. The inset map of Figure 7 shows location of the plantation subset used for this section of the paper.”

L212 – 214: Could you provide relevant reference(s) for application of NBR. However, I find this statement somewhat inconsistent with the preceding text in the paragraph.

Response: Reference to Equation 5 added to this line for NBR. We have added 2 references to support our choice of threshold value and the use of NBR to indicate harvesting.

White, J.C.; Saarinen, N.; Wulder, M.A.; Kankare, V.; Hermosilla, T.; Coops, N.C.; Holopainen, M.; Hyyppä, J.; Vastaranta, M. Assessing spectral measures of post-harvest forest recovery with field plot data. International Journal of Applied Earth Observation and Geoinformation 2019, 80, 102–114.

Haywood, A.; Verbesselt, J.; Baker, P.J. MAPPING DISTURBANCE DYNAMICS IN WET SCLEROPHYLL FORESTS USING TIME SERIES LANDSAT. International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences 2016, 41.

L215: What are Adjacent pixels? I find this statement incomprehensible.

Response: Clarified as “Adjacent Landsat pixels that were classed as being harvested in the same year were grouped together into patches, and the patch boundaries converted into shapefiles, with the time since the most recent harvest used as the age of each shapefile”

L217: …..classified into 6 groups - …… I suggest the use of colon to initiate a list.

Response: changed to colon

In general, there is not mention about where (software) the data processing, Landsat image-preprocessing and the classification were conducted.

Response: Clarified in various sections (Section 2.3.4 (Random Forest):” RF classification was carried out using scikit-learn package in Python. [51]”. Added to Section 2.3.5 (Plantation Age): “Landsat processing was carried out using Python.”

L224 – 226: This is part of the methodology, or at least, I perceive this only as an exploratory result.

Response: Lines deleted

L226 – 232. I can grasp the main ideas in these sentences; they are quite difficult to read and understand.

Response: Section 3.1.1, reordered the line for clarity, now reads “…with mean ± standard deviation values for the VH band falling from -12.1±0.6 dB pre-harvest to -14.5±1.2dB immediately after harvest. Similarly, for the VV band, values fell from -6.9±0.7 dB to -8.6± 1.2dB.”

L233 – 244: It is confusing to find the figure sub-label (a, b, c, etc.) on the charts; these are expected at the bottom of the respective charts.

Response: Sub-labels moved to top of respective figures.

Response: We are basing these lines entirely upon Fig. 3c and 3d. We were unable to find any references to optical time series in acacia plantation. We have added an inset ‘zoomed’ section to Fig 3c so that the relative values of the selected bands can be more clearly seen.

Response: Switched to table format as requested, with rows and columns clearly labelled.

Response: We have added a discussion of ancillary data importance to the Discussion:” On most of the flatter, lower-elevation sites in the study area the natural forest had been overwhelmingly replaced by acacia plantation, while natural forest dominated the steeper, higher elevation land. This meant that the classification accuracy using only ancillary data was extremely high at 84%, though the addition of S-1 or S-2 features raised accuracy significantly by 3% and 7% respectively.”

Response: Discussion section added to paper, We added the following: ”Our time series (Fig 3c and 3d) showed a greater difference in S-2 bands between plantation and natural forest in the summer than the winter months, but we found no significant difference between the S-2 classification accuracies for a summer image, taken towards the end of the dry season, and a winter image, taken shortly after the end of the monsoon. However, using both these images for classification did significantly raise accuracy. Therefore even for evergreen forests such as found in our study area the use of multiple images can prove useful.”

Response: added map coordinates, north arrow and map scale.

L343: Please provide the geographic orientation and coordinates on the map. I suggest to specify the year of classification to serve as reference for other studies or experiments in the future.

Response: co-ordinates and north arrow added. Date of classification (February 2018) added to the Figure caption

Response: Changed to table format

L385 -387: The presented results do not provide evidence about the SWIR being consistently darker in acacia plantation than in natural forests.

Response: changed line to “with the ratio of the Short Wave Infra Red band to the Red Band the most important feature.”

Response: Addressed sources of error by switching from ascending/descending orbits to descending orbits only. Added statistical tests to check accuracy.

Reviewer 4 Report

This work combined C-Band Synthetic Aperture Radar (Sentinel-1, S-1) and optical satellite imagery (Sentinel-2, S-2) and examine Random Forest (RF) classification of acacia plantations and natural forest in North-Central Vietnam.

Specific comments and recommendations

Need to follow the journal guideline to prepare the manuscript.

Please make sure all dash, font size, space, a hyphen, en dash, and capital words would be appropriate thought out the manuscript.

Please make sure the font size in all figures.

Need to check English word throughout the manuscript.

Abstract:

Line 12: add “an ability”

Line 15: add “the natural forest”

Line 18: “a” or “the combination”

Introduction

Line 25: add “natural forests”

Line 27: add “so that”

Line 28: add “the ongoing”

Line 28: add “the plantation”

Line 30: add “, and local livelihood”

Line 33: add plantation forests” and natural forests

Line 44: add “In Vietnam ,”

Line 46 to 48: re-write the sentence “SAR…

Line 50: add “Canopies”

Line 50: “High-frequency”

Line 53: “,therefore,”

Line 56: “ and, china”

Line 59: “( 5 years)”

Line 65: “radar, or”

2.1. Study Site

Need to explain more about study area.

Need to add forest cover map for better understanding if possible.

Line 84 to 89: Should part of database. Re-adjust in data sources section

Line 94: Need to re-adjust the figure 2. Prepossessing is not clear to me.

Line 95 to 220: Lots of ideas are merging. Need to prepare two section regarding database, data sources, and methodology.

Line 236: Make sure about font size in the figures 2.

Line 294: Need to keep equal range of feature importance in figure 5 for better understanding.

Author Response

Reviewer 4: We thank the Reviewer for their useful comments on our paper. In the following, their comments are followed by our responses in italics.

Specific comments and recommendations

Need to follow the journal guideline to prepare the manuscript.

Please make sure all dash, font size, space, a hyphen, en dash, and capital words would be appropriate thought out the manuscript.

Please make sure the font size in all figures.

Response: Increased the size of font for all axis labels, titles and legends. Used Microsoft remotesensing-template.dot

Need to check English word throughout the manuscript.

Abstract:

Line 12: add “an ability”

Response: changed

Line 18: “a” or “the combination”

Response: added ‘a’

Introduction

Line 25: add “natural forests”

Response: changed

Line 27: add “so that”

Response: added “that”

Line 28: add “the ongoing”

Response: added “the”

Line 28: add “the plantation”

Response: changed plantation to plantations.

Line 30: add “, and local livelihood”

Response: added comma

Line 33: add plantation forests” and natural forests

Response: changed

Line 44: add “In Vietnam ,”

Response: added comma

Line 46 to 48: re-write the sentence “SAR…

Response: Broke this line in 2 so that it now reads: “The SAR imagery from forest areas is dependent on the electrical properties and internal and external moisture content of the vegetation. The forest’s 3-dimensional structure, such as the roughness, size and orientation of the leaves and branches also affects the SAR backscatter.”

Line 50: add “Canopies”

Response: changed

Line 50: “High-frequency”

Response: changed

Line 53: “,therefore,”

Response: added ‘Therefore,’ to start of line

Line 56: “ and, china”

Response: added comma

Line 59: “( 5 years)”

Response: changed to “less than 5 years”

Line 65: “radar, or”

Response: Added comma

2.1. Study Site

Need to explain more about study area.

Response: We have added the following section to include information on the state of the natural forest in the study area: “There is no virgin forest left within the study area, with all the natural forest affected to a varying degree by war, resource extraction and logging for high value timber. In many areas the natural forest is secondary forest, developing after cessation of agriculture. A Birdlife Report [27] gives a detailed overview of the remaining natural forest.”

Added citation for climate data.

Need to add forest cover map for better understanding if possible.

Response: Figure 5 provides a map of natural and plantation forest in the study area. We have added co-ordinates and north arrow to the Figure to enhance comprehension of the data.

Line 84 to 89: Should part of database. Re-adjust in data sources section

Response: Moved this paragraph to ancillary data section (Section 2.2.3.)

Line 94: Need to re-adjust the figure 2. Prepossessing is not clear to me.

Response: Completely reformulated figure so that it is hopefully clearer: flowchart runs vertically down the page, legend added, flowchart shapes reorganized.

Line 95 to 220: Lots of ideas are merging. Need to prepare two section regarding database, data sources, and methodology.

Response: divided up in Data Sources section and Methodology. We have divided up the S-1 and S-2 processing sections into 2 different subsections – one in the data sources section (2.2.1. and 2.2.2. for S-1 and S-2 respectively) and one in the Methodology section 2.3.2 and 2.3.3 for S-1 and S-2 respectively.

Line 236: Make sure about font size in the figures 2.

Response: We have increased the fontsize of the axis labels in all Figures.

Line 294: Need to keep equal range of feature importance in figure 5 for better understanding.

Response: Changed so that y-axis limits are the same in each subplot for (former) Figure 5, now Figure 4.

Round 2

Reviewer 2 Report

Dear Dr. Spracklen and the co-author,

You have taken into account all my comments and given responses exceptional thoroughly. It is my pleasure to thank you for your good work. The manuscript looks very good now. I propose the acceptance in its current form. Only two minor things that do not cause any additional work: 1) I think that production line numbers the equations using the journal’s standard, 2) post-stratified error estimation is missing but let’s leave it out. There is much enough and well-analysed and well-written material about the uncertainties already now.

Reviewer 4 Report

Thank you for revised manuscript.

LINE 97- 101: Need to re-write the sentences. seems incomplete sentences.

LINE 101- 102: Data source?

LINE 115-118: Please re-write the sentences.

Please make sure the references list, and font size in the figures.

good luck.

Article Menu

Synergistic Use of Sentinel-1 and Sentinel-2 to Map Natural Forest and Acacia Plantation and Stand Ages in North-Central Vietnam

Further Information

Guidelines

MDPI Initiatives

Follow MDPI