Assessment of Land Desertification and Its Drivers in Semi-Arid Alpine Mountains: A Case Study of the Qilian Mountains Region, Northwest China

Abstract

Land desertification associated with climate change and human activities significantly impacts ecosystem functioning in semi-arid alpine mountains. However, accurately revealing the state of desertification risk and the drivers of its evolution is frequently difficult, especially in the semi-arid alpine mountains. A new theoretical framework that combined qualitative and quantitative concepts has been developed to enhance ecological risk assessment in semi-arid alpine mountains and reveal the causes of desertification. The PSR model, multi-layer hierarchical theory, hierarchical analysis, inverse cloud generating principles, field surveys, structured questionnaires, and remote sensing techniques are all combined in this method. Our results showed that the risk of desertification in the study area exhibited a fluctuating trend between 2000 and 2020, with a period of decrease, followed by an increase, and then a subsequent decrease. However, the risk status remained overall stable, remaining at a light desertification level during the entire period. Desertification risk is driven primarily by climate warming and humidification, which can cause the melting of ice/snow. Additionally, increased rainfall and freeze–thaw cycles can enhance soil erosion, further exacerbating the risk. Conversely, the implementation of environmental protection projects, such as the establishment of protected areas, efforts to restore forests and grasslands, and initiatives to conserve soil and water, has been effective in limiting the increase in desertification risk. These efforts serve as a counterforce to the negative impacts of climate change and human activity, highlighting the beneficial effects of human intervention in preventing desertification. High-altitude, high-topographic relief places have considerable desertification risk, mainly in the alpine desert. Due to geography, grazing, rodent and pest infestation, and wildlife, there is still a risk of desertification expanding in low elevation areas. There will be a greater urgency in the future to enhance the management of anthropogenic activities in the local environment in order to handle the growing threat of desertification caused by climate change. This study combined the interactions of the natural environment and human activities, filled a research gap in assessing desertification risk, and revealed its driving mechanisms, as well as provided a theoretical foundation for improving the integrity and sustainability of ecosystems in semi-arid alpine regions and elsewhere.

1. Introduction

Desertification is a common form of land degradation that happens primarily in arid and semi-arid settings [1,2]. Desertification impacts approximately 46% of the worldwide geographical area, and the problem is growing [3,4]. Among the various driving factors, climate change and human activities have proven to be the most important drivers of desertification evolution. Soil erosion is a major problem that is exacerbated by changes in conditions such as precipitation, temperature, and human production methods. These changes ultimately contribute to desertification [5,6,7,8,9].

Over the past few decades, there has been a significant amount of research conducted by scholars focusing on the driving mechanisms of desertification, the relationship with media, spatial risks, and evolutionary processes. These studies have yielded substantial research outcomes and findings. The geographical scope of the studies conducted revolved primarily around desert ecological zones and arid ecologically sensitive areas. In Ref. [10], the data mining approach was applied to investigate the impact of climate change on the spatial and temporal patterns of land desertification in the Agro-Pastoral Ecological Zone of Northern China (APENC). In Ref. [11], the multi-scale trend analysis approach was applied to analyze the long-term trends (2000–2020) of four desert ecosystem services and their associated drivers in the Hunsandak Sandy Land of Inner Mongolia. In Refs. [12,13], through the enhancement of conventional vegetation monitoring techniques and the addition of meteorological variables, the characteristics of desertification and land degradation in the Mongolian steppe were assessed. In Ref. [14], based on data from the yearly and growing season normalized vegetation index (NDVI), trends and drivers of desertification progression in the arid region of Sindh Province, Pakistan, were investigated. These studies have thoroughly analyzed the risk, evolutionary pattern, and drivers of desertification in the desert ecoregions and ecologically sensitive areas characterized by arid and semi-arid conditions. However, in cold semi-arid regions, water and heat resources have been shown to be the most essential variables governing the evolution of regional ecosystems and soil and water environment changes, which have a direct impact on the evolutionary changes of the ecological environment, particularly in semi-arid alpine mountains [15,16]. Strong vertical zonation in these areas is frequently accompanied by extensive hydrothermal exchange and is linked to soil hydrological properties [17,18]. Ultimately, it directly drives the evolutionary changes in the region’s soil and water environment, which has an impact on the ecological barrier function of the alpine ranges [19,20]. This physicochemical pattern of coupled water–thermal–soil–vegetation is frequently mirrored in water phase transition, water–heat exchange, soil physicochemical property reorganization, soil erosion, and land use change processes [21,22]. Human activities also profoundly impact desertification development in both positive and negative ways [23,24]. Different climate patterns and changes in human activities are gradually changing the native soil and water environment in these regions, leading to the gradual degradation of land resources, primarily desertification, which seriously threatens the ecological barrier function of alpine mountains [25,26,27]. As a result, the large-scale desertification differentiation process in the semi-arid alpine highlands is a complex process combining multi-media participation and multi-level driving. Compared with the traditional studies on desertification in dry land habitats, desertification in semi-arid alpine ranges at numerous temporal and spatial scales is more intricate, distinctive, and unique [28,29]. It is surprising that the dynamics and underlying processes of desertification risk in semi-arid alpine highlands have received little attention to far. With the intensification of climate change, an accurate assessment of the desertification risk under the combined influence of climate change and human activities is an important necessity to increase desertification management and adaptation to climate change in the alpine highlands.

As far as existing methods for desertification risk assessment are concerned, two main types are commonly used: those based on remote sensing and those based on questionnaires. The remote sensing method is used primarily for spatial analysis of desertification status by creating a physical relationship between the degree of desertification and the ground-level characteristics. This is a standard quantitative analysis approach for accurately assessing the danger of desertification on a large scale [30,31]. Remote sensing parameters, such as the Normalized Difference Vegetation Index (NDVI), albedo, Net Primary Production (NPP), Fractional Vegetation Cover (FVC), Desertification Degree Index (DDI), Modified Soil Adjusted Vegetation Index (MSAVI), and Soil Moisture Index, are used to assess quantitatively the desertification in different scale areas by developing desertification evaluation models [11,23,32,33,34,35,36]. However, this method cannot be combined with human activities and their subjective perception of the evolution of the desert environment. It can only quantitatively monitor the spatial characterization of the state of desertification, which can only be quantitatively and spatially analyzed statistically.

The questionnaire-based desertification evaluation approach first evaluates the driving forces before establishing an evaluation system based on the background condition of desertification in the target area. The indicator weights are then determined using several weight determination methods based on experts’ or local inhabitants’ perceptions of the desertification in the target area, and the degree of desertification is calculated. For example, sociological surveys are used to measure the impact of desertification on the environment and the local population’s level of living [37]. The Environmental Change Perception and Reflection Process Model were used to analyze the public’s environmental perception of the desertification in Minqin, China [38]. The sensitivity of land desertification in the Mediterranean region was analyzed using this method by developing an evaluation system based on vegetation, soil, climate, and land management parameters [39,40]. Multiple disturbances in the Guilin World Heritage Site karst terrain were investigated, taking into account the impact of human activities and natural changes on the rocky desert landscape [41]. However, questionnaire-based research generally yields subjective conclusions that do not objectively assess desertification risk.

In addition to these deficiencies, the lack of a thorough and accurate technique to estimate desertification risk is the main difficulty facing desertification assessment. Desertification is a non-linear process influenced by many factors. Remote sensing is a quantitative tool for desertification evaluation, but it overemphasizes remote sensing indicators and emphasizes the spatial significance and danger. The complex non-linear relationship between human activity and the natural environment is often overlooked. Traditional questionnaire methods can reflect the perceptions of target area residents on the evolution of the desert environment and provide more reliable empirical evidence, but their findings are often subjective and uncertain. Both techniques of depicting desertification risk and its causes have pros and cons. Therefore, a new theoretical model of the desertification environment in the target area, as a way to reveal the desertification risk state and its driving mechanisms, is still an urgent issue in desertification assessment. The cloud model, proposed by academician Deyi Li, is a modern mathematical theoretical model that combines traditional probability statistical theory with fuzzy theory. This model offers significant advantages over other methods in terms of addressing uncertainty problems. In Ref. [42], the Innovative Driver-Pressure-Engineering water shortage-State–Ecological basis–Response-Management (DPESBRM) conceptual and cloud models were utilized for assessing the water resource carrying capacity in China’s karst regions. In Ref. [43], the cloud model was used to investigate strategies for enhancing the differential measurement of urban resilience in China, particularly from a social–ecological systems perspective. In Ref. [44], the Improved Entropy Weights–Cloud Model (IEW-CM) was used to evaluate the level of energy sustainability development in 30 provinces across China. It has been shown that cloud models have significant advantages over other methods in solving uncertainty problems [45]. Cloud models may precisely and efficiently explain complex system stochasticity and uncertainty, allowing quantitative values and qualitative notions to be converted [46,47]. This method combines questionnaire and remote sensing data and yields accurate evaluation findings and probability values.

The Qilian Mountains are a typical semi-arid alpine region, a water-conserving oasis along China’s Hexi Corridor [48]. Due to the rapid pace of climate change, the ecological environment of the region has undergone a series of changes, particularly with regard to desertification. This issue has become a national concern in recent years. Therefore, to preserve the Qilian Mountains’ ecosystem integrity, it is crucial to identify precisely the region’s desertification trends and subsequently manage the desert ecosystems. The primary aims of this study were as follows: (1) Using a structured questionnaire, investigate the desertification risk and its driving factors in the study area from 2000 to 2020; (2) Obtain three remote sensing indicators characterizing desertification risk based on the Google Earth Engine (GEE) platform, albedo, FVC, and Green Normalized Difference Vegetation Index (GNDVI) to reveal the spatial and temporal changes in desertification status; and (3) Introduce a cloud model, calculate three numerical cloud model features derived from survey questionnaire and remote sensing index data, and build a comprehensive cloud model taking both qualitative and quantitative concepts into account to reveal changes in the desertification risk and its driving mechanism in the study area from 2000 to 2020.

2. Materials and Methods

2.1. Study Area

The Qilian Mountains are located on the northeastern edge of the Tibetan Plateau, (lat. 93.51–103.90° E, long. 35.83–39.98° N), with the elevation increasing from southeast to northwest and a highest altitude of 5791 m, belonging to the typical semi-arid alpine continental climate. The average annual temperature is −2.1 °C, and the precipitation varies significantly at different altitudes, with an average annual precipitation of 366 mm. The strong vertical zonation and topographic relief of the region result in distinct geographical differences in temperature and precipitation, ultimately leading to the creation of various vegetation zones. The region’s vegetation zones are categorized based on altitude, ranging from low to high. The zones include desert steppe, montane shrub–steppe, montane forest–steppe, subalpine shrub–steppe, alpine meadow/steppe, alpine desert, and permanent snow/glacier [49]. Due to the complex topography and climatic conditions of the region, its ecological fragility is high, particularly in light of drastic climate change. The area is highly susceptible to degradation of soil and water resources on a regional scale [50].

We selected two counties within the Qilian Mountains National Park in the study area: Menyuan Hui Autonomous County and Qilian County (Figure 1). The elevation of the study area is between 2352 and 5218 m. The study area is characterized by a semi-arid alpine climate that is primarily controlled by the highland continental climate. The climate in this area is characterized by high solar radiation, long hours of sunshine, low temperatures year-round, and relatively high humidity levels. Key data from the survey and specific information about the study area are shown in

Table 1. Basic information regarding the study area.

County	Area	Average Elevation	Average Annual Temperature	Average Annual Precipitation	No. of Surveyed Villages 1	Major Ecosystem Type 2
Menyuan Hui Autonomous County	6902.26 km2	2866 m	0.8 °C	520 mm	4 (7)	① Desert steppe ② Montane shrub–steppe ③ Montane forest–steppe ④ Subalpine shrub–steppe ⑤ Alpine meadow/steppe ⑥ Alpine desert ⑦ Permanent snow/glaciers
Qilian County	13,886 km2	3196 m	1 °C	420 mm	7 (12)

¹ Figures in brackets are the number of townships in the surveyed counties. ² ① to ⑦ indicate the seven major ecosystems present in the study area and the elevation of their distribution area increases with increasing serial number.

.

2.2. Methodology

The process of investigating and assessing the risk of desertification in semi-arid alpine mountains is shown in Figure 2. The process can be divided into five steps.

Step 1: By combining the Pressure–State–Response (PSR) model from environmental quality assessment, which evaluates ecosystem health, with multi-level fuzzy hierarchy theory, conduct a survey using the questionnaire to gather qualitative data.

Step 2: Analyze spatiotemporal evolution trends in albedo, FVC, and GNDVI data. Use randomly derived raster attribute datasets from different time periods to input into the inverse cloud generator. The generator will produce numerical features of desertification risk state clouds with quantitative characteristics.

Step 3: Use the qualitative data gathered from the questionnaire to calculate desertification risk state cloud digital features using the inverse cloud generator.

Step 4: Integrate the qualitative and quantitative ideas of cloud digital signatures to develop integrated parameters. Analyze the desertification risk profile in distinct time series.

Step 5: Apply the 3δ stability concept to the cloud model assessment. If He < En/3, the evaluation outcomes will be more stable and less variable. If He > En/3, variability and uncertainty will increase.

2.2.1. Quantitative Raster Data Extraction

Using the Create Random Points module in ArcGIS, 1000 points were generated randomly within the vector range of the study area. The raster image element values corresponding to each of these random points were extracted from the albedo, FVC, and GNDVI raster data from 2000, 2005, 2010, 2015, and 2020, respectively. These data were used as the input source for the quantitative cloud model. It is important to note that different indicators have both positive and negative effects on the characterization of the desert environmental risk status. Outliers were rejected from the extracted 1000 random point image element values, due mainly to individual null points. The three remote sensing quantitative data sources were then normalized to eliminate the influence of the indicator scale. The equation used for normalization is as follows:

For positive indictors:

$$y_{ij}=\left({\mathrm{x}}_{ij}-\min x_j\right)/\left(\max x_j-\min x_j\right)$$

(1)

For negative indictors:

$$y_{ij}=\left(\max x_j-x_{ij}\right)/\left(\max x_j-\min x_j\right)$$

(2)

where y_ij is the standardized value of the jth indicator in the ith sample; x_ij (i = 1, 2, …, n; j = 1, 2, …, m) is the jth indicator in the ith sample; and maxx_j and minx_j are the maximum and minimum of the jth indicator, respectively.

2.2.2. Cloud Model

The cloud model, proposed by academician Deyi Li of the Chinese Academy of Engineering in 1995, is a mathematical model that facilitates the conversion of qualitative concepts into quantitative values. This model combines traditional mathematical and statistical models with fuzzy theory and is particularly useful in addressing problems characterized by strong stochasticity and uncertainty [42,51]. Desertification risk is a case in point, as it involves multi-level drivers and multi-factor coupling, which further exacerbate uncertainty problems.

The inverse cloud generator principle was introduced in this study to assimilate and fuse qualitative assessment data of desertification risk status obtained through questionnaires and remote sensing index raster data into a quantitative feature. This was achieved using the cloud model, which employed three cloud numerical feature parameters: expectation (Ex), entropy (En), and superentropy (He). Ex represents the central value of spatial distribution of the cloud model and reflects the rank fall of assessment results. En measures the uncertainty of the assessment results, which is determined by randomness and ambiguity [38]. He is a measure of the uncertainty of En and reflects the agglomeration and stability of the point distribution of all assessment results. Matlab 2022b software was used to implement all computations and visual representations of the cloud model.

The Desertification Risk Status (DRS) cloud model produced by this process is denoted by the symbol C (Ex, En, He). The following algorithm outlines the detailed computational procedure:

(1)

Calculation of sample means for survey indicators and quantitative spatial indicator data:

$$Ex=\overline{X}=\frac{1}{n}{\displaystyle \sum _{i=1}^{n}D}R{S}_i$$

(3)

where Ex is the expectation, $\overline{X}$ represents the mean value, and DRS_i is the desertification status characterization indicator (i includes questionnaire indicator factors and spatial data indicator factors).

(2)

Entropy is obtained from the central moments of the sample:

$$d=\frac{1}{n}{\displaystyle \sum _{i=1}^n\left|DR{S}_i-\overline{X}\right|}$$

(4)

$$En=\sqrt{\frac{\pi }{2}}d$$

(5)

where En represents the entropy and d represents the first-order sample absolute central moment.

(3)

Calculation of superentropy:

$$S^2=\frac{1}{n-1}{\displaystyle \sum _{i=1}^n{\left(DR{S}_i-\overline{X}\right)}^2}$$

(6)

$$He=\sqrt{\left|S^2-E_n^2\right|}$$

(7)

where S² represents the sample variance and He represents superentropy.

2.2.3. Scale Judgment Matrix of Cloud Mode

U = x = 1, 2, …, 5 can be represented by Ex, En, and He and is denoted as R = (Ex, En, He). Nine cloud models were used to establish the importance decision scale: R₁(Ex₁, En₁, He₁), R₂(Ex₂, En₂, He₂), R₃(Ex₃, En₃, He₃), R₄(Ex₄, En₄, He₄), and R₅(Ex₅, En₅, He₅). The entropy and hyper entropy of the five cloud models can be obtained using the golden section method, Define Enmin = 0.618Enmax and Hemin = 0.618Hemax between adjacent classes [52]. Ex ∈ [0, 1], and when Ex is close to 0.5, En, He, and R belong to the moderate level. For “slight risk of desertification R₄” and “no risk of desertification R₅”, the following was calculated:

Extreme risk of desertification R₁:

$$E{x}_1=0$$

(8)

$$E{n}_1=E{n}_2/0.618=0.103$$

(9)

$$H{e}_1=H{e}_2/0.618=0.0131$$

(10)

Severe risk of desertification R₂:

$$E{x}_2=0.618(x_{\max }+x_{\min })/2=0.309$$

(11)

$$E{n}_2=(1-0.618)(x_{\max }-x_{\min })/6=0.064$$

(12)

$$H{e}_2=H{e}_3/0.618=0.0081$$

(13)

where x_max represents the maximum value of the expected Ex distribution interval and x_min represents the minimum value of the expected Ex distribution interval, Ex ∈ [0, 1].

Moderate risk of desertification R₃:

$$E{x}_3=\frac{x_{\max }+x_{\min }}{2}=0.5$$

(14)

$$E{n}_3=E{n}_2\times 0.618=0.039$$

(15)

$$H{e}_3=0.005$$

(16)

Light risk of desertification R₄:

$$E{x}_4=E{x}_3+(1-0.618)(x_{\max }+x_{\min })/2=0.691$$

(17)

$$E{n}_4=(1-0.618)(x_{\max }-x_{\min })/6=0.064$$

(18)

$$H{e}_4=H{e}_3/0.618=0.0081$$

(19)

No risk of desertification R₅:

$$E{x}_5=1$$

(20)

$$E{n}_5=E{n}_4/0.618=0.103$$

(21)

$$H{e}_5=H{e}_4/0.618=0.0131$$

(22)

The characteristic parameters of the cloud model for each of the five desertification risk assessment levels were calculated as R₁(0, 0.103, 0.0131), R₂(0.309, 0.064, 0.0081), R₃(0.50, 0.039, 0.005), R₄(0.691, 0.064, 0.0081), and R₅(1, 0.103, 0.0131). The standard assessment cloud can be visualized as in Figure 3.

2.2.4. Analytic Hierarchy Process Based on the Cloud Model Scale Judgment Matrix

Assuming that there are two cloud models CR₁(Ex₁, En₁, He₁) and CR₂(Ex₂, En₂, He₂) in the same quantitative domain, such that the operation of CR₁ and CR₂ results in CR(Ex, En, He), for CR = CR₁/CR₂, the following equation is used to calculate the three cloud parameters [53]:

$$Ex=\frac{E{x}_1}{E{x}_2}$$

(23)

$$En=\left|\frac{E{x}_1}{E{x}_2}\right|\sqrt{{\left(\frac{E{n}_1}{E{x}_1}\right)}^2+{\left(\frac{E{n}_2}{E{x}_2}\right)}^2}$$

(24)

$$He=\left|\frac{E{x}_1}{E{x}_2}\right|\sqrt{{\left(\frac{H{e}_1}{E{x}_1}\right)}^2+{\left(\frac{H{e}_2}{E{x}_2}\right)}^2}$$

(25)

where Ex₁ represents the expected value of cloud model 1 and Ex₁ represents the expected value of cloud model 2; En₁ represents the entropy of cloud model 1 and En₂ represents the entropy of cloud model 2; and He₁ represents the superentropy of cloud model 1 and He₂ represents the superentropy of cloud model 2.

The traditional Satty scales cannot eliminate the uncertainty and dispersion problems from the scoring process when constructing the judgment matrix. The conventional hierarchical analysis method 1 to 9 scales were used to create the cloud-model-based scaling guidelines for assessing the risk status of desertification according to the cloud model theory [5,54].

The nine cloud models (

Table 2. The nine comparison cloud models.

Relative Importance between Indicators	Cloud Model
Xi is slightly more important than Yj	C1(Ex1, En1, He1)
Xi is significantly more important than Yj	C2(Ex2, En2, He2)
Xi is strongly more important than Yj	C3(Ex3, En3, He3)
Xi is definitely more important than Yj	C4(Ex4, En4, He4)
Xi is slightly less important than Yj	C5(Ex5, En5, He5)
Xi is obviously less important than Yj	C6(Ex6, En6, He6)
Xi is strongly less important than Yj	C7(Ex7, En7, He7)
Xi is definitely less important than Yj	C8(Ex8, En8, He8)
Xi and Yj are equally important	C0(Ex0, En0, He0)

) were defined based on the importance of the Satty’s scale value indicators compared between the two indicators, X_i and Y_j.

The cloud model characteristics parameters are determined. ① Expectation Ex. According to the Satty scale, Ex₄ = 9, Ex₃ = 7, Ex₂ = 5, Ex₁ = 3, Ex₅ = 1/3, Ex₆ = 1/5, Ex₇ = 1/7, Ex₈ = 1/9, and Ex₀ = 1 can be used to represent the expectation value. ② Entropy En. Combining the normal distribution 3En principle, the entropy corresponding to cloud models C₁, C₂, C₃, and C₄ are obtained as follows: En₁ = 0.33, En₂ = 0.33, En₃ = 0.33, and En₄ = 0.33; cloud models C₅, C₆, C₇, and C₈ are reciprocal to cloud models C₁, C₂, C₃, and C₄ and are considered to be the standard cloud model CR₀(1, 0, 0) to C₁, C₂, C₃, and C₄. En for cloud models C₅, C₆, C₇, and C₈ can be obtained according to Equation (9) as En₅ = 0.33/9, En₆ = 0.33/25, En₇ = 0.33/49, and En₈ = 0.33/81, respectively. The entropy of cloud model C₀ (1, 0, 0) with equal importance of X_i and Y_j is En₀ = 0. ③ Superentropy He. The corresponding superentropy of cloud models C₁, C₂, C₃, and C₄ He can be taken as an empirical value based on existing research results and defined as He₁ = 0.01, He₂ = 0.01, He₃ = 0.01, and He₄ = 0.01 [5,45,52,54]. In line with the calculation of entropy, the superentropy of cloud models C₅, C₆, C₇, and C₈ can be calculated by Equation (10) as He₅ = 0.01/9, He₆ = 0.01/25, He₇ = 0.01/49, and He₈ = 0.01/81; for the hyperentropy of the equally important cloud models X_i and Y_j, C₀(1, 0, 0), we have He₀ = 0. A binary comparison of the nine cloud models is shown in

Table 3. Importance scales for binary comparisons of nine cloud models.

Comparison between Indicators	Level of Importance	Cloud Model
Xi is more important than Yj	Absolutely	C4(9, 0.33, 0.01)
	Strongly	C3(7, 0.33, 0.01)
	Obviously	C2(5, 0.33, 0.01)
	Slightly	C1(3, 0.33, 0.01)
Xi and Yj are equally important		C0(1, 0, 0)
Xi is less important than Yj	Slightly	C5(1/3, 0.33/9, 0.01/9)
	Obviously	C6(1/5, 0.33/25, 0.01/25)
	Strongly	C7(1/7, 0.33/49, 0.01/49)
	Absolutely	C8(1/9, 0.33/81, 0.01/81)

.

Following the establishment of the judgment matrix (R_ij)_n×n and the use of the square root method to calculate the weights of the response indicators for each element in the judgment matrix [53,55], the cloud model weights W_i (Ex_i, En_i, He_i) were constructed by the two-by-two comparison of the importance of the indicators for assessing the environmental risk status of deserts in the study area, respectively:

$$E{x}_i=\frac{E{x}_i}{{{\displaystyle \sum }}_{i=1}^{n}E{x}_i}=\frac{{\left({{\displaystyle \prod }}_{j=1}^{n}E{x}_{ij}\right)}^{1/n}}{{{\displaystyle \sum }}_{i=1}^n{\left({{\displaystyle \prod }}_{j=1}^{n}E{x}_{ij}\right)}^{1/n}}$$

(26)

$$E{n}_i=\frac{{\left[{{\displaystyle \prod }}_{j=1}^{n}E{x}_{ij}\sqrt{{{\displaystyle \sum }}_{j=1}^n{\left(\frac{E{n}_{ij}}{E{x}_{ij}}\right)}^2}\right]}^{1/n}}{{{\displaystyle \sum }}_{i=1}^n{\left[{{\displaystyle \prod }}_{j=1}^{n}E{x}_{ij}\sqrt{{{\displaystyle \sum }}_{j=1}^n{\left(\frac{E{n}_{ij}}{E{x}_{ij}}\right)}^2}\right]}^{1/n}}$$

(27)

$$H{e}_i=\frac{{\left[{{\displaystyle \prod }}_{j=1}^{n}E{x}_{ij}\sqrt{{{\displaystyle \sum }}_{j=1}^n{\left(H{e}_{ij}/E{x}_{ij}\right)}^2}\right]}^{1/n}}{{{\displaystyle \sum }}_{i=1}^n{\left[{{\displaystyle \prod }}_{j=1}^{n}E{x}_{ij}\sqrt{{{\displaystyle \sum }}_{j=1}^n{\left(H{e}_{ij}/E{x}_{ij}\right)}^2}\right]}^{1/n}}$$

(28)

where Ex_i represents the central value of the evaluation results, indicating the rank distribution of the evaluation; En_i is the uncertainty value of the evaluation result, meaning the ambiguity and dispersion of the value distribution; He_i is the uncertainty value of En_i, representing the degree of cohesion of the value distribution; Ex_ij represents the central value of indicator i in column j in the judgment matrix; En_ij represents the uncertainty value of indicator i in column j in the judgment matrix; He_ij represents the uncertainty value of En_ij; and i (i = 1, 2,…, n) and j (j = 1, 2,…, n) are the number of rows and columns, respectively, of the judgment matrix (R_ij)_n×n.

2.3. Remote Sensing Data

Desertification in semi-arid alpine mountains is a complicated process involving numerous layers and media, characterized by visible complexity and distinctiveness at different temporal and spatial scales. Spatially, it is expressed mainly as a process of change between vegetation growth state and land cover. Albedo, GNDVI, and FVC are three remote sensing indicators that capture changes in plant status and land cover well [56,57]. In addition, in order to explore the relationship between environmental and climate change in the region and the evolution of desert environments, land cover data were introduced as a basis for discussion. Therefore, this study used albedo, Green Normalized Difference Vegetative Index (GNDVI), FVC, and land cover data to reveal the process and state of spatial and temporal evolution of desertification.

Referring to the methodology of the Refs. [58,59], we utilized the Google Earth Engine (GEE) platform to access Landsat 5/7/8 remote sensing data for the entire year. This was achieved by employing online coding in the GEE application programming interface. The data obtained underwent various preprocessing steps, including georeferencing, radiometric calibration, atmospheric correction, batch de-clouding, and image cloud content filtering. Additionally, we computed remote sensing metrics and vectorially cropped the data. Finally, the processed data were exported to Google Cloud Drive. This allowed us to acquire annual data for FVC, albedo, and GNDVI, all at a pixel resolution of 30 m, where GNDVI was the maximum value of GNDVI in each image year from 2000 to 2020. The above data were derived from the United States Geological Survey (USGS) and the National Aeronautics and Space Administration (NASA) (https://lpdaac.usgs.gov/). All Landsat 5, 7, and 8 images were accessed on 1 January to 31 December corresponding to the 2000, 2005, 2010, 2015, 2020 year, respectively.

Our study utilized the 30 m precision annual land cover product (CLCD) created by Jie Yang and Xin Huang at Wuhan University [60]. This updated product spans from 1985 to 2020 and has undergone validation, demonstrating superior data accuracy in China compared with other comparable data products [61]. The land cover classification system for the China Land Cover Database categorized the land cover in our study area into nine types. Digital elevation data was obtained through the geospatial data cloud (http://gscloud.cn, accessed on 1 January 2015). We utilized topographic data to analyze the spatial distribution of desertification in terms of topographic factors.

To ensure consistency and accuracy in our geographic analysis, we conducted projection coordinate conversion, resampling, and spatial alignment of all data sources. We projected all data using the GCS WGS 1984 geographic coordinate system and the WGS_1984_UTM_Zone_48N projective coordinate system, both at a 30 m resolution. This allowed us to match data sources and ensure accurate analysis.

2.4. Questionnaire

In the study area, the inhabitants still maintain a nomadic summer lifestyle. They have witnessed the changes of desertification in the study area and have a deep perception of this environmental change. We conducted preliminary interviews with herders to gain a qualitative understanding of the changes in desertification in the study area from their perspective. This helped us to improve the accuracy of the formal questionnaire preparation and increase the validity of the survey. Based on the preliminary findings, we introduced the PSR model and multi-level fuzzy grading theory to prepare the formal questionnaire [62,63]. The questionnaire’s content was divided into four layers: assessment, driven, factor, and indicator. The factor layer included eight aspects: topography, climate, natural disasters, grassland carrying capacity, habitat quality, environmental quality, ecological governance, and policy guidance. Each factor layer corresponded to four formal survey questions, totaling 32 questions. The survey questions were simplified using specialized terminology. The system and content of the entire questionnaire are shown in Figure 4.

We conducted a formal survey from July to August 2022 using a stratified random sampling method to select villages for the study [64]. We selected 11 villages out of 19 in 2 counties and distributed 115 questionnaires. Of these, 109 were returned, and 6 invalid questionnaires were excluded. We retained 103 valid questionnaires, resulting in a valid return rate of 89.6%. We conducted a Cronbach’s alpha reliability test on the collected data, which yielded a Cronbach’s alpha value of 0.841, indicating high reliability [65]. The test statistic’s Kaiser–Meyer–Olkin (KMO) score of 0.853 was also satisfactory [66]. These results suggested that the questionnaire accurately reflected the true characteristics and feelings of the pastoralists in the study area [67]. The questionnaires were conducted in high mountain pasture areas with altitudes above 2000 m. Due to various factors, such as language barriers, each questionnaire took an average of 45 to 60 min to complete. The interviews were conducted in respondents’ tents, on roads, grasslands, and other convenient locations. In addition to the closed survey questions, we included some open-ended questions to supplement our data. We used the Likert 5-point scale to score the survey responses [68], which were then divided into five levels: (0–0.2) for extreme desert risk R₁, (0.2–0.4) for severe desert risk R₂, (0.4–0.6) for moderate desert risk R₃, (0.6–0.8) for light desert risk R₄, and (0.8–1.0) for no desert risk R₅. This improved scoring method provided more detailed risk assessments.

3. Results

3.1. Quantitative Indicator Characteristics of Remote Sensing

3.1.1. Spatial and Temporal Variation of Quantitative Indicators

The decoding results of albedo, FVC, and GNDVI were categorized into five levels using the natural interruption grading method. These levels corresponded to regions of low value, second-lowest value, middle value, second-highest value, and high value. The statistical results of the three desertification risk characterization factors in the study area from 2000 to 2020 are shown in

Table 4. Albedo/FVC/GNDVI changes in the study area from 2000 to 2020 (Area unit: km²).

Index	Index Characteristic Area	2000	2005	2010	2015	2020	Change (2000–2020)
albedo	Low-value area	3510.14	4074.51	4074.51	2972.09	4709.45	1199.31
	Sub-low-value area	9047.52	8939.77	8939.77	8552.35	13,568.38	4520.86
	Medium-value area	7347.27	7121.01	7121.01	8580.24	1211.53	−6135.74
	Sub-high-value area	708.25	556.44	556.44	546.41	868.47	160.22
	High-value area	175.09	96.54	96.54	137.17	430.43	255.35
FVC	Low-value area	3673.87	3609.20	3616.48	3530.59	3440.58	−233.29
	Sub-low-value area	2321.58	2353.00	2233.65	2468.57	2410.37	88.79
	Medium-value area	2946.44	2907.46	2803.84	3239.42	3187.22	240.78
	Sub-high-value area	5240.73	5382.02	5096.53	5051.34	4852.04	−388.69
	High-value area	6605.64	6536.58	7037.75	6498.34	6898.06	292.42
GNDVI	Low-value area	2614.91	2720.94	2808.82	2370.66	1471.06	−1143.85
	Sub-low-value area	2437.63	2296.33	2334.93	2669.16	2655.66	218.03
	Medium-value area	2961.21	3002.78	2800.12	3316.57	3125.30	164.09
	Sub-high-value area	5766.77	6132.67	5618.69	5821.94	5741.61	−25.16
	High-value area	7007.74	6635.54	7225.70	6609.93	7794.63	786.89

. According to these results, albedo exhibited a significant increase in the low-value area and sub-low-value area, a significant decrease in the medium-value area, and a non-significant increase in the sub-high-value and high-value areas. The total area of change for the five characteristic zones was 1199.31 km², 4520.86 km², −6135.74 km², 160.22 km², and 255.35 km², respectively. Albedo is a measure of the amount of solar radiation reflected by a surface. The significant increase in the low-value and sub-low-value zones from 2000 to 2020 suggested changes in subsurface conditions in the study area during this time. The decrease in albedo may be due to snow melt and surface greening, which are negatively correlated with surface albedo. These factors were likely the main cause of the observed decrease in albedo. Between 2000 and 2020, there was a significant decrease in the low-value zone for both FVC and GNDVI in the study area, accompanied by an increase in the medium-value zone and the high-value zone. The FVC low-value area decreased from 17.67% to 16.55%, which represented a decrease in area of 233.29 km². In contrast, the sub-low-value, medium-value, and high-value areas increased by 88.79 km², 240.78 km², and 292.42 km², respectively. Between 2000 and 2020, the low-value area of GNDVI decreased by 1143.85 km², while the high-value area increased by 786.89 km². The trend in the low-value area of albedo showed an initial increase, followed by a decrease, and ending with an increase. This same inter-annual trend was observed in the high-value areas of FVC and GNDVI, which corresponded to the trend in albedo. The correlation between remote sensing indicators and desert environmental risk status suggested that the region’s environmental risk should follow a decreasing, then increasing, and finally decreasing trend over time.

The spatial distribution of the remote sensing indicators is shown in Figure 5. The study area’s albedo distribution is concentrated in low- and sub-low-value areas, while high- and sub-high-value areas are situated primarily in the northwest of Menyuan County and southwest of Qilian County. The study area’s elevation distribution was combined with the spatial distribution of FVC and GNDVI, which revealed significant differences in spatial distribution. There was a decreasing trend from southeast to northwest, due mainly to vegetation conditions and hydrothermal characteristics. The statistics of FVC and GNDVI’s elevation bands indicated that the vegetation cover in the study area was characterized by a trend of increasing and then decreasing with the elevation. The maximum vegetation cover was observed in the elevation range of 2700~3200 m, which is the primary distribution area of grassland and woodland. The cold desert zone and the glacial distribution area above 3800 m in elevation were characterized mainly by a high-value area of albedo and low-value areas of FVC and GNDVI. Between 2000 and 2020, the regions that experienced an increase in high values of albedo and low values of FVC and GNDVI were concentrated mainly in specific areas. These areas are particularly vulnerable to soil erosion due to the effects of glacial/snow melt and rainfall erosion, which are exacerbated by warming and humidification. These factors contributed to an increased risk of desert environmental evolution in the region. Despite these challenges, the spatial distribution of all three remote sensing indicators suggested a low risk of desertification.

3.1.2. Variation in Quantitative Indicator Factor Cloud Model Characteristics

Based on the image metadata extracted from the raster data of albedo, FVC, and GNDVI, respectively, their cloud feature parameters in 2000, 2005, 2010, 2015, and 2020 were calculated by data pre-processing using a cloud model. The results are shown in

Table 5. Albedo, FVC, and GNDVI cloud feature values in 2000, 2005, 2010, 2015, and 2020.

Year	Albedo	FVC	GNDVI
2000	(0.8582, 0.0489, 0.0192)	(0.6129, 0.3670, 0.1110)	(0.6328, 0.2566, 0.0820)
2005	(0.8806, 0.0479, 0.0201)	(0.6156, 0.3571, 0.0921)	(0.6573, 0.2548, 0.0835)
2010	(0.8690, 0.0444, 0.0105)	(0.6253, 0.3674, 0.1187)	(0.5187, 0.2220, 0.0659)
2015	(0.7885, 0.0968, 0.0324)	(0.6062, 0.3541, 0.1067)	(0.5535, 0.2065, 0.0658)
2020	(0.8246, 0.0883, 0.0297)	(0.6252, 0.3464, 0.1029)	(0.5889, 0.2122, 0.0577)

and Figure 6. The cloud characteristics parameters of all three quantitative indicators, albedo, FVC, and GNDVI, showed that the desertification risk states characterized by them all showed a trend of increasing, then decreasing, and finally rising over the period 2000–2020, and that the cloud droplet distribution dispersion was low, with He/En < 1/3. The years 2005 and 2015 showed two inflection points for albedo and GNDVI, respectively, while FVC showed inflection points in 2010 and 2015. This phenomenon can be attributed to the fact that FVC is based on the year-to-year maximum NDVI, which is less sensitive to areas with dense vegetation compared with GNDVI and albedo. This also explains the overall insignificant changes in the FVC cloud feature parameters. In relation to the central distribution of cloud features, the three indicators showed different results. The highest Ex was observed in relation to albedo, which suggested a lower risk of desertification. FVC’s Ex fell between albedo and GDNVI, indicating a moderate desertification risk. The highest desertification risk was indicated by GNDVI’s Ex. The combined indicators of albedo, FVC, and GNDVI were significant for remote sensing in determining desertification risk. Albedo directly indicates the ability of the ground to absorb and reflect solar radiation. In the study area, desertified areas, bare land, and snow-covered areas have strong reflectance, while areas with vegetation cover have low reflectance. Therefore, albedo is a better indicator for visualizing desertification risk status. FVC and GNDVI indirectly reflect desertification risk through vegetation health and growth, and they provide better results during the middle of the plant growth stage. However, at the beginning and end of the growing season, they may be over-indicated by chlorophyll content images in the vegetation, leading to false indications of desertification risk.

3.2. Qualitative Indicators of Factor Cloud Characteristics

The CM-AHP model calculated 32 indicators and 8 components to obtain cloud distinctive parameters of stress, state, and response in different years by pre-processing questionnaire responses.

Table 6. Cloud feature values for Pressure/State/Response from in 2000, 2005, 2010, 2015, and 2020.

Year	Pressure	State	Response
2000	(0.8323, 0.0323, 0.0085)	(0.8123, 0.0297, 0.0068)	(0.3308, 0.0301, 0.0051)
2005	(0.8029, 0.0288, 0.0058)	(0.7734, 0.0255, 0.0092)	(0.4021, 0.0214, 0.0051)
2010	(0.7719, 0.0255, 0.0064)	(0.7767, 0.0221, 0.0049)	(0.4611, 0.0343, 0.0077)
2015	(0.7323, 0.0282, 0.0056)	(0.7419, 0.0250, 0.0059)	(0.5757, 0.0413, 0.0082)
2020	(0.6757, 0.0395, 0.0115)	(0.7205, 0.0356, 0.0094)	(0.7557, 0.0337, 0.0081)

and Figure 7 indicate the projected cloud characteristic parameters of stress drivers in 2000, 2005, 2010, 2015, and 2020: 0.8323, 0.8029, 0.7719, 0.7323, and 0.6757, respectively. This trend implies that stress-related impacts, notably climate, natural hazard, and grassland carrying, were growing in the research area between 2000 and 2020 and positively affected the desertification risk enhancement. The expected cloud characteristic parameters for state drive in 2000, 2005, 2010, 2015, and 2020 were 0.8123, 0.7734, 0.7767, 0.7419, and 0.7205, respectively. These values showed a slow decreasing trend, indicating a decreasing trend of state drive effect on habitat quality and environmental quality. However, the decreasing trend was not significant based on the cloud characteristic expectations. All values belonged to the light desertification level. The response-driven effect, which was centered on ecological governance and policy guidance, showed the most significant change. The cloud characteristic parameter expectations for this effect were 0.3308, 0.4021, 0.4611, 0.5757, and 0.7557 in 2000, 2005, 2010, 2015, and 2020, respectively, indicating a clear increasing trend. This suggested that the local government had been intensifying its efforts towards desertification control between 2000 and 2020. By combining the affiliation between the indicator layer and the driver layer, this trend of improvement was reflected mainly in ecological investment, rodent and pest control, forest and grass restoration, soil and water conservation, ecological publicity and education, mine development control, protected area construction, and improvements in the rotational grazing system.

3.3. Comprehensive Assessment of Desertification Risk State Cloud Model Features

The CM-AHP model was used to calculate and analyze the qualitative and quantitative cloud feature parameters and their corresponding cloud model weights. This process was combined with the one used to construct the desertification risk state cloud model (Figure 2). The questionnaire was used to calculate the qualitative cloud feature parameters, while the metadata extraction of remote sensing indicators was used to calculate the quantitative cloud feature parameters. The results are shown in

Table 7. Cloud feature values for C^Qual, C^Quan, and C^Com from in 2000, 2005, 2010, 2015, and 2020.

Year	CQual	CQuan	CCom
2000	(0.6151, 0.0310, 0.0068)	(0.7641, 0.1468, 0.0480)	(0.6896, 0.0889, 0.0274)
2005	(0.6275, 0.0252, 0.0061)	(0.7830, 0.1441, 0.0453)	(0.7052, 0.0846, 0.0257)
2010	(0.6400, 0.0287, 0.0067)	(0.7502, 0.1376, 0.0441)	(0.6951, 0.0832, 0.0254)
2015	(0.6649, 0.0332, 0.0067)	(0.7050, 0.1572, 0.0525)	(0.6872, 0.0992, 0.0296)
2020	(0.7171, 0.0365, 0.0098)	(0.7376, 0.1595, 0.0486)	(0.7273, 0.0980, 0.0292)

Note: In this table, C^Qual, C^Quan, and C^Com represent the qualitative, quantitative, and comprehensive cloud feature values, respectively.

. The qualitative cloud feature parameters for the years 2000, 2005, 2010, 2015, and 2020 were as follows: C^Qual(0.6151, 0.0310, 0.0068), C^Qual(0.6275, 0.0252, 0.0061), C^Qual(0.6400, 0.0287, 0.0067), C^Qual(0.6649, 0.0332, 0.0067), and C^Qual(0.7171, 0.0365, 0.0098). The expected Ex of the qualitative cloud characteristics parameters indicated that the overall risk of desertification in the study area had been continuously and steadily decreasing from 2000 to 2020, with a more pronounced change in the improvement trend after 2010. This result suggested that local pastoralists had generally perceived a positive trend in the risk status of desertification in the study area from 2000 to 2020. The quantitative cloud feature parameters of C^Quan(0.7641, 0.1468, 0.0480), C^Quan(0.7830, 0.1441, 0.0453), C^Quan(0.7502, 0.1376, 0.0441), C^Quan(0.7050, 0.1572, 0.0525), and C^Quan(0.7376, 0.1595, 0.0486) were recorded for the years 2000, 2005, 2010, 2015, and 2020. Based on the expected Ex of the quantitative cloud characteristic parameters, the study area’s desertification risk exhibited a decreasing trend from 2000 to 2005, followed by an increasing trend until 2015, and then a decreasing trend once again until 2020. The identified temporal inflection points were in 2005 and 2015, which aligned with the findings of the previous paper on desertification risk using albedo and GNDVI. The fusion of the two cloud feature parameters resulted in composite cloud figures for desertification risk status for the years 2000, 2005, 2010, 2015, and 2020, with the respective values of C^Com(0.6896, 0.0889, 0.0274), C^Com(0.7052, 0.0846, 0.0257), C^Com(0.6951, 0.0832, 0.0254), C^Com(0.6872, 0.0992, 0.0296), and C^Com(0.7273, 0.0980, 0.0292). The trends of the integrated cloud numerical features aligned with those of the quantitative cloud numerical features, showing a decreasing trend followed by an increasing trend.

The qualitative, quantitative, and integrated cloud maps are shown in Figure 8. The qualitative cloud feature parameters revealed that the distribution of cloud droplets ranged from a moderate risk of desertification to a slight chance of desertification, gradually transitioning to a remote desertification state. Among the qualitative, quantitative, and composite cloud feature distributions, the qualitative cloud feature distribution had the smallest En value. Additionally, the cloud droplets were more densely concentrated on the cloud map, which served as reverse validation of the questionnaire design and the survey results’ validity. Between 2000 and 2020, the quantitative cloud feature parameters indicated that cloud droplets fell between slightly desertified and non-desertified, oscillating between these two states. The results of the quantitative cloud characterization suggested a lower risk of desertification when compared with the results of the qualitative cloud characterization. However, the cloud droplet distribution was more discrete, possibly due to the compatibility of different remote sensing indicators and the relatively small number of extracted random points. Nevertheless, all the cloud droplet distributions fell within the effective range, which was less than He/En < 1/3. The cloud droplet distribution in the composite cloud feature C^Com was concentrated mainly in the rank distribution points of slight desertification, and the overall desertification risk remained stable, with minor changes between 2000 and 2015. However, there was a transition trend from slight desertification to non-desertification from 2015 to 2020.

4. Discussion

4.1. Spatial and Temporal Changes in the Risk Profile of Desertification

The land use/cover data helped us to understand the changes in desertification (

Table 8. Land use/cover changes in the study area from 2000 to 2020 (Area unit: 10³ km²).

Land Use/Cover Type	Cropland	Forest	Shrub	Grassland	Water	Snow/Ice	Barren
2000	551.24	887.12	320.38	16,449.23	23.99	264.86	2291.09
2005	568.61	873.92	269.84	16,466.57	53.63	471.74	2083.62
2010	549.54	931.50	184.34	16,631.19	44.83	489.23	1957.37
2015	507.61	940.39	166.08	16,434.14	40.45	391.91	2307.30
2020	487.07	922.58	226.09	16,281.70	57.68	274.51	2538.03
Change (2000–2020)	−64.18	35.46	−94.29	−167.53	33.69	9.66	246.94

). The study area was largely dominated by grassland and barren lands, which together accounted for more than 90% of the total area (Figure 9). The barren land in the study area was composed primarily of sandy land, Gobi, and saline land, making it a high-risk indicator for desertification [69]. This land type posed a significant potential source of desertification risk and warrants further discussion. The barren land was concentrated in the central and western regions of Qilian County, as well as the northwestern part of Menyuan County. In addition to the aforementioned types of land, the primary land types in the study area were cropland, forest, shrub, and glacier/snow. Water bodies, wetlands, and impervious were present in very small areas. The altitude strip revealed that the proportion of land types varied across different altitude gradients. Below 3700 m elevation, there was little to no distribution of glacier/snow. Forests were found mainly in the range of 2700–3200 m above sea level, likely due to the challenging terrain and steep slopes at higher elevations. Between 3200 m and 3700 m, the dominant land types were shrub and cropland, while impervious and wetlands were present in small areas at lower elevations [70,71]. Grasslands had a wide range of distribution, spanning almost all elevations except those above 4700 m [72]. Barren lands were located predominantly in the alpine desert, with elevations above 3700 m. These areas were characterized by steep topography and sparse vegetation. They were transitional zones between grasslands and glacier/snow [73]. Additionally, a small area of barren lands was found at altitudes ranging from below 2700 m to 2700–3200 m. Results from field surveys indicated that barren lands in these areas were caused primarily by grassland degradation. In Ref. [74], research on grassland ecosystems in the Qilian Mountains revealed that desert grasslands, being highly sensitive to temperature and precipitation, were more prone to degradation due to climate warming and the frequent occurrence of extreme weather events.

While the structure of land-use types in the study area, dominated by grassland and barren, remained constant from 2000 to 2020, changes in the scope of different land-use types were observed over time (Table 8). These changes suggested a range of transformations in the desert environment [74,75]. Between 2000 and 2005, the study area experienced a decrease in barren area and an increase in grassland and glacier/snow area. These changes were reflected in the qualitative, quantitative, and comprehensive assessments of desertification risk status. The increase in cropland and the decrease in forest and shrub suggested that human activity was relatively intense during this period. In Ref. [76], the study indicated that during this timeframe, the Hexi region of China, including the study area, witnessed a significant expansion of arable land. This expansion was driven primarily by the growing demand for agricultural production. Between 2005 and 2015, cropland in the study area continued to decrease, while forests expanded, particularly in the southwest and central regions of Menyuan County. This expansion could be attributed to various ecological projects implemented by the government during this period, such as the “Return of Cultivated Land to Forests”, the “Natural Forest Protection Project”, and the “Three North Protection Forest Project” [77,78]. In Ref. [71], the study on the trend of vegetation restoration in the Qilian Mountains highlighted that the implementation of various projects was successful in preventing the further of degraded grassland areas. Additionally, these initiatives played a crucial role in promoting positive development in vegetation restoration efforts. The grasslands in the study area remained relatively stable, while barren land expanded, and ice/snow experienced a decrease. The primary reason behind this phenomenon was the temperature increase during the study period [79]. In Ref. [80], according to the study, negative glacier mass was the result of temperature fluctuations and a decrease in precipitation recharge during the snowmelt season. In Ref. [81], the findings of this study supported this theory and indicated that the ecosystems of glaciers and bare ground at an altitude of approximately 4000 m have been significantly impacted by the rise in temperature. This time period was also thought to be the main contributor to the increased risk of desertification in the study area.

During the period of 2015–2020, the study area experienced the highest risk of desertification. This was due to the significant conversion of snow/ice to barren, caused primarily by grassland degradation and the melting of snow/ice, as well as the rapid conversion of the snow/ice area to barren due to increased temperatures. These factors resulted in an expansion of barren and an increase in the risk of desertification in areas with retreating snowlines [74,82]. Additionally, in grassland areas with high topographic relief, the risk of soil erosion was exacerbated due to the increased erosive power of rainfall in the context of warming and humidification [74]. Despite increasingly severe changes in climatic conditions, the establishment of the Qilian Mountains National Park effectively implemented a series of ecological protection measures. This led to significant positive effects on the ecological restoration of mainly grasslands and forests, particularly at low altitudes and in areas with gentle terrain. The park’s measures also improved the hydrothermal conditions and supported the overall restoration efforts [83]. In Refs. [81,84], this was supported by every one of their findings, which also highlighted the excellent contribution made by the creation of the Qilian Mountains National Park to ecology protection.

4.2. Factors Influencing the Evolution of Desertification Risk

We calculated the weights of 32 indicators and 8 factors using the CM-AHP model (

Table 9. Comparison of indicator weights.

Index	E			F			G			H			Wi (Exi, Eni, Hei)	Grade
	Ex	En	He	Ex	En	He	Ex	En	He	Ex	En	He
X	0.0411	0.0409	0.0409	0.1065	0.1091	0.1091	0.2097	0.2003	0.2003	0.0712	0.0708	0.0708
E1	0.0895	0.0927	0.0927										(0.0037, 0.0038, 0.0038)	32
E2	0.3533	0.3480	0.3480										(0.0145, 0.0142, 0.0142)	23
E3	0.2040	0.2113	0.2113										(0.0084, 0.0086, 0.0086)	30
E4	0.3533	0.3480	0.3480										(0.0145, 0.0142, 0.0142)	24
F1				0.3505	0.3446	0.3446							(0.0373, 0.0376, 0.0376)	9
F2				0.2663	0.2618	0.2618							(0.0284, 0.0286, 0.0286)	12
F3				0.2663	0.2618	0.2618							(0.0284, 0.0286, 0.0286)	13
F4				0.1168	0.1318	0.1318							(0.0124, 0.0144, 0.0144)	26
G1							0.2908	0.2854	0.2854				(0.0610, 0.0572, 0.0572)	5
G2							0.2908	0.2854	0.2854				(0.0610, 0.0572, 0.0572)	6
G3							0.1276	0.1437	0.1437				(0.0267, 0.0288, 0.0288)	16
G4							0.2908	0.2854	0.2854				(0.0610, 0.0572, 0.0572)	7
H1										0.3436	0.3381	0.3381	(0.0245, 0.0239, 0.0239)	17
H2										0.1984	0.2053	0.2053	(0.0141, 0.0145, 0.0145)	25
H3										0.1145	0.1185	0.1185	(0.0082, 0.0084, 0.0084)	31
H4										0.3436	0.3381	0.3381	(0.0245, 0.0239, 0.0239)	18
Index	I			J			K			L			Wi (Exi, Eni, Hei)
	Ex	En	He	Ex	En	He	Ex	En	He	Ex	En	He
X	0.0714	0.0788	0.0788	0.0714	0.0788	0.0788	0.2143	0.2106	0.1053	0.2143	0.2106	0.1053
I1	0.4121	0.4121	0.4121										(0.0294, 0.0325, 0.0325)	11
I2	0.3131	0.3131	0.3131										(0.0224, 0.0247, 0.0247)	19
I3	0.1374	0.1374	0.1374										(0.0098, 0.0108, 0.0108)	27
I4	0.1374	0.1374	0.1374										(0.0098, 0.0108, 0.0108)	28
J1				0.2908	0.2854	0.2854							(0.0208, 0.0225, 0.0225)	20
J2				0.2908	0.2854	0.2854							(0.0208, 0.0225, 0.0225)	21
J3				0.1276	0.1437	0.1437							(0.0091, 0.0113, 0.0113)	29
J4				0.2908	0.2854	0.2854							(0.0208, 0.0225, 0.0225)	22
K1							0.2908	0.2854	0.2854				(0.0623, 0.0601, 0.0301)	2
K2							0.1276	0.1437	0.1437				(0.0273, 0.0303, 0.0151)	14
K3							0.2908	0.2854	0.2854				(0.0623, 0.0601, 0.0301)	3
K4							0.2908	0.2854	0.2854				(0.0623, 0.0601, 0.0301)	4
L1										0.1424	0.1427	0.1427	(0.0305, 0.0301, 0.0150)	10
L2										0.2467	0.2601	0.2601	(0.0529, 0.0548, 0.0274)	8
L3										0.4855	0.4775	0.4775	(0.1040, 0.1006, 0.0503)	1
L4										0.1254	0.1197	0.1197	(0.0269, 0.0252, 0.0126)	15

W_i (Ex_i, En_i, He_i) represents the combined cloud weight results for each factor, while all the remaining letters have the same meaning as in Figure 4.

). Based on the ranking results of the weights, we identified the factors that had the greatest impact on the study area at the factor level. These were primarily ecological governance (K), policy guidance (L), natural disasters (G), and climate (F). Ecological governance and policy guidance are response-driven, while natural disasters and climate were pressure-driven. Our results suggested that the risk of desertification in alpine mountains was influenced mainly by the antagonistic effect between stress load and response feedback within the environment. In Ref. [48], the study highlighted the substantial improvement in the ecological environment of the study area, which could be attributed to the robust support from government departments in terms of ecological management and macro-policy direction. In addition, in Refs. [74,85], the study’s findings consistently emphasized that the degradation of ecosystems in the study area was the outcome of severe weather conditions, climate change, and natural disasters. These investigations provided substantial evidence to support our conclusions.

The results of the weighting analysis at the indicator level revealed the driving factors influencing the evolution of desertification in semi-arid alpine mountains. The factors were ranked from largest to smallest: protected area construction (L₃), eco-investment (K₁), forest and grass restoration (K₃), soil and water conservation (K₄), extreme precipitation (G₁), extreme drought (G₂), flood disaster (G₄), control of mine development (L₂), annual precipitation (F₁), publicity and education (L₁), vegetable coverage (I₁), annual evaporation (F₂), annual temperature (F₃), rodent management (K₂), rotational grazing (L₄), frequency of earthquakes (G₃), stock capacity (H₁), soil erosion (H₄), grassland grass amount (I₄), rainfall erosion (J₁), glacier/snow melt erosion (J₂), freeze–thaw erosion (J₃), slope (E₂), topographic relief (E₄), rodent and insect pests (H₂), temperature difference (F₄), soil fertility (I₂), biodiversity science (I₃), grassland degradation (J₄), aspect (E₃), wildlife (H₃), and elevation (E₁).

It was evident that the construction of protected areas, the increased investment in ecological management, and the implementation of forest and grass restoration as well as soil and water conservation projects had a significant inhibiting effect on the expansion of desertification. This was the main reason why the qualitative cloud characteristic state continued to improve from 2000 to 2020 [77,78,83]. During this period, the local government increased its control over these areas, mainly through the establishment of Qilian Mountains National Park, the restoration of cultivated land to forest, natural forest protection, and the implementation of projects such as the Three North Protection Forest Project [86]. The improvement in these areas has become more evident since the establishment of Qilian Mountains National Park in 2017. Simultaneously, the continuous improvement in mine development control, ecological education, rodent and pest management, and rotational grazing systems also had a positive effect on the control of desertification. These positive results were evident in the continuous improvement in the response-driven dimension of the qualitative cloud maps. The dominating factors that increased the risk of desertification in the study area were climatic factors and natural hazards, such as extreme precipitation, extreme drought, flood hazards, annual precipitation, annual evapotranspiration, and average annual temperature [74,79,82]. One example of the negative impacts of desertification was the increasing rate of glacier/snow melt caused by the rising temperatures. This intensified the occurrence of ice melt and freeze–thaw erosion, leading to a decrease in the organic matter content of the soil and erosion, which harmed the vegetation growth. Additionally, the melting of glaciers/snow also directly transformed glacial–snow-covered territory into barren land, increasing the risk of local desertification expansion [86].

The frequency of extreme precipitation, drought, and earthquakes was also on the rise in the region, posing a significant threat to the soil and water environment. This destabilization had a negative impact on the ecological environment in the area. With an increasing trend of extreme weather events, the risk of soil erosion, flooding, and landslides also increased, leading to further environmental degradation. It is important to take measures to mitigate the effects of these extreme events and promote sustainable practices to ensure the long-term health of the ecosystem [87,88]. Rodent infestation also played a significant role in grassland degradation and the increased risk of desertification. The severe infestation of rodents not only reduced grassland productivity but also led to the excavation of holes and mounds, causing soil organic matter and parent material to be pushed to the surface. This eventually resulted in the formation of secondary bare ground after wind or water erosion, which weakened the resistance and resilience of the grassland ecosystems [85,89]. The proliferation of wild animals in the region could be attributed to the development of the wildlife protection policies. Specifically, herbivorous animals, such as wild deer and rock sheep, experienced a rapid population growth due to favorable growth conditions and the lack of natural predators. Unfortunately, this population explosion had a detrimental effect on the grasslands in the rangeland, resulting in extreme pasture degradation [85,90]. This degradation increased the risk of desertification, posing a serious threat to the ecosystem.

Overall, the study area is experiencing a significant increase in the risk of desertification due to changing climatic conditions. However, human activities are having a positive impact on mitigating this phenomenon and controlling its occurrence.

4.3. Advantages and Stability of the Cloud Model Assessment

Qualitative and quantitative cloud characteristics exhibited some variability over time. Notably, the qualitative clouds showed a trend of continuous elevation, particularly after 2015, coinciding with the establishment of Qilian Mountains National Park. The quantitative cloud showed a trend of increasing, then decreasing, and finally increasing. This pattern was determined by the attributes represented by the two cloud characteristics. Qualitative clouds, on the other hand, expressed mainly the intuitive feelings of the herders. These clouds focused on factors such as climate, environment, living comfort, economic income, and policy changes within their living area. The herders tend to concentrate their living range in areas with relatively good grass-bearing status, low altitude, and relatively flat terrain. Based on the results of the remote sensing scale analysis, the areas with higher altitude and greater topographic relief were found to have a higher risk of desertification. These areas also experienced harsh water–soil–thermal–environmental conditions, which were further detailed in the quantitative cloud. Qualitative clouds, on the other hand, provided insight into the role of human activities in controlling the risk of desertification in the region. They highlighted factors such as overgrazing, deforestation, and improper land-use practices. By identifying these factors, policymakers can take appropriate measures to mitigate desertification. While quantitative cloud data can provide an objective assessment of desertification risk, it often overlooks the impact of human activities on this phenomenon. However, by integrating both quantitative cloud data and information on human activities using an improved cloud model, we were able to overcome the limitations of each type of data and achieve more accurate and objective assessments of desertification risk status. This approach greatly enhanced the usefulness of the data and improved our ability to accurately assess and monitor the desertification risk over time.

Entropy (En) and superentropy (He) are crucial parameters for assessing the reliability of the cloud model evaluation results. The dispersion of the cloud droplet distribution was directly determined by He, and the lesser its value, the better the model’s stability performance. The opposite was true: the more discrete the assessment model, the less trustworthy the assessment outcomes. The effect of changing En and He on the estimated stability for a constant Ex is shown in Figure 10. When He < En/3, 99.7% of the cloud droplets lay within the area defined by the two boundary curves, i.e., y = exp[−(x − Ex)²/(2(En − 3He)²)] (minimum boundary) and y = exp[−(x − Ex)²/(2(En + 3He)²)] (maximum boundary) [91]. As a result, the model evaluation for He < En/3 was exceptionally stable and accurate. However, when He > En/3, the cloud droplet distribution had very high spatial dispersion characteristics, indicating that the evaluation results were relatively unstable and that disclosing the risk features was questionable.

The cloud droplet distributions of the qualitative, quantitative, and composite clouds passed the En/3 test in all years. The qualitative and comprehensive clouds had concentrated droplet distributions and exhibited better aggregation. In contrast, the quantitative clouds had dispersed droplet distributions and a higher degree of droplet atomization compared with the qualitative and comprehensive clouds. Considering the year 2000 as an example, it can be seen that all three distributions of cloud droplets fell within the standard line, which suggested that the assessment results were valid and reliable (Figure 11). In addition to albedo, FVC, and GNDVI, there are several other quantitative remote sensing indicators that can be used to characterize the risk of desertification in semi-arid alpine mountains. These include environmental factors such as soil moisture, surface temperature, meteorological conditions, and snow cover [35,36,92,93]. By analyzing these various indicators, researchers can gain a more comprehensive understanding of the factors that contribute to desertification in these regions. To improve the survey questionnaire, it is recommended to incorporate quantitative indicators and consider the actual conditions of the target study area. By enriching the survey’s perspective and dimensions, we can better understand the correlation mechanism among the economic income of farmers and herders, their production and lifestyle, population density, and desert environment. This will provide a more comprehensive understanding of the survey results [94,95]. To further improve the combination of quantitative and qualitative data in assessing desertification risk in semi-arid alpine mountains, the use of system dynamics and structural equation models can be beneficial. These models can aid in assessing and identifying the current status of the desertification risk, as well as predicting the future evolution using various climate models and scenarios [96]. By utilizing these models, we can systematically predict and assess dynamic changes in the ecosystem service functions and ecological security patterns, providing a more comprehensive understanding of the potential impacts of desertification.

5. Conclusions

This paper introduces a new theoretical framework for assessing the risk of desertification in semi-arid alpine regions. The framework aims to address the convergence between qualitative and quantitative evaluation methods. The study reveals that the risk of desertification in the study area experienced three distinct phases from 2000 to 2020, characterized by a decline, an increase, and another decline. However, the risk status remained stable overall, maintaining a light desertification level during the entire period. The primary cause of the increased risk of desertification in the region is the pressure-driven effect brought about by climate change and natural disasters. This effect is due mainly to the increased glacial and snow melt, rainfall erosion force, and freeze–thaw erosion force resulting from climate warming and humidification. As a result, the local expansion of desertification has been observed in high-altitude areas with significant topographic relief. On the other hand, response-driven enhancements, such as the construction of protected areas, forest restoration, soil and water conservation, and other environmental protection projects, have been effective in curbing the increase in desertification risk. These measures have been successful in controlling the phenomenon despite the adverse climatic conditions in the study area, demonstrating the positive impact of human activity.

The combination of topographical and climatic variables has resulted in high desertification risk areas located in the central and western portions of Qilian County and the northwestern region of Menyuan County, particularly in the alpine desert situated at altitudes exceeding 3700 m. Meanwhile, lower altitude areas are also at risk of desertification due to various factors, such as topography, rodent and insect infestation, wildlife, and poor livestock systems. These factors contribute to the overall risk of desertification in the region.

Although there has been a recent decrease in the risk status of desertification in the region, the risk of desertification is still expected to increase in the future due to the impact of climate change. Therefore, it is imperative to implement measures that will enhance the control of human activities on the ecological environment and strengthen ecological management to prevent the risk of future desertification expansion. The research findings have shed light on the evolution pattern and driving mechanism of the desertification risk status, providing a solid theoretical foundation for improving the stability of regional ecosystems in the semi-arid alpine mountains. Such measures are needed to ensure the sustainability of the region’s natural resources and secure the livelihoods of its inhabitants.