Monitoring the Spatiotemporal Dynamics of Aeolian Desertification

remote sensing

Article Monitoring the Spatiotemporal Dynamics of Aeolian Desertiﬁcation Using Google Earth Engine

Ang Chen 1, Xiuchun Yang 1,*, Bin Xu 1,2, Yunxiang Jin 2, Jian Guo 3,4, Xiaoyu Xing 2, Dong Yang 2, Ping Wang 1 and Libo Zhu 5

1 School of Grassland Science, Beijing Forestry University, Beijing 100081, China; [email protected] (A.C.); [email protected] (B.X.); [email protected] (P.W.) 2 Key Laboratory of Agri-informatics, Ministry of Agriculture and Rural Affairs/Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Beijing 100081, China; [email protected] (Y.J.); [email protected] (X.X.); [email protected] (D.Y.) 3 State Key Laboratory of Remote Sensing Science, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China; [email protected] 4 Beijing Key Laboratory for Remote Sensing of Environment and Digital Cities, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China 5 Hulunbuir Institute of Animal Husbandry Science, Hulunbuir 021008, China; [email protected] * Correspondence: [email protected]

Abstract: Northern China has been long threatened by aeolian desertification. In recent years, all levels of the Chinese government have performed a series of ecological protection and sand control projects. To grasp the implementation effects of these projects and adjust policies in time, it is necessary to understand the process of aeolian desertification quickly and accurately. Remote sensing technologies play an irreplaceable role in aeolian desertification monitoring. In this study, the Zhenglan Banner, which is in the hinterland of the Hunshandake Sandy Land, was considered as the research area. Based on unmanned aerial vehicle (UAV) images, ground survey data, and Citation: Chen, A.; Yang, X.; Xu, B.; Landsat images called in Google Earth Engine (GEE), the aeolian desertified land (ADL) in 2000, Jin, Y.; Guo, J.; Xing, X.; Yang, D.; 2004, 2010, 2015, and 2019 was extracted using spectral mixture analysis. A desertification index Wang, P.; Zhu, L. Monitoring the Spatiotemporal Dynamics of Aeolian (DI) was constructed to evaluate the spatial and temporal dynamics of the ADL in the Zhenglan Desertification Using Google Earth Banner. Finally, a residual analysis explored the driving forces of aeolian desertification. The results Engine. Remote Sens. 2021, 13, 1730. showed that (1) the ADL area in the Zhenglan Banner has been trending downwards over the past https://doi.org/10.3390/rs13091730 20 years but rebounded from 2004 to 2010; (2) over the past 20 years, the area of slightly, moderately, and severely desertified land has decreased at annual rates of 0.4%, 2.7%, and 3.4%, respectively; Academic Editor: Guangxing Wang (3) human activities had significantly positive and negative impacts on the aeolian desertification trend for 20.0% and 21.0% of the study area, respectively, but not for the rest. This paper explored Received: 5 March 2021 new techniques for rapid aeolian desertification monitoring and is of great significance for controlling Accepted: 28 April 2021 and managing aeolian desertification in this region. Published: 29 April 2021

Keywords: aeolian desertiﬁcation; Google Earth Engine (GEE); unmanned aerial vehicle (UAV); Publisher’s Note: MDPI stays neutral human activity; climate change with regard to jurisdictional claims in published maps and institutional afﬁl- iations.

1. Introduction According to the definition of the United Nations Convention to Combat Desertifi- cation, desertification is defined as land degradation in arid, semi-arid, or dry subhumid Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. regions as caused by various factors including climate variability and human activities [1]. This article is an open access article Desertification can be divided into aeolian desertification, water erosion, soil salinization, distributed under the terms and and freezing. In China, most of the desertification belongs to aeolian desertification, which conditions of the Creative Commons is land degradation caused by the uncoordinated human–land relationship and is marked Attribution (CC BY) license (https:// by sandstorm activities [2]. China is among the countries most seriously affected by aeolian creativecommons.org/licenses/by/ desertification throughout the world. The results of the Fifth Chinese National Desertifi- 4.0/). cation Monitoring from 2014 showed that the area of aeolian desertification land (ADL)

Remote Sens. 2021, 13, 1730. https://doi.org/10.3390/rs13091730 https://www.mdpi.com/journal/remotesensing Remote Sens. 2021, 13, 1730 2 of 27

was 1,721,200 km2, which accounts for 17.93% of the national land area [3]. In the face of severe aeolian desertification, all levels of government have made tremendous efforts and a series of ecological projects for sand prevention have been launched such as the Three North Shelterbelt Project, the Beijing–Tianjin Sandstorm Source Control Project, and the Return Grazing to Grassland Project. Many studies [4,5] have shown that the trend of aeolian desertification in several regions have been effectively curbed, the overall ecological situation has been continuously improved, and the ADL area has continued to decrease. Data released by NASA in 2019 show that China has increased the global “greening” over the past 20 years [6]. Remote sensing technologies have become the primary method of desertification monitoring due to the fact of its large-scale data acquisition and continuous observation capabilities as well as the advantages of large and repeated coverage, substantial information, and low cost. The main analysis methods include visual interpretation, supervised classification, and use of single-index or multi-index methods. Visual interpretation is a traditional but widely used method to monitor desertification [7]. It is performed through the establishment of interpretation signs and the comprehensive use of the spectrum, tex- ture, shape, and other characteristics of imagery. However, visual interpretation relies on the interpreter’s understanding of aeolian desertification in the study area, which is time-consuming and has difficulty meeting the quantitative development needs of desertification monitoring. The supervised classification method uses known areas as the basis for image classification and classifies imagery using various classifiers including machine learning algorithms [8,9]. In recent years, more complex deep learning algorithms have been applied in desertification monitoring [10]. However, supervised classification has high requirements for the quantity and quality of training samples, which are prone to problems of poor sample representativeness, high sample subjectivity, and insufficient quantitative evaluation capabilities. Many researchers have chosen a representative indicator to classify ADL, where the most widely used are the fractional vegetation coverage [11], normalized difference vegetation index [12], modified soil adjusted vegetation index [13], and others. However, aeolian desertification is a complex and dynamic process and a single indicator can only reflect changes in one aspect of the desertification, making it difficult to achieve high accuracies. Some researchers have comprehensively used multiple indicators for desertification monitoring including decision trees [14], feature space [15], principal component analysis [16], and the analytic hierarchy process [17]. However, a relatively unified comprehensive indicator system and monitoring method has not yet been formulated. The causes of aeolian desertification have long been a research focus, and the explo- ration of the driving forces helps adjust sand prevention and control strategies. Climate change and human activities have been recognized as the two primary driving forces of aeolian desertification. However, accurately and quantitatively analyzing the relative effects of human activities and climate change in aeolian desertification is still difficult. Commonly used methods include qualitative and semi-quantitative assessments [18], regression modeling [19], principal component analysis [20], and residual analysis [21]. Qualitative and semi-quantitative assessments analyze meteorological and social factors based on changes in aeolian desertification and use experienced experts to judge the relative effects. However, this method is subjective and may cause significant differences in the results. The method based on regression modeling analyzes the regression relationship between desertification and climate and social factors. However, the development process of aeolian desertification is complex and nonlinear, making it difficult to simulate using simple regression models. Although the method of principal component analysis has a solid mathematical foundation, it only analyzes the driving forces and does not consider desertification development. In addition, principal component analysis considers the study area without incorporating differences in the continuous space. Residual analysis simulates aeolian desertification conditions without the influence of human factors and uses differences in actual situations to determine the influence of human activities. Remote Sens. 2021, 13, 1730 3 of 27

The Google Earth Engine (GEE) cloud platform was jointly developed by Google, the United States Geological Survey, and Carnegie Mellon University. The GEE platform stores MODIS data from 2000 to present, Landsat data from 1984 to present, Sentinel data from 2014 to present, and massive amounts of other data such as elevation, meteorology, and land use. The remote sensing data in the GEE platform can be processed in parallel with millions of servers, which significantly reduces the computational time and cost. With its powerful computing capabilities, abundant remote sensing data sources, and convenient processing methods, GEE has advantages in long-term time series and large- scale research and has been widely used to monitor forests [22], agriculture [23], urban areas [24], water [25], etc. Unmanned aerial vehicles (UAVs) have been developed in recent years as a new type of near-ground remote sensing platform. UAVs have the advantages of low cost and fast acquisition of imagery with high spatial resolutions. They are suitable for acquiring images in areas with complex topographical conditions and that are difficult to access. They effectively compensate for the gap between the regional scale of satellite remote sensing and the sample scale of ground surveys. UAVs can be used for high-precision monitoring of ADL in small areas, which can largely replace traditional manual ground surveys and reduce workloads. To date, UAVs are widely used in vegetation monitoring, habitat monitoring, wildlife protection, etc. [26,27]. However, few studies have applied them to research aeolian desertification [28]. This paper takes the Zhenglan Banner as the research area, which is located within the source of Beijing–Tianjin sandstorms and is in the hinterland of Hunshandake Sandy Land. The Landsat5 and Landsat8 satellite images collected from GEE were used as the primary data resource. Based on the sample points and UAV images obtained in the field, the ADL over the past 20 years was quickly and accurately extracted through spectral unmixing, and a desertification index was established to analyze dynamic changes in the ADL. Finally, the driving forces of aeolian desertification were analyzed.

2. Materials and Methods The methodological framework of this study is shown in Figure1, with the key steps of calculation in GEE cloud platform environment. All abbreviations and deﬁnitions are shown in Table A1.

2.1. Study Area The Zhenglan Banner is in the central part of Inner Mongolia and the hinterland of the Hunshandake Sandy Land. It is within the scope of the Beijing–Tianjin sandstorm source control project and is only 180 km away from Beijing. It is approximately 138.7 km long from north to south (41◦46’–43◦7’N) and 122 km from east to west (114◦55’–116◦38’ E) with a total area of 10,182 km2 (Figure2). The average annual precipitation of the study area is approximately 365 mm, where the precipitation season is unevenly distributed and concentrated from June to August. The terrain is high in the southwest and low in the northeast with an average altitude of approximately 1300 m. The south part of the study area is mostly low mountains and hills, which is the intersection between the southern edge of Daxinganling and the northern edge of Yanshan Mountains. The north part of the study area is the middle and eastern section of the Hunshandake Sandy Land, which is composed primarily of ﬁxed and semi-ﬁxed sand dunes with a small number of crescent-shaped mobile dunes and low-vegetation coverage. Remote Sens. 2021, 13, 1730 4 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 4 of 26

Figure 1. Research flow chart. The first part shows the preprocessing step of Landsat and UAV Figure 1. Research flow chart. The first part shows the preprocessing step of Landsat and UAV images. The second part shows the image processing steps to obtain the extent of the ADL. The images. The second part shows the image processing steps to obtain the extent of the ADL. The third third part shows the dynamic change evaluation of the ADL based on the desertification index part(DI) shows. The last the part dynamic mainly change focuses evaluation on the driving of the forces ADLbased of aeolian on the desertification desertification. index (DI). The last part mainly focuses on the driving forces of aeolian desertification. 2.1. Study Area The vegetation in the Hunshandake Sandy Land is sensitive to climate change and showsThe the Zhenglan effects of Banner human is influence in the central and control, part of givingInner Mongolia it obvious and ecological the hinterland vulnera- of bilitythe Hunshandake [29]. The relatively Sandy arid Land. climate It is within and long-term the scope uncontrolled of the Beijing grazing–Tianjin in thesandstorm study areasource have control made project aeolian and desertification is only 180 km increasingly away from serious. Beijing. TheIt is approximately research results 138.7 on the km developmentlong from north status to south and driving (41°46’– forces43°7’N) of and aeolian 122 desertificationkm from east to in west the study(114°55 area’–116°38 will ’ provideE) with aa referencetotal area forof 10,182 sand prevention km2 (Figure projects 2). The over average the entire annual Hunshandake precipitation Sandy of the Landstudy andarea the is approximately Beijing–Tianjin 365 Sandstorm mm, where Project the precipitation area. season is unevenly distributed and concentrated from June to August. The terrain is high in the southwest and low in the northeast with an average altitude of approximately 1300 m. The south part of the study area is mostly low mountains and hills, which is the intersection between the southern edge of Daxinganling and the northern edge of Yanshan Mountains. The north part of the study area is the middle and eastern section of the Hunshandake Sandy Land, which is Remote Sens. 2021, 13, x FOR PEER REVIEW 5 of 26

composed primarily of fixed and semi-fixed sand dunes with a small number of crescent- shaped mobile dunes and low-vegetation coverage. The vegetation in the Hunshandake Sandy Land is sensitive to climate change and shows the effects of human influence and control, giving it obvious ecological vulnerability [29]. The relatively arid climate and long-term uncontrolled grazing in the study area have made aeolian desertification increasingly serious. The research results on the development status and driving forces of aeolian desertification in the study area will provide Remote Sens. 2021, 13, 1730 5 of 27 a reference for sand prevention projects over the entire Hunshandake Sandy Land and the Beijing–Tianjin Sandstorm Project area.

FigureFigure 2. 2. LocationLocation and and topographic topographic map map of of the the Zhenglan Zhenglan Banner Banner..

2.2.2.2. Construction Construction of of the the ADL ADL Classification Classification System System BasedBased on on the the field field survey andand previouslypreviouslyrelated related research, research, a a land land classification classification system sys- temin the in the Zhenglan Zhenglan Banner Banner was was established established to include to include terrain, terrain, vegetation, vegetation, and soiland characteris- soil char- acteristicstics (Table (Table1). The 1). land The was land divided was divided into non-desertified into non-desertified land (NDL), land (NDL), slightly slightly desertified des- ertifiedland (SlDL), land moderately(SlDL), moderately desertified desertified land (MDL), land and (MDL), severely and desertified severely desertified land (SeDL). land (SeDL). Table 1. Land classification system in the Zhenglan Banner.

ADL ClassificationTable Terrain 1. Land classification system in Vegetationthe Zhenglan Banner. Soil ADL There is no obvious sand-drift TerrainPsammophytes are companionVegetation or Soil The terrainClassification is generally flat activity, the biological crust occasional species, and the NDL with open grasslands or coverageThere and is no silt obvious and clay sand content-drift The terrain isvegetation gener- Psammophytes coverage is greater are compan- lowlands between hills are high,activity, and the the mobile biological sand crust area ally flat with open ionthan or 80%occasional species, and NDL coverageis less and than silt 15% and clay con- grasslands or low- the vegetation coverage is A smalltent are amount high, and of sand-drift the mobile Psammophytes become the main The topography is not stronglands between hills greater than 80% activity occurs, which is mostly companion species, and the sand area is less than 15% SlDL and is usually flat inter-dunes covered by a biological crust or soil vegetation coveragePsammophytes decreases become to the A small amount of sand-drift or fixed dunes The topography is layer; the area of mobile sand is betweenmain 50% companion and 80% species, and activity occurs, which is mostly not strong and is between 15% and 40% SlDL the vegetation coverage de- covered by a biological crust or usually flat inter- Psammophytescreases becometo between the 50% andSand-drift soil layer; activity the area is strong, of mobile the Semi-fixed dunes withdunes small or fixeddominant dunes species, and the biological crust area is greatly MDL 80% sand is between 15% and 40% blowouts vegetation coverage is from 30% reduced, and the mobile sand area Psammophytes become the Sand-drift activity is strong, the to 50% is between 40% and 60% Semi-fixed dunes dominant species, and the biological crust area is greatly MDL Sand-drift activity is strong, the The topography is undulating,with small blowouts vegetation coverage is from reduced, and the mobile sand content of silt and clay particles is which includes semi-mobile There are only psammophytes30% to 50% area is between 40% and 60% SeDL extremely low, the mobile sand area and mobile sand dunes with with a coverage less than 30% exceeds 60%, and the crust surface large blowouts is small

2.3. Monitoring Period Selection The five periods of 2000, 2004, 2010, 2015, and 2019 were selected to analyze the aeolian desertification dynamics in the study area. These dates were selected for the following two reasons: First, precipitation is one of the most important factors that affect vegetation conditions and plays an important role in the development of aeolian desertification. Thus, Remote Sens. 2021, 13, x FOR PEER REVIEW 6 of 26

The topography is Sand-drift activity is strong, the undulating, which There are only psammo- content of silt and clay particles includes semi-mobile SeDL phytes with a coverage less is extremely low, the mobile and mobile sand than 30% sand area exceeds 60%, and the dunes with large crust surface is small blowouts

2.3. Monitoring Period Selection

Remote Sens. 2021, 13, 1730 The five periods of 2000, 2004, 2010, 2015, and 2019 were selected to analyze the6 aeo- of 27 lian desertification dynamics in the study area. These dates were selected for the following two reasons: First, precipitation is one of the most important factors that affect vegetation conditions and plays an important role in the development of aeolian desertification. Thuit iss, only it is meaningfulonly meaningful to analyze to analyze changes changes in desertification in desertification under under the same the same precipitation precipi- tationlevel, level otherwise, otherwise it is difficult it is difficult to determine to determine land land degradation degradation within within a certain a certain period. period. The Thestandardized standardized precipitation precipitation index index (SPI) (SPI) from from 2000 2000 to 2019to 2019 was was calculated calculated in Rin studio R studio by bycollecting collecting monthly monthly precipitation precipitation data data for meteorological for meteorological stations stations in the in study the areastudy as area shown as shownin Figure in 3Figure. We found 3. We thatfound the that mean the SPImean values SPI value of vegetations of vegetation growth growth for 2000, for 2004,2000, 2010,2004, 2010,2015, 2015, and 2019 and were2019 allwere moderately all moderately moist. moist. Second, Second, 2000 was 2000 the was first the year first of theyear Beijing– of the BeijingTianjin– SandstormTianjin Sandstorm Source ControlSource Control Project. Project The five. The time five periods time periods were relatively were relatively evenly evenlyspaced spaced with a differencewith a difference of approximately of approximately 5 years. 5 Thus,years. these Thus, years these represent years represent the overall the overallaeolian aeolian desertification desertification conditions conditions over the over 20 yearsthe 20 of years project of implementation.project implementation.

Figure 3. SPISPI results results:: The The SPI SPI is is drought drought index index that that is is widely widely used used including including by by the the World World Meteorological Meteorological Organization. It reflectsreﬂects the regional regional drought drought and and flood ﬂood situation situation in in a a given given time time scale scale by by calculating calculating the the cumulat cumulativeive probability probability of of rainfall. rainfall. The time scale in this study of the SPI was 3 months. The time scale in this study of the SPI was 3 months.

2.4.2.4. Data Data Sets Sets and and Preprocessing Preprocessing ThisThis study study was was conducted conducted at ata county a county level level in Zhenglan in Zhenglan Banner Banner where where medium medium--res- olutionresolution remote remote sensing sensing data datawas wasappropriate. appropriate. The Landsat The Landsat series seriesof satellites of satellites have a havelong runninga long running time, comprehensive time, comprehensive image coverage, image coverage, and rich band and richinformation. band information. The surface The re- flectancesurface reflectance products for products Landsat5 for and Landsat5 Landsat8 and are Landsat8 provided are in providedGEE, and inthe GEE, pixel and values the ofpixel each values band of are each the band corrected are the surface corrected reflectance, surface reflectance, which can which be used can directly.be used directly. In this study,In this the study, Landsat5 the Landsat5 SR Tier1 SRand Tier1 Landsat8 and Landsat8SR Tier1 products SR Tier1 were products used werefor the used “greenest” for the composite.“greenest” First, composite. all images First, from all images June to from September June to in September the study inarea the were study filtered. area were The cloudsfiltered. and The shadows clouds and in all shadows images in were all images then mas wereked then. Finally, masked. the Finally,normalized the normalized difference vegetationdifference vegetationindex (NDVI) index of (NDVI)the images of the was images calculated was calculatedand added and to the added image to theas a image new band.as a new The band. “qualityMosaic” The “qualityMosaic” function was function used was to composite used to composite all images all basedimages on based the on the maximum NDVI, and the annual remote sensing image was obtained by cropping the boundary of the Zhenglan Banner. As this paper focused on the ADL, water bodies, salinized land, and human settlements were extracted and masked in advance to avoid interference from these land-use types. Precipitation and temperature are important meteorological factors in the aeolian desertification development process. To perform the residual analysis in Section 2.6, this study collected Chinese monthly total precipitation and average temperature images from 2000 to 2017, with a resolution of 250 m. These images were based on the data of more than 1000 meteorological stations and interpolated by AUSPLINE. We selected images of vegetation growth period (May to September) in Zhenglan Banner for further analysis. The Remote Sens. 2021, 13, x FOR PEER REVIEW 7 of 26

maximum NDVI, and the annual remote sensing image was obtained by cropping the boundary of the Zhenglan Banner. As this paper focused on the ADL, water bodies, salinized land, and human settlements were extracted and masked in advance to avoid interference from these land-use types. Precipitation and temperature are important meteorological factors in the aeolian desertification development process. To perform the residual analysis in Section 2.6, this study collected Chinese monthly total precipitation and average temperature images from Remote Sens. 2021, 13, 1730 2000 to 2017, with a resolution of 250 m. These images were based on the data of 7more of 27 than 1000 meteorological stations and interpolated by AUSPLINE. We selected images of vegetation growth period (May to September) in Zhenglan Banner for further analysis. The MOD13Q1 NDVI data were called in GEE, which was also used for the residual anal- MOD13Q1ysis. In addition, NDVI statistical data were data called from in GEE,Zhenglan which Banner was also meteorological used for the residualstation from analysis. 2000 In addition, statistical data from Zhenglan Banner meteorological station from 2000 to 2019 to 2019 were collected for analysis. were collected for analysis. A field investigation of the Zhenglan Banner was performed along with photographs A field investigation of the Zhenglan Banner was performed along with photographs collected using a UAV in early August of 2019. Global positioning system (GPS) equip- collected using a UAV in early August of 2019. Global positioning system (GPS) equipment ment was used to locate various ADL types as classified using Table 1, and a total of 130 was used to locate various ADL types as classified using Table1, and a total of 130 sample sample points were collected. In addition, a same field investigation was performed in the points were collected. In addition, a same field investigation was performed in the Zhenglan Zhenglan Banner in early September 2010 and 12 sample points were collected. In the Banner in early September 2010 and 12 sample points were collected. In the follow-up follow-up study, these 142 sample points were used to verify the accuracy of ADL classi- study, these 142 sample points were used to verify the accuracy of ADL classification in fication in Landsat images. Landsat images. A variety of studies [30,31] have used the DJI Phantom Series UAVs for collection of A variety of studies [30,31] have used the DJI Phantom Series UAVs for collection ofaerial aerial photographs photographs due due to tothe the low low cost, cost, simple simple operation, operation, and and mature mature system system.. The The DJI DJIPhantom4 Phantom4 pro prov2 UAV v2 UAV was wasselected selected for image for image acquisition, acquisition, and 10 and representative 10 representative aerial aerialsample sample plots with plots different with different ADL ADLtypes typeswere wereselected selected (Figure (Figure 4). A4 total). A totalof 20 of aerial 20 aerial pho- photographytography sorties sorties were were performed. performed. The UAV The UAV operated operated at a height at a height of 80 ofm 80with m a with 75% a over- 75% overlaplap of the of heading the heading and anda 65% a 65% side side overlap. overlap. A total A total of 4382 of 4382 images images were were obtained, obtained, and and the theUAV UAV image image preprocessing preprocessing was was based based on on the the structure structure from from motion motion (SfM) (SfM) method inin thethe pix4dmapper.pix4dmapper. TheThe digitaldigital orthophotoorthophoto mapmap (DOM)(DOM) andand digitaldigital surfacesurface modelmodel (DSM)(DSM) werewere generated,generated, andand thethe pixelpixel sizesize waswas within 4 cm. Finally, 1313 high-qualityhigh-quality UAVUAV imagesimages werewere obtainedobtained thatthat coveredcovered aa totaltotal areaarea ofof 1.851.85 kmkm22..

FigureFigure 4.4. SampleSample plotplot distributiondistribution taken taken with with a a UAV. UAV The. The 10 10 aerial aerial sample sample plots plots included include alld all levels levels of theof the ADL ADL and and were were evenly evenly distributed distributed in in the the study study area, area which, which were were representative representative of of the the aeolian aeo- desertiﬁcationlian desertification status status in Zhenglan in Zhengl Banner.an Banner.

2.5.2.5. UAVUAV ImageImage ProcessingProcessing The UAV images can attain high-precision classification results because of their high spatial resolution. In this study, UAV images were classified as mobile sandy land and non- mobile sandy land. The classification results were then upscaled to obtain the mobile sand ratio (MSR) of the corresponding pixels in the Landsat images, which were used to verify the accuracy of the linear spectral mixture model (LSMM). A total of 11 features were used for the UAV image classification: the original RGB bands, DSM, gray level co-occurrence matrix (GLCM) mean, and 5 vegetation indices (Table A2). Before classification, visual interpretation was used to select samples based on the UAV images. A total of 255 samples were obtained, including 131 mobile and 124 non-mobile sand samples. In R studio, the samples were randomly divided into two subsets at a ratio of 1:2 and 75 verification Remote Sens. 2021, 13, 1730 8 of 27

samples, and 180 training samples were obtained. Then, the random forest (RF) model was built based on the “randomForest” package in R studio. This package has two primary parameters to adjust: the number of trees (ntree) and the number of variables (mtry). The grid search method was used to optimize the parameters, showing that the optimal model is obtained when ntree is 600 and mtry is 1. RF can obtain high accuracy classification results when the number of samples is limited, which is suitable for UAV image classification in this study. A total of 13 UAV images were classified in R studio using the optimal RF model. The method developed by Kattenborn et al. [32] was used to upscale the classification results to match the Landsat pixels. First, the mobile sand pixels were assigned a value of 1 and the non-mobile sand pixels were assigned 0. Second, the results were input into GEE, and the “reduceResolution” command was used to calculate the mean value of it in the corresponding Landsat pixels to obtain the UAV-MSR.

2.6. ADL Extraction Based on Spectral Mixture Analysis The southern part of the study area is part of the farming pastoral ecotone, where a large area of bare soil appeared from the field survey. Although the vegetation coverage in these areas was relatively low, there was no sand-drift activity, so it could not be classified as ADL. The spectral mixture analysis (SMA) can obtain proportions of various land-use types to give an abundance of each endmember. This method can distinguish mobile sand from other land cover types, and the increase in MSR can represent the enhancement of sand-drift activity and the aggravation of aeolian desertification. In this paper, the range of ADL in the study area was extracted from the endmember abundance of the mobile sand obtained from the LSMM (LSMM-MSR). In an LSMM, the reflectance of each pixel over the spectral bands is presented as a linear combination of the reflectance of each endmember and its relative abundance, which is defined as m ρ(λi) = ∑ Fjρj(λi) + ε(λi) (1) j=1

where j = 1,2, ... , m is the endmember; i = 1,2, ... , n is the spectral band; ρ(λi) is the reflectance of mixed pixels; ρj(λi) is the reflectance of endmember j in band i; Fj is the abundance of endmember j in the pixel; and ε(λi) is the difference between the actual and modeled reflectance. Field investigations suggested that the three endmembers of mobile sandy land, vegetation, and bare soil can represent the land cover in this area. To find “pure” spectral endmembers, the preprocessed Landsat8 images in 2019 and Landsat5 images in 2010 were output from GEE and were processed using the minimum noise fraction (MNF) rotation in ENVI5.3. Finally, the abundance images of the three endmembers were calculated using the “unmix” command in GEE.

2.7. Dynamic Change Evaluation of ADL 2.7.1. Construction of Desertification Index To select the most suitable indicators for desertification monitoring and construct the desertification index (DI), 14 indicators (Table A3) were initially selected from the aspects of vegetation, soil, humidity, and surface radiation. These indicators were calculated in GEE based on the Landsat image of 2019 (Figure A1). However, selecting too many redundant indicators reduces the accuracy and efficiency of aeolian desertification monitoring. A total of 130 samples obtained from a field investigation in 2019 were used for the gain ratio (Figure A2) and RF (Figure A3) feature selection, and a correlation analysis (Figure A4) was performed for all indicators. Finally, 4 indicators (albedo, NDVI, wetness, and TGSI) with high importance and low correlation were selected to construct the DI. The albedo is one of the most frequently used indicators in ADL monitoring [33,34] and represents the ratio of the solar radiation flux that is reflected by the surface to the incident solar radiation flux. The topsoil grain size index (TGSI) was proposed by Xiao et al. [35], which reflects the mechanical composition of the topsoil and is positively Remote Sens. 2021, 13, 1730 9 of 27

correlated with the fine sand content. Land surface humidity also has a strong correlation with the degree of aeolian desertification. The wetness of the tasseled cap transform can reflect the humidity status of the soil and vegetation, which has been often used to monitor ecological environments [36]. The NDVI is sensitive to vegetation changes and can reflect the growth status of vegetation on the ADL. The analytic hierarchy process (AHP) has been widely used in the study of aeolian desertification [37,38] and can reveal the relative importance of different desertification indicators based on the study area. The weight of each variable in the AHP is determined from the matrix of pairwise comparisons, where the importance of indices is ranked from 1 to 7, otherwise they are assigned 1/2 to 1/7 (Table2). Based on many relevant experts and previous research results [17], the final comparison matrix is as shown in Table3.

Table 2. Relative importance of two indicators.

1/7 1/6 1/5 1/4 1/3 1/2 1 2 3 4 5 6 7 Extremely low Strong Moderate Equal Moderate Strong Extreme important

Table 3. Index comparison matrix.

Albedo NDVI TGSI Wetness Weights Albedo 1 2 3 5 0.482 NDVI 1/2 1 2 3 0.272 TGSI 1/3 1/2 1 2 0.157 Wetness 1/5 1/3 1/2 1 0.088

As the dimensions of each index are not uniﬁed, they should be standardized before any analysis. The standardization methods for the NDVI sand wetness are shown in Equation (2), those for the TGSI and albedo are shown in Equation (3), and the DI is shown in Equation (4): maxXi − Xi Pi = (2) maxXi − minXi

Xi − minXi Pi = (3) maxXi − minXi n DI = ∑ Wi × Pi (4) i=1

where Xi represents the ith index, Pi is the standardization value of Xi, Wi is the weight of the ith indicator, and DI is a dimensionless value. A larger DI gives a higher degree of aeolian desertiﬁcation and a worse ecological status.

2.7.2. Classification and Dynamic Change in ADL The ADL extracted in Section 2.4 was classified into SlDL, MDL, and SeDL based on the DI. The accuracy of the classification results was verified using 142 samples obtained from the field investigation, and the area of each ADL type was counted in GEE. In addition, the annual change rate for each type of ADL was calculated as

U(b,i) − U(a,i) 1 ∆U = × × 100% (5) U(a,i) T

where ∆U is the change rate of a land type, U(a,i) is the area of type i at the beginning of a certain period, U(b,i) is the area of type i at the end of a certain period, and T is the duration of the study period. To further analyze the spatiotemporal dynamic change of the ADL in the study area over the past 20 years, the change patterns were classiﬁed into signiﬁcantly reserved, Remote Sens. 2021, 13, 1730 10 of 27

reserved, stable, developed, and seriously developed. Reserved is the desertification degree in a certain pixel decreased by a level. Significantly reserved is the desertification degree in a certain pixel decreased by two or more levels (e.g., a change from SeDL to SlDL). Stable is the desertification degree in a certain pixel with no change. Developed is the desertification degree in a certain pixel increased by one level. Seriously developed is a certain pixel with a cross-level increase in the degree of desertification.

2.8. Residual Analysis Residual analysis judges the effects of human activities on aeolian desertification. The impact of human activities is reflected primarily on vegetation. Based on previous studies [21,39], this paper selected the NDVI to characterize the effects of human activities on aeolian desertification. The mean of the NDVI in the growing season from 2000 to 2017 was calculated in GEE. The regression model for the meteorological factors (average temperature and total precipitation during the growing season) and the NDVI was established using the random forest algorithm in GEE to obtain the NDVI prediction (NDVIpredict). The NDVIpredict represents the aeolian desertification as affected only by climate. The calculation method of ResidualNDVI is shown in Equation (6), which represents the degree of influence of human activities on aeolian desertification.

ResidualNDVI = NDVI − NDVIpredict (6)

The Sen slope of the ResidualNDVI from 2000 to 2017 was then calculated, and the Mann–Kendall (M–K) significance test was performed. This paper divides human activities into significant positive, insignificant, and significant negative effects based on the impact on the ecological environment. When the Sen slope is >0 and p < 0.05 in the M–K test, it was considered a significant positive effect, which indicates that human activities positively influence ecological conditions and may promote the reversal of aeolian desertification. When p > 0.05, it was considered an insignificant effect, indicating that human activities have no meaningful impact on aeolian desertification. When the Sen slope is <0 and p < 0.05, it was considered as a significant negative effect, indicating that human activities negatively influence ecological conditions, and unreasonable land use may promote aeolian desertification.

3. Results 3.1. UAV-Based Mapping of MSR Based on the input features and the RF model after parameter adjustments, the overall classification accuracy of 13 UAV images reached 97.6% and the kappa coefficient reached 0.95. The high-precision classification results helped to obtain accurate UAV-MSR samples (Figure5). After manually deleting pixels with abnormal values and incomplete coverage, a total of 2428 pixels were used to determine the UAV-MSR samples, which was then used to verify the accuracy of the LSMM. Compared with selecting samples from the field and visually determining the MSR in previous aeolian desertification studies [40,41], UAVs provide a larger number of more objective and accurate samples with faster acquisition speeds. This effectively compensates for the scale gap between the samples and remote sensing image pixels. Remote Sens. 2021, 13, 1730 11 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 11 of 26

Figure 5. Results of the UAV-MSR: (A) and (B) are two UAV images and their UAV-MSR. Image (A) Figure 5. Results of the UAV-MSR: (A) and (B) are two UAV images and their UAV-MSR. Image is(A composed) is composed of large of large mobile mobile dunes. dunes. In image In image (B), ( theB), the semi-fixed semi-fixed dunes dunes and and inter-dune inter-dune land land with higherwith higher vegetation vegetation coverage coverage are the are main the land main types. land types. 3.2. Extraction of the ADL 3.2. Extraction of the ADL 3.2.1. Accuracy Analysis 3.2.1. Accuracy Analysis Based on the UAV-MSR sample pixels obtained in Section 3.1, we obtained the cor- respondingBased LSMM-MSRon the UAV-MSR pixels sample for the pixels Landsat obtained images in inSection GEE. 3.1, Then, we we obtained analyzed the thecor- LSMM-MSRresponding basedLSMM on-MSR the UAV-MSRpixels for the to verify Landsat the images accuracy in of GEE. the LSMM.Then, we The analyzed regression the analysisLSMM- resultsMSR based show on that the the UAV two-MSR data setsto verify had a the good accuracy linear relationshipof the LSMM. with The an regressionR2 value ofanalysis 0.7648 andresults RMSE show of that 0.1383 the (Figure two data6). Thissets had indicates a good that linear the relationship LSMM accurately with an obtained R2 value theof 0.7648 MSR. and RMSE of 0.1383 (Figure 6). This indicates that the LSMM accurately obtained the MSR.The LSMM-MSR was used to extract the ADL. The experiments showed that the ADL of the study area could be accurately extracted when the LSMM-MSR was 10–20%. To further determine the optimal threshold, 142 samples obtained from field surveys in 2019 and 2010 were used to verify the classification accuracy as shown in Table4. The highest classification accuracy was obtained when the LSMM-MSR was greater than 18%. Based on this threshold, the preprocessed images were used to extract the ADL of the study area in GEE. Remote Sens. 2021, 13, 1730 12 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 12 of 26

1 0.9 0.8 0.7

0.6

MSR -

0.5 UAV 0.4 0.3 0.2 R² = 0.7648 RMSE=0.1383 0.1 0 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 LSMM-MSR

FigureFigure 6. Linear 6. Linear regression regression analysis analysis of the of theUAV UAV-MSR-MSR and and LSMM LSMM-MSR.-MSR.

TableThe 4. LSMMExtraction-MSR accuracy was used of the to extract ADL. the ADL. The experiments showed that the ADL of the studyLSMM-MSLA area could be accurately10% extracted 12% when the 14% LSMM 16%-MSR was 18% 10–20%. 20% To further determine the optimal threshold, 142 samples obtained from field surveys in 2019 Correctly classified samples 108 115 120 127 129 121 and 2010 were used to verify the classification accuracy as shown in Table 4. The highest Accuracy 76.1% 81.0% 84.5% 89.4% 90.8% 85.2% classification accuracy was obtained when the LSMM-MSR was greater than 18%. Based on this threshold, the preprocessed images were used to extract the ADL of the study area in GEE.3.2.2. Extent of the ADL The ADL extraction results of the study area from 2000, 2004, 2010, 2015, and 2019 Tableare 4. shown Extraction in Figure accuracy7. ADL of the was ADL mainly. distributed in the north of Zhenglan Banner and concentrated in the eastern and western parts. Over the past 20 years, the ADL area in the LSMM-MSLA 10% 12% 14% 16% 18% 20% east decreased significantly, and now it is mainly concentrated in the northwest. Statistical Correctly classified samples 108 115 120 127 129 121 results of the ADL, shown in Figure8, indicate a decreasing trend in the study area over Accuracy 76.1% 81.0% 84.5% 89.4% 90.8% 85.2% the past 20 years. In 2000, the ADL reached 5161.1 km2, which accounts for 50.5% of the study area. From 2000 to 2004, the ADL rapidly decreased to 3097.5 km2, and its proportion 3.2.2. Extent of the ADL decreased to 30.3%. From 2004 to 2010, the desertification showed a rebounding trend, andThe the ADL ADL extraction grew to 4462.9results kmof 2t,he which study accounts area from for 2000, 43.7%. 2004, After 2010, 2010, 2015, desertification and 2019 arecontinued shown in toFigure reverse. 7. ADL In 2015, was themainly ADL distributed decreased toin 3736.6the north km 2of, which Zhenglan accounts Banner for and 36.6%, concentratedand in 2019, in itthe decreased eastern and to 3179.7 western km part2, whichs. Over accounts the past for20 years, 31.1%. the Although ADL area there in the were eastsome decreased fluctuations, significantl the desertificationy, and now it is trend mainly in Zhenglanconcentrated Banner in the was northwest. noticeably Statistical reversed. results of the ADL, shown in Figure 8, indicate a decreasing trend in the study area over the past 20 years. In 2000, the ADL reached 5161.1 km2, which accounts for 50.5% of the study area. From 2000 to 2004, the ADL rapidly decreased to 3097.5 km2, and its proportion decreased to 30.3%. From 2004 to 2010, the desertification showed a rebounding trend, and the ADL grew to 4462.9 km2, which accounts for 43.7%. After 2010, desertification continued to reverse. In 2015, the ADL decreased to 3736.6 km2, which accounts for 36.6%, and in 2019, it decreased to 3179.7 km2, which accounts for 31.1%. Although there were some fluctuations, the desertification trend in Zhenglan Banner was noticeably reversed. Remote Sens. 2021, 13, 1730 13 of 27

RemoteRemote Sens. Sens. 2021 2021, 13, ,13 x ,FOR x FOR PEER PEER REVIEW REVIEW 13 13of of26 26

FigureFigureFigure 7. 7.7.Extent ExtentExtent of ofofthe thethe ADL ADL.ADL. .

DesertifiedDesertified land land Non-desertifiedNon-desertified land land

100%100%

80%80%

60%60%

40%40%

20%20%

Percentage of total area Percentage of total area 0%0% 20002000 20042004 20102010 20152015 20192019 YearYear

Figure 8. Proportion of ADL in the total area of the Zhenglan Banner. FigureFigure 8. 8.Proportion Proportion of ofADL ADL in inthe the total total area area of ofthe the Zhenglan Zhenglan Banner Banner. . 3.3. Aeolian Desertiﬁcation Dynamics from 2000 to 2019 3.3.3.3. Aeolian AeolianBased Deserti onDeserti theficationfi resultscation Dynamics Dynamics in Section from from 3.2.2 2000 2000, the to to2019 NDL 2019 was masked and the DI was used to evaluateBasedBased theon on ADLthe the results dynamicsresults in inSection inSection the study3.2.2, 3.2.2, the area. the NDL TheNDL was DI was results masked masked are and shownand the the DI in DI Figurewas was used9 used. to to evaluateevaluate the the ADL ADL dynamics dynamics in inthe the study study area area. The. The DI DI results results are are shown shown in inFig Figureure 9. 9. Remote Sens. 2021, 13, 1730 14 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 14 of 26

Figure 9. Calculation results of the DI. The DI comprehensively reflects the status of aeolian desertification in the Zhenglan FigureBanner. 9. A Calculation larger DI represents results of the a higher DI. The degree DI comprehensively of aeolian desertification. reflects the status of aeolian desertification in the Zhenglan Banner. A larger DI represents a higher degree of aeolian desertification. 3.3.1. Classification and Time Variation Characteristics of ADL 3.3.1. AccordingClassification to experiments, and Time Variation ADL could Characteristics be classified intoof ADL 3 types by DI: SlDL (DI ≤ 0.35), MDLAccording (0.35 < DI to≤ 0.41),experiments and SeDL, ADL (0.41 could < DI). be The classified confusion into matrix 3 types of theby DI: four SlDL land (DI types ≤ 0.35),(Table MDL5) was (0.35 obtained, < DI ≤ 0.41 showing), and anSeDL overall (0.41 accuracy < DI). The of confusion 78.9%. The matrix classification of the four results land typesover the(Table considered 5) was obtained, years are showing shown in an Figure overall 10 .accuracy of 78.9%. The classification results over the considered years are shown in Figure 10. TableFig 5. ureClassification 11 shows accuracythe proportions of ADL. of various land types. The SeDL area decreased from 1410.2 km2 in 2000 to 497.8 km2 in 2019, with a decrease of 9.0% of the total area. From NDL SlDL MDL SeDL 2010 to 2015, the SeDL area decreased by 5.4%, and the restoration of aeolian desertification wasNDL significant. The 39 MDL area exhibited 9 a decreasing 0 trend; however, 0 this area SlDL 4 27 32 2 surgedMDL by 8.1% from 2004 0to 2010. The SlDL 4 area decreased 21from 1977.1 km in 4 2000 to 2 1810 kmSeDL in 2019; it has changed 0 little but continued 1 to increase 3 from 2004 to 2015. 25 OverallThe accuracyannual change rates (Figure 12) of SlDL, MDL,78.9% and SeDL from 2000 to 2019 were –0.4%, –2.7%, and –3.4%. The annual change rates of SeDL for the four periods were – 5.5%, 0.2%, –10.0%, and –2.6%. It was the ADL type with the fastest decline over the 20 Figure 11 shows the proportions of various land types. The SeDL area decreased from years. The annual MDL change rate fluctuated greatly; it decreased annually by 12.4% 1410.2 km2 in 2000 to 497.8 km2 in 2019, with a decrease of 9.0% of the total area. From 2010 from 2000 to 2004 and increased annually by 15.3% from 2004 to 2010. The annual rates of to 2015, the SeDL area decreased by 5.4%, and the restoration of aeolian desertification was SlDLsignificant. were – The11.1 MDL%, 8.0 area%, 6.6 exhibited%, and a–4.2 decreasing% for the trend; four however,periods. The this three area surgedtypes of by ADL 8.1% showfromed 2004 a downward to 2010. The trend SlDL over area the decreased past 20 from years, 1977.1 especially km2 in the 2000 area tos 1810of SeDL km2 andin 2019; MDL it droppedhas changed rapidly. little T buthis continued indicates tothat increase the aeolian from 2004desertification to 2015. of Zhenglan Banner reversed and the grassland vegetation status achieved marked improvement.

Table 5. Classification accuracy of ADL.

NDL SlDL MDL SeDL NDL 39 9 0 0 SlDL 4 27 3 2 MDL 0 4 21 4 SeDL 0 1 3 25 Overall accuracy 78.9% Remote Sens. 2021, 13, 1730 15 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 15 of 26 Remote Sens. 2021, 13, x FOR PEER REVIEW 15 of 26

FigureFigure 10. 10.ClassificationClassiﬁcation map map in the in the Zhenglan Zhenglan Banner Banner.. Figure 10. Classification map in the Zhenglan Banner.

25% 25% 20% 20% 15% 15% 10% 10% 5%

5% Proportion in Proportionin total area

Proportion in Proportionin total area 0% 0% 2000 2004 2010 2015 2019 2000 2004 Year2010 2015 2019 Year Slight Moderate Severe Slight Moderate Severe . . Figure 11. Proportions of the ADL area at all levels. FigureFigure 11. 11. ProportionsProportions of of the the ADL ADL area area at at all all levels levels.. The annual change rates (Figure 12) of SlDL, MDL, and SeDL from 2000 to 2019 were –0.4%, –2.7%, and –3.4%. The annual change rates of SeDL for the four periods were –5.5%, 0.2%, –10.0%, and –2.6%. It was the ADL type with the fastest decline over the 20 years. The annual MDL change rate ﬂuctuated greatly; it decreased annually by 12.4% from 2000 to 2004 and increased annually by 15.3% from 2004 to 2010. The annual rates of SlDL Remote Sens. 2021, 13, 1730 16 of 27

were –11.1%, 8.0%, 6.6%, and –4.2% for the four periods. The three types of ADL showed a downward trend over the past 20 years, especially the areas of SeDL and MDL dropped rapidly. This indicates that the aeolian desertiﬁcation of Zhenglan Banner reversed and the Remote Sens. 2021, 13, x FOR PEER REVIEWgrassland vegetation status achieved marked improvement. 16 of 26

20% 15% 10% 5% 0% -5%

Annual Annual change rate -10% -15% 2000-2004 2004-2010 2010-2015 2015-2019 2000-2019 Year Slight Moderate Severe

Figure 12. Annual change rate of the ADL area. Figure 12. Annual change rate of the ADL area. 3.3.2. Spatial Variation Characteristics of ADL 3.3.2. Spatial Variation Characteristics of ADL The spatial dynamics of the ADL in the study area are shown in Figure 13. The statisticsThe spatial in Figure dynami 14 showcs of thatthe ADL the area in the of study stable area during are theshown two in periods Figure is 13. the The largest, sta- tisticswhich in suggests Figure 14 that show the that change the area in aeolian of stab desertificationle during the two was periods relatively is the smooth largest, and which there suggestswere no that particularly the change drastic in aeolian changes. desertification Between 2000 was and relatively 2004, there smooth was and a strong there reverse were notrend partic inularly aeolian drastic desertification, changes. Between and the areas2000 and of reversed 2004, there and was significantly a strong reverse reversed trend land inwere aeolian larger desertification, than in other and years. the areas The areasof reversed of developed and significantly and seriously reversed developed land were land largerwere smallerthan in thanother in years. other years The areas and wereof developed concentrated and mostly seriously in the developed eastern region. land were From smaller2004 to than 2010, in bothother the years developed and were and concentrated seriously developedmostly in the land eastern areas region. were significantly From 2004 tolarger 2010, thanboth thosethe developed in other years, and seriously which took developed up the majority land area ofs thewere image significantly as illustrated larger in thanFigure those 13. in The other reversed years, areawhich was took distributed up the majority mostly inof thethe originalimage as SeDL illustr inated the northwestin Figure 13.and The east. reversed From 2010area towas 2015, distributed the reversed mostly area in accountedthe original for SeDL 20.9% in ofthe the northwest total, and and the east.significantly From 2010 reversed to 2015, areathe reversed accounted area for accounted 6.5%. Few for developed 20.9% of the areas total were, and distributedthe signif- icantlyin the reversed middle of area the studyaccounted region. for From6.5%. 2015Few todeveloped 2019, the areas stable w areaere distributed was larger thanin the in middleother years, of the and study aeolian region. desertification From 2015 stillto 2019, showed the astable reversed area trend. was larger However, than there in other were years,some and developed aeolian desertification areas in the northwest. still showed In general,a reversed the trend. total However, proportion there of reversed were some and developedsignificantly areas reversed in the landnorthwest. from 2000 In general, to 2019 reachedthe total 37%. proportion The reverse of reversed of the original and signifi- SeDL cantlyarea in reversed the east land and northwestfrom 2000 wasto 2019 the reached most significant. 37%. The Inreverse addition, of the the original total proportion SeDL area of indeveloped the east and and northwest seriously developedwas the most land significant. was only 5.4%In addition, and was the scattered total proportion at the eastern of developedand western and edges. seriously developed land was only 5.4% and was scattered at the eastern and western edges. Remote Sens. 2021, 13, 1730 17 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 17 of 26 Remote Sens. 2021, 13, x FOR PEER REVIEW 17 of 26

FigureFigure 13. 13. AeolianAeolian deserti desertiﬁcationfication spatial spatial distribution distribution dynamic dynamic-change-change maps maps.. Figure 13. Aeolian desertification spatial distribution dynamic-change maps.

90% 90% 80% 80% 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20%

20% Proportionin total area Proportion in Proportionin total area 10% 10% 0% 0% 2000-2004 2004-2010 2010-2015 2015-2019 2000-2019 2000-2004 2004-2010 2010-2015 2015-2019 2000-2019 Year Year Seriously developed Developed Stable Reversed Significantly reversed Seriously developed Developed Stable Reversed Significantly reversed

Figure 14. Statistical results of spatial dynamic change. Figure 14. StatisticalFigure results 14. ofStatistical spatial dynamic results of change spatial. dynamic change. 3.4.3.4. Impact Impact of of Human Human Activities Activities on on Aeolian Aeolian Desertifi Desertificationcation 3.4. Impact of Human Activities on Aeolian Desertification FigFigureure 15 15 shows shows that that areas areas with with significant significant positive positive effects effects of of human human activities activities were were Figure 15 shows that areas with significant positive effects of human activities were distributeddistributed primarily primarily in in the the middle middle eastern eastern and and western western parts parts of of the study area area.. These These distributed primarily in the middle eastern and western parts of the study area. These areasareas have have the the most most serious serious aeolian aeolian desertification desertification and and are are the the significant significant reversal reversal areas, areas, areas have the most serious aeolian desertification and are the significant reversal areas, showshowinging that that various various desertification desertification control control projects projects have have played played important important roles roles in in eco- eco- showing that various desertification control projects have played important roles in eco- logicallogical restoration. restoration. Some Some areas areas in in the the south south also also showed showed significant significant positive positive effects effects from from logical restoration. Some areas in the south also showed significant positive effects from humanhuman activities. activities. The The visual visual interpretation interpretation indicates indicates that that th theseese areas areas were were newly newly increased increased human activities. The visual interpretation indicates that these areas were newly increased cultivated land over the past 20 years, and the vegetation status was improved from crop cultivated land over the past 20 years, and the vegetation status was improved from crop planting, which also explains the rationality of the residual analysis. Human activities in planting, which also explains the rationality of the residual analysis. Human activities in Remote Sens. 2021, 13, x FOR PEER REVIEW 18 of 26

the eastern and northwestern parts of the study area have significant negative effects that may be related to overgrazing. Thus, increased human activities in these areas played a role in promoting desertification. The significant negative effects in the south are mostly Remote Sens. 2021, 13, 1730 18 of 27 related to the decreased cultivated land area. The statistical results showed that the area with insignificant effects accounted for 59.1% of the total, indicating that human activities had no serious impact on aeolian des- ertificationcultivated landon more over than the pasthalf of 20 the years, land. and The the proportion vegetation of status the area was with improved significant from pos- crop itiveplanting, effects which was 20.0%, also explains and the the proportion rationality of ofsignificant the residual negative analysis. effects Human was 21.0%. activities This in indicatesthe eastern that and the northwestern area with positive parts ofand the negative study area effects have of significant human activities negative in effectsthe study that areamay were be related nearly to the overgrazing. same over the Thus, past increased 20 years. human With the activities development in these of areas various played eco- a logicalrole in projects, promoting the desertification.original SeDL had The significantly significant negative improved, effects but inthere the are south still are large mostly ar- easrelated that suffer to the decreasedfrom the unreasonable cultivated land human area. activities.

FigureFigure 15. 15. SpatialSpatial distribution distribution of of human human activities activities on on aeolian aeolian d desertificationesertification in in the Zhenglan Ban- Banner. nerHuman. Human activities activities in the in greenthe green areas areas were were conducive conducive to the to reversal the reversal of desertification. of desertification. Human Human activities activitiesin the red in areasthe red made areas desertification made desertification more serious. more serious. Human Human activities activities in the in yellow the yellow areas areas had no haobviousd no obvious impact, impact, and climate and climate factors factors played played a major a role.major role.

4. DiscussionThe statistical results showed that the area with insignificant effects accounted for 59.1%In this of the study, total, a ground indicating survey, that UAV human images activities, and hadsatellite no seriousimages impactwere used on to aeolian ana- lyzedesertification the dynamic on of more aeolian than desertification. half of the land. Data The from proportion three different of the area scales with were significant effec- tivelypositive integrated. effects was In 20.0%,other stud andi thees of proportion vegetation of coveragesignificant [3 negative2], the satellite effects– wasaviation 21.0%.– groundThis indicates integration that method the area w withas often positive used, and but negativeit was rarely effects used of humanin desertification activities mon- in the itoring.study area In addition, were nearly MSR the is samemore overeffective the past than 20 vegetation years. With coverage the development in extracting of various ADL. Someecological existing projects, studies the in originalthe Zhenglan SeDL Banner had significantly used the vegetation improved, coverage but there [4 are2,43 still] to clas- large sifyareas the that ADL, suffer which from gave the incorrect unreasonable classifications human activities. for most of the bare soil in the south as SlDL4. Discussion or even MDL. Moreover, a multi-index method can more reasonably reflect the degree of ADL. As desertification is a complex process, a single index can only show one In this study, a ground survey, UAV images, and satellite images were used to analyze aspect of change [17]. The methods used in this paper can provide reference for rapid and the dynamic of aeolian desertification. Data from three different scales were effectively accurate monitoring of aeolian desertification in other areas. integrated. In other studies of vegetation coverage [32], the satellite–aviation–ground integration method was often used, but it was rarely used in desertification monitoring. In addition, MSR is more effective than vegetation coverage in extracting ADL. Some existing studies in the Zhenglan Banner used the vegetation coverage [42,43] to classify the ADL, which gave incorrect classifications for most of the bare soil in the south as SlDL or even Remote Sens. 2021, 13, x FOR PEER REVIEW 19 of 26

The results showed that the area of ADL and the degree of desertification in the study area decreased from 2000 to 2019. This is similar to many related studies in recent years. Remote Sens. 2021, 13, 1730 Yu et al. [44] studied the vegetation change trend in the Beijing–Tianjin Sandstorm Source19 of 27 Control Project area, and the results showed that 83.3% of the vegetation belonged to restoration state from 2005 to 2015. Moreover, the monitoring results of the desertification process in Hunshandake Sandy Land by Gou et al. [45] were basically consistent with thoseMDL. in Moreover, this paper. a multi-indexAs the reversal method trend can of moreaeolian reasonably desertification reflect is the obvious degree in of this ADL. area, As itdesertification is important for is a us complex to judge process, the most a single important index driving can only factor show behind. one aspect of change [17]. The methodsThe results used of inthe this residual paper analysis can provide show reference that the foraeolian rapid desertification and accurate monitoring was not sig- of nificantlyaeolian desertification affected by human in other activities areas. for 59.1% of the land area in the study region. This meansThe that results the dynamic showed thatof aeolian the area desertification of ADL and thein most degree areas of desertification was affected inby the climate. study Sareatatistical decreased data (Figure from 2000s 16 to–18 2019.) from This Zhenglan is similar Banner to many meteorological related studies station in recent in the years. past 20Yu years et al. show [44] studied that the the annual vegetation precipitation change trendand average in the Beijing–Tianjin temperature present Sandstorm an upward Source Control Project area, and the results showed that 83.3% of the vegetation belonged to trend. The warm and humid climate was beneficial to the reversal of aeolian desertifica- restoration state from 2005 to 2015. Moreover, the monitoring results of the desertification tion [46]. Although the average annual wind speed increased significantly from 2005 to process in Hunshandake Sandy Land by Gou et al. [45] were basically consistent with those 2011, it showed a downward trend over the entire 20 years, which reduces wind erosion. in this paper. As the reversal trend of aeolian desertification is obvious in this area, it is Thus, we believe that climate is the most important factor in the change of aeolian deser- important for us to judge the most important driving factor behind. tification in the study area, which is consistent with recent relevant studies. For example, The results of the residual analysis show that the aeolian desertification was not Yang et al. [47] considered the meteorological factor in Hunshandake Sandy Land was the significantly affected by human activities for 59.1% of the land area in the study region. key factor that affects aeolian desertification, while human activities may be the most im- This means that the dynamic of aeolian desertification in most areas was affected by portant factor in other sandy lands throughout China. The increase in ADL area in 2010 climate. Statistical data (Figures 16–18) from Zhenglan Banner meteorological station in may be related to the arid climate in 2009. The total precipitation in 2009 was the lowest the past 20 years show that the annual precipitation and average temperature present in 20 years. At the same time, the annual average wind speed also reached the maximum an upward trend. The warm and humid climate was beneficial to the reversal of aeolian indesertification 2009 and 2010. [46 ]. Although the average annual wind speed increased significantly from 2005Human to 2011, activities it showed also a downward had an impo trendrtant overimpact the on entire aeolian 20 years,desertification which reduces in the study wind areaerosion.. The Thus, proportion we believe of the that total climate land isarea the with most significant important factorpositive in theeffects change from of human aeolian activitiesdesertification was 20.0%, in the most study of area,which which was in is the consistent SeDL. This with indicates recent relevantthat since studies. 2000, sand For controlexample, measures Yang et, al. such [47] aconsidereds aerial seeding the meteorological and grazing factor prohibition in Hunshandake, have played Sandy an Land im- portantwas the role key in factor promoting that affects the reversal aeolian desertification,of SeDL. However, while 21.0% human of activitiesthe area was may still be theaf- fectedmost importantby the significant factor in negative other sandy effects lands of throughouthuman activities, China. which The increase indicates in that ADL there area arein 2010still large may beare relatedas of damaging to the arid land climate-use practices in 2009. in The the total study precipitation area. It is still in 2009necessary was the to reducelowest inunreasonable 20 years. At human the same activities time, the in the annual future average and adjust wind speed resource also utilization reached the to maintainmaximum the in restoration 2009 and 2010. of ecological environment and further reduce the ADL area.

) 4.5 ℃ 4

3.5

2.5

1.5 Annual average temperature( Annual average 1 2000 2003 2006 2009 2012 2015 2018 Year FigureFigure 16. 16. AnnualAnnual average average temperature temperature change change curve curve.. Remote Sens. 2021, 13, 1730 20 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 20 of 26 Remote Sens. 2021, 13, x FOR PEER REVIEW 20 of 26

550 550 500 500 450 450 400 400 350 350 300 300 250 Annual Annual precipitation(mm) 250 Annual Annual precipitation(mm) 200 200 150 1502000 2003 2006 2009 2012 2015 2018 2000 2003 2006 2009 2012 2015 2018 Year Year FigureFigure 17. 17. AnnualAnnual precipitation precipitation change change curve curve.. Figure 17. Annual precipitation change curve. 4.6 4.6 4.4 4.4 4.2 4.2 4 4 3.8 3.8 3.6 3.6 3.4 3.4 3.2 3.2

Annual average wind speed(m/s) Annual average 3 Annual average wind speed(m/s) Annual average 32000 2003 2006 2009 2012 2015 2018 2000 2003 2006 2009Year 2012 2015 2018 Year

Figure 18. Annual average wind speed change curve. FigureFigure 18. AnnualAnnual average average wind wind speed speed change curve curve.. 5. Conclusions 5. ConclusionHuman activitiess also had an important impact on aeolian desertification in the study This paper took the Zhenglan Banner as the research area and explored the processes area.This The paper proportion took the of Zhenglan the total land Banner area as with the research significant area positive and explored effects the from processes human andactivities driving was forces 20.0%, of aeolian most of desertification which was in thefrom SeDL. 2000 Thisto 2019. indicates The main that sinceconclusions 2000, sand are and driving forces of aeolian desertification from 2000 to 2019. The main conclusions are ascontrol follows: measures, such as aerial seeding and grazing prohibition, have played an important as follows: role(1) in promotingUAVs can quickly the reversal provide of SeDL. the MSR. However, In this 21.0%paper, ofthe the UAV area images was still were affected divided by (1) UAVs can quickly provide the MSR. In this paper, the UAV images were divided intothe mobile significant sandy negative land and effects non-mobile of human sandy activities, land with which an accuracy indicates that that reached there are97.6%. still into mobile sandy land and non-mobile sandy land with an accuracy that reached 97.6%. Thelarge classification areas of damaging results were land-use used practices for resampling in the study in GEE, area. and It isa total still necessaryof 2428 UAV to reduce-MSR The classification results were used for resampling in GEE, and a total of 2428 UAV-MSR samplesunreasonable corresponding human activities to the Landsat in the future pixels and were adjust obtained resource to verify utilization the LSMM to maintain accuracy the. samples corresponding to the Landsat pixels were obtained to verify the LSMM accuracy. restoration(2) The ofADL ecological area in environmentthe study area and decreas furthered reduce over the the past ADL 20 area. years. The ADL area (2) The ADL area in the study area decreased over the past 20 years. The ADL area decreased from 5161.1 km2 in 2000 to 3179.7 km2 in 2019 but increased by 1365.4 km2 from decreased from 5161.1 km2 in 2000 to 3179.7 km2 in 2019 but increased by 1365.4 km2 from 20045. Conclusions to 2010. 2004 to 2010. (3)This Over paper the tookpast the20 years, Zhenglan the Banneraeolian asdesertification the research areaof the and study explored area showed the processes a re- (3) Over the past 20 years, the aeolian desertification of the study area showed a re- versingand driving trend. forces The area of aeolian of SlDL desertification, MDL, and SeDL from decreased 2000 to 2019. at ann Theual main rates conclusions of 0.4%, 2.7%, are versing trend. The area of SlDL, MDL, and SeDL decreased at annual rates of 0.4%, 2.7%, andas follows: 3.4%. In terms of spatial dynamic changes, the total proportion of reversed and sig- and 3.4%. In terms of spatial dynamic changes, the total proportion of reversed and significantly(1) UAVs reversed can quicklylands from provide 2000 theto 2019 MSR. reached In this paper,37%, and the the UAV total images proportion were divided of de- nificantly reversed lands from 2000 to 2019 reached 37%, and the total proportion of de- velopedinto mobile and sandyseriously land developed and non-mobile lands was sandy only land 5.4% with. an accuracy that reached 97.6%. veloped and seriously developed lands was only 5.4%. The(4 classification) Climate change results and were human used activities for resampling have jointly in GEE, promoted and a total the reversal of 2428 UAV-MSRof aeolian (4) Climate change and human activities have jointly promoted the reversal of aeolian desertificationsamples corresponding in the study to thearea Landsat, where pixelsclimate were factors obtained played to a verify major the role. LSMM The results accuracy. of desertification in the study area, where climate factors played a major role. The results of residual(2) Theanalyses ADL showed area in thethat study aeolian area desertification decreased over was the positively past 20 years. affected The b ADLy human area residual analyses showed2 that aeolian desertification2 was positively affected by human2 activitiesdecreased over from 20.0% 5161.1 of km the inland 2000 area to, 3179.7 and 21.0% km in was 2019 affected but increased by the bynegative 1365.4 effects km from of activities over 20.0% of the land area, and 21.0% was affected by the negative effects of human2004 to activities. 2010. human activities. In order to further control aeolian desertification in the study area, it is still necessary In order to further control aeolian desertification in the study area, it is still necessary to reduce unreasonable human activities and increase project investment in SeDL areas. to reduce unreasonable human activities and increase project investment in SeDL areas. Remote Sens. 2021, 13, 1730 21 of 27

(3) Over the past 20 years, the aeolian desertification of the study area showed a reversing trend. The area of SlDL, MDL, and SeDL decreased at annual rates of 0.4%, 2.7%, and 3.4%. In terms of spatial dynamic changes, the total proportion of reversed and significantly reversed lands from 2000 to 2019 reached 37%, and the total proportion of developed and seriously developed lands was only 5.4%. (4) Climate change and human activities have jointly promoted the reversal of aeolian desertification in the study area, where climate factors played a major role. The results of residual analyses showed that aeolian desertification was positively affected by human activities over 20.0% of the land area, and 21.0% was affected by the negative effects of human activities. In order to further control aeolian desertification in the study area, it is still necessary to reduce unreasonable human activities and increase project investment in SeDL areas.

Author Contributions: Conceptualization, A.C. and X.Y.; methodology, A.C. and X.Y.; software, A.C.; validation, A.C.; formal analysis, A.C.; investigation, A.C., B.X., X.Y., Y.J., J.G., X.X., D.Y., P.W. and L.Z.; writing—original draft preparation, A.C.; visualization, A.C.; funding acquisition, X.Y. All authors have read and agreed to the published version of the manuscript. Funding: This work was supported by the National Key R&D Program of China (2017YFC0506504) and National Natural Science Foundation of China (No. 41571105, 41861019, and 31372354). Data Availability Statement: Landsat data are openly available via the Copernicus Open Access Hub and the GEE. The GEE codes developed in this research and the meteorological data are available, upon any reasonable request, by emailing the authors. Acknowledgments: We would like to thank all the reviewers and editors for their helpful and constructive comments on this paper. Conﬂicts of Interest: The authors declare no conﬂict of interest.

Appendix A

Table A1. Deﬁnition of main abbreviations.

Abbreviations Definition GEE Google Earth Engine UAV Unmanned aerial vehicle ADL Aeolian desertification land DI Desertification index NDL Non-desertified land SlDL Slightly desertified land MDL Moderately desertified land SeDL Severely desertified land DSM Digital surface model DOM Digital orthophoto map SPI Standardized precipitation index GLCM Gray level co-occurrence matrix MSR Mobile sand ratio LSMM Linear spectral mixture model RF Random forest SMA Spectral mixture analysis MNF Minimum noise fraction AHP Analytic hierarchy process

Table A1 prevents the main abbreviations and their deﬁnitions in this paper. The abbreviations and deﬁnitions of all indexes are given in Tables A2 and A3. Remote Sens. 2021, 13, 1730 22 of 27

Table A2. Calculation formulas of UAV vegetation indices.

UAV Indexes Calculation Formula Visible band difference vegetation index (VDVI) [48] 2green−red−blue 2green+red+blue Normalized green–red difference index (NGRDI) [49] green−red green+red Normalized green–blue difference index (NGBDI) [50] green−blue NGBDI = green+blue red red–green ratio index (RGRI) [50] green Excess gen index (EXG) [51] 2green − red − blue Color index of vegetation (CIVE) [51] 0.441red − 0.881green + 0.385blue + 18.78745

Table A3. Calculation formula of Landsat indices.

Landsat Indexes Calculation Formula nir−red Normalized difference vegetation index (NDVI) [33] nir+red 0.356blue+0.130red+0.373nir+0.085swir+0.072swir2−0.018 Albedo [33] 1.016 q 2 Modiﬁed soil adjusted vegetation index (MSAVI) [14] 2nir+1− (2nir+1) −8(nir−red) 2 NDVI−NDWI Normalized difference drought index (NDDI) [52] NDVI+NDWI √ Soil moisture monitoring index (SMMI) [53] nir2√+swir2 2 red−blue Topsoil grain size index (TGSI) [35] red+blue+green swir−nir normalized difference soil index (NDSI) [54] swir+nir Bare soil index (BSI) [46] (swir+red)−(nir+blue) × + (swir+red)+(nir+blue) 100 100 I [55] lg(red)−lg(blue) Fe2O3 lg(red)+lg(blue)

Tables A2 and A3 are the calculation formulas of the indexes mentioned in the paper. In Table A2, green is green band reflectance, red is red band reflectance, and blue is blue band reflectance. In Table A3, blue, green, red, nir, swir, and swir2 represent the surface reflectance derived from Landsat8 bands 2, 3, 4, 5, 6, and 7 and Landsat5 bands 1, 2, 3, 4, 5, and 7. In addition to the above indices, wetness, brightness, and greenness were obtained by using tasseled cap transform in GEE. GLCM mean was obtained using the “glcmTexture” command in GEE. The LST images were from the Landsat land surface temperature App developed by Parasatidis et al. [56]. Remote Sens. 2021, 13, 1730 23 of 27

RemoteRemote Sens.Sens. 2021 2021,, 13 13,, x x FORFOR PEERPEER REVIEWREVIEW 2323 ofof 2626

FigureFigure A1 A1.A1.. ResultsResults of of the the indicators indicators for for aeolian aeolian desertification desertiﬁcationdesertification inin 20192019 inin thethe ZhenglanZhenglan Banner.Banner.

FigureFigure A2A2.A2.. SelectionSelection resultsresults ofof thethethe gain gaingain ratio. ratioratio.. Gain GainGain ratio ratioratio is isis the thethe ratio ratioratio of ofof information informationinformation gain gaingain to to theto thethe intrinsic in-in- trinsicinformation,trinsic information,information, which which iswhich used isis in usedused the C4.5 inin thethe algorithm C4.5C4.5 algorithmalgorithm for feature forfor featurefeature selection. selection.selection. Remote Sens. 2021, 13, 1730 24 of 27

Remote Sens. 2021, 13, x FOR PEER REVIEW 24 of 26 Remote Sens. 2021, 13, x FOR PEER REVIEW 24 of 26

FigureFigure A3 A3.. SelectionSelection results results of ofthe the RF. RF. RF RF evaluate evaluatess all allfeatures features by out by outof bag of bag(OOB) (OOB) data. data. If the If the accaccuracyFigureuracy A3of of .OOB Selection OOB data data resultsis is greatly greatly of thereduced reduced RF. RF after afterevaluate permuting permutings all features a feature a feature by randomly, out randomly, of bag indicating(OOB) indicating data. that If that thethis this accuracy of OOB data is greatly reduced after permuting a feature randomly, indicating that this featurefeature has has a a great great influence inﬂuence on on the the classification classiﬁcation results. results. feature has a great influence on the classification results.

Figure A4. Correlation coefficients of the desertification index. The blank grid indicates that the p- Figure A4. Correlation coefficients of the desertification index. The blank grid indicates that the p- valueFigure in A4.the correlationCorrelation analysis coefﬁcients of the of two the indicators desertiﬁcation is greater index. than The 0.05. blank grid indicates that the value in the correlation analysis of the two indicators is greater than 0.05. p-value in the correlation analysis of the two indicators is greater than 0.05. References References 1. UNEP. Development of guidelines for assessment and mapping of desertification and degradation in Asia/Pacific. In Proceedings 1. UNEP. Development of guidelines for assessment and mapping of desertification and degradation in Asia/Pacific. In Proceedings of the Draft Report of the Expert Panel Meeting, Paris, France, 17 June 1994; United Nations Environment Programme: Robbie, Kanyanaof the Draft, 1994 Report. of the Expert Panel Meeting, Paris, France, 17 June 1994; United Nations Environment Programme: Robbie, 2. Wang,Kanyana T.; Zhu,, 1994 Z.D.. Study on sandy desertification in China-1. Definition of sandy desertification and its connotation. J. Desert

2. ResWang,. 2003 T., 23; Zhu,, 209 –Z.D.214. Study on sandy desertification in China-1. Definition of sandy desertification and its connotation. J. Desert Res. 2003, 23, 209–214. Remote Sens. 2021, 13, 1730 25 of 27

References 1. UNEP. Development of guidelines for assessment and mapping of desertification and degradation in Asia/Pacific. In Proceedings of the Draft Report of the Expert Panel Meeting, Paris, France, 17 June 1994; United Nations Environment Programme: Robbie, Kanyana, 1994. 2. Wang, T.; Zhu, Z.D. Study on sandy desertification in China-1. Definition of sandy desertification and its connotation. J. Desert Res. 2003, 23, 209–214. 3. State Forestry Administration of China A Bulletin of Status of Desertification and Sandification in China. Available online: http://www.forestry.gov.cn/main/69/content-831684.html (accessed on 29 December 2015). 4. Bryan, B.A.; Gao, L.; Ye, Y.Q.; Sun, X.F.; Connor, J.D.; Crossman, N.D.; Stafford-Smith, M.; Wu, J.G.; He, C.Y.; Yu, D.Y.; et al. China’s response to a national land-system sustainability emergency. Nature 2018, 559, 193–204. [CrossRef][PubMed] 5. Yang, X.; Xu, B.; Jin, Y.; Qin, Z.; Ma, H.; Li, J.; Zhao, F.; Chen, S.; Zhu, X. Remote sensing monitoring of grassland vegetation growth in the Beijing-Tianjin sandstorm source project area from 2000 to 2010. Ecol. Indic. 2015, 51, 244–251. [CrossRef] 6. Chen, C.; Park, T.; Wang, X.; Piao, S.L.; Xu, B.D.; Chaturvedi, R.K.; Fuchs, R.; Brovkin, V.; Ciais, P.; Fensholt, R.; et al. China and India lead in greening of the world through land-use management. Nat. Sustain. 2019, 2, 122–129. [CrossRef][PubMed] 7. Zhang, C.L.; Li, Q.; Shen, Y.P.; Zhou, N.; Wang, X.S.; Li, J.; Jia, W.R. Monitoring of aeolian desertification on the Qinghai-Tibet Plateau from the 1970s to 2015 using Landsat images. Sci. Total Environ. 2018, 619, 1648–1659. [CrossRef] 8. Dharumarajan, S.; Bishop, T.F.A.; Hegde, R.; Singh, S.K. Desertification vulnerability index-an effective approach to assess desertification processes: A case study in Anantapur District, Andhra Pradesh, India. Land Degrad. Dev. 2018, 29, 150–161. [CrossRef] 9. Wu, J.A.; Liu, Y.L.; Wang, J.; He, T. Application of Hyperion data to land degradation mapping in the Hengshan region of China. Int. J. Remote Sens. 2010, 31, 5145–5161. [CrossRef] 10. Zerrouki, Y.; Harrou, F.; Zerrouki, N.; Dairi, A.; Sun, Y. Desertification Detection Using an Improved Variational Autoencoder- Based Approach Through ETM-Landsat Satellite Data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 202–213. 11. Fan, Z.M.; Li, S.B.; Fang, H.Y. Explicitly Identifying the Desertification Change in CMREC Area Based on Multisource Remote Data. Remote Sens. 2020, 12, 3170. [CrossRef] 12. Hellden, U.; Tottrup, C. Regional desertification: A global synthesis. Glob. Planet. Change 2008, 64, 169–176. [CrossRef] 13. Li, B.L. Sandy desertification trend in western Northeast China Plain in the past 10 years. Acta Geogr. Sin. 2001, 2, 54–61. 14. Duan, H.; Wang, T.; Xue, X.; Yan, C. Dynamic monitoring of aeolian desertification based on multiple indicators in Horqin Sandy Land, China. Sci. Total Environ. 2019, 650, 2374–2388. [CrossRef] 15. Wei, H.; Wang, J.; Cheng, K.; Li, G.; Ochir, A.; Davaasuren, D.; Chonokhuu, S. Desertification Information Extraction Based on Feature Space Combinations on the Mongolian Plateau. Remote Sens. 2018, 10, 1614. [CrossRef] 16. Zhang, Y.; Zhang, C.; Wang, Z.; An, R.; Li, J. Comprehensive Research on Remote Sensing Monitoring of Grassland Degradation: A Case Study in the Three-River Source Region, China. Sustainability 2019, 11, 1845. [CrossRef] 17. Jiang, L.; Jiapaer, G.; Bao, A.; Kurban, A.; Guo, H.; Zheng, G.; De Maeyer, P. Monitoring the long-term desertification process and assessing the relative roles of its drivers in Central Asia. Ecol. Indic. 2019, 104, 195–208. [CrossRef] 18. Liu, S.L.; Wang, T. Primary study on sandy desertification in Otindag sandy land and its surrounding regions. J. Soil Water Conserv. 2004, 18, 99–103. 19. Xu, X.L.; Yan, J.P. Research on quantitative relations between human factors and desertification in northern Shanxi sandy area. J. Arid Land Resour. Environ. 2005, 19, 38–41. 20. Li, S.; Zheng, Y.; Luo, P.; Wang, X.; Li, H.; Lin, P. Desertification in Western Hainan Island, China (1959 to 2003). Land Degrad. Dev. 2007, 18, 473–485. [CrossRef] 21. Kundu, A.; Patel, N.R.; Saha, S.K.; Dutta, D. Desertification in western Rajasthan (India): An assessment using remote sensing derived rain-use efficiency and residual trend methods. Nat. Hazards. 2017, 86, 297–313. [CrossRef] 22. Hansen, M.C.; Potapov, P.V.; Moore, R.; Hancher, M.; Turubanova, S.A.; Tyukavina, A.; Thau, D.; Stehman, S.V.; Goetz, S.V.; Loveland, T.R.; et al. High-Resolution Global Maps of 21st-Century Forest Cover Change. Science 2013, 342, 850–853. [CrossRef] [PubMed] 23. Defourny, P.; Bontemps, S.; Bellemans, N.; Cara, C.; Dedieu, G.; Guzzonato, E.; Hagolle, O.; Inglada, J.; Nicola, L.; Rabaute, T.; et al. Near real-time agriculture monitoring at national scale at parcel resolution: Performance assessment of the Sen2-Agri automated system in various cropping systems around the world. Remote Sens. Environ. 2019, 221, 551–568. [CrossRef] 24. Liu, X.P.; Hu, G.H.; Chen, Y.M.; Li, X.; Xu, X.C.; Li, S.Y.; Pei, F.S.; Wang, S.J. High-resolution multi-temporal mapping of global urban land using Landsat images based on the Google Earth Engine Platform. Remote Sens. Environ. 2018, 209, 227–239. [CrossRef] 25. Pekel, J.-F.; Cottam, A.; Gorelick, N.; Belward, A.S. High-resolution mapping of global surface water and its long-term changes. Nature 2016, 540, 418–422. [CrossRef][PubMed] 26. Vermeulen, C.; Lejeune, P.; Lisein, J.; Sawadogo, P.; Bouche, P. Unmanned Aerial Survey of Elephants. PLoS ONE 2013, 8, e54700. [CrossRef] 27. Yin, D.; Wang, L. Individual mangrove tree measurement using UAV-based LiDAR data: Possibilities and challenges. Remote Sens. Environ. 2019, 223, 34–49. [CrossRef] Remote Sens. 2021, 13, 1730 26 of 27

28. Solazzo, D.; Sankey, J.B.; Sankey, T.T.; Munson, S.M. Mapping and measuring aeolian sand dunes with photogrammetry and LiDAR from unmanned aerial vehicles (UAV) and multispectral satellite imagery on the Paria Plateau, AZ, USA. Geomorphology 2018, 319, 174–185. [CrossRef] 29. Xiao, Y.; Xie, G.D.; Zhen, L.; Lu, C.X.; Xu, J. Identifying the Areas Benefitting from the Prevention of Wind Erosion by the Key Ecological Function Area for the Protection of Desertification in Hunshandake, China. Sustainability 2017, 9, 1820. [CrossRef] 30. Lucas, R.; Kerchove, R.V.D.; Otero, V.; Lagomasino, D.; Dahdouh-Guebas, F. Structural characterisation of mangrove forests achieved through combining multiple sources of remote sensing data. Remote Sens Environ. 2020, 237, 111543. [CrossRef] 31. Pearce, S.; Ljubiˇci´ljubiˇcić, R.; Peña-Haro, S.; Perks, M.; Manfreda, S. An Evaluation of Image Velocimetry Techniques under Low Flow Conditions and High Seeding Densities Using Unmanned Aerial Systems. Remote Sens. 2020, 12, 232. [CrossRef] 32. Kattenborn, T.; Lopatin, J.; Foerster, M.; Braun, A.C.; Fassnacht, F.E. UAV data as alternative to field sampling to map woody invasive species based on combined Sentinel-1 and Sentinel-2 data. Remote Sens. Environ. 2019, 227, 61–73. [CrossRef] 33. Guo, B.; Wen, Y. An Optimal Monitoring Model of Desertification in Naiman Banner Based on Feature Space Utilizing Landsat8 OLI Image. IEEE Access 2020, 8, 4761–4768. [CrossRef] 34. Wei, H.S.; Wang, J.L.; Han, B.M. Desertification Information Extraction along the China-Mongolia Railway Supported by Multisource Feature Space and Geographical Zoning Modeling. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2020, 13, 392–402. [CrossRef] 35. Xiao, J.; Shen, Y.; Tateishi, R.; Bayaer, W. Development of topsoil grain size index for monitoring desertification in arid land using remote sensing. Int. J. Remote Sens. 2006, 27, 2411–2422. [CrossRef] 36. Xu, H.; Wang, M.; Shi, T.; Guan, H.; Fang, C.; Lin, Z. Prediction of ecological effects of potential population and impervious surface increases using a remote sensing based ecological index (RSEI). Ecol. Indic. 2018, 93, 730–740. [CrossRef] 37. Nandy, S.; Singh, C.; Das, K.K.; Kingma, N.C.; Kushwaha, S.P.S. Environmental vulnerability assessment of eco-development zone of Great Himalayan National Park, Himachal Pradesh, India. Ecol. Indic. 2015, 57, 182–195. [CrossRef] 38. Sadeghravesh, M.H.; Khosravi, H.; Ghasemian, S. Application of fuzzy analytical hierarchy process for assessment of combating- desertification alternatives in central Iran. Nat. Hazards. 2015, 75, 653–667. [CrossRef] 39. Evans, J.; Geerken, R. Discrimination between climate and human-induced dryland degradation. J. Arid Environ. 2004, 57, 535–554. [CrossRef] 40. Li, J.; Xu, B.; Yang, X.; Qin, Z.; Zhao, L.; Jin, Y.; Zhao, F.; Guo, J. Historical grassland desertification changes in the Horqin Sandy Land, Northern China (1985–2013). Sci. Rep. 2017, 7, 3009. [CrossRef] 41. Li, J.Y.; Yang, X.C.; Jin, Y.X.; Yang, Z.; Huang, W.G.; Zhao, L.N.; Gao, T.; Yu, H.D.; Ma, H.L.; Qin, Z.H.; et al. Monitoring and analysis of grassland desertification dynamics using Landsat images in Ningxia, China. Remote Sens. Environ. 2013, 138, 19–26. [CrossRef] 42. Wu, J.; Shi, M.C.; Ding, G.D. On the spatio-temporal pattern of desertification in Zhenglan banner, Hunshandake sandy land. Sci. Soil Water Conserv. 2019, 17, 110–119. 43. Yang, T.T.; Liu, P.T.; Liu, T.P.; Li, P.; Wu, X.H.; Lv, S.H. The remote sensing monitoring on desertification distribution pattern and its dynamic changes in Zhenglanqi area from 2002 to 2011. J. Arid Land Resour. Environ. 2013, 27, 181–185. 44. Yu, L.; Wu, Z.; Du, Z.; Zhang, H.; Liu, Y. Insights on the roles of climate and human activities to vegetation degradation and restoration in Beijing-Tianjin sandstorm source region. Ecol. Eng. 2021, 159, 106105. [CrossRef] 45. Gou, F.; Liang, W.; Sun, S.B.; Jin, Z.; Zhang, W.B.; Yan, J.W. Analysis of the desertification dynamics of sandy lands in Northern China over the period 2000–2017. Geocarto Int. 2019, 1–22. [CrossRef] 46. Guo, Q.; Fu, B.; Shi, P.; Cudahy, T.; Zhang, J.; Xu, H. Satellite Monitoring the Spatial-Temporal Dynamics of Desertification in Response to Climate Change and Human Activities across the Ordos Plateau, China. Remote Sens. 2017, 9, 525. [CrossRef] 47. Yang, X.; Ding, Z.; Fan, X.; Zhou, Z.; Ma, N. Processes and mechanisms of desertification in northern China during the last 30 750 years, with a special reference to the Hunshandake Sandy Land, eastern Inner Mongolia. Catena 2008, 71, 2–12. [CrossRef] 48. Li, D.; Gu, X.; Pang, Y.; Chen, B.; Liu, L. Estimation of forest aboveground biomass and leaf area index based on digital aerial photograph data in Northeast China. Forests 2018, 9, 275. [CrossRef] 49. Jannoura, R.; Brinkmann, K.; Uteau, D.; Bruns, C.; Joergensen, R.G. Monitoring of crop biomass using true colour aerial photographs taken from a remote controlled hexacopter. Biosyst. Eng. 2015, 129, 341–351. [CrossRef] 50. Xu, J.; Gu, H.; Meng, Q.; Cheng, J.; Liu, Y.; Sheng, J.; Deng, J.; Bai, X. Spatial pattern analysis of Haloxylon ammodendron using UAV imagery-A case study in the Gurbantunggut Desert. Int. J. Appl. Earth Obs. Geoinf. 2019, 83, 101891. [CrossRef] 51. Torres-Sánchez, J.; Pena, J.M.; de Castro, A.I.; López-Granados, F. Multi-temporal mapping of the vegetation fraction in early- season wheat fields using images from UAV. Comput. Electron. Agric. 2014, 103, 104–113. [CrossRef] 52. Gu, Y.; Brown, J.F.; Verdin, J.P.; Wardlow, B. A five-year analysis of MODIS NDVI and NDWI for grassland drought assessment over the central Great Plains of the United States. Geophys. Res. Lett. 2007, 34, L06407. [CrossRef] 53. Liu, Y.; Wu, L.X.; Ma, B.D. Remote sensing monitoring of soil Moisture on the basis of TM/ETM+ spectral space. J. China Univ. Min. Technol. 2013, 42, 296–301. 54. Sambah, A.; Miura, F. Spatial data analysis and remote sensing for observing tsunami-inundated areas. Int. J. Remote Sens. 2016, 37, 2047–2065. [CrossRef] Remote Sens. 2021, 13, 1730 27 of 27

55. Dogan, H.M. Applications of remote sensing and Geographic Information Systems to assess ferrous minerals and iron oxide of Tokat province in Turkey. Int. J. Remote Sens. 2008, 29, 221–233. [CrossRef] 56. Parastatidis, D.; Mitraka, Z.; Chrysoulakis, N.; Abrams, M. Online Global Land Surface Temperature Estimation from Landsat. Remote Sens. 2017, 9, 1208. [CrossRef]