Anthropogenic Activities Dominate Vegetation Improvement in Arid Areas of China

Abstract

Arid regions, while providing essential ecosystem services, are among the most ecologically vulnerable worldwide. Understanding and monitoring their long-term vegetation dynamics is essential for accurate environmental assessment and climate adaptation strategies. This study examined the spatiotemporal variations and driving forces of the vegetation dynamics in arid Northwestern China during 2000 to 2020, using the annual peak fractional vegetation cover (FVC) as the primary indicator. The Sen’s slope estimator with the Mann–Kendall test and the coefficient of variation were employed to assess the spatiotemporal variations in FVC, while the Pearson correlation, geographic detector model and random forest model were applied to identify the dominant driving factors for FVC. The results indicated that (1) overall vegetation cover was low (averaged peak FVC = 0.191), showing a spatial pattern of higher values in the northwest and lower values in the southeast; high FVC values were primarily observed in mountainous areas and river corridors; (2) the annual peak FVC increased significantly at a rate of 0.0508 yr⁻¹, with 33.72% of the region showing significant improvements and 5.49% degradation; (3) the spatial pattern of FVC was shaped by the distribution of land use types (59.46%), while the temporal dynamics of FVC were driven by land use changes (16.37%) and the land use intensity (37.56%); (4) both the spatial pattern and the temporal dynamics were limited by the environmental conditions. These findings highlight the critical role of anthropogenic activities in shaping the spatiotemporal variations in FVC, particularly emphasizing the distinct contributions of changes in land use types and land use intensity. This study could provide a scientific basis for sustainable land management and restoration strategies in arid regions facing global changes.

1. Introduction

Vegetation, as a primary component of terrestrial ecosystems, plays a critical role in carbon, water, and energy cycles [1,2]. Quantifying the spatiotemporal variation in vegetation changes and identifying their driving factors is fundamental in understanding ecosystem responses to environmental change and in supporting ecological restoration and sustainable land management. Such assessments have been greatly facilitated by the advent of satellite remote sensing technologies, which offer high-resolution and large-scale observations of vegetation cover across regional and global scales. In recent years, numerous studies have explored vegetation changes across multiple spatial and temporal scales by utilizing long-term satellite-derived indicators, such as the normalized difference vegetation index (NDVI), fractional vegetation cover (FVC), enhanced vegetation index (EVI) and soil-adjusted vegetation index (SAVI). Their findings have revealed significant transformations in global vegetation coverage over the past several decades. Increased trends were observed in Europe [3]; the Eastern United States [4]; East, South and Southeast Asia [5,6,7]; and Western and Southern Australia [8], while decreased trends were observed in Southern Africa [9,10] and Southern South America [11,12].

Arid regions and semi-arid regions, which cover approximately 41% of the global land area, provide essential ecosystem services that support the livelihoods of more than two billion people [13,14]. However, vegetation in dryland is typically sparse and highly vulnerable to desertification driven by climate change and anthropogenic disturbances. Once degraded, these ecosystems are difficult to restore due to limited water availability and poor soil quality [15,16,17]. Recent studies on vegetation coverage have identified hotspots with spatially heterogeneous trends in arid regions. Land cover changes were widely confirmed as the primary driving factors [18,19,20]. However, anthropogenic activities other than land use change have rarely been discussed.

Northwest China represents a typical arid region, characterized by the mean annual precipitation being lower than the mean annual potential evapotranspiration. Particularly in the Xinjiang Uygur Autonomous Region, rugged terrain environments lead to diverse types of vegetation, with bare areas, cropland, grassland and urban land being the dominant land cover types [21,22]. Recent studies have highlighted the roles of precipitation, temperature and glacier melt in promoting vegetation recovery [23,24,25,26,27]. Furthermore, government-led revegetation initiatives and land reclamation have also significantly contributed to vegetation greening in recent decades [26,28,29]. However, critical livelihood activities such as grazing and cultivation place strong but spatially heterogeneous pressures on vegetation patterns [30,31,32,33]. Despite their importance, these human factors have not been comprehensively assessed at the regional scale using remote sensing techniques. This limitation leaves a critical gap in our understanding of human–environment interactions in these fragile ecosystems.

To address this gap, this study proposes an integrated framework to disentangle the contributions of natural and anthropogenic factors to the vegetation dynamics in arid environments. Specifically, we derived high-resolution FVC (250 m) from MODIS NDVI time series (2000 to 2020) to characterize vegetation changes in Xinjiang. FVC was selected as the primary vegetation indicator due to its intuitive quantification of vegetation coverage and its higher accuracy in sparsely vegetated areas [34,35,36]. Ten potential drivers, including terrain conditions (elevation, slope and aspect), climate (air temperature, precipitation), soil water content and changes in land use types and land use intensity (population density, fertilization and grazing) were assessed by Pearson correlation analysis, the geographical detector model and a random forest model. Compared to traditional studies focusing on climate or land cover alone, this multifactor attribution framework offers a more nuanced and spatially explicit understanding of vegetation dynamics, especially in regions experiencing rapid environmental and socioeconomic transitions. Moreover, comparing the quantified importance of each driver across three complementary methods allowed for a more robust and comprehensive evaluation of their influence. Based on the spatiotemporal patterns of FVC and the quantified potential driving factors, this study offers high-resolution insights into vegetation dynamics, thereby providing scientific support for ecological conservation and region-specific sustainable development in arid Northwest China under future climate and human pressures.

2. Materials and Methods

2.1. Study Area

The Xinjiang Uygur Autonomous Region (34°22′N to 49°33′N, 73°22′E to 96°21′E) is situated in the Northwest of China, at the heart of Asia, with elevation ranging from −155 to 8611 m (Figure 1). The region is dominated by a temperate continental climate with a mean annual temperature of 6.2 to 9.0 °C and annual precipitation of approximately 150 mm. Meanwhile, the topography has shaped a unique mountain basin system in arid and semi-arid regions, resulting in diverse local climatic conditions. The spatial variability in the hydrothermal conditions has led to the typical and complete vertical zonation of mountain natural belts in arid regions [37]. In recent years, with the establishment and development of the Central Asia Economic Corridor, Xinjiang has experienced population growth and economic expansion, leading to increasing ecological and environmental pressures [38]. Despite the Chinese government’s efforts in implementing projects such as the Three-North Shelterbelt Program and inter-basin water transfer, which have achieved notable accomplishments in ecological restoration, the vegetation dynamics in Xinjiang remain uncertain. Specifically, there is a limited understanding of the spatiotemporal variations in vegetation changes and their driving factors in the region.

2.2. Data Sources and Data Processing

FVC was utilized to investigate the spatiotemporal dynamics of vegetation cover. To identify the potential driving factors underlying these changes, ten associated factors (

Table 1. Descriptions and origins of the map variables used in this study.

Category	Variable	Spatial Resolution	Resource
Vegetation	FVC	250 m	MOD13Q1 NDVI
Environmental conditions
Terrain	Elevation	1 arc-second	DEM (https://www.usgs.gov/; accessed on 22 July 2025)
	Slope
	Aspect
Climate	Temperature	1 km	https://cstr.cn/18406.11.Meteoro.tpdc.270961 (accessed on 22 July 2025)
	Precipitation	1 km
Soil	Soil water content	1 km	https://cstr.cn/18406.11.RemoteSen.tpdc.272760 (accessed on 22 July 2025)
Anthropogenic activities
Land cover changes	Land cover	300 m	https://www.esa-landcover-cci.org/(accessed on 22 July 2025)
Land use intensity	Population	1 km	https://landscan.ornl.gov/(accessed on 22 July 2025)
	Nitrogen fertilization	5 km	https://cstr.cn/15732.11.nesdc.ecodb.pa.2022.13 (accessed on 22 July 2025)
	Grazing index	250 m	https://cstr.cn/15732.11.nesdc.ecodb.rs.2024.024 (accessed on 22 July 2025)

) were incorporated into the analysis. Environmental conditions included the terrain (elevation, slope and aspect), climate conditions (annual mean temperature and cumulative precipitation) and soil water content, whereas anthropogenic activities were represented by land cover changes and the land use intensity (population density, fertilizer application rates and grazing pressure). All datasets were projected to the GCS_WGS_1984 coordinate system, resampled to a spatial resolution of 250 m and clipped to the study area using the official boundary provided by the standard map of China (https://www.tianditu.gov.cn/; accessed on 22 July 2025).

2.2.1. FVC Data

The FVC data were calculated based on the NDVI datasets derived from the Moderate-Resolution Imaging Spectroradiometer (MODIS) aboard NASA’s Terra satellite. The MODIS NDVI datasets, sourced from the MOD13Q1 product, were obtained through the Google Earth Engine (GEE) platform. The datasets offer global coverage at 16-day intervals with a spatial resolution of 250 m from February 2000 to present [39]. FVC was estimated using the dimidiate pixel model as follows:

$$\begin{array}{c}FVC={\displaystyle \frac{NDVI-{NDVI}_{soil}}{{NDVI}_{veg}-NDVI}}\end{array}$$

(1)

where NDVI represent the pixel value, and NDVI_veg and NDVI_soil represent the NDVI values of pure vegetation and bare soil in the study area, respectively. Pixels with NDVI values outside the range of [0, 1] were assigned 0 before calculation. Moreover, NDVI_veg and NDVI_soil were defined as the 98th and 2nd percentiles of the NDVI distribution within the study area to reduce the influence of outliers [27,35].

2.2.2. Environmental Conditions

1.

Terrain data

Terrain features were characterized based on a digital elevation model (DEM) derived from the Shuttle Radar Topography Mission [40] and the Advanced Spaceborne Thermal Emission and Reflection Radiometer [41]. The datasets provide near-global coverage at a spatial resolution of 1 arc-second. The slope and aspect were calculated using the Spatial Analyst tool in ArcGIS 10.7.

2.

Climate data

The 1 km monthly temperature and precipitation datasets for China developed by Peng et al. were used to calculate the annual mean temperature and cumulative precipitation in this study [42]. The datasets were spatially downscaled from CRU TS v4.02 with WorldClim datasets based on the Delta downscaling method and evaluated via 496 national weather stations across China. The datasets provide monthly mean temperature and cumulative precipitation data with 0.0083333 arc-degrees for China from January 1901 to December 2023, obtained from the National Tibetan Plateau Data Center—Third Pole Environment Data Center.

3.

Soil water content data

In this study, the annual maximum value of each pixel was selected according to the global daily surface soil moisture dataset at a 1 km resolution (2000–2020) developed by Zheng et al. to represent the annual soil water content [43]. The datasets were generated using a machine learning-based approach based on the ESA Climate Change Initiative (ESA-CCI) dataset and verified by 2346 ground observation stations around the world. The datasets were obtained from the National Tibetan Plateau Data Center—Third Pole Environment Data Center.

2.2.3. Land Use Type Data

Land cover data were obtained from the European Space Agency Climate Change Initiative Land Cover (ESA CCI-LC) dataset, which provides annual global land cover maps at a spatial resolution of 300 m from 1992 onwards [44]. The dataset is generated using a consistent classification scheme aligned with the United Nations Land Cover Classification System (LCCS). The land cover types were reclassified into 15 categories based on the Level 1 classification system of ESA CCI-LC. These categories include cropland, cropland–natural vegetation mosaics, deciduous broadleaved forests, evergreen needle-leaved forests, deciduous needle-leaved forests, mixed forests, tree–shrub–herbaceous cover, shrubland, grassland, lichens and mosses, sparse vegetation, wetlands, urban areas, bare areas and non-vegetation areas.

2.2.4. Land Use Intensity Data

Annual population data were obtained from the LandScan Global Population Database published by the Oak Ridge National Laboratory. The dataset provides ambient population distributions at a 30-arc-second spatial resolution, representing the average population presence over a 24 h period [45].

Fertilizer application rates were represented by nitrogen fertilization data in this study. The historical nitrogen fertilizer application rate datasets developed by Yu et al. [46] were accessed from the National Science and Technology Infrastructure of China (https://www.geodata.cn; accessed on 22 July 2025), a national platform supporting long-term data sharing in environmental and resource sciences. The dataset provided annual nitrogen application rates at a spatial resolution of 5 km, expressed in units of g N m⁻² yr⁻¹, covering the period from 1952 to 2018. It was developed by integrating historical cropland distribution maps with multisource statistical and survey data, enabling the reconstruction of historical nitrogen inputs across agricultural lands in China.

Annual grazing intensity datasets were developed by [47]. The dataset provides the grazing intensity in seven pastoral provinces in Western China with a spatial resolution of 250 m, in units of SU ha⁻¹, from 1980 to 2022. It was developed by integrating livestock census statistics with satellite-derived vegetation indices to estimate the spatially explicit grazing intensity.

2.3. Methods

The methods and technical framework employed in this study are illustrated in Figure 2, comprising the following steps: (1) calculating annual maximum FVC data from 2000 to 2020 and preprocessing the auxiliary datasets to ensure consistency in spatial resolution and geographic extent; (2) analyzing the spatiotemporal variations in FVC and associated factors; and (3) quantifying the driving factors contributing to FVC changes. All computations and statistical analyses were conducted using Python 3.12.

2.3.1. Data Preprocessing and Standardization

As the datasets used in this study originated from multiple sources with varying spatial resolutions, a harmonization process was implemented to ensure spatial consistency and data compatibility prior to analysis. All auxiliary data were resampled to match the spatial resolution of the FVC (250 m). During the resampling process, bilinear interpolation was used for continuous variables, and nearest-neighbor interpolation was applied for categorical data. To control the data quality and reduce the propagation of resampling errors, (1) all input datasets were projected to a consistent coordinate system (WGS84); (2) post-resampling comparisons were made between the original and processed data using correlation and RMSE metrics; and (3) zonal statistics were cross-checked against officially reported administrative statistical data. All statistical outliers were identified and removed. Although resampling inevitably introduces uncertainty, particularly when upscaling low-resolution socioeconomic data, aligning all layers to the FVC resolution allows for the better preservation of vegetation signals and more robust driver attribution at the landscape scale.

2.3.2. Spatiotemporal Analysis

1.

Trend analysis of FVC and associated driving factors

The trend analysis of the FVC data was conducted using Sen’s slope estimator, which is a widely recognized and robust method for the analysis of long-term time series data [48,49,50]. The statistical significance of the identified trends was assessed using the Mann–Kendall (MK) test, a non-parametric method that does not require the data to follow a normal distribution and is robust to missing values and outliers. The formula was as follows:

$$\begin{array}{c}Se{n}^{\prime }sslope(S)=median{\displaystyle \frac{{FVC}_j-{FVC}_i}{j-i}}\forall j>i\end{array}$$

(2)

$$\begin{array}{c}Z=\left\{\begin{array}{c}{\displaystyle \frac{SS-1}{\sqrt{Var\left(SS\right)}}},S>0\\ 0,SS=0\\ {\displaystyle \frac{SS+1}{\sqrt{Var\left(SS\right)}}},S<0\end{array}\right.\end{array}$$

(3)

$$\begin{array}{c}SS={\sum }_i^{n-1}{\sum }_{i+1}^{n}sgn\left({FVC}_j-{FVC}_i\right)\end{array}$$

(4)

$$\begin{array}{c}sgn\left({FVC}_j-{FVC}_i\right)=\left\{\begin{array}{c}1,{FVC}_j-{FVC}_i>0\\ 0,{FVC}_j-{FVC}_i=0\\ -1,{FVC}_j-{FVC}_i<0\end{array}\right.\end{array}$$

(5)

$$\begin{array}{c}Var\left(SS\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)}{18}}\end{array}$$

(6)

where i and j denote the years under investigation in this study (2000–2020), and n represents the total number of years. Sen’s slope (S) represents the estimator trend in temporal FVC, where a positive slope indicates an increasing trend in FVC, whereas a negative slope indicates a decreasing trend. Z represents the test statistics of the MK statistical test. A two-tailed MK test was employed with a significance level of 0.05 (Z = 1.96), and the pixel-scale FVC trends were classified based on the results of Sen’s slope (S) and Z, as presented in

Table 2. Classification criteria for FVC trends.

Sen’s Slope (S)	Z	Classification
S > 0.005	Z > 1.96	Significant increase
S > 0.005	Z < 1.96	Slight increase
−0.005 < S < 0.005	-	Stable
S < −0.005	Z < 1.96	Slight decrease
S < −0.005	Z > 1.96	Significant decrease

.

2.

Stability analysis of FVC

The coefficient of variation (CV) was applied to assess the pixel stability and relative dispersion of the temporal variability in FVC over the long-term time series. The formula was as follows:

$$\begin{array}{c}CV={\displaystyle \frac{\sqrt{\frac{\sum {\left({FVC}_i-\overline{FVC}\right)}^2}{n-1}}}{\overline{FVC}}}\end{array}$$

(7)

where CV is the coefficient of variation of FVC. n represents the duration of the study. $\overline{FVC}$ denotes the average of FVC, and i represents the year. The CV values were stratified into five categories based on threshold values of 0.10, 0.15, 0.20 and 0.30, based on previous studies of vegetation dynamics in arid regions [23,51].

2.3.3. Driving Factor Detection

1.

Pearson correlation coefficients among FVC and associated driving factors

The Pearson correlation coefficient was used to analyze the linear relationships among the annual peak FVC and associated annual driving factors. The significance of the correlations was analyzed based on the t-test.

$$\begin{array}{c}r={\displaystyle \frac{\sum \left(x-\overline{x}\right)\times \left(y-\overline{y}\right)}{\sqrt{\sum {\left(x-\overline{x}\right)}^2\times {\left(y-\overline{y}\right)}^2}}}\end{array}$$

(8)

$$\begin{array}{c}t={\displaystyle \frac{r\sqrt{n-2}}{\sqrt{1-r^2}}}\end{array}$$

(9)

$$\begin{array}{c}df=n-2\end{array}$$

(10)

where r is the correlation coefficient between two variables. n is the sample size. df is the degree of freedom.

2.

Geographical detector

The geographical detector (GeoDetector) was applied to assess the initial conditions and interannual changes in these drivers for FVC temporal dynamics. GeoDetector, developed by [52,53,54], is a widely used spatial statistical method for the investigation of spatial variation and the quantification of the driving factors vegetation changes [24,55,56]. In this study, the factor detector module was used to quantify the explanatory power of each driving factor through the q-statistic (Equation (11)). Meanwhile, the interaction detection module was used to assess the interactions between pairs of factors (

Table 3. Interaction between two explanatory variables and their interaction impacts.

Interaction Relationship	Interaction Types	Description
q(X1 ∩ X2) < Min(q(X1), q(X2))	Weaken, nonlinear	The interaction of two variables nonlinearly weakens the impacts of a single variable.
Min(q(X1), q(X2)) < q(X1 ∩ X2) < Max(q(X1), q(X2))	Weaken, univariate	The interaction univariably weakens the impacts of single variables.
q(X1 ∩ X2) > Max(q(X1), q(X2))	Enhanced, bivariate	The interaction of two variables bivariable enhances the impacts of a single variable.
q(X1 ∩ X2) = q(X1) + q(X2)	Independent	The impacts of the two variables are independent.
q(X1 ∩ X2) > q(X1) + q(X2)	Enhance, nonlinear	The impacts of the variables are nonlinearly enhanced.

). Since GeoDetector is designed to work with categorical data, all selected continuous factors were discretized using the Jenks natural breaks classification method. This method minimizes within-class variance and reduces subjective bias in classification [57], making it widely applicable in GeoDetector-based studies [55,58,59,60]. For each factor, the classification scheme yielding the highest q-value was selected, ensuring that the spatial stratification best reflected the explanatory power of the variable.

$$\begin{array}{c}q=1-{\displaystyle \frac{\sum _{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}\end{array}$$

(11)

where h denotes the stratification of the study area based on FVC and the driving factors. N_h and N represent the number of units in stratum h and in the entire study area, respectively. ${\sigma }_h^2$ and ${\sigma }^2$ represent the variance in FVC within stratum h and the entire study area, respectively. SSW refers to the sum of within-group variances, while SST refers to the total variance of the entire study area. A higher q-value, ranging from 0 to 1, indicates a greater influence for the driving factor.

The interaction between two explanatory variables was assessed using the interaction detector module of the geographical detector. This allowed for the identification of whether the combined influence of the two factors enhanced, weakened or was independent of their individual effects on the spatial distribution of the dependent variable.

3.

Random forest model

A random forest model (RFM) was applied to quantify the annual drivers of FVC spatial variation, as well as to assess the initial conditions and interannual changes in these drivers for FVC temporal dynamics. RF is an ensemble learning algorithm that combines multiple regression trees using a bagging approach to improve the predictive performance and reduce overfitting [61]. The relative importance of the input individual variables was assessed following the development of the optimal evaluation model. The cumulative contribution of each categorized factor was quantified by summing the relative importance of all individual variables within the category.

$$\begin{array}{c}{I}_i={\sum }_1^K{\displaystyle \frac{1}{K}}\left({MSE}_k^i-{MSE}_k\right)\end{array}$$

(12)

where $I_i$ is the importance of variable i; K is the number of trees in the forest; ${MSE}_k^i$ is the estimation error, with predictor x being eliminated for the k decision tree; and ${MSE}_k$ is the forecasting error, with all predictors included in the k-th decision tree. The score $I_i$ indicated the relevant importance of the variable. The importance scores of all input variables were summed to be 1.

3. Results

3.1. Temporal and Spatial Variation in FVC

FVC exhibited pronounced seasonal dynamics and spatial heterogeneity. The intra-annual mean FVC ranged from 0.054 in February to 0.191 in July (Figure 3A). Over half of the total region was dominated by sparse vegetation (FVC < 0.2), which showed negligible seasonal dynamics. In contrast, the vegetation exhibiting strong seasonal variability was predominantly concentrated in ephemeral and perennial river systems, notably in the Tianshan and Altai Mountains of Northern Xinjiang, as well as the Tarim River Basin in the south (Figure 3B,C).

Over the past two decades, the annual peak FVC has exhibited a stepwise and consistent increase, characterized by the expansion of high-FVC areas and the contraction of low-FVC areas (Figure 3D). An increase in FVC was observed across approximately 355,700 km², while a decline was observed in about 100,400 km² of the region (Figure 3E). The overall conversion led to a significant increase in the annual peak FVC for the total region at a rate of 0.0508 yr⁻¹ (p < 0.01; Figure 3A).

3.2. Spatiotemporal Heterogeneity and Stability of FVC

The results of Sen’s slope estimator and the Mann–Kendall statistical test revealed pronounced spatiotemporal heterogeneity in FVC changes (Figure 4A,B). Notably, areas exhibiting significant improvements encompassed approximately 561,400 km², accounting for 33.72% of the total region. These improvements primarily spanned along the basins of rivers, including the Irtysh River, Ili River and Tarim River, where vegetation has benefited from abundant water resources. In contrast, significant degradation was observed across about 91,400 km², representing only 5.49% of the total region. These degradations were largely concentrated around rapidly urbanizing areas and the foothills (1000–2000 m) within the Ili River Basin.

The CV served as an indicator of interannual variability in FVC. An average CV of 0.163 was observed across the region, reflecting mild variability within basin regions and heightened fluctuations in mountainous zones. (Figure 4C). The regions with extremely high and low stability accounted for 37.39% and 9.4%, respectively (Figure 4D). Overall, the spatial pattern demonstrated a strong correlation with the FVC values and trends. Higher-FVC regions, including the Altai and Central Tianshan ranges, are typically associated with lower rates of increase and more stable conditions. However, in regions with low FVC values, the CV was dependent on the altitude. In the basins (<3500 m), the FVC values, CV and slope were all notably low. Meanwhile, in high-altitude regions (>3500 m), vegetation exhibited pronounced high CVs but insignificant long-term trends.

3.3. Changes in the Influencing Factors

The annual mean temperature during 2000 to 2020 ranged from −29.2 to 15.9 °C, which was associated with the latitude and altitude (Figure 5A). Meanwhile, the precipitation during 2000 to 2020 ranged from 3.9 to 753.2 mm and was predominantly concentrated in the mountainous areas (Figure 5B). There was a trend of warming and increased humidity in the eastern part of Xinjiang (Figure 5C,D), although the overall average temperature and cumulative precipitation in Xinjiang did not exhibit significant linear changes from 2000 to 2020 (Figure 5E,F). The soil water content depended on both the climate and terrain conditions, which was positive for precipitation and negative for the temperature at elevations below 5000 m (Figure 6). No significant changes in soil water content were observed between 2000 and 2020.

The land use types in Xinjiang were dominated by bare area, grassland, cropland and forests (Figure 7A). The FVC values for each land use type from highest to lowest were in the order of forests, cropland, urban areas, grassland and bare areas (Figure 7C). Between 2000 and 2020, the FVC of cropland, evergreen needle-leaved trees, shrubland, grassland, wetlands, lichens and mosses and urban areas increased significantly (p < 0.05). In addition, transitions in land cover types were mainly observed around natural and artificial water bodies, characterized by the expansion of cropland, forests and urban areas and a reduction in bare land (Figure 7A,B).

The population in Xinjiang was concentrated in urban areas along the Tarim River, the Ili River and the northern slopes of the Tianshan Mountains (Figure 8A), experiencing significant growth during 2000 to 2020 (Figure 8B). The managed areas, represented by grazing and fertilization, were distributed around the urban areas (Figure 8C,E). The total amount of livestock generally increased, accompanied by changes in the proportions of pigs, cattle and sheep (Figure 8D), and the fertilization rates increased significantly before 2015 and then remained stable afterwards.

3.4. Driving Mechanism of FVC

Vegetation growth was influenced by the topographical conditions, with a positive correlation observed with the slope and contributing a total of 4.7% importance in the RFM. The FVC of vegetation exhibited a positive correlation with precipitation and the soil water content and a negative correlation with the temperature. The climate conditions and soil water content contributed 8.23% and 10.29% to the overall importance, respectively. Besides the topographical and environmental conditions, anthropogenic activities play a major role in the FVC of vegetation in Xinjiang. Specifically, the land use type emerged as a predominant determinant of FVC, with vegetation cover showing a positive association with the land use intensity (Figure 9).

In contrast to the drivers of spatial heterogeneity in FVC values, the spatial heterogeneity of FVC trends was shaped by a combination of environmental conditions and anthropogenic activities, as well as their changes over time (Figure 10A,B). The fertilization rate yielded the highest q-value according to GeoDetector, whereas changes in the grazing index exhibited the highest importance in the RFM. These findings jointly suggest that the land use intensity serves as the predominant driver of desertification processes. The terrain conditions were the least important driving factors for FVC trends in both GeoDetector and the RFM. A comparative analysis of both individual and interactive effects revealed a synergistic enhancement between factors (Figure 10C). Notably, these enhancement effects were primarily non-linear. Among these factors, the combination of the fertilization rate and changes in soil water content had the greatest influence (q-value = 0.217), followed by the fertilization rate and land use type (q-value = 0.210).

4. Discussion

4.1. Spatiotemporal Variation in FVC

The observed spatial variation in vegetation coverage is consistent with previous studies, showing a pattern of higher values in the northwest and lower values in the southeast [19,23,26,29]. The Tianshan Mountains intercept moisture from the Atlantic under the influence of the westerly circulation. This results in a relatively humid environment in the northwestern mountainous areas and long-term drought in southeastern basins [62,63]. Additionally, the water has been spatiotemporally redistributed through the migration of meltwater and groundwater, as well as artificial irrigation [64,65,66]. Given that water availability is widely confirmed as the primary constraint on vegetation development [67,68], the vegetation distribution closely mirrored the precipitation patterns and exhibited significant spatial clustering near major natural watercourses and areas sustained by artificial irrigation.

The annual peak FVC for the total region has increased at a rate of 0.0508 yr⁻¹ over the past two decades, showing pronounced spatial clustering. The spatial analysis revealed that the spatial distribution of the FVC values, their CVs and the long-term trends shared similar spatial patterns. Due to prolonged drought and limited precipitation, desert regions in the basins were characterized by the absence of substantial vegetation cover, resulting in persistently low FVC values, low CVs and minimal long-term trends, as reported previously [15,69]. In contrast, high-altitude mountainous regions (>3500 m) with similarly low FVC values exhibited high interannual variability (CV > 0.5) but weak long-term trends, confirming that alpine ecosystems are particularly sensitive to environmental disturbances. In low-altitude regions (<3500 m) where vegetation was present, higher CVs were generally associated with significant FVC trends. Notably, these regions showing significant trends often overlapped with zones targeted by ecological restoration programs and land reclamation activities [27,70]. These pronounced and directional changes induced by anthropogenic environmental interventions would mask the intrinsic variability and time-lagged responses of vegetation to natural environmental drivers.

4.2. Driving Mechanism of Spatiotemporal Variation in FVC

The observed spatiotemporal variations in vegetation coverage in Xinjiang were the result of the interplay between the climatic conditions and human activities [5,20,71]. Notably, variations in land cover types are closely associated with differences in vegetation coverage (59.46%, Figure 9B), and land cover change has been widely recognized as an important factor driving the temporal dynamics of FVC (16.37%, Figure 10A), which was also demonstrated in previous studies [72,73]. Government-led ecological restoration programs, such as the ‘Grain-for-Green’ initiative, the water diversion projects in the Tarim River Basin and large-scale afforestation, have been widely acknowledged as among the most influential drivers of global greening during the early 21st century [74,75]. In these areas, land conversion from barren land to cropland or shrubland has significantly elevated FVC. However, changes in land use types driven by human activities exert multifaceted impacts on vegetation dynamics. In some cases, the conversion of cropland to grassland would result in decreased FVC, considering the differences in the vegetation structure and management intensity [1,74]. Meanwhile, population growth and associated urban expansion have driven the encroachment of built-up areas into ecologically valuable cropland and grassland. Although the FVC within urban areas increased significantly, such land conversion resulted in abrupt and spatially localized declines in the regional FVC. This encroachment of high-economic-output urban areas on limited land resources in arid regions underscores the intensifying conflict between rapid economic development and ecological conservation.

In addition to land cover changes, the land use intensity have been increasingly recognized as a critical driver of temporal changes in vegetation dynamics (37.6%, Figure 10A). Spatial variation in cropland vegetation cover was closely linked to nitrogen application rates, with a clear positive linear correlation observed between FVC increases and the total fertilizer input [35,76]. These findings highlight the critical role of improved water and fertilizer management in the planting density and in enhancing productivity, which is further corroborated by the steadily increasing agricultural output values reported in Xinjiang over the past two decades. Meanwhile, the increase in FVC in urban areas can also be attributed to the expansion of urban green spaces and enhanced water supply for urban vegetation maintenance. In addition, grazing plays a crucial role in land management practices, especially in grassland ecosystems. Intensified grazing pressure has markedly elevated the risk of vegetation degradation, manifested in recurrent transitions between grassland and barren land, which reflect the ecological instability induced by unsustainable land use [71,77,78].

The impacts of anthropogenic activities and the environmental conditions on FVC were not isolated (Figure 10C). The joint impact of the topographic and climatic conditions determined spatial variations in the temperature and hydrothermal resources across Xinjiang, characterized by an enhanced non-linear relationship, as revealed by the GeoDetector results and Pearson correlation analyses. As the elevation increases, there is a decrease in air pressure and an increase in radiation; combined with the varying distribution of precipitation and groundwater resources, this has resulted in the development of a diverse mountain vegetation ecosystem, ranging from forests to alpine tundra or lowland deserts [79,80,81]. The vegetation types with high FVC were primarily concentrated in the northwestern mountainous regions, where precipitation exceeds approximately 150 mm yr⁻¹, and along riverbanks, which allow sustainable water availability. Additionally, favorable temperature and hydrothermal conditions provided the necessary environmental foundations for anthropogenic-driven transformations in FVC, as evidenced by enhanced correlations in the GeoDetector results (Figure 10C). The expansion of both cropland and urban areas, leading to rapidly increasing and decreasing FVC, respectively, was observed in low-altitude (<3500 m) and flat regions (<30°) with habitable temperatures and abundant water resources, while relatively high precipitation and sustainable irrigation could support sustained vegetation growth [36]. Some studies have suggested that improvements in vegetation will enhance surface evapotranspiration, which could exceed the regional precipitation, potentially leading to water deficits [23,82,83]. Fortunately, the warming and wetter climate, along with the cross-regional water resource redistribution, can partially offset the losses from evapotranspiration [27,84,85].

4.3. Limitations and Future Perspectives

Multisource remote sensing data were utilized in this study to analyze the impacts of the environmental conditions and anthropogenic activities on vegetation coverage. The spatial resolutions of these datasets varied considerably, ranging from 30 m to 5 km. To ensure the reliability of the core results regarding vegetation cover changes, all datasets were resampled to match the spatial resolution of the FVC data. Meanwhile, efforts such as RMSE comparison, zonal statistics cross-checking and outlier removal were used to ensure data quality. However, the loss of fine-scale spatial information for the driving factors during the process of spatial resolution was inevitable. The annual mean temperature and cumulative precipitation were used as the climate indicators in this study. Besides annual data, the seasonal (May–October) climate conditions and annual extremes were also tested in the preliminary analysis. Among the indicators examined, the annual average temperature and cumulative precipitation exhibited the strongest correlations with vegetation cover across Xinjiang. The superior performance of annual-scale indicators could be attributed to their ability to capture integrated measures of thermal and hydrological availability throughout the year [86,87,88]. The temperature and precipitation during the non-growing season may indirectly influence vegetation growth by affecting soil moisture retention, snowmelt timing and the accumulation of thermal energy, especially in arid regions such as Xinjiang. However, these indirect effects require further investigation. In addition, this study conducted statistical analyses based on the annual peak FVC values, which may have overlooked the lagged responses of vegetation to variations in hydrothermal conditions, as well as the impacts of extreme climatic events [89,90,91]. Recognizing these constraints, future investigations will be undertaken to refine the current approach and provide more comprehensive insights.

5. Conclusions

The Xinjiang Uygur Autonomous Region, located in Northwestern China, is one of the typical arid regions, where vegetation is highly sensitive to global environmental changes. This study investigated the spatial and temporal features and driving factors of FVC in Xinjiang from 2000 to 2020. The main conclusions are as follows:

(1)

The vegetation coverage in Xinjiang was high in the northwest and low in the southeast, primarily distributed in mountainous regions and along river systems, where favorable hydrothermal conditions support vegetation growth;

(2)

The annual peak FVC increased significantly at a rate of 0.0508 yr⁻¹ over the study period, with 33.72% of the region showing significant improvements and 5.49% experiencing significant degradation;

(3)

Anthropogenic activities were the primary driving factors for the spatiotemporal variation in FVC, where the land use type divided the FVC values, and management was the most important factor for FVC trends.

In summary, due to the harsh natural conditions, characterized by prolonged droughts, intense evaporation, limited precipitation and frequent extreme weather events, the overall FVC in Xinjiang remained at a relatively low level. Nevertheless, despite the growing tension between humans and land use caused by economic development and population growth, vegetation cover has increased in the arid regions of Xinjiang. This improvement can be attributed to the combined effects of ecological restoration programs, improved land management practices and climate warming and humidification. Moreover, the ecological consequences induced by changes in vegetation cover call for further investigation, encompassing the ecosystem’s structure, function and service provision.