Land Cover Change as a Critical Driver of Vegetation Restoration in Water-Scarce Northern China

Abstract

The implementation of vegetation restoration projects in the Three-North Sandstorm Region has provided a critical opportunity for ecological rehabilitation. A thorough exploration of the mutual feedback mechanisms between land use and vegetation dynamics holds significant guiding value for ecological restoration practices. This study innovatively integrates the Hurst index, trend analysis, and significance testing to systematically analyze the characteristics of vegetation dynamics and their responses to land-use changes from 1990 to 2020. By constructing a multidimensional evaluation system, including the comprehensive land-use index (L) and land-use change rate (R), this research is the first to quantitatively reveal the relationship between land-use characteristics and the Normalized Difference Vegetation Index (NDVI) across different desertification regions. Key findings include: (1) Vegetation in the study area exhibited a significant improvement trend, with 92.12% of regions showing a Hurst index > 0.5, indicating strong persistence in vegetation recovery. (2) Land-use transitions revealed a consistent increase in construction land across all sandy regions, accompanied by the conversion of unused land to grassland. (3) The fluctuation of R gradually narrowed from 1990 to 2020, while the L demonstrated a significant upward trend. (4) The R generally exerted a negative influence on NDVI, whereas the L exhibited a threshold effect. Specifically, NDVI increased with rising L when L < 215 (e.g., MWSD, KBQD regions), but declined when L exceeded 215 (e.g., SNS region). This suggests that frequent land-use changes hinder vegetation growth, while moderate increases in land-use intensity initially promote biomass accumulation—until a critical threshold is surpassed, beyond which further intensification exerts adverse effects. One-way ANOVA and significance testing further demonstrated that climate predominantly drove NDVI increases in cropland, forest, shrubland, and grassland, whereas human activities played a decisive role in vegetation growth within unused or sparsely vegetated areas. This study not only quantitatively identifies critical thresholds for land-use impacts on vegetation but also provides a scientific foundation for targeted ecological restoration strategies, offering practical insights for reconciling ecological conservation with land-use demands.

1. Introduction

Desertification ranks among the world’s most critical ecological challenges. Across the globe, drylands extend over about 41% of the Earth’s terrestrial area and support over 38% of the world’s population, which totals 6.5 billion people [1]. In response to desertification, governments around the world have adopted a range of interventions, including sand stabilization, revegetation [2], the establishment of windbreaks, and natural sealing techniques [3,4]. As a pivotal component of the Groundwater-Soil-Plant-Atmosphere Continuum (GSPAC), vegetation plays an indispensable role in global carbon-water cycles and energy fluxes [5,6,7]. Its dynamic changes profoundly influence regional ecological security, particularly in arid and semi-arid climatic zones [8]. Since the late 1990s, China has implemented a series of ecological restoration initiatives [9], including the “Grain for Green Program” (GGP) and the “Grazing Withdrawal Program” (GWP) in its northwestern regions, aimed at combating desertification and enhancing vegetation coverage [10,11,12]. The execution of these programs has also triggered significant transformations in land-use and land cover [13,14,15]. With the advancement of vegetation restoration projects, surface vegetation coverage in China’s arid and semi-arid regions has progressively improved, making related studies a focal point in ecological and remote sensing research [8,16,17].

Over the past few decades, the Normalized Difference Vegetation Index (NDVI) has served as a pivotal indicator for vegetation monitoring, extensively employed to elucidate vegetation responses to environmental changes. Significant breakthroughs have been made in understanding vegetation dynamics and ecosystem adaptability [18,19,20,21,22]. Numerous studies have documented a global greening trend [23,24], particularly pronounced in mid-to-high latitude regions of the Northern Hemisphere [25], with similar patterns observed elsewhere [26,27,28,29,30]. Investigating vegetation dynamics is crucial for unraveling climate-vegetation feedback mechanisms at regional scales [31,32,33]. Concurrently, human activities—such as urbanization, land-use changes, and ecological restoration projects—are profoundly altering vegetation growth patterns [34,35], driving regional land-cover transformations [36,37], and enhancing vegetation coverage [38,39]. This impact is particularly evident in China’s vegetation restoration project areas [37,40,41,42]. However, research indicates that anthropogenic vegetation restoration does not always promote growth and may even exacerbate environmental degradation in some ecologically fragile regions [43]. Moreover, human-induced land-use changes have significantly influenced vegetation distribution and composition [44,45,46], making land-use modification a key pathway through which human activities affect vegetation dynamics [47]. Studies reveal divergent vegetation growth trends under different land-use types [48]. While afforestation can positively impact local vegetation [49], agricultural land conversion may have adverse effects [50]. Nevertheless, quantitative assessments of land-use impacts on vegetation restoration remain limited, particularly in identifying critical intensity thresholds that govern vegetation growth responses—a central challenge in current research. Although existing studies have explored the general relationship between land-use change and vegetation indices, three critical research gaps remain in the field. First, most analyses focus on changes in single land-use types or localized regions, lacking systematic quantification and assessment of “comprehensive land-use intensity”. Second, current research often fails to clearly distinguish between the differential impacts of land-use “change” and “stable states” on vegetation dynamics, making it difficult to reveal the true mechanisms by which long-term human activities affect ecological resilience. Third, there is still insufficient empirical support and quantitative identification regarding whether ecological thresholds exist for land-use intensity—specifically, the phenomenon where vegetation responses undergo abrupt changes once a certain critical intensity is exceeded. Therefore, this study aims to develop a comprehensive land-use intensity indicator, analyze its nonlinear relationship with vegetation indices, identify potential thresholds, and separately quantify the contributions of land-use change and stability to vegetation dynamics, thereby addressing the aforementioned theoretical and empirical gaps.

The Three-North Afforestation Program (TNAP) region primarily comprises four key zones: the arid Northwest Desert, the Loess Plateau with its hilly and gullied terrain, the Sandy Desertification Area, and the agricultural plains of Northeast and North China [51]. Among these, the Sandy Desertification Area represents China’s most severely desertified region and serves as the critical frontline for TNAP’s desertification control efforts. The study area, located within China’s largest temperate water-limited region, represents a distinctive and ecologically sensitive zone dominated by extensive sandy land [52]. Over the past three decades, this region has undergone significant vegetation greening driven by large-scale ecological restoration initiatives [53]. This ecologically fragile zone has experienced pronounced impacts from both climate change and anthropogenic activities [54,55], with subsequent ecological restoration initiatives demonstrating significant improvements in vegetation conditions [20,56]. This study aims to: (1) investigate the spatiotemporal evolution of vegetation indices across different phases through integrated Hurst analysis, trend analysis, and significance testing; (2) examine land-use dynamics during vegetation recovery, including transitional patterns and temporal trends in land-use characteristics (e.g., land-use intensity and change rates); and (3) quantitatively assess the relationship between land-use and vegetation indices, specifically evaluating the impact and threshold effects of comprehensive land-use intensity on vegetation indices, while differentiating the quantitative influences of land-use changes versus stability on vegetation dynamics.

2. Materials and Methods

2.1. Study Area

The study area is located within the desertification-prone region of China’s Three-North Shelterbelt Program, encompassing five key areas: the Mu Us Desert (MWSD), the Kubuqi Desert (KBQD), the Horqin Sandy Land (KEQS), the Hunshandake Sandy Land (HSDKS), and the Songnen Sandy Land (SNS). These regions predominantly fall within arid and semi-arid climate zones, characterized by an annual precipitation range of 20–450 mm, significant diurnal temperature variations, and an arid climate where evaporation exceeds precipitation. Due to limited rainfall, the primary land-use types consist of grasslands and deserts, with most areas exhibiting sparse vegetation and extremely fragile ecosystems before the initiation of ecological restoration efforts. Furthermore, the ecological environment in these regions demonstrates high sensitivity, influenced by both climatic conditions and human activities (Figure 1).

2.2. Datasets

NDVI Data: The NDVI dataset was derived from Landsat series imagery (Landsat 5, 7, and 8) spanning 1990–2020, specifically covering the periods of 1990–2011, 2012, and 2013–2020, with a spatial resolution of 30 m. Image preprocessing involved study area extraction and cloud removal. The maximum value composite (MVC) algorithm [57,58,59] was applied to obtain annual NDVI values for the vegetation growing season (June–August), followed by vegetation coverage calculations. Land-Use Data: The China Land Cover Dataset (CLCD) provides annual land cover information for China from 1985 to 2021 [60]. The dataset categorizes land-use into nine classes with the following codes: Cropland (1), Forest (2), Shrub (3), Grassland (4), Water (5), Snow/Ice (6), Barren (7), Impervious (8), and Wetland (9). Desert (sandy land) distribution data, obtained from the National Tibetan Plateau Data Center (http://data.tpdc.ac.cn) (accessed on 15 May 2025), comprises 1:100,000-scale desert (sandy land) distribution data for China [61]. Five arid and semi-arid regions were selected for analysis: the Kubuqi Desert (KBQD), Mu Us Sandy Land (MWSD), Hunshandake Sandy Land (HSDKS), Horqin Sandy Land (KEQS), and Songnen Sandy Land (SNS).

2.3. Methods

2.3.1. Hurst Exponent

The core function of the Hurst exponent is to quantify the long-term memory of a time series, determining whether its dynamic behavior is random, trend-enhancing, or mean-reverting [62]. Its significance lies in revealing the hidden persistence patterns within natural and social systems. This metric breaks through the limitations of the traditional random walk hypothesis, providing critical insights for fields such as water resource management and climate change analysis, thereby serving as an important tool for understanding the intrinsic dynamical mechanisms of complex systems [63]. The Hurst exponent can be used to assess the validity of time series consistency in vegetation indices such as NDVI based on their historical variation patterns. This study employs the R/S analysis method to estimate the Hurst exponent. For a given NDVI time series ${NDVI}_i$, where i = 1, 2, 3, …, n, and for any positive integer t ≥ 1, the mean series of this time sequence is defined as follows:

$$\overline{∆NDVI\left(t\right)}={\displaystyle \frac{1}{t}}{\sum }_{i=1}^t{NDVI}_i\left(t=\mathrm{1,2},\dots ,n\right)$$

(1)

$$X\left(m\right)={\sum }_{i=1}^t\left(∆{NDVI}_i-\overline{∆NDVI\left(t\right)}\right)\left(1\le \mathrm{m}\le \mathrm{t}\right)$$

(2)

$$R\left(t\right)=\max \left(X\left(m\right)\right)-\min \left(X\left(m\right)\right)\left(t=\mathrm{1,2},\dots ,n\right)$$

(3)

$$S\left(t\right)={[{\displaystyle \frac{1}{t}}{\sum }_{i=1}^t{(∆{NDVI}_i-∆\overline{NDVI\left(t\right)})}^2]}^{{\displaystyle \frac{1}{2}}}\left(t=\mathrm{1,2},\dots ,n\right)$$

(4)

where X(m) represents the cumulative deviation, R(t) denotes the range, and S(t) is the standard deviation. By comparing R(t)/S(t) ≅ R/S, if a relationship R/S ∝ tH exists, it indicates the presence of the Hurst phenomenon in the analyzed time series. The parameter H, known as the Hurst exponent, reflects the randomness or persistence characteristics of the time series. When 0 < H < 0.5, the time series exhibits correlation, but its future overall trend will be opposite to the past. The closer H is to 0, the stronger the anti-persistence. Conversely, when 0.5 < H < 1, the time series demonstrates persistence. The closer H is to 1, the stronger the persistence and the greater the correlation.

2.3.2. Trend Analysis

This study aims to reveal vegetation dynamics, including the detection of “increase” or “decrease” trends, by analyzing the trends in the Normalized Difference Vegetation Index (NDVI). The research focuses on the trend analysis of the maximum NDVI values during the growing season (June to October) in several sandy regions from 1990 to 2020. A univariate linear regression analysis was applied to process the multi-year NDVI values for each grid cell, with the regression slope reflecting the spatial variation trend of NDVI over the study period. The slope is calculated as follows:

$$Slope={\displaystyle \frac{n\times \sum _{i=1}^n\left(i\times NDV{I}_i\right)-\sum _{i=1}^{n}i\sum _{i=1}^{n}NDV{I}_i}{n\times \sum _{i=1}^n{i}^2-{\left(\sum _{i=1}^{n}i\right)}^2}}$$

(5)

where Slope represents the slope of the NDVI regression equation for each grid cell, n denotes the number of years, and NDVI_i is the NDVI value for the i-th year. If Slope > 0, it indicates an increasing trend in vegetation coverage; if Slope = 0, it suggests no change in vegetation coverage; if Slope < 0, it signifies a declining trend in vegetation coverage. Regarding the determination of statistical significance for the trend slope, a linear regression model was fitted using R. The significance of the slope coefficient was directly assessed based on the p-value (reported as Pr(>|t|)). The conventional standard was applied whereby a p-value less than 0.05 is considered to indicate statistical significance at the 95% confidence level.

2.3.3. F-Test

To further assess the changes in vegetation coverage, an F-test was employed to analyze the significance of NDVI trends, which indicates the confidence level of the observed trend variations.

$$a=avg\left(y\right)-slpoe\times avg\left(x\right)$$

(6)

$$\widetilde{{\mathrm{y}}_{\mathrm{i}}}=\mathrm{a}+\mathrm{s}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}\times {\mathrm{x}}_{\mathrm{i}}$$

(7)

$$\mathrm{U}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}\left(\widetilde{{\mathrm{y}}_{\mathrm{i}}}-\stackrel{\mathrm{‾}}{\mathrm{y}}\right)$$

(8)

$$\mathrm{Q}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\left({\mathrm{y}}_{\mathrm{i}}-\widetilde{{\mathrm{y}}_{\mathrm{i}}}\right)}^2$$

(9)

$$\mathrm{F}=\mathrm{U}\times {\displaystyle \frac{\mathrm{n}-2}{\mathrm{Q}}}$$

(10)

where n represents the study time series; U denotes the sum of squared errors; Q indicates the regression sum of squares; $\widetilde{{\mathrm{y}}_{\mathrm{i}}}$ stands for the fitted regression value; $\stackrel{\mathrm{‾}}{\mathrm{y}}$ is the mean value over n years; yi represents the value for the xi-th year.

2.3.4. Land-Use Change Characteristics

Land-Use Change Indexes

To quantitatively analyze land-use changes in the study area, the comprehensive land-use degree index and the land-use degree change rate were selected, with reference to land-use classification types.

(1) The comprehensive land-use degree index quantitatively describes the intensity of regional land utilization, expressed as follows:

$$L_i={\sum }_{i=1}^n{A}_i{C}_i \mathrm{\ast } 100\mathrm{\%}$$

(11)

where L_i represents the comprehensive land-use index of the study area; A_i denotes the percentage of area for the i-th level of land-use intensity within the study region; C_i indicates the grading index for the i-th level of land-use intensity in the study area; and n stands for the total number of land-use classification levels.

(2) The degree of land-use change rates can reflect the rate of change in regional land-use intensity (unit: %), making it suitable for comprehensive evaluation of land-use intensity, with its expression being written as follows:

$$R={\displaystyle \frac{L_b-L_a}{L_a}} \mathrm{\ast } 100\mathrm{\%}$$

(12)

where R represents the rate of land-use change, L_b denotes the comprehensive land-use index of the region at time b, and L_a indicates the comprehensive land-use index of the region at time a.

Relationship Between Land-Use Change Characteristics and Vegetation Index

This study categorizes the research area into three types: Type I represents land-use change areas where land-use types altered in every year between 1990 and 2020; Type II denotes stable land-use areas where land-use types remained unchanged during this period; and Type III encompasses the entire study region.

For Type I, Type II, and Type III regions, the average NDVI values and land-use status indicators were calculated for each corresponding area. Using the average NDVI as the dependent variable and the comprehensive land-use index, along with the land-use intensity change rate, as independent variables, a regression model was constructed to analyze the relationship between land-use changes and NDVI. The impact of land-use changes on NDVI was investigated through statistical analysis. By comparing the differences in average NDVI values between Type II and Type I regions, the influence of land-use changes on NDVI was assessed, and one-way ANOVA was employed to test the significance of this impact (Figure 2).

2.3.5. Statistical Analysis and One-Way ANOVA

The impact of climatic and human activity factors on NDVI from 1990 to 2020 was analyzed using statistical analysis and one-way ANOVA. The vegetation distribution areas were classified into two categories: (1) Climate-Dominated regions (CD), where land-use types remained unchanged throughout the 30 years; and (2) Human Activity-Dominated regions (HD), where land-use types exhibited annual variations. When calculating the average NDVI values for CD and HD regions, areas with elevation differences exceeding 50 m were excluded to ensure comparability between the NDVI values of CD and HD regions. The difference in NDVI between human activity-dominated and climate-dominated regions was calculated using the following formula:

$$\mathrm{F}={\sum }_{\begin{array}{c}0\le i \le m\\ 0<j<n \end{array}}({NDVI}_{Hij}-{NDVI}_{Cij})/n$$

(13)

where “i” and “j” denote specific years and different land types, respectively, NDVI_Cij represents the average NDVI value for a specific land type in a given year in the CD region, while NDVI_Hij represents the average NDVI value for a specific land type in the HD region. “n” and “m” denote the total number of years and land types examined, respectively. A positive F value suggests an increase in the NDVI value for a specific land type, indicating a positive impact of human activities on vegetation growth after excluding the influence of climate. Conversely, a negative F value suggests that climate dominates the growth of vegetation.

The statistical significance of changes induced by the factor was assessed using one-way analysis of variance (ANOVA). Regions were classified into two categories based on land-use cover: High Vegetation Coverage areas (HVCs), including cropland, forests, grasslands, and shrublands, and Low Vegetation Coverage areas (LVCs), encompassing snow areas, impervious surfaces, and barren land. ANOVA was employed to determine significant differences between CD and HD under varying vegetation covers (HVC or LVC) through Formulas (4) and (5).

$$SST={\sum }_{i=1}^m{\sum }_{j=1}^{ni}{\left(x_{ij}-\mu \right)}^2$$

(14)

$$DT=N-1$$

(15)

where SST indicates the total variation, which is the discrepancy between measurement x_ij and the total mean μ; DT is the degrees of freedom; m is the number of observation groups; n_i refers to the number of observations in each group; and N is the total number of observations.

3. Results

3.1. Vegetation Trend Analysis

3.1.1. Vegetation Coverage and NDVI Changes

Based on the vegetation coverage classification [64] (

Table 1. Vegetation coverage classification.

Grade	Vegetation Coverage (%)	Surface Landscape Characteristics
High coverage (V)	≥70	Dense vegetation such as grasslands and forests
Relatively high coverage (IV)	50–70	Patchy sandy areas, medium-to-high-yield grasslands, and forests
Moderate coverage (III)	30–50	Fixed sand dunes, farmland, and forests
Relatively low coverage (II)	10–30	Semi-mobile sand dunes, low-yield grasslands, and sparse forests
Low coverage (I)	<10	Mobile sand dunes, residential areas, water bodies, transportation routes, and construction sites

), this study calculated the area distribution of several sandy lands under different vegetation coverage levels. The results (Figure 3) indicate that over the past 30 years, SNS exhibited the highest proportion of high-coverage areas (V), followed by moderate coverage (III), low coverage (II), and minimal coverage (I). In HSDKS and KBQD, low-coverage areas (II) accounted for the largest proportion. Conversely, KEQS and MWSD were dominated by moderate-coverage areas (III), with minimal-coverage areas (I) being the least prevalent.

This study calculated the regional average NDVI values and their temporal trends for five sandy areas from 1990 to 2020, with the results illustrated in Figure 4. The NDVI values across all sandy areas generally exhibited an increasing trend over time. Among them, SNS and KEQS showed the highest average NDVI values, reaching 0.36, while MWSD had the lowest average at 0.19. The average NDVI values for KBQD and HSDKS were 0.3 and 0.27, respectively. Regarding the annual growth rate of NDVI, SNS, and MWSD, demonstrated the fastest increases at 0.0041/yr and 0.004/yr, respectively. KBQD and KEQD followed with growth rates of 0.0039/yr and 0.0029/yr, while HSDKS had the slowest annual growth rate at just 0.001/yr.

Through analyzing the areal extent of NDVI trend variations during different periods (

Table 2. Area proportion of NDVI trend significance at different research stages.

Year	Percentage of Change Area (%)	MWSD	KEQS	KBQD	SNS	HSDKS
1990–2000	Significantly increased	55.08	42.57	56.25	29.88	42.75
	Not significantly increased	6.12	4.73	6.25	3.32	4.75
	Significantly decreased	36.86	50.07	35.63	63.46	49.88
	Not significantly decreased	1.94	2.64	1.88	3.34	2.63
2000–2010	Significantly increased	78.03	54.54	64.98	79.83	44.73
	Not significantly increased	8.67	6.06	7.22	8.87	4.97
	Significantly decreased	22.14	37.43	26.41	20.24	47.79
	Not significantly decreased	1.17	1.97	1.39	1.07	2.52
2010–2020	Significantly increased	85.77	82.26	85.59	77.04	74.34
	Not significantly increased	9.53	9.14	9.51	8.56	8.26
	Significantly decreased	4.47	8.17	4.66	13.68	16.53
	Not significantly decreased	0.24	0.43	0.25	0.72	0.87
1990–2020	Significantly increased	89.46	73.89	85.86	80.91	58.86
	Not significantly increased	9.94	8.21	9.54	8.99	6.54
	Significantly decreased	0.57	17.01	4.37	9.60	32.87
	Not significantly decreased	0.03	0.90	0.23	0.51	1.73

and

Table 3. Area of significant trends in the NDVI across different phases.

Year	Changes in Area (km2)	KEQS	SNS	MWSD	KBQD	HSDKS
1990–2000	Significantly increased	53,202.36	15,391.99	44,899.92	49,063.75	112,755.75
	Not significantly increased	5911.37	1710.22	4988.88	5451.52	12,528.41
	Significantly decreased	62,662.14	32,759.03	30,103.84	31,032.18	131,317.22
	Not significantly decreased	3133.10	1637.95	1505.19	1551.60	6565.86
2000–2010	Significantly increased	68,212.19	41,164.53	63,690.04	56,614.83	117,871.97
	Not significantly increased	7579.13	4573.83	7076.67	6290.53	13,096.88
	Significantly decreased	46,818.42	5554.68	10,269.82	23,061.60	125,916.76
	Not significantly decreased	2464.12	292.35	540.51	1213.76	6627.19
2010–2020	Significantly increased	102,908.81	39,760.76	69,983.12	74,581.85	195,873.92
	Not significantly increased	11,434.31	4417.86	7775.90	8286.87	21,763.76
	Significantly decreased	10,194.21	7036.43	3627.14	4096.41	43,581.37
	Not significantly decreased	509.71	351.82	181.35	204.82	2179.06
1990–2020	Significantly increased	92,471.76	41,731.10	73,000.96	74,885.22	155,016.24
	Not significantly increased	10,274.64	4636.78	8111.21	8320.58	17,224.02
	Significantly decreased	21,211.10	4956.64	441.63	3776.19	86,708.92
	Not significantly decreased	1116.37	260.87	23.24	198.74	4563.62

) and their spatial distribution patterns (Figure 5), the results demonstrate that the overall growth rate of NDVI significantly outpaced the degradation rate. Decadal statistics reveal that the SNS experienced its most rapid NDVI growth during 2000–2010, while other regions achieved their peak growth rates in 2010–2020. Over time, the area showing a gradual decrease gradually diminished. During 1990–2000, the SNS exhibited the largest area of NDVI reduction, followed by KEQS, HSDKS, KBQD, and MWSD. Between 1990 and 2020, the MWSD displayed the most extensive area of NDVI increase, with 89.4% of its territory showing vegetation improvement by the end of 2020. Throughout the three-decade vegetation restoration process, the proportions of NDVI-increasing areas in KBQD, SNS, KEQS, and HSDKS reached 85.8%, 80.9%, 73.8%, and 58.86%, respectively.

3.1.2. Persistence of NDVI Change Trends

The spatial distribution of the Hurst index, when overlaid with NDVI change trends, reveals the persistence characteristics of these trends. The Hurst index is classified into four levels: strong anti-persistence (Level I: 0–0.25), weak anti-persistence (Level II: 0.25–0.5), weak persistence (Level III: 0.5–0.75), and strong persistence (Level IV: 0.75–1). Trend significance is categorized into five levels based on slope values: significant degradation (Slope ≤ −0.0119), slight degradation (−0.0119 < Slope ≤ −0.0014), no significant change (−0.0014 < Slope ≤ 0.0014), slight increase (0.0014 < Slope ≤ 0.0052), and significant increase (Slope > 0.0052).

Analysis of the five regions shows the following persistence patterns and area proportions for NDVI trends: strong persistence with significant increase (31.77%) and strong persistence with slight increase (39.40%) dominate, followed by strong persistence with no significant change (9.70%), strong persistence with slight degradation (4.95%), and weak persistence with slight increase (3.21%). Overall, strong persistent growth accounts for 71.2% of the total vegetation recovery trend in these regions, indicating its predominance (Figure 6).

3.2. Land-Use Change

3.2.1. Land-Use Transition

An analysis of land-use transitions and distribution in the study area from 1990 to 2020 (Figure 7 and Figure 8) reveals that in the SNS, cropland covers the largest area, followed by forestland, shrubland, and grassland, with grassland occupying the smallest proportion. Between 1990 and 2020, cropland decreased by 2782 km², primarily transitioning into forestland and grassland, while forestland also experienced a decline. In contrast, construction land and grassland expanded. In the HSDKS, grassland dominates, followed by shrubland and forestland. By 2020, cropland and forestland decreased by 2524 km² and 375 km², respectively, largely converted into grassland. Grassland and construction land continued to expand, increasing by 2227 km² and 259 km², respectively. The KEQS region saw reductions in forestland, cropland, and unused land, declining by 1821 km², 4375 km², and 865 km², respectively, with most transitions favoring grassland. Construction land expanded by approximately 143 km² due to conversions from grassland. In the KBQD, cropland, grassland, and construction land exhibited net growth, expanding by 958 km², 8840 km², and 383 km², respectively, while other land-use types diminished. Forestland, grassland, and water bodies supplemented cropland, and unused land was primarily converted into construction land. Compared to other sandy regions, the MWSD has the smallest proportion of cropland. Grassland, forestland, cropland, and construction land increased by 11,740 km², 1117 km², 484 km², and 64 km², respectively, with grassland and cropland mainly derived from shrubland and forestland, while construction land originated from unused land.

Overall, as human activities intensify the utilization of land resources in sandy regions, construction land continues to expand across all areas, and unused land is increasingly converted into grassland. With improvements in land-use conditions in these regions, cropland in MWSD and KBQD demonstrates a growing trend. Conversely, under the Grain for Green policy, cropland in SNS and KEQS has been steadily declining.

3.2.2. Characteristic of Land-Use Change

As illustrated in Figure 9, analysis of the land-use change characteristics of the study area revealed that SNS exhibited the highest land-use intensity (L) (Figure 9A), followed by KEQS, while MWSD has the lowest average L value. Influenced by the Grain for Green policy, the land-use intensity of SNS has gradually declined, with an annual average decrease rate of −0.482/yr (p < 0.001) in the comprehensive land-use index. KEQS and HSDKS showed a stable yet slight downward trend in land-use intensity (Figure 9B). In contrast, KBQD and MWSD demonstrated an overall increase in land-use intensity, with annual average growth rates of 0.325/yr (p < 0.001) and 0.4895/yr (p < 0.001), respectively. Over time, land-use changes in the sandy area have gradually stabilized. Among these regions, SNS and MWSD exhibited more pronounced fluctuations in land-use, while HSDKS displayed the least variability.

3.3. The Relationship Between Land-Use and Vegetation

3.3.1. Correlation Between Land-Use Characteristic Indices and NDVI

According to the correlation analysis results (Figure 10), there is an overall significant negative correlation between R and NDVI, with a correlation coefficient of −0.27, indicating that higher R values correspond to lower NDVI values. It should be noted that the relationship between L and NDVI varies across different L levels: when L is below 215, the two exhibit a significant positive correlation, whereas when L exceeds 215, the correlation becomes negative. This suggests that within a certain range of L values, there may exist a critical threshold at which the relationship between L and NDVI shifts from positive to negative.

3.3.2. Quantifying the Impact of Land-Use Characteristic Values on NDVI

Multiple linear regression models were established using all data to analyze the influence of L and R on NDVI, including the NDVI of the entire study area (A_NDVI) and the NDVI of regions with land-use changes (C_NDVI). The results are presented in

Table 4. The results of multiple linear regression analysis, including accuracy measures and fitted equations. Here, “Class” denotes vegetation areas corresponding to changed or unchanged land-use types, “R²” indicates modeling accuracy, “Fitted Equation” represents the regression equation, and “p-value” shows the significance of regression coefficients (with smaller values indicating higher significance). “Numbers” refers to the count of data points used for modeling.

Class	R2	Fitted Equation	p Value
C_NDVI	0.615	C_NDVI = 0.419 ∗ L − 0.042 ∗ R + 0.807	4.59 × 10−31
A_NDVI	0.651	A_NDVI = 0.492 ∗ L − 0.056 ∗ R + 0.860	7.13 × 10−34

.

The modeling accuracy for A_NDVI and C_NDVI, as indicated by R² values of 0.61 and 0.65, respectively, demonstrates that the models exhibit strong fitting and predictive performance. The results reveal that L generally exerts a positive influence on both A_NDVI and C_NDVI (as evidenced by positive regression coefficients of L), whereas R predominantly shows a negative impact (reflected by negative regression coefficients of R). Notably, the absolute effect of L on NDVI significantly outweighs that of R, as indicated by the substantially larger absolute value of L’s regression coefficient compared to that of R.

Analysis of the correlation coefficient matrix (Figure 10) reveals a distinct segmented characteristic in the relationship between L and NDVI. Given the varying NDVI growth patterns and land-use conditions across different sandy areas, separate modeling analyses were conducted for L and R in relation to NDVI for each sandy region, with the following results (

Table 5. The accuracy of multiple linear regression analysis and fitted equations across different aeolian sand regions. Here, “area” denotes distinct aeolian sand zones, “Class” indicates land-use type changes or the entire study area, “R²” represents modeling accuracy, “Fitted Equation” shows the regression equation, “p-value” indicates the significance of regression coefficients (with smaller values denoting higher significance), and “Numbers” refers to the count of data points used for modeling.

Area	Class	R2	Fitted Equation	p Value
MWSD	C_NDVI	0.696	C_NDVI = 0.761 ∗ L − 0.023 ∗ R + 1.023	1.87 × 10−7
	A_NDVI	0.616	A_NDVI = 0.671 ∗ L − 0.023 ∗ R + 0.939	3.94 × 10−6
KBQD	C_NDVI	0.628	C_NDVI = 0.970 ∗ L − 0.048 ∗ R +1.031	1.60 × 10−6
	A_NDVI	0.589	A_NDVI = 0.902 ∗ L − 0.056 ∗ R + 1.025	6.20 × 10−6
HSDKS	C_NDVI	0.015	C_NDVI = 0.704 ∗ L − 0.056 ∗ R + 0.603	8.22 × 10−1
	A_NDVI	0.031	A_NDVI = 0.626 ∗ L + 0.206 ∗ R + 0.750	6.67 × 10−1
KEQS	C_NDVI	0.029	C_NDVI = 0.936 ∗ L + 0.023 ∗ R + 1.157	6.68 × 10−1
	A_NDVI	0.049	A_NDVI = 0.717 ∗ L + 0.800 ∗ R + 1.244	5.09 × 10−1
SNS	C_NDVI	0.338	C_NDVI = −0.645 ∗ L + 0.0199 ∗ R + 2.599	3.84 × 10−3
	A_NDVI	0.376	A_NDVI = −0.817 ∗ L − 0.005 ∗ R + 3.123	1.72 × 10−3

, Figure 11):

In the MWSD, KBQD, HSDKS, and KEQS sandy areas, L exhibits a positive effect on NDVI, with regression coefficients all exceeding 0.5, indicating that NDVI increases with rising L. Particularly in the MWSD and KBQD regions, the modeling accuracy of L and R for NDVI is relatively high, with an average R² exceeding 0.5, while the L values in these areas are generally low (L < 215). In contrast, in SNS sandy areas, L shows a negative contribution to NDVI, with a regression coefficient below −0.5, and the L values here are notably higher (L > 215). By statistically analyzing different L values and the coefficients of the L variable in the regression models (Figure 11), it was observed that the influence of L on NDVI may exhibit a threshold effect. When L is low, its increase positively affects NDVI; however, once L surpasses a certain threshold, further increases lead to an inverse relationship with NDVI.

3.3.3. Extent and Significance of Land-Use Change Impacts on Vegetation

To validate the above findings, we conducted statistical analysis and significance tests on NDVI values in areas with and without land-use changes. Vegetated regions were classified into two categories: climate-dominated (CD) areas, where land-use types remained unchanged over the 30 years, and human-dominated (HD) areas, where land-use types changed annually. The NDVI characteristics of both regions were analyzed. The results show that the total area of annually changing land-use types is significantly larger than that of unchanged land-use types (Figure 12a). Subsequently, the average NDVI values for CD and HD regions across different land-use types were calculated, along with their F-values (Figure 12b). For cropland, forest, shrubland, and grassland, the F-values were predominantly negative, indicating significantly lower NDVI in human-dominated areas compared to climate-dominated areas. Conversely, for other land-use types, the F-values were mostly positive, suggesting significantly higher NDVI in human-dominated areas. This implies that climate factors play a more substantial role in vegetation growth for cropland, forest, shrubland, and grassland, whereas human activities exert greater influence on vegetation growth in other land-use types.

Given the substantial NDVI differences between high-NDVI land-use types (cropland, forest, shrubland, and grassland) and low-NDVI land-use types (barren land, snow/ice, and impervious surfaces), we further categorized them into high vegetation cover (HVC) regions (cropland, forest, shrubland, and grassland) and low vegetation cover (LVC) regions (all other land-use types). One-way ANOVA results (Figure 13) further corroborate the above conclusions. In HVC regions, the NDVI of CD areas is significantly higher (p < 0.01) than that of HD areas, confirming that climate predominantly drives NDVI variations in high vegetation cover regions under similar conditions. In contrast, human activities play a decisive role in vegetation growth in barren and sparsely vegetated areas.

4. Discussion

4.1. Challenge and Insight of Ecological Restoration Projects

The impact of vegetation restoration on local sustainable development remains contentious. Conventional perspectives suggest that ecological restoration projects can enhance vegetation coverage and mitigate desertification [12,13]. However, emerging research indicates that such initiatives may not uniformly benefit vegetation growth and could even exacerbate ecosystem degradation [43]. Human activities have been demonstrated to contribute to regional ecosystem deterioration [65], manifesting as grassland degradation and soil erosion [66]. Over recent decades, intensive land-use changes in arid and semi-arid regions have accelerated land degradation [67,68], particularly in Northwest China, where anthropogenic pressures have intensified landscape fragmentation and vegetation decline, further aggravating desertification. Moreover, evaluations of China’s ecological restoration outcomes reveal that large-scale afforestation efforts have often fallen short of expectations [49], with anthropogenic land-use activities emerging as critical determinants of vegetation dynamics in drylands [69,70]. Constrained by limited water and soil resources, vegetation growth faces significant limitations. Water availability and sustainability constitute primary considerations for restoration projects in these moisture-deficient regions [71,72]. The prevailing hot and arid climatic conditions may induce severe water stress, compromising plant survival rates [72]. Scarce surface water resources coupled with declining groundwater tables create inhospitable conditions for surface vegetation [49]. Furthermore, continued groundwater extraction to meet intensive agricultural demands exacerbates water resource conflicts, threatening the sustainability of agricultural systems. Consequently, ensuring the long-term viability of ecological restoration requires careful consideration of water and soil resource regeneration capacities maintaining land-use intensity within sustainable thresholds. Restoration represents a multifaceted challenge influenced by diverse mechanisms, necessitating context-specific implementation strategies. A comprehensive assessment of local land-use patterns serves as a fundamental prerequisite for formulating effective vegetation restoration policies [73].

4.2. Limitations

Vegetation growth is influenced not only by land-use activities but also by climatic and geographical factors such as topography [74], soil texture [75,76], CO₂ concentration, fertilization effects, and extreme weather events [77,78,79], which were not accounted for in this study. Additionally, land-use activities encompass a wide range of human behaviors, including ecological engineering, grazing, reclamation, and anthropogenic land degradation [22,54,80,81,82,83]. In this study, all human activities are unified into a concept of land-use change, but future research should further differentiate these activities. Given the complex interplay of natural and anthropogenic factors, the process of vegetation change is intricate, making it challenging to quantitatively assess the contribution of human activities to vegetation dynamics [24,84]. Future studies should refine the temporal and spatial resolution of data, increase the number of monitoring sites, and further investigate the impact of land-use changes on vegetation growth.

5. Conclusions

This study employed the Hurst index, trend analysis, and significance testing to evaluate the consistency of NDVI time series, while utilizing land-use transfer matrices and characteristic indicators to assess changes in different sandy land-use types. By establishing the relationship between land-use characteristic indicators and NDVI, the quantitative impact of land-use on NDVI was investigated. The findings reveal: (1) The vegetation dynamics in the Three-North Sandy Area exhibit high consistency, with a significant increasing trend in vegetation growth. (2) As human activities intensify land resource utilization in sandy regions, construction land continues to expand, while unused land is progressively converted into grassland. (3) In the MWSD, KBQD, HSDKS, and KEQS regions, land-use intensity positively influences vegetation growth, whereas in the SNS, increased land-use composite indices suppress vegetation development. (4) Within the Three-North Sandy Area, when the land-use intensity index (L) exceeds a threshold of 225, NDVI demonstrates an inverse trend. (5) The study demonstrates that from 1990 to 2020, climatic factors predominantly drove the NDVI growth trends in cropland, forestland, shrubland, and grassland, while human activities played a decisive role in vegetation restoration or expansion in unused land and sparsely vegetated areas.