An Erosion-Based Approach Using Multi-Source Remote Sensing Imagery for Grassland Restoration Patterns in a Plateau Mountainous Region, SW China
Round 1
Reviewer 1 Report
Dear Authors,
The manuscript has the potential to gain interest from a wide audience.
However, I suggest some changes and give suggestions in the revised manuscript (see attached).
One of my major concerns is some English language issues all over the text.
Please consider rephrasing some sentences, and pay much bigger attention to your captions of tables and figures, also pay attention to self-explanatory manner of the figures and tables (No unexplained abbreviations).
My final major concern is the discussion part. You need to discuss your results against other people results!
See some minor issues in the text.
Best regards,
Comments for author File: 
Comments.pdf
Author Response
List of Responses
Dear editor and reviewer:
Thank you for your letter and for the reviewer’ s first round comments concerning our manuscript (manuscript ID: remotesensing-2278468). Those comments are very valuable and helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in the paper. The main corrections in the paper and the responds to the reviewer’s comments are listed as follows.
Reviewer #1:
Comments and Suggestions for Authors：
Dear Authors,
The manuscript has the potential to gain interest from a wide audience. However, I suggest some changes and give suggestions in the revised manuscript (see attached). One of my major concerns is some English language issues all over the text. Please consider rephrasing some sentences, and pay much bigger attention to your captions of tables and figures, also pay attention to self-explanatory manner of the figures and tables (No unexplained abbreviations). My final major concern is the discussion part. You need to discuss your results against other people results!
Point 1: Please check English, I am not sure that it is the correct formula. (Line 100).
Response 1: We have deleted the sentence to avoid misunderstanding.
Point 2: No, they are not! (Line 111).
Response 2: We are very sorry for our negligence of the word ‘interdependent’ and we have made correction according to your comments. (Line 20, 21)
Point 3: Please rephrase! If it is Vegetation Restoration Potential then use it as VRP, otherwise if it should be the vegetation's Restoration Potential. I think. It is just a very basic grammar issue, I think. (Line 135).
Response 3: We sincerely apologize for our incorrect writing, and we have modified as the vegetation's Restoration Potential (RP).
Point 4: Check English, I feel like there is too many "to" in a row! (Line 141).
Response 4: We are very sorry for the incorrect use of grammar and we have corrected them.
Point 5: Fix erratic automatic separation of this word! (Line 147).
Response 5: Dear reviewer, the automatic separation of words is actually caused by the word template (required format) provided by the journal, and we have tried to avoid it.
Point 6: What’s grassland potential? (Lines 149-150).
Response 6: Again, we are very sorry for the inappropriate use of “grassland potential” and we have replaced it with ‘grassland restoration potential’.
Point 7: Any information about the soils? The article is about soil erosion!!! (Line 163).
Response 7: Thank you for your comments, we are sorry about the insufficient description of soils of the study area. And words like “red soil” mentioned in the manuscript refers to the “Acrisols” in the FAO classification. Since the content of the paper is mostly focused on soil erosion-based restoration, so we didn’t mention much about soil, but in the calculation of soil erodibility, we do use soil information such as silt, sand and clay percentage in soils, as well as soil organic matter content (%) just as the USLE equation did.
Point 8: It would be nice to know a little bit more about the field investigation units on the map! (Line 165).
Response 8: Considering your valuable suggestion, we have added a map describe the field investigated primary sample units (PSUs) and some sentences about other basic information on PSU. Actually, the Ministry of Water Resources of China (MWRC) conducted the Fourth National Soil Erosion Survey by using unequal probability sampling methods and the Chinese Soil Loss Equation (CSLE) model. It is the first ever and only national survey to incorporate sampling survey, field investigation, model-based calculation and latest spatial information techniques in China’s history to quantify soil erosion on site. Sampling investigation is the main method of the survey, and a total of 33,966 primary sample units (PSUs) were determined and allocated in the field with sampling densities of 4%, 1%, 0.25% and 0.0625%. Each PSU is a 1 km × 1km grid in the plain area or a small watershed with area from 0.2 to 3.0 km2 in the mountainous area, which was selected for investigation. Thousands of data gathered in the country were instructed to visit every land parcel (defined as land with same land use and conservation practice) in each PSU and collect indicators and information of soil erosion in the field. The data collected concerning the erosion affecting factors share uniform standards and specifications and was well-examined in every step of the survey. After a comprehensive analysis of rainfall, soil, terrain, land use and soil conservation practice, the soil erosion rate was computed by CSLE model with a spatial resolution of 10 m for each land parcel, then statistical methods were applied to evaluate the distribution, area and intensity of soil erosion at PSU, county, provincial and national levels. Yunnan, as one of the mountainous provinces suffering from severe soil erosion, received a lot of attention in the national survey for its complex erosion conditions and irreplaceable ecological value. Based on the PSUs with grassland parcels, information such as FVC of every 15 days, rotations, terraces and grassland type were also investigated onsite. For different elevation zones and grassland types, we generated FVC dynamics curves across the year, which is helpful to fulfill the absence of image data caused by cloudy and rainy weather, as well as the complex terrain condition in our study area.
Point 9: Why past tense? Is it not true any more? (Line 172).
Response 9: We have corrected this tense issue and we apologize for the grammar mistake.
Point 10: I would rather say range of annual ... as 11 and 21 degrees celsius is way too wide to be a single annual average. Or, at least, say that the average is 11 in the north and 21 in the south, or something like that! (Line 174).
Response 10: We have revised the part based on your suggestion.
Point 11: Same her, as in case of the temperature. (Line 175).
Response 11: We have revised the part based on your suggestion.
Point 12: Zhaotong what? Zhaotong municipality, region, subregion or. . . ???? (Line 185).
Response 12: Zhaotong is actually a municipality unit and thank you for your suggestion.
Point 13: I suggest rephrasing the caption as it is not only but . . . e.g. there is also the Yangtze River not only the Jinsha River! (Line 185).
Response 13: We have revised this part according to the Reviewer’s suggestion in the Figure.
Point 14: were? seems like the sentence is lacking some part! (Line 195).
Response 14: We are very sorry for our incorrect writing and we have corrected it.
Point 15: you listed 6 types of land cover and 7 figures! What is missing here? (Lines 232-233).
Response 15: We apologize for the ambiguity created here and have made changes.
Point 16: (Line 281).
Response 16: Dear reviewer, like the previous suggestion, thank you for your suggestion, the automatic separation of words is actually caused by the word template (required format) provided by the journal, and we have tried to avoid it.
Point 17: (Line 284).
Response 17: Same as Point 16.
Point 18: Please rephrase! (Line 322).
Response 18: We have rephrased this sentence according to your suggestion.
Point 19: Please consider citing the soil classification system! What is yellow and red soil? It needs to be detailed a little bit more. (Lines 350-351).
Response 19: We are very sorry for our incorrect quoting of “yellow and red soil ”. We added the sentence “The major soil groups (FAO/UNESCO classifications) in the study area are Acrisols, Cam-bisols, Luvisols and Alisols” in the Study Area section and made some revision here too.
Point 20: a detailed. . . I think (Fig 5e).
Response 20: We appreciate your suggestion and have corrected the error on Figure 5.
Point 21: How did hm2 became km2??? (Fig 5f).
Response 21: The difference between these two units is actually 100 times, since in the field of soil erosion square kilometers is an prevalent unit, and 500t/km2·a is actually an important thresholds value of soil erosion in this region, which is proposed by the Ministry of Water Resources of China (MWRC) for long-term sustainable soil productivity. So we made some conversion in the manuscript.
Point 22: Please, rephrase! (Line 404).
Response 22: With your suggestion, we have rephrased the sentence.
Point 23: consider some rephrasing here as well, e.g. grasslands or grassland areas . . . or . . . (Line 412).
Response 23: The sentence has been rephrased thanks to your suggestion.
Point 24: all figures and tables must be understandable as stand-alone features, so I would suggest explaining RP as you did on Fig. 7! (Lines 474-475).
Response 24: We are sorry about the insufficient description of RP and we have added some sentences here.
Point 25: In the discussion part you should discuss your results with other scientists' results! (Line 504).
Response 25: We have reformulated the Discussion section based on your suggestion and other reviewers.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Dear authors,
Please find attached my comments in the word doc.
Best wishes,
Comments for author File: Comments.pdf
Author Response
List of Responses
Dear editor and reviewer:
Thank you for your letter and for the reviewer’ s first round comments concerning our manuscript (manuscript ID: remotesensing-2278468). Those comments are very valuable and helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in the paper. The main corrections in the paper and the responds to the reviewer’s comments are listed as follows.
Reviewer #2:
This study brings a fair contribution to the literature as they are many examples in the literature on grassland monitoring and restoration via remote sensing. The study proposes a framework (however this is missing) for restoring grassland ecosystems in mountainous regions, which takes into account both above-ground biomass and soil productivity. The study focuses on grasslands in southwestern China and uses indicators such as soil erosion conditions, net primary productivity, and regrowth rate to assess restoration possibilities. Results show that there is a need for urgent action to address grassland degradation and restore its sustainability.
The grammar and spelling errors require attention. Please consider using Grammarly or any other platform to quickly detect the errors.
Point 1: The sentence is too long, and it does not make any sense, please make it succinct and clear. “Above-ground biomass soils information/data” you might consider adding. (Lines 13-16).
Response 1: Thank you for your suggestion, we have rephrased the sentence.
Point 2: Too long sentence. I recommend a full stop after you provide the information on accuracy. (Lines 22-25).
Response 2: We appreciate your suggestion and have rephrased the sentence.
Point 3: Please rephase the sentence between this lines, especially People want everything from grassland….. (Lines 71-73).
Response 3: We are very sorry for the inappropriate use of “People want everything from grassland” in the manuscript and considering the Reviewer’s suggestion, we have reformulated the sentence..
Point 4: I would add between brackets Topography parameters as these are also important for understanding grassland degradation. (Line 80).
Response 4: we have made correction according to the Reviewer’s comments and added topographic characteristics here.
Point 5: What do you mean by as far as I know remote sensing is a key data for mapping grassland, or conventional remote sensing methods however, the challenge is the terminology and how we actually understanding the types of grassland.
In addition, the cloudy and rainy weather, results in poor remote sensing data availability. From my point of view data availability is not really an issue anymore, if you take into account all satellites launched and the amount of data. Indeed, rainy weather is an obstacle, but we can use SAR satellite data. (Line 94).
Response 5: We are very sorry for the inappropriate use of “conventional remote sensing methods” in the manuscript and we didn’t explain its meaning earlier, actually we want to point out the drawbacks of traditional optical and medium-low spatial resolution images. And we also used SAR satellite data in our work, but still it has its shortage in this region, and not all the indicators and parameters about soil erosion can we acquired only through remote sensing, especially information like the complex soil conservation measures. So we didn’t describe the grassland species or types much in the area, but we did investigated the grassland types, which can be found in Figure 5(e).
Point 6: The sentence is too long having a mixture of thoughts, please better structure it. In plus some other of the commonly used vegetation indices for NPP estimation are: Enhanced Vegetation Index (EVI) Soil-Adjusted Vegetation Index (SAVI) Fraction of Photosynthetically Active Radiation (FPAR) Leaf Area Index (LAI). Use them in the text and a suitable reference after each.
General comment on the introduction- above the comments above I would suggest reflecting on the grassland types in your region. Indeed, you provided some types between lines 153-155, but those are more linked to the quality of the grassland.
Also, there is a strong link between soil erosion and grassland degradation (as you stated) but reflect more on soil erosion condition and the remote sensing capabilities on this topic. (Lines 143-150).
Response 6: We sincerely appreciate the valuable comments. We have rephrased the sentences, checked the literatures（quoted relevant bibliographical references）carefully and added more references on the commonly used vegetation index for NPP estimation into the Introduction part in the revised manuscript, as well as the soil erosion condition and remote sensing capabilities.
Point 7: Yunnan Province owns the largest proportion of natural grassland in southwest China. Could you please provide the exact size, ideally in percentage? (Line 164).
Response 7: We have added the exact size and percentage of grassland in the Study Area section.
Point 8: There are also other LULC products, such as Dynamic World (https://dynamicworld.app/) and ESA Land Cover 2020 and 2021 https://esa-worldcover.org/en ESRI Land cover (https://livingatlas.arcgis.com/landcover/). Why not using and considering better spatial resolution products? (Lines 187-198)
Response 8: In our previous research of ‘identification of rubber plantation’ (doi.org/10.3390/rs15051228) published on Remote Sensing, we did actually consider those three LULC products with better spatial resolution and achieved good performance. The reason why we didn’t choose here in this work is because the significant differences in the definition of grasslands among them. As a result, we found great differences and uncertainties in the distribution and area of grasslands they provided. The LULC products we used for sample selection in this study are basically from China and they share similar or same definitions of grasslands.
Point 9: 3. Results
All graphs and maps need a slightly better resolution. When you create the final draft you can do some settings in Word to not alter the image quality (as below)
Response 9: Thank you for your suggestion, we have replaced the images with better resolution.
Point 10: About 3.4. Grassland Zoning and Grassland Restoration Pattern
The overlay analysis and all classes must be explained in a separate subsection 2.4. part of Material and Methods and 3.4 should only cover the results of applying this method. Also, in the subsection 2.4. you need to support the classifications (tolerate/unsuitable, and the thresholds for restoration potential and regrowth rate) made in Table 7 with the literature.
Even though I agree with the methodology used it has intrigued me the accuracy rates obtained. This is because it has been used different spatial resolution products and when resampling is performed the accuracy decrease. Will be the coding and data analysis available as a FAIR procedure?
Response 10: We have reformulated this part according to your suggestion in those sections and quoted with references. As for the accuracy issue, we want to make some explanations here. Considering your valuable suggestion, we have added a map describe the field investigated primary sample units (PSUs) and some sentences about other basic information on PSU. Actually, the Ministry of Water Resources of China (MWRC) conducted the Fourth National Soil Erosion Survey by using unequal probability sampling methods and the Chinese Soil Loss Equation (CSLE) model. It is the first ever and only national survey to incorporate sampling survey, field investigation, model-based calculation and latest spatial information techniques in China’s history to quantify soil erosion on site. Sampling investigation is the main method of the survey, and a total of 33,966 primary sample units (PSUs) were determined and allocated in the field with sampling densities of 4%, 1%, 0.25% and 0.0625%. Each PSU is a 1 km × 1km grid in the plain area or a small watershed with area from 0.2 to 3.0 km2 in the mountainous area, which was selected for investigation. Thousands of data gathered in the country were instructed to visit every land parcel (defined as land with same land use and conservation practice) in each PSU and collect indicators and information of soil erosion in the field. The data collected concerning the erosion affecting factors share uniform standards and specifications and was well-examined in every step of the survey (So we didn’t mention the data quality analysis). After a comprehensive analysis of rainfall, soil, terrain, land use and soil conservation practice, the soil erosion rate was computed by CSLE model with a spatial resolution of 10 m for each land parcel, then statistical methods were applied to evaluate the distribution, area and intensity of soil erosion at PSU, county, provincial and national levels. Yunnan, as one of the mountainous provinces suffering from severe soil erosion, received a lot of attention in the national survey for its complex erosion conditions and irreplaceable ecological value. Based on the PSUs with grassland parcels, information such as FVC of every 15 days, rotations, terraces and grassland type were also investigated onsite. For different elevation zones and grassland types, we generated FVC dynamics curves across the year, which is helpful to fulfill the absence of image data caused by cloudy and rainy weather, as well as the complex terrain condition in our study area.
As you mentioned different spatial resolution products have been used and when resampling is performed the accuracy decrease. We are very sorry for our negligence of analyzing the scale issue of different parameters, as well as the knowledge gap between data gathers, we did consider the spatial heterogeneity so we improve the resolution similar to the model requirements for most indicators. However, the fact is that since most of the soil erosion models are based on a large number of point or slope observation data, when applied to the regional scale or larger scale, the establishment conditions of the model, the characteristics of the applicable object, the requirements of the input variables or parameters, etc., have to be simplified accordingly, and the reliability of the results is often questioned. At present, the contradiction between the relatively low resolution of available data and the high resolution of runoff erosion process is the biggest difficulty and challenge in the quantitative study of regional soil erosion for mountainous regions, so we discussed the uncertainties in the Discussion part which we hope can meet your approval.
Point 11: 4. Discussion
In this section is missing the framework. You have explained and showed in this paper a way for grassland monitoring and restoration via remote sensing, and even the title of the paper states framework. Yet there is no framework presented and explained in the study. Framework In general are important for policy makers and serves as guidelines for key stakeholders (e.g. ecologist, land owners, conservation managers, etc).
These are the key sentences you need to expand and elaborate.
The quantitative restoration guide is of the utmost importance for 521 policy makers. However, any results estimated by the model will have certain uncertainty.
Response 11: We are so sorry for our incorrect quoting of the concept “Framework”, so we changed the word into “Approach” in both the title and the context, we have also re-written the Discussion part according to your suggestion and other reviewers.
Author Response File: Author Response.pdf
Reviewer 3 Report
1. In the method section, provide a table describing the data collected in the field. This table also may include the important statistics related to the ground truth data and labels.
2. Random Forest; I highly recommend that you provide the hyper parameters used for training random forest model. And, describe the important features (independent variables) selected by random forest based on feature selection product of random forest.
3. The treatments in Table 8. How did you come up with these treatments? Discuss.
4. The discussion section lacks important information. It is recommended that the author discuss the findings offered in the results section. The discussion section is not acceptable and needs much more work. For example. Why does Fig.8 (a) show a high inter-annual variation in RP? How about intra-annual variation? Do you consider that? Discuss all figures and tables.
5. Discuss the uncertainties of the model and remote sensing products that you have you used in this research.
6. Discuss how this research could contribute future works and benefit other research studies nationwide and worldwide.
Author Response
List of Responses
Dear editor and reviewer:
Thank you for your letter and for the reviewer’ s first round comments concerning our manuscript (manuscript ID: remotesensing-2278468). Those comments are very valuable and helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in the paper. The main corrections in the paper and the responds to the reviewer’s comments are listed as follows.
Reviewer #3:
Point 1: In the method section, provide a table describing the data collected in the field. This table also may include the important statistics related to the ground truth data and labels.
Response 1: Considering your valuable suggestion, we have added a map describe the field investigated primary sample units (PSUs) and some sentences about other basic information on PSUs. Actually, the Ministry of Water Resources of China (MWRC) conducted the Fourth National Soil Erosion Survey by using unequal probability sampling methods and the Chinese Soil Loss Equation (CSLE) model. It is the first ever and only national survey to incorporate sampling survey, field investigation, model-based calculation and latest spatial information techniques in China’s history to quantify soil erosion on site. Sampling investigation is the main method of the survey, and a total of 33,966 primary sample units (PSUs) were determined and allocated in the field with sampling densities of 4%, 1%, 0.25% and 0.0625%. Each PSU is a 1 km × 1km grid in the plain area or a small watershed with area from 0.2 to 3.0 km2 in the mountainous area, which was selected for investigation. Thousands of data gathered in the country were instructed to visit every land parcel (defined as land with same land use and conservation practice) in each PSU and collect indicators and information of soil erosion in the field. The data collected concerning the erosion affecting factors share uniform standards and specifications and was well-examined in every step of the survey. After a comprehensive analysis of rainfall, soil, terrain, land use and soil conservation practice, the soil erosion rate was computed by CSLE model with a spatial resolution of 10 m for each land parcel, then statistical methods were applied to evaluate the distribution, area and intensity of soil erosion at PSU, county, provincial and national levels. Yunnan, as one of the mountainous provinces suffering from severe soil erosion, received a lot of attention in the national survey for its complex erosion conditions and irreplaceable ecological value. Based on the PSUs with grassland parcels, information such as FVC of every 15 days, rotations, terraces and grassland type were also investigated onsite. For different elevation zones and grassland types, we generated FVC dynamics curves across the year, which is helpful to fulfill the absence of image data caused by cloudy and rainy weather, as well as the complex terrain condition in our study area.
Point 2: Random Forest; I highly recommend that you provide the hyper parameters used for training random forest model. And, describe the important features (independent variables) selected by random forest based on feature selection product of random forest.
Response 2: Thank you for your suggestion and we made some revision in this part. The hyper parameters used to train the Random Forest model in the process of grassland classification and recognition in Zhaotong include numberOfTrees, variablesPerSplit, minLeafPopulation, bagFraction, maxNodes and seed. The other parameters in the experiment remain the default setting. The numberOfTrees start number is set to 10, the termination number is set to 200, and the loop iteration is performed with 10 as the step size. When numberOfTrees = 150, the classification accuracy reaches the maximum, so the optimal value of the decision tree is set to 150. In the study, various characteristic variables such as spectrum, vegetation index, texture, vegetation coverage, terrain (elevation, slope and aspect) were constructed. According to the characteristics of the random forest algorithm, the feature importance is sorted. Finally, VV, VH, Blue, Green, Red, NIR, SWIR, NDVI, EVI, FVC, Brightness, Greenness and Humidity were used as the input feature variables for the final classification. The detailed information about features can be found in our previous work published in Remote Sensing (“Identifying grassland distribution in a mountainous region in southwestern China using multi–source remote sensing images”, doi.org/10.3390/rs14061472), but the major difference here is we used a lot of field investigation data to generate monthly grassland distribution information.
Point 3: The treatments in Table 8. How did you come up with these treatments? Discuss.
Response 3: We are sorry about the insufficient description of the treatments in Table 8. Zhaotong lies in the upper Yangtze River, which is recognized as a National Key Management Area of Soil Erosion. The land degradation restoration practice in this area is basically started from the perspective of soil erosion. We actually referred to several local research articles and had a certain literature source and combination for the selection of each threshold. However, there are still some problems in this integrated zoning approach, and we discussed them in the Discussion section.
Point 4: The discussion section lacks important information. It is recommended that the author discuss the findings offered in the results section. The discussion section is not acceptable and needs much more work. For example. Why does Fig.8 (a) show a high inter-annual variation in RP? How about intra-annual variation? Do you consider that? Discuss all figures and tables.
Response 4: Thank you for your comments, we are sorry about the insufficient description of this part. We deleted Figure 8a Inter-annual grassland RP trend for all grasslands because our focus is on analyzing the trends and differences in restoration potential of four typical grasslands. We found that the overall trend of the restoration potential of the four types of typical grasslands showed a large fluctuation. The underlying reasons include climate change, such as the frequent occurrence of extreme precipitation and drought, the inconsistent protection focus of the decision-making department each year, and the continuous change of the scale of grazing. The reason why we only the discussed inter-annual variation in RP is that the data availability of long-term time series data at a monthly scale is insufficient. And the fluctuation of single-year grassland restoration potential is too large, so we choose the multi-year average as a reference. We have also rewritten the Discussion part according to your suggestion and other reviewers’.
Point 5: Discuss the uncertainties of the model and remote sensing products that you have you used in this research.
Response 5: We have reformulated this part according to your suggestion.
Point 6: Discuss how this research could contribute future works and benefit other research studies nationwide and worldwide.
Response 6: Thank you for your comments, we are sorry about the insufficient description of potential. We have revised this part according to your suggestion.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
No further comments. Happy to get published.
Best wishes, A.
