The Driving Mechanism and Spatio-Temporal Nonstationarity of Oasis Urban Green Landscape Pattern Changes in Urumqi

Abstract

The green landscapes of oasis cities play an important role in maintaining ecological security. However, these ecosystems face increasing threats from desertification and fragmentation, driven by intensifying climate change and rapid urbanization. Understanding the characteristics and driving mechanisms behind changes in green landscape patterns is crucial for advancing sustainable urban green space management. This study explores the spatio-temporal changes in the green landscape pattern in Urumqi during 1990–2020 using a random forest classifier. This study also applies geographical detectors and geographically weighted regression to comprehensively determine the driving mechanism and spatio-temporal nonstationarity. The results are as follows: (1) The landscape types are primarily dominated by unused land, urban green spaces, and construction land, accounting for more than 80%. The areas of urban green spaces, water bodies, cropland, and unused land decreased by 0.38%, 37.41%, 0.57%, and 4.58%, respectively, from 1990 to 2020. With rapid urbanization, construction land exhibited a significant expansion trend, and the degree of fragmentation of urban green spaces increased spatially over these 30 years. (2) From 1990 to 2020, each landscape index exhibited fluctuating characteristics. Overall, the Shannon’s diversity and evenness indices of the urban green landscapes exhibited an increasing trend. The contagion and connectivity indices exhibited a decreasing trend, decreasing from 50.894 and 99.311 in 1990 to 46.584 and 99.048 in 2020, respectively. (3) During these 30 years, the dynamics of urban greenery were affected by a combination of natural and social factors, with elevation determining the overall urban green distribution pattern. Precipitation and temperature dominate the urban green space changes in the north and south of Urumqi. Socioeconomic factors such as GDP, population, river distance, and town distance regulate the urban green space changes in the central built-up area.

1. Introduction

Urban green landscape patterns refer to the spatio-temporal distribution states formed through the geographical differentiation of and anthropogenic influences on urban green spaces, as expressed through their compositional, morphological, structural, and scalar attributes. On the one hand, they are manifested in the quantity, spatial configuration, and changes of urban green spaces. On the other hand, they are interrelated with the “process–service–function” of urban green spaces, indirectly affecting residents’ well-being [1]. However, rapid urbanization and extreme climate change have led to the scarcity of urban greenery, fragmentation of landscape patterns, and supply–demand imbalance, hindering the efficient exertion of ecosystem services. Therefore, exploring the changes in urban green landscape patterns and their driving mechanisms is the basis for understanding the cascade effect of the “composition–pattern–process–service–function” of the urban green system, which can provide a basis for the sustainable management of the system [2].

The temporal and spatial dynamics of urban green landscape patterns and their driving mechanisms have been a research hotspot. Changes in urban green landscape patterns are influenced by both natural and anthropogenic factors [3], with anthropogenic factors typically playing a dominant role. Rapid urbanization, rapid urban population growth, and urban planning have typically resulted in the conversion of arable land, grassland, and unused land into construction land to meet expansion demands. Land use changes contributed to the scarcity and fragmentation of urban green spaces [4,5]. As socioeconomic urbanization advances, increased investments in funding and technologies have improved living environments across urban and rural areas. Influenced by various green space planning policies such as “Green Space Construction Planning,” “Ecological Garden Cities,” and “Ecological Park Cities,” the construction of small green spaces has gradually increased the green coverage rate [6,7]. From a natural perspective, the key influencing factors include elevation, slope, precipitation, evaporation, road distance, and land type [8]. Influenced by topography and slope, urban suburbs tend to have higher vegetation cover, and the green landscape pattern shows gradient characteristics along the “urban center–urban–rural intersection–suburbs” [9]. Affected by factors such as precipitation and evaporation, the urban green pattern also exhibits heterogeneity. With the decrease in precipitation and the increase in surface temperature and evaporation, the normalized difference vegetation index (NDVI) exhibited a significant downward trend. Urban green landscape patterns are affected by climate, historical evolution, and human geography. Notably, oasis urban green spaces in arid regions experience particularly severe water stress compared with coastal developed cities, resulting in lower vegetation survival rates and higher fragmentation levels [10]. The spatio-temporal changes and driving mechanisms of landscape patterns tend to exhibit special model types under these constrained conditions.

Many scholars have conducted systematic research on the evolution of urban green landscape patterns and driving mechanisms [11,12,13], proposing various research methods, including correlation analysis, ecological footprint models, gray correlation models, system dynamics models, and geographically weighted regression (GWR) models [3,14,15,16]. Conventional statistical methods such as correlation and regression analyses have been used to explore the driving mechanisms of landscape changes. However, they typically face challenges in terms of the complexity, aggregation, discontinuity, and linear/nonlinear relationships of urban green landscape spatial changes [17,18,19]. Furthermore, current research fails to sufficiently address the spatial heterogeneity of these driving mechanisms [20], with most studies focusing on European, North American, and eastern Chinese coastal hotspots while neglecting northwestern China’s extreme arid zones. As the western arid areas face multiple constraints such as extreme climate, water scarcity, low plant survival rates, and poverty alleviation, the implementation of economic development and ecological protection policies, such as “Urbanization,” the “Western Development Strategy,” the “Silk Road Economic Belt,” “Three North Protective Forests,” and the “10,000 mu Corridor Greening Project,” has established unique driving mechanisms for urban green landscape changes in northwestern China’s arid regions. The adaptability of the results of urban green landscape evolution characteristics and landscape planning concepts from the coastal areas of China to oasis cities in arid areas is worth further discussion. What is the law of the urban green landscape’s spatial–temporal change in the oasis cities? What are the spatial–temporal dynamics change mechanisms of the urban green landscape? What are the similarities and differences between urban green landscape change patterns and driving mechanisms in arid and coastal areas? There is still a lack of comprehensive data and scientific support for this series of questions, and in-depth research is urgently required.

This study investigates urban green landscape changes in Urumqi, a typical oasis city, from 1990 to 2020 using a random forest classifier and multiple analytical techniques, including spatial analysis, landscape index, geographic detector, and GWR. This study aims to (1) reveal the quantitative changes in and spatial distribution of urban green landscapes from 1990 to 2020; (2) select landscape pattern metrics at the type and landscape levels to reveal the spatio-temporal variation characteristics of urban green landscape metrics from 1990 to 2020; and (3) construct a multidimensional driving force index system incorporating “natural–economic–social” factors while accounting for urban planning and green space policies to assess the contribution and spatio-temporal nonstationarity of driving factors for green landscape changes. This study explores the evolution mechanism of landscape patterns and their background policies in oasis cities, providing theoretical and practical references for sustainable landscape planning in arid areas.

2. Study Area

Urumqi is located in the extremely arid area of northwestern China (86°37′–88°58′ E, 42°45′–44°08′ N), serving as the center of the Asian continent and an important node city in the core area of the Silk Road Economic Belt (Figure 1). The total area of the study area is about 1.42 × 10⁴ km², with seven districts and one county, including Tianshan District, Shuimogou District, Xinshi District, Shayibake District, Toutunhe District, Midong District, Dabancheng District, and Urumqi County. Urumqi is located in the northern foothills of the Tianshan Mountains, at the southern edge of the Gurbantunggut Desert, with an average altitude of 800 m. From north to south, it presents a gradient pattern of “desert-farmland-city-mountains”. The annual average precipitation is 280 mm, the annual average evaporation is 2830 mm, and the annual average temperature is 6.4 °C, meaning it has a mid-temperate arid continental climate. Due to its geographical particularity, water resource limitations, and topographical constraints, coupled with the impacts of the “Western Development Strategy” and “Rapid Urbanization”, problems such as urban green space fragmentation, land desertification, and soil salinization have become increasingly prominent. Under the background of ecological civilization city construction, Urumqi not only faces new development opportunities, but is also confronted by numerous ecological and environmental challenges.

3. Data Resources and Method

3.1. Data Resources

According to the phenological period and ecological status of arid areas, remote sensing image data (12 scenes, 30 m resolution) spanning from 1990 to 2020 were obtained from the U.S. Geological Survey (https://www.usgs.gov/, accessed on 10 July 2021) and Geospatial Data Cloud (http://www.gscloud.cn/search, accessed on 10 July 2021). These included data from Landsat 4–5 TM and Landsat 8 OLI-TIRS sensors. The imaging time was from July to September, and the cloud cover was generally lower than 10%. Considering Urumqi’s green policy and social and economic development plan, 1990, 2000, 2010, and 2020 were selected as representative years. In addition, the research data included natural factor data such as DEM, slope, aspect, precipitation, evaporation, temperature, and rivers and socioeconomic factor data such as population, GDP, and roads. Precipitation, evaporation, and temperature were derived from the long-term meteorological data (1990–2019) of Urumqi and surrounding counties collected by the National Meteorological Science Data Center (http://data.cma.cn/, https://www.aqistudy.cn/, accessed on 27 August 2025). The population, GDP, etc., were sourced from the Resource and Environment Science and Data Center of the Chinese Academy of Sciences (https://www.resdc.cn/, accessed on 27 August 2025). The population and GDP data of the Environmental Resource Science Data Center of the Chinese Academy of Sciences integrate land use, night-time light index, and settlement density data on a large scale. Therefore, to improve accuracy, this study normalized the existing raster data and corrected them using the actual population and GDP data of seven districts and one county from the statistical yearbooks. Combined with the 30 m digital elevation data, the slope and aspect were extracted via terrain analysis in ArcGIS 10.1. Vector standard map data such as administrative boundaries and roads in the study area were obtained from the Xinjiang Geospatial Information Public Service Platform (http://xinjiang.tianditu.gov.cn/, accessed on 27 August 2025). All types of data were resampled to a 30 m resolution.

Given the mutual transformation relationship between the evolution of urban green landscapes in arid areas and construction land, cropland, unused land, etc., this study investigates the spatio-temporal evolution of urban green landscape patterns due to policies such as urbanization, agriculturalization, and the return of cropland to forest and grassland. Referring to the classification methods of previous studies [12], this study defines the urban green landscape as land covered by vegetation such as forests and grasslands and divides the landscape into forestland, grassland, construction land, cropland, water bodies, and unused land. Among these categories, forestland and grassland are classified as urban green landscapes. The study area is a typical desert-oasis mosaic landscape with complex and diverse landscape types. Based on Ge’s research results on oasis cities, the accuracy of random forests is as high as 96.92% [21]. The random forest method was selected for the urban green landscape interpretation (Figure 2). Radiometric calibration and atmospheric correction were performed separately for each landscape remote sensing image using ENVI 5.3. Based on the “Random Forest Classification Tool” in ENVI 5.3, the regions of interest for each landscape type were constructed by comparing the color, texture, and spectral information of each image. To objectively assess the classification accuracy, 100 ground observation points and 1010 validation points of 91 satellite maps (with a spatial resolution of 0.27 m) were collected for each landscape type. The overall accuracy and kappa coefficient were selected to compare the training and validation data. The results show that the OA is above 85% and the kappa coefficient is much higher than 0.70. The specific calculation formulas are as follows:

(1)$\mathrm{O}\mathrm{A}={\displaystyle \frac{1}{\mathrm{N}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{x}}_{\mathrm{i}\mathrm{i}}$

(2)$\mathrm{K}\mathrm{a}\mathrm{p}\mathrm{p}\mathrm{a}={\displaystyle \frac{\mathrm{N}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{x}}_{\mathrm{i}\mathrm{i}}-{\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{x}}_{\mathrm{i}+}\times {\mathrm{x}}_{+\mathrm{i}})}{{\mathrm{N}}^2-{\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{x}}_{\mathrm{i}+}\times {\mathrm{x}}_{+\mathrm{i}})}}$

In the formula, OA is the overall accuracy; N is the total number of observation points; n is the matrix row and column number; ${\mathrm{x}}_{\mathrm{i}\mathrm{i}}$ is the total number of observation points in the i row and i column of the confusion matrix; ${\mathrm{x}}_{\mathrm{i}+}$ and ${\mathrm{x}}_{+\mathrm{i}}$ are the total number of observation points in the i row or i column.

Figure 2

Remote sensing image processing and interpretation flow chart.

[Figure omitted. See PDF]

3.2. Landscape Pattern Metrics

Landscape pattern metrics emphasize the exploration of the quantity, composition, structure, and spatial distribution characteristics of landscape elements at both the landscape and patch levels [22,23,24]. This study aimed to uncover the composition, configuration, and dynamic evolution patterns of urban green landscapes by integrating landscape pattern metrics. The landscape pattern spatio-temporal changes were analyzed at both the patch and landscape levels to determine the quantity, edge shape characteristics, adjacency relationships, agglomeration characteristics, and diversity of the urban green landscapes. Adhering to the principles of scale appropriateness, comprehensiveness, representativeness, typicality, and system integrity, the class- and landscape-level metrics were analyzed separately to identify the temporal trends of each metric (Table 1). The transformation of urban green spaces involves both transfer in and out. Transfer in refers to the conversion of other types of land, such as construction land, unused land, and water bodies, into urban forests and grasslands. Transfer out refers to the conversion of urban forests and grasslands into other types such as construction land, unused land, and water bodies.

3.3. GeoDetector

Geodetector is a spatial statistical method for detecting spatial heterogeneity of geographic objects and driving factors [25]. In this study, factor detection in Geodetector was selected for screening the relative importance of main factors and expressing spatial heterogeneity. Taking the area of urban green conversion (inflow/outflow) as the dependent variable, 10 independent variables were constructed from two aspects, natural factors (elevation, slope, precipitation, evaporation, temperature, distance from rivers) and socioeconomic factors (population, GDP, distance from roads, distance from towns), to reveal the dominant factors of urban green landscape pattern changes and their interaction relationships. Geodetector measures the explanatory power of independent variables on dependent variables by constructing a “q” statistic, and normalizes the “q” statistic to [0, 1]. The larger the “q”-value, the higher the contribution of the independent variable to the dependent variable and the stronger the explanatory power. The calculation formula is as follows:

(3)$\mathrm{q}=1-{\displaystyle \frac{1}{\mathrm{N}{\sigma }^2}}\sum _{\mathrm{h}=1}^{\mathrm{L}}{\mathrm{N}}_{\mathrm{h}}{\sigma }_{\mathrm{h}}^2$

In the formula, h = 1, 2, …; L represents the classification levels of independent/dependent variables; ${\mathrm{N}}_{\mathrm{h}}$ and $\mathrm{N}$ denote the numbers of grid cells in each classification level and the entire region respectively; ${\sigma }_{\mathrm{h}}^2$ and ${\sigma }^2$ are the variances of each classification level and the entire region, respectively.

3.4. Geographically Weighted Regression

The geographically weighted regression (GWR) model is based on the principle of Ordinary Least Squares (OLS), introducing geographical location and spatial weights into the model to explore the spatial heterogeneity of the influence of independent variables on dependent variables [26,27]. To reveal the influence of spatial nonstationarity of dominant factors on the evolution of urban green landscape patterns, this study used ArcGIS to construct a 1 km × 1 km window size, and uses the dominant factors screened by the Geodetector as independent variables, takes the change in the area of urban green inflow/outflow in each grid as the dependent variable, and combines GWR to quantitatively describe the spatial variation characteristics of the interaction relationship, so as to reveal the spatial heterogeneity of the influence of driving factors. Among them, determining the function weight and bandwidth is particularly critical. This study uses the fixed kernel function and the AIC criterion bandwidth method to fit the explanatory degree of independent variables to dependent variables. When the AIC reaches the minimum value, the corresponding bandwidth is the optimal bandwidth. The GWR operation results are evaluated and analyzed using the Coefficient of Determination (R²) and Root Mean Square Error (RMSE). The specific principles and steps of GWR are as follows:

(4)$\mathrm{GWR}\mathrm{model}:{\mathrm{Y}}_{\mathrm{i}}={\sigma }_0\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)+{\sum }_{\mathrm{j}}^{\mathrm{n}}{\sigma }_{\mathrm{j}}({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}){\mathrm{x}}_{\mathrm{i}\mathrm{j}}+{\epsilon }_{\mathrm{i}}$

(5)$\mathrm{Fixed}\mathrm{kernel}\mathrm{function}:{\mathrm{W}}_{\mathrm{i}\mathrm{j}}=\left\{\begin{array}{c}1,,,{\mathrm{x}}_{\mathrm{i}\mathrm{j}}\le \delta \\ 1,,,{\mathrm{x}}_{\mathrm{i}\mathrm{j}}\ge \delta \end{array}\right.$

(6)$\mathrm{AIC}\mathrm{criterion}\mathrm{bandwidth}\mathrm{method}:\mathrm{A}\mathrm{I}\mathrm{C}=2\mathrm{n}\mathrm{l}\mathrm{n}\left({\displaystyle \frac{\mathrm{R}\mathrm{S}\mathrm{S}}{\mathrm{n}-\mathrm{t}\mathrm{r}\left(\mathrm{S}\right)}}\right)+\mathrm{n}\mathrm{l}\mathrm{n}\left(2\pi \right)+\mathrm{n}\left. [{\displaystyle \frac{\mathrm{n}+\mathrm{t}\mathrm{r}\left(\mathrm{S}\right)}{\mathrm{n}-2-\mathrm{t}\mathrm{r}\left(\mathrm{S}\right)}}\right]$

In the formula, ${\mathrm{Y}}_{\mathrm{i}}$ is the weighted regression coefficient at point i; $\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ represents the geographic coordinates at point i; ${\sigma }_{\mathrm{j}}({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}})$ denotes the regression parameter of the j-th variable at point i; ${\mathrm{x}}_{\mathrm{i}\mathrm{j}}$ is the value of the j-th driving factor variable at point i; ${\epsilon }_{\mathrm{i}}$ is the random deviation at point i; D is the model parameter; $\mathrm{t}\mathrm{r}\left(\mathrm{S}\right)$ is the function of bandwidth; and $\mathrm{R}\mathrm{S}\mathrm{S}/\mathrm{n}-\mathrm{t}\mathrm{r}\left(\mathrm{S}\right)$ is the maximum likelihood estimate of the random variance.

4. Results

4.1. Quantity and Spatial Distribution Characteristics of Landscape Patterns

The landscape types in Urumqi are dominated by unused land, urban green spaces, and construction land, accounting for more than 80% of the total area. Cropland and construction land follow, and water bodies occupy the smallest area, only 1.43–3.14% (Figure 3). In terms of temporal changes, different times exhibit different change characteristics. From 1990 to 2000, the areas of urban greenery, cropland, water bodies, and construction land increased by 98.24, 29.83, 49.11, and 78.93 km², respectively, whereas only the area of unused land exhibited a decreasing trend. During 2000–2010, construction land and unused land continued to grow, increasing by 130.09 and 1889.82 km², respectively, whereas water bodies, cropland, and urban green spaces exhibited decreasing characteristics. From 2010 to 2020, urban green spaces, cropland, water bodies, and construction land showed an increasing trend, with construction land and urban green spaces increasing most significantly, by 74.31% and 55.11%, respectively. Ecological land dominated by urban green spaces exhibited a slight recovery trend, whereas the unused area decreased by 1975.11 km². Overall, with rapid urbanization, the areas of urban greenery, water bodies, cropland, and unused land exhibited a decreasing trend from 1990 to 2020, decreasing by 0.38%, 37.41%, 0.57%, and 4.58%, respectively.

From the spatial distribution perspective of the landscape type (Figure 4), an urban green landscape dominated by forests and grasslands and water bodies is concentrated along the Tianshan Mountains in high-altitude mountainous areas. The urban greenery in the urban central area also exhibits a fragmented distribution. Unused land covers a wide range and is extensively distributed within the study area. Cropland is concentrated in the basins of the Toutun River and Urumqi River, and other river basins. Construction land is concentrated in the central area of Urumqi, forming a “T”-shaped distribution. From 1990 to 2020, construction land exhibited a significant expansion trend, focusing on northward expansion. The urban green area in the central urban area has also gradually increased, especially on Yamalik Mountain, Hetan Corridor, and Spider Mountain. Along with the northward expansion of construction land, cropland in northern Urumqi was increasingly reduced from 1990 to 2020, and the degree of spatial fragmentation was obvious.

4.2. Temporal and Spatial Dynamic Changes in Landscape Pattern Metrics

Changes in landscape pattern metrics at various levels can reveal the characteristics of the number of landscapes, edge shape, aggregation and dispersion, and adjacency relationships. The results demonstrate that each landscape index presented a fluctuating change feature from 1990 to 2020 (Figure 5). The urban green landscape exhibited a “rise–fall–rise” trend in terms of all landscape metrics. The values of the landscape metrics varied among the different landscape types, with unused land and urban green spaces having the largest patch density and index. The edge density presented the characteristic of “unused land > green space > construction land > cropland > water bodies.” Unused land exhibited the largest average proximity, followed by urban green spaces. Regarding the landscape shape index, unused land (36.36–39.81) and urban green spaces (46.62–49.46) maintained high values, indicating that the shapes of unused land and urban green spaces were diverse, with a relatively high degree of fragmentation and complexity.

The richness, Shannon’s diversity, and Shannon’s evenness indices reflect the richness of the landscape types. As shown in Table 2, the Shannon’s diversity and evenness indices showed an overall increasing trend from 1990 to 2020, rising from 1.211 and 0.745 in 1990 to 1.298 and 0.779 in 2020, where the richness index remained unchanged. This indicates that landscape diversity and evenness have increased, with the landscape gradually exhibiting greater diversity and heterogeneity. The contagion and connectivity indices decreased, dropping from 50.894 and 99.311 in 1990 to 46.584 and 99.048 in 2020, respectively. This suggests that the connectivity and aggregation of landscape types have declined, and the landscape exhibits a trend of fragmentation under the dual influences of natural and human activities.

4.3. Driving Mechanism and Spatio-Temporal Nonstationarity of Urban Green Landscape Pattern Changes

4.3.1. Temporal Variation Characteristics of Driving Forces for Urban Green Landscape Patterns

Regarding the overall conversion of urban green spaces from 1990 to 2020, evaporation and temperature were the main driving factors from 1990 to 2000, particularly the evaporation factor, with a q-value as high as 0.082, whereas socioeconomic factors, dominated by population and GDP, were relatively weak (Figure 6). Compared with the trends during 1990–2000, elevation, distance from roads, and distance from towns gradually played a driving role during 2000–2010, with q-values all greater than 0.100. From 2010 to 2020, socioeconomic factors dominated by population (0.109 **) and GDP (0.083 **) gradually became dominant, with overall q-values higher than those of natural factors, indicating that with the rapid growth in population and economy, the demand for construction land and cropland increased, gradually encroaching on urban green spaces and unused land. Overall, the conversion of urban green spaces over the past 30 years has been comprehensively influenced by natural and social factors, with natural factors as the main driving force from 1990 to 2000 and a combination of natural and socioeconomic factors as the main driving force from 2000 to 2020.

Regarding the overall urban green transfer in from 1990 to 2020, elevation, road distance, temperature, and precipitation were the main driving forces from 1990 to 2000, whereas socioeconomic factors dominated by population and GDP exhibited insignificant driving effects (p > 0.05) (Figure 7). During 2000–2010, natural factors such as elevation, temperature, and precipitation drove urban green transfer in conversion, and GDP (0.073 **) also gradually played a driving role. From 2010 to 2020, the driving factors exhibited a characteristic of “elevation > precipitation > temperature greater than road distance > urban distance > slope > evaporation” in terms of explanatory power. In particular, the q-values of natural driving factors such as elevation, precipitation, and temperature were higher than 0.200 (p < 0.01). Urban green transfer in was affected by the comprehensive influence of natural and social factors over these 30 years, among which natural factors were the main driving force from 1990 to 2000, and natural factors and socioeconomic factors collectively drove the conversion from 2000 to 2020.

4.3.2. Spatial Nonstationarity of Driving Forces for Urban Green Landscape Patterns

To further reveal the spatial nonstationarity of the driving forces of urban green space changes, we conducted principal component analysis based on correlation analysis. A collinearity diagnosis was performed for each principal component factor. After removing the collinear factors, seven main factors were obtained. Finally, a GWR model was used to analyze the spatial nonstationarity of the urban green space change driving forces. The correlation analysis results (Figure 8) demonstrated that from 1990 to 2020, both the urban green transfer in and out exhibited significant relationships with various driving factors to different degrees. Specifically, urban green transfer out/in was significantly correlated with elevation, slope, precipitation, temperature, evaporation, distance to roads, distance to rivers, and distance to towns at a level of 0.001 and with GDP and population at a level of 0.050.

A GWR model was used to analyze the driving factors and spatial nonstationarity of urban greenery. The results (Figure 9) demonstrated that the R² average value of the entire region was 0.588. The R² values exhibited a decreasing trend from south to north. The high-value areas were primarily located in the southwestern region, whereas the low-value regions were primarily concentrated in the margin of the Gurbantünggüt Desert in the north. Among all driving factors, precipitation and evaporation had the greatest impact on urban green transfer out, with the main impact areas distributed in the south, primarily in the Nanshan Mountain region and the northern part of the Tianshan Mountains. Among the driving factors, elevation and GDP were the main factors inhibiting urban greenery. The inhibitory effect of GDP exhibited a divergent pattern from the central urban area to the surrounding areas, whereas the inhibitory effect of elevation exhibited gradient differences along the north–south altitude. Driving factors such as population, distance to rivers, and distance to roads exhibited similar characteristics of spatial heterogeneity, generally presenting a pattern of high values in the south and low values in the north. According to the comparison of the maximum values of the regression coefficients of all driving forces, precipitation and temperature dominated urban green transfer out in the northern and southern parts of the study area, and GDP, population, and distance to towns also played important driving roles in the central part. The results revealed the spatial heterogeneity of the strength of the driving forces for urban greenery.

For urban green transfer in (Figure 10), the adjusted R² value of the model was 0.535. The R² value exhibited a decreasing trend from south to north. The high-value areas were primarily located in the southwest of the study area, the medium-value areas were in central Urumqi, concentrated in Tianshan District, Shayibake District, Xinshi District, and Shuimogou District, and the low-value areas were primarily concentrated in the southern margin of the Gurbantünggüt Desert, adjacent to Fukang City and Wujiaqu City. Similar to urban green transfer out, precipitation and evaporation had the greatest impact on urban green transfer in, and the main impact areas were distributed in the south and central parts, primarily in the northern part of the Tianshan Mountains and the central urban area. Among the driving factors, elevation and GDP were the main factors inhibiting urban green transfer in. The inhibitory effect of GDP was primarily distributed in the northern desert and southern mountainous areas, whereas that of elevation was concentrated in the southeastern high-altitude areas. The driving factors of population, distance to rivers, and distance to roads generally exhibit the characteristic of being “high in the south and low in the north” in terms of their role in urban green transfer in, primarily because high-altitude areas have abundant precipitation, sparse population, and low urban development intensity. Under the condition of low-intensity human disturbance, urban green transfer in is more susceptible to the influence of altitude, precipitation, and evaporation. Overall, the role of each driving factor in urban green transfer in exhibits spatial nonstationarity characteristics. Elevation determines the overall distribution pattern of urban greenery. Precipitation and temperature dominate urban green transfer in in the north and south, and socioeconomic factors, primarily GDP, population, distance to rivers, and distance to towns, regulate urban green transfer in in the central urban area.

5. Discussion

5.1. Landscape Pattern Changes and Driving Mechanisms

This study explores the spatio-temporal evolution characteristics and driving mechanisms of urban green landscapes in Urumqi from 1990 to 2020. According to the landscape pattern evolution results, (1) urban green spaces exhibited a “rise–fall–rise” trend from 1990 to 2020, with an increase of 98.24 km² from 1990 to 2000; along with the rapid implementation of the “Western Development Strategy” in 2000, construction land gradually expanded and encroached on urban green spaces. Coupled with grassland degradation on the northern slope of the Tianshan Mountains, the area of urban green spaces gradually decreased from 2000 to 2010. From 2010 to 2020, with the introduction of the “Implementation Opinions on Urumqi’s Creation of Garden City,” “Returning Cropland to Forest/Grassland,” “Barren Mountain Greening Projects,” and other measures were gradually implemented in the central urban area and suburbs, such as the “Yamalik Mountain Reconstruction Project,” “Nanshan Greening Project,” and “Hetan Thousand-Mu Green Corridor Project” [6]. Consequently, the urban green spaces in suburban areas such as Nanshan and Yamalik Mountain were gradually recovered, and the urban green area increased by 55.11%, which is consistent with the research results of Zhao et al. [28]. Overall, the ecological space of the green areas exhibited a continuous shrinking trend, with a faster degradation rate from 2000 to 2010. Subsequently, due to the implementation of policies such as urban green protection policies, ecological function zoning, and ecological protection red lines, the transfer-out rate gradually slowed down from 2010 to 2020, exhibiting similarity to the research results of Zhang et al. [29]. From a spatial distribution perspective, Urumqi presents a complete ecosystem unit of “mountain-water-forest-grass-urban-desert” from south to north, which differs from the “concentric circle” structure of cities such as Beijing. Coupled with the influence of altitude gradients, the spatial differentiation of landscape types in Urumqi is more obvious. (2) The results of the driving mechanisms of urban green landscape pattern changes in Urumqi demonstrate that urban green landscape changes are comprehensively influenced by natural, social, economic, and policy factors. The natural factors are dominated by elevation, slope, precipitation, and evaporation, whereas the socioeconomic factors primarily include GDP, population, distance to roads, and distance to towns. Huang found that the driving forces behind forest and grassland landscape changes in Urumqi are primarily slope and distance to water systems and settlements, reflecting that urban green space changes are jointly driven by natural and socioeconomic factors [30]. The current study further revealed through interaction detection that each factor exhibits different interaction relationships in different urban green transfer out/in modes. Overall, natural factors dominated by elevation, slope, precipitation, and evaporation and socioeconomic factors dominated by GDP, population, road distance, and urban distance exhibited a linear enhancement effect. In comparison, the research results of Zhao et al. revealed that the dual-factor enhancement is most obvious when NDVI and elevation act together [28]. In addition, the current study further combined the urban development of Urumqi and the urban green protection policy to qualitatively analyze the potential impact of the policy. From 1990 to 2000, the policy was primarily focused on the natural dynamic evolution of urban green spaces. From 2000 to 2010, influenced by the “Western Development Strategy,” “Urumqi Urban Master Plan,” and “Xinjiang Uygur Autonomous Region Action Plan for Promoting New Urbanization,” the population and construction land in the central urban area surged, gradually encroaching on the urban greenery, and the degradation rate of the urban green landscape increased. From 2010 to 2020, with the introduction of policies such as the “Urumqi Barren Mountain Greening Project,” “Urumqi Ecological Function Zoning,” and “Three Lines and One List,” the green space changes in Urumqi were affected by both urbanization and urban green protection, and the degradation rate of urban green spaces gradually slowed down. Note that although previous studies have revealed the impacts of natural, social, economic, and policy factors on regional landscape patterns, there is a lack of research on the spatial nonstationarity of the strength of each driving factor. This study further found using the GWR model that the elevation of Urumqi determines the overall distribution pattern of urban green spaces, regulating the differences in the quantity and spatial allocation of urban green spaces. Precipitation and temperature dominate the transfer in/out of urban green spaces in the north and south, and socioeconomic factors, primarily GDP, population, river distance, and urban distance, regulate the transfer in/out of urban green spaces in the central urban area. Compared with other oasis cities in arid areas, such as Turpan City, Liu found that the basic factors driving landscape changes in Turpan are natural factors such as the arid climate background, topography, and water resource distribution, and the decisive factors are socioeconomic driving factors such as socioeconomic levels and policies [31]. Against the extreme arid background, the driving mechanisms among the oasis cities exhibit a certain degree of similarity.

5.2. Contributions and Limitations

This study combined multisource data categories such as topography, climate, social economy, and statistical yearbooks, analyzed the spatio-temporal evolution characteristics of urban green spaces from 1990 to 2020, and identified the driving mechanisms and spatio-temporal nonstationarity of urban green space changes from multiple dimensions, including natural, social, economic, and policy factors. Although many studies have been conducted on urban green landscapes and ecosystem services, existing research primarily relies on mathematical statistical analysis methods, such as the quantity and indices of urban green landscape patterns, to evaluate the evolution of single spatio-temporal landscape patterns. There is a lack of research on the spatio-temporal correlation patterns of urban green landscapes, the spatial display of landscape patterns, the multidimensional driving mechanisms of “natural–social–economic-policy,” and spatial nonstationarity. By focusing on oasis cities with extreme aridity, water scarcity, high desertification, and scarce green spaces, this study adopted multisource data types and revealed the spatial dependence and scale effects of the driving factors on the evolution of the urban green landscape, thereby enhancing the understanding of the complexity, discontinuity, and nonlinear relationships in the spatial changes in the urban green landscapes in arid oasis cities. However, this study also has certain limitations in terms of data accuracy and method optimization. On the one hand, this study uses Landsat 30 m remote sensing images. Due to data resolution limitations, urban green space was only classified into forestland and grassland, making it impossible to conduct a more refined classification of urban green spaces, and the degree of refinement of the research results was relatively low. On the other hand, due to the lack of field observation stations for climate and meteorology in arid areas, the results of the spatial interpolation of data such as precipitation and evaporation are prone to accuracy errors compared with the actual situation. Therefore, in future studies, it is imperative to use high-resolution remote sensing images, increase ground-based measured data, perform refined classification of urban green spaces in oasis cities according to park, protective, road, and affiliated green spaces, and reveal the spatio-temporal change laws among different green space types to provide refined guidance for the scientific and reasonable planning and layout and structural design of green spaces.

6. Conclusions

This research integrates multi-source data including remote sensing, climate, topography, and socioeconomic data from 1990 to 2020, and uses the random forest method for remote sensing interpretation and verification of urban green landscapes. It clarifies the spatio-temporal change characteristics of urban green landscapes and the dynamic evolution characteristics of landscape metrics from 1990 to 2020, and quantitatively analyzes the spatio-temporal changes and driving mechanisms of urban green landscape patterns by integrating natural, social, economic, and policy factors. This study found that the landscape types in Urumqi are complex and diverse, covering a complete ecosystem unit of “mountain–water-forest-field-lake-grass-desert”. From 1990 to 2020, the area of urban green spaces showed a “rise-fall-rise” trend, with an overall decrease of 0.38%. The urban green landscape metrics also presented the “rise-fall-rise” trend, featuring a large number of patches, a high landscape shape index, a relatively high fragmentation degree, and degradation characteristics. The dynamic changes in urban green spaces over the past 30 years have been comprehensively influenced by natural and social factors. Elevation determines the overall distribution pattern of urban green spaces, precipitation and temperature dominate the conversion in/out of urban green spaces in the northern and southern parts of Urumqi, and socioeconomic factors, mainly including GDP, population, river distance, and urban distance, regulate the conversion in/out of urban green spaces in the central built-up area. The research results can provide a reference for landscape pattern optimization and ecological risk management and control in Urumqi and related oasis cities.

Author Contributions

Conceptualization, L.S. and Ü.H.; methodology, L.S. and X.Z.; software, L.S.; validation, L.S.; formal analysis, X.Z.; investigation, L.S.; resources, L.S.; data curation, L.S. and X.Z.; writing—original draft preparation, L.S. and Ü.H.; writing—review and editing, X.Z. and L.S.; visualization, L.S. and X.Z.; supervision, L.S.; project administration, L.S.; funding acquisition, L.S. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The data supporting the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgments

The authors thank the anonymous referees for comments that improved the content as well as the presentation of this paper.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1 Location of study area. (a) The location of Urumqi in China. (b) Distribution of the central urban area in the study area. (c) Distribution of landscape types in the study area.

Figure 3 Area changes in landscape types from 1990 to 2020.

Figure 4 Spatial distribution of landscape types in Urumqi.

Figure 5 Change in landscape metrics at category level.

Figure 6 Factor detection results of driving force factors of urban green transfer out from 1990 to 2020. Note: * and ** represents significant at the 0.05 and 0.01 level, respectively.

Figure 7 Factor detection results of driving force factors of urban green transfer in from 1990 to 2020. Note: * and ** represents significant at the 0.05 and 0.01 level, respectively.

Figure 8 Correlation between urban green transfer out (a) and transfer in (b) and driving factors from 1990 to 2020. DEM denotes elevation; P denotes precipitation; Tem denotes temperature; T denotes evaporation; Slope denotes slope; Urban, Road, and River denote distance from towns, roads, and rivers, respectively; POP and GDP denote population and gross domestic product (GDP), respectively; and GZ and ZG denote greenfield transfer out and in, respectively.Note: *, ** and *** represents significant at the 0.05, 0.01 and 0.001 level, respectively.

Figure 9 Driving forces and spatial instability of urban green transfer out from 1990 to 2020.

Figure 10 Driving forces and spatial instability of urban green transfer in from 1990 to 2020.

Table 1

Selection and significance of landscape metrics.

Level	Landscape Metrics	Connotation of Metrics
Classlevel	Patch number (NP)	The number of patches of a certain landscape type.
	Patch density (PD)	The number of patches per unit area of a certain landscape.
	Edge density (ED)	Describes the boundary relationship between a certain landscape and its adjacent elements.
	Largest patch index (LPI)	The proportion of the largest patch of a certain type to the landscape area.
	Landscape shape index (LSI)	The deviation degree of the patch shape from a circle of the same area.
	Average proximity index (API)	Characterizes the degree of proximity and fragmentation within a certain landscape.
Landscapelevel	Connectivity index (COHESION)	Explores the degree of connection between patches in the landscape.
	Richness index (PI)	Characterizes the richness among landscapes.
	Contagion index (CONTAG)	Reflects the agglomeration characteristics and spread trends among landscapes.
	Shannon diversity index (SHDI)	Reveals the spatio-temporal heterogeneity characteristics of the landscape.
	Shannon equitability index (SHEI)	Describes the uniformity of the spatio-temporal distribution of the landscape.

Table 2

Change in landscape metrics at landscape level.

Years	CONTAG	COHESION	PI	SHDI	SHEI
1990	50.894	99.311	6.000	1.211	0.745
2000	47.525	99.249	6.000	1.274	0.771
2010	54.796	99.436	6.000	1.079	0.635
2020	46.584	99.048	6.000	1.298	0.779

References

1. Wang, Z.; Yu, Y.; Zhou, R. A longitudinal exploration of the spatiotemporal coupling relationship and driving factors between regional urban development and ecological quality of green space. Ecol. Indic.; 2024; 164, 112134. [DOI: https://dx.doi.org/10.1016/j.ecolind.2024.112134]

2. Zhang, L.L.; Meng, Q.Y.; Yao, S.; Qian, J.K.; Gao, J.F.; Wu, J.H. How to optimize urban blue space to maximize its cooling benefits? A case study in megacity of Beijing. Build. Environ.; 2025; 270, 112502. [DOI: https://dx.doi.org/10.1016/j.buildenv.2024.112502]

3. Jiang, S.M.; Feng, F.; Zhang, X.N.; Xu, C.Y.; Jia, B.Q.; Lafortezza, R. Driving factors of fragmentation in urban landscapes: Local contributions, spatial relationships, and causal effects. Ecol. Indic.; 2025; 174, 113454. [DOI: https://dx.doi.org/10.1016/j.ecolind.2025.113454]

4. Tian, Y.H.; Jim, C.Y.; Wang, H.Q. Assessing the landscape and ecological quality of urban green spaces in a compact city. Landsc. Urban Plan.; 2014; 141, pp. 97-108. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2013.10.001]

5. Bian, H.Y.; Gao, J.; Wu, J.G.; Sun, X.; Du, Y. Hierarchical analysis of landscape urbanization and its impacts on regional sustainability: A case study of the Yangtze River Economic Belt of China. J. Clean. Prod.; 2020; 279, 123267. [DOI: https://dx.doi.org/10.1016/j.jclepro.2020.123267]

6. Shi, L.; Halik, Ü.; Abliz, A.; Mamat, Z.; Welp, M. Urban Green Space Accessibility and Distribution Equity in an Arid Oasis City: Urumqi, China. Forests; 2020; 11, 690. [DOI: https://dx.doi.org/10.3390/f11060690]

7. Liu, M.; Li, J.; Song, D.; Dong, J.; Ren, D.; Wei, X. Spatiotemporal Dynamics Effects of Green Space and Socioeconomic Factors on Urban Agglomeration in Central Yunnan. Forests; 2024; 15, 1598. [DOI: https://dx.doi.org/10.3390/f15091598]

8. Liu, C.J.; Zhang, F.; Johnson, V.C.; Duan, P.; Kung, H. Spatio-temporal variation of oasis landscape pattern in arid area: Human or natural driving. Ecol. Indic.; 2021; 125, 107495. [DOI: https://dx.doi.org/10.1016/j.ecolind.2021.107495]

9. Kong, F.H.; Nakagoshi, N. Spatial-temporal gradient analysis of urban green spaces in Jinan, China. Landsc. Urban Plan.; 2006; 3, pp. 147-164. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2005.07.006]

10. Gonzalez-Mendez, B.; Chavez-Garcia, E. Re-thinking the Technosol design for greenery systems: Challenges for the provision of ecosystem services in semiarid and arid cities. J. Arid. Environ.; 2020; 179, 104191. [DOI: https://dx.doi.org/10.1016/j.jaridenv.2020.104191]

11. Pan, Y.J.; Teng, T.W.; Wang, S.P.; Wang, T.T. Impact and mechanism of urbanization on urban green development in the Yangtze River Economic Belt. Ecol. Indic.; 2024; 158, 111612. [DOI: https://dx.doi.org/10.1016/j.ecolind.2024.111612]

12. Yuan, Y.Y.; Tang, S.Q.; Zhang, J.Q.; Guo, W. Quantifying the relationship between urban blue-green landscape spatial pattern and carbon sequestration: A case study of Nanjing’s central city. Ecol. Indic.; 2023; 154, 110483. [DOI: https://dx.doi.org/10.1016/j.ecolind.2023.110483]

13. Liu, X.Y.; Li, Y.F.; Zhang, S.; Niu, Q. Spatiotemporal patterns, driving mechanism, and multi-scenario simulation of urban expansion in Min Delta Region, China. Ecol. Indic.; 2024; 158, 111312. [DOI: https://dx.doi.org/10.1016/j.ecolind.2023.111312]

14. Duan, Z.; Huang, L.; Zhu, Z.; Long, S.; Liu, Y. Transformation and Inequity of Urban Green Space in Guangzhou: Drivers and Policy Implications Under Rapid Urbanization. Sustainability; 2025; 17, 2217. [DOI: https://dx.doi.org/10.3390/su17052217]

15. Zhu, Y.; Ling, G.H.T. Spatio-Temporal Changes and Driving Forces Analysis of Urban Open Spaces in Shanghai between 1980 and 2020: An Integrated Geospatial Approach. Remote Sens.; 2024; 16, 1184. [DOI: https://dx.doi.org/10.3390/rs16071184]

16. He, X.; Zhou, Y.Q. Urban spatial growth and driving mechanisms under different urban morphologies: An empirical analysis of 287 Chinese cities. Landsc. Urban Plan.; 2024; 248, 105096. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2024.105096]

17. Yang, H.; Jin, C.; Li, T. Megacity urban green space equity evaluation and its driving factors from supply and demand perspective: A case study of Tianjin. J. Clean. Prod.; 2024; 474, 143583. [DOI: https://dx.doi.org/10.1016/j.jclepro.2024.143583]

18. Li, X.Q.; Xiao, L.M. The impact of urban green business environment on FDI quality and its driving mechanism: Evidence from China. World Dev.; 2024; 175, 106494. [DOI: https://dx.doi.org/10.1016/j.worlddev.2023.106494]

19. Gao, C.; Dang, Q.; Li, C.; Fan, Y. Analysis of Landscape Fragmentation Evolution Characteristics and Driving Factors in the Wei River Basin, China. Land; 2025; 14, 538. [DOI: https://dx.doi.org/10.3390/land14030538]

20. Wang, H.; Lin, C.; Ou, S.; Feng, Q.; Guo, K.; Wei, X.; Xie, J. Multilevel Change of Urban Green Space and Spatiotemporal Heterogeneity Analysis of Driving Factors. Sustainability; 2024; 16, 4762. [DOI: https://dx.doi.org/10.3390/su16114762]

21. Ge, G.B.; Shi, Z.J.; Zhu, Y.J.; Yang, X.H.; Hao, Y.G. Land use/cover classification in an arid desert-oasis mosaic landscape of China using remote sensed imagery: Performance assessment of four machine learning algorithms. Glob. Ecol. Conserv.; 2020; 22, e00971. [DOI: https://dx.doi.org/10.1016/j.gecco.2020.e00971]

22. McGarigal, K.; Marks, B.J. FRAGSTATS: Spatial Pattern Analysis Program for Quantifying Landscape Structure (General Technical Report PNW-GTR-351); USDA Forest Service, Pacific Northwest Research Station: Corvallis, OR, USA, 1995; 122.

23. Dai, E.; Wu, Z.; Du, X. A gradient analysis on urban sprawl and urban landscape pattern between 1985 and 2000 in the Pearl River Delta, China. Front. Earth Sci.; 2018; 12, pp. 791-807. [DOI: https://dx.doi.org/10.1007/s11707-017-0637-0]

24. Chen, Y.; Zhen, W.; Li, Y.; Zhang, N.; Shi, Y.; Shi, D. Spatiotemporal Analysis of Urban Blue Space in Beijing and the Identification of Multifactor Driving Mechanisms Using Remote Sensing. Remote Sens.; 2023; 15, 5182. [DOI: https://dx.doi.org/10.3390/rs15215182]

25. Wang, J.F.; Xu, C.D. Geodetector: Principle and prospective. Acta Geo Sin.; 2017; 72, pp. 116-134. (In Chinese)

26. Fotheringham, A.S.; Charlton, M.E.; Brunsdon, C. Geographically weighted regression: A natural evolution of the expansion method for spatial data analysis. Environ. Plan. A Econ. Space; 1998; 30, pp. 1905-1927. [DOI: https://dx.doi.org/10.1068/a301905]

27. Yan, Z.; Wang, Y.; Wang, Z.; Zhang, C.; Wang, Y.; Li, Y. Spatiotemporal Analysis of Landscape Ecological Risk and Driving Factors: A Case Study in the Three Gorges Reservoir Area, China. Remote Sens.; 2023; 15, 4884. [DOI: https://dx.doi.org/10.3390/rs15194884]

28. Zhao, Y.Y.; Kasim, A.; Gao, W.P.; Liang, H.W. Quantitative analysis of urban expansion and response factors in Urumqi City based on random forest algorithm and geographical detector. Arid. Land. Geo; 2021; 44, pp. 1729-1739. (In Chinese)

29. Zhang, Q.Q.; Shao, Z.L.; Lin, J.; Meng, L. Analysis on the temporal and spatial pattern evolution of land use function transformation in western oasis cities from the perspective of ‘Production-living-Ecological Space’: A case study of Urumqi. J. Chin. Agric. Mech.; 2022; 43, pp. 176-185+194. (In Chinese)

30. Huang, L. Analysis of Temporal and Spatial Pattern Changes and Driving Factors of Oasis in Turpan City. Master’s Thesis; Xinjiang University: Xinjiang, China, 2019; (In Chinese)

31. Liu, Y.W. The Changes of Landscape Pattern and Driving Forces in Urumqi County. Master’s Thesis; Xinjiang University: Xinjiang, China, 2019; (In Chinese)