Landscape Ecological Risk Assessment and Multi-Scenario Simulation of Land Use Based on the Markov-FLUS Model: A Case Study of the Hexi Corridor

Abstract

As a major ecological safeguard in northwestern China and an important corridor for the Belt and Road Initiative, the Hexi Corridor holds strategic significance for improving landscape structure and enhancing regional ecological security. Focusing on the Hexi Corridor, this study develops a landscape ecological risk (LER) index based on land use (LU) data from 2000, 2010, and 2020. The study employs ArcGIS spatial analysis and XGBoost-SHAP, an interpretable machine learning method, to analyze the spatiotemporal evolution of LU and LERs, as well as their driving factors. Furthermore, the Markov-FLUS model is utilized to simulate and predict LU and LER spatial patterns under multiple scenarios for 2030. The results show that: (1) The dominant land type in the Hexi Corridor is unused land, accounting for 67.33%. During the research period, the extents of unused land, grassland, and forestland showed a steady decline, while built-up land and cropland increased. (2) LERs are categorized into five types, with high risk being the most prevalent, accounting for 52.02%. Between 2000 and 2020, the total area of higher and high risks decreased by 4312 km², indicating an overall decrease in LER across the region. (3) LER is primarily influenced by annual rainfall, population density, distance to main roads, and distance to rivers. (4) Marked variations in LU patterns and LER are observed across different development scenarios projected for 2030.

1. Introduction

Landscape ecological risk (LER) denotes the likelihood that ecosystem structure becomes less stable and functional integrity is compromised under the combined effects of natural processes and human activities [1,2]. As a fundamental driver of landscape pattern formation, land use (LU) plays a central role in the assessment of LER through its dynamic changes [3,4]. This allows for the accurate characterization of regional LER spatial patterns, which is an effective means of identifying weak links in regional ecological security [5]. Such assessments are critical for strengthening ecological security barriers, promoting global ecological environment governance, and supporting high-quality economic development. Arid and semi-arid regions cover approximately 42% of the Earth’s land surface and support nearly 38% of the global population [6]. These areas are highly vulnerable due to their ecosystems’ fragility, making them hotspots for ecological degradation and LERs. The northwest region of China, as a significant part of the global arid and semi-arid zone, faces typical challenges such as arid climates, loose soil, weak disturbance resistance, and high ecological degradation risks [7,8]. Its ecological condition influences not only China’s ecological security but also plays a pivotal role in global material cycling and energy flows within arid ecosystems, thereby contributing to overall ecological balance [9]. The “Proposal of the Central Committee of the Communist Party of China for the 15th Five-Year Plan for National Economic and Social Development” clearly emphasizes the need to “strengthen ecological security barriers and enhance green development momentum” and to “improve ecological environment standards, monitoring, evaluation, and assessment systems.” This aligns with both China’s internal ecological protection needs and the global consensus on ecological governance and sustainable development.

As the green barrier against desertification and sandstorms in northwest China, the Hexi Corridor not only serves as the water conservation and ecological security area for the Qilian Mountains but also as a key region for implementing desertification prevention and comprehensive ecological projects like the “Three-North Shelterbelt” program [9,10]. Furthermore, it is a vital passage for East–West cultural exchanges under the Belt and Road Initiative. Therefore, evaluating and controlling its ecological risks is not only essential for the stability of regional ecological barriers but also influences global ecological governance in arid regions and the advancement of the Belt and Road green development initiative. The 14th Five-Year Plan for the Development of the Hexi Corridor Economic Belt identifies the substantial improvement of ecological civilization as a key objective and positions ecological construction as a lifeline to the economic development of the Hexi Corridor, thereby emphasizing the importance of regional ecological protection. This goal also aligns with global trends in ecological restoration and desertification control in arid regions. However, under the combined influence of global climate change and human activities, ecosystem vulnerability in the Hexi Corridor has intensified [6]. The region still faces widespread land desertification, water scarcity, and other ecological issues. The task of protecting and restoring oasis wetlands is daunting, and understanding the evolving patterns of LER and control pathways is an urgent and significant issue in global ecological governance in arid regions. Thus, conducting LER assessment and future scenario simulation research for the Hexi Corridor not only responds proactively to national policy requirements but also addresses the practical need for strengthening regional ecological security and supporting high-quality economic development. Moreover, this research provides a valuable case for LER assessment, driving mechanism analysis, and scenario simulations in the globally ecologically fragile arid and semi-arid regions, contributing to collaborative global ecological governance and helping the international community address ecological degradation in arid regions and achieve sustainable development goals [11,12].

In recent years, growing ecological pressures driven by global climate change and human activities have made ecological security a central focus in Geography, Ecology, and Sustainable Development research. Many scholars have delved into topics such as the construction of ecological security patterns [5,13,14], ecological environment quality assessment [15,16], ecosystem services trade-offs and synergies [17,18], ecological resilience impacts [19,20], and ecological governance [21]. For example, Huang examined the interaction mechanisms between LU changes and ecosystem services, suggesting that changes in LU significantly regulate ecological resilience [17]. Li et al., using LU data from their study area, developed an ecological security model to predict and assess the regional resilience of ecological security [5]. Similarly, LER assessment and scenario simulation prediction, key elements of ecological security management, have seen notable progress, especially in the development of evaluation indicators, identification of driving mechanisms, and prediction of future scenarios. Kang, Wang, and others constructed an LER index by combining indicators such as ecological vulnerability and loss degree to evaluate the level of regional LER [2,22]. In addition, Yang, Hu, and their team explored the driving mechanisms of LER in different regions [23]. Geographically, research tends to exhibit clear biases. In China, studies focus predominantly on regions with relatively favorable ecological conditions and well-established data, such as the Yangtze River Economic Belt, Yellow River Basin, Beijing-Tianjin-Hebei region, and Yangtze River Delta, as well as representative lake wetlands such as Poyang Lake and Dongting Lake [5,11,24]. Internationally, research often centers on ecological hotspots like the Amazon Rainforest, Southeast Asia’s tropical regions, and major river basins in Europe and North America [25,26,27]. Methodologically, scholars frequently adopt landscape index methods to develop LER assessment models, using models such as PLUS, FLUS, and CA-Markov for scenario simulation, or applying techniques like geographic detectors and LASSO regression to identify driving factors [6,26,28,29]. According to existing studies, LU change is a critical driver of regional ecosystem evolution, with adjustments in LU patterns and transformations between LU types affecting ecosystem structural stability, functional integrity, and ecosystem service provision [30,31]. Thus, evaluating regional LU changes is crucial for formulating urban development plans, optimizing LU policies, and achieving a balance between ecological protection and socio-economic development. However, notable gaps remain in the existing body of research. First, although extensive research has been conducted on arid and semi-arid regions in Africa, the Americas, and Australia, studies on LER in Asian arid regions remain limited, particularly in northwestern China. The Hexi Corridor, a pivotal node of the Belt and Road Initiative and a representative region in global arid zones, has not been given sufficient attention. Second, most research on arid regions focuses on singular ecological issues, such as water scarcity and desertification, neglecting comprehensive LER analysis based on LU. Third, although conventional linear and spatial analytical approaches are widely used to identify driving factors, relatively few studies apply machine learning methods to capture the nonlinear relationships between LER and its determinants. Despite Zhao Wenzhi, Lei Jiaqiang, and others recognizing the importance of the Hexi Corridor in reinforcing the ecological barrier of northwest China, and Li Ji’s preliminary analysis of its oasis social-ecological resilience, research on the LERs in the Hexi Corridor remains fragmented. A comprehensive study incorporating LU change, risk assessment, driving mechanisms, and scenario simulation has not yet been developed, which limits its ability to meet the region’s ecological governance needs.

Overall, the Hexi Corridor holds a prominent ecological position, yet existing research lacks a focus on its LU changes, LER assessment, driving factor identification, and future scenario simulations. Building on this context, this study takes the Hexi Corridor as a case study and aims to address the following research objectives. First, using LU grid data from 2000, 2010, and 2020, the study constructs LU transition matrices and calculates the LER index. Using spatial analysis tools in ArcGIS 10.7, this study examines the spatiotemporal dynamics of LU and LER. In addition, Kriging is applied to classify LER into distinct zones. Second, in contrast to traditional methods like geographic detectors, this study uses XGBoost-SHAP, an interpretable machine learning technique, to identify the driving factors of LER. This approach allows for accurately capturing the nonlinear impacts of these factors, while quantifying their contributions and the direction of their positive or negative effects. Lastly, using the Markov-FLUS model, the study simulates and predicts LU patterns and LER zones for 2030, considering three different scenarios: the natural development scenario, cropland protection scenario, and urban development scenario. Through these efforts, the study elucidates the spatiotemporal dynamics of LU and LER in the Hexi Corridor, while identifying their key driving factors. This not only provides valuable reference for the Chinese government in rationally planning LU and formulating differentiated ecological prevention and control strategies, but also offers a typical case and methodological insights for the LER evaluation and management of ecologically fragile regions in global arid and semi-arid zones. The findings aim to contribute to global ecological governance and the achievement of sustainable development goals.

2. Materials and Methods

2.1. Study Area

The Hexi Corridor is situated in the arid and semi-arid regions of northwestern China. It extends west of the Yellow River, bounded by the Qilian Mountains to the south and the Badain Jaran Desert and Tengger Desert to the north [6]. The region mainly comprises five cities in northwestern Gansu Province: Wuwei, Jinchang, Zhangye, Jiuquan, and Jiayuguan (Figure 1). The region not only has significant geographical advantages and profound cultural heritage, but also possesses a solid foundation for development. By 2024, the five cities within the Hexi Corridor had a total resident population of 4.31 million and a combined GDP of 346.34 billion CNY, accounting for 26.63% of the total GDP of Gansu Province. The per capita GDP was 27,500 CNY higher than the provincial average. The total import and export value of goods reached 29.027 billion CNY, making up 47.38% of the province’s total, positioning the Hexi Corridor as a crucial growth pole for Gansu’s economic development and external openness. Promoting ecological conservation alongside high-quality economic growth in the Hexi Corridor is essential not only for achieving sustainable regional development, but also for reinforcing the ecological security barrier in western China.

2.2. Data Sources

The LU data used in this study were sourced from the Geospatial Data Cloud, with a spatial resolution of 1 km × 1 km and a consistent coordinate system of WGS 1984 UTM Zone 48N. A reclassification procedure was employed to group the data into six landscape categories: cropland, forestland, grassland, water body, built-up land, and unused land. The study boundary data and vector data for driving factors were obtained from the National Geomatics Center of China (https://www.webmap.cn/, accessed on 1 January 2026), the Baidu Maps Coordinate Picker (https://lbs.baidu.com/, accessed on 3 January 2026), the National Geomatics Center of China (https://www.ngcc.cn/, accessed on 6 January 2026), and the Resource and Environment Science Data Center (https://www.resdc.cn/, accessed on 11 January 2026).

2.3. Research Methods

2.3.1. LER Assessment

First, the total landscape area and number of patches in the study area were calculated using Fragstats 4.2, and the mean patch area was derived accordingly [32,33]. Based on the empirical principle that the grid scale should be two to five times the mean patch area, the suitable grid size was determined to range from 6000 to 15,000 hectares [34,35]. Considering computational efficiency, the study area was ultimately divided into 4167 regular grids of 8 km × 8 km. Fragstats was then used to calculate the area and patch number for each LU type. The results were further processed in Excel to derive landscape metrics, including fragmentation, separation, dominance, disturbance, vulnerability, and loss indices (Table 1). Finally, the LER index was calculated according to Equation (1).

$$ER{I}_i={\displaystyle \sum _{i=1}^N\frac{A_{ki}}{A_k}}{R}_i$$

(1)

where ERI_i is the LER index for the i-th grid; A_ki is the area of the i-th landscape type in the k-th grid, and A_k is the total area of the k-th grid; R_i is the loss degree index for the i-th landscape type.

Landscape disturbance captures the degree to which ecosystems, as represented by various landscape types, respond to external stressors [36]. In the construction of landscape disturbance, the weights of fragmentation, separation, and dominance are set at 0.5, 0.3, and 0.2, respectively [9,37]. This weighting scheme is derived from existing studies on LER assessment in the Hexi Corridor and is further justified by the ecological characteristics of the study area. The ecosystems of the Hexi Corridor are highly fragile and sensitive. Landscape fragmentation exerts a particularly strong influence on ecological processes. On the one hand, human activities such as road construction, oasis development, and urban expansion intensify landscape fragmentation, breaking originally continuous natural patches into smaller units [38]. On the other hand, under arid climatic conditions, ecosystem recovery is limited. Once fragmentation occurs, its negative effects on ecological functions become more pronounced [39]. Therefore, fragmentation is assigned the highest weight. Separation reflects the spatial isolation among patches of the same landscape type [40]. It affects ecological connectivity and the exchange of matter and energy. The Hexi Corridor exhibits a typical “oasis–desert” pattern. Landscape patches are naturally discrete. However, excessive separation weakens ecological corridor functions and increases LER. Thus, separation is given the second-highest weight. By contrast, dominance captures the controlling effect of dominant landscape types on overall patterns [4]. The landscape in the Hexi Corridor is relatively homogeneous, with unused land being predominant. Changes in dominance have a comparatively limited marginal effect on ecosystem stability. Therefore, a lower weight is assigned to this factor.

Landscape vulnerability describes the ability of different landscape types to withstand external disturbances as well as their sensitivity to environmental changes [41]. A higher vulnerability index indicates weaker resistance and a greater likelihood of ecosystem degradation [42]. In contrast, a lower value suggests a relatively stable landscape structure. In constructing landscape vulnerability, this study builds on existing LER assessment studies in the Hexi Corridor and assigns values of 4, 2, 3, 5, 1, and 6 to the six landscape types, respectively, based on the ecological characteristics of the study area [9,37]. Unused land accounts for the largest proportion of the study area. It mainly consists of Gobi and desert landscapes, characterized by low vegetation cover and loose soil structure. Ecosystem stability is weak. Under arid climatic conditions and frequent aeolian processes, these areas are highly sensitive to disturbance [38]. Once affected by human activities, degradation occurs easily and recovery is slow. Therefore, unused land is assigned the highest vulnerability. Water body constitutes a fundamental component of arid ecosystems and plays a vital role in sustaining regional ecological security [43]. However, water resources in the Hexi Corridor are limited and unevenly distributed. The ecosystem is highly sensitive to changes in water availability. Shrinkage or pollution can trigger cascading ecological risks. Thus, water body is assigned a relatively high vulnerability. Cropland and grassland are mainly distributed in oases and their margins. They are strongly influenced by human activities. Cropland relies on irrigation and is subject to frequent surface disturbance. Risks such as soil salinization and degradation are common [32]. Grassland, dominated by desert steppe and saline meadow, shows some resilience but is vulnerable to overgrazing and water scarcity [26]. Both are therefore assigned moderate vulnerability, with cropland slightly higher than grassland. Forestland is primarily composed of artificial shelterbelts and natural shrublands. Its vegetation structure is relatively stable, and it exhibits strong capacities for windbreak and sand stabilization, as well as water conservation [29]. Consequently, it is characterized by relatively low vulnerability. Built-up land is protected by artificial infrastructure and has a high degree of surface sealing [44]. It is the least affected by natural disturbances and exhibits the highest stability. Therefore, it is assigned the lowest vulnerability.

Table 1. Calculation Formula for LER Index.

Number	Index	Formula	Meaning
1	Landscape Fragmentation	$C_i=n_i/A_i$	ni represents the number of patches of landscape type i; Ai represents the area of landscape type i [7,29].
2	Landscape Isolation	$N_i={\displaystyle \frac{A}{2A_i}}\sqrt{{\displaystyle \frac{n_i}{A}}}$	A represents the total area of the landscape [11].
3	Landscape Dominance	$D_i={\displaystyle \frac{Q_i+M_i}{4}}+{\displaystyle \frac{L_i}{2}}$	Qi represents the number of patches i appearing in the sample plots/total sample plots; Mi represents the number of patches i/total patches; Li represents the area of patch i/total sample plot area [40].
4	Landscape Disturbance	$E_i=a{C}_i+b{N}_i+c{D}_i$	a, b, and c represent the weights of fragmentation, isolation, and dominance, respectively, with values assigned as 0.5, 0.3, and 0.2 based on existing literature [9,37].
5	Landscape Vulnerability	Normalized data	The six landscape types are assigned the values 4, 2, 3, 5, 1, and 6, respectively, and then normalized to obtain the vulnerability of each landscape type [9,37].
6	Landscape Loss Degree	$R_i=E_i*F_i$	Ri represents the product of the disturbance degree and the loss degree [36].

Table 1. Calculation Formula for LER Index.

Number	Index	Formula	Meaning
1	Landscape Fragmentation	$C_i=n_i/A_i$	ni represents the number of patches of landscape type i; Ai represents the area of landscape type i [7,29].
2	Landscape Isolation	$N_i={\displaystyle \frac{A}{2A_i}}\sqrt{{\displaystyle \frac{n_i}{A}}}$	A represents the total area of the landscape [11].
3	Landscape Dominance	$D_i={\displaystyle \frac{Q_i+M_i}{4}}+{\displaystyle \frac{L_i}{2}}$	Qi represents the number of patches i appearing in the sample plots/total sample plots; Mi represents the number of patches i/total patches; Li represents the area of patch i/total sample plot area [40].
4	Landscape Disturbance	$E_i=a{C}_i+b{N}_i+c{D}_i$	a, b, and c represent the weights of fragmentation, isolation, and dominance, respectively, with values assigned as 0.5, 0.3, and 0.2 based on existing literature [9,37].
5	Landscape Vulnerability	Normalized data	The six landscape types are assigned the values 4, 2, 3, 5, 1, and 6, respectively, and then normalized to obtain the vulnerability of each landscape type [9,37].
6	Landscape Loss Degree	$R_i=E_i*F_i$	Ri represents the product of the disturbance degree and the loss degree [36].

2.3.2. XGBoost-SHAP Interpretable Machine Learning

XGBoost is an ensemble learning method characterized by efficient training and strong capacity to model nonlinear relationships [45,46]. However, its inherent “black-box” nature limits interpretability, making it challenging to quantify both the magnitude and direction of individual driving factors influencing LER. To overcome this limitation, this study incorporates SHAP for interpretability analysis, allowing for a clearer understanding of the driving factors behind LER [46,47]. The formulation is given as follows:

$$Obj={\displaystyle \sum _{i=1}^{n}l}(y_i,{\tilde{y}}_i)+{\displaystyle \sum _{k=1}^k\Omega }(f_k)$$

(2)

$$f(x)=E\left[f(x)\right]+{\displaystyle \sum _{i=1}{\phi }_i}$$

(3)

where ${\displaystyle \sum _{i=1}^{n}l}(y_i,{\tilde{y}}_i)$ is the training loss; ${\displaystyle \sum _{k=1}^k\Omega }(f_k)$ is the regularization term; $f(x)$ is the model’s prediction for sample x; $E\left[f(x)\right]$ is the baseline value (average prediction); and ${\displaystyle \sum _{i=1}{\phi }_i}$ is the SHAP value for feature i.

2.3.3. Markov-FLUS Model

The Markov-FLUS model is designed for LU simulation and projection and is based on the Markov chain approach [26,48]. It comprises two core components: an Artificial Neural Network (ANN) module and a Cellular Automata (CA) module [26]. The ANN module is capable of producing LU suitability probability maps for a given year by integrating LU data with corresponding driving factors [29]. The CA module integrates LU data with suitability probability maps to simulate and project LU scenarios under different conditions. This process is driven by key parameters, including the number of future LU categories, conversion cost matrices, and neighborhood weights [29]. The formulas are as follows:

$$S_{t+1}=P_{ij}*S_t$$

(4)

$$T{P}_{p,k}^t=P_{p,k}\times {\Omega }_{p,k}^t\times I_k^t\times S{C}_{c\to k}$$

(5)

where $S_{t+1}$ and $S_t$ is the area of each landscape type at time t + 1 and time t, respectively; $P_{ij}$ is the transition probability matrix for the transformation of landscape type i into landscape type j; $T{P}_{p,k}^t$ is the overall transition probability for cell p from the original landscape type to the k-th landscape type at time t; $P_{p,k}$ is the development probability of cell p transitioning to the k-th landscape type; ${\Omega }_{p,k}^t$ is the neighborhood influence factor; $I_k^t$ is the adaptive driving coefficient; and $S{C}_{c\to k}$ is the conversion cost for transforming landscape type c into landscape type k.

2.4. Technical Framework

To clearly present the research logic, the technical roadmap is illustrated in Figure 2.

3. Results

3.1. Spatiotemporal Evolution of LU and LER

3.1.1. Spatiotemporal Evolution of LU

The LU structure of the Hexi Corridor in 2000, 2010, and 2020 is shown in

Table 2. LU Structure for 2000, 2010, and 2020.

LU Type	2000		2010		2020
	Area/km2	Proportion/%	Area/km2	Proportion/%	Area/km2	Proportion/%
Cropland	14,402	5.76%	15,523	6.21%	15,996	6.40%
Forestland	7467	2.99%	7432	2.97%	7433	2.97%
Grassland	53,782	21.52%	53,794	21.53%	53,655	21.47%
Water body	2696	1.08%	2673	1.07%	2776	1.11%
Built-up land	1175	0.47%	1236	0.49%	1783	0.71%
Unused land	170,372	68.18%	169,236	67.72%	168,251	67.33%

. Among these, unused land is the dominant landscape type, accounting for over 67%. The second largest area is grassland, comprising over 21%, while built-up land has the smallest area, less than 1%. During the study period, the LU structure underwent substantial changes, with unused land, cropland, and built-up land exhibiting the greatest fluctuations in area. Specifically, cropland increased by 1594 km², rising to 6.40% of the total area; built-up land increased by 608 km², reaching 0.71%; water body increased by 80 km², accounting for 1.11%; unused land decreased by 2121 km², dropping to 67.33%; grassland decreased by 127 km², reducing to 21.47%; and forestland decreased by 34 km², accounting for 2.97%.

Further analysis of LU spatial patterns and transition trends is presented in Figure 3. Figure 3a–c illustrates the LU spatial patterns in the Hexi Corridor for 2000, 2010, and 2020. Unused land is primarily concentrated in the eastern and northern regions, with the largest distribution found in Jiuquan and Wuwei cities. Forestland and grassland are concentrated along the southern boundary, with the largest distribution in Zhangye city. Cropland and built-up land are mainly located in the central area, distributed across all five cities. Figure 3d shows the LU transition trends in the Hexi Corridor. From 2000 to 2010, the total transition area was 1944 km². Among these, unused land transitioned most significantly to cropland and grassland, with areas of 784 km² and 376 km², respectively. Cropland transitioned to forestland and built-up land, with areas of 81 km² and 16 km², respectively. Grassland transitioned to cropland and built-up land, with areas of 398 km² and 8 km², respectively. From 2010 to 2020, the total transition area was 1547 km². In this period, unused land transitioned most significantly to cropland and built-up land, with areas of 393 km² and 414 km², respectively. Cropland and grassland transitioned to built-up land, with areas of 81 km² and 46 km², respectively.

3.1.2. Spatiotemporal Evolution of LER

The LER structure of the Hexi Corridor in 2000, 2010, and 2020 is shown in

Table 3. LER Structure of the Study Area in 2000, 2010, and 2020.

Risk Types	2000		2010		2020
	Area/km2	Proportion/%	Area/km2	Proportion/%	Area/km2	Proportion/%
Low Risk	32,987	13.20%	34,139	13.66%	34,907	13.97%
Lower Risk	18,739	7.50%	19,251	7.70%	19,507	7.80%
Moderate Risk	23,670	9.47%	23,542	9.42%	25,294	10.12%
Higher Risk	41,639	16.66%	40,956	16.39%	40,213	16.09%
High Risk	132,906	53.17%	132,053	52.83%	130,020	52.02%

. Among these, high risk is the prime risk type, comprising more than 52% of the total, followed by higher risk at over 16%, with lower risk being the smallest, at less than 10%. Over the study period, the structure of LER exhibited substantial changes. The areas of higher and high risk decreased, while the areas of moderate risk and lower risk increased, indicating an overall improvement in LER. Specifically, the areas of high risk and higher risk reduced by 2885 km² and 1426 km², respectively, with their proportions dropping to 52.02% and 16.09%. Meanwhile, the areas of moderate risk, lower risk, and low risk increased by 1624 km², 768 km², and 1920 km², respectively, with their proportions rising to 10.12%, 7.8%, and 13.97%.

Further analysis of the LER spatial patterns and transition trends is shown in Figure 4. Figure 4a–c shows the LER spatial patterns of the Hexi Corridor for 2000, 2010, and 2020. Areas characterized by high and higher risk are mainly clustered in the eastern and northern parts of the study region, with the most extensive distributions observed in Jiuquan and Wuwei. Lower- and low-risk areas are concentrated along the southern boundary, with the largest distributions found in Zhangye and Wuwei cities. Moderate risk areas are primarily concentrated in the central part of the study region and are distributed across all five cities. Figure 4d shows the LER transition trends in the Hexi Corridor. From 2000 to 2020, the total transition area was 88,885 km². Among these, the largest transitions from high risk occurred to higher and moderate levels, with areas of 15,391 km² and 5738 km², respectively. The largest transitions from higher risk occurred to moderate and higher levels, with areas of 6462 km² and 17,357 km², respectively. For moderate risk, the largest transitions were to higher and high levels, with areas of 4831 km² and 6437 km², respectively. Lower risk transitioned mostly to low and moderate levels, with areas of 3816 km² and 3619 km², respectively. Low risk transitioned mainly to higher and moderate levels, with areas of 3980 km² and 1737 km², respectively.

3.2. Driving Factors of LER

To examine the effects of both natural geographic and socio-economic factors on LER in the Hexi Corridor, ten secondary indicators were selected in accordance with the specific characteristics of the study area [10,49]. Among these, natural geographic factors include elevation (X1), slope (X2), annual rainfall (X3), average annual temperature (X4), and distance to rivers (X5). Socio-economic factors include population density (X6), per capita GDP (X7), nighttime light (X8), distance to main roads (X9), and distance to county government offices (X10). First, to eliminate the potential interference of multicollinearity on model estimation, this study conducts VIF tests on ten driving factors. The results show that all variables have VIF values below 10, indicating no severe multicollinearity and satisfying the requirements for model construction [50,51]. On this basis, Pearson correlation analysis is further employed to examine the linear relationships among variables. As shown in Figure 5, all correlation coefficients are below 0.8, which again confirms the absence of significant multicollinearity [50,52]. Second, to test the spatial dependence of LER, this article applies Global Moran’s I for spatial autocorrelation analysis. The results indicate that Moran’s I = 0.000208 (p = 0.968), which is not statistically significant. This indicates that LER does not exhibit significant global spatial autocorrelation [53]. Therefore, no additional controls for spatial dependence are introduced in subsequent modeling. To ensure model generalization and the reliability of the results, five-fold cross-validation was performed on the training set to evaluate model stability. The results show that the mean R² is 0.7060 (±0.0328) and the mean RMSE is 0.0045 (±0.0002), indicating strong robustness across different data subsets. Based on these results, the final model was trained using the full training dataset. It achieved an RMSE of 0.0045 and an R² of 0.7200 on the test set, which is highly consistent with the average performance of cross-validation. The results indicate that the model exhibits no obvious overfitting, and the discrepancy between predicted and actual ecological risk values is relatively small [45]. The model can explain 72% of the variance in landscape ecological risk, demonstrating satisfactory goodness of fit and predictive accuracy, and meets the modeling requirements and data conditions of this study.

The feature dependency diagram and contribution results shown in Figure 6 indicate that annual rainfall, population density, distance to main roads, and distance to rivers are the main driving factors, with contributions of 29.7%, 25.8%, 10.8%, and 10.7%, respectively. Following these, distance to county government offices, slope, elevation, and average annual temperature contribute 6.6%, 5.5%, 4.4%, and 4.0%, respectively. Per capita GDP and nighttime light have a relatively minor impact on LER, contributing less than 3%. Thus, it can be concluded that natural geographic factors are the primary influences on LER in the Hexi Corridor, with water resource abundance, particularly represented by annual rainfall and distance to rivers, being the core driving force. This aligns with the region’s “oasis-desert” natural base. From the perspective of natural ecological processes, water resources are the key constraint on landscape pattern evolution. The Hexi Corridor is characterized by low precipitation and intense evaporation, resulting in ecosystems that remain under persistent water limitation [54]. In recent years, meltwater supply from the Qilian Mountains has remained relatively stable. In addition, unified watershed water allocation and the promotion of water-saving irrigation technologies have improved water-use efficiency in downstream oasis areas and enhanced ecological water security. These changes have increased soil moisture and vegetation cover, effectively curbing desertification expansion and contributing to a decline in LER.

In terms of human activities, regulated LU and ecological restoration are important drivers of landscape evolution. In the Hexi Corridor, farmland reclamation and urban construction are mainly concentrated in oasis areas, which reduces the disorderly expansion into ecologically sensitive margins [10]. At the same time, ecological restoration measures, such as sand control engineering, afforestation, and enclosure for grassland recovery, have been continuously implemented. These efforts have strengthened the ecological buffering capacity at oasis edges, ensured oasis stability, and effectively reduced LER. From the perspective of governance, ongoing ecological projects and regional regulatory policies play a critical role in guiding landscape evolution [6]. Initiatives such as comprehensive desertification control and the Three-North Shelterbelt Program have promoted the restoration of degraded ecosystems. Moreover, regulatory instruments, including ecological redline policies and territorial spatial planning, have effectively curbed irrational development. These measures provide long-term institutional support for reducing LER. Therefore, the reduction in LER in the Hexi Corridor cannot be attributed to a single factor; rather, it reflects the combined effects of multiple interacting mechanisms, including water resource constraints, human activity responses, and policy regulation. This process highlights a broader transition in arid regions, from predominantly natural control to a coupled human–environment regulatory system.

Further analysis examines the nonlinear effects of four key driving factors on LER (Figure 7). Overall, all variables exhibit pronounced nonlinear relationships and clear threshold effects. X3 shows a distinct threshold pattern. At low precipitation levels, SHAP values are positive, indicating that water scarcity exacerbates LER. This is consistent with the semi-arid conditions of the Hexi Corridor, where long-term water constraints lead to low vegetation cover and weak ecosystem resilience. Once precipitation exceeds the threshold of 0.33, SHAP values turn negative. This indicates a shift from a risk-enhancing to a risk-mitigating effect. Increased water availability promotes vegetation recovery and improves ecosystem stability, thereby reducing LER. X6 has an overall negative effect on LER, with a gradually diminishing magnitude. Mechanistically, the population tends to concentrate in oasis core areas. This concentration increases investments in ecological projects per unit area, including water-saving irrigation, shelterbelt construction, and land consolidation. As a result, human activities shift from passive adaptation to active regulation, which helps reduce ecological risk. As population density continues to increase, the negative effect weakens, indicating diminishing marginal effects [30]. Therefore, in the study area, population density primarily reflects adaptive agglomeration in ecologically favorable regions, rather than simple human disturbance intensity. X5 also exhibits a clear threshold effect. In areas close to major roads, SHAP values are negative. This suggests that high accessibility is associated with better infrastructure and stronger ecological management, which helps reduce LER. When the distance exceeds the threshold of −0.70, SHAP values become positive and remain at a relatively low level. This indicates a transition from human-dominated areas to desert or Gobi regions. In these areas, human intervention weakens, ecosystem stability declines, and LER increases [55,56]. This finding suggests that accessibility influences LER indirectly by shaping the intensity of human activities and ecological regulation capacity. X9 shows a similar threshold pattern. In areas close to rivers, SHAP values are negative, indicating that rivers act as key ecological corridors in arid regions and play a critical role in reducing LER. When the distance exceeds the threshold of −0.49, SHAP values turn positive. This reflects a rapid decline in water availability, reduced vegetation cover, and increased ecosystem vulnerability, all of which contribute to higher LER [12]. This process highlights the fundamental role of water resources in maintaining regional ecological stability.

3.3. Multi-Scenario LU and LER Simulation and Prediction

3.3.1. Model Accuracy Verification

To ensure the reliability of LU simulation and projection across scenarios, model accuracy was rigorously validated [33]. First, using LU data from 2000 and 2010 and ten driving factors, the Markov module in the FLUS model was applied to estimate the demand for six LU types in 2020 [32]. Then, based on the predicted demand, neighborhood weights, and land conversion cost matrices, the CA module was used to simulate the 2020 LU spatial pattern [29]. Finally, observed LU data for 2020 were compared with simulated results to calculate the Kappa coefficient and overall accuracy [3]. A Kappa value closer to 1 indicates higher accuracy, and values above 0.8 are generally considered satisfactory. The results show a Kappa coefficient of 0.952 and an overall accuracy of 0.976, indicating strong model performance and supporting the reliability of the Markov–FLUS framework for LU simulation and projection.

3.3.2. Multi-Scenario LU Simulation and Prediction

Given that LU change is jointly influenced by natural conditions and policy regulation, this study develops future LU scenarios for the Hexi Corridor according to the binding indicators and development goals outlined in the Gansu Province Territorial Spatial Planning (2021–2035) and the 14th Five-Year Plan for the Economic Belt of the Hexi Corridor. Specifically, the goals include maintaining cropland area at no less than 77.03 million mu by 2035, achieving an urbanization rate of over 60% for the permanent population, and optimizing urban layout. Accordingly, the future LU scenarios were set as the natural development scenario, cropland protection scenario, and urban development scenario [2,57].

Before the simulation, key parameters were specified. In the ANN module, uniform sampling, single-precision floating-point, and non-normalization were adopted, with a sampling rate of 10 and 12 neurons in the hidden layer. In the CA module, the maximum iterations, neighborhood size, and acceleration coefficient were set to 300, 3, and 0.1, respectively. Nature reserves in the Hexi Corridor were defined as restricted areas. The conversion cost matrix and neighborhood weights are reported in

Table 4. LU conversion cost matrix and neighborhood weights.

LU Type	Cropland	Forestland	Grassland	Water Body	Built-Up Land	Unused Land
Cropland	1,1,1	1,0,1	1,0,1	1,0,1	1,0,1	1,0,1
Forestland	1,1,1	1,1,1	1,1,1	0,0,0	1,0,1	1,0,0
Grassland	1,1,1	1,1,1	1,1,1	1,1,1	1,0,1	1,0,1
Water body	0,1,1	0,0,0	1,1,1	1,1,1	0,0,1	0,0,0
Built-up land	1,1,0	1,1,0	1,1,0	0,0,0	1,1,1	1,0,0
Unused land	1,1,1	1,1,1	1,1,1	1,1,1	1,0,1	1,1,1
Neighborhood weights	0.176	0.062	0.02	0.238	0.491	0.012

, with weights determined based on existing literature and local conditions [6,36]. The three parameter sets correspond to the natural development, cropland protection, and urban development scenarios. A value of 1 indicates allowed LU conversion, whereas 0 indicates restriction [29]. Scenario-specific adjustments were implemented based on policy requirements. Under the cropland protection scenario, the transition probability from cropland to built-up land was reduced by 70%, and those to grassland and water body by 40%, while the probability of converting unused land to cropland was increased by 50% [32]. Under the urban development scenario, the transition probabilities from built-up land to cropland, forest land, grassland, water body, and unused land are reduced by 30% [31]. At the same time, the probabilities of converting cropland, forest land, and grassland into built-up land are increased by 20%, in line with urban expansion objectives [26,32]. No additional adjustments are applied under the natural development scenario.

Changes in LU area and proportion for 2030 under different scenarios are shown in

Table 5. LU Area Under Multi-Scenario Simulation.

LU Type	Natural Development Scenario		Cropland Protection Scenario		Urban Development Scenario
	Area/km2	Proportion/%	Area/km2	Proportion/%	Area/km2	Proportion/%
Cropland	15,965	6.39%	16,122	6.45%	15,966	6.39%
Forestland	7434	2.97%	7434	2.97%	7434	2.97%
Grassland	53,517	21.42%	53,926	21.58%	53,507	21.41%
Water body	2876	1.15%	2875	1.15%	2876	1.15%
Built-up land	1847	0.74%	1781	0.71%	1973	0.79%
Unused land	168,255	67.33%	167,756	67.13%	168,138	67.28%

. Under the natural development scenario, compared to 2020, cropland and grassland areas decrease by 31 km² and 138 km², respectively, with their proportions dropping to 6.39% and 21.42%. Water body and built-up land areas increase by 100 km² and 64 km², respectively, with their proportions rising to 1.15% and 0.74%. Under the cropland protection scenario, cropland, grassland, and water body areas increase by 126 km², 271 km², and 99 km², respectively, with their proportions increasing to 6.45%, 21.58%, and 1.15%. Built-up land and unused land areas decrease by 2 km² and 495 km², respectively, with their proportions dropping to 0.71% and 67.13%. Under the urban development scenario, cropland, grassland, and unused land areas will reduce by 30 km², 148 km², and 113 km², respectively, with their proportions dropping to 6.39%, 21.41%, and 67.28%. Water body and built-up land areas will increase by 100 km² and 190 km², with their proportions rising to 1.15% and 0.79%. Forestland area and proportion show no significant changes across the three scenarios.

The spatial patterns of LU across the three scenarios are presented in Figure 8. Using ArcGIS regional analysis tools, this study derives the spatial allocation of different LU categories at the city level. For LU types, under the natural development scenario, the city with the highest proportion of cropland area to total cropland area is Wuwei, at 40.05%. Within the urban development scenario, Zhangye exhibits the largest share of forestland relative to the total forest area, accounting for 53.18%. In the cropland protection scenario, Jiuquan accounts for the largest shares across multiple LU categories, including grassland, water bodies, built-up land, and unused land, representing 52.68%, 61.29%, 30.11%, and 80.48% of their respective totals. At the city scale, the highest proportions of unused land area in each city are found in Jiayuguan, Jinchang, Wuwei, and Jiuquan. Under the natural development Scenario, these proportions are 50.00%, 46.70%, 44.94%, and 79.67%, respectively. The second-highest proportion is grassland, with the highest proportions under the cropland protection scenario being 26.87%, 28.45%, 26.01%, and 16.71%, respectively, for the four cities. Zhangye has the highest proportion of grassland area to total city area, followed by unused land. Within the urban development scenario, grassland and unused land account for 37.62% and 36.31% of the total area, respectively.

3.3.3. Multi-Scenario LER Simulation and Prediction

Based on the LU simulation predictions for 2030 under each scenario, the LER Index was recalculated for each scenario. Kriging interpolation was then used to delineate the LER zones of the study area. The results are shown in

Table 6. LER Area Under Multi-Scenario Simulation.

Risk Types	Natural Development Scenario		Cropland Protection Scenario		Urban Development Scenario
	Area/km2	Proportion/%	Area/km2	Proportion/%	Area/km2	Proportion/%
Low Risk	35,514	14.21%	35,706	14.29%	34,420	13.77%
Lower Risk	18,798	7.52%	18,670	7.47%	19,387	7.76%
Moderate Risk	25,589	10.24%	25,845	10.34%	25,375	10.15%
Higher Risk	40,546	16.22%	40,484	16.20%	41,516	16.61%
High Risk	129,478	51.81%	129,219	51.70%	129,226	51.71%

. Under the natural development scenario, the areas of high risk and low risk will decrease by 542 km² and 709 km², respectively, with proportions decreasing by 0.21% and 0.28%. Meanwhile, the areas of higher risk, moderate risk, and low risk will increase by 333 km², 295 km², and 607 km², respectively, with proportions increasing by 0.13%, 0.12%, and 0.24%. Within the cropland protection scenario, the extents of both high-risk and low-risk areas are projected to decline by 801 km² and 837 km², respectively, corresponding to reductions of 0.32% and 0.33% in their shares. Conversely, the areas of higher risk, moderate risk, and low risk will increase by 271 km², 551 km², and 799 km², with proportions increasing by 0.11%, 0.22%, and 0.32%. Within the urban development scenario, the extents of high-, lower-, and low-risk areas are expected to contract by 794 km², 120 km², and 487 km², respectively, corresponding to proportional declines of 0.31%, 0.05%, and 0.19%. Concurrently, the extents classified as higher risk and moderate risk are projected to expand by 1303 km² and 81 km², respectively, corresponding to increases of 0.52% and 0.03% in their shares.

Figure 9 illustrates the spatial configuration of LER across the three scenarios. Similarly, using ArcGIS regional analysis tools, the spatiotemporal distribution of each land type in the cities was calculated. In terms of LER types, low risk is primarily concentrated in Wuwei and Zhangye, with their proportions reaching 38.64% and 42.85%, respectively, under the urban development scenario. Lower risk is mainly concentrated in Zhangye and Jiuquan, with proportions of 35.67% and 42.05% under the same scenario. Moderate risk is primarily found in Zhangye and Jiuquan, with proportions of 22.57% and 63.85%, respectively, under the natural development scenario. Higher and high risks are concentrated in Jiuquan, with proportions of 72.36% and 86.43% under the urban development scenario. At the city scale, Jiayuguan is mainly characterized by higher and lower risks, with their proportions reaching 29.44% and 32.91% under the natural development scenario. Jinchang is mainly characterized by low and higher risks, with proportions of 23.26% and 26.31% under the urban development scenario. Jiuquan is dominated by higher and high risks, with proportions of 17.18% and 65.83%, respectively, within the natural development scenario. Wuwei is characterized by high and low risks, with proportions of 31.52% and 42.60% within the urban development scenario. Zhangye is dominated by lower and low risks, with proportions of 17.24% and 39.62% under the urban development scenario.

4. Discussion and Conclusions

4.1. Discussion

4.1.1. Analysis of Spatiotemporal Evolution of LU and LER

The patterns of spatiotemporal change in LU and LER identified for the Hexi Corridor in this study are broadly consistent with prior research on oasis systems in arid regions. From the perspective of LU types, during the study period, significant changes occurred in LU in the Hexi Corridor, primarily marked by a reduction in unused land and an increase in cropland and built-up land [6,28]. This trend closely matches the recent expansion of oasis agriculture and accelerated urbanization in the northwestern arid region. It reflects the ongoing implementation of national western development strategies and land rehabilitation policies and highlights the rigid demand for land resources driven by regional urbanization, industrialization, and agricultural modernization [58,59]. Particularly under the backdrop of the “Green Belt and Road Initiative,” the Hexi Corridor, as a key gateway for northwest China’s opening-up, requires coordinated efforts between economic development and ecological security. From a theoretical perspective, this study reveals the characteristics and evolutionary patterns of LU in arid regions from a dynamic viewpoint. It enriches the understanding of human–environment interactions in oasis areas. From a practical perspective, the findings indicate that future regional development should prioritize stricter boundary regulation alongside structural adjustments in the transformation of unused land.

With respect to LER, high-risk areas constitute the predominant LER category in the Hexi Corridor. This is in line with the ecological foundation of the region, where “oases exist where there is water, and deserts prevail where there is none.” High-risk zones are primarily located in areas with a high proportion of unused land, where simple landscape configurations and weak disturbance resistance give rise to elevated LER. During the study period, LER underwent significant changes. Although the area of high risk has gradually decreased during the study period, it still accounts for over 50%, indicating that LER remains a serious issue in the region. This trend objectively reflects the results of ecological protection projects such as sand control, desertification prevention, and oasis restoration, while also emphasizing the vulnerability of ecosystems in arid areas and the urgency of strengthening ecological protection [60,61]. From a theoretical perspective, these results reveal a typical pattern in the evolution of LER in arid regions. Under inherently fragile ecological conditions, the risk pattern is characterized by a “high baseline with gradual improvement.” Ecosystem recovery shows clear stage-based and long-term dynamics. This insight deepens the understanding of ecosystem stability and recovery mechanisms in arid environments. From a practical perspective, the findings suggest that regional ecological governance should shift from short-term, project-driven approaches to long-term, system-oriented management. Priority should be given to areas with concentrated high-risk levels. Continuous efforts in ecological restoration and protection are required. It is also important to avoid overestimating localized improvements while the overall risk level remains high.

4.1.2. Analysis of LER Drivers

In contrast to the geographic detector approach adopted by Sha et al., this study utilizes an interpretable machine learning framework that integrates XGBoost with SHAP to identify the key determinants of LER [62,63]. This method not only accurately quantifies the contribution of each driving factor but also reveals their non-linear impact on LER. The results show that the primary drivers of LER in the Hexi Corridor are natural geographical factors. This conclusion aligns with Wang et al.’s findings but contrasts with the view of Yang et al., who argued that “human factors have a significantly greater driving force on LER than natural factors” [22,55]. The discrepancy in conclusions arises because Yang et al.’s study areas were mostly population-dense and economically active regions, where socio-economic factors significantly outweigh natural geographical factors in their impact on ecosystems. By contrast, in the Hexi Corridor—a representative ecologically fragile region within arid and semi-arid environments—water resource endowment is closely associated with LER, thereby rendering the constraining influence of natural factors more pronounced. From a theoretical perspective, these results indicate that the driving mechanisms of LER exhibit significant regional heterogeneity. It is therefore inappropriate to simply apply a generalized framework in which human activities dominate LER. The differences in dominant driving factors fundamentally reflect variations in the ecological baseline and human–environment interactions across regions. This finding extends the theoretical understanding of LER formation in arid contexts. From a practical perspective, this implies that in arid regions such as the Hexi Corridor, LER management should prioritize natural constraints, particularly water resource allocation and use efficiency, rather than relying solely on restricting human activities. This approach provides a more targeted basis for regional ecological governance and resource management.

Among natural geographic factors, annual rainfall and distance to rivers are the dominant drivers. In the Hexi Corridor, unused land accounts for more than 50% of the total area, and desertified and barren landscapes are widely distributed. The spatial pattern and supply capacity of water resources determine vegetation cover, LU suitability, and the resistance of landscape systems to disturbance [10]. These factors, in turn, shape the level and spatial differentiation of LER. Specifically, annual rainfall reflects natural vegetation cover and soil moisture conditions. It serves as a key indicator of the regional ecological carrying capacity. Distance to rivers represents the spatial accessibility of surface water. It is a critical constraint for the persistence of artificial oases and the implementation of ecological restoration projects in arid regions [5]. Areas with higher rainfall and closer proximity to rivers tend to exhibit more complete vegetation structures and more stable landscape patterns. These ecosystems have stronger resistance to disturbance and greater self-recovery capacity, resulting in lower risk [60]. In contrast, desert and Gobi areas with low rainfall and limited access to water resources are characterized by sparse vegetation and simple landscape structures [28]. Their ecosystems are highly fragile and prone to irreversible degradation once disturbed, leading to a higher risk. From a theoretical perspective, these findings further confirm the fundamental principle of “water-resource dominance” in arid ecosystems. This insight deepens the understanding of the coupling relationship between water availability and landscape structure, and provides an important basis for interpreting arid-region ecosystems. From a practical perspective, the results highlight the central role of water resources in LER management. They support the adoption of a “water-determined LU” principle in LU planning and ecological governance. By optimizing water allocation, improving water-use efficiency, and protecting river ecological corridors, it is possible to enhance ecosystem stability and reduce LER.

Among the socioeconomic variables, population density and distance to main roads make the largest contributions. Nie et al. pointed out that human activities are a major cause of LER, yet the results of this study show a negative correlation between population density and LER, which is contrary to their conclusion [9,22]. The core reason for this discrepancy lies in the ecological vulnerability of the Hexi Corridor: the region is predominantly unused land, with a high proportion of desertification and desert areas. To adapt to the extreme arid environment, the population has concentrated in ecologically favorable areas. High population density areas are core oasis regions with relatively abundant water resources and suitable conditions for cultivation, meaning population density in these areas is not simply an “interference source” but rather an “indicator” of human adaptation to superior natural conditions over the long term [12]. The true driver of LER in this region is the overexploitation of resources to maintain and expand high-density settlement patterns, which exacerbates LER. Further, from the perspective of on-the-ground conditions, in the desert–oasis ecotones of the Heihe and Shiyang River basins, settled farmers have actively participated in sand control initiatives under the Three-North Shelterbelt Program. Through extensive afforestation efforts aimed at sand stabilization, together with the implementation of water-efficient irrigation practices such as drip systems, areas previously degraded by intense wind erosion and elevated ecological risk have been converted into shelterbelt networks that provide stable ecological protection functions. At the same time, the establishment of public ecological stewardship positions has enabled local residents to transition into full-time environmental stewards. This has created a positive feedback mechanism characterized by “settling in one place, managing one area, and stabilizing one region.” This process represents a typical pathway through which active human intervention promotes the positive succession of fragile ecosystems. Therefore, the results of this article enhance the understanding of human–environment interactions. In arid regions with fragile ecological conditions, human settlement can play a constructive role. Through proactive ecological management, optimized resource allocation, and large-scale ecological restoration, human activities can enhance ecosystem stability and ultimately reduce LER.

4.1.3. Future LU and LER Simulation Prediction Analysis

Based on the 2030 multi-scenario simulation results of the Markov-FLUS model, data support is provided for LU optimization and LER prevention in the Hexi Corridor. In the natural development scenario, cropland and grassland areas will decrease by 31 km² and 138 km², respectively, while built-up land will increase by 64 km², and high-risk areas will decrease by 542 km². In this scenario, regional LER shows a slow improvement trend; however, the potential pressure on the local ecological environment caused by the expansion of built-up land, especially in ecologically vulnerable areas such as the Heihe River Basin and Shiyang River Basin, should still be monitored. The combination of glacier meltwater reduction and human water use pressure may still lead to vegetation degradation and landscape fragmentation, potentially triggering a rebound in LER. In the cropland protection scenario, cropland and grassland areas expand by 126 km² and 271 km², respectively, whereas built-up land contracts by 2 km² and high-risk areas decline by 801 km². This scenario aligns with the objectives of the Gansu Provincial Land and Space Planning. By strictly controlling the encroachment of built-up land on ecological areas and ensuring the preservation of cropland, the ecosystem’s resilience to risks can be significantly enhanced. In the urban development scenario, cropland and grassland decline by 30 km² and 148 km², respectively, whereas built-up land expands by 190 km², and high-risk areas increase by 794 km². In this scenario, the rapid industrialization and urbanization of the region will drive the expansion of built-up land, which will squeeze ecological spaces and increase pressure on LER. From a theoretical perspective, this study employs a multi-scenario comparative analysis to reveal the differentiated responses of LER under various LU regulation pathways. The results indicate that the evolution of LER is not only constrained by the natural baseline but is also highly sensitive to human LU decisions and management strategies. This finding enriches the scenario-based research framework for human–environment coupling in arid regions.

4.1.4. Policy Implications

First, differentiated spatial regulation strategies should be implemented according to LER levels. For higher and high-risk areas (e.g., Jiuquan and Wuwei), priority should be given to ecological conservation and restoration. In desert areas far from oases, protection of native vegetation should be strengthened, and human disturbance should be minimized [38]. In contrast, desert areas surrounding oases should adopt site-specific restoration strategies. Drought- and cold-tolerant species, such as Elaeagnus angustifolia, Populus euphratica, and Haloxylon ammodendron, can be introduced to enhance surface cover and improve windbreak and sand stabilization capacity. In terms of implementation timing, sand stabilization and planting activities should follow seasonal patterns [27]. Sand compaction is more suitable in autumn, while afforestation should be emphasized in spring. At the same time, multiple approaches—including photovoltaic-based sand control, mechanized sand control, engineering measures, and public welfare programs—should be integrated to generate synergistic effects in comprehensive desertification control [54]. For areas with moderate risk, priority should be given to optimizing LU structure and regulating the intensity of human activities. Cropland protection policies and permanent basic farmland redlines should be strictly enforced, and non-agricultural land conversion should be regulated. Simultaneously, efforts should be made to advance the development of high-standard farmland and promote water-efficient irrigation technologies, thereby reducing agricultural water consumption and easing pressure on regional water resources and ecosystems [64]. Spatially, the expansion of rural settlements, transport infrastructure, and industrial land should be properly controlled. This will reduce landscape fragmentation and enhance ecological connectivity. For areas with low and lower risk, the focus should be on maintaining ecological functions and improving resource-use efficiency. Water-saving agriculture and intensive LU should be promoted under the premise of ecological security to maintain landscape stability [28]. In addition, priority should be given to the protection and restoration of upstream water conservation areas. Human disturbance should be strictly controlled to enhance regional water regulation capacity and provide stable support for downstream oasis ecosystems.

Second, key elements should be regulated based on the identified driving factors. First, the rigid constraint of water resources should be strengthened. The functional roles of the three inland river systems—the Heihe River, Shiyang River, and Shule River—should be clearly defined. Water allocation patterns and total water use should be rationally distributed across upstream, midstream, and downstream regions. In upstream regions, priority should be placed on preserving water conservation functions, with water utilization constrained to a maximum of 15%. Midstream areas should primarily support socioeconomic functions while maintaining ecological stability, with water use controlled at 50%. Downstream areas should prioritize ecological security, with production and domestic water use as secondary functions, and total water use limited to 35%. At the same time, long-term mechanisms for water conservation and control should be improved. Water-saving technologies should be developed and promoted, and water-efficient infrastructure should be upgraded [44]. Efforts should focus on improving agricultural water efficiency, reducing industrial water consumption and emissions, minimizing urban water losses, and strengthening ecological water management. Second, human activities should be regulated in an orderly manner to promote coordinated human–environment development. Population and industrial activities should be directed toward oasis zones and other areas with comparatively favorable ecological conditions [43]. Measures such as water-saving agriculture, refined land management, and ecological restoration should be strengthened to enhance the ecological regulation function of human activities. At the same time, human activity intensity in ecologically fragile areas should be strictly restricted to prevent disturbances caused by uncontrolled development in desert–oasis transition zones. Finally, ecological constraints on transportation infrastructure should be reinforced. LER assessment should be incorporated as a prerequisite for project approval. Road expansion into high-risk areas, such as Jiuquan and Wuwei, as well as ecologically fragile downstream regions of the Heihe River and Shiyang River, should be strictly prohibited [65]. For roads crossing high-risk areas, sand-fixing vegetation belts should be constructed. Road construction and maintenance should be coordinated with desertification control efforts. In addition, differentiated and precise management should be applied to rural roads, fire access routes, and tourism corridors.

Third, institutional safeguards for LER prevention and control should be improved based on scenario simulation results. First, constraints on the expansion of built-up land should be strengthened. For major river basins, including the Heihe and Shiyang Rivers, provincial natural resources authorities should establish caps on the total extent of built-up land [36]. This quota should then be allocated to prefecture-, county-, and township-level governments. At the same time, a negative list system for built-up land should be established. Built-up land and non-agricultural development in oasis margins and desert areas should be strictly restricted [36]. A grid-based dynamic monitoring platform, supported by high-resolution remote sensing imagery, should be developed to track built-up land expansion and LER changes in real time. In addition, mechanisms for revitalizing existing built-up land should be promoted. Regular surveys of inefficient LU should be conducted, and an inventory of existing land should be established [28]. The effectiveness of land revitalization should be incorporated into local government performance evaluation systems. Second, cropland protection policies should be strictly enforced. Annual LU plans should prioritize land allocation for high-standard farmland construction and farmland quality improvement. Regions that meet or exceed cropland protection targets should receive preferential support in terms of built-up land quotas and fiscal resource allocation [60]. At the same time, for large-scale grain producers and agricultural operators, subsidy mechanisms for farmland quality protection should be improved, along with supporting policies. Finally, dynamic monitoring and adaptive management mechanisms for ecological risk should be strengthened. A monitoring network integrating satellite remote sensing, unmanned aerial vehicle inspections, and ground-based surveys should be established [14,34]. This system should enable continuous monitoring of vegetation cover, desertification dynamics, water resource changes, and extreme climate events. Based on these data, regular evaluations of ecological restoration outcomes, management effectiveness, and ecological benefits should be conducted [12,31]. In response to intensified glacier melt in the Qilian Mountains and the increasing frequency of extreme climate events, protection of water conservation areas should be strengthened. Monitoring, early warning, and emergency response systems for hazards such as flash floods and debris flows should also be improved. These efforts will enhance the foresight and resilience of regional ecological risk management.

4.1.5. Limitations of the Study

This study has several limitations that should be addressed in future research. First, regarding the evaluation framework, the LER index is constructed based on a classical landscape pattern approach. This method cannot fully capture risks at the ecosystem function level. Key processes, such as water quality degradation and biodiversity loss, are not explicitly represented. Future studies could integrate LER assessment with ecosystem service evaluation models to develop a more comprehensive risk assessment framework. In addition, the weights assigned to landscape disturbance and vulnerability are derived from existing studies. Future work could further examine their robustness through sensitivity analysis, robustness tests, or comparative experiments. Second, in terms of driving factors, this study does not fully account for certain human activities, such as cross-regional infrastructure connectivity and industrial transfer. Future research could incorporate indicators reflecting the intensity of interregional economic linkages in order to better capture the mechanisms driving LER. Finally, with respect to scenario design, the current settings do not fully consider uncertainties such as extreme climate events or abrupt ecological policy adjustments. Future studies could introduce additional scenarios, including extreme drought, enhanced ecological restoration, and strict water resource constraints. These scenarios would help simulate the evolution of LER under different response strategies and improve the policy relevance of the findings.

4.2. Conclusions

In terms of LU and LER, the Hexi Corridor is dominated by unused land, and substantial changes in LU occurred during the study period. From 2000 to 2020, the areas of cropland, water body, and built-up land increased by 1594 km², 80 km², and 608 km², respectively, with their shares rising to 6.40%, 1.11%, and 0.71%. In contrast, forestland, grassland, and unused land decreased by 34 km², 127 km², and 2121 km², with their proportions declining to 2.97%, 21.47%, and 67.33%, respectively. Regarding LER, the dominant category is high risk, and the overall risk pattern changed markedly over the study period. Between 2000 and 2020, the areas of higher and high risk decreased by 1426 km² and 2885 km², accounting for 16.09% and 52.02%, respectively. Meanwhile, the areas of low, lower, and moderate risk increased by 1920 km², 768 km², and 1624 km², with shares rising to 13.97%, 7.80%, and 10.12%. In terms of driving factors, natural geographic variables play a dominant role in shaping LER. Among them, annual precipitation and distance to rivers show the highest contributions, with SHAP values of 29.7% and 10.7%. Among socio-economic factors, population density and distance to main roads are the most influential, with SHAP values of 25.8% and 10.8%. With respect to scenario-based simulation and prediction, LU and LER exhibit pronounced spatial differences under alternative development pathways. Under the natural development scenario, cropland decreases by 31 km² (to 6.39%), while built-up land increases by 64 km² (to 0.74%). The areas of higher and moderate risk increase by 333 km² and 295 km², with proportions rising by 0.13% and 0.12%. Under the cropland protection scenario, cropland increases by 126 km² (to 6.45%), while built-up land decreases by 2 km² (to 0.71%). The area of high risk decreases by 801 km², with a reduction of 0.32%. Under the urban development scenario, cropland decreases by 30 km² (to 6.39%), whereas built-up land increases by 190 km² (to 0.79%). The areas of higher and moderate risk increase by 1303 km² and 81 km², with proportions rising by 0.52% and 0.03%, respectively.