Spatiotemporal Changes of Soil Salinization in the Yellow River Delta of China from 2015 to 2019

Round 1

Reviewer 1 Report

Dear authors

This new version of the manuscript has improved, good job. In any case,  I still do not believe that comparing two single images acquired with a time lap of 4 years may allow to infer the spatiotemporal trend of any parameter. You defined a method with its accuracy level, hence, why does not apply it on a longer series of OLI data?

Regards

Author Response

Point 1: This new version of the manuscript has improved, good job. In any case, I still do not believe that comparing two single images acquired with a time lap of 4 years may allow to infer the spatiotemporal trend of any parameter. You defined a method with its accuracy level, hence, why does not apply it on a longer series of OLI data?

 

Response 1: Thank you for your comment and constructive suggestions. For parameter referencing, using a remote sensing method is a smart way to obtain their spatial-temporal distribution because annual sampling and in-situ monitoring are time- and money-consuming and even unsustainable in some regions with a harsh environment. Usually, we cannot obtain ground sampling data support and only have historical images. Thus, finding a suitable model for parameter referencing using the valuable sampling data and the image together is a key issue. In this study, two images in the spring season (high salinization level and low vegetation cover) were selected with the same period of ground sampling data in 2015 and 2019. Thus, it is feasible to find the model and obtain the spatio-temporal distribution of salinization. If this model works well in this region, it can be referenced by other users in different regions of the world for soil salinization. We have added this to the Discussion section. Meanwhile, we will use this model for long-term monitoring in this coastal zone region in the next step. It was added in the last part of the Conclusion section.

We also revised the paper in whole, enhanced the method and discussion section deeply, and increase the language quality.

Reviewer 2 Report

Dear authors,

Thanks for interesting paper. I have few suggestion to make paper better readable .

Please describe in more details Feature space theory and how this Feature Space Theory was Applied. (line 153). In the next text you are working with this theory, but here some description is needed.

Please describe in more details Albedo-MSAVI feature space, Salinity Indices-Surface albedo(SI-Albedo) feature space, SI-NDVI feature space or probably move text starting from line of 180 in front of this feature space description. It will make text better redable.

Please describe The Jenks Natural Breaks Classification method.

What do you think about relatively high difference in accuracy for SI NDVI between 2014 and 2018.

Author Response

Thanks for interesting paper. I have few suggestions to make paper better readable.

 

Point 1: Please describe in more details Feature space theory and how this Feature Space Theory was Applied. (line 153). In the next text, you are working with this theory, but here some description is needed.

 

Response 1: Thank you for your suggestion. We have supplemented the description of feature space theory in the Methods section. The spectral feature space is a spatial system composed of two or more bands or spectral indicators to measure the spectral characteristics of ground objects in remote sensing images. The pixel points are distributed in a certain geometric shape and the ground objects with the same attributes form a relatively clustered point group, while the ground objects with different attributes are separated and distributed in different positions in the feature space because of different spectral characteristics. Thus, different land surface parameters can be effectively separated using feature space.

 

Point 2: Please describe in more details Albedo-MSAVI feature space, Salinity Indices-Surface albedo (SI-Albedo) feature space, SI-NDVI feature space or probably move text starting from line of 180 in front of this feature space description. It will make text better readable.

 

Response 2: Thank you for your suggestion. We have moved line 180 to line 138 to make it more readable. Furthermore, we expanded the discussion on the advantages and disadvantages of Albedo-MSAVI, SI-Albedo, and SI-NDVI in the Discussion section.

 

Point 3: Please describe The Jenks Natural Breaks Classification method.

 

Response 3: Thank you for your suggestion. We added the description of the Jenk natural breakpoint grading method at the beginning of the Result section. The Jenk natural breakpoint classification method is based on the statistical Jenk optimization method. This method is a self-similar clustering method for a one-dimensional data series. According to the intrinsic statistical characteristics of data samples, this method calculates the relatively large jumps between arrays under the condition of a specified number of categories. On this basis, the data with the greatest similarity is divided into a group (category) and the data samples with the greatest difference are divided into different categories.

 

Point 4: What do you think about relatively high difference in accuracy for SI-NDVI between 2015 and 2019.

 

Response 4: Thank you for your question. SI-NDVI reflects the interaction between salinity and vegetation. High vegetation coverage will affect the accuracy of soil salinity inversion in this region. Although the minimum vegetation coverage season in spring is selected in this study, there is still vegetation or crop growing here in the coastal region. However, vegetation coverage is different in different years due to the influence of local climate. Compared with the two periods of images, the ground vegetation growth in 2015 was denser and the vegetation coverage was larger, resulting in relatively lower accuracy in 2015. This also indicates that the SI-NDVI model is not suitable for coastal areas with large vegetation growing regions. Inversely, this model is more suitable for arid or semi-arid areas with sparse vegetation coverage. This point is highlighted in the Discussion section.

Reviewer 3 Report

This is an interesting manuscript presenting the assessment of salinization spatial distribution using remote sensing techniques and field data. The methodology used isn’t really new and in general the novelty of the manuscript isn’t its strongest advantage; however, the presentation of the implementation of the methodology, the comparison between the obtained results with the various versions of the feature space method and the characteristics of the studied region make the paper interesting for the readers.

However, the manuscript in its present form has some important problems. A key problem is the language that should be improved. Some parts are a little confusing and there are some strange wordings and phrases.

Another problem is that the results and the discussion of the manuscript, as it is right now, focus mainly on the spatial variability of salinity and its change in the study region and the measures that should be taken to improve the situation. The performance of the methodology, the reasons behind this performance, the comparison with other alternative approaches etc. are very shortly presented. This makes the manuscript to seem mostly as a case study, but I believe that there is much more in it that can be presented. For example, in table 2 there is only the overall accuracy statistic. Other statistics could be also presented (please also check why there is the year 2014 instead of 2015 in table 2). Furthermore, some more analysis on the performance would be more informative. For example, why only comparisons of classes? Scatterplots of observed - predicted salinization or scatterplots of feature space distances from the reference point vs. salinity of field observations could provide more information. I would expect also that differences in classification (Table 1) could improve the accuracy. Did you try different classes? Finally, the spatial distribution of errors may provide a better insight of the reasons behind the performance.

Another example is that there is a discussion of the advantages of the followed approach but there isn’t any comparison with other methods accuracy. Accordingly, the discussion of the performance in comparison with other methods is a little vague.

The methodology should be presented a little clearer as well. Please try to improve the quality of figure 2 if possible. Check equations 5,6 and also present the distance from the reference point with D and a subscript for the corresponding space in all these equations (5,6,7). Why is there a different reference point between equations 5 and 7? In figures 3 and 4 they seem as similar cases.

Finally, I would propose to shorten any reference in the introduction and discussion to policies and suggestions that are not related to the findings of this study. I believe that emphasis should be given to measures that were taken between 2015 and 2019 that their spatial footprint can be identified in the observed changes. Also, any proposals should be associated with the findings (spatial variability) of this study.

Based on the above comments my recommendation is major revision but I believe that the study has a good potential.

Author Response

Point 1: This is an interesting manuscript presenting the assessment of salinization spatial distribution using remote sensing techniques and field data. The methodology used isn’t really new and in general the novelty of the manuscript isn’t its strongest advantage; however, the presentation of the implementation of the methodology, the comparison between the obtained results with the various versions of the feature space method and the characteristics of the studied region make the paper interesting for the readers.

 

Response 1: Thank you for this comment. Your comment will be of great benefit to our future scientific research and thesis writing.

 

Point 2: However, the manuscript in its present form has some important problems. A key problem is the language that should be improved. Some parts are a little confusing and there are some strange wordings and phrases.

 

Response 2: Thank you very much for your suggestion. We have improved the language of the article. At the same time, an English native professional editor helped us polish the manuscript.

 

Point 3: Another problem is that the results and the discussion of the manuscript, as it is right now, focus mainly on the spatial variability of salinity and its change in the study region and the measures that should be taken to improve the situation. The performance of the methodology, the reasons behind this performance, the comparison with other alternative approaches etc. are very shortly presented. This makes the manuscript to seem mostly as a case study, but I believe that there is much more in it that can be presented. For example, in table 2 there is only the overall accuracy statistic. Other statistics could be also presented (please also check why there is the year 2014 instead of 2015 in table 2).

 

Response 3: Thank you for your comments and suggestions. We have enhanced the Discussion section for the approach comparison and the reasons for their performances. Compared with supervised classification, unsupervised classification, neural network automatic extraction, and other methods, the model index of the feature space method is simple, feasible, and easy to extract. The application of two-dimensional feature space theory to develop quantitative methods and indicators of salinization remote sensing monitoring has great potential. Based on the feature space approach study, the SI-Albedo inversion model has higher accuracy compared to the dryland region studies of soil salinity using the Albedo-MSAVI feature space by Feng et al. and the soil salinity study using the NDVI-SI feature space by Wang et al. In addition, I am sorry for our negligence in writing 2015 as 2014 in Table 2; we have revised this.

 

Point 4: Furthermore, some more analysis on the performance would be more informative. For example, why only comparisons of classes? Scatterplots of observed-predicted salinization or scatterplots of feature space distances from the reference point vs. salinity of field observations could provide more information. I would expect also that differences in classification (Table 1) could improve the accuracy. Did you try different classes? Finally, the spatial distribution of errors may provide a better insight of the reasons behind the performance.

 

Response 4: Thank you for your suggestion. We use the Jenk natural discontinuity classification method to divide the data into five grades and compare the salinization grade with the measured data. According to the correct proportion of the sample data, the accuracy of the model can be determined. The classification of salinization grade in Table 1 is in line with China's national standards and is suitable for the region. According to the regional characteristics of different countries, corresponding classification indices can be formulated. We added this in the Discussion and Conclusion sections.

 

Point 5: Another example is that there is a discussion of the advantages of the followed approach but there isn’t any comparison with other methods accuracy. Accordingly, the discussion of the performance in comparison with other methods is a little vague.

 

Response 5: Thank you for your comments. We have updated the discussion on the comparison of methods in the Discussion section. This can also be seen in the response to Comment 3 above.

 

Point 6: The methodology should be presented a little clearer as well. Please try to improve the quality of figure 2 if possible. Check equations 5,6 and also present the distance from the reference point with D and a subscript for the corresponding space in all these equations (5,6,7). Why is there a different reference point between equations 5 and 7? In figures 3 and 4 they seem as similar cases.

 

Response 6: Thank you for your suggestion. We updated the introduction of the methods, including the details of the feature space theory and related equations’ description. Because the background image in figure 2 is from Google Earth, we are sorry we can’t edit it.

The distance from any point in the characteristic spatial scatter plot to a reference point gives an indication of the salinity level of the soil. Usually, the reference point is chosen to reflect the extreme values of salinity, like extreme high value or extreme low value. Due to different methods of parameter calculation, there exists diffenent empirical choices of reference point. In equation 5, the empirical reference point in the Albedo-MSAVI feature space is the extreme point with the most severe degree of salinization. In equation 7, the SI-NDVI feature space reference point is the least salinized extreme point. This is the difference of reference point between the two equations.

Because the distance D doesn’t show on the Figures, thus we delete that related text in the manuscript. The quality of Figure3 and 4 looks good, so we don’t update them again.

 

Point 7:  Finally, I would propose to shorten any reference in the introduction and discussion to policies and suggestions that are not related to the findings of this study. I believe that emphasis should be given to measures that were taken between 2015 and 2019 that their spatial footprint can be identified in the observed changes. Also, any proposals should be associated with the findings (spatial variability) of this study.

 

Response 7: Thank you for your helpful suggestions. We have shortened the policy and suggestions irrelevant to the results of this study in the Introduction and Discussion sections. More salinization control measures in local region are revealed and highlighted in the Discussion section, including the large-scale rice cultivation supported by China “Bohai Granary” project.

 

Point 8: Based on the above comments my recommendation is major revision but I believe that the study has a good potential.

 

Response 8: Thank you very much for your guidance. We also revised the paper thoroughly in whole, enhanced the method and discussion section deeply, and increase the language quality.

Round 2

Reviewer 1 Report

Thanks

Reviewer 3 Report

The revised version is adequately improved. The authors corrected some important errors and improved the clarity of the methodology and the discussion. The responses to the reviewers comments are also sufficient. 

There are still possibilities for improvement and some editorial corrections are required but I believe that all those can be done during the production phase based on the advise of the editors.

 

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

Thank you for sending this paper to me for review. The paper is useful but requires major modification to make it acceptable for publication.

1. 'Sustainability' is a broad journal, but the authors assume too much knowledge on remote sensing - in fact the main effort seems to have been in demonstrating the best method for analyzing salinization in coastal areas. Very little of the work is actually about analyzing the reasons why salinization has changed in Kenli County.  
2. There needs to be clear statement about the purpose of the paper at the end of the introduction section.
3. In the methods section there is no description of how the 47 validation sites sampled in 2019 were chosen. The validity of the conclusions about which feature space is best to use depends on having a representative dataset that adequately samples a range of landscape positions. Even less information is provided about the 2015 sample sites.
4. The authors need to explain to a non-remote sensing audience what the "feature space principle" is. They seem to be hypothetical relationships between remote sensing parameters and the land properties of interest (in this case salinization). 
5. The methods section has no statement as to how the classification was performed. What is logic that the Jenks Natural Breaks classification should yield the same classes as the Chinese Salinization classification, which are simply breaks in a continuous variable? The fact that it works in this study could simply be luck or may depend on poor choice of the sample sites. Why not simply regress D against salinity? 
6. The authors need to explain why they wish to reduce a 2D feature space to one dimension.
   1. In the case of SI-Albedo, the two parameters are so highly correlated that there is nothing to be gained by using D that could not have been done using either SI or albedo alone. What happens if the authors simply classify SI (or albedo) and compare it to the sample sites?
   2. In the cases of Albedo-MSAVI and SI-NDVI, reducing the feature space to one parameter losses whatever additional information is provided by MSAVI or NDVI that is not already provided by SI and albedo. I.e. it would have been better to use a 2D classification of the feature space, possibly after some sort of transformation to get two uncorrelated variables.
7.  On line 323 - 330 the authors try to interpret the changes in the amount of salinization in terms of a plan for sustainable development. This should have been their starting hypothesis: namely that the plan has resulted in less salinization. However, they have presented no evidence to show that the plan caused the lower salinization observed. I would have expected to see some spatial analysis of where particular land use changes recommended by the plan had been implemented in relation to areas with less/more/unchanged salinity.

in conclusion I am left uncertain whether the focus of the paper is about the use of the feature space to map salinity or about what caused the changes in salinity in Kenli County between 2015 and 2019. The paper requires considerable clarification before it can be published.  

Minor comments

• What is the area of Kenli County? (About 1800km² ? But don;t make the reader work it out).
• Whilst hm^2 could be considered a valid unit, I suggest that ha or km^2 are more familiar to most readers.
• Equations 2 and 3 for MSAVI and NDVI, respectively, require references.
• The caption for figure 3 is uninformative-The authors need to explain what they are trying to show. What are the ellipses? It is also not clear what point B represents.  
• Equations 5 and 6 seem to be incorrect. Both are missing ² on the last term (should be MSAVI² and Albedo² respectively)
• In equations 5, 6, and 7 and Table 1, the variable derived from the feature space needs to expressed differently: e.g. as D(Albedo, MSAVI). "Albedo - MSAVI" reads as Albedo minus MSAVI.
• The descriptions of the changes in Figure 6 is incorrect. I suggest:
  • Ford > Flooded or Water body
  • Unchange > Unchanged
  • Lighten > Less salinized
  • Aggravate > More salinized

Many more comments and edits are included in the attached manuscript.

Comments for author File: Comments.pdf

Reviewer 2 Report

Manuscript presented for review Spatiotemporal changes of soil salinization in the Yellow River Delta of China from 2014 to 2019 presents the way to choose the right (best) method for measuring changes of soil salinization. The work is rather methodical, so it can be classified as a technical note rather than an original article. However, this does not diminish the quality of the research. I have some comments on the manuscript.

1. In my opinion, the title of the manuscript does not quite match the text of the work. The authors test various measurement methods to select the best one. This is not reflected in the manuscript title.
2. line 40 - I propose to write 15th
3. lines 49.50.52 - I suggest using as a unit of area hectare (ha) instead of hm2
4. The manuscript lacks a clear purpose of the study / research hypothesis. In lines 92-94, the Authors write about the test method being tested rather than the purpose of the work.
5. In the methods and results chapter, add sections as recommended by the journal.
6. The introduction and discussion lack references to international literature.

Reviewer 3 Report

The paper deals with the analysis of the spatial variation of soil salinization in a subset of the Yellow River Delta by investigating two OLI-Landsat8 imagery, one acquired in 2014 and the other in 2019. To this aim the feature space method was used analyzing scatter plot relationship between different indicators: Albedo, MSAVI and NDVI. The relationship between SI and Albedo was found as the best one for the inversion of salinization in the investigated area.
The topic is quite interesting, because soil salinization is a common issue worldwide affecting land degradation process.
I have several concerns with the paper, described in the following.

1) The paper seems a sort of duplication/extension of the paper [25] where at least one of the authors of this submitted work was present. The area investigated here is a spatial subset of the one analyzed in [25], and the same sampling network of in-situ data have been used. A similar methodological approach was also used considering only at the relationship between MSAVI and SI in two different schemes, applied for two Landsat images, a TM-Landsat5 of 1987 and an OLI-Landsat8 of 2016. It is worth noting that different results have been achieved in the two papers, but in any case the level of similarity is quite high. Moreover, paper [25] discuss better the feature space model implementation, as well as all the preliminary steps. Summarizing this main point, for sure the authors should try to emphasize the difference, if existing, with the previous paper, otherwise, they should clearly state that is a further extension of that work. 

2) As above-mentioned, the level of clarity of the paper is quite poor, even because of the used English that should be carefully revised. Methods are described with a low level of details, as well as the classification scheme applied.

3) I do not believe that comparing two single images acquired with a time lap of 5 years may allow to infer the spatiotemporal trend of any parameter. The authors defined a method with its accuracy level, hence, why does not apply it on a longer series of OLI data?

4) Section 3.1 still belongs to the methods section.