Spatio-Temporal Pattern of Land Degradation along the China-Mongolia Railway (Mongolia)
, Haishuo Wei 1,3, Kai Cheng 1,4, Ge Li 1,3, Altansukh Ochir 5, Lingling Bian 1,3, Davaadorj Davaasuren 6, Sonomdagva Chonokhuu 5 and Elbegjargal Nasanbat 7Round 1
Reviewer 1 Report
General Comments:
Title of the paper can be improved as paper is discussing spatial distribution of land degradation patterns.
Abbreviations of the terms should be consistence throughout the paper.
This paper is using GIS and remote sensing as a tool for the mapping of land degradation patterns however GIS analysis can be improved to produce better results.
Object oriented classification method is based on the spectral reflectence of pixels of an image, please explain in the paper how you have done the mapping of the land gradation patters and what is the accuracy and precision level?
Specific comments:
Line 23: mention the name of the images along spatial resolution.
Line 43: Please explain what is ‘Belt and Road initiative’.
Line 59: Write full name of NOAA and MODIS.
Line 77: replace the word ‘advanced’ with ‘higher’.
Line 100” Add detail of ‘forests with dominant plants’.
Line 105: write full name of Landsat TM and Landsat OLI images.
Line 107: write Digital Elevation Model (DEM).
Line106: there is a difference between image generation and image acquisition. Use word ‘acquisition’ instead of ‘generation’.
Line 122: Write full name of NDVI.
Line 126: write full name of NIR and SWIR1.
Line 140-141: where is confusing matrix and what is the accuracy?
Line 142: change subheading ‘Land degradation information and processes’.
Line 151: write full name of GIS.
Author Response
Comments and Suggestions for Authors
General Comments:
Title of the paper can be improved as paper is discussing spatial distribution of land degradation patterns.
Response: Thank you for your comments. We have changed the title to “Spatio-Temporal Pattern of Land Degradation Along the China–Mongolia Railway (Mongolia)” (lines 2–4).
Abbreviations of the terms should be consistence throughout the paper.
Response: We have unified the abbreviations of terms throughout the paper. In addition, we have provided full names for the abbreviations in this article, such as TM, OLI, GIS, NOAA, MODIS, NIR, and SWIR, NDVI (lines 25-26, 28, 75–76, 143-144, 191).
This paper is using GIS and remote sensing as a tool for the mapping of land degradation patterns however GIS analysis can be improved to produce better results.
Response: With technical support from a GIS spatial analysis module, we superposed land cover data, and generated map variations in land cover for regions along the Railway during the years spanning 1990-2010 and 1990-2015, as shown in Figure 2. In this study, a spatial overlay analysis was used to superpose the land cover data of 1990 with the land cover data of 2010 and 2015 to analyze the changes. The areas of changed land cover types with land degradation were then extracted, and the distribution maps of newly-increased land degradation along the railway from 1990 to 2010 and from 1990 to 2015 were obtained. We have added the corresponding information in lines 241–248.
Object oriented classification method is based on the spectral reflectence of pixels of an image, please explain in the paper how you have done the mapping of the land gradation patters and what is the accuracy and precision level?
Response: We obtained the land cover maps of this study area based on an object-oriented method, which included the preprocessing, segmentation, and selection of different indexes; setting thresholds for all indexes; and accuracy analysis. With technical support from a GIS spatial analysis module, we superposed land cover data, and generated map variations in land cover for regions along the Railway. The overall classification accuracy of the land cover product was 82.26% for 1990, 92.34% for 2010, and 92.75% for 2015. We have added the corresponding information in Section 2.3.1 (lines 160–171), Section 2.3.2 (lines 173–224), Section 2.3.3 (lines 241–248), and Section 3.1 (lines 254–269).
Specific comments:
Line 23: ention the name of the images along spatial resolution.
Response: We have added “Landsat TM and Landsat OLI images” to lines 25-26.
Line 43: Please explain what is ‘Belt and Road initiative’.
Response: In September and October 2013, Chinese President Jinping Xi put forward the cooperation initiatives of building ‘the Silk Road Economic Belt’ and ‘21st-Century Maritime Silk Road’, also known collectively as the ‘Belt and Road Initiative’. In June 2016, the leaders of China, Russia, and Mongolia witnessed the signing of the Outline for the Construction of China–Mongolia–Russia Economic Corridor in Tashkent. It is the first multilateral cooperation outline of the ‘Belt and Road Initiative’. The China–Mongolia Railway is located at the intersection of the China–Mongolia–Russia economic corridor and Mongolia's Steppe Road. We have added this to lines 46–53.
Line 59: Write full name of NOAA and MODIS.
Response: We have added their full names to lines 75–76.
Line 77: replace the word ‘advanced’ with ‘higher’.
Response: We have replaced the word in line 107.
Line 100: Add detail of ‘forests with dominant plants’.
Response: This is a clerical error. We have revised this in lines 132–133. This sentence means that the main types of land cover in the northern part of the study area are real steppe and forest.
Line 105: write full name of Landsat TM and Landsat OLI images.
Response: We have added their full names to lines 25–26.
Line 107: write Digital Elevation Model (DEM).
Response: We have added this to line 149.
Line 106: there is a difference between image generation and image acquisition. Use word ‘acquisition’ instead of ‘generation’.
Response: We have revised this in line 139-140.
Line 122: Write full name of NDVI.
Response: We have added the full name to line 191.
Line 126: write full name of NIR and SWIR1.
Response: We have added their full names to lines 143–144.
Line 140-141: where is confusing matrix and what is the accuracy?
Response: The overall classification accuracy of the land cover product was 82.26% for 1990, 92.34% for 2010, and 92.75% for 2015. The corresponding text description is in lines 254–262.
Line 142: change subheading ‘Land degradation information and processes’.
Response: We have changed it to “Land degradation information and processing” in line 225.
Line 151: write full name of GIS.
Response: We have added the full name to line 28.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The article is well conceived and quite interesting, however, requires several changes. My recommendations are the following
The abstract and the introduction, both have to be rewritten. Several words and phrases of the two overlap.
The keywords should not find a place in the title of the manuscript. Make necessary changes.
I am okay with all the maps, however, figures 4&5 have to be redone. Please don't use excel.
In the methods section, include a flowchart, explicitly demonstrating all the parameters.
Get the manuscript checked by a native English speaker for all the English language related corrections, which is quite important to get rid of all the weaknesses in the manuscript.
References have been quite scanty in the entire manuscript. Please include relevant references where necessary.
Author Response
Comments and Suggestions for Authors
The article is well conceived and quite interesting, however, requires several changes. My recommendations are the following
The abstract and the introduction, both have to be rewritten. Several words and phrases of the two overlap.
Response: Thank you for your comments. We revised the abstract and the introduction.
The keywords should not find a place in the title of the manuscript. Make necessary changes.
Response: We have changed the title and keywords in lines 2–4, 38–39.
I am okay with all the maps, however, figures 4&5 have to be redone. Please don't use excel.
Response: Figures 4 and 5 have been changed to Figures 5 and 6, respectively, and lines 413 and 457–458 have been updated accordingly.
In the methods section, include a flowchart, explicitly demonstrating all the parameters.
Response: We have detailed the selection and application of various indices in the interpretation process (lines 196–215), provided a technical flowchart (line 189), and explained the threshold range of various indices (lines 196–215, and 224). With technical support from a GIS spatial analysis module, we superposed the land cover data and generated variation maps of land cover for regions along the railway during 1990-2010 and 1990-2015. We have added the corresponding information to Section 2.3.3 (lines 241–248).
Get the manuscript checked by a native English speaker for all the English language related corrections, which is quite important to get rid of all the weaknesses in the manuscript.
Response: The manuscript has been revised and polished by a native English speaker.
References have been quite scanty in the entire manuscript. Please include relevant references where necessary.
Response: We have added 11 relevant references to this manuscript.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors present an article entitled “Land degradation pattern along the China-Mongolia Railway from 1990 to 2015 in Mongolia”. The subject seem interesting but a careful reading of the article shows, in my opinion, that it needs a lot of work before being considered for publication.
First, in one hand, Land degradation is a process in which the value of the biophysical environment is affected by combination of human-induced processes acting upon the land. In addition, environmental degradation is the gradual destruction or reduction of the quality and quantity of human activities animals’ activities or natural means example water causes soil erosion, wind, etc. It is viewed as any change or disturbance to the land perceived to be deleterious or undesirable. Natural hazards are excluded as a cause; however, human activities can indirectly affect phenomena such as floods and bush fires. On the other hand, desertification is the process of fertile irrigated lands transformation into dry dead desert with loss of fertility and vegetation
In this article is not clear if the study is about land degradation or desertification? Authors use both designations as synonymous and clearly, they are not.
Then authors performed an analysis for the 1990-2010 and 1990-2015? Since we have almost continuous Landsat data availability from 1990 towards why just these dates? And why precisely these dates? Also is never explained why 1990-2010 and then 1990-2015 repeating the 1990-2010 period. Why not 1990-2010 and 2000-2015?
When describing the data, in line 106 forward, author said that used images from June to September but we do not know if were several images for each year or just one image per year. The decision for choosing that specific time of the year is never justified. We do not know the date of the images, the number of the images, which bands, or even where they come from? USGS? Also, in this section, Thematic Mapper (TM) and Operational Land Imager (OLI) sensor must be defined.
OLI and TM are different sensors and they slightly differ in the bandwidth. To get similar surface reflectances for OLI and TM images they need to be cross-calibrated. Did authors performed sucha a task? And what about georreferencing? Images are already georeferenced or that was done by authors? What was the coordinate system?
Starting to specify the image classification process (line 115) we need to know what kind of algorithm was used for image segmentation, e.g. region growing? Multiresolution? And with what band and what software? The segmentation was one for each year or one for all the year.
What were the segmentation parameters in terms of scale and homogenization? I expect that several different combinations were testes. How authors decided what was the best result? Also the parameters for classifying the polygons in the several LULC classes are not explicit. One should have a table with all the parameter for each LULC classe. These parameters are the same for all the 3 years?
In explaining some of the parameters, authors start referring to NDVI (1), NDWI (2), NDSI (3)
Without previously saying that they intent to calculate them and how they did it. Equations 1, 2 and 3 should precede this part of the text.
Acurracy was measured with confusion matrix (line 141) but with what validation samples? How are they collected? What indexes? Overall accuracy? Producer accuracy? Consumer accuracy? We need to have the matrix with the partial errors for each LULC class.
In section 2.3.2 (line 142) why choosing 200 km, 100 km, 50 km, 30 km, 10 km, and 5 km? There are some references supporting those distances? 200 km from the railroad is significant?
The LULC changes in table 2 and figure 3 are difficult to read. The table should be a double entrance one, i.e. changing from to, figure should have just increase, decrease, as it is there are too many classes, and it is unreadable. In addition, figures must be retouched because the study are looks like an island. I also do not understand what is “non”.
Since we have 82% accuracy in 1995 and 92% in 2010 and 2015 and we know that the error is cumulative, we have 75% accuracy between 1995-2010 and 1995-2015 (0.82 * 0.92 * 100). This means that almost all the changes highlighted in table 2 can be explained by classification mismatches.
Regarding the Natural factors, (line 287) the climate of a region cannot be defined by a 15 year’s time series and surely not using one single station (we do not know its location, authors did not provide it). Like this and without a spatial representation of the values it is impossible to try to relate this data with the land degradation.
Finally, about Socioeconomic factors (line 318). They are general for the all area? The are no data for each of the subregions? Like this one can mask several regional disparities. Also the analysis if for the buffer areas and people can live all outside those buffer, we need to have the LULC class urban that disappeared (the one done by visual interpretation). Following the same line of thought, what happened too DEM and slope?
Some more detailed notes:
In Line 50 authors have to specify that Uvs, Dundgovi and Dornogovi are regions from Mongolia. We should have figure 1 here.
In Line 77 “Higher spatial resolution” should replace “advanced spatial resolution”
Line 103 figure 1 is missing the location of the study area in a broader context.
Line 108 http://datamirror.csdb.cn link does not work
Line 126 (SWIR) short-wave infrared and (NIR) near infrared. Authors’ haven not defined it before
Line 134 SW are infrared bands? Or authors meant SWIR?
Line 151 GIS is not defined
Line 212 until now TOV was referred as Tov
So these comments and suggestions for continuing improving your work.
Best regards
Author Response
Comments and Suggestions for Authors
The authors present an article entitled “Land degradation pattern along the China-Mongolia Railway from 1990 to 2015 in Mongolia”. The subject seems interesting, but a careful reading of the article shows, in my opinion, that it needs a lot of work before being considered for publication.
First, in one hand, Land degradation is a process in which the value of the biophysical environment is affected by combination of human-induced processes acting upon the land. In addition, environmental degradation is the gradual destruction or reduction of the quality and quantity of human activities animals’ activities or natural means example water causes soil erosion, wind, etc. It is viewed as any change or disturbance to the land perceived to be deleterious or undesirable. Natural hazards are excluded as a cause; however, human activities can indirectly affect phenomena such as floods and bush fires. On the other hand, desertification is the process of fertile irrigated lands transformation into dry dead desert with loss of fertility and vegetation
In this article is not clear if the study is about land degradation or desertification? Authors use both designations as synonymous and clearly, they are not.
Response: Thank you for your comments and explanations. This is a study about land degradation based on land cover change monitoring data. However, we mistakenly wrote “land degradation” as “desertification.” We have unified the terms throughout the paper. Our research idea is to interpret the land cover of the study area based on remote-sensing technology and then analyze the change in the obtained land cover data with the help of GIS spatial analysis technology, so as to obtain the land pattern and development trend of this region based on the change in land cover type. We have provided the corresponding explanations in the abstract (lines 24–29), introduction (lines 109–113), method (lines 173–224, 241–248), and conclusion (lines 490–493).
Then authors performed an analysis for the 1990-2010 and 1990-2015? Since we have almost continuous Landsat data availability from 1990 towards why just these dates? And why precisely these dates? Also is never explained why 1990-2010 and then 1990-2015 repeating the 1990-2010 period. Why not 1990-2010 and 2000-2015?
Response: The land cover data of the three stages mentioned in this study were obtained by manual scene-by-scene interpretation using eCognition software. A total of 93 landscape images were interpreted. This is a data-intensive job that requires much manpower. Most of these studies rely on data with coarse spatial resolution (e.g., 1 km) and therefore can only acquire the macroscale distribution and general trends for changes in desertification or land degradation. In this study, for the first time, we obtained land cover data at a 30-m resolution based on Landsat TM and Landsat OLI images. At present, we do not have data from 2000, but have only completed the interpretation of land cover in 1900, 2010, and 2015. In the future, a longer time series can be established to continue research in this region.
When describing the data, in line 106 forward, author said that used images from June to September but we do not know if were several images for each year or just one image per year. The decision for choosing that specific time of the year is never justified. We do not know the date of the images, the number of the images, which bands, or even where they come from? USGS? Also, in this section, Thematic Mapper (TM) and Operational Land Imager (OLI) sensor must be defined.
Response: In this study, 62 Landsat TM (Land Satellite Thematic Mapper) and 31 Landsat OLI (Land Satellite Operational Land Imager) remote-sensing images were acquired. Among them, there were 31 Landsat TM images in 1990 and 2010, and 31 Landsat OLI images in 2015. The imaging data were taken between June and September in three different years. Most images have short imaging time intervals. Their spatial resolution is 30 m, and the cloud coverage is less than 5%. Blue, green, red, near infrared (NIR), short-wave infrared 1 (SWIR1), short-wave infrared 2 (SWIR2), and middle infrared (MIR) are the main bands in these images. Landsat TM and OLI images were obtained from the United States Geological Survey (USGS) website (http://earthexplorer.usgs.gov/). We have added the corresponding information in lines 139–146.
OLI and TM are different sensors and they slightly differ in the bandwidth. To get similar surface reflectances for OLI and TM images they need to be cross-calibrated. Did authors performed such a task? And what about georeferencing? Images are already georeferenced or that was done by authors? What was the coordinate system?
Response: The OLI data was selected for interpretation of land cover in 2015 because the OLI sensor was used after the successful launch of Landsat8 in 2013, and the TM sensor no longer works. Therefore, we can only select OLI data. However, the band designs of TM and OLI are consistent and continuous, and their data processing is synchronous. We have preprocessed all the images before remote-sensing interpretation, including radiometric correction and atmospheric correction. The images are already georeferenced, and their coordinate system is WGS-1984. We have provided the corresponding information in Section 2.3.1 (lines 160–171).
Starting to specify the image classification process (line 115) we need to know what kind of algorithm was used for image segmentation, e.g. region growing? Multiresolution? And with what band and what software? The segmentation was one for each year or one for all the year.
Response: In this study, eCognition software was used for multiresolution segmentation and spectral difference segmentation of each image in different years. We have added the corresponding information in lines 173–184.
What were the segmentation parameters in terms of scale and homogenization? I expect that several different combinations were testes. How authors decided what was the best result? Also the parameters for classifying the polygons in the several LULC classes are not explicit. One should have a table with all the parameter for each LULC classe. These parameters are the same for all the 3 years?
Response: (1) The scale parameter of multiresolution segmentation that we selected was 30, the homogenization of shape was 0.1, and the homogenization of compactness was 0.5. The maximum spectral difference of spectral difference segmentation was 4. Before determining the above parameters, we selected different parameters for the experiments and recorded the segmentation results and time consumption under different parameter conditions. Then the parameters above are determined to be the best parameters after considering the segmentation effect and time. We have added the corresponding information to lines 174–180. (2) We employed a remote-sensing classification system with 30-m resolution for land cover in Mongolia that was developed by Wang et al. (lines 185–188). These parameters are in the same range for all 3 years. We have added a statistical table for parameter selection of different land cover types to line 224.
In explaining some of the parameters, authors start referring to NDVI (1), NDWI (2), NDSI (3)
Without previously saying that they intent to calculate them and how they did it. Equations 1, 2 and 3 should precede this part of the text.
Response: Equations 1, 2, and 3 have been placed before this part of the text (lines 191–195).
Accuracy was measured with confusion matrix (line 141) but with what validation samples? How are they collected? What indexes? Overall accuracy? Producer accuracy? Consumer accuracy? We need to have the matrix with the partial errors for each LULC class.
Response: The data sources for this precision verification are mainly composed of three parts: the field test verification points in 2013 (50 points), longitude and latitude intersection verification points downloaded from the Degree Confluence Program (119 points), and high-resolution sample points obtained from Google (140 points). A total of 309 verification points was collected. We have included the corresponding information in lines 256–262.
In section 2.3.2 (line 142) why choosing 200 km, 100 km, 50 km, 30 km, 10 km, and 5 km? There are some references supporting those distances? 200 km from the railroad is significant?
Response: Combined with previous studies, it is found that people mostly choose the regions within the range of 2 km, 5 km, 10 km, 30 km, 50 km and 100 km on both sides of the railway as the study area (Zhang et al., 2004; Ding et al., 2006). We set up the range of 200km to expand the study area and try to ensure that the human activities and climate change impacts adjacent to the railway can be reflected. We have added the corresponding information in lines 234–239.
The LULC changes in table 2 and figure 3 are difficult to read. The table should be a double entrance one, i.e. changing from to, figure should have just increase, decrease, as it is there are too many classes, and it is unreadable. In addition, figures must be retouched because the study are looks like an island. I also do not understand what is “non”.
Response: Table 2 has been changed to Table 3 and Figure 3 has been changed to Figure 4. In Figure 4, the dark color represents the newly increased land, whereas the light color represents the land restoration areas. Through Figure 4 and Table 3, we want to show the land degradation types in different regions, show the proportion of different land degradation types, and analyze which types of land degradation are the main types and their spatial distribution characteristics, so as to achieve a detailed analysis of the land degradation process in this region. In the figure and table, “non” means “non-degraded land”.
Since we have 82% accuracy in 1995 and 92% in 2010 and 2015 and we know that the error is cumulative, we have 75% accuracy between 1995-2010 and 1995-2015 (0.82 * 0.92 * 100). This means that almost all the changes highlighted in table 2 can be explained by classification mismatches.
Response: In the land cover data, the classification accuracy of the built area, cropland, and meadow steppe is low, whereas the classification accuracy of desert steppe, barren, sand and desert, which account for the majority of the study area, is high and stable. Therefore, the accuracy of land degradation dominated by these land cover types is stable and reliable.
Regarding the Natural factors, (line 287) the climate of a region cannot be defined by a 15 year’s time series and surely not using one single station (we do not know its location, authors did not provide it). Like this and without a spatial representation of the values it is impossible to try to relate this data with the land degradation.
Response: We downloaded the data of all provinces and a total of 12 meteorological stations covered by the study area from the Mongolian Statistical Information Service website, and then calculated their average and sum to obtain the meteorological data in this paper. These data can better express the climate and other natural conditions in this study area.
Finally, about Socioeconomic factors (line 318). They are general for the all area? The are no data for each of the subregions? Like this one can mask several regional disparities. Also the analysis if for the buffer areas and people can live all outside those buffer, we need to have the LULC class urban that disappeared (the one done by visual interpretation). Following the same line of thought, what happened too DEM and slope?
Response: The population and goat number in Section 4.2.2 of this paper are the sum of the population and number of goats in the provinces covered by this study, and represent the whole study area. At present, the influence of DEM and slope is not considered in this study because the land relief along the railway is flat in the Mongolia plateau.
Some more detailed notes:
In Line 50 authors have to specify that Uvs, Dundgovi and Dornogovi are regions from Mongolia. We should have figure 1 here.
Response: We have revised lines 64–66.
In Line 77 “Higher spatial resolution” should replace “advanced spatial resolution”
Response: We have revised line 107.
Line 103 figure 1 is missing the location of the study area in a broader context.
Response: We have revised Figure 1 in line 135.
Line 108 http://datamirror.csdb.cn link does not work
Response: We apologize; this is a clerical error. We have revised in lines 149–151.
Line 126 (SWIR) short-wave infrared and (NIR) near infrared. Authors’ haven not defined it before
Response: We have revised Figure 1 in lines 143–144.
Line 134 SW are infrared bands? Or authors meant SWIR?
Response: SW are infrared bands. They can be replaced by NIR (Near Infrared) as well. We have revised lines 191–195 accordingly.
Line 151 GIS is not defined
Response: We have revised this in line 28.
Line 212 until now TOV was referred as Tov
Response: Tov is the name of a province in Mongolia. The province surrounds Ulaanbaatar, the capital of Mongolia.
Author Response File: Author Response.pdf
Reviewer 4 Report
General comments
The manuscript focuses on the impact of Land Degradation processes in Mongolia on the sustainability of a very important infrastructure for the area, the China-Mongolia Railway. The general aim and the basic outline of the study are very interesting for Sustainability journal. The authors analyze land degradation patterns at high resolution (30m) evaluating both natural (temperature, precipitation) and socioeconomic (population, pastoralism) factors.
However, there many point to be addressed before publication.
There are two main issues: (1) the reliability of analyzed precipitation data and (2) the impact of vegetation phenology on the analyzed satellite image, particularly on vegetation index (NDVI) values.
1) Precipitation data show the same value of rainfall for the first ten years (2000-2010) and another constant value for the period 2012-2015. Even if the shown data are annual mean it is not reliable that rained the same exact amount (different from zero) for ten consecutive years. Therefore, results and comments based on these data are not consistent.
2) Land degradation is evaluated on basis of changes in land covers; these maps were generated from satellite images collected during the period June-September. During this three-month period, vegetation phenology is highly variable and, consequently, also NDVI values are uneven. Since the authors used thresholds on NDVI for classifying land covers, differences in phenological status can “erroneously” be interpreted as different land cover classes. Thus, the spurious changes due to phenology have to be avoided.
Specific comments
Keywords: “distribution patter”, It seems a fairly vague keywords
Line 21: It is not clear "They" what it refers to. The influence of ecosystem and sustainable development on traffic arteries (???) … perhaps, land degradation processes can have influence on traffic arteries.
Line 44: “Mongolia is a hotspot of global land degradation.”The cited reference don not compare Mongolian degradation at global scale. Please modify the sentence or select an appropriate reference.
Line 64: reference [8], Please add a link or more information about this reference.
Line 97: “temperate grassy climate”. The adjective “grassy” is more related to ecosystems than a climate characteristic. Do you refer to a biogeographic region “Temperate grassland” ? ….Please verify !!!
Lines 105-106: More details are required about the satellite TM/OLI images. Please specify the number and acquisition dates of the images used per year (1990, 2010, and 2015). Such information are very relevant to understand the vegetation phenology status at the time of acquisition. The land covers for the three years were generated from satellite images collected during the period June-September. If you compare the vegetation status of an image acquired close to the phenological maximum with the status obtained from images collected in other months, their difference can “erroneously” ascribed as vegetation degradation. Such a point is particularly relevant for land covers having slight and sparse vegetation (such as desert steppe).
To avoid a false detection of land degradation you can evaluate these options:
· to compare land cover maps obtained by applying a classification including images acquired in more dates in one year covering the phenology peak;
· to verify the status of phenology at the time of Landsat image acquisition for the considered year by analyzing the NDVI curve from low resolution data (e.g. MODIS 250m)
· to increment the number of year for the analysis, i.e. by considering also land cover maps for 2011, 2012, 2013, 2014; this last point will at least reduce the probability of errors.
Moreover, information on the characteristic of the collected Landsat TM and OLI images have to be added. Do you acquired Radiance or reflectance? What is the level of preprocessing? From which database were the images downloaded?
Line 113: The aim and type of field investigations are not clear; please, provide more information on what kind of data were sampled, how the surveys were carried out, etc.
Lines 116-121: The implemented object-oriented classification method has to be better clarified. The authors provide some references for the identification of spectral heterogeneity of polygons and their classification, but some of them are not accessible (e.g., thesis and meteorological press book) and the others are focused only on some of the index (e.g., TS-NDVI or NDVI-DEM) selected by the authors. Therefore, you need to specify from which TM and OLI bands the polygons were identified as well as the parameters and rules applied for their classification.
Line 122: The NDVI is not strictly a tool. From Remote sensing point of view it is a multispectral index; from mathematical point of view it is simply an algorithm.
Lines 134-136: Add in brackets the Landsat TM and OLI bands near the spectral acronyms; e.g. R (TM=B3, OLI=B4).
Line 138: The reference (Wang et l., [22]) describing the classification method is uncorrected. It was not possible to find this title in the indicated journal or anywhere in internet.
Line 140-141: For the computation of the accuracy of land cover maps, which data are used as truth map in the confusion matrix?
Line 161-162: “this remote-sensing interpretation data set is considered to be more refined and accurate.”, please, better specify the improvements compared to other studies (in term of labeling detail, number of classes, spatial resolution, etc. ???).
Line 203: Section 3.2 Spatial distribution pattern of land degradation. The analyses of land cover changes are based on the difference 1990-2010 and 1990-2015. As shown in Table2 large percentages of the area are involved in restoration and degradation processes, but with the adopted approach it is not possible to identify if there are zones where the degradation process persist and where there are recovery dynamics.
In order to better follow the dynamics and separate the areas where land degradation persist from those recently involved in the process, it would be better to evaluate the difference 1990-2010 and 2010-2015, and then build a compendious map of persistent increase and decrease.
The percentages, as they are, make little meaning as, abstractly, the degraded areas can be completely different between the two time steps. Moreover, the analysis of persistent changes can help to minimize the effect of classification errors induced by vegetation phenology (see comments on lines 105-106) and to better support comments of section 4.1 (Spatiotemporal distribution characteristics of land cover and newly-increased land degradation along the railway).
Line 210: The authors correctly define Land Degradation into the introduction. On the basis of this definition, it is more appropriate to indicate the classes not affected by land degradation processes as “non-degraded land” rather than “non-desertification land”. Such a change of term must to be applied throughout the result section.
Line 288: “with obvious” it is not appropriated.
Lines 291-293: Figure 4(a) does not show “a rising trend”. The estimated trend line is rather flat (a slope coefficient of 0.0028 means a temperature increase of 0.042°C in 15 years). The temperature for 2010 seems to be one of the coldest of the 2000-2015 time series, jointly with 2011 and 2012.
The commented 1.34 °C, obtained as pure difference between the temperature of 2015 and 2000, has no climatic relevance in term of trend. Therefore, it cannot be compared with the global mean temperature trend. The correct comparison is with the value 0.042 derived from the estimated trend line.
Lines 297-301: These comments are related to data of precipitation that seems to be unreliable. Figure 4(b) shows a constant (exactly the same rain amount) precipitation value between 2000 2010, similarly for the period 2012-2015. Please, verify the reliability of rainfall data.
The authors can check precipitation patterns shown in a study they cited (Hansen et al 2015 ref n.[3]), in the same period (2000-2010) rainfalls show a higher variability also in the areas buffering the railway line (see fig 7a in [3]). Furthermore, by taking into account the high spatial variability of climate and ecosystems along the north-south direction, it would be better to avoid an analysis based on a mean of the whole area and, instead, consider the spatial distribution of positive and negative rainfall trends (see fig 6 in [3]).
Line 347: change “degenerated” with “degraded”.
Line 361: Conclusions. Comments on decreased rainfall are not supported by the current unreliable precipitation data.
Author Response
Comments and Suggestions for Authors
General comments
The manuscript focuses on the impact of Land Degradation processes in Mongolia on the sustainability of a very important infrastructure for the area, the China-Mongolia Railway. The general aim and the basic outline of the study are very interesting for Sustainability journal. The authors analyze land degradation patterns at high resolution (30m) evaluating both natural (temperature, precipitation) and socioeconomic (population, pastoralism) factors.
However, there many point to be addressed before publication.
There are two main issues: (1) the reliability of analyzed precipitation data and (2) the impact of vegetation phenology on the analyzed satellite image, particularly on vegetation index (NDVI) values.
1) Precipitation data show the same value of rainfall for the first ten years (2000-2010) and another constant value for the period 2012-2015. Even if the shown data are annual mean it is not reliable that rained the same exact amount (different from zero) for ten consecutive years. Therefore, results and comments based on these data are not consistent.
Response: Thank you for your comments. We found the mistake in the statistics. The precipitation data should be the average annual precipitation. The average annual precipitation from 2000 to 2010 was 2494.20 mm, and the average annual precipitation from 2011 to 2015 was 2361.42 mm. We have revised this information in lines 417–422.
2) Land degradation is evaluated on basis of changes in land covers; these maps were generated from satellite images collected during the period June-September. During this three-month period, vegetation phenology is highly variable and, consequently, also NDVI values are uneven. Since the authors used thresholds on NDVI for classifying land covers, differences in phenological status can “erroneously” be interpreted as different land cover classes. Thus, the spurious changes due to phenology have to be avoided.
Response: NDVI is only one of several indexes selected in the interpretation process. We also use NDWI, NDSI, Brightness, Compactness, DEM, and visual interpretation for land cover interpretation. The results of land cover interpretation using multiple indexes are more accurate and reliable. The corresponding information is in lines 191–224.
Specific comments
Keywords: “distribution patter”, It seems a fairly vague keywords
Response: “Distribution patter” has been changed to “land degradation pattern” in line 38.
Line 21: It is not clear "They" what it refers to. The influence of ecosystem and sustainable development on traffic arteries (???) … perhaps, land degradation processes can have influence on traffic arteries.
Response: “They” referred to land degradation. We have revised this in line 22.
Line 44: “Mongolia is a hotspot of global land degradation.” The cited reference don not compare Mongolian degradation at global scale. Please modify the sentence or select an appropriate reference.
Response: We have added another reference in lines 548–551.
Line 64: reference [8], Please add a link or more information about this reference.
Response: We have added more information in lines 562–563.
Line 97: “temperate grassy climate”. The adjective “grassy” is more related to ecosystems than a climate characteristic. Do you refer to a biogeographic region “Temperate grassland” ? ….Please verify !!!
Response: We apologize for this clerical error. It should be “temperate continental climate.” We have revised lines 129–130 accordingly.
Lines 105-106: More details are required about the satellite TM/OLI images. Please specify the number and acquisition dates of the images used per year (1990, 2010, and 2015). Such information are very relevant to understand the vegetation phenology status at the time of acquisition. The land covers for the three years were generated from satellite images collected during the period June-September. If you compare the vegetation status of an image acquired close to the phenological maximum with the status obtained from images collected in other months, their difference can “erroneously” ascribed as vegetation degradation. Such a point is particularly relevant for land covers having slight and sparse vegetation (such as desert steppe).
To avoid a false detection of land degradation you can evaluate these options:
· to compare land cover maps obtained by applying a classification including images acquired in more dates in one year covering the phenology peak;
· to verify the status of phenology at the time of Landsat image acquisition for the considered year by analyzing the NDVI curve from low resolution data (e.g. MODIS 250m)
· to increment the number of year for the analysis, i.e. by considering also land cover maps for 2011, 2012, 2013, 2014; this last point will at least reduce the probability of errors.
Moreover, information on the characteristic of the collected Landsat TM and OLI images have to be added. Do you acquired Radiance or reflectance? What is the level of preprocessing? From which database were the images downloaded?
Response: We have added the corresponding information in lines 139–146 and 160–171. After comprehensive consideration of the imaging time, cloud cover, and other factors, most images have shorter imaging time intervals. Only a small number of images have a relatively large imaging time interval.
Line 113: The aim and type of field investigations are not clear; please, provide more information on what kind of data were sampled, how the surveys were carried out, etc.
Response: We have added information about field investigations in lines 259–262.
Lines 116-121: The implemented object-oriented classification method has to be better clarified. The authors provide some references for the identification of spectral heterogeneity of polygons and their classification, but some of them are not accessible (e.g., thesis and meteorological press book) and the others are focused only on some of the index (e.g., TS-NDVI or NDVI-DEM) selected by the authors. Therefore, you need to specify from which TM and OLI bands the polygons were identified as well as the parameters and rules applied for their classification.
Response: We have supplemented the corresponding information in Section 2.3.2 (lines 173–224).
Line 122: The NDVI is not strictly a tool. From Remote sensing point of view it is a multispectral index; from mathematical point of view it is simply an algorithm.
Response: We have changed “tool” to “index” in line 197.
Lines 134-136: Add in brackets the Landsat TM and OLI bands near the spectral acronyms; e.g. R (TM=B3, OLI=B4).
Response: We have added the corresponding information in lines 193–195.
Line 138: The reference (Wang et l., [22]) describing the classification method is uncorrected. It was not possible to find this title in the indicated journal or anywhere in internet.
Response: We apologize; this is a clerical error. We have changed it to the correct title in lines 602–603.
Line 140-141: For the computation of the accuracy of land cover maps, which data are used as truth map in the confusion matrix?
Response: We have added the corresponding information in lines 256–262.
Line 161-162: “this remote-sensing interpretation data set is considered to be more refined and accurate.”, please, better specify the improvements compared to other studies (in term of labeling detail, number of classes, spatial resolution, etc. ???).
Response: This study obtained the land cover data of Mongolia with a resolution of 30 m, which was much higher than the results of Wei et al of 1 km resolution. The overall classification accuracy of the results obtained by Tian et al. is 72.66%, which is lower than the accuracy obtained in this study. We have added the corresponding information in lines 264–267.
Line 203: Section 3.2 Spatial distribution pattern of land degradation. The analyses of land cover changes are based on the difference 1990-2010 and 1990-2015. As shown in Table2 large percentages of the area are involved in restoration and degradation processes, but with the adopted approach it is not possible to identify if there are zones where the degradation process persist and where there are recovery dynamics.
In order to better follow the dynamics and separate the areas where land degradation persist from those recently involved in the process, it would be better to evaluate the difference 1990-2010 and 2010-2015, and then build a compendious map of persistent increase and decrease.
The percentages, as they are, make little meaning as, abstractly, the degraded areas can be completely different between the two time steps. Moreover, the analysis of persistent changes can help to minimize the effect of classification errors induced by vegetation phenology (see comments on lines 105-106) and to better support comments of section 4.1 (Spatiotemporal distribution characteristics of land cover and newly-increased land degradation along the railway).
Response: Land degradation is a long process and will not change significantly in just a few years. It is better to select the data with same period in long term to monitor the land degradation in this area. However, land cover interpretation requires many human and material resources, it is a very data intensive work. At present, we just finished the data in 1990, 2010, 2015, thus we find the land degradation in 1990-2010, and 1990-2015. Because the land degradation process is not easy to be discovered in a short time period, so we don’t analysis the change in 2010-2015 dependently in this paper. In the future, we plan to conduct encryption analysis at intervals of 5 or 10 years, and strengthen the dynamic study of land degradation at different internal distances within the study area.
Line 210: The authors correctly define Land Degradation into the introduction. On the basis of this definition, it is more appropriate to indicate the classes not affected by land degradation processes as “non-degraded land” rather than “non-desertification land”. Such a change of term must to be applied throughout the result section.
Response: We have revised this in the results section.
Line 288: “with obvious” it is not appropriated.
Response: We have deleted it (line 400).
Lines 291-293: Figure 4(a) does not show “a rising trend”. The estimated trend line is rather flat (a slope coefficient of 0.0028 means a temperature increase of 0.042°C in 15 years). The temperature for 2010 seems to be one of the coldest of the 2000-2015 time series, jointly with 2011 and 2012.
The commented 1.34 °C, obtained as pure difference between the temperature of 2015 and 2000, has no climatic relevance in term of trend. Therefore, it cannot be compared with the global mean temperature trend. The correct comparison is with the value 0.042 derived from the estimated trend line.
Response: The temperature showed a slow rising trend and significant fluctuation during 2000-2015 (as shown in Figure 5), and the difference between the maximum annual average temperature and the minimum annual average temperature is 3.14 °C. The temperature fluctuations create an adverse effect on the normal growth and succession of vegetation, and can cause a reduction in vegetation coverage and productivity, severe grassland degradation, and can thereby accelerate the land degradation process. We have revised (lines 401–407).
Lines 297-301: These comments are related to data of precipitation that seems to be unreliable. Figure 4(b) shows a constant (exactly the same rain amount) precipitation value between 2000 2010, similarly for the period 2012-2015. Please, verify the reliability of rainfall data.
The authors can check precipitation patterns shown in a study they cited (Hansen et al 2015 ref n.[3]), in the same period (2000-2010) rainfalls show a higher variability also in the areas buffering the railway line (see fig 7a in [3]). Furthermore, by taking into account the high spatial variability of climate and ecosystems along the north-south direction, it would be better to avoid an analysis based on a mean of the whole area and, instead, consider the spatial distribution of positive and negative rainfall trends (see fig 6 in [3]).
Response: Thank you for comment. We found this mistake in the statistics. Precipitation data should be the average annual precipitation. The average annual precipitation from 2000 to 2010 was 2494.20 mm, and the average annual precipitation from 2011 to 2015 was 2361.42 mm in this study area. We have revised lines 417–422 accordingly.
Line 347: change “degenerated” with “degraded”.
Response: “Degenerated” has been changed to “degraded” in line 474.
Line 361: Conclusions. Comments on decreased rainfall are not supported by the current unreliable precipitation data.
Response: Precipitation data should be the average annual precipitation. According to the statistics, the average annual precipitation from 2000 to 2010 was 2494.20 mm, and the average annual precipitation from 2011 to 2015 was 2361.42 mm. Therefore, it also presents an overall declining trend.
Author Response File: Author Response.pdf
Reviewer 5 Report
In this paper the authors obtained the pattern of land degradation at a 30-m resolution along China-Mongolia railway (Mongolia section) during 1990-2010 and 1990-2015, and explore the joint effect of natural factors and socioeconomic factor resulting the land degradation. This paper is interesting and easy to read. My major concern is the originality and the method. Some detailed revisions are given below.
1. The authors selected the China-Mongolia Railway as the study area to analyze the land degradation, which is an interesting case study, but there has no further analysis to clarify how the railway affects land degradation.
2. In the introduction part, the authors mentioned that some previous studies analyzed the land degradation using limited data with coarse spatial resolution, and lacked of application of data in recent years. Besides the data description, efforts should be made to describe the research status of method with new data, or the advantage using new method or new data. Furthermore, the mechanism of railway's influence on land degradation should be elaborated in this part, as stated above.
3. In the method part, although the authors used the land cover classification system in Mongolia from Wang et al., the description and application of the data should be stated, as well as the accuracy evaluation.
4. In the Section 3.1, the authors mentioned land cover data were produced using object-oriented method. Is the method used in this paper? If yes, the description of the object-oriented method should be added in the methods part, and the results and accuracy should be added in the Results.
5. In the Section 3.1, the authors mentioned the remote-sensing interpretation data is considered to be more refined and accurate. A quantitative comparison with previous studies is needed to draw this conclusion.
6. In the Discussion part, the yearly changes of temperature and precipitation, population and goat number were analyzed. These factors are indeed related to land degradation, but more detailed quantitative analysis is required. For example, the overlay analysis could be applied to study the spatio-temporal relationship between the land degradation and influencing factors. Hence, detailed explanation or extra work should be added.
7. In the Conclusions part, again, the present analysis between land degradation and natural and socioeconomic factors is insufficient to draw the clear conclusion.
In my opinion, the research needs a critical analysis of the data, explore the innovation besides the 30-m resolution data.
Author Response
Comments and Suggestions for Authors
In this paper the authors obtained the pattern of land degradation at a 30-m resolution along China-Mongolia railway (Mongolia section) during 1990-2010 and 1990-2015, and explore the joint effect of natural factors and socioeconomic factor resulting the land degradation. This paper is interesting and easy to read. My major concern is the originality and the method. Some detailed revisions are given below.
1. The authors selected the China-Mongolia Railway as the study area to analyze the land degradation, which is an interesting case study, but there has no further analysis to clarify how the railway affects land degradation.
Response: We are grateful to the learned reviewer for the insightful comments. The Railway has historically been the traffic artery between China, Mongolia, and Russia, and will become the core of traffic infrastructure in the China–Mongolia–Russia Economic Corridor with the implementation of the Belt and Road Initiative. Therefore, this study focuses more on the impact of land degradation status and the development trend in the region along the railway. This manuscript revealed the status and development trend of land degradation along the China-Mongolia railway, and found significant areas of land degradation according to its development and changes. It can provide quantitative and spatial support for the management and control of land degradation risk in this region.
2. In the introduction part, the authors mentioned that some previous studies analyzed the land degradation using limited data with coarse spatial resolution, and lacked of application of data in recent years. Besides the data description, efforts should be made to describe the research status of method with new data, or the advantage using new method or new data. Furthermore, the mechanism of railway's influence on land degradation should be elaborated in this part, as stated above.
Response: We have added part of the research status on land degradation based on data with higher resolution and updated the imaging time in lines 91–104. This study focuses more on the impact of land degradation status and the development trend in the region along the railway.
3. In the method part, although the authors used the land cover classification system in Mongolia from Wang et al., the description and application of the data should be stated, as well as the accuracy evaluation.
Response: We obtained the land cover maps of this study area based on an object-oriented method, which included the preprocessing, segmentation, selection of different indexes, setting of thresholds for all indexes, and accuracy analysis. We have added the corresponding information in Section 2.3.1 (lines 160–171), Section 2.3.2 (lines 173–224), and Section 3.1 (lines 254–267).
4. In the Section 3.1, the authors mentioned land cover data were produced using object-oriented method. Is the method used in this paper? If yes, the description of the object-oriented method should be added in the methods part, and the results and accuracy should be added in the Results.
Response: We used object-oriented method to obtain the land cover data in this manuscript. We have added the corresponding information of the object-oriented method in Section 2.3.2 (lines 173–224) and the accuracy in Section 3.1 (lines 254–267). And we have described the results in Section 3.1 (lines 267-310).
5. In the Section 3.1, the authors mentioned the remote-sensing interpretation data is considered to be more refined and accurate. A quantitative comparison with previous studies is needed to draw this conclusion.
Response: We have added the corresponding information in lines 264–267.
6. In the Discussion part, the yearly changes of temperature and precipitation, population and goat number were analyzed. These factors are indeed related to land degradation, but more detailed quantitative analysis is required. For example, the overlay analysis could be applied to study the spatio-temporal relationship between the land degradation and influencing factors. Hence, detailed explanation or extra work should be added.
Response: Quantitative analysis requires synchronous observation based on a large number of observation stations. There are not yet such conditions in Mongolia. Because there are no such observation stations, we can only perform a qualitative driving-force analysis using existing stations and statistical data. In the future, stations can be set up in some locations along the railway or in the north–south sample belt to strengthen the quantitative analysis between land degradation and temperature, precipitation, and human activities in the long term.
7. In the Conclusions part, again, the present analysis between land degradation and natural and socioeconomic factors is insufficient to draw the clear conclusion.
Response: We have only analyzed the driving force, and we have inferred several obvious influencing factors. We have also revised the relevant content of the discussion and conclusion sections.
In my opinion, the research needs a critical analysis of the data, explore the innovation besides the 30-m resolution data.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors of the article “Spatio-Temporal Pattern of Land Degradation Along the China-Mongolia Railway (Mongolia)” made a tremendous effort on improving their work. In my modest opinion, the article is now much more consistent and explicit. Nevertheless, six main issues still persist.
OLI and TM images need to be cross-calibrated the atmospheric calibration is insuficiente when using such a long term time-series and different sensors and specially with a sensible index as NDVI. If authors used MSAVI the problem would be smaller.
Still, there is no justification for using 1990, 2010 and 2015. Why 20 years difference and then 5? Why this specific dates? Are their core dates on the region?
The reference Zhang et al., 2004 is not searchable in the internet and Ding et al., 2006 just use a 100 km buffer arbitrarily chosen, with no references supporting that option. (page 209-211 and not 234-239 as stated in the response)
Authors state that “the classification accuracy of the built area, cropland, and meadow steppe is low”. So, the built area was achieved by visual interpretation and still has low classification accuracy? Authors still have to present an error matrix with overall, producer and consumer accuracy for each of the classes. At least for the ones that have influence in land degradation. It is the only way that readers have to evaluate the quality of the methodology and the usefulness of the final data.
Some provinces are less than 50% covered by the study area. What can prove to readers that the population and the goats are only in the part covered by the study? Or this is true or authors have to do some kind of dasymetric analysis.
Finally, using 12 meteorological stations resolves part of the problem but still it is impossible to characterize some region climate using a 15 years time-series. One should have at least a 30 years time-serie, i.e. climatological normal.
Best regards
Author Response
Response to the review of English language and style: Thank you for the comments. The manuscript has been revised and polished again by a native English speaker.
Comments and Suggestions for Authors:
The authors of the article “Spatio-Temporal Pattern of Land Degradation Along the China-Mongolia Railway (Mongolia)” made a tremendous effort on improving their work. In my modest opinion, the article is now much more consistent and explicit. Nevertheless, six main issues still persist.
OLI and TM images need to be cross-calibrated the atmospheric calibration is insuficiente when using such a long term time-series and different sensors and specially with a sensible index as NDVI. If authors used MSAVI the problem would be smaller.
Response: Thank you for the comments. In this study, we did not carry out cross-calibration for TM and OLI, which may indeed have an impact on NDVI. However, the images of Landsat series are consistent and continuous; the two sensors had good contact and consistency in image processing. Moreover, our remote sensing interpretation was carried out scene by scene, meaning that not only the three years images were interpreted independently but also each scene image in each year was processed independently. This more time-consuming interpretation method can ensure that data processing is not influenced by different sensors and therefore could reduce the error effect to a certain extent.
Still, there is no justification for using 1990, 2010 and 2015. Why 20 years difference and then 5? Why this specific dates? Are their core dates on the region?
Response: This is a good point. Initially, we tried to capture the general change in land degradation in the period from 1990 to 2010. As our research advanced, we added 2015 to show the more recent status. Although the image interpretation is very time-consuming, we plan to increase the land cover monitoring period every 5 years in the future. Thus, we can analyze the land degradation dynamics in every 5 years from 1990.
The reference Zhang et al., 2004 is not searchable in the internet and Ding et al., 2006 just use a 100 km buffer arbitrarily chosen, with no references supporting that option. (page 209-211 and not 234-239 as stated in the response)
Response: (1) The reference Zhang et al., 2004 is a paper in Chinese. It can be found through this link (http://www.ere.ac.cn/CN/abstract/abstract9024.shtml). In 2018, the journal name has been changed from Rural Ecological Environment to Journal of Ecology and Rural Environment. We have revised in lines 641-643. (2) In the cited paper, Ding et al. clearly analyzed the spatial distribution and dynamic change in land cover along the Qinghai-Tibet Highway and Railway by setting a buffer of 100 km. Therefore, the buffer of 100 km can be used as a reference. The 200 km is an empirical value of our choice. After comparing with relevant studies, to better reflect the land degradation pattern in this region, we not only selected the buffer range set by previous studies but also added a larger buffer range (200 km). Therefore, the study area we chose can reflect the pattern of land degradation along the railway.
Authors state that “the classification accuracy of the built area, cropland, and meadow steppe is low”. So, the built area was achieved by visual interpretation and still has low classification accuracy? Authors still have to present an error matrix with overall, producer and consumer accuracy for each of the classes. At least for the ones that have influence in land degradation. It is the only way that readers have to evaluate the quality of the methodology and the usefulness of the final data.
Response: Thank you for this comment. This is a clerical error. The classification accuracy of the built area is not low. The overall classification accuracy of the land cover data was 82.26%, 92.34% and 92.75% for 1990, 2010, and 2015, respectively (lines 254-256). We have added a table containing the producer and consumer accuracy for each land cover types in 1990 and 2010 (lines 256 and 271).
Some provinces are less than 50% covered by the study area. What can prove to readers that the population and the goats are only in the part covered by the study? Or this is true or authors have to do some kind of dasymetric analysis.
Response: At first, we wanted to reflect the overall situation of population and goat numbers in the study area. Therefore, we collected the data of most provinces covered by the study area except Ovorhangay and Omnogovi, because the areas involved in Ovorhangay and Omnogovi are very small. Based on this comment, now we exclude the provinces with an area less than 50% covered by the study area. We have made revisions in lines 463-466, 472-475, and 478-479.
Finally, using 12 meteorological stations resolves part of the problem but still it is impossible to characterize some region climate using a 15 years time-series. One should have at least a 30 years time-serie, i.e. climatological normal.
Response: The availability of regional data has brought obstacles to this study. At present, we only have the data from 2000 to 2015 from the National Statistical Office of Mongolia, who has not released the data of 30 years or longer. Before data selection, we compared data released by the National Statistical Office of Mongolia, NOAA, and other organizations, finding that only three sites published by NOAA could provide a 30-year or longer time series in this study area. Compare there, we use the 12 stations’ records from the National Statistical Office of Mongolia for this study. In the future, if we can get more open station data, we can further strengthen the climate change study in the region.
Author Response File: Author Response.pdf
Reviewer 4 Report
General comments
The manuscript on the impact of Land Degradation processes in Mongolia Railway area is substantially improved by the revision. In particular, the method section is more detailed and clear.
At the same time, the fundamental issue related to classification errors induced by the phenological status of vegetation is not addressed. The authors analyze land degradation patterns comparing land cover maps for three years.
Previous comment: 2) Land degradation is evaluated on basis of changes in land covers; these maps were generated from satellite images collected during the period June-September. During this three-month period, vegetation phenology is highly variable and, consequently, also NDVI values are uneven. Since the authors used thresholds on NDVI for classifying land covers, differences in phenological status can “erroneously” be interpreted as different land cover classes. Thus, the spurious changes due to phenology have to be avoided.
If you compare the vegetation status of an image acquired close to the phenological maximum with the status obtained from images collected in other months, their difference can “erroneously” ascribed as vegetation degradation. Such a point is particularly relevant for land covers having slight and sparse vegetation (such as desert steppe).
Authors’ Response: NDVI is only one of several indexes selected in the interpretation process. We also use NDWI, NDSI, Brightness, Compactness, DEM, and visual interpretation for land cover interpretation. The results of land cover interpretation using multiple indexes are more accurate and reliable. The corresponding information is in lines 191– 224.
Comment: As reported in your flowchart you applied (or not) some indices as a function of the condition of the previous index. Therefore, as an example, if you have an NDVI of 0.38 during the senescence period of a Forest, you do not apply the condition on distance and DEM and classify it as Real Steppe. Moreover, the phenological status of vegetation influences all the reflectance bands, therefore also other indices, such as NDWI, NDSI, Brightness, are directly influenced.
In support of the hypothesis of the presence of some errors induced by the peculiar status of vegetation at the acquisition time, there are large barren areas that appear in the period 1990-2010 and suddenly disappear between 2010 and 2015 (see red circles in the attached images; sustainability-476523-FigComment.pdf).
Desertification processes, as also stated by the authors, act on a long period and it is not represented by the peculiar condition of a single year. Therefore, to evaluate the land degradation in the area, I suggest to analyze the persistence of changes between the change maps 1990-2010 and 1990-2015 (i.e., to show the areas where the same change persist in both the maps, such as Desert Steppe to Barren or Barren to non).
The implementation of such a map takes just one hour of work in ENVI or GIS environment and makes your results more consistent, and the article publishable in Sustainability journal.
Specific comments
Line 76: Please verify the use of Reference in the text for MDPI journal standard (e.g. Zhuo)
Lines 83-85: “With the continuous improvement of spectral resolution, time resolution and spatial resolution of remote sensing data, people began to utilize more sensors to discover more accurate land degradation monitoring results [12-14]” … The reported references used only Landsat data, which have the same time and multispectral spatial resolution since 1972. Such references [12-14] are better positioned close to the next sentence “Landsat series data has been widely used.”
Here some reference with improved spatial and spectral resolution:
- Pignatti, S., Acito, N., Amato, U., Casa, R., Castaldi, F., Coluzzi, R., ... & Matteoli, S. (2015, July). Environmental products overview of the Italian hyperspectral prisma mission: The SAP4PRISMA project. In 2015 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (pp. 3997-4000). IEEE. (DOI: https://0-doi-org.brum.beds.ac.uk/10.1109/IGARSS.2015.7326701)
- Mansour, K., Mutanga, O., Adam, E., & Abdel-Rahman, E. M. (2016). Multispectral remote sensing for mapping grassland degradation using the key indicators of grass species and edaphic factors. Geocarto International, 31(5), 477-491. (DOI: https://0-doi-org.brum.beds.ac.uk/10.1080/10106049.2015.1059898)
- Meusburger, K., Bänninger, D., & Alewell, C. (2010). Estimating vegetation parameter for soil erosion assessment in an alpine catchment by means of QuickBird imagery. International Journal of Applied Earth Observation and Geoinformation, 12(3), 201-207. (DOI: https://0-doi-org.brum.beds.ac.uk/10.1016/j.jag.2010.02.009)
Figure 2: in the flowchart some corrections are needed:
Please verify the condition “NDWI > 0” and the reference threshold for water in table 1 “NDWI >-0.4”.
The condition on “Distance” for Forest and Medow Steppe is not discussed in the text.
Only the first condition on NDWI is numerically expressed, please uniform the logic flow by adding values for each condition.
Figure3: Please verify the correctness of year label for the land cover maps. If you compare these maps with those of changes in figure 4, you can see some discrepancies. See, as an example the area with blue circle in the attached file (sustainability-476523-FigComment.pdf). This area is reported as recovery from Desert Steppe to Non degraded classes in figure 4a and 4b (light blue pixels), but the map of 1990 in figure 3 shows a white area, i.e. already non degraded classes; no Desert Steppe are present in 1990 in this circled are.
Please check which map is uncorrected.
Figure 4: Please substitute the change maps of the two pairs of years with the suggested map of change persistence that can be more representative of the analyzed land degradation processes (see an example https://0-doi-org.brum.beds.ac.uk/10.3390/rs70608154). The map should contain the areas where the same change between 1990-2010 is preserved also in the change map 1990-2015 in both the directions (increase and decrease of degraded conditions, such as Desert Steppe to Barren or Barren to non). Also no change area have to be identifiable respect to non persistent changes.
Consequently, Table 3 have to be rearranged.
Lines 382-386: “the precipitation distribution has obvious zonal characteristics: specifically, it increases gradually from south to north. Vegetation growth is suppressed with decreased precipitation along the Railway, where precipitation mainly occurs from June to September.” … You are discussing precipitation data from only one station … how you derive information on spatial distribution of rainfall?
It would be better to support the discussion with previous study in the region, you already cited.
Comments for author File: Comments.pdf
Author Response
Response to the review of English language and style: Thank you for the comments. The manuscript has been revised and polished again by a native English speaker.
General comments
The manuscript on the impact of Land Degradation processes in Mongolia Railway area is substantially improved by the revision. In particular, the method section is more detailed and clear.
At the same time, the fundamental issue related to classification errors induced by the phenological status of vegetation is not addressed. The authors analyze land degradation patterns comparing land cover maps for three years.
Previous comment: 2) Land degradation is evaluated on basis of changes in land covers; these maps were generated from satellite images collected during the period June-September. During this three-month period, vegetation phenology is highly variable and, consequently, also NDVI values are uneven. Since the authors used thresholds on NDVI for classifying land covers, differences in phenological status can “erroneously” be interpreted as different land cover classes. Thus, the spurious changes due to phenology have to be avoided.
If you compare the vegetation status of an image acquired close to the phenological maximum with the status obtained from images collected in other months, their difference can “erroneously” ascribed as vegetation degradation. Such a point is particularly relevant for land covers having slight and sparse vegetation (such as desert steppe).
Authors’ Response: NDVI is only one of several indexes selected in the interpretation process. We also use NDWI, NDSI, Brightness, Compactness, DEM, and visual interpretation for land cover interpretation. The results of land cover interpretation using multiple indexes are more accurate and reliable. The corresponding information is in lines 191– 224.
Comment: As reported in your flowchart you applied (or not) some indices as a function of the condition of the previous index. Therefore, as an example, if you have an NDVI of 0.38 during the senescence period of a Forest, you do not apply the condition on distance and DEM and classify it as Real Steppe. Moreover, the phenological status of vegetation influences all the reflectance bands, therefore also other indices, such as NDWI, NDSI, Brightness, are directly influenced.
In support of the hypothesis of the presence of some errors induced by the peculiar status of vegetation at the acquisition time, there are large barren areas that appear in the period 1990-2010 and suddenly disappear between 2010 and 2015 (see red circles in the attached images; sustainability-476523-Fig Comment.pdf).
Response: Thank you for the comments. The phenological state of vegetation does affect remote sensing interpretation, especially in transition zones, such as the interface between vegetation zones and non-vegetation zones, as well as the interface between the grassland and barren. We have made corresponding explanations in lines 415-420. In the future, strengthening the study on the phenological effects of vegetation may be the key to solve this problem and to improve the classification accuracy.
Desertification processes, as also stated by the authors, act on a long period and it is not represented by the peculiar condition of a single year. Therefore, to evaluate the land degradation in the area, I suggest to analyze the persistence of changes between the change maps 1990-2010 and 1990-2015 (i.e., to show the areas where the same change persist in both the maps, such as Desert Steppe to Barren or Barren to non).
The implementation of such a map takes just one hour of work in ENVI or GIS environment and makes your results more consistent, and the article publishable in Sustainability journal.
Response: We have added the distribution map of the same and persistent change region in the two periods of 1990-2010 and 1990-2015 (line 366), as well as a table about the area and proportion of the persistent change region (lines 373-374). We have also added the corresponding description in lines 348-363 and 397-398.
Specific comments
Line 76: Please verify the use of Reference in the text for MDPI journal standard (e.g. Zhuo)
Response: We have added “et al.” in lines 66 and 75. “Zhuo” is correct because there is only one author of this paper.
Lines 83-85: “With the continuous improvement of spectral resolution, time resolution and spatial resolution of remote sensing data, people began to utilize more sensors to discover more accurate land degradation monitoring results [12-14]” … The reported references used only Landsat data, which have the same time and multispectral spatial resolution since 1972. Such references [12-14] are better positioned close to the next sentence “Landsat series data has been widely used.”
Here some reference with improved spatial and spectral resolution:
- Pignatti, S., Acito, N., Amato, U., Casa, R., Castaldi, F., Coluzzi, R., ... & Matteoli, S. (2015, July). Environmental products overview of the Italian hyperspectral prisma mission: The SAP4PRISMA project. In 2015 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (pp. 3997-4000). IEEE. (DOI: https://0-doi-org.brum.beds.ac.uk/10.1109/IGARSS.2015.7326701)
- Mansour, K., Mutanga, O., Adam, E., & Abdel-Rahman, E. M. (2016). Multispectral remote sensing for mapping grassland degradation using the key indicators of grass species and edaphic factors. Geocarto International, 31(5), 477-491. (DOI: https://0-doi-org.brum.beds.ac.uk/10.1080/10106049.2015.1059898)
- Meusburger, K., Bänninger, D., & Alewell, C. (2010). Estimating vegetation parameter for soil erosion assessment in an alpine catchment by means of QuickBird imagery. International Journal of Applied Earth Observation and Geoinformation, 12(3), 201-207. (DOI: https://0-doi-org.brum.beds.ac.uk/10.1016/j.jag.2010.02.009)
Response: Thanks for the suggestion. We have listed the three references you mentioned as references “[12-15]”. And we have changed the original references “[12-14]” to “[15-17]” as references for the next sentence (i.e., “Landsat series data has been widely used.”). We have revised in lines 92 and 582-596.
Figure 2: in the flowchart some corrections are needed:
Please verify the condition “NDWI > 0” and the reference threshold for water in table 1 “NDWI >-0.4”.
The condition on “Distance” for Forest and Medow Steppe is not discussed in the text.
Only the first condition on NDWI is numerically expressed, please uniform the logic flow by adding values for each condition.
Response: We have revised Figure 2 (line 183), added the corresponding description (lines 192-218), and improved Table 1 accordingly (line 219).
Figure3: Please verify the correctness of year label for the land cover maps. If you compare these maps with those of changes in figure 4, you can see some discrepancies. See, as an example the area with blue circle in the attached file (sustainability-476523-FigComment.pdf). This area is reported as recovery from Desert Steppe to Non degraded classes in figure 4a and 4b (light blue pixels), but the map of 1990 in figure 3 shows a white area, i.e. already non degraded classes; no Desert Steppe are present in 1990 in this circled are.
Please check which map is uncorrected.
Response: We apologize for this error. The previous Figure 3 (a) was an outdate version in our experiments. It has been replaced by with the final figure in line 307 in this revised manuscript.
Figure 4: Please substitute the change maps of the two pairs of years with the suggested map of change persistence that can be more representative of the analyzed land degradation processes (see an example https://0-doi-org.brum.beds.ac.uk/10.3390/rs70608154). The map should contain the areas where the same change between 1990-2010 is preserved also in the change map 1990-2015 in both the directions (increase and decrease of degraded conditions, such as Desert Steppe to Barren or Barren to non). Also no change area have to be identifiable respect to non persistent changes.
Consequently, Table 3 have to be rearranged.
Response: We have added the distribution map of the same and persistent change region in the two periods of 1990-2010 and 1990-2015 (line 366), as well as a table about the area and proportion of the persistent change region (lines 373-374). We have also added the corresponding description in lines 348-363 and 397-398.
Lines 382-386: “the precipitation distribution has obvious zonal characteristics: specifically, it increases gradually from south to north. Vegetation growth is suppressed with decreased precipitation along the Railway, where precipitation mainly occurs from June to September.” … You are discussing precipitation data from only one station … how you derive information on spatial distribution of rainfall?
It would be better to support the discussion with previous study in the region, you already cited.
Response: We have downloaded the data of all provinces and a total of 12 meteorological stations covered by the study area from the Mongolian Statistical Information Service website. Using these data, we have obtained the meteorological data in this study area by calculation. We have added corresponding information in lines 429-432 and 440-443. A related reference has been cited to explain that the precipitation distribution of the entire Mongolian plateau has obvious zonal characteristics and increases gradually from south to north (lines 446-448 and 652-653).
Author Response File: Author Response.pdf
Reviewer 5 Report
The authors have responded to my questions and made the necessary changes to the manuscript. I recommend that this paper be accepted for publication.
Author Response
The authors have responded to my questions and made the necessary changes to the manuscript. I recommend that this paper be accepted for publication.
Response: Thank you very much.
Author Response File: Author Response.pdf
Round 3
Reviewer 3 Report
Thank you for the effort and sorry for any inconvenience.
Reviewer 4 Report
The manuscript on the impact of Land Degradation processes in Mongolia Railway area was largely improved by the revision. In particular, the addition of the map of persistence of changes provided the needed confidence in the identification of degraded areas.
The manuscript is ready to be published.
