An Explanation of the Differences in Grassland NDVI Change in the Eastern Route of the China–Mongolia–Russia Economic Corridor
by
and
Zhengfei Wang
1,
Jiayue Wang
1,*,
Wenlong Wang
1,
Chao Zhang
1,
Urtnasan Mandakh
2,
Danzanchadav Ganbat
2,3
Nyamkhuu Myanganbuu
2,41
College of Land Science and Technology, China Agricultural University, Beijing 100083, China
2
Division of GIS and Remote Sensing, Institute of Geography and Geoecology, Mongolian Academy of Sciences, Ulaanbaatar 15170, Mongolia
3
College of Geographical Science, Inner Mongolia Normal University, Hohhot 010022, China
4
Department of Geography, School of Art and Sciences, National University of Mongolia, Ulaanbaatar 14200, Mongolia
*
Author to whom correspondence should be addressed.
Remote Sens. 2025, 17(5), 867; https://doi.org/10.3390/rs17050867
Submission received: 8 January 2025
/
Revised: 18 February 2025
/
Accepted: 26 February 2025
/
Published: 28 February 2025
(This article belongs to the Section Environmental Remote Sensing)
Download keyboard_arrow_down
Figure 1
<p>An overview of the areas along the main railway of the eastern line of the China–Mongolia–Russia Economic Corridor. (<b>a</b>) The elevation distribution within the study area and the geographical locations of China, Mongolia, and Russia; (<b>b</b>) the spatial distribution of average NDVI during growing seasons from 2000 to 2020 for grasslands in the study area. (<a href="https://maps.elie.ucl.ac.be/CCI" target="_blank">https://maps.elie.ucl.ac.be/CCI</a>, accessed on 20 September 2024) (<a href="https://www.webmap.cn/" target="_blank">https://www.webmap.cn/</a>, accessed on 21 August 2024).</p> "> Figure 2
<p>Average spatial distribution characteristics of major environmental and socioeconomic elements in study area: (<b>a</b>) downward shortwave radiation (DSR); (<b>b</b>) land surface temperature; (<b>c</b>) soil moisture; (<b>d</b>) livestock density (in sheep units); (<b>e</b>) population; (<b>f</b>) annual precipitation.</p> "> Figure 3
<p>(<b>a</b>) Spatial pattern of Theil–Sen slope analysis showing NDVI trends; (<b>b</b>) frequency distribution of Theil–Sen slopes.</p> "> Figure 4
<p>Spatiotemporal patterns of grassland NDVI in eastern route of China–Mongolia–Russia Economic Corridor from 2000 to 2020. (<b>a</b>) Spatial distribution of emerging hotspot analysis and its statistical composition in three countries; (<b>b</b>) percentage of different spot types in each country; (<b>c</b>) annual variations in NDVI in different spot types for China, Mongolia, and Russia.</p> "> Figure 5
<p>Importance of driving factors for persistent cold spots and hotspots in whole study area—China, Mongolia, and Russia—analyzed using MRMR method. Numbers in brackets represent ranking of importance.</p> "> Figure 6
<p>Comparison of model performance (R<sup>2</sup>) across different machine learning algorithms using test set (<b>a</b>) and validation set (<b>b</b>) in different grassland clusters. Black line represents mean R<sup>2</sup> across all clusters. Legend explains regions and clusters as follows: CPCS (China Persistent Cold Spot), MPCS (Mongolia Persistent Cold Spot), RPCS (Russia Persistent Cold Spot), CPHS (China Persistent Hotspot), MPHS (Mongolia Persistent Hotspot), RPHS (Russia Persistent Hotspot), PCS (persistent cold spot for entire study area), and PHS (persistent hotspot for entire study area).</p> "> Figure 7
<p>Partial dependence plots of driving factors affecting grassland NDVI in CPHS. (<b>a</b>) Land surface temperature (°C); (<b>b</b>) soil pH; (<b>c</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>d</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>e</b>) soil moisture (mm); (<b>f</b>) soil organic carbon (g/kg); (<b>g</b>) air temperature (°C); (<b>h</b>) annual precipitation (mm/year).</p> "> Figure 8
<p>Partial dependence plots of driving factors affecting grassland NDVI in CPCS. (<b>a</b>) Land surface temperature (°C); (<b>b</b>) population density (person/km<sup>2</sup>); (<b>c</b>) annual precipitation (mm/year); (<b>d</b>) soil organic carbon (g/kg); (<b>e</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>f</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>g</b>) soil moisture (mm); (<b>h</b>) air temperature (°C).</p> "> Figure 9
<p>Partial dependence plots of driving factors affecting grassland NDVI in MPHS. (<b>a</b>) Soil organic carbon (g/kg); (<b>b</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>c</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>d</b>) land surface temperature (°C); (<b>e</b>) livestock density (sheep units); (<b>f</b>) annual precipitation (mm/year); (<b>g</b>) soil moisture (mm); (<b>h</b>) soil pH.</p> "> Figure 10
<p>Partial dependence plots of driving factors affecting grassland NDVI in MPCS. (<b>a</b>) Annual precipitation (mm/year); (<b>b</b>) soil organic carbon (g/kg); (<b>c</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>d</b>) land surface temperature (°C); (<b>e</b>) soil moisture (mm); (<b>f</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>g</b>) livestock density (sheep units); (<b>h</b>) population density (person/km<sup>2</sup>).</p> "> Figure 11
<p>Partial dependence plots of driving factors affecting grassland NDVI in RPHS. (<b>a</b>) Land surface temperature (°C); (<b>b</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>c</b>) soil organic carbon (g/kg); (<b>d</b>) annual precipitation (mm/year); (<b>e</b>) air temperature (°C); (<b>f</b>) soil bulk density (10 kg/m<sup>3</sup>; (<b>g</b>) population density (person/km<sup>2</sup>); (<b>h</b>) livestock density (sheep units).</p> "> Figure 12
<p>Partial dependence plots of driving factors affecting grassland NDVI in RPCS. (<b>a</b>) Livestock density (sheep units); (<b>b</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>c</b>) soil organic carbon (g/kg); (<b>d</b>) population density (person/km<sup>2</sup>); (<b>e</b>) soil pH; (<b>f</b>) land surface temperature (°C); (<b>g</b>) soil bulk density (10 kg/m<sup>3</sup>); (<b>h</b>) soil moisture (mm).</p> "> Figure 13
<p>Response intensity of NDVI to key factors.</p> ">
Versions Notes
<p>An overview of the areas along the main railway of the eastern line of the China–Mongolia–Russia Economic Corridor. (<b>a</b>) The elevation distribution within the study area and the geographical locations of China, Mongolia, and Russia; (<b>b</b>) the spatial distribution of average NDVI during growing seasons from 2000 to 2020 for grasslands in the study area. (<a href="https://maps.elie.ucl.ac.be/CCI" target="_blank">https://maps.elie.ucl.ac.be/CCI</a>, accessed on 20 September 2024) (<a href="https://www.webmap.cn/" target="_blank">https://www.webmap.cn/</a>, accessed on 21 August 2024).</p> "> Figure 2
<p>Average spatial distribution characteristics of major environmental and socioeconomic elements in study area: (<b>a</b>) downward shortwave radiation (DSR); (<b>b</b>) land surface temperature; (<b>c</b>) soil moisture; (<b>d</b>) livestock density (in sheep units); (<b>e</b>) population; (<b>f</b>) annual precipitation.</p> "> Figure 3
<p>(<b>a</b>) Spatial pattern of Theil–Sen slope analysis showing NDVI trends; (<b>b</b>) frequency distribution of Theil–Sen slopes.</p> "> Figure 4
<p>Spatiotemporal patterns of grassland NDVI in eastern route of China–Mongolia–Russia Economic Corridor from 2000 to 2020. (<b>a</b>) Spatial distribution of emerging hotspot analysis and its statistical composition in three countries; (<b>b</b>) percentage of different spot types in each country; (<b>c</b>) annual variations in NDVI in different spot types for China, Mongolia, and Russia.</p> "> Figure 5
<p>Importance of driving factors for persistent cold spots and hotspots in whole study area—China, Mongolia, and Russia—analyzed using MRMR method. Numbers in brackets represent ranking of importance.</p> "> Figure 6
<p>Comparison of model performance (R<sup>2</sup>) across different machine learning algorithms using test set (<b>a</b>) and validation set (<b>b</b>) in different grassland clusters. Black line represents mean R<sup>2</sup> across all clusters. Legend explains regions and clusters as follows: CPCS (China Persistent Cold Spot), MPCS (Mongolia Persistent Cold Spot), RPCS (Russia Persistent Cold Spot), CPHS (China Persistent Hotspot), MPHS (Mongolia Persistent Hotspot), RPHS (Russia Persistent Hotspot), PCS (persistent cold spot for entire study area), and PHS (persistent hotspot for entire study area).</p> "> Figure 7
<p>Partial dependence plots of driving factors affecting grassland NDVI in CPHS. (<b>a</b>) Land surface temperature (°C); (<b>b</b>) soil pH; (<b>c</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>d</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>e</b>) soil moisture (mm); (<b>f</b>) soil organic carbon (g/kg); (<b>g</b>) air temperature (°C); (<b>h</b>) annual precipitation (mm/year).</p> "> Figure 8
<p>Partial dependence plots of driving factors affecting grassland NDVI in CPCS. (<b>a</b>) Land surface temperature (°C); (<b>b</b>) population density (person/km<sup>2</sup>); (<b>c</b>) annual precipitation (mm/year); (<b>d</b>) soil organic carbon (g/kg); (<b>e</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>f</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>g</b>) soil moisture (mm); (<b>h</b>) air temperature (°C).</p> "> Figure 9
<p>Partial dependence plots of driving factors affecting grassland NDVI in MPHS. (<b>a</b>) Soil organic carbon (g/kg); (<b>b</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>c</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>d</b>) land surface temperature (°C); (<b>e</b>) livestock density (sheep units); (<b>f</b>) annual precipitation (mm/year); (<b>g</b>) soil moisture (mm); (<b>h</b>) soil pH.</p> "> Figure 10
<p>Partial dependence plots of driving factors affecting grassland NDVI in MPCS. (<b>a</b>) Annual precipitation (mm/year); (<b>b</b>) soil organic carbon (g/kg); (<b>c</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>d</b>) land surface temperature (°C); (<b>e</b>) soil moisture (mm); (<b>f</b>) downward surface shortwave radiation (W/m<sup>2</sup>); (<b>g</b>) livestock density (sheep units); (<b>h</b>) population density (person/km<sup>2</sup>).</p> "> Figure 11
<p>Partial dependence plots of driving factors affecting grassland NDVI in RPHS. (<b>a</b>) Land surface temperature (°C); (<b>b</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>c</b>) soil organic carbon (g/kg); (<b>d</b>) annual precipitation (mm/year); (<b>e</b>) air temperature (°C); (<b>f</b>) soil bulk density (10 kg/m<sup>3</sup>; (<b>g</b>) population density (person/km<sup>2</sup>); (<b>h</b>) livestock density (sheep units).</p> "> Figure 12
<p>Partial dependence plots of driving factors affecting grassland NDVI in RPCS. (<b>a</b>) Livestock density (sheep units); (<b>b</b>) GDP (millions of 2017 US dollar/km<sup>2</sup>); (<b>c</b>) soil organic carbon (g/kg); (<b>d</b>) population density (person/km<sup>2</sup>); (<b>e</b>) soil pH; (<b>f</b>) land surface temperature (°C); (<b>g</b>) soil bulk density (10 kg/m<sup>3</sup>); (<b>h</b>) soil moisture (mm).</p> "> Figure 13
<p>Response intensity of NDVI to key factors.</p> ">
Abstract
:This study analyzed the spatiotemporal changes in grassland NDVI from 2000 to 2020 in the eastern route of the China–Mongolia–Russia Economic Corridor, a region with frequent ecological–economic interactions, and explained the main driving factors, influencing patterns, and degrees of grassland NDVI changes in different regions. Based on MODIS NDVI data, the study employs emerging spatiotemporal hotspot analysis, Maximum Relevance Minimum Redundancy (mRMR) feature selection, and Gaussian Process Regression (GPR) to reveal the spatiotemporal variation characteristics of grassland NDVI, while identifying long-term stable trends, and to select the most relevant and non-redundant factors to analyze the main driving factors of grassland NDVI change. Partial dependence plots were used to visualize the response and sensitivity of grassland NDVI to various factors. The results show the following: (1) From 2000 to 2020, the NDVI of grassland in the study area showed an overall upward trend, from 0.61 to 0.65, with significant improvement observed in northeastern China and northeastern Russia. (2) Spatiotemporal hotspot analysis indicates that 51% of the area is classified as persistent hotspots for grassland NDVI, mainly distributed in Russia, whereas 12% of the area is identified as persistent cold spots, predominantly located in Mongolia. (3) The analysis of key drivers reveals that precipitation and land surface temperature are the dominant climatic factors shaping grassland NDVI trends, while the effects of soil conditions and human activity vary regionally. In China, NDVI is primarily driven by land surface temperature (LST), GDP, and population density; in Mongolia, precipitation, LST, and GDP exert the strongest influence; whereas in Russia, livestock density and soil organic carbon play the most significant roles. (4) For the whole study area, in persistent cold spot areas of grassland NDVI, the negative effects of rising land surface temperature were most pronounced, reducing NDVI by 36% in the 25–40 °C range. The positive effects of precipitation on NDVI were most evident under low to moderate precipitation conditions, with the effects diminishing as precipitation increased. Soil moisture and soil pH have stronger effects in persistent hotspot areas. Regarding human activity factors, the livestock factor in Mongolia shows an inverted U-shaped relationship with NDVI, and increasing population density contributed to grassland degradation in persistent cold spots. Proper grazing intensity regulation strategy is crucial in these areas with inappropriate grazing intensity, while social and economic activities promoted vegetation cover improvement in persistent hotspots in China and Russia. These findings provide practical insights to guide grassland ecosystem restoration and ensure sustainable development along the eastern route of the China–Mongolia–Russia Economic Corridor. China should prioritize ecological compensation policies. Mongolia needs to integrate traditional nomadic grazing with modern practices. Russia should focus on strengthening regulatory frameworks to prevent the over-exploitation of grasslands. Especially for persistent cold spot areas of grassland NDVI in Mongolia and Russia that are prone to grassland degradation, attention should be paid to the significant negative impact of livestock on grassland.
1. Introduction
Grasslands are the second-largest terrestrial ecosystem globally, covering approximately one-fifth of the Earth’s land surface, and are crucial for human survival and development [1]. Grasslands are vulnerable to climate change and human activities. Therefore, identifying the driving factors behind their changes is extremely important for the protection and management of grassland resources [2].
The China–Mongolia–Russia Economic Corridor (CMREC), a pivotal component of the Belt and Road Initiative, establishes a critical link between the economic hubs of Europe and Northeast Asia [3]. This transnational corridor exhibits complementary land use patterns, with Russia focusing on mineral extraction in the Far East, China on industrialized agricultural zones, and Mongolia on maintaining its traditional nomadic systems [4]. The eastern route of the CMREC, predominantly characterized by grassland ecosystems, faces severe ecological challenges that threaten its stability and sustainability. The Mongolian Plateau, a major source of aeolian sediments for adjacent regions, is particularly vulnerable to climatic extremes, such as droughts, leading to significant grassland deterioration and lake contraction [5]. Concurrently, the northeastern regions of China, including Inner Mongolia, confront grassland degradation stemming from agricultural intensification, rapid urbanization, and industrial expansion in arid and semi-arid zones. The Russian Far East has likewise experienced substantial modifications in its grassland ecosystems, primarily attributed to climate change and resource exploitation. Consequently, the eastern route of the CMREC not only serves as an economic conduit but also represents a critical zone for grassland conservation and management, necessitating coordinated transboundary efforts to address these interconnected ecological challenges [6].
Researchers have primarily focused on the impacts of climate change and human factors on vegetation changes; typically, using traditional linear methods, such as regression and correlation analyses, fails to capture the complex interactions between climate variables, grazing intensity, and topographic factors, especially in the arid and semi-arid regions of Mongolia. These methods struggle to explain the non-linear dynamics of vegetation productivity, making them unsuitable for the accurate modeling of NDVI changes in this region [7,8,9]. Grassland NDVI is influenced not only by precipitation and temperature but also by the combined effects of grazing intensity and topographic complexity. However, linear models perform poorly in explaining these complex interactions [10]. These methods are unable to quantify the relative importance of multiple factors on vegetation greening and may overestimate or underestimate their effects, thus failing to reflect the true interannual changes in vegetation. Therefore, they are unsuitable for extracting non-linear relationships, leading to insufficient accuracy in assessing vegetation greening effects [11,12]. The interactions among climate variables, topographic factors, and management practices result in high variability in vegetation productivity, which traditional regression models struggle to capture due to these complex non-linear dynamics. To address these complex issues, the application of machine learning regression methods has provided solutions for exploring the complex dynamics of vegetation influenced by various factors [13].
There is a lack of comprehensive research on grassland changes and their attribution in the China–Mongolia–Russia Economic Corridor (CMREC) region, which is insufficient to meet the demands for the green economy and environmental protection within this transboundary area. Previous studies have not adequately addressed the comparative analysis of grassland NDVI dynamics and their driving factors across the CMREC region. The existing literature has typically focused on isolated regions, without fully exploring interregional differences and similarities. By utilizing machine learning techniques, this study provides a comprehensive cross-country comparison of grassland NDVI changes, shedding light on how various environmental and socioeconomic factors influence vegetation dynamics in China, Mongolia, and Russia. Through this detailed analysis, we aim to quantify the trends and spatial distribution of grassland NDVI changes from 2000 to 2020, identifying areas with consistently high and low NDVI values across the three countries. Additionally, machine learning attribution models are employed to reveal the impacts of key explanatory factors on grassland NDVI changes. By conducting this comparative analysis, our study will contribute to a more nuanced understanding of the complex dynamics that govern grassland ecosystems in this transboundary region. The cross-country comparison will highlight both the shared and distinct responses of grasslands to environmental and socioeconomic factors, thus informing more targeted and effective strategies for the sustainable management and conservation of these vulnerable ecosystems in each country within the CMREC.
2. Materials and Methods
2.1. Study Area
The study area, defined as a 300 km buffer zone along the railway of the China–Mongolia–Russia Economic Corridor (CMREC) eastern line and refined by administrative boundaries, extends from Dalian to Chita along the route of the former Chinese Eastern Railway (Figure 1). This corridor traverses the temperate grassland ecosystems of China, Mongolia, and Russia, regions that are ecologically significant and economically interconnected. Grasslands in these areas are essential for livestock production, ecological stability, and regional economies. The selection of a 300 km buffer zone was based on its ability to effectively capture key ecological transitions, such as shifts in soil property and climate gradients (Figure 2) along the CMREC eastern line. This distance ensured that the analysis included the core grassland ecosystems that were most relevant to the environmental changes being studied. Additionally, aligning the buffer zone with administrative boundaries enhanced its relevance for practical policy implementation, as these boundaries typically correspond to areas of jurisdiction for local management and conservation efforts.
This region exhibits significant climatic gradients and ecological transition zones: mean annual temperatures decrease from 0 °C to 15 °C in the Chinese section, to −5 °C to 5 °C in the Mongolian section, and down to −10 °C in the Russian section. Annual precipitation follows a southeast–northwest decreasing pattern of horizontal zonality, ranging from 800 to 1300 mm in eastern China, diminishing to 200–400 mm in Mongolia, and further reducing to 50–300 mm in the Russian Far East, forming a continuous transition from humid to semi-humid to semi-arid conditions. Regional economic development displays marked spatial heterogeneity: Northeast China is characterized by a highly industrialized and urbanized complex economic structure; eastern Mongolia is dominated by a primary industry economy based on animal husbandry and mineral resource extraction; and the Russian Far East, intermediary between the two, focuses on resource development and primary product processing. The latter two face infrastructure constraints limiting their economic potential. Grassland ecosystems, serving as the key landscape type in this region, exhibit different human–nature coupling characteristics across the three countries: grasslands in the Chinese section have undergone significant transformation due to urbanization and industrialization; Mongolian grasslands maintain traditional nomadic systems but face overgrazing pressures; and Russian grasslands remain relatively pristine but are experiencing dual stresses from climate change and economic development. The grassland ecosystems throughout the study area generally face increased vulnerability and biodiversity loss risks, potentially posing significant challenges to regional ecological security and sustainable development.
2.2. Data Sources
2.2.1. Land Cover Data
In this study, the European Space Agency (ESA) global land cover dataset (https://maps.elie.ucl.ac.be/CCI, accessed on 21 August 2024) was primarily utilized. This dataset adopts the United Nations Food and Agriculture Organization’s land cover classification system, categorizing global land into 22 distinct classes, and covers a time span from 1992 to 2020 with a spatial resolution of 300 m. The dataset integrates data from multiple remote sensing sources and employs change detection techniques to generate consistent annual land cover products, ensuring temporal continuity. In the ESA Global Land Cover dataset, category 150 represents pure grassland and category 130 represents shrub grassland (10–60% shrub cover and the rest is herbaceous). In this study, these two categories were combined to be considered as grasslands, not only because they are similar in ecological function and structure and are difficult to accurately differentiate in 300 m resolution remote sensing data, but also with reference to the categorization of the Globeland30 dataset, which has a much higher spatial resolution (30 m). The integration of Globeland30 data specifically addressed two limitations of the 300m ESA product: (1) the calibration of mixed pixels (e.g., distinguishing grassland–shrubland mosaics within a single ESA pixel) through higher-resolution spatial constraints, and (2) the identification of small grassland patches (below 10 hectares) that are spatially averaged in the coarser ESA data. This synergistic approach ensures a more comprehensive and accurate assessment of grassland resources in the study area. Based on this dataset and referencing the Globeland30 dataset (https://www.webmap.cn/, accessed on 20 September 2024) with a higher spatial resolution, this study reclassified the ESA land use categories, considering the original two categories (150, 130) as grasslands, which were subsequently used as grassland masks for further analysis.
2.2.2. Normalized Difference Vegetation Index (NDVI)
In this study, the NASA Earthdata Application for Extracting and Exploring Analysis Ready Samples (AppEEARS) (https://appeears.earthdatacloud.nasa.gov/, accessed on 15 October 2024) platform was used to obtain MODIS13Q1.061 Normalized Vegetation Index (NDVI) data from the Terra product for the period of 2000 to 2020. These data have a spatial resolution of 250 m and were obtained using a 16-day maximum composite method. The NDVI products were derived based on atmospherically corrected bi-directional surface reflectance and masked for waters, clouds, heavy aerosols, and cloud shadows to ensure data accuracy and availability. We used NDVI data from May to September and applied an annual maximum value composite to represent the grassland condition for each year. This approach was chosen to reduce the impact of seasonal variations and short-term environmental disturbances, thus more accurately reflecting the optimal vegetation status of grasslands and avoiding the potential loss of interannual variation information that may be caused by average-value synthesis methods.
2.2.3. Meteorological Data
In this study, we utilized multiple high-resolution remote sensing datasets via the Google Earth Engine platform to comprehensively analyze the impact of climate factors on grassland ecosystems from 2000 to 2020. Firstly, we used the Global Precipitation Measurement (GPM) v6 dataset (patterns from 2000 to 2020). The results showed that annual precipitation in the study area exhibits a southeast–northwest decreasing gradient, with higher precipitation in the southeastern region and lower precipitation in the northwestern region (Figure 2f). Secondly, the MOD11A1.061 Terra Land Surface Temperature (LST) dataset (https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD11A1, accessed on 5 November 2024), with a spatial resolution of 1 km and a daily temporal resolution, was used to obtain the average land surface temperature during the growing season for each year. Analysis revealed that the land surface temperature in the study area was higher in the southern region and lower in the northern region, showing a distinct south-to-north decreasing spatial distribution (Figure 2b). Additionally, long-term climate data, including soil moisture, wind speed, and downward shortwave radiation, were obtained from the TerraClimate dataset (https://www.climatologylab.org/terraclimate.html, accessed on 3 June 2024) for the same period, with a spatial resolution of 4 km. Soil moisture analysis showed higher levels in the eastern part of the study area and lower levels in the western part (Figure 2c), while downward shortwave radiation exhibited higher values in the southeastern region, gradually decreasing toward the northwest (Figure 2a).
2.2.4. Terrain and Soil Data
The NASADEM_HGT V001 30m dataset (https://lpdaac.usgs.gov/products/nasadem_hgtv001/, accessed on 12 December 2024), generated by integrating data from the Shuttle Radar Topography Mission (SRTM) and other supplementary elevation sources, provides high-precision elevation information at a global scale. The application of such data not only helped us understand the direct impact of terrain on vegetation distribution more accurately but also indirectly assess how climate change and soil conditions affect vegetation through topographical factors. We used a high-resolution soil dataset from OpenLandMap which predicted global soil attribute information for 2018 using various global soil databases from 1950 to 2017 (https://OpenLandMap.org, accessed on 8 July 2024). The dataset has a spatial resolution of 250 m. We selected three attributes from this dataset: soil pH value, soil organic carbon content, and bulk density. Each data point is provided at six standard depths (0, 10, 30, 60, 100, and 200 cm) within the range of 0 to 200 cm. We used the average values of surface soil and soil at a depth of 10 cm as the soil attribute values that influence grassland NDVI.
2.2.5. Socioeconomic Data
To comprehensively evaluate the impact of human activities and livestock on grassland ecosystems, this study utilized multiple important datasets. Firstly, LandScan Global Population Data (https://landscan.ornl.gov, accessed on 10 May 2024) was employed to provide high-resolution information on population distribution, accurately reflecting population density across the study area (Figure 2e). Analysis showed that population density was concentrated in the eastern regions of the study area, with sparse populations in the west, forming a distinct east–west gradient. Secondly, actual gross domestic product (GDP) data, derived from calibrated nighttime light data on a global 1 km × 1 km grid scale for the period 1992–2019, was used to capture detailed spatial economic activity patterns. In addition, Global Livestock Distribution Data (GLW3&4) (https://data.apps.fao.org, accessed on 18 June 2024), which includes information on various livestock species such as cattle, buffaloes, horses, sheep, and goats, was utilized for the years 2010 and 2015 (Figure 2d). This study focused on four main livestock types—cows, horses, sheep, and goats—to assess their distribution and ecological impact. Livestock density was found to be higher in the southern regions of the study area and lower in the northern regions, with some areas near zero density, showing a south–north gradient. To ensure uniform evaluation, all livestock data were converted into sheep units using the Livestock Unit (LSU) standard proposed by the European Union (https://ec.europa.eu/, accessed on 22 July 2024), which is widely applied in studies of ecological footprints and carrying capacity.
Figure 2.
Average spatial distribution characteristics of major environmental and socioeconomic elements in study area: (a) downward shortwave radiation (DSR); (b) land surface temperature; (c) soil moisture; (d) livestock density (in sheep units); (e) population; (f) annual precipitation.
Figure 2.
Average spatial distribution characteristics of major environmental and socioeconomic elements in study area: (a) downward shortwave radiation (DSR); (b) land surface temperature; (c) soil moisture; (d) livestock density (in sheep units); (e) population; (f) annual precipitation.
2.3. Methods
2.3.1. Spatiotemporal Analysis of Grassland Dynamics
This study utilized emerging hotspot analysis to identify trends in grassland NDVI over time and spatial clustering within the study area, such as new, intensifying, diminishing, and sporadic hotspots and cold spots. The fundamental principle was based on the Gi* statistic proposed by Getis and Ord to measure local spatial association between observed values and their neighboring values [14]. We conducted a Getis–Ord Gi* analysis on the NDVI time series cube over a one-year period and then combined it with the Z-score series obtained through the Mann–Kendall (MK) trend test along the temporal dimension to analyze whether there were clustered spatial patterns and if they showed a trend over time. To further determine the direction of grassland change trends, we also employed a method combining Sen’s median analysis and MK trend analysis. The Theil–Sen median analysis and the Mann–Kendall trend test can be used together to analyze trends in a time series without assuming any specific data distribution [15]. Mann–Kendall trend analysis [16,17] is a method commonly used in the fields of environmental science and meteorology to detect changes in trends in time series data [18,19,20,21,22,23]. It is widely applicable to the trend detection of changes in time series data, not only to assess the existence and direction of the trend, but also to effectively detect whether the time series has undergone a sudden change [24].
To accurately identify the key driving factors of grassland ecosystem changes, we employed the Emerging Spatiotemporal Hotspot Analysis tool in ArcGIS Pro 3.1, generating 16 spatiotemporal patterns. These patterns were systematically consolidated into five ecologically meaningful categories (Table 1), prioritizing analytical clarity while retaining essential spatiotemporal characteristics. For detailed definitions of the subtypes, please refer to the official guidance interface (https://pro.arcgis.com, accessed on 25 October 2024).
Subsequently, we focused on persistent hotspots and persistent cold spots as our research objectives. These two categories represent areas with consistently high or improving NDVI values and areas with consistently low or deteriorating NDVI values, respectively. The selection of these stable regions for subsequent analysis was justified by several considerations: First, these areas exhibited significant temporal persistence and spatial clustering, minimizing the influence of short-term fluctuations and random factors, thus more accurately reflecting long-term driving mechanisms. Second, persistent hotspots and persistent cold spots represent the optimal and poorest states of grassland ecosystems, respectively, and this stark contrast facilitates the identification of key driving factors leading to different evolutionary trajectories of ecosystems. Furthermore, the high-stability characteristics of these regions ensured that subsequent feature selection and modeling could capture more reliable driver–response relationships, providing a robust data foundation for identifying the non-linear effects and critical thresholds of driving factors in partial dependence plot analysis.
The formulas for Sen’s median analysis and the MK trend test are as follows:
where β is the slope, Median is the median function, and i and j represent years. If β > 0, it indicates a positive trend in the pixel’s NDVI; if β < 0, it means there is a negative trend in the pixel’s NDVI.
where
The trend test is conducted using the test statistic Z, which is calculated as follows:
can be expressed as follows:
2.3.2. mRMR Sorting Algorithm
The Maximum Relevance Minimum Redundancy (mRMR) algorithm was utilized to rank factors influencing grassland Normalized Difference Vegetation Index (NDVI). mRMR selects features from high-dimensional datasets that maximize relevance to the target variable while minimizing inter-feature redundancy. This approach identified the variables most strongly correlated with grassland NDVI changes, prioritizing explanatory features for subsequent modeling. The mRMR criterion combines maximizing relevance and minimizing redundancy, using mutual information to measure feature–target correlation and to quantify inter-feature redundancy.
Maximum Relevance: Select the features with the highest correlation to the target variable. Correlation D can be measured using mutual information. For each feature xi in the feature set G, its average mutual information with the target variable c is defined as follows:
Minimum Redundancy: Select features with minimal redundancy among themselves. Redundancy S can be measured using mutual information I(xi, xj) between features. For each pair of features in the feature set G, their average minimum redundancy is defined as follows:
Combining maximum relevance and minimum redundancy, we can calculate the mRMR score as follows:
where represents the mRMR score. We selected the top 8 features with the highest mRMR scores as the feature subset for subsequent machine learning.
2.3.3. Machine Learning Regression
Machine learning algorithms, particularly supervised learning techniques, have demonstrated high effectiveness in solving regression problems for large-scale observational data that violate traditional statistical assumptions. To model NDVI change drivers, we employed various machine learning models, including regression trees, Gaussian Process Regression (GPR), neural networks, Support Vector Regression (SVR), and Ensemble Learning. Details on the model variants and configurations are provided in Appendix A. Five-fold cross-validation was used to prevent overfitting, with a 10% test set for evaluating model generalization. The optimal model was determined by comparing the coefficient of determination (R2) and RMSE among different models. To elucidate the relationships between feature variables and NDVI changes, we used partial dependence plots (PDPs) to both visualize the marginal effects of the eight most important features for each grassland region and calculate the response intensity of NDVI to these features. PDPs are a machine learning interpretability tool that shows the average predicted response of a target variable (NDVI) to a particular feature while keeping other features constant. This helps isolate and understand the individual impact of each feature on NDVI, making complex models more interpretable. The response intensity, defined as the difference between the maximum and minimum values of the PDPs, helps measure how strongly NDVI responds to changes in each factor. This approach provides a clear picture of both the effects and the strength of each driving factor on vegetation changes across different grassland ecosystems.
3. Results
3.1. Spatiotemporal Patterns of Grassland NDVI
3.1.1. Temporal Variation Trends of Grassland NDVI
From 2000 to 2020, the NDVI of grassland in the study area showed an overall upward trend, from 0.61 to 0.65, with significant improvement observed in northeastern China and northeastern Russia. Approximately 67% of the study area exhibited increasing NDVI trends from 2000 to 2020 (Figure 3a). Regional variations in these trends were evident: In China, the changes in grassland NDVI were relatively stable, with slopes primarily concentrated between −100 and 100, indicating a balanced pattern of NDVI dynamics. Mongolia demonstrated a more pronounced positive trend, with slopes predominantly ranging from 0 to 200, reflecting a significant overall increase in NDVI. In contrast, Russia exhibited the greatest variability, with slopes ranging from −300 to 300, showing a nearly equal distribution of positive and negative trends and highlighting the region’s spatial heterogeneity. The slope distribution histogram (Figure 3b) further illustrates these differences, with China and Russia displaying near-normal distributions while Mongolia exhibits a right-skewed distribution, indicating a stronger tendency toward NDVI increase.
Time series analysis (Figure 4c) revealed detailed temporal dynamics across different hotspot types. In persistent hotspot regions, all three countries maintained relatively stable but distinctly different NDVI levels: Russian grasslands maintained the highest levels (NDVI: 0.75–0.8), showing stability during 2002–2004, followed by a slight decrease in 2005 (to 0.73), before stabilizing post-2010 with minimal interannual variations (±0.05); China consistently ranked second (NDVI: 0.7–0.75), experiencing a notable increase–decrease cycle during 2006–2008 followed by increased stability after 2010 despite minor fluctuations (±0.07); Mongolia maintained the lowest levels (NDVI: 0.65–0.7) with the most pronounced interannual variations, particularly exhibiting significant troughs in 2006 and 2014 (declining to 0.65 and 0.63, respectively). Unstable hotspot regions (NDVI: 0.5–0.8) experienced widespread decline in 2006, with Mongolia showing the most significant decrease (below 0.5), while Russia and China demonstrated rapid recovery during 2007–2009 and maintained higher levels after 2010. Non-significant areas (NDVI: 0.4–0.7) exhibited synchronized changes, with all three countries showing simultaneous decreases in 2007 (Russia from 0.65 to 0.55, China from 0.6 to 0.5, and Mongolia from 0.55 to 0.35) followed by relative stability during 2010–2015, before experiencing another synchronized fluctuation in 2016.
3.1.2. Spatial Distribution Characteristics of Grassland NDVI
The grassland NDVI in the study area exhibits significant spatial heterogeneity. The higher NDVI values (0.79–0.93) appear in the low-elevation northeastern region, medium NDVI values (0.54–0.70) show a belt-like distribution in the central region, while the lower NDVI values (0.10–0.40) are found in the high-elevation southwestern region. Further hotspot analysis (Figure 4a) reveals significant spatial clustering characteristics. A total of 51% of the area comprises persistent hotspots for grassland NDVI, mainly distributed in the Russian part, while 12% of the area is mad up of persistent cold spots for grassland NDVI, mainly distributed in the Mongolian part. Persistent hotspots are primarily distributed in the northeastern region, covering the Russian Far East and northeastern China; persistent cold spots are concentrated in the south–central region, mainly covering southern Mongolia; and the central part forms a transition zone composed of unstable hot and cold spots. At the national scale (Figure 4b), the three countries show distinct hotspot–-cold spot spatial compositions: Russia exhibits the highest proportion of persistent hotspots (approximately 77.3%), reflecting the stability of its grassland ecosystem; China shows a complex spatial pattern with persistent hotspots accounting for about 51.7%, along with significant north–south differentiation; and Mongolia displays the highest proportion of cold spots (approximately 33.9%) and the largest proportion of non-significant areas.
3.2. The Importance of the Driving Factors of Grassland NDVI Sorting
Based on the Maximum Relevance Minimum Redundancy (mRMR) algorithm, we quantified the relative importance of potential driving factors affecting grassland NDVI variations across the China–Mongolia–Russia Economic Corridor. The mRMR scores, denoted in parentheses, represent the correlation strength between each factor and NDVI, as well as the degree of non-redundancy among factors.
Precipitation and land surface temperature were identified as the dominant climatic factors influencing grassland NDVI changes (Figure 5), while the importance of soil properties and human activities varied significantly across regions. In China, grassland NDVI changes were primarily influenced by land surface temperature, GDP, and population. For persistent cold spots, the mRMR analysis determined that soil organic carbon (0.0019), population density (0.0025), and land surface temperature (0.1878) were the most influential factors, balancing relevance to NDVI with minimal redundancy. For persistent hotspots, land surface temperature (LST, 0.375), soil pH (0.2922), and GDP (0.2824) emerged as key contributors, reflecting their significant combined impact on grassland dynamics.
In Mongolia, grassland NDVI changes were predominantly driven by precipitation and population density. For persistent cold spots, the mRMR algorithm highlighted precipitation (0.2745), GDP (0.1452), and soil organic carbon (0.1987) as the most critical factors, effectively capturing the interplay of relevance and redundancy. In persistent hotspots, soil organic carbon (0.1908), downward shortwave radiation (0.1603), and GDP (0.0272) demonstrated the strongest combined influence.
In Russia, grassland NDVI changes were largely shaped by livestock-related factors. For persistent cold spots, livestock density (0.4625), GDP (0.3233), and soil organic carbon (0.1525) were identified by the MRMR analysis as the most important drivers, integrating their relevance and minimal redundancy. For persistent hotspots, soil organic carbon (0.059), GDP (0.0654), and land surface temperature (LST, 0.0773) were determined to be the primary influencing factors.
3.3. Comparative Analysis of Driving Factors for NDVI Changes
3.3.1. Comparison of Regression Model Performance
The experimental results reveal distinct performance variations among modeling tasks for China, Mongolia, and Russia (Figure 6). The Gaussian Process Regression (GPR) model consistently demonstrated superior and stable performance across all experiments, particularly for the Chinese and Mongolian datasets. For Russian data, while Ensemble Learning initially performed best (R2 = 0.891), GPR showed consistent performance in subsequent experiments. Cross-country comparisons indicate more stable model performances for China and Mongolia, while models for Russia show higher performance ceilings but greater fluctuations. GPR’s consistent stability and predictive capability across all three countries affirm its adaptability and accuracy in handling diverse datasets and grassland NDVI regions. Consequently, employed the results from the GPR model for further analysis in the subsequent stages of our research.
3.3.2. Regional Response Patterns of Grassland NDVI to Main Driving Factors
Based on the partial dependence plot analysis, this study quantified how grassland NDVI responds to multiple natural and socioeconomic factors. By comparing the CPCS, CPHS, MPCS, MPHS, RPCS, and RPHS, we revealed distinct regional differences and the combined influence of various driving forces on grassland ecosystems.
Natural factors in the CPHS displayed distinct patterns. Air temperature (Figure 7g) maintained an increasing trend within 12–20 °C, with NDVI rising from 0.80 to 0.83. Soil organic carbon (Figure 7f) showed a significant “U-shaped” curve in the 25–125 g/kg range, with an inhibition valley (NDVI = 0.818) around 50 g/kg. Downward shortwave radiation (Figure 7d) exhibited a unimodal pattern in the 190–250 W/m2 range, peaking at approximately 220 W/m2. Land surface temperature (Figure 7a) from 20 to 35 °C resulted in a 17.6% reduction in NDVI (0.82→0.675 or 0.80→0.66, depending on the exact baseline). Moisture conditions demonstrated optimal values at about 400 mm of soil moisture (Figure 7e) and 1100 mm of annual precipitation (Figure 7h), both showing typical “inverted U-shaped” distributions. In terms of socioeconomic factors, GDP (Figure 7c) drove moderate NDVI increases (by about 0.008) within 0–15 × 105 USD/km2, while soil pH (Figure 7b) in the 5.5–8.0 range showed negative correlations with NDVI, causing an approximate 0.03 reduction.
Natural factors in the CPCS exhibited different influence mechanisms from the CPHS. Air temperature (Figure 8h) formed a typical “inverted U-shape” within 16–22 °C, with an optimal temperature around 18 °C; by contrast, the CPHS showed a continuous increase at lower temperatures (12–20 °C). Soil organic carbon (Figure 8d) followed a fluctuating upward trend in the 0–40 g/kg range, whereas the CPHS had a pronounced “U-shaped” curve at higher SOC values. Downward shortwave radiation (Figure 8e) within 220–255 W/m2 exerted inhibitory effects, and land surface temperature (Figure 8a) caused the most significant NDVI reductions in the CPCS (up to 36% drop within 25–40 °C). Annual precipitation (Figure 8c) from 200 to 1000 mm in CPCS continually promoted vegetation, while soil moisture (Figure 8g) exhibited a “V-shaped” fluctuation in the 0–300 mm range; CPHS, on the other hand, reached optimal precipitation (1100 mm) and soil moisture (400 mm) thresholds. Socioeconomic factors in the CPCS showed a strong positive response to GDP (Figure 8f) within 0–15 × 105 USD/km2, and population density (Figure 8b) in the 0–4 × 104 people/km2 range had an S-shaped effect, cumulatively boosting NDVI by 0.08. Compared to the CPHS, the CPCS is more vulnerable to high land surface temperatures but more responsive to socioeconomic inputs.
In the MPHS, natural factors included livestock density (Figure 9e) exhibiting a mild negative linear correlation; NDVI decreased only from about 0.785 to 0.77 when density exceeded 2 × 104 sheep units/km2. Downward shortwave radiation (Figure 9b) was negatively correlated with NDVI in the 215–245 W/m2 range; however, the MPHS still maintained relatively high NDVI levels (0.76–0.81). Land surface temperature (Figure 9d) from 20 to 35 °C lowered NDVI from 0.82 to 0.74, indicating moderate sensitivity. Annual precipitation (Figure 9h) reached water-use saturation around 700 mm, and soil pH (Figure 9h, another axis or subfigure) in the 6–7 range promoted NDVI up to 0.81. Soil organic carbon (Figure 9a) between 6 and 18 g/kg steadily raised NDVI from 0.775 to 0.805. Socioeconomic factors showed GDP (Figure 9c) improving NDVI to 0.799 after passing 5 × 105 USD/km2.
In the MPCS, livestock density (Figure 10g) demonstrated a significant “inverted U-shape”, peaking at NDVI ≈ 0.40 when density reached about 2 × 104 sheep units/km2. Downward shortwave radiation (Figure 10f) also showed negative correlations, with NDVI ranging 0.35–0.40 within 220–260 W/m2. Annual precipitation (Figure 10f, another axis or subfigure) from 200 to 600 mm consistently improved NDVI from 0.3 to 0.4, underscoring progressive water-use efficiency. Land surface temperature (Figure 10d) in the 28–42 °C interval caused a steep NDVI decline (0.45→0.25), more pronounced than in the MPHS (20–35 °C). Socioeconomic factors indicated stepped increases in NDVI when population density (Figure 10h) reached 3 × 104 people/km2 (0.35→0.55), while soil organic carbon (Figure 10b) in the 0–10 g/kg range manifested an “N-shaped” fluctuation. Relative to the MPHS, the MPCS is more sensitive to higher temperatures and lower precipitation, whereas the MPHS maintains a higher NDVI in more favorable hydrothermal conditions.
In the RPHS, natural factors showed considerable complexity. Livestock density (Figure 11h) across 0–1200 sheep units/10 km2 displayed an “N-shaped” pattern, with noticeable inhibition around 800 sheep units/10 km2 and then recovery afterward. Soil bulk density (Figure 11f) in the 80–120 kg/m3 range substantially promoted NDVI, while soil organic carbon (Figure 11c) in the 10–25 g/kg interval had a negative correlation (NDVI decreasing from 0.695 to 0.67). Land surface temperatures (Figure 11a) from 15 to 35 °C caused continuous linear reductions. Annual precipitation (Figure 11d) within 300–900 mm formed an “inverted U-shape”, peaking around 750 mm. As for socioeconomic factors, GDP (Figure 11b) correlated negatively with NDVI, and population density (Figure 11g) also exhibited a persistent negative effect, collectively reducing NDVI by about 0.04.
In the RPCS, livestock density (Figure 12a) displayed an “inverted U-shape” between 0 and 400 sheep units/km2, peaking at about 0.446 NDVI around 180 sheep units/km2. Soil bulk density (Figure 12g) followed an “N-shaped” trend within 20–140 kg/m3, whereas soil organic carbon (Figure 12c) in 0–30 g/kg ranged in a clear “inverted U-shape”, with an optimal value near 18 g/kg. Land surface temperature (Figure 12f) exhibited a “unimodal–steep decline” pattern from 10 to 50 °C, peaking at about 20 °C (NDVI = 0.46) before a marked drop. Socioeconomic factors indicated GDP (Figure 12b) having a positive impact, while population density (Figure 12d) over 0–5000 people/km2 showed a “U-shaped” relationship with NDVI. Soil moisture (Figure 12h) steadily promoted NDVI in the 0–800 mm range, but soil pHs (Figure 12e) of 6–8 caused a steep NDVI decline. Compared with the RPHS, the RPCS underwent more acute NDVI reduction under thermal stress, whereas the hot region faced negative or complex constraints primarily linked to soil organic carbon and GDP.
Across the entire study area, to evaluate the strength of each factor’s influence on grassland NDVI variations across different regions, we define response intensity as the magnitude of NDVI change attributable to a given factor while controlling for the effects of all other variables within the partial dependence framework (Figure 13). Specifically, response intensity is derived from partial dependence plots (PDPs), where it is quantified as the difference between the maximum and minimum NDVI values on the response curve for each factor. This metric is designed to provide a standardized measure of how strongly each predictor influences NDVI, while accounting for the influence of other variables in the statistical model. By using this approach, we can isolate the unique contribution of each factor to NDVI variation.
In our study, the response intensity values (ranging from 0.021 to 0.30) reflect the degree of variability in NDVI driven by each factor, with climatic variables generally exhibiting the highest influence across the study area. This metric is important because it allows us to assess the relative impact of different environmental and socioeconomic drivers on NDVI, independent of other confounding factors. In the PCS, precipitation displays the highest response intensity (0.30), followed by surface temperature at 0.29. In contrast, soil-related factors have lower values, such as soil organic carbon at only 0.04. In the PHS, surface temperature likewise reaches a response intensity of 0.30, whereas air temperature and precipitation are lower, at 0.20 and 0.07, respectively. Notably, soil moisture in the PHS shows a response intensity of 0.021, which, although modest in absolute terms, is still more pronounced than in the PCS. Overall, climatic factors consistently exhibit high response intensities throughout the study area, whereas soil factors, though generally lower in influence, play a relatively more important role in the PHS.
Among China, Mongolia, and Russia, the response intensity of natural factors showed noticeable regional differences. In the CPCS, land surface temperature had a response intensity of 0.18, making it the main factor, while soil organic carbon showed a lower response intensity of 0.04. In the CPHS, precipitation showed an increased response intensity compared to the CPCS, reaching 0.1, while surface temperature maintained a high response intensity of 0.3. In the MPCS, surface temperature and population density both had response intensities of 0.21, while in the MPHS, soil moisture and soil pH were more prominent, with response intensities of 0.035 each. In the RPCS, surface temperature had the highest response intensity at 0.29, followed by soil bulk density at 0.11. In the RPHS, surface temperature had a response intensity of 0.14, and precipitation showed a response intensity of 0.026, while soil-related factors generally had lower response intensities.
For human activity factors, the response intensity in the PCS was generally higher than in the PHS. In the MPCS, population density had the highest response intensity among all regions, reaching 0.21, while livestock density had a response intensity of 0.06, slightly higher than 0.015 in the RPCS. In the CPCS, population density had a lower response intensity of 0.018. In the hotspots, socioeconomic factors exhibited lower response intensities. GDP response intensity was 0.007 and 0.008 in the CPHS and MPHS, respectively, while livestock density showed no significant differences across the PHS. In the RPHS, the response intensities of population density and livestock density remained low, not exceeding 0.016. Overall, socioeconomic factors showed stronger response intensities in the PCS, while their influence in the PHS was relatively weak. Especially for NDVI cold spot areas in Mongolia and Russia that are prone to grassland degradation, attention should be paid to the significant negative impact of livestock on grassland.
Based on the above results, to effectively protect the grasslands along the eastern route of the China–Mongolia–Russia Economic Corridor, a series of targeted measures must be implemented to achieve a balance between ecological conservation and economic development. In Mongolia, it is recommended to advance a national policy promoting a “zoned rotational grazing” system to reduce overgrazing in specific areas, while integrating traditional nomadic practices with modern ecological pastoralism to optimize grazing intensity and mitigate grassland degradation. Additionally, increased investment in rainwater harvesting systems and small reservoir construction is necessary, with priority given to water-saving irrigation technologies in low-NDVI areas, such as the central Mongolian Plateau, to ensure sustainable water use during drought seasons. In China, grassland ecological compensation policies should be further improved and expanded in scope, linking ecological conservation with economic incentives to encourage local pastoral communities to participate in grassland protection.
Concurrently, urban expansion must be coordinated with grassland conservation, and green industries should be promoted through the ecological transformation of agriculture and the development of eco-tourism, reducing the pressure of industrialization and urbanization on grassland ecosystems. In Russia, it is essential to prioritize the support of low-carbon economic activities in grassland regions and strengthen environmental assessments and regulatory oversight of grassland resource development to prevent over-exploitation and ecological degradation. Furthermore, vegetation restoration and ecological barrier construction should be emphasized to enhance the resilience of grassland ecosystems to temperature fluctuations and precipitation changes. These comprehensive measures will provide robust policy support for the sustainable development and conservation of grasslands along the eastern route of the China–Mongolia–Russia Economic Corridor.
4. Discussion
4.1. Spatiotemporal Distribution and Trends of Grassland NDVI
The situation in China is complex. Grassland NDVI in plain areas such as Ulan hot City, Qian’an County, and Tongyu County shows a clear downward trend. However, grasslands in plateau areas like Zhalantun City and Arun Banner exhibit an upward trend. This is closely related to climatic factors and human activities. The impact of human activities on grasslands is complex [25,26]; improving grassland utilization efficiency and reducing livestock intensity can decrease environmental pressure, meanwhile, agricultural expansion, urbanization, and the inappropriate application of pasture fertilizers can still lead to a decline in grassland quality in China [27,28,29]. Moreover, there were significant downward shifts in growing season grassland NDVI within the study area in 2007 and 2016. This may be attributed to the impact of extreme climate events, such as heat waves and heavy rainfall, on grasslands. Summer droughts can lead to a decrease in the coverage of perennial herbaceous plants, while heavy rainfall can promote the growth of some short-lived plants [30]. In the grassland regions of northern China, the sensitivity of grasslands to high-intensity precipitation is far greater than to medium and low-intensity precipitation. This further increases the potential impact of extreme climate events on grasslands [31].
Previous studies have often classified data based on the patterns of vegetation change curves or grassland types before conducting subsequent analyses [32,33]. This approach may overlook the spatial heterogeneity of vegetation growth conditions. In this study, we identified high-value and low-value cluster areas of grasslands in the study region based on emerging spatiotemporal hotspot analysis and SEN slope trend analysis. We found that changes in grassland NDVI within the study area showed significant spatiotemporal differences. Overall, grassland NDVI gradually transitioned from low-value clusters to high-value clusters from the southwest to the northeast of the study area. Cold spots mainly appeared in the Mongolian region, such as areas near the Mongolian Plateau like Matad and Bayandelger. It is worth noting that although grasslands on the Mongolian Plateau generally have lower NDVI values, possibly due to precipitation variability, grazing intensity, and reduced shortwave radiation, there is a widespread greening phenomenon. However, the increased frequency of extreme droughts, particularly the severe droughts of 2000 and 2010 in Mongolia, have notably reduced vegetation coverage on the Mongolian Plateau [34,35]. During severe droughts, net primary productivity (NPP) dropped by up to 75% in affected areas, with 60.55% to 87.75% of the land experiencing drought conditions, particularly in the western regions of the plateau where water availability is already limited [34,36,37,38]. In contrast, grassland NDVI in the Russian region generally showed a downward trend. This may be because grasslands in the Russian region often have higher coverage and are more susceptible to disturbances caused by variations in precipitation and temperature. Moreover, disturbances from human grassland management practices are also important influencing factors, such as overgrazing, unsustainable agricultural development measures, and the accumulation of dead grass resulting from ceased management [39,40,41].
4.2. Attribution of Grassland NDVI Changes Using Machine Learning Methods
Before establishing a machine learning attribution model, it is often necessary to select the correct and reasonable factors that may influence changes in grassland NDVI as inputs for the model. Climate change and human activities often jointly dominate vegetation changes [42,43], previous studies have typically employed residual trend analysis methods to attribute unexplained climate changes to human activities [44]. However, such approaches frequently neglect other natural factors, including elevation, soil properties, and livestock density, leading to an incomplete understanding of their relative importance. Consequently, it is particularly crucial to account for the combined effects of multiple environmental and anthropogenic factors on NDVI dynamics and to quantify their contributions. In this study, the MRMR (Minimum Redundancy Maximum Relevance) factor selection algorithm was used to screen redundant factors in high-dimensional data, which helped us effectively reduce the dimensionality of the data, overcome problems caused by redundancy factors, and improve model accuracy [26]. Our results indicate that surface temperature and annual accumulated precipitation consistently emerge as the most influential variables, reaffirming their critical role in shaping grassland NDVI. Notably, the interplay between precipitation and land surface temperature is vital, as both factors directly influence vegetation growth and soil moisture availability. In arid and semi-arid regions, precipitation directly determines soil moisture levels, thereby regulating vegetation productivity. Conversely, rising land surface temperatures accelerate evapotranspiration, exacerbating soil moisture deficits and further limiting vegetation growth [45,46].
Our results show that soil organic carbon (SOC) has a higher mRMR score than soil bulk density due to its more dynamic role in vegetation productivity. SOC contributes more significantly to vegetation growth through its influence on soil fertility and carbon cycling, while soil bulk density is a structural attribute that changes more gradually in response to environmental factors. In persistent hotspots in China, soil pH exhibits a higher mRMR score than SOC. This is because soil pH directly regulates nutrient solubility, microbial activity, and root nutrient uptake, making it an immediate determinant of plant productivity. As a result, even minor fluctuations in pH can have a pronounced impact on vegetation growth, especially in regions with high NDVI values. In contrast, SOC levels in high-NDVI areas tend to be more stable due to consistent vegetation productivity, which reduces their variability over time.
Additionally, land surface temperature (LST) consistently exhibited higher mRMR scores than downward shortwave radiation (DSR) across both persistent cold spots and hotspots. This is primarily due to LST’s direct impact on plant metabolic processes and water stress regulation. Higher temperatures increase evapotranspiration rates, reducing soil moisture availability and imposing stress on vegetation, leading to noticeable NDVI fluctuations [47]. In contrast, DSR serves as an energy input for photosynthesis but has a more indirect influence on vegetation health, as photosynthesis is also constrained by soil moisture, temperature, and nutrient availability. Furthermore, seasonal temperature extremes drive vegetation stress responses [48]. Consequently, LST exerts a more dynamic influence on NDVI changes than DSR. However, in semi-arid regions like Mongolia, where water availability is often the primary limiting factor, the importance of DSR increases. In Mongolia’s persistent hotspots, where NDVI is already high, variations in DSR may exert a stronger influence on vegetation productivity, particularly during the growing season when temperature alone does not limit photosynthesis as much. This suggests that while LST dominates NDVI fluctuations in moisture-limited ecosystems, DSR can become more critical in regions where precipitation is less of a constraint [49].
4.3. The Non-Linear Relationships Between Different Factors and Grassland NDVI
Partial dependence plots can help quantify the impact of each input feature on the prediction results. By plotting the partial dependence plots for individual features, one can visually observe how changes in that feature affect the predicted values [50]. Based on the partial dependence plots obtained from Gaussian Process Regression, we found that land surface temperature is the most important driving factor in all four regions. The increase in land surface temperature leads to the negative growth of NDVI, indicating that in areas with less moisture, rising land surface temperature causes a decrease in NDVI as high temperatures increase evapotranspiration and reduce soil moisture [51]. In contrast, the NDVI of cold region grasslands in Russia shows a relationship with land surface temperature that is first positively correlated and then negatively correlated. This may be due to the positive effect of land surface temperature on grasslands when vegetation cover is low [52].
It is noteworthy that the impact patterns of air temperature on grassland NDVI in the cold and hot regions of China are the opposite. We speculate that vegetation in hot regions may be in a relatively stable ecological environment, possessing stronger temperature adaptability and buffering capacity [53]. Temperature increases may directly promote vegetation growth by extending the growing season, enhancing photosynthesis, and increasing biomass [54,55]. In comparison, vegetation in NDVI cold regions may be more susceptible to environmental stress. Direct changes in air temperature in these areas may have positive effects, such as relieving low-temperature constraints, increasing water availability, or promoting nutrient cycling. However, when the temperature exceeds a certain threshold, these positive effects may turn negative.
On the other hand, in sparsely populated regions, population growth is negatively correlated with grassland NDVI, suggesting the initial pressures of human activities on the natural environment. In Inner Mongolia, for instance, rapid urbanization in cities such as Hohhot and Baotou has not only led to direct land conversion but has also induced a series of indirect impacts on the surrounding grasslands [56]. Increased urban populations drive higher demands for resources, including water, energy, and food, which in turn leads to the intensification of agricultural practices, overgrazing, and the expansion of industrial activities into previously untouched areas [57,58]. These indirect effects contribute to the degradation of grassland ecosystems, as natural habitats are fragmented or subjected to higher levels of resource extraction [59]. Similarly, in Mongolia, Ulaanbaatar’s urban expansion and the intensification of mining activities have not only altered land cover directly but have also exacerbated environmental pressures in surrounding areas. The increase in urban-generated waste, pollution, and the demand for water resources further contribute to the degradation of nearby grasslands.
These indirect effects, combined with overgrazing and deforestation, have led to significant reductions in NDVI, underscoring the multi-faceted relationship between urbanization, resource use, and grassland degradation [60,61]. In densely populated areas, the two show a positive non-linear relationship, possibly due to more intensive land management and ecological protection measures. Low-NDVI value areas in China and Mongolia show a positive correlation with population density, while Russian regions exhibit more complex “U” or “N” shaped relationships. This is closely related to grassland policies in the different countries. Due to China’s rapid urbanization and environmental degradation, in Mongolia, policymakers’ strategies must adapt to balance economic development and environmental protection. China has implemented a series of policies, including the Grazing Withdrawal Program and the Grassland Ecological Compensation Policy (GECP). The Grazing Withdrawal Program improves grassland conditions by reducing livestock numbers, effectively controlling regional grassland degradation and promoting increases in grassland NDVI and NPP [62,63]. The GECP aims to provide financial incentives for herders to adopt sustainable grazing practices, thereby promoting the restoration and protection of grassland ecosystems. In contrast, Mongolia has retained its traditional nomadic grazing practices, which differ from the fixed pasture and mechanized agricultural policies of China and Russia. Studies have shown that Mongolia’s mobile pastoral approach is more effective in reducing grassland degradation [37]. In addition, Mongolia’s flexible land tenure system allows for rotational grazing, which reduces overgrazing pressure on specific grassland areas and supports better soil health, contributing to improved vegetation cover in certain regions.
The influence of soil properties on grassland NDVI exhibits significant spatial heterogeneity and non-linear characteristics, reflecting the complex response mechanisms of grassland ecosystems to environmental factors. The contrasting effects of soil moisture in high- and low-NDVI areas of Mongolian grasslands are particularly noteworthy. This difference may stem from variations in the composition of vegetation functional groups and their water utilization strategies. High-NDVI areas may be dominated by species with higher water-use efficiencies, while low-NDVI areas may be subject to the dual impacts of water stress or excess [64]. The diverse non-linear relationships between soil organic carbon content and NDVI reveal the multiple ecological functions of organic matter in nutrient cycling, water regulation, and soil structure improvement. In high-SOC regions, the effect of soil organic carbon on vegetation growth is likely driven by improved soil water retention and nutrient availability, promoting greater vegetation productivity. However, in regions with excessive SOC content, nutrient imbalances may limit plant growth, leading to the observed non-linear relationships. This also reflects the varying degrees of dependency on organic matter among different grassland ecosystems. Higher soil organic matter content improves soil structure and nutrient cycling, supporting grassland growth and maintaining higher productivity [65,66]. Notably, the strong explanatory relationship between soil pH and NDVI spans all studied regions. This may be due to the role of soil acidity in regulating nutrient availability and microbial activity. Previous studies have suggested that acidic soil conditions typically lead to nutrient imbalances and restricted vegetation growth [10,67]. However, in this study, grasslands in acidic environments generally grew better than those under alkaline conditions.
4.4. Limitations of This Study
This study employed Sen’s slope trend analysis and emerging spatiotemporal hotspot analysis methods to investigate grassland change trends along the eastern route of the China–Mongolia–Russia Economic Corridor, and analyzed the potential driving factors of NDVI changes in stable high-value and low-value regions. However, several limitations and uncertainties need to be addressed. Primarily, there are temporal continuity constraints in our data sources, with some critical data such as livestock numbers and soil parameters being single-year predicted values. The reliability of our results thus rests on the assumption that these parameters remained relatively stable between 2000 and 2020. From a methodological perspective, while the Gaussian Process Regression model demonstrates high accuracy and robustness, it has limitations in handling spatial autocorrelation effects compared to spatial random forest models. This may introduce bias when evaluating driving factors with significant spatial dependence, such as precipitation and temperature. Although we considered multiple potential driving factors, our analysis did not fully explore factor interactions or the impacts of abrupt events (such as extreme climate events) on grassland ecosystems. In terms of socioeconomic factor representation, our reliance on GDP and population density as proxy indicators for human activities may not adequately capture the effects of diverse grassland management policies and practices across countries. For instance, the specific impacts of China’s ecological compensation policies, Mongolia’s traditional nomadic systems, and Russia’s land use policies are difficult to assess accurately within the current analytical framework. Furthermore, our spatial scale choice of a 300 km buffer zone may overlook broader-scale ecological processes and transboundary effects, while data resolution limitations may have prevented the full capture of small-scale ecological processes and local characteristics. These limitations suggest future research directions: establishing more comprehensive long-term monitoring systems, developing models better suited to handle spatial autocorrelation, considering factor interactions and abrupt events, and more thoroughly evaluating the impact of socioeconomic policies on grassland ecosystems.
5. Conclusions
The eastern route of the China–Mongolia–Russia Economic Corridor is a highly interactive area between the ecological environment and socioeconomic development. Studying the changes in the NDVI of its grasslands and their underlying reasons can contribute to the sustainable development of the local economy and ecosystem. By integrating spatiotemporal hotspot analysis, the mRMR algorithm, and advanced machine learning models, this study analyzed the spatiotemporal changes in grassland NDVI from 2000 to 2020 in the eastern route of the China–Mongolia–Russia Economic Corridor and explained the main driving factors, influencing patterns, and variations in grassland NDVI in different regions. The findings provide insights into how climate and socioeconomic factors interact in grassland ecosystems and offer actionable guidance for sustainable management.
(1). Regarding the temporal and spatial patterns of grasslands, the results show that grassland NDVI in the study area exhibited an overall upward trend from 2000 to 2020, with approximately 67% of the area experiencing varying degrees of growth. Significant spatial heterogeneity was observed: grassland NDVI remained relatively stable in China, while Mongolia’s grasslands showed significant growth but with fluctuations, and Russia’s grasslands displayed the greatest variability, with both positive and negative changes coexisting.
(2). The analysis of driving factors reveals that precipitation, soil organic carbon, and land surface temperature are the dominant climatic factors influencing grassland NDVI changes, while GDP is the key socioeconomic factor. In China, NDVI changes are primarily driven by land surface temperature, GDP, and population; in Mongolia, NDVI changes are more dependent on precipitation, land surface temperature, and GDP; and in Russia, livestock density, GDP, and soil organic carbon emerge as key contributors.
(3). Partial dependence plot analysis further reveals the non-linear relationships between NDVI and these factors, showing significant regional differences: In the CPHS, NDVI increases with precipitation up to 1100 mm, after which the growth rate stabilizes; in the RPHS, NDVI peaks when precipitation reaches 750 mm, with further increases in precipitation having minimal impact. The MPCS shows continuous NDVI improvement as precipitation rises from 200 mm to 600 mm, reflecting its adaptation to arid conditions. Land surface temperature fluctuations show significant differences in sensitivities across regions: the CPHS and MPHS maintain stable NDVI trends with rising temperatures, while the RPCS experiences a sharp decline in NDVI, indicating higher vulnerability to temperature increases. The effects of soil organic carbon are region-specific: in the CPHS and MPHS, higher soil organic carbon content positively impacts NDVI, while in the RPCS, when soil organic carbon exceeds approximately 20 g/kg, its effect on NDVI shifts from positive to negative. Additionally, the effects of socioeconomic factors such as GDP and population density also differ: the CPHS and MPHS benefit from moderate economic development and improved land management, while the RPCS suffers from over-exploitation and economic development constraints, leading to a decline in grassland NDVI.
These results highlight the importance of tailored grassland management strategies. To enhance ecological resilience, China should prioritize ecological compensation policies and sustainable urban expansion. Mongolia needs to integrate traditional nomadic grazing with modern practices to mitigate overgrazing and water resource shortages. Russia should focus on developing a low-carbon economy and strengthening regulatory frameworks to prevent over-exploitation. Moreover, cross-border cooperation and adaptive management frameworks are essential to address shared challenges and ensure the sustainable utilization of grasslands along the corridor. Particular attention should be given to the cold spot areas of grassland NDVI in Mongolia and Russia, which are prone to grassland degradation, to mitigate the significant negative impacts of livestock on grasslands. By implementing these strategies, the grassland ecosystems along the China–Mongolia–Russia Economic Corridor can achieve long-term sustainability, balancing ecological conservation with economic development.
Author Contributions
Conceptualization, Z.W. and J.W.; methodology, Z.W. and W.W.; software, Z.W.; validation, Z.W., J.W. and W.W.; formal analysis, Z.W. and J.W.; investigation, C.Z., J.W., U.M., D.G. and N.M.; resources, Z.W. and J.W.; data curation, Z.W. and J.W.; writing—original draft preparation, Z.W.; writing—review and editing, J.W.; visualization, Z.W.; supervision, J.W. and C.Z.; project administration, J.W.; funding acquisition, C.Z. All authors have read and agreed to the published version of the manuscript.
Funding
This research was funded by the National Key Research and Development Program of China, grant number 2022YFE0197300.
Data Availability Statement
The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author.
Conflicts of Interest
The authors declare no conflicts of interest.
Appendix A
We systematically compared different machine learning models and their variants to determine the optimal approach for analyzing grassland NDVI changes. The model types include the following: Tree (Decision Trees), with fine (4 splits), medium (12 splits), and coarse (36 splits) variants; SVM (Support Vector Machines), including linear, quadratic, cubic, and Gaussian kernels (fine, medium, and coarse, with kernel scales of sqrt(P)/4, sqrt(P), and sqrt(P)*4, respectively, where P represents the number of predictors); NN (Neural Networks) ranging from narrow (10 neurons), medium (25 neurons), and wide (100 neurons) to bilayer and trilayer configurations, all excluding the final regression layer; GPR (Gaussian Process Regression), employing quadratic rational, square exponential, Matern 5/2, and exponential kernels; and EL (Ensemble Learning) methods comprising boosted trees using the LSBoost algorithm, bagged trees using bootstrap aggregation, and a complete ensemble utilizing all available ensemble models. Each variant was run 10 times in a MATLAB 2023b-based environment, and the average validation set R2 and test set R2 were used as performance metrics.
References
- Yan, Z.; Gao, Z.; Sun, B.; Ding, X.; Gao, T.; Li, Y. Global degradation trends of grassland and their driving factors since 2000. Int. J. Digit. Earth 2023, 16, 1661–1684. [Google Scholar] [CrossRef]
- Chen, T.; Bao, A.; Jiapaer, G.; Guo, H.; Zheng, G.; Jiang, L.; Chang, C.; Tuerhanjiang, L. Disentangling the relative impacts of climate change and human activities on arid and semiarid grasslands in Central Asia during 1982–2015. Sci. Total Environ. 2019, 653, 1311–1325. [Google Scholar] [CrossRef] [PubMed]
- Dong, S.; Li, Y.; Li, Y.; Li, S. Spatiotemporal Patterns and Drivers of Land Use and Land Cover Change in the China-Mongolia-Russia Economic Corridor. Pol. J. Environ. Stud. 2021, 30, 1–15. [Google Scholar] [CrossRef]
- Dong, S.; Zheng, J.; Li, Y.; Li, Z.; Li, F.; Jin, L.; Yang, Y.; Bilgaev, A. Quantitative Analysis of the Coupling Coordination Degree Between Urbanization and Eco-environment in Mongolia. Chin. Geogr. Sci. 2019, 29, 861–871. [Google Scholar] [CrossRef]
- Tao, S.; Fang, J.; Zhao, X.; Zhao, S.; Shen, H.; Hu, H.; Tang, Z.; Wang, Z.; Guo, Q. Rapid loss of lakes on the Mongolian Plateau. Proc. Natl. Acad. Sci. USA 2015, 112, 2281–2286. [Google Scholar] [CrossRef]
- Choe, H.; Thorne, J. Climate exposure of East Asian temperate forests suggests transboundary climate adaptation strategies are needed. Clim. Change 2019, 156, 51–67. [Google Scholar] [CrossRef]
- Song, Y. A statistical analysis of the relationship between climatic factors and the Normalized Difference Vegetation Index in China. Int. J. Remote Sens. 2011, 32, 3947–3965. [Google Scholar] [CrossRef]
- Duo, A.; Zhao, W.; Qu, X.; Jing, R.; Xiong, K. Spatio-temporal variation of vegetation coverage and its response to climate change in North China plain in the last 33 years. Int. J. Appl. Earth Obs. Geoinf. 2016, 53, 103–117. [Google Scholar] [CrossRef]
- Jong, R.; Schaepman, M.; Furrer, R.; Bruin, S.; Verburg, P. Spatial relationship between climatologies and changes in global vegetation activity. Glob. Change Biol. 2013, 19, 1953–1964. [Google Scholar] [CrossRef]
- Zhao, Y.; Peth, S.; Krümmelbein, J.; Horn, R.; Wang, Z.; Steffens, M.; Hoffmann, C.; Peng, X. Spatial variability of soil properties affected by grazing intensity in Inner Mongolia grassland. Ecol. Model. 2007, 205, 241–254. [Google Scholar] [CrossRef]
- Kawabata, A.; Ichii, K.; Yamaguchi, Y. Global monitoring of interannual changes in vegetation activities using NDVI and its relationships to temperature and precipitation. Int. J. Remote Sens. 2001, 22, 1377–1382. [Google Scholar] [CrossRef]
- Korb, J.E.; Fule, P.Z. Intra-and Interannual Vegetation Change: Implications for Long-Term Research. Restor. Ecol. 2008, 16, 5–11. [Google Scholar] [CrossRef]
- Green, R.K. Search of Greener Pastures: Advancements in Modeling for Vegetation Dynamics, Climate-Driven Human Migration, and Disaster Classification; University of California: Santa Barbara, CA, USA, 2024. [Google Scholar]
- Getis, A.; Ord, J.K. The analysis of spatial association by use of distance statistics. Geogr. Anal. 1992, 24, 189–206. [Google Scholar] [CrossRef]
- William, M.B. Statistical Methods in the Atmospheric Sciences. J. Am. Stat. Assoc. 2007, 102, 380. [Google Scholar] [CrossRef]
- Henry, B.M. Nonparametric tests against trend. Econometrica 1945, 13, 245–259. [Google Scholar] [CrossRef]
- Maurice, G.K. Rank Correlation Methods. 1948. Available online: https://openlibrary.org/books/OL6064969M/Rank_correlation_methods (accessed on 21 January 2024).
- Alireza, K.; Alireza, K.; Ali Reza, K.; Mahmood, K.; Mohsen, H. Spatial–temporal analysis of net primary production (NPP) and its relationship with climatic factors in Iran. Environ. Monit. Assess. 2020, 192, 718. [Google Scholar] [CrossRef]
- Li, Z.; Huffman, T.; McConkey, B.; Townley-Smith, L. Monitoring and modeling spatial and temporal patterns of grassland dynamics using time-series MODIS NDVI with climate and stocking data. Remote Sens. Environ. 2013, 138, 232–244. [Google Scholar] [CrossRef]
- Lili, X.; Lili, X.; Baolin, L.; Baolin, L.; Yecheng, Y.; Xizhang, G.; Tao, Z.; Tao, Z.; Tao, Z.; Qingling, S.; et al. Detecting Different Types of Directional Land Cover Changes Using MODIS NDVI Time Series Dataset. Remote Sens. 2016, 8, 495. [Google Scholar] [CrossRef]
- Taifeng, D.; Jiangui, L.; Jiali, S.; Budong, Q.; Ted, H.; Yinsuo, Z.; Catherine, C.; Bahram, D. Assessing the Impact of Climate Variability on Cropland Productivity in the Canadian Prairies Using Time Series MODIS FAPAR. Remote Sens. 2016, 8, 281. [Google Scholar] [CrossRef]
- Youjia, L.; Lijun, L.; Lijun, L.; Lijun, L.; Shizuka, H. Spatiotemporal analysis of trends in vegetation change across an artificial desert oasis, Northwest China, 1975–2010. Arab. J. Geosci. 2020, 13, 742. [Google Scholar] [CrossRef]
- Zohre, E.-K.; Maryam, M.; Mohsen, E.-K.; Ruhollah, T.-M. Impacts of vegetation anomalies and agricultural drought on wind erosion over Iran from 2000 to 2018. Appl. Geogr. 2020, 125, 102330. [Google Scholar] [CrossRef]
- Noel, A.C.; Christopher, K.W. Statistics for Spatio-Temporal Data; John Wiley & Sons: Hoboken, NJ, USA, 2011. [Google Scholar]
- Li, C.; Dou, T.; Wang, Y.; Zhu, T.; Yin, H.; Zhou, M.; Liu, L.; Wu, X. A Method for Quantifying the Impacts of Human Activities on Net Primary Production of Grasslands in Northwest China. Remote Sens. 2021, 13, 2479. [Google Scholar] [CrossRef]
- Yang, H.; Yao, L.; Wang, Y.; Li, J. Relative contribution of climate change and human activities to vegetation degradation and restoration in North Xinjiang, China. Rangel. J. 2017, 39, 289–302. [Google Scholar] [CrossRef]
- Chen, T.; Tang, G.; Yuan, Y.; Guo, H.; Xu, Z.; Jiang, G.; Chen, X. Unraveling the relative impacts of climate change and human activities on grassland productivity in Central Asia over last three decades. Sci. Total Environ. 2020, 743, 140649. [Google Scholar] [CrossRef]
- Seabloom, E.W.; Adler, P.B.; Alberti, J.; Biederman, L.; Buckley, Y.M.; Cadotte, M.W.; Collins, S.L.; Dee, L.; Fay, P.A.; Firn, J.; et al. Increasing effects of chronic nutrient enrichment on plant diversity loss and ecosystem productivity over time. Ecology 2021, 102, e03218. [Google Scholar] [CrossRef]
- Bi, X.; Li, B.; Zhang, L.; Nan, B.; Zhang, X.; Yang, Z. Response of grassland productivity to climate change and anthropogenic activities in arid regions of Central Asia. PeerJ 2020, 8, e9797. [Google Scholar] [CrossRef]
- Morecroft, M.D.; Masters, G.J.; Brown, V.K.; Clarke, I.P.; Taylor, M.E.; Whitehouse, A.T. Changing precipitation patterns alter plant community dynamics and succession in an ex-arable grassland. Funct. Ecol. 2004, 18, 648–655. [Google Scholar] [CrossRef]
- Yuan, X.; Li, L.; Chen, X.; Shi, H. Effects of Precipitation Intensity and Temperature on NDVI-Based Grass Change over Northern China during the Period from 1982 to 2011. Remote Sens. 2015, 7, 10164–10183. [Google Scholar] [CrossRef]
- Wang, Z.; Fu, B.; Wu, X.; Wang, S.; Li, Y.; Feng, Y.; Zhang, L.; Hu, Y.; Cheng, L.; Li, B. Distinguishing Trajectories and Drivers of Vegetated Ecosystems in China’s Loess Plateau. Earth’s Future 2024, 12, e2023EF003769. [Google Scholar] [CrossRef]
- Hou, Q.; Ji, Z.; Yang, H.; Yu, X. Impacts of climate change and human activities on different degraded grassland based on NDVI. Sci. Rep. 2022, 12, 15918. [Google Scholar] [CrossRef]
- John, R.; Chen, J.; Ouyang, Z.; Xiao, J.; Becker, R.; Samanta, A.; Ganguly, S.; Yuan, W.; Batkhishig, O. Vegetation response to extreme climate events on the Mongolian Plateau from 2000 to 2010. Environ. Res. Lett. 2012, 8, 035033. [Google Scholar] [CrossRef]
- Lu, Z.; Peng, S.; Slette, I.; Cheng, G.; Li, X.; Chen, A. Soil moisture seasonality alters vegetation response to drought in the Mongolian Plateau. Environ. Res. Lett. 2020, 16, 014050. [Google Scholar] [CrossRef]
- Li, G.; Yu, L.; Liu, T.; Bao, Y.; Yu, J.; Xin, B.; Bao, L.; Li, X.; Chang, X.; Zhang, S. Spatial and temporal variations of grassland vegetation on the Mongolian Plateau and its response to climate change. Front. Ecol. Evol. 2023, 11, 1067209. [Google Scholar] [CrossRef]
- Miao, L.; Sun, Z.; Ren, Y.; Schierhorn, F.; Müller, D. Grassland greening on the Mongolian Plateau despite higher grazing intensity. Land Degrad. Dev. 2021, 32, 792–802. [Google Scholar] [CrossRef]
- Nanzad, L.; Zhang, J.; Tuvdendorj, B.; Yang, S.; Rinzin, S.; Prodhan, F.; Sharma, T.P.P. Assessment of Drought Impact on Net Primary Productivity in the Terrestrial Ecosystems of Mongolia from 2003 to 2018. Remote. Sens. 2021, 13, 2522. [Google Scholar] [CrossRef]
- Chadaeva, V.; Gorobtsova, O.; Pshegusov, R.; Tsepkova, N.; Tembotov, R.; Khanov, Z.; Gedgafova, F.; Zhashuev, A.; Uligova, T.; Khakunova, E.; et al. Stages of grassland degradation in subalpine ecosystems of the Central Caucasus, Russia. Chil. J. Agric. Res. 2021, 81, 630–642. [Google Scholar] [CrossRef]
- Chendev, Y.G.; Sauer, T.J.; Ramirez, G.H.; Burras, C.L. History of East European Chernozem Soil Degradation; Protection and Restoration by Tree Windbreaks in the Russian Steppe. Sustainability 2015, 7, 705–724. [Google Scholar] [CrossRef]
- Mathar, W.P.; Kämpf, I.; Kleinebecker, T.; Kuzmin, I.; Tolstikov, A.; Tupitsin, S.; Hölzel, N. Floristic diversity of meadow steppes in the Western Siberian Plain: Effects of abiotic site conditions, management and landscape structure. Biodivers. Conserv. 2016, 25, 2361–2379. [Google Scholar] [CrossRef]
- Peng, J.; Jiang, H.; Liu, Q.; Green, S.M.; Quine, T.A.; Liu, H.; Qiu, S.; Liu, Y.; Meersmans, J. Human activity vs. climate change: Distinguishing dominant drivers on LAI dynamics in karst region of southwest China. Sci. Total Environ. 2021, 769, 144297. [Google Scholar] [CrossRef]
- Jiang, M.; Tian, S.; Zheng, Z.; Zhan, Q.; He, Y. Human Activity Influences on Vegetation Cover Changes in Beijing, China, from 2000 to 2015. Remote Sens. 2017, 9, 271. [Google Scholar] [CrossRef]
- Xiong, Q.; Xiao, Y.; Halmy, M.W.A.; Dakhil, M.A.; Liang, P.; Liu, C.; Zhang, L.; Pandey, B.; Pan, K.; El Kafraway, S.B.; et al. Monitoring the impact of climate change and human activities on grassland vegetation dynamics in the northeastern Qinghai-Tibet Plateau of China during 2000–2015. J. Arid Land 2019, 11, 637–651. [Google Scholar] [CrossRef]
- Lu, T.; Zheng, C.; Zhou, B.; Wu, J.; Wang, X.; Zhao, Y.; Liu, X.; Yuan, W. Response of Liquid Water and Vapor Flow to Rainfall Events in Sandy Soil of Arid and Semi-Arid Regions. Agronomy 2023, 13, 2424. [Google Scholar] [CrossRef]
- Hui, Y.; Zhang, Q.; Xu, C.-Y.; Xiong, L.; Chen, J.; Xie, P. Vegetation’s role in controlling long-term response of near ground air temperature to precipitation change in a semi-arid region. J. Arid Environ. 2018, 152, 83–86. [Google Scholar] [CrossRef]
- Sun, Y.; Wang, C.; Chen, H.; Ruan, H. Response of Plants to Water Stress: A Meta-Analysis. Front. Plant Sci. 2020, 11, 978. [Google Scholar] [CrossRef]
- Xu, B.-C.; Arain, M.; Black, T.; Law, B.; Pastorello, G.; Chu, H. Seasonal variability of forest sensitivity to heat and drought stresses: A synthesis based on carbon fluxes from North American forest ecosystems. Glob. Change Biol. 2019, 26, 901–918. [Google Scholar] [CrossRef]
- Zhang, Z. Impact of drought on vegetation over inner Mongolia autonomous region. J. Phys. Conf. Ser. 2024, 2798, 012034. [Google Scholar] [CrossRef]
- Dearmon, J.; Smith, T.E. Gaussian Process Regression and Bayesian Model Averaging: An Alternative Approach to Modeling Spatial Phenomena. Geogr. Anal. 2016, 48, 82–111. [Google Scholar] [CrossRef]
- Stas, M.; Orshoven, J.V.; Dong, Q.; Heremans, S.; Zhang, B. A comparison of machine learning algorithms for regional wheat yield prediction using NDVI time series of SPOT-VGT. In Proceedings of the 2016 Fifth International Conference on Agro-Geoinformatics (Agro-Geoinformatics), Tianjin, China, 18–20 July 2016; pp. 1–5. [Google Scholar]
- Tin, Z. Relationship between vegetation index and ground surface temperature on the Tibetan Plateau alpine grassland. J. Glaciol. Geocryol. 2015, 37, 58–69. [Google Scholar]
- Zhang, Y.; Gao, J.; Liu, L.; Wang, Z.; Ding, M.; Yang, X. NDVI-based vegetation changes and their responses to climate change from 1982 to 2011: A case study in the Koshi River Basin in the middle Himalayas. Glob. Planet. Change 2013, 108, 139–148. [Google Scholar] [CrossRef]
- Jeong, S.-J.; Ho, C.-H.; Jeong, J.-H. Increase in vegetation greenness and decrease in springtime warming over east Asia. Geophys. Res. Lett. 2009, 36, L02710. [Google Scholar] [CrossRef]
- Sun, R.; Chen, S.; Su, H. Climate Dynamics of the Spatiotemporal Changes of Vegetation NDVI in Northern China from 1982 to 2015. Remote Sens. 2021, 13, 187. [Google Scholar] [CrossRef]
- Shang, C.; Wu, T.; Huang, G.; Wu, J. Weak sustainability is not sustainable: Socioeconomic and environmental assessment of Inner Mongolia for the past three decades. Resour. Conserv. Recycl. 2019, 141, 243–252. [Google Scholar] [CrossRef]
- Zhang, Y.; Wang, Q.; Wang, Z.; Yang, Y.; Li, J.-L. Impact of human activities and climate change on the grassland dynamics under different regime policies in the Mongolian Plateau. Sci. Total Environ. 2020, 698, 134304. [Google Scholar] [CrossRef] [PubMed]
- Wang, Y.; Yang, Y. Analysis of the Heterogeneous Coordination between Urban Development Levels and the Ecological Environment in the Chinese Grassland Region (2000–2020): A Case Study of the Inner Mongolia Autonomous Region. Land 2024, 13, 951. [Google Scholar] [CrossRef]
- Behnisch, M.; Krüger, T.; Jaeger, J.A. Rapid rise in urban sprawl: Global hotspots and trends since 1990. PLoS Sustain. Transform. 2022, 1, e0000034. [Google Scholar] [CrossRef]
- Gantumur, B.; Wu, F.; Vandansambuu, B.; Tsegmid, B.; Dalaibaatar, E.; Zhao, Y. Spatiotemporal dynamics of urban expansion and its simulation using CA-ANN model in Ulaanbaatar, Mongolia. Geocarto Int. 2020, 37, 494–509. [Google Scholar] [CrossRef]
- Fan, P.; Chen, J.; John, R. Urbanization and environmental change during the economic transition on the Mongolian Plateau: Hohhot and Ulaanbaatar. Environ. Res. 2016, 144 Pt B, 96–112. [Google Scholar] [CrossRef]
- Zhou, T.; Yang, H.; Qiu, X.; Sun, H.; Song, P.; Yang, W. China’s grassland ecological compensation policy achieves win-win goals in Inner Mongolia. Environ. Res. Commun. 2023, 5, 031007. [Google Scholar] [CrossRef]
- Chen, H.; Shao, L.; Zhao, M.; Zhang, X.; Zhang, D. Grassland conservation programs, vegetation rehabilitation and spatial dependency in Inner Mongolia, China. Land Use Policy 2017, 64, 429–439. [Google Scholar] [CrossRef]
- Jiang, P.; Ding, W.; Yuan, Y.; Ye, W. Diverse response of vegetation growth to multi-time-scale drought under different soil textures in China’s pastoral areas. J. Environ. Manag. 2020, 274, 110992. [Google Scholar] [CrossRef]
- Wiesmeier, M.; Steffens, M.; Kölbl, A.; Kögel-Knabner, I. Degradation and small-scale spatial homogenization of topsoils in intensively-grazed steppes of Northern China. Soil Tillage Res. 2009, 104, 299–310. [Google Scholar] [CrossRef]
- Su, Y.Z.; Li, Y.L.; Zhao, H.L. Soil Properties and their Spatial Pattern in a Degraded Sandy Grassland under Post-grazing Restoration, Inner Mongolia, Northern China. Biogeochemistry 2006, 79, 297–314. [Google Scholar] [CrossRef]
- Qi, Y.-C.; Dong, Y.-S.; Jin, Z.; Peng, Q.; Xiao, S.-S.; He, Y.-T. Spatial Heterogeneity of Soil Nutrients and Respiration in the Desertified Grasslands of Inner Mongolia, China. Pedosphere 2010, 20, 655–665. [Google Scholar] [CrossRef]
Figure 1.
An overview of the areas along the main railway of the eastern line of the China–Mongolia–Russia Economic Corridor. (a) The elevation distribution within the study area and the geographical locations of China, Mongolia, and Russia; (b) the spatial distribution of average NDVI during growing seasons from 2000 to 2020 for grasslands in the study area. (https://maps.elie.ucl.ac.be/CCI, accessed on 20 September 2024) (https://www.webmap.cn/, accessed on 21 August 2024).
Figure 1.
An overview of the areas along the main railway of the eastern line of the China–Mongolia–Russia Economic Corridor. (a) The elevation distribution within the study area and the geographical locations of China, Mongolia, and Russia; (b) the spatial distribution of average NDVI during growing seasons from 2000 to 2020 for grasslands in the study area. (https://maps.elie.ucl.ac.be/CCI, accessed on 20 September 2024) (https://www.webmap.cn/, accessed on 21 August 2024).
Figure 3.
(a) Spatial pattern of Theil–Sen slope analysis showing NDVI trends; (b) frequency distribution of Theil–Sen slopes.
Figure 3.
(a) Spatial pattern of Theil–Sen slope analysis showing NDVI trends; (b) frequency distribution of Theil–Sen slopes.
Figure 4.
Spatiotemporal patterns of grassland NDVI in eastern route of China–Mongolia–Russia Economic Corridor from 2000 to 2020. (a) Spatial distribution of emerging hotspot analysis and its statistical composition in three countries; (b) percentage of different spot types in each country; (c) annual variations in NDVI in different spot types for China, Mongolia, and Russia.
Figure 4.
Spatiotemporal patterns of grassland NDVI in eastern route of China–Mongolia–Russia Economic Corridor from 2000 to 2020. (a) Spatial distribution of emerging hotspot analysis and its statistical composition in three countries; (b) percentage of different spot types in each country; (c) annual variations in NDVI in different spot types for China, Mongolia, and Russia.
Figure 5.
Importance of driving factors for persistent cold spots and hotspots in whole study area—China, Mongolia, and Russia—analyzed using MRMR method. Numbers in brackets represent ranking of importance.
Figure 5.
Importance of driving factors for persistent cold spots and hotspots in whole study area—China, Mongolia, and Russia—analyzed using MRMR method. Numbers in brackets represent ranking of importance.
Figure 6.
Comparison of model performance (R2) across different machine learning algorithms using test set (a) and validation set (b) in different grassland clusters. Black line represents mean R2 across all clusters. Legend explains regions and clusters as follows: CPCS (China Persistent Cold Spot), MPCS (Mongolia Persistent Cold Spot), RPCS (Russia Persistent Cold Spot), CPHS (China Persistent Hotspot), MPHS (Mongolia Persistent Hotspot), RPHS (Russia Persistent Hotspot), PCS (persistent cold spot for entire study area), and PHS (persistent hotspot for entire study area).
Figure 6.
Comparison of model performance (R2) across different machine learning algorithms using test set (a) and validation set (b) in different grassland clusters. Black line represents mean R2 across all clusters. Legend explains regions and clusters as follows: CPCS (China Persistent Cold Spot), MPCS (Mongolia Persistent Cold Spot), RPCS (Russia Persistent Cold Spot), CPHS (China Persistent Hotspot), MPHS (Mongolia Persistent Hotspot), RPHS (Russia Persistent Hotspot), PCS (persistent cold spot for entire study area), and PHS (persistent hotspot for entire study area).
Figure 7.
Partial dependence plots of driving factors affecting grassland NDVI in CPHS. (a) Land surface temperature (°C); (b) soil pH; (c) GDP (millions of 2017 US dollar/km2); (d) downward surface shortwave radiation (W/m2); (e) soil moisture (mm); (f) soil organic carbon (g/kg); (g) air temperature (°C); (h) annual precipitation (mm/year).
Figure 7.
Partial dependence plots of driving factors affecting grassland NDVI in CPHS. (a) Land surface temperature (°C); (b) soil pH; (c) GDP (millions of 2017 US dollar/km2); (d) downward surface shortwave radiation (W/m2); (e) soil moisture (mm); (f) soil organic carbon (g/kg); (g) air temperature (°C); (h) annual precipitation (mm/year).
Figure 8.
Partial dependence plots of driving factors affecting grassland NDVI in CPCS. (a) Land surface temperature (°C); (b) population density (person/km2); (c) annual precipitation (mm/year); (d) soil organic carbon (g/kg); (e) downward surface shortwave radiation (W/m2); (f) GDP (millions of 2017 US dollar/km2); (g) soil moisture (mm); (h) air temperature (°C).
Figure 8.
Partial dependence plots of driving factors affecting grassland NDVI in CPCS. (a) Land surface temperature (°C); (b) population density (person/km2); (c) annual precipitation (mm/year); (d) soil organic carbon (g/kg); (e) downward surface shortwave radiation (W/m2); (f) GDP (millions of 2017 US dollar/km2); (g) soil moisture (mm); (h) air temperature (°C).
Figure 9.
Partial dependence plots of driving factors affecting grassland NDVI in MPHS. (a) Soil organic carbon (g/kg); (b) downward surface shortwave radiation (W/m2); (c) GDP (millions of 2017 US dollar/km2); (d) land surface temperature (°C); (e) livestock density (sheep units); (f) annual precipitation (mm/year); (g) soil moisture (mm); (h) soil pH.
Figure 9.
Partial dependence plots of driving factors affecting grassland NDVI in MPHS. (a) Soil organic carbon (g/kg); (b) downward surface shortwave radiation (W/m2); (c) GDP (millions of 2017 US dollar/km2); (d) land surface temperature (°C); (e) livestock density (sheep units); (f) annual precipitation (mm/year); (g) soil moisture (mm); (h) soil pH.
Figure 10.
Partial dependence plots of driving factors affecting grassland NDVI in MPCS. (a) Annual precipitation (mm/year); (b) soil organic carbon (g/kg); (c) GDP (millions of 2017 US dollar/km2); (d) land surface temperature (°C); (e) soil moisture (mm); (f) downward surface shortwave radiation (W/m2); (g) livestock density (sheep units); (h) population density (person/km2).
Figure 10.
Partial dependence plots of driving factors affecting grassland NDVI in MPCS. (a) Annual precipitation (mm/year); (b) soil organic carbon (g/kg); (c) GDP (millions of 2017 US dollar/km2); (d) land surface temperature (°C); (e) soil moisture (mm); (f) downward surface shortwave radiation (W/m2); (g) livestock density (sheep units); (h) population density (person/km2).
Figure 11.
Partial dependence plots of driving factors affecting grassland NDVI in RPHS. (a) Land surface temperature (°C); (b) GDP (millions of 2017 US dollar/km2); (c) soil organic carbon (g/kg); (d) annual precipitation (mm/year); (e) air temperature (°C); (f) soil bulk density (10 kg/m3; (g) population density (person/km2); (h) livestock density (sheep units).
Figure 11.
Partial dependence plots of driving factors affecting grassland NDVI in RPHS. (a) Land surface temperature (°C); (b) GDP (millions of 2017 US dollar/km2); (c) soil organic carbon (g/kg); (d) annual precipitation (mm/year); (e) air temperature (°C); (f) soil bulk density (10 kg/m3; (g) population density (person/km2); (h) livestock density (sheep units).
Figure 12.
Partial dependence plots of driving factors affecting grassland NDVI in RPCS. (a) Livestock density (sheep units); (b) GDP (millions of 2017 US dollar/km2); (c) soil organic carbon (g/kg); (d) population density (person/km2); (e) soil pH; (f) land surface temperature (°C); (g) soil bulk density (10 kg/m3); (h) soil moisture (mm).
Figure 12.
Partial dependence plots of driving factors affecting grassland NDVI in RPCS. (a) Livestock density (sheep units); (b) GDP (millions of 2017 US dollar/km2); (c) soil organic carbon (g/kg); (d) population density (person/km2); (e) soil pH; (f) land surface temperature (°C); (g) soil bulk density (10 kg/m3); (h) soil moisture (mm).
Figure 13.
Response intensity of NDVI to key factors.
Table 1.
Classification framework of spatiotemporal hotspot analysis.
| Category | Subtypes Included | Ecological Interpretation |
|---|---|---|
| Persistent Hotspots | Persistent, Intensifying | Stable high NDVI/continuous improvement |
| Unstable Hotspots | New, Consecutive, Diminishing, Sporadic, Oscillating, Historical Hotspots | High NDVI with significant fluctuations |
| Non-Significant | No Pattern | Undetectable temporal trends |
| Unstable Cold Spots | New, Consecutive, Diminishing, Sporadic, Oscillating, Cold spots | Low NDVI with marked variability |
| Persistent Cold Spots | Persistent, Intensifying | Stable low NDVI/continuous degradation |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
MDPI and ACS Style
Wang, Z.; Wang, J.; Wang, W.; Zhang, C.; Mandakh, U.; Ganbat, D.; Myanganbuu, N. An Explanation of the Differences in Grassland NDVI Change in the Eastern Route of the China–Mongolia–Russia Economic Corridor. Remote Sens. 2025, 17, 867. https://doi.org/10.3390/rs17050867
AMA Style
Wang Z, Wang J, Wang W, Zhang C, Mandakh U, Ganbat D, Myanganbuu N. An Explanation of the Differences in Grassland NDVI Change in the Eastern Route of the China–Mongolia–Russia Economic Corridor. Remote Sensing. 2025; 17(5):867. https://doi.org/10.3390/rs17050867
Chicago/Turabian StyleWang, Zhengfei, Jiayue Wang, Wenlong Wang, Chao Zhang, Urtnasan Mandakh, Danzanchadav Ganbat, and Nyamkhuu Myanganbuu. 2025. "An Explanation of the Differences in Grassland NDVI Change in the Eastern Route of the China–Mongolia–Russia Economic Corridor" Remote Sensing 17, no. 5: 867. https://doi.org/10.3390/rs17050867
APA StyleWang, Z., Wang, J., Wang, W., Zhang, C., Mandakh, U., Ganbat, D., & Myanganbuu, N. (2025). An Explanation of the Differences in Grassland NDVI Change in the Eastern Route of the China–Mongolia–Russia Economic Corridor. Remote Sensing, 17(5), 867. https://doi.org/10.3390/rs17050867
Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.