Predicting Soil Organic Carbon and Soil Nitrogen Stocks in Topsoil of Forest Ecosystems in Northeastern China Using Remote Sensing Data

Round 1

Reviewer 1 Report

I believe that the authors addressed successfully and/or responded to all the reviewers' comments. The manuscript is adequately improved and it is now suitable for publication apart from some minor corrections that are mostly linguistic. The language should be also checked one more time. Please find my minor comments in the commented pdf file.

Comments for author File: Comments.pdf

Author Response

Response to reviewer 1 comments on the manuscript remotesensing- 757741 “Predicting soil organic carbon and soil nitrogen stocks in topsoil of forest ecosystems in Northeast ern China using remote sensing data”

 

I believe that the authors addressed successfully and/or responded to all the reviewers' comments. The manuscript is adequately improved and it is now suitable for publication apart from some minor corrections that are mostly linguistic. The language should be also checked one more time. Please find my minor comments in the commented pdf file.

Response: We appreciate your help. With this submission, we provided a version (Track Changes) of the revised manuscript and detailed our responses to your comments as below.

 

1. L19 “prediction” change to “developed”.

Response: We have modified it according to your comment. L19

2. L20 “shows that” change to “developed”.

Response: We have modified it. L20

3. L73 “Is” inserted “which” before

Response: We have inserted it. L76

4. L77 “Our” change to “The”.

Response: We have modified it. L81

5. L145 “was” change to “were”.

Response: We have modified it. L151

6. L395 “a more reliable” why it is considered more reliable.

Response: We have eliminated this inaccurate statement. L447

 

 

We thank the reviewer 1’s constructive comments, which significantly help improve our manuscript.

 

Reviewer 2 Report

Most of the minor issues raised in previous rounds have been corrected in the resubmitted version of the manuscript, but the major issue related to the Landsat product level and the atmospheric correction has not been addressed properly. The recommendation made in the previous round was not to use the digital number in the models of soil properties, but the Botton-Of-Atmosphere (BOA) reflectance, which is the most appropriate remote sensing variable for this research. According to the resubmitted manuscript (L166-L168), an atmospheric correction has been performed to obtain the BOA reflectances. As no references, nor a detailed description of the atmospheric correction, nor a change in the results have been found in the resubmitted version of the manuscript, I have serious doubts on the atmospheric correction of the data. Thus, I would kindly like to ask the authors to provide complementary information, in order to clarify this issue. Specifically, a report including a detailed description of the process, software and tools used to convert the digital numbers and radiances to BOA reflectances, and the raw data used to generate the results in Table 3 and 4 in the previous version of the manuscript (remotesensing-733788-v2, models based on digital numbers) and to generate the same tables in the current version of the resubmitted manuscript (remotesensing-757741-v1, models based on BOA reflectances).

Other major comments:

1) L24 and L448. ‘This study provides a robust and efficient method to predict SOC and STN stocks of dense forest ecosystems.’ I honestly think that the method is not robust nor efficient, if we analyse the accuracy obtained in the validation. The Ratio of Performance to Deviation (RPD=SD/RMSE) is 0.56 for SOC and 1.33 for STN. The results for SOC are very poor, as the RMSE almost doubles the SD. Thus, I suggest avoiding these confusing sentences.

2) L172-L176. The topographic correction has not been explained in sufficient detail to allow the reproduction of the method. Please explain the method in detail or provide relevant references.

3) Equations 10 and 11. In a multiple stepwise linear regression (MLSR), it is expected to include uncorrelated variables. The MLSR models for the prediction of SOC and STN (Equations 10 and 11) include correlated variables, which is not appropriate. For instance, models include the NDVI, DVI and RDVI (RDVI=sq(NDVIxDVI)). These variables are correlated according to the information contained in Table 3, with a Pearson correlation coefficient of 0.47 (correlation between NDVI and RDVI), and 0.43 (between DVI and RDVI).

Minor comments:

L16. Please remove one of the brackets after ‘BRT’.

L62. Please consider replacing ‘remote sensing image data’ with ‘images’.

L64. Space required between ‘estimate’ and ‘SOC’.

L74. ‘source and provides’.

L77. ‘The objectives of this research’.

Figure 1. The labels have not been corrected in the resubmitted manuscript. ‘km’ in lowercase according to the international System of Units (SI).

L97. Please add a space between the value and the units: ‘1100 mm’. This is also applicable to other parts of the manuscript, for instance in L138, L139, L147, L156, L157, L286, L288, L292, L299, etc. Please check the manuscript carefully to correct this error.

L105. ‘of’ is missing. ‘Variation of SOC’.

L145. ‘were’ instead of ‘was’.

L146. ‘was’ instead of ‘is’.

L166-L168. Please rewrite this sentence. Do you mean ‘digital numbers were converted to radiances’?

L192. Please explain the variable ‘L’ on first use.

Equations (10) and (11). Please include not only the regression coefficients, but also the error of them, for instance 15.33±0.06.

L265. ‘of the BRT model’.

L265-L266. What is the meaning of ‘LR’, ‘TC’, ‘BF’ and ‘NT’?

L274. Please remove one period after ‘models’.

L290-L292. Please rephrase this sentence and avoid using ‘vs’.

L292. Please add a period after ‘(Table 2)’.

L320. ‘forecasting’ is not adequate, as the models estimate the SOC and STN content, but do not forecast the future content of these soils properties.

L335. ‘SD, standard deviation’ has not been included in Table 4.

Figure 4. Please include in each plot a label with the R² and RMSE values. The variable represented in each plot (SOC or STN) cannot be distinguished, as this information is not included in a label nor in the figure caption.

L359. ‘variables’ instead of ‘ variable’.

L359. ‘100 times simulations’, do you mean ‘100 simulations’?

Table 5. Please check the column names, as two columns have the same name (‘Average STN stocks (kg m^-2)’).

Figures 7 and 8. Please use ‘km’ in lowercase in the scale labels and integer divisions in the scale (for instance 0, 5, 10, 20, 30 and 40 km).

Author Response

Please see the attach below.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The major issue raised in previous rounds related to the atmospheric correction has been clarified in the revised manuscript, and apparently, the research has been conducted properly. A detailed description of the process, software and tools used to convert the digital numbers and radiances to BOA reflectances has been provided in the cover letter, which helped to clarify the issue. The description of the atmospheric and topographic corrections has been improved in the methods section of the manuscript and now the research is more reproducible. The statistical analysis has been reworded to use BOA reflectances instead of digital numbers, and the results in Tables 1, 3 and 4 have been modified accordingly. For this research, the BOA reflectance is more appropriate than the digital number and it is reflected by the new model performance results, which improved with respect to those in the previous versions of the manuscript.

In this round, I would only like to mention an important issue that still needs to be clarified. According to the cover letter (major issue #3), the multiple stepwise linear regression (MLSR) models have been rebuilt and modified to avoid the inclusion of correlated variables. However, the MLSR models for the prediction of SOC and STN (Equations 12 and 13) still include correlated variables, which is not appropriate. For instance, the models include the NDVI, DVI and RDVI (RDVI=sq(NDVIxDVI)). These variables are correlated according to the information contained in Table 3, with a Pearson correlation coefficient of 0.46 (correlation between NDVI and RDVI), and 0.41 (between DVI and RDVI). In MLSR, it is expected to include uncorrelated variables, thus I suggest rebuilding these models.

Minor comments:

Equations 12 and 13. Please include not only the regression coefficients, but also the error of them, for instance '(13.31±0.06) – (0.842±0.002)*B_GREEN + (0.33±0.05)*B_NIR'.

L315-318. Please consider to include ‘respectively’. For instance, ‘the average SOC and STN stocks were 13.14 kg m^-2 and 1.43 kg m^-2, respectively’.

Figure 4. The variable represented in each plot (SOC or STN) cannot be distinguished properly, as this information is not included in a label nor in the figure caption. Do plots (a-c) correspond to SOC and plots (d-f) to STN? Please clarify.

Author Response

Response to reviewer 2 comments on the manuscript remotesensing-757741 “Predicting soil organic carbon and soil nitrogen stocks in topsoil of forest ecosystems in Northeast ern China using remote sensing data”

 

The major issue raised in previous rounds related to the atmospheric correction has been clarified in the revised manuscript, and apparently, the research has been conducted properly. A detailed description of the process, software and tools used to convert the digital numbers and radiances to BOA reflectances has been provided in the cover letter, which helped to clarify the issue. The description of the atmospheric and topographic corrections has been improved in the methods section of the manuscript and now the research is more reproducible. The statistical analysis has been reworded to use BOA reflectances instead of digital numbers, and the results in Tables 1, 3 and 4 have been modified accordingly. For this research, the BOA reflectance is more appropriate than the digital number and it is reflected by the new model performance results, which improved with respect to those in the previous versions of the manuscript.

Response: We appreciate your help and your patience.

 

In this round, I would only like to mention an important issue that still needs to be clarified. According to the cover letter (major issue #3), the multiple stepwise linear regression (MLSR) models have been rebuilt and modified to avoid the inclusion of correlated variables. However, the MLSR models for the prediction of SOC and STN (Equations 12 and 13) still include correlated variables, which is not appropriate. For instance, the models include the NDVI, DVI and RDVI (RDVI=sq(NDVIxDVI)). These variables are correlated according to the information contained in Table 3, with a Pearson correlation coefficient of 0.46 (correlation between NDVI and RDVI), and 0.41 (between DVI and RDVI). In MLSR, it is expected to include uncorrelated variables, thus I suggest rebuilding these models.

Response: Thank you for your careful and patient comments. Based on your comments, we have revised equations 12 and 13 again. Table 4, figure 3, 4, 7 and 8 involved in the manuscript have been modified accordingly. L267-276, Table 4, Figure 3,4,7and8.

 

Minor comments:

 

Equations 12 and 13. Please include not only the regression coefficients, but also the error of them, for instance '(13.31±0.06) – (0.842±0.002)*BGREEN + (0.33±0.05)*BNIR'.

Response: We have modified equations 12 and 13 based on your comments. L267-276

L315-318. Please consider to include ‘respectively’. For instance, ‘the average SOC and STN stocks were 13.14 kg m-2 and 1.43 kg m-2, respectively’.

Response: According to your comment, we have modified them. L319-321

Figure 4. The variable represented in each plot (SOC or STN) cannot be distinguished properly, as this information is not included in a label nor in the figure caption. Do plots (a-c) correspond to SOC and plots (d-f) to STN? Please clarify.

Response: We modified Figure 4 to avoid this confusion. Figure 4

 

We thank the reviewer 2 constructive comments, which significantly help improve our manuscript.

 

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

The article deals with the important issue of developing methods for monitoring soil carbon and nitrogen content. The authors proposed mapping of C and N content in soils under forests based on field survey data, satellite data and digital soil mapping technologies.

The article needed to be further developed before a decision could be made on whether to publish it.

1. The title of the article does not correspond to its content. The article does not refer to the full content of carbon and nitrogen in soils under forests. It only refers to a 30 cm layer of soil.
2. The claim that soil carbon and nitrogen are important indicators of forest soil productivity (line 35) is too categorical.
3. There is too little information about soils in the description of the object of study. It is necessary to provide data on their basic properties and the content of carbon and nitrogen in soils of the region.
4. Not all factors of soil formation are listed and taken into account (lines 110-112). For example, important factors are soil parent material and the level of groundwater, which may predetermine the spatial variation of soils, but they are not taken into account by the authors. Besides, as one of the factors of soil formation they use NDVI, which is wrong. The soil forming factor is vegetation, but not NDVI. As a result, zoning that does not take into account important soil formation factors can hardly be considered reliable.
5. The methods of soil sampling and analysis raise many questions. It is not clear why the authors singled out soil fraction >2mm. After all, coarse fragments in soils can have a smaller size. Besides, it is very doubtful to take a sample to a depth of 30 cm. Why exactly 30 cm and not from all horizons of soils containing organic matter?
6. The units of all variables on formulas 1 and 2 should be indicated.
7. Section 2.2.3 is not clear at all. It is necessary to describe in detail what was done with satellite images, what does the phrase "were generated from 11 images" mean?
8. Section 2.3 can be removed from the manuscript without damage. Such a description of methods can be found in the cited papers. There's no point in repeating it.
9. The C soil content can only be expressed in kg/m2 if the C content of all soil horizons is taken into account (all C is taken into consideration). When only 0-30 cm layer is considered (not all C is taken into consideration), it is necessary to refer to it.
10. The authors use the term “scale” incorrectly (lines 49-50). Large scale maps show a smaller amount of area with a larger amount of detail. Small scale maps show a larger geographic area with few details on them.

Author Response

Response to reviewer 1 comments on the manuscript remotesensing-733788 “Using remote sensing data to predict soil organic carbon and soil nitrogen stocks in Northeastern Chinese forests”

 

The article deals with the important issue of developing methods for monitoring soil carbon and nitrogen content. The authors proposed mapping of C and N content in soils under forests based on field survey data, satellite data and digital soil mapping technologies.

 

The article needed to be further developed before a decision could be made on whether to publish it.

Response: We appreciate your help and your patience. With this submission, we provided a version (Track Changes) of the revised manuscript. Below we detail how we have followed your comments in our revision.

1.The title of the article does not correspond to its content. The article does not refer to the full content of carbon and nitrogen in soils under forests. It only refers to a 30 cm layer of soil.

Response: Thanks for the comment. To better reflect the content, we have revised the title to “Predicting soil organic carbon and soil total nitrogen in topsoil of forest ecosystems in Northeastern China using remote sensing data”. L2-4

2. The claim that soil carbon and nitrogen are important indicators of forest soil productivity (line 35) is too categorical.

Response: We deleted the sentence.

3. There is too little information about soils in the description of the object of study. It is necessary to provide data on their basic properties and the content of carbon and nitrogen in soils of the region.

Response: Following your comments, we have added the description of soil type to the manuscript in L137-158. In addition, to enrich the content, we also added Figure 2b and Table 2 to the manuscript.

4.  Not all factors of soil formation are listed and taken into account (lines 110-112). For example, important factors are soil parent material and the level of groundwater, which may predetermine the spatial variation of soils, but they are not taken into account by the authors. Besides, as one of the factors of soil formation they use NDVI, which is wrong. The soil forming factor is vegetation, but not NDVI. As a result, zoning that does not take into account important soil formation factors can hardly be considered reliable.

Response: Thanks for your comments. Due to the difficulty of data collection, we did not collect the information of groundwater and parent material in the early sampling scheme design process, but in the sampling scheme design process, we considered the soil type and topographic Wetness index. Indeed, we agree that one of the five soil forming factors is vegetation, not NDVI. However NDVI is a good indicator for vegetation in terms of plant productivity and thus litters from plants, which will affect soil carbon and nitrogen. We revised the manuscript according to your comment to make this clear. L171-176

5. The methods of soil sampling and analysis raise many questions. It is not clear why the authors singled out soil fraction >2mm. After all, coarse fragments in soils can have a smaller size. Besides, it is very doubtful to take a sample to a depth of 30 cm. Why exactly 30 cm and not from all horizons of soils containing organic matter?

Response: The coarse fragments in soils can have a smaller size, but we recorded gravel content greater than 2mm only for later calculation of SOC and STN density. See Formulas 1 and 2. In addition, limited by funds and personnel, it is very difficult to sample deeper (such as 1 m) in such a large regional forest area, so this study is only to sample surface soil samples (0-30cm). In our next research, we will select a relatively small forest area to carry out deeper sampling so as to obtain more accurate SOC and STN stocks estimation of forest soil. In this revision, we expanded the uncertainty analysis for the manuscript. L179-185

 6. The units of all variables on formulas 1 and 2 should be indicated.

Response: We added missing unit information based on your comment. L198-202

7. Section 2.2.3 is not clear at all. It is necessary to describe in detail what was done with satellite images, what does the phrase "were generated from 11 images" mean?

Response: We added a description of processing satellite image. In addition, regarding the phrase "were generated from 11 images", we revised it to "were produced using 11 satellite imageries" in this revision. L207-216

 8. Section 2.3 can be removed from the manuscript without damage. Such a description of methods can be found in the cited papers. There's no point in repeating it.

Response: Thank you for your patience and support. We considered that Section 2.3 is necessary with more detailed description of our research models.

 9. The C soil content can only be expressed in kg/m2 if the C content of all soil horizons is taken into account (all C is taken into consideration). When only 0-30 cm layer is considered (not all C is taken into consideration), it is necessary to refer to it.

Response: Thank you for your detailed comments. We added a 0-30cm level description to the manuscript based on your comments. The whole manuscript.

10. The authors use the term “scale” incorrectly (lines 49-50). Large scale maps show a smaller amount of area with a larger amount of detail. Small scale maps show a larger geographic area with few details on them.

Response: We have revised it following your comments. L60

 

 

We thank the reviewer 1 constructive comments, which significantly help improve our manuscript.

 

Reviewer 2 Report

The manuscript focuses on comparing the performance of three modelling techniques (Geographical Weighted Regression, GWR; Multiple Linear Stepwise Regression, MLSR; and Boosted Regression Trees, BRT) to estimate Soil Organic Carbon (SOC) and Soil Total Nitrogen (STN) stocks in the topsoil of forests located in Liaoning Province (Northeastern China). 513 topsoil (0-30 cm) samples were collected to train the models. Nine environmental variables (three bands and six spectral indices) generated from 11 Landsat-8 images were used as predictor variables.

Soil carbon and nitrogen are important indicators of soil quality and productivity and are of a great significance to global carbon and nitrogen balance and global climate change. The progress in the estimation of soil carbon and nitrogen using remote sensing is very valuable. The topic of the manuscript is interesting, but there is a number of issues that need to be clarified.

Major comments:

1. The method to estimate the soil carbon and nitrogen in the forests of the study area is based on the use of environmental variables related to the vegetation that were obtained from Landsat-8 images. The results show that there is a certain relationship between the soil carbon and nitrogen and the reflectance of the forest vegetation. The manuscript is based in this relationship, however the physical reasons of this link are not explained at all. Why the carbon and nitrogen in the forest soils are related to the reflectance of the forest vegetation? I would suggest providing a deeper explanation of this relationship, not only from a statistical point of view (showing the R or R² among variables), but also from a physical point of view.

2. Three Landsat-8 bands from the optical region of the spectrum (VIS-Green, VIS-Red, and NIR) were used as predictor variables. According to Table 1, the units of these bands are degree Celsius, which cause me a great concern. Does the surface range between 21.8 and 106.7 degrees Celsius? What processing steps have been applied to the Landsat-8 images? What Landsat-8 product has been used?

3. The most accurate models for estimating SOC and STN stocks are based on the BRT method (SOC: R²=0.52 and RMSE=1.09 kg/m²; STN: R²=0.46 and RMSE=0.35 kg/m2). The RMSE values of the best models are even greater than the standard deviation of the soil samples (SOC: SD=0.53 kg/m2; STN: 0.32 kg/m2). Thus the accuracy of the models is very poor and the following sentences do not seem reasonable:

(1) ‘BRT models … could be used to accurately predict SOC and STN stocks in densely vegetated areas’ (Abstract, L23-L25)

(2) ‘The results showed that the BRT model had lower uncertainty in the prediction of SOC and STN stocks, with an average SDs of 0.86 and 0.26 kg/m2, respectively’ (Results, L264-L266)

(3) ‘the lower SDs value of BRT showed that the BRT had a good prediction performance in forest areas of Norheastern China’ (Conclusions, L377-L378).

These sentences should be avoided as they contain information that is not derived from the results.

4. Important details about the remote sensing data and processing steps are missing. The reader could not reproduce the research using the information provided in the manuscript. It is needed a deeper explanation on the Landsat-8 product that have been used, its processing level, spatial resolution, atmospheric correction, cloud removal and shadow processing. The study area includes mountainous regions, has a topographic correction been applied?

5. What R package have been used to perform the statistical analysis? ‘demo’ package has not been found in the R repositories (L288).

6. The cross-validation method usually underestimates the error. Why the cross-validation method has been selected instead of a validation based on an independent dataset? The number of samples seems large enough to split them into two datasets (train and test).

7. Density plots in Figure 3 are interesting to show the prediction performance of the three methods, but the plots of predicted versus observed values are of a great importance. I strongly recommend adding the plots of predicted versus observed values for the six models in Table 3 and Figure 3.

8. It would be useful to include a land use map when describing the land use types in Section 2.1. It would help the reader to understand better the study area.

Minor comments:

1. L24. Please consider adding some indicators of the predictive quality of the models (R² and RMSE for instance).

2. L61. Do you mean ‘KOMPSAT’ instead of ‘KOMPAST’?

3. L86. ‘9 remote sensing based environmental variables’. Please be concise and indicate that Landsat-8 band reflectances and spectral indices have been used.

4. Figure 1. North arrow and coordinate grids are missing.

5. L115. Please indicate in the text that soil samples were collected in 2015.

6. L188. Please include the accuracy of the GPS used to record the location of the soil samples.

7. Equation 3. Why ‘L’ value has been fixed to 0.5?

8. L169. Typo. ‘from’.

9. L205. ‘in a complex’.

10. L216. Please remove the period before ‘have’.

11. Table 2. Please indicate the variable that is showed in this table. Is it the Pearson correlation coefficient? ‘Relationships’ is ambiguous.

12. Figures 5 and 6. ‘a’ is missing.

13. Table 4. Do these summary statistics correspond to the area within the black rectangle in Figures 5 and 6 or to the entire Liaoning Province?

14. L392. Typo. ‘MAE’.

Author Response

Response to reviewer 2 comments on the manuscript remotesensing-733788 “Using remote sensing data to predict soil organic carbon and soil nitrogen stocks in Northeastern Chinese forests”

 

The manuscript focuses on comparing the performance of three modelling techniques (Geographical Weighted Regression, GWR; Multiple Linear Stepwise Regression, MLSR; and Boosted Regression Trees, BRT) to estimate Soil Organic Carbon (SOC) and Soil Total Nitrogen (STN) stocks in the topsoil of forests located in Liaoning Province (Northeastern China). 513 topsoil (0-30 cm) samples were collected to train the models. Nine environmental variables (three bands and six spectral indices) generated from 11 Landsat-8 images were used as predictor variables.

 

Soil carbon and nitrogen are important indicators of soil quality and productivity and are of a great significance to global carbon and nitrogen balance and global climate change. The progress in the estimation of soil carbon and nitrogen using remote sensing is very valuable. The topic of the manuscript is interesting, but there is a number of issues that need to be clarified.

Response: We appreciate your help and your patience. Below we detail how we have revised the manuscript following your comments and suggestions.

 

Major comments:

1. The method to estimate the soil carbon and nitrogen in the forests of the study area is based on the use of environmental variables related to the vegetation that were obtained from Landsat-8 images. The results show that there is a certain relationship between the soil carbon and nitrogen and the reflectance of the forest vegetation. The manuscript is based in this relationship, however the physical reasons of this link are not explained at all. Why the carbon and nitrogen in the forest soils are related to the reflectance of the forest vegetation? I would suggest providing a deeper explanation of this relationship, not only from a statistical point of view (showing the R or R2 among variables), but also from a physical point of view.

Response: Following your suggestions, we have added some supplementary contents to the manuscript. L66-72

2. Three Landsat-8 bands from the optical region of the spectrum (VIS-Green, VIS-Red, and NIR) were used as predictor variables. According to Table 1, the units of these bands are degree Celsius, which cause me a great concern. Does the surface range between 21.8 and 106.7 degrees Celsius? What processing steps have been applied to the Landsat-8 images? What Landsat-8 product has been used?

Response: This was an error. Its correct unit is “digital number”. In addition, the processing steps have been added to the manuscript. The selected remote sensing data include 9 variables, including Landsat TM green band (B_GREEN)(0.52-0.60μ m), Landsat TM red band (B_RED)(0.63-0.69μ m), Landsat TM near-infrared band (B_NIR)(0.77-0.90μ m), difference vegetation index (DVI), enhanced vegetation index (EVI), ratio vegetation index (RVI), normalized difference vegetation index (NDVI), renormalization difference vegetation index (RDVI), and soil adjusted vegetation index (SAVI). See Table 1, L207-218

3. The most accurate models for estimating SOC and STN stocks are based on the BRT method (SOC: R2=0.52 and RMSE=1.09 kg/m2; STN: R2=0.46 and RMSE=0.35 kg/m2). The RMSE values of the best models are even greater than the standard deviation of the soil samples (SOC: SD=0.53 kg/m2; STN: 0.32 kg/m2). Thus the accuracy of the models is very poor and the following sentences do not seem reasonable:

Response: Thanks for the comments. We double-checked and revised the manuscript accordingly.

 

(1) ‘BRT models … could be used to accurately predict SOC and STN stocks in densely vegetated areas’ (Abstract, L23-L25)

Response: We have revised it to “T This study provides a robust and efficient method to predict SOC and STN stocks of dense forest ecosystems. Our findings can be used to evaluate soil quality and facilitate land policy and regulation making by the government in the region.”. L30-32

 

(2) ‘The results showed that the BRT model had lower uncertainty in the prediction of SOC and STN stocks, with an average SDs of 0.86 and 0.26 kg/m2, respectively’ (Results, L264-L266)

Response: We have revised to “We found that the BRT model had a lower uncertainty compared with GWR and MLSR models.”. L383-386

 

(3) ‘the lower SDs value of BRT showed that the BRT had a good prediction performance in forest areas of Norheastern China’ (Conclusions, L377-L378).

Response: We have modified it to “BRT performed best in comparison with GWR and MLSR models to predict SOC and STN in forest areas of Northeastern China.”. L514-517

 

These sentences should be avoided as they contain information that is not derived from the results.

Response: Thanks for your comments, and the manuscript has been revised to state what we have found from this study.

4. Important details about the remote sensing data and processing steps are missing. The reader could not reproduce the research using the information provided in the manuscript. It is needed a deeper explanation on the Landsat-8 product that have been used, its processing level, spatial resolution, atmospheric correction, cloud removal and shadow processing. The study area includes mountainous regions, has a topographic correction been applied?

Response: Following your comments, we have added "The spatial resolution of the remote sensing data is 30 meters, and the data level is L1T, which has gone through geometric precision correction. Therefore, it is not necessary to use the ground control points or digital elevation model (DEM) data to do geometric precision correction again. In MATLAB software, we used a homomorphic filtering method to weaken the thin cloud in the remote sensing imageries. In addition, because of the sun's altitude angle, some remote sensing image will appear having mountain shadow, so we used a ratio method to eliminate it in ENVI software. Furthermore, we also carried out topographic correction because the study region is a mountainous area. Then the region of interest (ROI) method was used to cut the remote sensing imagery data, removed the overlapped parts, and then assembled the multiple imageries to form into a single remote sensing imagery for the region." to the manuscript. L207-216

5. What R package have been used to perform the statistical analysis? ‘demo’ package has not been found in the R repositories (L288).

Response: We have revised it to “gbm”. L310

6. The cross-validation method usually underestimates the error. Why the cross-validation method has been selected instead of a validation based on an independent dataset? The number of samples seems large enough to split them into two datasets (train and test).

Response:  In this revision, we have divided the dataset into an independent verification dataset and a training dataset, and have revised the whole manuscript accordingly. L180-183, 318-319

7. Density plots in Figure 3 are interesting to show the prediction performance of the three methods, but the plots of predicted versus observed values are of a great importance. I strongly recommend adding the plots of predicted versus observed values for the six models in Table 3 and Figure 3.

Response: Based on your comment, we have added the plots of predicted versus observed values for the six models in Figure 3. See Figure 4.

8. It would be useful to include a land use map when describing the land use types in Section 2.1. It would help the reader to understand better the study area.

Response: We have added the land use map in the manuscript. See Figure 2

 

Minor comments:

 1. L24. Please consider adding some indicators of the predictive quality of the models (R2 and RMSE for instance).

Response: We have added those indicators to the manuscript. L25

2. L61. Do you mean ‘KOMPSAT’ instead of ‘KOMPAST’?

Response: We have revised to “KOMPSAT”. L77

3. L86. ‘9 remote sensing based environmental variables’. Please be concise and indicate that Landsat-8 band reflectances and spectral indices have been used.

Response: We have added “Landsat TM green band (B_GREEN), Landsat TM red band (B_RED), Landsat TM near-infrared band (B_NIR), difference vegetation index (DVI), enhanced vegetation index (EVI), ratio vegetation index (RVI), normalized difference vegetation index (NDVI), renormalization difference vegetation index (RDVI), and soil adjusted vegetation index (SAVI)” to the manuscript. L113-117

4. Figure 1. North arrow and coordinate grids are missing.

Response: Based on your comments, we have modified Figure 1. See Figure 1.

5. L115. Please indicate in the text that soil samples were collected in 2015.

Response: We have added “in 2015” to the manuscript. L179

6. L188. Please include the accuracy of the GPS used to record the location of the soil samples.

Response: The positioning accuracy is 5m. We have added it to the manuscript. L189

7. Equation 3. Why ‘L’ value has been fixed to 0.5?

Response: Huete’s research [29] was conducted for a well-vegetated area, indicating that when L is 0.5, SAVI has a better effect on eliminating soil reflectance. L233-234

Huete, A. R. A soil-adjusted vegetation index (SAVI). Remote Sens. Enviro. 1988, 25(3), 295-309.

8. L169. Typo. ‘from’.

Response: We have changed “form” to “from”. L256

9. L205. ‘in a complex’.

Response: “a” have been added to the manuscript. L307

10. L216. Please remove the period before ‘have’.

Response: The period had been removed according to your comment. L324

11. Table 2. Please indicate the variable that is showed in this table. Is it the Pearson correlation coefficient? ‘Relationships’ is ambiguous.

Response: There is a Pearson correlation coefficient, and we have revised in the manuscript. See Table 2

12. Figures 5 and 6. ‘a’ is missing.

Response: We have added the missing “a” to Figures 7 and 8.

13. Table 4. Do these summary statistics correspond to the area within the black rectangle in Figures 5 and 6 or to the entire Liaoning Province?

Response: Table 4 is summary statistics corresponding to the area within the forest topsoil (0-30 cm).  We have revised it in the manuscript. See Table 4

14. L392. Typo. ‘MAE’.

Response: We have revised to “MAE”. L532

 

We thank the reviewer 2 constructive comments, which significantly help improve our manuscript.

 

Reviewer 3 Report

This is an interesting study presenting the utilization of remote sensing data in the regression of soil organic carbon stocks and soil total nitrogen stocks in an extensive forested region in China. The study isn’t very novel but it is very interesting and informative and could be useful for the readers of the journal. Generally, it is well written but some improvements are needed to be suitable for publication. The main improvements needed regard the description of the methodology and especially the description of the regression models and their application in this study. It should be also clarified in the manuscript that the related environmental variables are all coming from remote sensing. An idea would be to use a phrase such as “"remotely sensed environmental variables" instead of the phrases “related environmental variables” or “remote sensing related environmental variables” etc. throughout the text. In general, the language should be improved in order to avoid confusion as there are many other confusing phrases in the manuscript. I provide detailed comments in the commented pdf manuscript file.

Based on the above I suggest the publication of the manuscript after a (successful) moderate revision.

Comments for author File: Comments.pdf

Author Response

Response to reviewer 3 comments on the manuscript remotesensing-733788 “Using remote sensing data to predict soil organic carbon and soil nitrogen stocks in Northeastern Chinese forests”

 

 

This is an interesting study presenting the utilization of remote sensing data in the regression of soil organic carbon stocks and soil total nitrogen stocks in an extensive forested region in China. The study isn’t very novel but it is very interesting and informative and could be useful for the readers of the journal. Generally, it is well written but some improvements are needed to be suitable for publication. The main improvements needed regard the description of the methodology and especially the description of the regression models and their application in this study. It should be also clarified in the manuscript that the related environmental variables are all coming from remote sensing. An idea would be to use a phrase such as “"remotely sensed environmental variables" instead of the phrases “related environmental variables” or “remote sensing related environmental variables” etc. throughout the text. In general, the language should be improved in order to avoid confusion as there are many other confusing phrases in the manuscript. I provide detailed comments in the commented pdf manuscript file.

 

Based on the above I suggest the publication of the manuscript after a (successful) moderate revision.

Response: We appreciate your help and your patience. With this submission, we provided a version (Track Changes) of the revised manuscript. Below we detail how we have revised the paper following your comments.

1. L18 “9 related environment variables” This phrase creates confusion as only remote sensing data were used. I believe that it would be clearer if you used another expression such as "remotely sensed environmental variables" or something similar that would make clear that you used only remote sensing data. Please correct this throughout the manuscript.

Response: Following your comments, we changed “9 related environment variables” to “9 remotely-sensed environmental variables”. L22

2. L21 “remote sensing related” change to remotely sensed

Response: “remote sensing related” have been changed to “remotely-sensed”. L27

3. L24 delete “related”

Response: Based on your comment, we have deleted “related”. L33

 4. L40 “However” change to “Furthermore”

Response: We have changed “However” to “Furthermore”. L50

5.L49 “simulate” change to “estimate”.

Response: “simulate” had been revised to “estimate”. L59

6. L66 resource???

Response: We eliminated “resource” to avoid ambiguity. L88

7. L67-68. “It was successfully launched by Atlas V rocket on February 11, 2013 at Vandenberg Air Force base, California, initially known as "Landsat data continuity mission.” Is this information needed?

Response: We have deleted unnecessary information. L88-92

8. L71 “spatial surface data of the earth's land surface” please rephrase.

Response: We have eliminated inaccurate statements. L96

9. L71-74 Please check the language. Confusing.

Response: In this revision, we have changed it to "Remote sensing data have been used to predict SOC and STN stocks with DSM. Furthermore, the topsoil SOC and STN stocks in the natural environment proved to have a good correlation with the topsoil biomass". L92-94

10. L92 “was” change to “is”.

Response: “was” was revised to “is”. L125

11. L116-117 Please correct this phrase.

Response: We have revised to “100 cm³ of undisturbed soil cores was collected from topsoil layer for subsequent laboratory determination of soil bulk density”. L186-187

12. L122 “the SOC and STN stocks of calculation formula were as follows:” rephrase

Response: We have revised “the SOC and STN stocks of calculation formula were as follows:” to “SOC and STN stocks were calculated using the following formula:”. L194

13. L124 This isn't clear. Please try to explain it better. Explain also what i and k represent and provide the units.

Response: According to your comment, we have revised those sentences. L198-202

14. L125 “density of SOC and STN” explain what this density represents and the units.

Response: We have added the missing information. L198

15. L170 section 2.3 Clearer descriptions and some more specific information about how these methods were used in this study are needed as this is the main focus of this study.

Response: According to your comments, we have added those parts to the manuscript. L258-315

16. L182 deleted “(Foster et al., 1986)”

Response: We have delated “(Foster et al., 1986)” according to your comment. L270

17. L213 LCCC is more appropriate

Response: We have revised “LUCC” to “LCCC”. L322

18. L233 Table 1 “degree celsius” ???

Response: Thank you for your patience and support. This was an error. We have revised it to "digital number". Table 1

19. L259 “mapped” change to “created”

Response: “mapped” had been changed to “created”. L377

20. L273 Density of what.???

Response: This is “Density plots of the predicted and measured values of topsoil (0-30cm) soil organic carbon stocks (SOC) (kg m^-2) and soil total nitrogen (STN) (kg m^-2). The predicted data are derived from geographically weighted regression (GWR) (a,d), multiple stepwise linear regression (MSLR) (b,e), and boosted regression trees (BRT) (c,f).” L394-397

21. Figure 4 please explain the graph, i.e. what the angles and radius represent.

Response: We have added these explanations to the manuscript. L409-410

22. L302 “GWR, MLSR and BRT” Previously you mentioned that you selected BRT for the analysis.

Response: In order to avoid ambiguity, we excluded GWR and MLSR. L429

23. L321 “indicating remote sensing data is critical to predicting SOC and STN stocks” But you only used remote sensing data, with what are they compared and found to be better?

Response: We have eliminated inaccurate statements. L455

24. L324 images in general??

Response: We have changed “images” to “data”. L458

25. L338 “cubist,” model or what??

Response: “cubist” is a model. We have added “cubist model” to the manuscript. L472

26. L355-356 “and increases with the increase of vegetation. Its value increased with the increase of vegetation density.” please rephrase

Response: We have eliminated inaccurate statements. L491-492

27. L358 cubist kriging?

Response: Cubist model is a decision tree model. We have added “cubist model” to the manuscript. L472

28. Revise the language of the whole manuscript.

Response: We have revised the manuscript thoroughly.

 

We thank the reviewer 3 constructive comments, which significantly help improve our manuscript.

 

Round 2

Reviewer 2 Report

The manuscript has been revised properly and most of the issues raised in round #1 have been corrected or clarified. However, I should like to mention a major issue that cause me a great concern. According to the newly added information in Table 1, the units of the Landsat-8 images are digital numbers. Digital numbers do not have a physical meaning and they are not appropriate for quantitative modelling of soil properties. The same digital number could correspond to different radiances and reflectances in the different images you combined to generate the mosaic. Thus, the Landsat L1T product you have chosen is not suitable for this analysis. The most adequate remote sensing variable to use in the SOC and STN models is the BOA (Botton-Of-Atmosphere) reflectance. Therefore, I strongly recommend converting digital numbers to radiances and then performing an atmospheric correction to obtain the BOA reflectances. The effect of the atmosphere is not negligible in the blue and green bands and should be corrected.

The study area is a mountainous region and a topographic correction has been applied to the images (L171). What topographic correction method has been applied? Please specify.

Two variables are represented in each graph of Figure 6, one using the length of the radius and other using the central angle of the sector. The meaning of the length of the radius is clear, but what is the meaning of the central angles?

Minor comments:

L17-L18. Abstract. I recommend adding the mean and standard deviation of SOC and STN in the soil samples. This would help to understand the relevance of the RMSE values obtained in the models included in L20.

L20. Units are missing in both RMSE values.

L48. Do you mean ‘region’ instead of ‘regional’?

L57. Please remove ‘and nitrification’.

L65. I suggest using ‘estimate’ instead of ‘simulate’.

L64-L75. ‘the eighth satellite of the 74 United States Land Satellite Program (Landsat)’. This clarification does not add any relevant information, please remove it.

Figure 1. ‘km’ in lowercase according the International System of Units (SI). The size of the labels of the coordinate grid is too small and they can not be clearly seen in the printed version of the manuscript. 0’ and 0’’ labels are not necessary.

L122. Please add a period after ‘activities’.

L123. Please add a period after ‘content’.

L133. ‘slope aspect’. Do you mean ‘aspect’?

L143. Please consider using ‘remaining’ instead of ‘rest’.

L163, L168 and L173. Do you mean ‘images’ instead of ‘imageries’?

L164. ‘single remote sensing imagery for the region’ Do you mean a mosaic of the region?

L303. The table number is wrong. Please replace ‘Table 2’ with ‘Table 3’.

Table 4. Please add the number of samples (n) in this table.

L368. Wrong figure number. ‘Figure 8’ instead of ‘Figure 9’.

L442. Please remove the period after ‘since’.