Integrating Multi-Source Data to Explore Spatiotemporal Dynamics and Future Scenarios of Arid Urban Agglomerations: A Geodetector–PLUS Modelling Framework for Sustainable Land Use Planning

Abstract

Land use and landscape changes undermine the balance between humans and the environment, threatening sustainable regional development, yet their driving mechanisms and future trends remain insufficiently understood, particularly in arid areas. This study establishes a long-term analytical framework for the temporal evolution and driving mechanisms of land use and landscape patterns in arid areas, based on Landsat remote sensing imagery and socio-economic data. We investigate spatiotemporal evolution trends, driving mechanisms, and spatial non-stationarity of regional landscapes, and apply the Patch-generating Land Use Simulation (PLUS) model to predict future landscape changes under business-as-usual (BAU), economic development (ED), and ecological protection (EP) scenarios. The results show that: (1) Grassland and unused land together account for over 80% of the total area. From 1990 to 2020, built-up land expanded by 1471.58 km², an increase of 190.09%. The comprehensive land use dynamic degree in the Urumqi–Changji–Shihezi (UCS) region was 0.22%, with the highest value observed between 2000 and 2010. (2) At the class level, spatial heterogeneity and fragmentation of different landscape types increased, enhancing regional landscape diversity. (3) Spatiotemporal changes in land use and landscape patterns were driven by the combined effects of natural factors, socio-economic conditions, and policy influences. (4) By 2030, under all three scenarios, unused land is expected to decrease, with the most significant reduction under the EP scenario. Grassland will increase most notably under the EP scenario, built-up land will expand, especially under the ED scenario, and cropland will also grow, mainly under the EP scenario. Forest and water areas will show slight decreases with minimal fluctuations. Overall, the proposed framework effectively captures the spatiotemporal dynamics and driving forces of land use and landscape changes, providing support for the formulation of long-term sustainable development policies.

1. Introduction

Changes in the landscape patterns are key factors influencing ecosystem processes such as soil conservation, carbon sequestration, eco-hydrology, and biogeochemical cycling [1], thereby affecting the structure and function of ecosystems [2]. Since the last century, the world has been experiencing an unprecedented process of urbanization at a global scale [3]. With economic growth and technological advancements, human modification of natural landscapes has intensified [4], leading to extensive transformation and large-scale expansion of artificial landscapes. However, these changes have also resulted in land degradation and ecosystem loss [5], restricting regional development particularly in ecologically fragile areas such as arid and semi-arid regions, which collectively cover approximately 40% of the Earth’s terrestrial surface [6,7]. Under the combined pressures of urbanization, landscape transformation, and ongoing ecological degradation, natural ecological spaces are being increasingly lost, posing a severe threat to sustainable development [8]. Consequently, there is growing interest among researchers in effectively characterizing spatiotemporal land use changes and their driving mechanisms.

Under the influence of human activities, land use/cover change has intensified, continuously encroaching upon and altering habitat patches, leading to modifications in landscape structure composition and spatial configuration. These changes in landscape patterns have triggered a series of ecological problems [9]. Landscape patterns uniquely reflect the composition and spatial structure of landscape elements, providing multiple representations of landscape ecology and geographical characteristics [10]. Landscape patterns reflect regional environmental background conditions, illustrate the characteristics of the natural and human environment, and, at the same time, characterize the influence mechanism in the process of spatial development and change through development trajectories, evolutionary direction, and intensity, thereby enabling a deeper analysis of complex ecological processes [11]. Landscape patterns also reveal the types and levels of ecosystem services, demonstrate the diversity and dominance of landscape elements, and reflect the connectivity and fragmentation among different landscape patches, as well as the spatiotemporal heterogeneity of the overall landscape [12]. Research on landscape patterns encompasses nearly all terrestrial ecosystems, spanning diverse focal ecosystems and research scales, including forest ecosystems, grassland ecosystems, agricultural and urban ecosystems, wetland landscape ecosystems, desert and Gobi ecosystems, and comprehensive watershed ecosystems [13]. Currently, with the continuous improvement of socio-economic development levels, human well-being and demands have become increasingly prominent. Consequently, a growing body of landscape pattern research has focused on regions at various administrative levels, investigating the spatiotemporal changes and influencing factors of landscape patterns during their development, as well as landscape pattern optimization and restoration, and the impacts of these changes on regional ecology and the environment [14,15,16]. Research on the spatiotemporal changes in landscape patterns primarily relies on technical methods based on fundamental metrics and model-based predictive simulations for spatiotemporal assessment, predominantly from the perspectives of human geography and landscape ecology. However, studies integrating ecosystem services and the theoretical framework of landscape perception ecology with multi-scenario settings remain relatively limited, often failing to account for predictive changes under diverse development and management scenarios. Scenario settings are also predominantly based on ecological redlines and environmental protection policies, with limited consideration of alternative scenarios [17,18]. Additionally, research on landscape patterns has predominantly focused on economically developed coastal regions, urban clusters in central and eastern China, and early-stage resource-based cities [19,20,21]. In contrast, oasis urban agglomerations in arid regions, characterized by underdeveloped socio-economic conditions and lagging development, have received limited attention. These areas face inherent challenges, including fragile natural environments, extreme water scarcity, resource shortages, and data deficiencies, severely limiting research efforts. This lack of focus severely hampers the protection of natural resources and the pursuit of green, sustainable development in arid regions [22].

Xinjiang Uighur Autonomous Region (Xinjiang), located in the arid northwestern China, has been experiencing a more intense urbanization process due to its relatively late development. The Urumqi–Changji–Shihezi (UCS) urban agglomeration, comprising Urumqi City, Changji Hui Autonomous Prefecture, and Shihezi City, serves as the political and economic core area of Xinjiang. This region distinctly exemplifies the impacts of national policies such as the “Western Development Strategy”, while simultaneously grappling with pressing issues such as land degradation, urbanization, and the contraction of ecological spaces. These challenges have progressively developed into significant threats to regional sustainable development and human well-being [23]. The UCS region highlights the inherent conflict between ecological fragility in arid-zone urban agglomerations and the pressures of rapid socio-economic development over a short time period. Although scholars have conducted extensive research on developed cities, the differences in land use landscape evolution patterns and driving mechanisms in the UCS region have yet to be fully clarified due to distinct socio-economic–natural composite driving factors and divergent development trajectories. Therefore, this study seeks to answer the following three scientific questions: (1) What are the spatiotemporal evolution patterns of land use landscapes in arid regions? (2) What factors regulate land use changes, and does spatial heterogeneity exist? (3) How will regional land use landscapes evolve under different scenarios? Based on multi-source data, including land use records, socio-economic statistics, and statistical yearbooks, this study examines the spatiotemporal evolution patterns of landscapes from 1990 to 2020. The Geodetector model was employed to investigate the driving mechanisms and spatial heterogeneity of ecological risk patterns. Finally, the PLUS model was employed to simulate the spatiotemporal evolution patterns and development trends of land use landscapes under three scenarios: BAU, ED, and EP. This study used multi-source data to construct a comprehensive long time series model framework for arid zone urban agglomerations, formed a multi-scale and multi-scenario coupled analysis framework, revealed the multi-factor nonlinear synergistic driving mechanism of the landscape pattern evolution of arid zone urban agglomerations, and systematically analysed the nonlinear synergistic effects of natural, socio-economic and policy factors, which compensated for the deficiencies of the traditional model in the analysis of the driving mechanism and the portrayal of spatial heterogeneity. The multi-scenario simulation based on the PLUS model proposes a transition path for future land use in the region, which provides a scientific basis and new ideas for ecological security, sustainable development, and land use optimization in arid zones. Figure 1 illustrates the research framework of this study, including data sources, research methods, content, and technical pathways.

2. Materials and Methods

2.1. Study Area

The UCS region (comprising Urumqi City, Changji Hui Autonomous Prefecture, and Shihezi City) is located in northern Xinjiang (42°52′N–45°28′N, 88°40′E–91°33′E), situated within the ecological transition zone between the southern margin of the Junggar Basin and the northern slopes of the Tianshan Mountains. The total area of this region is approximately 86,000 km² (Figure 2), and it serves as a strategic hub within the core area of the Silk Road Economic Belt. The regional geomorphology is predominantly alluvial fan plains, with a topography that slopes from high elevations in the south to lower elevations in the north. The elevation varies from 450 to 2200 m. The region experiences a typical temperate continental arid climate, characterized by an annual precipitation of 160–220 mm, high evaporation rates of 2000–2800 mm, and an aridity index exceeding 4.0. Water resources are primarily dependent on glacial meltwater and groundwater recharge. Vegetation cover exhibits significant heterogeneity, with natural vegetation primarily comprising desert steppe and halophytic shrubs, whereas oasis agriculture is concentrated along riverbanks. Deserts and gravelly Gobi surfaces account for a large proportion of the land area. Over the past four decades, urbanization in the region has been accelerating, with a continuous expansion of built-up land from 1990 to 2020. The regional GDP maintained an average annual growth rate of 9.8%, contributing 45.2% of Xinjiang’s total economic output [24,25].

However, rapid urbanization has exacerbated the fragmentation of the oasis–desert transition zone, resulted in severe groundwater overextraction, and increased soil salinization [26]. Additionally, areas where the NDVI has declined account for 31% of the total oasis area [27]. As the core of the “Northern Slope of the Tianshan Mountains Urban Agglomeration”, this region accounts for 60% of Xinjiang’s industrial output and 52% of its energy consumption. The intensifying contradictions between land use and population growth, coupled with heightened ecological risks, necessitate in-depth research on land use dynamics and landscape pattern evolution to promote sustainable regional development [24,28].

2.2. Data Sources

The Landsat satellite series, initiated by the National Aeronautics and Space Administration (NASA) in 1972, represents one of the longest-running Earth observation programs globally. To date, nine satellites have been successfully deployed, generating a continuous multitemporal dataset spanning five decades. Characterized by systematic global coverage, specialized spectral bands for terrain feature extraction, and moderate spatial resolution (30 m), Landsat imagery has demonstrated exceptional applicability in land use/cover mapping and landscape pattern evolution analysis. These attributes provide critical technical support for investigating urbanization dynamics in the UCS urban agglomeration. In this study, Landsat products for 1990, 2000, 2010, and 2020 were systematically acquired from the Geospatial Data Cloud Platform (http://www.gscloud.cn, accessed on 8 March 2021) and the United States Geological Survey Earth Explorer (https://earthexplorer.usgs.gov/). A total of 48 scenes were collected following rigorous quality control criteria: cloud cover <10% and acquisition during peak vegetation phenological phases (July–September). Adjacent temporal/spatial scenes were employed for data gap-filling when necessary. Socio-economic drivers, such as population density, GDP spatialization, and Euclidean distances to administrative centers and transportation networks, were obtained from the Resource and Environment Science and Data Center (RESDC, https://www.resdc.cn) and National Geographic Information Public Service Platform (Tianditu, http://www.tianditu.gov.cn). Natural climatic variables included temperature/precipitation grids (National Meteorological Science Data Center, http://data.cma.cn), 30 m DEM-derived topographic parameters (slope, aspect) using ArcGIS tools, and hydrological features from national geospatial databases. Ancillary data included administrative boundaries and cartographic basemaps sourced from the National Catalogue Service For Geographic Information (https://www.webmap.cn/main.do?method=index ). All geospatial datasets underwent standardized preprocessing including coordinate unification (WGS84), resampling (30 m pixel alignment), and mask extraction using the UCS boundary vector. A summary of the data description is shown in Table S1.

2.3. Methods

2.3.1. Remote Sensing Image Processing and Interpretation

Remote Sensing Image Preprocessing

This study used the ENVI operating platform to perform radiometric calibration of images to eliminate the sensor errors of image imaging. In the process, the brightness gray value of the image is converted into physical quantities such as radiometric brightness value; the atmospheric correction is realized by using the FLAASH atmospheric correction model tool; after atmospheric correction, the edge line and feathering are set to obtain the image mosaic results of each period; finally, the administrative boundary vector of the study area is clipped to obtain the regional image data of the UCS region in 1990, 2000, 2010, and 2020 (Figure 3). All image data are uniformly found in the WGS84 (World Geodetic System 1984) geographic coordinate system and Albers projection.

Remote Sensing Image Interpretation

This study examines spatiotemporal landscape pattern changes in the UCS urban agglomeration, focusing on land use evolution dynamics and future change simulations. The methodology follows GB/T 21010-2017 land classification standards [29], integrating remote sensing interpretation frameworks from the Chinese Academy of Sciences’ Resource and Environment Science Data Center with established scholarly approaches. Taking into account the actual conditions of the study area, land use landscape types in the UCS region are classified into six categories: cropland, referring to land used for crop cultivation, including cultivated and newly reclaimed land; forest, which includes areas covered by trees, shrubs, bamboo, and coastal mangroves; grassland, denoting land dominated by herbaceous vegetation; water, encompassing natural terrestrial water bodies and water-related infrastructure such as rivers, canals, lakes, reservoirs, and permanent snow and glaciers; built-up land, referring to land used for urban and rural settlements, transportation, and other infrastructure; and unused land, which includes currently unused land such as deserts, Gobi areas, and saline–alkali land.

In the process of applying remote sensing image interpretation to obtain land use classifications, the choice of classification method determines the accuracy of the classification results. Jozdani et al. compared deep neural networks (DNNs), ensemble classifiers (ECs), and the support vector machine (SVM) [30], while Kasahun et al. examined four classification techniques [31]: maximum likelihood classifier (MLC), artificial neural networks (ANNs), random forest (RF), and object-oriented classification. Comparative studies on the interpretation performance of different algorithms have shown that neural network algorithms achieve the highest classification accuracy, followed closely by the random forest classifier, with minimal difference between the two. Both outperform other classification methods [32]. Considering its ease of use of algorithm application platforms, stability in handling large datasets, and complexity in parameter selection and configuration, as well as computational efficiency and processing time, this study selects the random forest algorithm for land use landscape classification. To further enhance the accuracy of the random forest classifier, this study integrates multiple data sources based on the characteristics of the study area and research objectives. These data sources include preliminary unsupervised land use classification data (ISO data), satellite imagery spectral bands (Blue, Green, Red, NIR, SWIR), synthetic spectral indices (normalized difference vegetation index, normalized build index, modified normalized water index, modified soil-adjusted vegetation index 2), and topographic and geomorphological data (DEM, slope, aspect). Finally, ENVI 5.3, combined with high-resolution satellite imagery and visual interpretation, is used to perform local corrections and validation of the land use landscape types after supervised classification.

To scientifically evaluate the accuracy of the classification results, a confusion matrix was used as the evaluation basis, with overall accuracy (OA) and the kappa coefficient selected for validation. A validation sample set (

Table 1. The validation sample of land use classification from 1990 to 2020.

Year Category	1990	2000	2010	2020
Cropland	400	393	409	388
Forest	175	181	189	168
Grassland	243	257	234	263
Water	120	104	111	132
Built-up land	275	286	271	293
Unused land	419	417	404	422
Total	1632	1638	1618	1666

) was established by combining field verification with high-resolution satellite imagery. The final overall classification accuracy and kappa coefficients for the years 1990, 2000, 2010, and 2020 were 88.36% (0.86), 91.54% (0.90), 91.30% (0.89), and 92.59% (0.91), respectively. These accuracy levels meet the classification precision requirements and are suitable for subsequent research in this study.

2.3.2. Land Use Dynamic Degree

Single Dynamic Degree of Land Use

It is of great significance to analyze the changes in and trends of land use in time series for regional sustainable management [33]. The single dynamic degree analysis model of land use can reflect the intensity of changes in different land types over a period of time. The formula is as follows [33]:

$$L_s={\displaystyle \frac{S_b-S_a}{S_a}}\times {\displaystyle \frac{1}{T}}\times 100\%$$

(1)

In the formula, $L_s$ is the dynamic degree of a certain type of land use; $S_a$ is the initial area of a land use type; $S_b$ is the later area of a land use type; and $T$ is the research time interval, and this study represents the inter-annual change rate in annual units.

Comprehensive Land Use Dynamic Degree

The comprehensive land use dynamic degree can reflect the overall change in land use in the region. The greater the value of the comprehensive land use dynamic degree, the more severe the land use change in the study period, and the more stable and the smaller the change. The formula is as follows [33]:

$$L_o=\left|{\displaystyle \frac{\sum _{i=1}^n{∆S}_i}{2\sum _{i=1}^n{S}_i}}\right|\times {\displaystyle \frac{1}{T}}\times 100\%$$

(2)

In the formula, $L_o$ is the comprehensive land use dynamic degree; ${∆S}_i$ is the area number of the $i$th landscape type transformed into other types within a certain time interval; $S_i$ is the area number of land landscape type $i$ at the beginning of a certain time interval; and $T$ is the research time interval, and this study represents the inter-annual change rate in annual units.

2.3.3. Landscape Pattern Index

Landscape pattern indices represent a direct and effective means of characterizing landscape configurations in landscape ecology research. They consolidate and quantitatively reflect the state information of landscape patterns, enabling the effective exploration of characteristics at the patch, class, and landscape levels—including the number, composition, and spatial configuration of landscape elements across different scales [34]. This study adheres to the principle of capturing the state and trends of landscape change while avoiding indicator redundancy. By integrating previous research and the specific regional context [35,36,37], the selected landscape pattern indices are presented in

Table 2. Examination of necessary landscape pattern indices and their ecological implications.

Scale Level	Name	Index Type	Ecological Significance
Class level	Number of patches (NP)	Landscape quantity	The number of patches of a certain type of landscape
	Patch density (PD)	Aggregation	Describes the degree of patch differentiation or fragmentation of the overall landscape
	Landscape shape index (LSI)	Edge shape	Describes the complexity of landscape shape
	Largest patch index (LPI)	Landscape quantity	The influence of dominant types on the whole landscape coverage pattern
	Mean nearest-neighbor distance (MNN)	Edge shape	Describes the degree of dispersion of landscape space
Landscape level	Patch richness (PR)	Diversity	Describes the richness of different landscapes
	Patch Cohesion index (COHESION)	Adjacency relation	Characterizes the link degree of patches in the landscape
	Contagion index (CONTAG)	Aggregation	Reflects the degree of agglomeration of different plaque types
	Shannon’s diversity index (SHDI)	Diversity	Describes the richness and complexity of landscape types. The richer the landscape components, the more serious the fragmentation
	Shannon’s evenness index (SHEI)	Diversity	Describes the uniformity of landscape type distribution

. Utilizing the Fragstats, the study employs the moving window method by selecting an appropriate window scale to compute and analyze the chosen landscape pattern indices. Based on existing research outcomes [38,39] and considering similar regional scales as well as research findings in arid regions [40,41,42], a 250 m scale was ultimately selected as the appropriate scale for the investigation of landscape pattern indices in the UCS region.

2.3.4. Driving Mechanism of Landscape Pattern

Revealing the evolution process and law of landscape pattern change and analyzing its driving mechanism provide scientific theoretical basis for regional sustainable development management. Geodetectors use a series of statistical methods to detect spatial differences and reveal the driving forces behind variables [43]. The geographical detector model is a nonlinear model that does not require consideration of multicollinearity between factors. The influence of different independent variables on the final model results is also independent, and the interaction between the factors can be calculated. This is an advantage and a highlight of the geographical detector model [44]. In this study, factor detection and interactive detection were used to analyze the influence of each factor on landscape change. The specific formula is as follows:

(1) The factor detector is used to detect the explanatory power of different independent variable factors to the dependent variable and the spatial differentiation of the dependent variable. The greater the explanatory power of the results, the greater the influence and contribution of the independent variables on the change in landscape pattern, and vice versa. Measured by $q$ value, the expression is [43]:

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

(3)

$$\begin{gathered} SSW={\sum }_{h=1}^L{N}_h{\sigma }_h^2, \\ SST=N{\sigma }^2 \end{gathered}$$

(4)

The range of $q$ is [0, 1], and the larger the value is, the more obvious the spatial differentiation of the dependent variable is. If the stratification is generated by the independent variable, the larger the value of $q$ indicates the stronger the explanatory power of the independent variable to the dependent variable. In some special cases, a $q$ value of 1 indicates that the spatial distribution of the dependent variable is completely controlled by the independent variable, and a $q$ value of 0 indicates that it has no relationship.

(2) The interaction detector is used to identify the strength of the interaction between the factors, that is, to analyze whether the explanatory power of the two factors (X1 and X2) on the dependent variable will be enhanced or weakened, or the influence of the factor on the dependent variable is independent of each other. Calculate the $q$ value of a single factor: $q$(X1) and $q$(X2); the interaction between the two is the $q$ value: $q$(X1∩X2). Then, compare $q$(X1), $q$(X2), and $q$(X1∩X2). The interaction relationship of factors can be divided into five categories, as shown in

Table 3. Types of two-factor interactions.

No.	Interaction Types	Judgement Basis
1	Nonlinear weakening	$q$(X1∩X2) < Min($q$(X1), $q$(X2)
2	Single factor nonlinear weakening	Min($q$(X1), $q$(X2) < $q$(X1∩X2) < Max($q$(X1), $q$(X2)
3	Two-factor enhancement	$q$(X1∩X2) > Max($q$(X1), $q$(X2)
4	Independence	$q$(X1∩X2) = $q$(X1) + $q$(X2)
5	Nonlinear enhancement	$q$(X1∩X2) > $q$(X1) + $q$(X2)

[43].

In the process of selecting driving factors, this study references previous research results [45,46,47] based on the availability, reliability, correlation, and consistency of data. Taking into account the natural conditions and development status of the study area, 11 factors were selected (

Table 4. Driving factors of landscape pattern change in the UCS.

	Variable	Factors	Data Type
Natural factors	X1	Temperature	Continuous grid
	X2	Precipitation
	X3	DEM
	X4	Slope
Socio-economic factors	X5	Population
	X6	GDP
Policy factors	X7	Distance to government
	X8	Distance to highway
	X9	Distance to railway
Spatial factors	X10	Distance to primary road
	X11	Distance to water

), including temperature, precipitation, digital elevation model (DEM), slope, population, gross domestic product (GDP), distance to government offices, distance to main roads, distance to water bodies, distance to highways, and distance to railways, where the seat of government is used to represent the regional centre. These factors span four categories: natural, socio-economic, policy, and spatial distance factors.

The independent variable X in the geographical detector model must be discretized. Therefore, this study uses ArcGIS to categorize the driving factors into five levels using the natural breaks method. The data for all driving factors across different periods were used as independent variables X, while landscape change was treated as the dependent variable Y. The geographical detector model was then employed to obtain the quantitative relationship between landscape change and the driving factors for different periods.

2.3.5. The PLUS Model

The PLUS (Patch-generating Land Use Simulation) model is an advanced tool for simulating land use change by integrating a cellular automata (CA) framework with a patch-generating simulation strategy [48]. It is designed to capture the drivers of land use transitions and simulate patch-level dynamics across multiple land categories. The model incorporates several modules, including the Land Expansion Analysis Strategy (LEAS) for extracting the expansion portions of various land use types, a demand prediction module based on Markov chains, and a CA component that utilizes multiple random seeds (CARS) to simulate spatially explicit patch generation. In our application, 15 driving factors (detailed in Table S1) were used to calculate the probability of occurrence for each landscape type. To ensure accuracy and efficiency, we configured the random forest (RF) algorithm with 50 regression trees, an RF feature sampling rate of 15, and a sampling rate of 0.01. The RF algorithm quantifies the contribution of each driver to the expansion of specific land types, while the CA mechanism, featuring a threshold-decreasing process, governs the automatic generation of new patches based on a probability surface and stochastic seeding [48] (The PLUS model download and detailed tutorial: https://github.com/HPSCIL/Patch-generating_Land_Use_Simulation_Model accessed on: 25 March 2025).

The PLUS model was employed to simulate the spatial distribution of land use in the UCS region for the year 2020. To evaluate the model’s performance, the simulated results were compared with actual land use data for the same year. The Kappa coefficient, a commonly used metric to assess spatial agreement, was used to validate simulation accuracy. This coefficient ranges from 0 to 1, with values closer to 1 indicating stronger spatial consistency. A Kappa value greater than 0.7 is generally considered indicative of high simulation accuracy. In this study, a random sampling rate of 10% was applied to compare the simulated and observed land use data, resulting in a Kappa coefficient of 0.93 and an overall accuracy of 0.96. This high level of agreement demonstrates that the PLUS model yields reliable simulation results and is suitable for projecting future land use patterns in the UCS region.

For future projections, three multi-scenario designs were established: business-as-usual (BAU), economic development (ED), and ecological protection (EP). The BAU scenario extrapolates recent trends observed from 2010 to 2020. In contrast, the ED scenario simulates rapid urban and economic expansion by increasing the probability of conversion from cropland, forestland, and grassland to built-up areas, while reducing the probability of deconversion from built-up areas. The EP scenario prioritizes sustainability by decreasing conversion probabilities from natural to built-up land and promoting the transformation of unused land into ecological and agricultural uses. These scenario configurations, informed by the relevant literature and consultations with land planning authorities [49,50], provide a comprehensive framework for assessing future land use trajectories and their associated ecosystem services.

3. Results

3.1. Spatial–Temporal Pattern Change in Land Use

3.1.1. The Landscape Quantity Changes in Land Use

An analysis of quantitative land use changes from 1990 to 2020 revealed significant trends in regional development and land use structure within the UCS region (Figure 4). Data from four time points indicated that cropland and built-up land increased steadily, while forest land and water areas declined. Grassland initially declined and subsequently recovered, while unused land underwent a sharp rise around 2010 before receding. Specifically, in the UCS region as a whole, cropland expanded by 2759.78 km² (an increase of 29.54%), and built-up land surged by 1471.58 km² (a 190.09% increase). In contrast, forest land shrank by 1211.50 km² (a 37.53% reduction) and grassland diminished by 4112.40 km² (a 10.69% decline), while unused land expanded by 1647.86 km² (a 4.63% increase). A more detailed examination of subregions revealed distinct patterns. In Urumqi City, all ecological land types—including cropland, forest land, grassland, and water area—declined. Specifically, cropland contracted by 233.56 km² (17.18%), forest land by 239.41 km² (36.86%), grassland by 695.21 km² (8.67%), and water area by 140.24 km² (38.40%). Conversely, built-up land expanded by 595.55 km² (204.58%) and unused land grew by 712.89 km² (22.54%). In Changji Prefecture, cropland increased by 3053.46 km² (39.79%) and built-up land by 786.59 km² (175.41%), while forest land, grassland, and water area decreased by 968.45 km² (37.64%), 3396.19 km² (11.19%), and 409.49 km² (44.21%), respectively; unused land showed a modest rise of 933.33 km² (2.88%). In Shihezi City, ecological areas declined sharply, with cropland reducing by 60.12 km² (19.41%), forest land by 3.62 km² (60.84%), and grassland by 20.94 km² (20.67%), while built-up land and unused land increased by 89.44 km² (258.42%) and 1.64 km² (136.67%), respectively.

Overall, the marked expansion of built-up land, along with the steady increase in cropland and significant reductions in forest, grassland, and water areas, underscores a rapid urbanization process. This trend suggests that escalating developmental pressures are increasingly displacing valuable ecological land, emphasizing the urgent need for sustainable land management and urban planning strategies.

Figure 5 illustrates the changes in the proportion of each land use type. Overall, owing to its unique geographical location, unused land and grassland constitute the largest proportion, exceeding 40% and 35%, respectively, together comprising nearly 80% of the total area. Cropland accounts for 10.50–13.60%, forest land for 2.30–3.60%, built-up land for 0.90–2.50%, and water area for 0.80–1.50%. This also indicates that water resources are limited, while unused land and grassland represent a dominant share, providing a fundamental advantage for the development of agricultural and animal husbandry industries.

From 1990 to 2020, the proportion of unused land rose from 40.10% in 1990 and 40.20% in 2000 to 45.30% in 2010, before declining to 42.00% in 2020. The proportion of grassland declined steadily from 43.30% in 1990 to 36.50% in 2010, before increasing slightly to 38.70% in 2020. The proportion of cropland consistently increased, rising from 10.50% in 1990 to 13.60% in 2020. The proportion of forest land gradually declined from 3.60% in 1990 to 2.30% in 2020. Meanwhile, the proportion of built-up land experienced steady growth, expanding from 0.90% to 2.50%. The proportion of water areas declined overall, dropping from 1.50% in 1990 to 0.80% in 2020.

3.1.2. Rate of Change in Land Use Landscapes

Analysis of Table S2 and Figure 6 indicates distinct temporal variations in land use within the UCS region. Between 1990 and 2000, grassland underwent the most dramatic reduction, shrinking by 745.37 km², while the dynamic degree of built-up land reached 2.14%, marking the most intense change. Both forest and grassland areas exhibited continuous declines. During 2000–2010, grassland experienced the most substantial transformation, decreasing by 5337.27 km², while the dynamic degree of built-up land surged to 5.85%, reflecting even more drastic changes. Forest, grassland, and water areas declined more sharply than in the previous decade. From 2010 to 2020, the most notable change was observed in unused land, which contracted by 2920.32 km², although built-up land continued to exhibit the highest dynamic degree at 5.08%. While forest and grassland continued to decline, their rates of reduction moderated compared to the previous period.

In Urumqi, during 1990–2000, built-up land exhibited the highest volatility, increasing by 59.46 km², with a dynamic degree of 2.04%. Meanwhile, cropland, forest, grassland, and unused land declined, except for water and built-up land. The 2000–2010 period saw a substantial increase in unused land (+470.58 km²) and a rapid rise in the dynamic degree of built-up land to 9.31%, while other land types, including cropland, forest, grassland, and water, underwent even sharper declines. During 2010–2020, grassland contracted by 318.16 km², while the dynamic degree of built-up land decelerated to 3.10%. Cropland diminished more drastically, though the reductions in forest, grassland, and water areas showed signs of moderation.

In Changji, grassland contracted by 717.48 km² from 1990 to 2000, while built-up land recorded a dynamic degree of 2.16%. During this period, forest and grassland declined, whereas cropland, water, built-up, and unused land expanded. Between 2000 and 2010, grassland further declined by 4968.00 km², with water area exhibiting the most dramatic contraction (dynamic degree of −4.42%). Built-up land, cropland, and unused land increased, while forest, water, and grassland declined more sharply. From 2010 to 2020, unused land contracted by 3194.39 km², while built-up land maintained a dynamic degree of 6.93%, remaining the most rapidly transforming land type. Other land types, with the exception of cropland, continued to decline.

In Shihezi, during 1990–2000, built-up land expanded by 9.14 km², while unused land recorded a dynamic degree of 13.17%, marking the most intense transformation, whereas cropland, forest, and grassland declined. Between 2000 and 2010, built-up land expanded by 39.20 km² (dynamic degree 8.96%), with concurrent growth in unused land and accelerated declines in water, grassland, forest, and cropland. From 2010 to 2020, built-up land again exhibited the greatest increase (+41.10 km², dynamic degree 4.95%), while most other land use types continued to decline.

Overall, these trends underscore pronounced urban expansion and the transformation of built-up land at the expense of ecological land uses, reflecting rapid urbanization and escalating developmental pressures across the region.

As shown in

Table 5. Comprehensive land use dynamic degree (%) in each period from 1990 to 2020.

Region		1990–2000	2000–2010	2010–2020	1990–2020
Comprehensive land use dynamic degree (%)	The UCS	0.09	0.79	0.34	0.22
	Urumqi	0.05	0.58	0.35	0.32
	Changji	0.10	0.84	0.44	0.21
	Shihezi	0.25	0.86	0.89	0.66

, between 1990 and 2020, the comprehensive land use dynamic degree in the UCS area averaged 0.22%, with values of 0.32% in Urumqi, 0.21% in Changji, and 0.66% in Shihezi. This indicates that the magnitude of comprehensive land use change was relatively significant in Shihezi and Urumqi. Across different time periods, the comprehensive land use dynamic degree peaked between 2000 and 2010, exhibiting a trend of initial increase followed by a decline. As shown in Figure 6, the dynamic degree of built-up land mirrored its overall trend.

3.1.3. Changes in the Spatial Distribution of Land Use

The overall transfer sources and destinations of various land use types in the UCS region from 1990 to 2020 are summarized as follows. The cropland area increased overall, with an inflow of 4894.89 km², mainly from grassland and unused land, and an outflow of 2135.11 km², primarily to grassland and built-up land. The forest land area decreased overall, with an inflow of 700.87 km² primarily from grassland and cropland, while 1912.36 km² was converted mainly to grassland and cropland. The grassland area also declined, with 8064.92 km² of land converted to grassland, mostly from unused land and forest land. Meanwhile, 386.57 km², mainly from unused land and grassland, was converted into water areas, while 942.69 km² of water was transformed into other land types, primarily unused land and grassland. The built-up land area expanded substantially, with 1756.30 km² converted from various land types, primarily from cropland and grassland. Conversely, only 284.72 km² of built-up land was converted into other land types, mostly cropland and grassland. A total of 8159.83 km² of land, mainly grassland and water areas, was converted into unused land, while 6511.57 km² of unused land was transformed into other land types, primarily grassland and cropland. As illustrated in Figure 7, between 1990 and 2000, land use changes mainly occurred between cropland and grassland, while the extent of unused land remained largely unchanged. During 2000–2010, land conversion intensified, particularly between cropland and grassland, as well as between unused land, grassland, and forest land. In the period from 2010 to 2020, the most pronounced shifts were observed between grassland and unused land. Over the entire period from 1990 to 2020, land conversion primarily took place between grassland, unused land, cropland, and forest land, whereas built-up land had a relatively low transfer volume. However, all land types contributed to the expansion of built-up land, with cropland, grassland, forest land, and unused land being the most significant sources.

Figure 8 illustrates the spatial distribution of land use transfer changes over different periods. The conversion of cropland primarily took place within existing cropland areas, driven mainly by the expansion of residential zones into the northern and southern flatlands. The conversion of forest land predominantly occurred in low- and mid-altitude regions on the northern slopes of the Tianshan Mountains and around urban residential areas. Grassland transfer was concentrated in the Gurbantunggut Desert in the north and in mountainous grassland regions along the southern slopes of the Tianshan Mountains. Water transfer was mainly observed in high-altitude glaciers of the Tianshan Mountains, as well as in rivers, small lakes, and wetlands within populated areas. The expansion of built-up land was most prominent in Urumqi, Changji, and Shihezi, along with villages and towns adjacent to National Highway 312 and the Zhundong Economic and Technological Development Zone. Unused land conversion was largely found in the southern mountainous regions, the northern deserts and Gobi, and in transitional zones where cropland, grassland, and water interlace.

3.2. Spatial and Temporal Evolution of Landscape Pattern Indices

Based on the Fragstats results (

Table 6. Landscape pattern index of the UCS from 1990 to 2020.

Indicator Level		Landscape Index	Year
			1990	2000	2010	2020
Type level	Cropland	NP	488.00	743.00	890.00	820.00
		PD	0.0055	0.0084	0.0100	0.0092
		LPI	3.4239	4.4252	6.6891	7.0863
		LSI	45.9083	55.0189	56.1233	56.1975
		MNN	1082.2848	959.2508	962.5855	941.1929
	Forest	NP	1459.00	1466.00	2279.00	2225.00
		PD	0.0164	0.0165	0.0257	0.0251
		LPI	0.1608	0.1608	0.1776	0.2098
		LSI	68.1824	67.7428	77.4429	80.2917
		MNN	922.5401	991.2205	682.7431	671.1117
	Grassland	NP	800.00	1049.00	1466.00	1631.00
		PD	0.0090	0.0118	0.0165	0.0184
		LPI	27.1833	35.7191	27.0696	29.6109
		LSI	51.8547	53.2683	59.5167	59.8307
		MNN	738.7584	737.9229	732.7951	717.5750
	Water	NP	371.00	401.00	901.00	931.00
		PD	0.0042	0.0045	0.0102	0.0105
		LPI	0.2194	0.2194	0.0557	0.0544
		LSI	31.9931	33.0369	40.5487	42.1187
		MNN	1585.1869	1590.8029	1110.0086	1307.6915
	Built-up land	NP	930.00	1230.00	1483.00	1823.00
		PD	0.0105	0.0139	0.0167	0.0205
		LPI	0.2249	0.2808	0.4867	0.7054
		LSI	34.2018	38.7114	37.9258	43.4289
		MNN	1732.2406	1592.9874	1484.1139	1325.8525
	Unused land	NP	352.00	429.00	300.00	327.00
		PD	0.0040	0.0048	0.0034	0.0037
		LPI	32.5136	32.8153	40.7706	36.7327
		LSI	23.9834	25.8279	18.7899	20.5994
		MNN	1224.3505	1178.3360	1233.9438	1256.5799
Landscape level		SPLIT	5.3443	4.1882	4.0486	4.3424
		COHESION	99.6523	99.7210	99.7048	99.6939
		CONTAG	58.3804	57.3395	58.1887	56.9465
		SHDI	1.1899	1.2065	1.1944	1.2230
		SHEI	0.6641	0.6734	0.6666	0.6826

), significant spatial and temporal variations in landscape pattern indices were observed from 1990 to 2020. At the class level, the patch density (PD) of cropland increased markedly before slightly declining by 2020, indicating an overall rise in spatial heterogeneity and fragmentation. The largest patch index (LPI) continuously increased, reflecting the growing dominance of certain patches and intensifying human influence. Meanwhile, the landscape shape index (LSI) exhibited sustained growth, indicating greater shape complexity and irregularity. The mean nearest neighbor distance (MNN) declined overall, implying reduced dispersion and a trend toward aggregation. For forest land, PD generally increased, followed by a slight decrease in 2020. Both LPI and LSI exhibited continuous growth, while MNN first increased (1990–2000) before decreasing, indicating an initial phase of dispersion followed by gradual clustering. Grassland displayed a steady rise in PD, with fluctuating LPI, increasing LSI, and declining MNN, all suggesting enhanced aggregation. In water areas, PD consistently increased, while LPI remained stable initially, then increased after 2010 before slightly declining in 2020. LSI exhibited a steady rise, whereas MNN increased, suggesting enhanced dispersion. For built-up land, both PD and LPI increased significantly, while LSI showed an oscillatory trend—rising, then falling, and rising again—indicating fluctuations in shape complexity. MNN declined, suggesting a shift toward a more clustered distribution. In contrast, unused land exhibited fluctuating PD and LSI, while LPI initially increased but later declined. MNN first decreased before increasing, reflecting periodic variations in spatial heterogeneity and fragmentation.

At the landscape level, the separation index (SPLIT) first decreased and then increased. The initial decrease in SPLIT indicates improved regional ecosystem stability and greater patch concentration. The contagion index (CONTAG) exhibited a fluctuating trend—declining initially, then increasing, followed by another decline—with only minor variations, suggesting relatively stable connectivity among dominant landscape patches. The Shannon diversity index (SHDI) showed an overall upward trend but declined in 2010, indicating increasing landscape diversity and enrichment, accompanied by a narrowing proportional difference among various patch types. Similarly, the Shannon evenness index (SHEI) followed the same pattern as SHDI, suggesting a more uniform distribution of different patch types within the landscape.

3.3. Mechanisms Driving Spatial and Temporal Landscape Change

3.3.1. Factor Detection Results and Analysis

The factor detection module of the geographical detector identifies key drivers of landscape change in the UCS region, quantifying the influence of each factor on landscape changes across different periods. Based on the p-value thresholds, this study excludes factors with p-values greater than 0.05 from ranking the q values, as shown in Tables S3–S8.

The influence of different factors on cropland landscape changes varies across periods. According to Table S3, the distance to the government center strongly explains cropland changes, while DEM and slope have negligible effects. Table S4 shows that temperature and precipitation have a strong impact on forestland changes, while slope, population, and GDP exert minimal influence. Table S5 reveals that precipitation, temperature, and distance to the government center significantly affect grassland changes, whereas slope, population, and GDP have little influence. As shown in Table S6, precipitation, temperature, and DEM are key drivers of water body changes, with minimal influence from distance to major roads, population, and GDP. Table S7 highlights that GDP, population, and distance to highways significantly influence built-up land changes, while DEM and distance to water bodies have no significant effect. Finally, Table S8 demonstrates that distance to the government center, distance to highways, and precipitation are the main factors driving changes in unused land, whereas slope and population have little explanatory power.

3.3.2. Interaction Factor Detection Results and Analysis

To further explore the driving forces behind landscape dynamics after accounting for interactions between factors, this study used the interactive detection module in the geodetector to examine the interactions of each factor. A total of 55 interactions were generated between each landscape change and 11 factors at different times. The detection results for each landscape type are shown in Figures S1–S6. The interaction analysis of landscape change reveals both nonlinear and two-factor enhancement effects across different land use types and periods.

For cropland, from 1990–2000, there were 45 nonlinear and 10 two-factor enhancements, with the strongest effect observed between temperature and distance to the government. From 2000–2010, the number of nonlinear enhancements increased to 53, with only 2 two-factor enhancements, and the interaction between temperature and precipitation became dominant. From 2010–2020, there were 50 nonlinear and 5 two-factor enhancements, with notable influences from the interactions between distance to the government and railway, as well as between railway and main road. Overall, from 1990 to 2020, cropland recorded 42 nonlinear and 13 two-factor enhancements, with the strongest interaction between distance to the railway and government.

For forest landscapes, 1990–2000 saw 52 nonlinear and 3 two-factor enhancements, with temperature and distance to the government exerting the strongest influence. From 2000–2010, 46 nonlinear and 9 two-factor enhancements were detected, with the interaction between distance to the expressway and precipitation being the most pronounced. From 2010–2020, 51 nonlinear and 4 two-factor enhancements were observed, with significant effects from both temperature–precipitation and temperature–distance to the railway interactions. Overall, from 1990 to 2020, forest landscapes experienced 55 nonlinear enhancements, with the temperature–precipitation interaction being paramount.

For grassland, during 1990–2000 there were 52 nonlinear and 3 two-factor enhancements, with the strongest interaction between precipitation and DEM. From 2000–2010, 54 nonlinear enhancements and 1 two-factor enhancement were recorded, with the interaction between distance to the expressway and precipitation being dominant. From 2010–2020, grassland showed 54 nonlinear enhancements and 1 two-factor enhancement, with temperature and distance to water playing a strong role. Over the entire period, grassland changes involved 47 nonlinear and 8 two-factor enhancements, with the temperature–precipitation interaction being the most influential.

For water areas, 1990–2000 exhibited 49 nonlinear and 6 two-factor enhancements, with precipitation and distance to the highway having the strongest effect. From 2000–2010, 51 nonlinear and 4 two-factor enhancements were noted, with significant interactions among distance to the expressway, precipitation, and temperature. From 2010–2020, 52 nonlinear and 3 two-factor enhancements were observed, with the temperature–precipitation interaction being dominant. Overall, water areas experienced 50 nonlinear and 5 two-factor enhancements, with temperature-related interactions showing greater significance.

For built-up land, the period from 1990–2000 recorded 48 nonlinear and 7 two-factor enhancements; from 2000–2010, 46 nonlinear and 9 two-factor enhancements were noted; and from 2010–2020, 45 nonlinear and 10 two-factor enhancements were observed. Cumulatively, from 1990 to 2020, there were 54 nonlinear enhancements and 1 two-factor enhancement, with interactions involving GDP emerging as the most significant.

For unused land, during 1990–2000 there were 41 nonlinear and 14 two-factor enhancements, with precipitation and DEM being the strongest interaction. From 2000–2010, 52 nonlinear and 3 two-factor enhancements were observed, with key effects from interactions among distance to water, precipitation, and temperature. From 2010–2020, there were again 52 nonlinear and 3 two-factor enhancements, with temperature and distance to the government having the strongest influence. Overall, from 1990 to 2020, unused land recorded 43 nonlinear and 12 two-factor enhancements, with the temperature–precipitation interaction dominating.

These results underscore the complex, dynamic interplay of environmental and socio-economic factors in shaping landscape changes over time in the UCS.

3.4. Prediction of Land Use Change Under Different Development Scenarios

To analyze future land use trends in the UCS region and assess potential changes in ecosystem services, trade-offs, and synergies, this study aims to develop scientific and effective ecological and environmental protection strategies. Additionally, it examines the evolution of ecosystem service trade-offs and synergies under different future scenarios and proposes recommendations for sustainable regional development. Three different development management scenarios—namely, the natural growth, economic development, and ecological environmental protection scenarios—were established to simulate and predict the regional land use pattern in 2030. To ensure the reasonableness and scientific accuracy of the scenarios, professional managers from the regional government’s Land Resources Bureau, Ecological Environment Protection Bureau, and other relevant departments were consulted to determine the transfer probabilities of land use types in the economic development and ecological protection scenarios, based on years of experience, the region’s natural conditions, and the development policy of the “Three Areas and Three Lines”. The specific settings of the three scenarios, based on the Markov model, are shown in Tables S8–S10. Using the 2020 land use data for the UCS region, the PLUS model was employed to simulate and predict land use changes for 2030 under the three distinct scenarios: natural growth, economic development, and ecological protection. This simulation produced spatial distribution patterns of land use under each scenario (Figure 9).

By comparing the simulated land use results under different scenarios with the 2020 land use data,

Table 7. The area of and variation in each type of land use in the UCS under different scenarios from 2020 to 2030.

Types	Area (km2)				Land Use Change (km2)
	2020	BAU	ED	EP	2020-BAU	2020-ED	2020-EP
Cropland	12,066.88	12,457.69	12,393.56	12,546.88	390.81	326.69	480.00
Forest	2014.81	1945.50	1943.81	1948.50	−69.31	−71.00	−66.31
Grassland	34,241.44	35,781.50	35,684.81	37,183.06	1540.06	1443.38	2941.63
Water	744.00	705.94	705.75	705.75	−38.06	−38.25	−38.25
Built-up land	2234.56	2925.50	3177.94	2765.81	690.94	943.38	531.25
Unused land	37,167.44	34,653.00	34,563.25	33,319.13	−2514.44	−2604.19	−3848.31

shows that unused land exhibited the most significant change across all three scenarios, consistently decreasing. The most substantial reduction occurred under the ecological protection scenario, with a decrease of 3848.31 km². Grassland area increased in all scenarios, with the greatest expansion observed under the ecological protection scenario, reaching 2941.63 km². Built-up land grew in all development scenarios, with the largest increase of 943.38 km² under the economic development scenario. Cropland also expanded in all scenarios, with the most significant growth of 480.00 km² under the ecological protection scenario. Conversely, forest land and water bodies showed a decreasing trend across all three scenarios, with relatively minor fluctuations in the magnitude of the reduction.

4. Discussion

4.1. The Trends in Regional Landscape Patterns

Due to global climate change and the ongoing development of human socio-economic activities, regional landscape patterns have undergone dramatic transformations, affecting both the structure and function of landscapes. The results indicate that from 1990 to 2020, the UCS region experienced an increase in cropland and built-up land, while forest land and water bodies declined. Grassland areas initially decreased and then increased, while unused land expanded significantly until 2010 before declining thereafter. The largest transfer area was observed for cropland, with a total increase of 4894.89 km², while the transfer-out area for built-up land was minimal. All land use types contributed to the conversion into built-up land, with cropland, grassland, forest land, and unused land making notable contributions. These observations are consistent with the findings of He et al. [51]. Due to the characteristic distribution of typical landscape types in arid regions, unused land and grassland account for a relatively high proportion of the total area. During socio-economic development, large-scale cropland reclamation and the continuous expansion of built-up land primarily resulted from the conversion of unused land and grassland. Additionally, landscape types with strong ecological functions, such as forest land and water bodies, have been increasingly consumed and encroached upon to meet the growing resource demands of urbanization. These findings align with the conclusions drawn by Yang et al. [52]. The cropland and built-up land areas in the study area increased by 29.54% and 190.09%, respectively, reflecting the largest and fastest growth among the land use types. This aligns with the results and conclusions of Li et al. [53], reflecting the rapid pace of urbanization in Xinjiang, where agriculture remains a vital and fast-developing industry. From the perspective of land use dynamics, it is evident that the intensity of land use changes increased significantly after 2000, with the most intense changes occurring between 2000 and 2010. This period coincides with the implementation of the “Western Development Strategy” that began in 2000, a pattern also identified by Shi et al. [54] and Chen et al. [55]. The spatiotemporal characteristics of landscape pattern indices show that the degree of landscape fragmentation and spatial heterogeneity in the UCS region has gradually increased. This reflects the inherent distribution characteristics of landscape types in Xinjiang’s arid regions, where long-term development and changes follow similar overall trends. The region also exhibits typical oasis urban landscape characteristics. The spatial distribution of land use landscape types follows a distinct pattern from south to north: “water bodies near mountains, forests near mountains, cities along roads, and cropland surrounding cities”. The changes in landscape types are primarily concentrated around urban residential areas.

4.2. The Drivers and Mechanisms of Landscape Pattern Changes

The study on the driving mechanisms behind landscape pattern changes in the UCS region reveals that natural, socio-economic, and policy factors exert varying influences on different landscape types across different periods. Among the natural factors, temperature, precipitation, and the digital elevation model (DEM) play significant roles. Regarding socio-economic factors, GDP has a substantial impact, while policy factors such as proximity to government centers, railways, and highways collectively exert considerable influence on landscape pattern changes. Zhang et al. investigated the driving forces of land use change in Urumqi and identified population growth and economic development as the primary factors influencing land use change in the city [56]. Similarly, Li Xiaolong et al. found that the main drivers of land use distribution and change were elevation, slope, temperature, precipitation, and distance to residential areas [57], which aligns closely with the conclusions of this study. Furthermore, this study employed interaction detection to explore the interactive relationships among various factors in the changes in different land use landscape types. Cropland was most strongly influenced by the interaction between the distance to railways and the distance to government centers, while forest land, grassland, water bodies, and unused land were most significantly affected by the interaction between temperature and precipitation. Built-up land exhibited strong interactions between GDP and other factors. Overall, the analysis of interaction effects across different periods indicates that the dynamic changes in regional land use landscape types are not driven by a single factor but rather by the combined effects of multiple interacting factors. This finding is consistent with the results of Lu et al. on the influencing factors of land use change in the Sangong River Basin oasis [58]. However, it differs from the findings of Kang Xuan and Wang Xuemei on the factors influencing landscape pattern changes in the Weigan–Kuqa River Delta in Xinjiang [59]. This discrepancy arises because the UCS region, compared to river basins in southern Xinjiang, has distinct ethnic compositions and cultural backgrounds, leading to significant differences in production practices and lifestyles. Consequently, the demand for different land use types varies, resulting in differences in the influencing factors. Landscape pattern changes are driven by a combination of natural, social, economic, and policy factors. In arid regions, due to similar natural geographical conditions and socio-economic development trajectories, the driving mechanisms exhibit a certain degree of overall similarity. With the introduction and implementation of policies such as new quality productivity under industrial restructuring after the New Crown Epidemic, as well as the influence of changes in demographic trends and other factors, the influencing factors and mechanisms for changes in the regional landscape pattern in the coming period will be different from those of today.

4.3. Differences in Landscape Pattern Changes Between Arid Regions and Coastal Areas

Wu et al. investigated the mechanisms of landscape pattern changes in the Beijing–Tianjin–Hebei urban agglomeration and found that, from 1990 to 2010 [60], the landscape patterns in this region underwent significant transformations, with clear temporal variations in the driving mechanisms across different landscape types. Compared to arid regions, the significant period of landscape change in the Beijing–Tianjin–Hebei urban agglomeration occurred earlier, and the factors influencing different landscape types also exhibited considerable differences. In the central and southeastern coastal developed regions of China, high vegetation coverage and a balanced distribution of landscape types are prominent features. These areas experienced earlier urbanization and higher development levels, with landscape changes primarily reflected in forest land and water bodies. In addition, processes such as land reclamation and river transportation have led to influencing factors that differ from those identified in this study [61,62]. Hainan, located at the northern edge of the tropics, experiences a tropical monsoon climate characterized by warm and humid natural conditions. These conditions foster unique landscape types and a relatively balanced landscape distribution, contributing to a thriving tourism industry. In this region, landscape changes are less influenced by natural factors and more significantly shaped by socio-economic factors, such as industrial structure and development [63]. In contrast, the UCS region, as a typical urban agglomeration in an arid zone, exhibits significant disparities in the distribution and proportion of land use landscape types. The industrial structure is predominantly based on agriculture and animal husbandry, and socio-economic development began relatively late. These factors collectively result in substantial differences in the quantitative structure, spatial distribution, change patterns, and driving factors of land use landscape types, compared to the developed regions of central and eastern China. This highlights the unique spatiotemporal characteristics of land use landscape changes in arid regions.

4.4. Restriction and Future Research

Due to limitations in the availability of long-term data, the spatial resolution of this study is constrained to 30 m, which limits the ability to conduct a more detailed analysis of urban core areas. The Geodetector model offers several advantages, including excellent compatibility with categorical data, the ability to identify multi-factor synergistic enhancements or nonlinear superposition effects through interaction detection, and suitability for multi-scale spatial pattern analysis. However, it can only characterize statistical correlations and is limited in its ability to verify causal mechanisms or analyze potential influencing factors. The PLUS model demonstrates significant advantages in fine-scale simulation and multi-scenario adaptability. However, parameters such as patch generation thresholds and seed point distribution have a substantial impact on the results, and subjective calibration biases may lead to unstable simulation outcomes. Additionally, more attention should be given to the setting of different development scenarios. Although multiple factors have been taken into account, the drivers of land use change are undoubtedly more complex. Future research should explore more diverse data sources to address the inherent limitations of the data and integrate other models to further analyze potential influencing factors. This will allow for the development of a more precise and robust comprehensive research framework, ultimately supporting the formulation of policies and plans that align with regional development goals based on medium- to long-term integrated research findings.

5. Conclusions

This study, based on Landsat series 30 m resolution remote sensing data from 1990, 2000, 2010, and 2020, integrates multi-source data and applies the random forest method for land use landscape type interpretation. Using theories and methods from landscape ecology, it reveals the spatiotemporal patterns of land use landscape changes in the UCS region. Natural, social, and economic factors were analyzed to explore the driving mechanisms and contributions across different periods. The main conclusions are as follows:

(1) The study area exhibits diverse landscape types, though their proportions vary significantly. From 1990 to 2020, the comprehensive land use dynamic degree was 0.22%, peaking during 2000–2010.

(2) At the class level, spatial heterogeneity and fragmentation increased, with human activities playing a growing role. Landscape shape complexity rose, leading to greater disorder. At the landscape level, types became more diversified and evenly distributed, with narrowing differences in patch proportions.

(3) Land use landscape changes were driven by a combination of natural, socio-economic, and policy factors. Single-factor analysis identified temperature, precipitation, DEM, GDP, and distances to government centers, railways, and highways as key contributors. Interaction detection revealed strong nonlinear enhancements, particularly between temperature and precipitation among natural factors, and among GDP, transportation accessibility, and administrative proximity among socio-economic factors. Thus, changes were the result of multiple interacting drivers rather than single factors.

(4) Under three development scenarios for 2030, unused land consistently decreased, with the largest reduction under the ecological protection scenario. Grassland area increased across all scenarios, most notably under ecological protection. Built-up land expanded in all scenarios, especially under economic development. Cropland also grew, with the highest increase under ecological protection. Forest land and water bodies declined slightly in all scenarios.

Future social development and ecological management should prioritize optimizing natural resource utilization, regulating ecological spaces, and establishing development boundaries to ensure the balanced and sustainable advancement of both economic and social systems.