Net Effect of Urbanization on Vegetation Dynamics in Arid Regions: Spatiotemporal Patterns and Driving Mechanisms

Junnan Gan 1,2,3 Emailganjn2023@lzu.edu.cn

Hongzhan Sun 1,2,3 Emailsunhzh2024@lzu.edu.cn

Ashraf Dewan 4 EmailA.Dewan@curtin.edu.au

Haoyan Zhang 1,2,3 Emailzhhaoyan2023@lzu.edu.cn

Xiaoyan Cao 5 Emailcaoxiaoyan19940603@163.com

Yaowen Xie 1✉,2,3,6 Emailxieyw@lzu.edu.cn

1 College of Earth and Environmental Sciences Lanzhou University Lanzhou Gansu China

2 Center for Remote Sensing of Ecological Environments in Cold and Arid Regions Lanzhou University Lanzhou China

3 Data Intelligence Laboratory of Tibetan Plateau Humanistic Environment Lanzhou University Lanzhou China

4 School of Earth and Planetary Sciences Curtin University Kent Street Bentley Perth Australia

5 School of Water Conservancy and Environment University of Jinan Jinan Shandong China

6 Center for Remote Sensing of Ecological Environments in Cold and Arid Regions Lanzhou University No. 222, South Tianshui Road Lanzhou, Lanzhou Gansu China, China

Junnan Gan^{a b c}, Hongzhan Sun ^{a b c}, Ashraf Dewan^d, Haoyan Zhang^{a b c}, Xiaoyan Cao^e, Yaowen Xie^{a b c*}

^a College of Earth and Environmental Sciences, Lanzhou University, Lanzhou, China

^b Center for Remote Sensing of Ecological Environments in Cold and Arid Regions, Lanzhou University, Lanzhou, China

^c Data Intelligence Laboratory of Tibetan Plateau Humanistic Environment, Lanzhou University, Lanzhou, China

^d School of Earth and Planetary Sciences, Curtin University, Kent Street, Bentley, Perth, Australia

^e School of Water Conservancy and Environment, University of Jinan, Jinan, Shandong, China.

All of the author's email address:

Junnan Gan: ganjn2023@lzu.edu.cn

Hongzhan Sun: sunhzh2024@lzu.edu.cn

Haoyan Zhang: zhhaoyan2023@lzu.edu.cn

Ashraf Dewan: A.Dewan@curtin.edu.au

Xiaoyan Cao: caoxiaoyan19940603@163.com

Yaowen Xie: xieyw@lzu.edu.cn

Corresponding author details

Yaowen Xie,

College of Earth and Environmental Sciences, Lanzhou University, Lanzhou, Gansu,, China

Center for Remote Sensing of Ecological Environments in Cold and Arid Regions, Lanzhou University, Lanzhou, Gansu,, China

No. 222, South Tianshui Road, Lanzhou,, China.

xieyw@lzu.edu.cn

Net Effect of Urbanization on Vegetation Dynamics in Arid Regions: Spatiotemporal Patterns and Driving Mechanisms

Abstract

Rapid urbanization significantly influences vegetation dynamics in ecologically fragile arid regions, but comprehensive quantification of its net effects remains limited. This study aims to analyze the spatiotemporal patterns and underlying driving mechanisms of urbanization’s net impact on vegetation growth, quantified as the Urban Background Difference (UBD = EVI_urban − EVI_rural), across China’s arid regions (Aridity Index < 0.5; precipitation < 500 mm/year) from 2000 to 2020.We integrated Landsat-derived Enhanced Vegetation Index (EVI), Global Artificial Impervious Area (GAIA) data, and multiple auxiliary factors including climate, socioeconomic variables, urban form, and topography. An urban–rural gradient framework was established to assess spatial variation. Driving factors were identified using an XGBoost machine learning model combined with SHapley Additive exPlanations (SHAP) for interpretability.Vegetation exhibited a consistent “core greening, peripheral browning” spatial pattern with significant heterogeneity: UBD was positive in western oasis cities (e.g., Kashi_West: 0.25), attributed to effective water management and economic growth, but negative in eastern grassland areas (e.g., Hulun_Buir: −0.16), constrained by conservation policies. UBD increased significantly between 2000 and 2010 (from 0.05 to 0.12), then diverged post-2010 with sharp declines linked to ecological zoning policies. SHAP analysis identified third-quarter precipitation and annual temperature as major negative drivers, while gross regional product, population density, built-up area proportion, and land-use diversity had positive effects. The barren land ratio accounted for 39.78% of UBD variance.Urbanization’s net impact on vegetation in arid regions is spatially heterogeneous and sensitive to policy interventions. Sustainable urban expansion in drylands requires tailored strategies that balance water resource management, economic development, and optimized urban form.

Keywords:

Urbanization

Arid urban regions

Interpretable machine learning (SHAP)

Ecological zoning policy

1. Introduction

Global urbanization is reshaping terrestrial ecosystems at an unprecedented pace. According to United Nations projections, 55% of the world's population currently resides in urban areas ( United Nations, 2019), a figure expected to rise to 68% by 2050 (Grimm et al., 2008). Rapid urbanization is characterized by the conversion of natural landscapes into built-up areas(Gong et al., 2020), leading to significant environmental and ecological challenges. These include extensive farmland loss and the deterioration of natural habitats(He et al., 2014). In arid regions, the conflict between urbanization and ecological conservation is particularly acute.

Urbanization impacts vegetation not only through direct land cover changes but also indirectly by altering multi-scale environmental factors that suppress vegetation growth(Zhuang et al., 2023). Most studies indicate a decline in vegetation growth along urban-rural gradients. This is primarily attributed to the replacement of natural vegetation with impervious surfaces, causing reduced vegetation vitality (Zhang et al., 2022; Zhong et al., 2019). Research has shown that land-use transitions in urban peripheries lead to significant vegetation and carbon loss, while older urban areas exhibit vegetation recovery, and eastern urban regions experience widespread vegetation degradation (Yang et al., 2021). However, some studies suggest that at the urban regional scale, most cities show enhanced vegetation growth and greening trends (Luo et al., 2021; S. Zhang et al., 2023).

Despite findings noted above, research on vegetation responses to urbanization in arid regions remains limited, with most studies focusing on humid or semi-humid areas. Arid ecosystems, characterized by scarce precipitation (< 500 mm annually) and high evaporation rates, are highly sensitive to water availability. For instance, urban expansion in the Hexi Corridor has caused groundwater levels to drop by 0.5 meters annually, threatening the survival of drought-tolerant vegetation like poplar forests (Feng et al., 2015). Additionally, unique land-use practices in arid regions, such as oasis agriculture and ecological migration, may lead to urbanization-vegetation dynamics distinct from those in humid regions (Hasan et al., 2020).

Although scholars worldwide have extensively studied the impacts of urbanization on vegetation from various perspectives(Chang et al., 2024; G. Li et al., 2025; Li et al., 2023; L. Li et al., 2025; You et al., 2024; Zhong, 2025), research on arid regions remains immensely limited. First, most studies focus on the whole city scale, examining vegetation growth trends across entire regions during urbanization but neglecting the complex relationships between vegetation growth and urban-rural gradients, which obscures the nuanced effects of urbanization. Second, while cities in humid and semi-humid regions dominate the research landscape, arid regions are often underrepresented, with only a few representative cities included. As a result, the unique patterns of urban expansion, vegetation growth trends, and their interactions in arid regions remain poorly understood (Yu and Yan, 2024; Z. Zhang et al., 2023). Third, traditional statistical methods, such as linear regression, fail to capture the nonlinear relationships and interactions between urbanization and vegetation, limiting the exploration of driving mechanisms(Zhang et al., 2025).

To address these gaps, this study examines the spatiotemporal patterns of urbanization and vegetation growth in the arid regions, proposing an urban-rural gradient framework to quantify the net effects of urbanization on vegetation dynamics. By employing interpretable machine learning methods, we further explore the driving mechanisms behind these effects. The study aims to address the following research questions:

What are the spatiotemporal patterns of urbanization and vegetation growth in arid regions?

How does urbanization affect vegetation growth in arid regions?

What are the driving mechanisms of Urban Background Difference (UBD) in arid regions?

Through this approach, we aim to provide a more comprehensive understanding of the complex interplay between urbanization and vegetation dynamics in arid ecosystems.

2. Materials and methods

2.1 Study area

The Aridity Index (AI) is a critical metric for drought monitoring and arid zone identification (Zomer et al., 2022), with precipitation serving as a direct indicator of drought severity. This study delineated China's arid regions using two criteria: AI < 0.5 and annual precipitation < 500 mm, ensuring the study area’s representativeness and typicality (Fig. 1). Within this region, five provinces predominantly located in arid zones were selected. Tibet was excluded due to limited urbanization–vegetation interaction in areas where elevation exceeds 50 meters above the highest urban point(Imhoff et al., 2010), as most Tibetan cities are in high-altitude mountain regions. The dominant land use types in these provinces are barren land and grasslands. To reflect fragmented urban development and avoid insufficient data from small urban areas, cities with impervious surface area over 50 km² in 2020 were selected as focal points. This ensures robust analysis of urbanization’s impact on vegetation in arid regions.

Fig. 1

Definition of the Arid Zone, land use (Yang and Huang, 2021) and geographic distribution of the study cities.

2.2 Data

2.2.1 Vegetation indices data

This study used the Enhanced Vegetation Index (EVI) to characterize vegetation growth (Huete et al., 2002, 1997). Landsat satellites provide 30m spatial and 16-day temporal resolution surface reflectance data suitable for large-scale, long-term analysis. We collected all available Landsat 5 TM (2000–2012), Landsat 7 ETM+ (2012–2013), and Landsat 8 OLI (2013–2020) images. Preprocessing on Google Earth Engine included cloud masking and radiometric correction using the QA band and CFMask algorithm to remove clouds, shadows, and snow/ice (Foga et al., 2017; Zhu et al., 2015; Zhu and Woodcock, 2012). Annual mean EVI was calculated as:

$\:\begin{array}{c}EVI=2.5\times\:\frac{NIR-RED}{NIR+6\times\:RED-7.5\times\:BLUE+1}\#\left(1\right)\end{array}$

where NIR, RED, and BLUE are Landsat reflectance bands.

2.2.2 Urbanization data

Impervious surface coverage (ISC) was derived from the 30m Global Artificial Impervious Area (GAIA) dataset (Gong et al., 2020), aggregated into 250m grids. GAIA integrates nighttime lights and Sentinel-1 SAR data, achieving > 90% accuracy in arid regions. Urban boundaries were extracted from the Global Urban Boundary (GUB) dataset (Li et al., 2020), consistent with nighttime light data to delineate urban extents.

2.2.3 Factors influencing urban background difference (UBD)

This study selected and quantified four categories of factors to explore driving forces behind UBD > They are: natural geographic factors, climatic factors, socioeconomic factors, and land use factors. The specific factors are detailed in Table 1:

Table 1

Types of factors, abbreviations, and units
Factor type	Factor	Abbreviation	Unit
Natural geographic	Mean elevation	ME	m
	Regional elevation difference	RED	m
	Mean alope	MS	degree
	Longitude	LON	decimal degrees
	Latitude	LAT	decimal degrees
Climatic	Annual mean precipitation	AMP	mm
	Q1 mean precipitation	Q1MP
	Q2 mean precipitation	Q2MP
	Q3 mean precipitation	Q3MP
	Q4 mean precipitation	Q4MP
	Annual mean temperature	AMT	degree
	Q1 mean temperature	Q1MT
	Q2 mean temperature	Q2MT
	Q3 mean temperature	Q3MT
	Q4 mean temperature	Q4MT
Socioeconomic	Total population	TP	person
	Population density	PD	person/km²
	Annual area growth rate	AAGR	km²/year
	Gross regional product	GRP	10⁴ CNY
Urban morphology	Urban compactness	UC	/
	Shannon diversity index	SDI	/
	Built-up area percentage	BAP	%
	Study area	SA	km²

2.3 Methods

2.3.1 Urban-rural gradient and buffer zone delineation

The impact of urbanization on surrounding vegetation varies in scope and intensity depending on city size. In this study, cities were categorized as small, medium, or large based on urban area. To quantify urbanization effects on vegetation, rural zones were delineated individually. A 20–25 km buffer was initially considered to represent rural areas, but due to the smaller spatial influence of cities in arid regions, buffer sizes were adjusted according to city type.

To isolate urbanization effects, the analysis excluded water bodies, croplands, and areas more than 50 meters higher than the urban core within each city’s buffer zone (Fig. 2). These exclusions helped eliminate interference from non-urban ecological factors. The urban–rural gradient was defined based on changes in impervious surface coverage, representing spatial differences in urbanization intensity. This gradient was derived using impervious surface data from 2000 and 2020. Specific classification criteria for buffer zones and urban–rural gradients are summarized in Table 2.

Table 2

Buffer and city gradient partition thresholds
Buffer zone classification			Urban-rural gradient division
Urban area (km²)	Urban scale	Buffer range (km)	Impervious surface coverage in 2000	Impervious surface coverage in 2020	Region type
0-200	Small	15	0.5-1	/	Urban core
200–400	Medium	20	0-0.5	0.5-1	Urbanized area
200–400	Medium	20	/	0.25–0.5	Suburbs
> 400	Large	25	/	0.05–0.25	Outer suburbs
> 400	Large	25	/	0-0.05	Rural (Background)

Fig. 2

The city of Urumqi is used as an example to demonstrate delineation of buffer zones and urban-rural gradients: (A) land use in Urumqi, (B) the study area after excluding water bodies and farmland, (C) delineation of urban-rural gradients in the city, excluding pixels whose elevation is more than 50m above the highest point of the urban core on the basis of (B)

2.3.2 Characterization of vegetation growth by urbanization

The net effect of urbanization on vegetation was quantified using UBD, calculated as:

$\:\begin{array}{c}UBD={EVI}_{Urban}-{EVI}_{rural}\#\left(2\right)\end{array}$

where

$\:{EVI}_{Urban}$

and

$\:{EVI}_{rural}$

denote the EVI (spatial average) of urban and rural areas, respectively, UBD > 0 indicates that urbanization promotes vegetation growth, and UBD < 0 indicates that it inhibits vegetation growth.

2.3.3 Statistical analysis methods

Temporal trends of EVI and ISC were assessed using Sen’s slope estimator and the Mann-Kendall test (Kendall, 1948; Mann, 1945). Sen’s method calculates the median of pairwise slopes to estimate the direction and magnitude of change, while Mann-Kendall tests the significance of monotonic trends without requiring normal distribution assumptions. A trend is considered significant when |Z| > 1.96 at the 95% confidence level..

2.3.4 Driving force inquiry methodology

To explore the driving mechanisms of UBD, we employed SHAP (SHapley Additive exPlanations) interpretability analysis based on the XGBoost model(Chen and Guestrin, 2016; Lundberg and Lee, 2017). XGBoost is a tree-based ensemble learning algorithm that achieves high predictive performance through optimized gradient boosting, with advantages such as nonlinearity handling, feature selection, and computational efficiency. SHAP, rooted in Shapley values from cooperative game theory, was applied to quantify the contribution of each input feature to the UBD predictions. It satisfies key mathematical properties like additivity and local accuracy, and the explanation model is expressed as:

$\:\begin{array}{c}g\left({z}^{{\prime\:}}\right)={\varnothing\:}_{0}+\sum\:_{j=1}^{M}{\varnothing\:}_{j}{Z}_{j}^{{\prime\:}}\#\left(7\right)\end{array}$

where

$\:g$

is the explanation model,

$\:M\:$

represents the number of input features,

$\:Z$

indicates the presence (1) or absence (0) of the corresponding feature, and

$\:\varnothing\:$

denotes the attribution value (Shapley value) of each feature. SHAP has become a widely adopted method for interpreting machine learning models, especially tree-based ones, and provides valuable insights into variable importance and feature interactions under varying datasets and spatiotemporal contexts.

3. Results

3.1 Urbanization process in arid areas

Based on previous studies and human activity intensity, barren lands were defined as Natural Vegetation Areas (NA), while croplands, forests, shrublands, and grasslands were classified as Semi-Natural Vegetation Areas (SA), due to their sensitivity to farming, restoration, and grazing. Urban Built-up Areas (UA) were designated as urban zones. Water bodies, wetlands, and snow/ice were excluded due to limited extent and relevance.

From 2000 to 2020, land use changes were analyzed at three levels: provincial, city, and aridity zone (Table 3). At the provincial scale, UA expanded in all provinces, especially Inner Mongolia (+ 265.4 km²) and Xinjiang (+ 190.03 km²). NA declined in most regions (e.g., Xinjiang − 915.39 km², Gansu − 410.66 km²), while SA increased, particularly in Xinjiang (+ 562.7 km²) and Gansu (+ 368.98 km²).

At the city level, UA increased across all cities, notably in Ningxia (+ 31.29 km²). NA declined in Gansu (–32.07 km²) and Ningxia (–23.03 km²), with corresponding SA gains (+ 31.25 km² and + 11.22 km², respectively).

In terms of aridity, hyper-arid and arid cities saw NA decline (–8.32 km², − 21.74 km²) and SA increase (+ 4.82 km², + 21.02 km²). In semi-arid and dry sub-humid zones, NA increased slightly (+ 0.34 km²), while SA decreased (–0.95 km²). UA expanded in all zones, with dry sub-humid cities showing the greatest growth (+ 1.98 km²).

Table 3

Land use transfer in the study area.
Region	Region	Natural Area ( $\:{km}^{2}/a$ )	Semi-Natural Area ( $\:{km}^{2}/a$ )	Urban Area ( $\:{km}^{2}/a$ )
Provincial all	Xinjiang	-915.39	562.7	190.03
	Gansu	-410.66	368.98	20.37
	Qinghai	28.18	-204.65	0.71
	Inner-Mongolia	-372.19	156.08	265.4
	Ningxia	-54.4	12.39	35.81
Provincial cities	Xinjiang cities	-11.43	11.22	0.41
	Gansu cities	-32.07	31.25	0.48
	Qinghai cities	0.34	-0.95	0.47
	Inner-Mongolia cities	-1	-6.1	7.21
	Ningxia cities	-23.03	-11.34	31.29
Urban drought	Hyper Arid	-8.32	4.82	3.47
	Arid	-21.74	21.02	0.58
	Semi-Arid	0.34	-0.95	0.47
	Dry sub-humid	0.02	-2.04	1.98

Figure 3 shows urbanization in arid region cities from 2000 to 2020. Urban growth in Northwest China varies greatly spatially. Most cities' built-up areas grew between 100% and 600%, with a few exceptional cases. Notably, Jiuquan (6921.86%) and Zhangye (2241.99%) are regional growth centers.

Low-growth cities (e.g., Alashan (66.23%), Guyuan (53.92%), Qingyang (49.43%)) are mainly in Inner Mongolia and Ningxia. Medium-low-growth cities spread wider, including Hotan_West (123.39%), Tacheng_Huyanghe (138.29%), Tacheng_Kelamayi (98.29%), and Chifeng (129.82%). Medium-high growth appears in parts of Xinjiang and Gansu, such as Kashi_East (592.08%), Bayingguoleng (630.28%), Tacheng_Changji (493.71%), and Turpan (540.88%). High-growth cities like Yili (907.59%), Wuwei (1085.37%), and Baiyin (1027.97%) exceed the regional average, reflecting their economic importance.

Ultra-high growth in Jiuquan and Zhangye likely results from policy, resources, or infrastructure. Xinjiang cities generally show higher growth, especially Kashi, Hotan, Akesu, and Turpan. In Gansu, Jiuquan and Zhangye are growth poles, while parts of Inner Mongolia and Ningxia have lower growth, showing clear regional differences..

Fig. 3

Process of urbanization

3.2 Spatial pattern of vegetation greenness in urban areas

10.

3.2.1 Trends in vegetation growth in arid cities

From 2000 to 2020, vegetation greening in arid cities showed clear spatial heterogeneity Fig. 4. Browning areas were less than 10% of each city’s study area, mostly at urban edges. In 11 cities, over 50% of areas showed no significant change, with Hulun_Buir highest at 81.42%. Only four cities had greening trends under 30%, while background regions of all cities generally showed greening. Urban cores mostly exhibited significant greening.

In short, vegetation greening in arid cities features strong greening in urban cores, clear greening in backgrounds, and browning at urban edges, illustrating complex spatial heterogeneity during urbanization.

Fig. 4

Trend of vegetation growth in each city of the study area

3.2.2 Vegetation greenness along urban-rural gradient

Figure 5 shows vegetation greenness along the urban-rural gradient from 2000–2010 and 2011–2020, reflecting urbanization’s impact (UBD). Vegetation forms an inverted U-shaped pattern: lower greenness in urban cores and background areas, and higher greenness in suburban/urbanized zones, revealing complex urban–vegetation interactions.

Background areas have low greenness due to arid climate—low rainfall, high evapotranspiration, and dominant land types (wasteland, grassland) limit growth. Urban cores also show low greenness, mainly from impervious surfaces, which reduce soil moisture and hinder vegetation. In contrast, suburban and urbanized areas show higher greenness, likely due to compensatory greening (e.g., artificial vegetation, restoration) and land-use optimization (e.g., green space planning).

Data from both periods show aligned mean and median curves, suggesting near-normal distribution and regular UBD effects. Statistical analysis can help clarify how urbanization affects vegetation, supporting green space planning and ecological resilience.

Fig. 5

Distribution of EVI values along urban-rural gradient in the study area, (a) 2000–2010 mean by region, (b) 2011–2020 mean by region

3.3 Spatial and temporal patterns of changing in UBD

3.3.1 Spatial pattern of UBD

Figure 6 shows spatiotemporal trends of Urbanization-induced Vegetation Dynamics (UBD) along east-west and north-south gradients. The heatmap (a) reveals clear spatial heterogeneity: UBD decreases from west to east, with Kashi_West (0.25) and Hotan_West (0.26) peaking in 2000. Hami maintained values above 0.13 (2000–2016), indicating limited urban impact on agriculture. In contrast, eastern cities like Hulun_Buir and Xilingol showed persistent negative UBD (min − 0.16 in 2018), reflecting grassland conservation. Cities along the Yellow River (e.g., Lanzhou, Yinchuan) had stable UBD within ± 0.05.

Temporally, UBD increased from 0.05 to 0.12 during 2000–2010, then diverged. After 2010, western cities (e.g., Hotan_East) retained high values (up to 0.21 in 2015), while eastern cities (e.g., Chifeng) turned negative after 2012.

Panel (b) shows turning points clustered around 2012. Akesu_West and Zhangye dropped by over 0.1 that year. Abrupt UBD shifts also appeared in Bayingguoleng (2012) and Wulumuqi_Changji (2017), indicating effects of regional policies and ecological projects.

Fig. 6

Heat map of spatial and temporal variation patterns of UBD in cities in the arid zone. The vertical axis of the heat map is the study cities in the arid zone, which are arranged from top to bottom in the order of west to east. (a) The distribution of the UBD values of each city in the corresponding year, and (b) Trend of the UBD change over time in each city, with the red color indicating that the trend of the UBD value of that city in the corresponding year has undergone a mutation, and the right hand side is the trend of the UBD change of each city, with ↑ representing that the city's UBD shows an increasing trend from 2000 to 2020, ↓ is the opposite, and = is that the city's UBD value has not changed significantly over the period.

Between 2000 and 2020, urbanization’s impact on vegetation (UBDC) varied significantly across regions (Fig. 7). Positive UBDC values were mainly found in oasis and resource-based cities, such as Tacheng_Kelamayi (0.067), Jinchang (0.015), and Haixi_Mongol (0.010), where urbanization enhanced vegetation growth. In contrast, negative UBDC values were concentrated in the Loess Plateau and Inner Mongolia grasslands, e.g., Qingyang (–0.21), Linxia (–0.26), and Xilingol (–0.08), suggesting negative or limited effects, possibly moderated by grazing bans.

Cities with higher wasteland ratios saw stronger UBDC effects. For example, Tacheng_Kelamayi (77.71%) and Jinchang (55.18%) improved vegetation via development or restoration. Meanwhile, cities with low wasteland ratios, like Lanzhou (1.72%) and Huhehaote (0.06%), had lower UBDC values (–0.058, − 0.177), possibly due to urban land-use change or conservation restrictions.

Overall, wasteland ratio explains 39.78% of UBDC variation (R² = 0.3978), highlighting its role, though other factors remain due to ecosystem complexity.

Fig. 7

Changes in UBD values and its spatial distribution in the study area

3.3.2 Exploring UBD impact factor based on explainable machine learning

Changes in urban vegetation greenness are influenced by a variety of factors. This study selected four categories of factors: as noted in Table 1.

Prior to inputting the factors into the model to investigate driving forces, a correlation matrix was multicolinearity. Although significant positive or negative correlations exist among some climate variables(Fig. 8), most explanatory variables exhibit low correlations across different categories of influencing factors. Nevertheless, considering that different variables may contribute uniquely to predictive capability and that the model adopted in this study can mitigate overfitting issues caused by multicollinearity to some extent, all variables were retained for subsequent model training and SHAP value-based analysis.

Fig. 8

Correlation matrix of the factors used in this work

To quantify the contribution of each feature to the model's predictions, this study employed the SHAP (SHapley Additive exPlanations) method to calculate and visualize feature importance. First, Bayesian Optimization was utilized to fine-tune the hyperparameters of the XGBoost model, ensuring optimal performance on both the training and testing datasets. Subsequently, the trained model was used to predict the test set samples, and the SHAP framework was applied to compute the contribution values (SHAP values) of each feature to the predictions. SHAP values provide a global interpretation of feature importance: the SHAP Summary Plot illustrates the distributional impact of each feature on the model's output, offering an intuitive representation of the relationship between feature values and SHAP values. Additionally, the feature importance Bar Plot displays the average absolute contribution of each feature to the model's output, thereby quantifying the global importance of each feature.

Fig. 9

Results of SHAP analysis. The left panel shows the Shapley values for each feature ranked by importance. Absolute value of the mean SHAP value of each feature is on the right, characterizing relative importance of each feature.

Based on SHAP analysis Fig. 9, this study assessed the effects of climate, socio-economic, urban form, and topographic factors on UBD. Climate variables like third-quarter precipitation (–0.000311) and annual mean temperature (–0.000920) negatively impacted UBD, suggesting limited water-use efficiency and the suppression of vegetation by urban heat islands. Socio-economic factors showed positive effects: gross regional product (+ 0.000077) and total population (+ 0.000469) indicated that economic development and population density promoted urban greening, especially in arid regions. Urban form indicators, including built-up area proportion (+ 0.000592) and land use diversity (SDI, + 0.000912), contributed positively by enhancing greening compensation and ecological resilience. In terms of topography, longitude (+ 0.000767) supported vegetation performance in eastern cities, while mean elevation (–0.000078) had a weak negative influence, reflecting its uncertain role in balancing urban expansion and rural vegetation protection.

4. Discussion

4.1 Common characteristics and spatial heterogeneity of urbanization and vegetation growth in arid areas

From 2000 to 2020, urbanization and vegetation growth in Northwest China's arid regions generally showed a pattern of "core greening and peripheral browning", consistent with earlier findings(Berdejo-Espinola et al., 2021; De la Sota et al., 2019; Rigolon and Németh, 2020; Z. Zhang et al., 2023). Urban cores improved due to greening compensation and restoration, while peripheries declined due to construction, water shortages, and land-use changes. Spatial heterogeneity emerged: western oasis cities (e.g., Kashi_West, Hotan_West) gained from water and agriculture, while eastern grassland cities (e.g., Hulun_Buir, Xilingol) suffered under ecological fragility and conservation constraints. Vegetation gradients also differed—more gradual in the west, more abrupt in the east—highlighting the complex interplay of environmental conditions and policy interventions.

4.2 Patterns of spatial and temporal changes in UBD: evolution from regional differences to policy implications

UBD displayed clear east–west contrasts(Chen and Shen, 2021; Zhao et al., 2024) : it was higher in western cities due to intensive agricultural practices and water inputs, and persistently negative in eastern regions dominated by semi-natural vegetation (Piao et al., 2015).. Temporally, UBD rose between 2000–2010 (e.g., from 0.05 to 0.12), suggesting positive urban effects. Post-2010, UBD declined significantly in some cities (e.g., Δ > 0.1 in Akesu_West, Zhangye), reflecting zoning controls and ecological redlining. After 2016, UBD values stabilized, suggesting a maturing balance between urbanization and regulation. These trends reflect how urban expansion, ecological capacity, and governance together shape vegetation dynamics(Bush et al., 2023; Yan et al., 2022; Zhao et al., 2013).

4.3 UBD driving mechanisms: multifactorial synergies and regional heterogeneity

The driving mechanisms of urbanization on vegetation dynamics in arid regions reflect a multidimensional synergy of climatic constraints, economic forces, urban morphology, and policy interventions, with dominant factors varying by region. Climatic variables such as third-quarter precipitation (Q3MP, SHAP = − 0.000311) and annual mean temperature (AMT, SHAP = − 0.000920) exerted notable negative effects on UBD, likely due to enhanced evapotranspiration from precipitation runoff and intensified urban heat island effects (Paudel and States, 2023; Vujovic et al., 2021). For instance, Hami maintained high UBD values (> 0.13) due to irrigation systems, whereas Huhehaote saw declines post-2012, coinciding with urban expansion and heat stress.

Socioeconomic drivers including gross regional product (GRP, SHAP = + 0.000077) and total population (TP, SHAP = + 0.000469) positively influenced UBD. Cities like Tacheng_Kelamayi (UBDC = 0.067) leveraged economic growth for ecological restoration in areas with > 77% wasteland. High population density also fostered urban greening, as seen in Wuwei, where a 35% population increase between 2000–2010 corresponded to a 0.19 rise in UBD(Chen and Jim, 2008; Haase et al., 2014; Jim and Chen, 2009; Kabisch et al., 2016).

Urban morphological variables—built-up area proportion (BAP, SHAP = + 0.000592) and Shannon Diversity Index (SDI, SHAP = + 0.000912)—also enhanced UBD. In Zhangye, policies mandating green space (e.g., 15% ratio) offset urban expansion impacts post-2012. Land-use diversity further supported ecological connectivity (Fahrig, 2003), evident in Turpan’s integration of viticulture with green infrastructure, maintaining UBD above 0.11.

Natural topography and policy effects interacted via longitude (LON, SHAP = + 0.000767) and elevation (ME, SHAP = − 0.000078). Eastern cities (e.g., Yinchuan) with humid climates exhibited stable UBD (± 0.05), while western cities relying on irrigation were more sensitive to policy shifts, particularly after the 2012 ecological zoning reforms. High-altitude cities (e.g., Qingyang) faced constrained urban growth and policy-protected rural vegetation (UBDC = − 0.21), though urbanization’s negative effects persisted. In grassland cities (e.g., Hulun_Buir), grazing bans (UBD = − 0.16) partially mitigated urbanization’s suppressive impact.

Overall, a "resource–economy–policy" triadic model explains regional variation in UBD. Western oases (e.g., Kashi_West) benefited from high GRP and water allocation, sustaining UBD peaks (0.25); eastern grasslands (e.g., Hulun_Buir) were constrained by conservation policies and climatic stress (UBD = − 0.16); transitional cities (e.g., Zhangye) experienced policy-induced UBD fluctuations (Δ > 0.1). Wasteland proportion explained 39.78% of UBD variance (R² = 0.3978), with residuals likely shaped by nonlinear factors (e.g., groundwater changes, project lag), requiring further validation. Policy recommendations include targeted water governance, dynamic compensation schemes, and morphology optimization based on regional geoclimatic conditions.

5. Conclusions

This study quantified the net effect of urbanization on vegetation growth (UBD) in arid regions, revealing the complex relationship between urbanization and vegetation dynamics and its driving mechanisms. The research finds that the impact of urbanization on vegetation in arid regions exhibits significant spatiotemporal heterogeneity, characterized by the common pattern of "greening in the core and browning on the periphery," with notable differences between the eastern and western regions. In the western oasis cities (e.g., Kashi_West), UBD values remain high and stable due to water resource allocation and economic drivers, while in the eastern grassland cities (e.g., Hulun_Buir), UBD values are consistently negative due to climatic constraints and policy restrictions. The analysis of driving mechanisms indicates that climatic factors (e.g., Q3MP, AMT) negatively influence UBD, while socio-economic factors (e.g., GRP, TP) and urban morphological factors (e.g., BAP, SDI) positively drive vegetation growth through economic strength and structural optimization. Policy interventions (e.g., ecological function zoning) significantly influenced the phased evolution of UBD after 2010, highlighting the critical role of policy regulation.

However, this study has some limitations: First, explanatory power of wasteland proportion for UBD is only R²=0.3978, suggesting that other factors (e.g., groundwater level changes, timeliness of ecological projects) may significantly impact vegetation dynamics. Second, data resolution limits in-depth analysis of vegetation changes at micro-scales (e.g., neighborhood level). Future research could integrate high-resolution remotely sensed data and social surveys to further explore the interactions between urbanization, climate change, and ecological policies, providing a more comprehensive scientific basis for the coordinated development of ecology and urbanization in arid regions.

Funding

This work was supported by the Gansu Provincial Natural Science Foundation (Grant No. 22JR5RA425).

Data Availability

The data that support the findings of this study are available on request from the corresponding author, Yaowen Xie, upon reasonable request.

Author Contribution

JG contributed to the study conception and design. Material preparation, data collection, and code development were performed by JG and HZ. Visualization was carried out by JG and HS. The first draft of the manuscript was prepared by JG. The manuscript was reviewed and edited by AD, YX, and XC. Supervision, project administration, and funding acquisition were carried out by YX. All authors reviewed and approved the final manuscript.

Reference

Berdejo-Espinola, V., Suárez‐Castro, A.F., Amano, T., Fielding, K.S., Oh, R.R.Y., Fuller, R.A., 2021. Urban green space use during a time of stress: A case study during the COVID‐19 pandemic in Brisbane, Australia. People Nat. 3, 597–609. https://doi.org/10.1002/pan3.10218

Bush, J., Oke, C., Dickey, A., Humphrey, J., Harrison, L., Amati, M., Fornari, G., Soanes, K., Callow, D., Van Der Ree, R., 2023. A decade of nature: Evolving approaches to Melbourne’s ‘nature in the city.’ Landsc. Urban Plan. 235, 104754. https://doi.org/10.1016/j.landurbplan.2023.104754

Chang, Y., van Strien, M.J., Zohner, C.M., Ghazoul, J., Kleinschroth, F., 2024. Effects of climate, socioeconomic development, and greening governance on enhanced greenness under urban densification. Resour. Conserv. Recycl. 206, 107624. https://doi.org/10.1016/j.resconrec.2024.107624

Chen, L., Shen, W., 2021. Spatiotemporal differentiation of urban-rural income disparity and its driving force in the Yangtze River Economic Belt during 2000–2017. PLOS ONE 16, e0245961. https://doi.org/10.1371/journal.pone.0245961

Chen, T., Guestrin, C., 2016. XGBoost: A Scalable Tree Boosting System, in: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Presented at the KDD ’16: The 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ACM, San Francisco California USA, pp. 785–794. https://doi.org/10.1145/2939672.2939785

Chen, W.Y., Jim, C.Y., 2008. Cost–benefit analysis of the leisure value of urban greening in the new Chinese city of Zhuhai. Cities 25, 298–309. https://doi.org/10.1016/j.cities.2008.06.002

De la Sota, C., Ruffato-Ferreira, V.J., Ruiz-García, L., Alvarez, S., 2019. Urban green infrastructure as a strategy of climate change mitigation. A case study in northern Spain. Urban For. Urban Green., Urban green infrastructure – connecting people and nature for sustainable cities 40, 145–151. https://doi.org/10.1016/j.ufug.2018.09.004

Fahrig, L., 2003. Effects of Habitat Fragmentation on Biodiversity. Annu. Rev. Ecol. Evol. Syst. 34, 487–515. https://doi.org/10.1146/annurev.ecolsys.34.011802.132419

Feng, Q., Ma, H., Jiang, X., Wang, X., Cao, S., 2015. What Has Caused Desertification in China? Sci. Rep. 5, 15998. https://doi.org/10.1038/srep15998

Foga, S., Scaramuzza, P.L., Guo, S., Zhu, Z., Dilley, R.D., Beckmann, T., Schmidt, G.L., Dwyer, J.L., Joseph Hughes, M., Laue, B., 2017. Cloud detection algorithm comparison and validation for operational Landsat data products. Remote Sens. Environ. 194, 379–390. https://doi.org/10.1016/j.rse.2017.03.026

Gong, P., Li, X., Wang, J., Bai, Y., Chen, B., Hu, T., Liu, X., Xu, B., Yang, J., Zhang, W., Zhou, Y., 2020. Annual maps of global artificial impervious area (GAIA) between 1985 and 2018. Remote Sens. Environ. 236, 111510. https://doi.org/10.1016/j.rse.2019.111510

Grimm, N.B., Faeth, S.H., Golubiewski, N.E., Redman, C.L., Wu, J., Bai, X., Briggs, J.M., 2008. Global Change and the Ecology of Cities. Science 319, 756–760. https://doi.org/10.1126/science.1150195

Haase, D., Larondelle, N., Andersson, E., Artmann, M., Borgström, S., Breuste, J., Gomez-Baggethun, E., Gren, Å., Hamstead, Z., Hansen, R., Kabisch, N., Kremer, P., Langemeyer, J., Rall, E.L., McPhearson, T., Pauleit, S., Qureshi, S., Schwarz, N., Voigt, A., Wurster, D., Elmqvist, T., 2014. A Quantitative Review of Urban Ecosystem Service Assessments: Concepts, Models, and Implementation. AMBIO 43, 413–433. https://doi.org/10.1007/s13280-014-0504-0

Hasan, S.S., Zhen, L., Miah, Md.G., Ahamed, T., Samie, A., 2020. Impact of land use change on ecosystem services: A review. Environ. Dev., Resources Use, Ecosystem Restoration and Green Development 34, 100527. https://doi.org/10.1016/j.envdev.2020.100527

He, C., Liu, Z., Tian, J., Ma, Q., 2014. Urban expansion dynamics and natural habitat loss in China: a multiscale landscape perspective. Glob. Change Biol. 20, 2886–2902. https://doi.org/10.1111/gcb.12553

Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ., The Moderate Resolution Imaging Spectroradiometer (MODIS): a new generation of Land Surface Monitoring 83, 195–213. https://doi.org/10.1016/S0034-4257(02)00096-2

Huete, A.R., Liu, H.Q., Batchily, K., van Leeuwen, W., 1997. A comparison of vegetation indices over a global set of TM images for EOS-MODIS. Remote Sens. Environ. 59, 440–451. https://doi.org/10.1016/S0034-4257(96)00112-5

Imhoff, M.L., Zhang, P., Wolfe, R.E., Bounoua, L., 2010. Remote sensing of the urban heat island effect across biomes in the continental USA. Remote Sens. Environ. 114, 504–513. https://doi.org/10.1016/j.rse.2009.10.008

Jim, C.Y., Chen, W.Y., 2009. Ecosystem services and valuation of urban forests in China. Cities 26, 187–194. https://doi.org/10.1016/j.cities.2009.03.003

Kabisch, N., Strohbach, M., Haase, D., Kronenberg, J., 2016. Urban green space availability in European cities. Ecol. Indic. 70, 586–596. https://doi.org/10.1016/j.ecolind.2016.02.029

Kendall, M.G., 1948. Rank correlation methods., Rank correlation methods. Griffin, Oxford, England.

Li, G., Cao, Y., Fang, C., Sun, S., Qi, W., Wang, Z., He, S., Yang, Z., 2025. Global urban greening and its implication for urban heat mitigation. Proc. Natl. Acad. Sci. 122, e2417179122. https://doi.org/10.1073/pnas.2417179122

Li, L., Li, G., Cao, Y., 2025. Long-term trends in urban vegetation greenness and its causal relationship with built-up land expansion: A case study of the Yangtze River Delta, China. Habitat Int. 156, 103286. https://doi.org/10.1016/j.habitatint.2025.103286

Li, L., Zhan, W., Ju, W., Peñuelas, J., Zhu, Z., Peng, S., Zhu, X., Liu, Z., Zhou, Y., Li, J., Lai, J., Huang, F., Yin, G., Fu, Y., Li, M., Yu, C., 2023. Competition between biogeochemical drivers and land-cover changes determines urban greening or browning. Remote Sens. Environ. 287, 113481. https://doi.org/10.1016/j.rse.2023.113481

Li, Xuecao, Gong, P., Zhou, Y., Wang, J., Bai, Y., Chen, B., Hu, T., Xiao, Y., Xu, B., Yang, J., Liu, X., Cai, W., Huang, H., Wu, T., Wang, X., Lin, P., Li, Xun, Chen, J., He, C., Li, Xia, Yu, L., Clinton, N., Zhu, Z., 2020. Mapping global urban boundaries from the global artificial impervious area (GAIA) data. Environ. Res. Lett. 15, 094044. https://doi.org/10.1088/1748-9326/ab9be3

Lundberg, S.M., Lee, S.-I., 2017. A Unified Approach to Interpreting Model Predictions, in: Advances in Neural Information Processing Systems. Curran Associates, Inc.

Luo, Y., Sun, W., Yang, K., Zhao, L., 2021. China urbanization process induced vegetation degradation and improvement in recent 20 years. Cities 114, 103207. https://doi.org/10.1016/j.cities.2021.103207

Mann, H.B., 1945. Nonparametric Tests Against Trend. Econometrica 13, 245–259. https://doi.org/10.2307/1907187

Paudel, S., States, S.L., 2023. Urban green spaces and sustainability: Exploring the ecosystem services and disservices of grassy lawns versus floral meadows. Urban For. Urban Green. 84, 127932. https://doi.org/10.1016/j.ufug.2023.127932

Piao, S., Yin, G., Tan, J., Cheng, L., Huang, M., Li, Y., Liu, R., Mao, J., Myneni, R.B., Peng, S., Poulter, B., Shi, X., Xiao, Z., Zeng, N., Zeng, Z., Wang, Y., 2015. Detection and attribution of vegetation greening trend in China over the last 30 years. Glob. Change Biol. 21, 1601–1609. https://doi.org/10.1111/gcb.12795

Rigolon, A., Németh, J., 2020. Green gentrification or ‘just green enough’: Do park location, size and function affect whether a place gentrifies or not? Urban Stud. 57, 402–420. https://doi.org/10.1177/0042098019849380

Vujovic, S., Haddad, B., Karaky, H., Sebaibi, N., Boutouil, M., 2021. Urban Heat Island: Causes, Consequences, and Mitigation Measures with Emphasis on Reflective and Permeable Pavements. CivilEng 2, 459–484. https://doi.org/10.3390/civileng2020026

Yan, Y., Jaung, W., Richards, D.R., Carrasco, L.R., 2022. Where did the ecosystem services value go? Adaptive supply, demand and valuation of new urban green spaces. Resour. Conserv. Recycl. 187, 106616. https://doi.org/10.1016/j.resconrec.2022.106616

Yang, J., Huang, X., 2021. The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019. Earth Syst. Sci. Data 13, 3907–3925. https://doi.org/10.5194/essd-13-3907-2021

Yang, K., Sun, W., Luo, Y., Zhao, L., 2021. Impact of urban expansion on vegetation: The case of China (2000–2018). J. Environ. Manage. 291, 112598. https://doi.org/10.1016/j.jenvman.2021.112598

You, Y.-J., Liu, M., Ni, H.-G., 2024. Dynamic indirect impacts of urbanization on vegetation growth in Chinese cities. Int. J. Appl. Earth Obs. Geoinformation 131, 103974. https://doi.org/10.1016/j.jag.2024.103974

Yu, F.-J., Yan, L., 2024. Trends in Urban Vegetation Growth in China from 2000 to 2022. Land 13, 1015. https://doi.org/10.3390/land13071015

Zhang, L., Yang, L., Zohner, C.M., Crowther, T.W., Li, M., Shen, F., Guo, M., Qin, J., Yao, L., Zhou, C., 2022. Direct and indirect impacts of urbanization on vegetation growth across the world’s cities. Sci. Adv. https://doi.org/10.1126/sciadv.abo0095

Zhang, S., Jia, W., Zhu, H., You, Y., Zhao, C., Gu, X., Liu, M., 2023. Vegetation growth enhancement modulated by urban development status. Sci. Total Environ. 883, 163626. https://doi.org/10.1016/j.scitotenv.2023.163626

Zhang, S., Zhu, H., Zeng, K., Zhang, Y., Jin, Z., Wang, Y., Zhang, R., Jürgen, B., Liu, M., 2025. From city to countryside: Unraveling the long-term complex effects of urbanization on vegetation growth in China. J. Environ. Manage. 380, 124975. https://doi.org/10.1016/j.jenvman.2025.124975

Zhang, Z., Zhao, W., Liu, Y., Pereira, P., 2023. Impacts of urbanisation on vegetation dynamics in Chinese cities. Environ. Impact Assess. Rev. 103, 107227. https://doi.org/10.1016/j.eiar.2023.107227

Zhao, J., Chen, S., Jiang, B., Ren, Y., Wang, H., Vause, J., Yu, H., 2013. Temporal trend of green space coverage in China and its relationship with urbanization over the last two decades. Sci. Total Environ. 442, 455–465. https://doi.org/10.1016/j.scitotenv.2012.10.014

Zhao, Z., Li, X., Liu, Y., Yang, L., Liu, M., 2024. Exploring Spatial Differentiation in Cultural Urban Land Use in China’s Ancient and Emerging Cities: A Comparative Case Study of Xi’an and Shenzhen. Sustainability 16, 9704. https://doi.org/10.3390/su16229704

Zhong, J., 2025. Unveiling the spatio-temporal patterns of vegetation growth influenced by diverse urban intensity gradients.

Zhong, Q., Ma, J., Zhao, B., Wang, X., Zong, J., Xiao, X., 2019. Assessing spatial-temporal dynamics of urban expansion, vegetation greenness and photosynthesis in megacity Shanghai, China during 2000–2016. Remote Sens. Environ. 233, 111374. https://doi.org/10.1016/j.rse.2019.111374

Zhu, Z., Wang, S., Woodcock, C.E., 2015. Improvement and expansion of the Fmask algorithm: cloud, cloud shadow, and snow detection for Landsats 4–7, 8, and Sentinel 2 images. Remote Sens. Environ. 159, 269–277. https://doi.org/10.1016/j.rse.2014.12.014

Zhu, Z., Woodcock, C.E., 2012. Object-based cloud and cloud shadow detection in Landsat imagery. Remote Sens. Environ. 118, 83–94. https://doi.org/10.1016/j.rse.2011.10.028

Zhuang, Q., Shao, Z., Li, D., Huang, X., Altan, O., Wu, S., Li, Y., 2023. Isolating the direct and indirect impacts of urbanization on vegetation carbon sequestration capacity in a large oasis city: evidence from Urumqi, China. Geo-Spat. Inf. Sci. 26, 379–391. https://doi.org/10.1080/10095020.2022.2118624

Zomer, R.J., Xu, J., Trabucco, A., 2022. Version 3 of the Global Aridity Index and Potential Evapotranspiration Database. Sci. Data 9, 409. https://doi.org/10.1038/s41597-022-01493-1

Yes