Bayesian Network Analysis: Assessing and Restoring Ecological Vulnerability in the Shaanxi Section of the Qinling-Daba Mountains Under Global Warming Influences

Abstract

Human activities, especially industrial production and urbanization, have significantly affected vegetation cover, water resource cycles, climate change, and biodiversity in the Qinling-Daba Mountain region and its surrounding areas. These activities contribute to complex and lasting impacts on ecological vulnerability. The Qinling Mountain region exhibits a complex interaction with human activities. The current research on the ecological vulnerability of the Qinling Mountain region primarily focuses on spatial distribution and the driving factors. This study innovatively applies the VSD assessment and Bayesian networks to systematically evaluate and simulate the ecological vulnerability of the study area over the past 20 years, which indicates that the integration of the VSD model with the Bayesian network model enables the simulation of dynamic relationships and interactions among various factors within the study areas, providing a more accurate assessment and prediction of ecosystem responses to diverse changes from a dynamic perspective. The key findings are as follows. (1) Areas of potential and slight vulnerability are concentrated in the Qinling-Daba mountainous regions. Over the past 20 years, areas of extreme and high vulnerability have significantly decreased, while areas of potential vulnerability and slight vulnerability have increased. (2) The key factors impacting ecological vulnerability during this period included industrial water use, SO₂ emissions, industrial wastewater, and ecological water use. (3) Areas primarily hindering the transition to potential vulnerability are concentrated in well-developed small urban regions within basins. Furthermore, natural factors like altitude and temperature, which cannot be artificially regulated, are the major impediments to future ecological restoration. Therefore, this paper recommends natural restoration strategies based on environmental protection and governance strategies that prioritize green development as complementary measures. The discoveries of the paper provide a novel analytical method for the study of ecological vulnerability in mountainous areas, offering valuable insights for enhancing the accuracy of ecological risk prediction, fostering the integration of interdisciplinary research, and optimizing environmental governance and protection strategies.

1. Introduction

In recent years, climate change has presented numerous challenges to natural environments and human social systems, particularly affecting ecologically complex mountainous regions [1,2,3]. Mountain areas, with their distinctive ecological environments and intricate human–land interactions, have become global ecological hotspots [4]. China, with its extensive mountainous terrain, is home to some of the world’s most complex and diverse mountain ecosystems. Studying China’s mountain ecosystems not only enhances our understanding of global ecological processes and biodiversity conservation but also holds significant practical implications for addressing climate change, conserving water resources, preventing natural disasters, supporting regional economic development, and sustaining the livelihoods and cultures of local communities. Moreover, ecological research in China’s mountainous regions offers valuable insights for global mountain ecological protection and management, particularly in achieving sustainable development goals and advancing ecological civilization. Thus, an in-depth study of China’s mountain ecosystems is not only of scientific importance but also of pressing practical necessity [5,6,7].

The Qinling-Daba Mountain Area serves as a representative example of China’s mountainous regions. With its abundant water resources and extensive network of rivers, this area provides essential water sources for the northern and southern slopes of the Qinling Mountains, playing a critical role in maintaining regional ecological balance, ensuring safe drinking water, and supporting agricultural and industrial water needs through its water conservation and soil retention functions. The Qinling-Daba Mountain Area’s rich biodiversity and extensive forest cover contribute significantly to soil and water conservation, climate regulation, and biodiversity habitat provision. However, in recent years, rapid population growth and economic expansion have driven sharp increases in water demand, particularly for agriculture and industry, placing strain on both water supply and ecosystem health [3]. Although industrial activities are key to economic growth, they have led to serious environmental pollution in certain areas due to wastewater and emissions, posing threats to water quality and ecosystem integrity [8,9,10].

Agricultural development in this region faces limitations due to natural and infrastructural constraints, with a lack of funding and slow progress in modernization, including limited advancement in green, ecologically sustainable agriculture. Furthermore, there is a pronounced imbalance in regional economic development; some areas lack necessary ecological risk prevention and management measures due to insufficient resources. The absence of such measures has negatively impacted residents’ quality of life and created challenges for the region’s sustainable development [11].

The Qinling-Daba Mountain Area is a vital ecological function zone and an essential component of China’s ecological security framework. While economic development is generally low in this region, the interrelationship between ecosystem services and resident welfare is complex, with notable conflicts among ecological, resource, and socio-economic development goals, underscoring issues of developmental imbalance. Recent years have seen ecological civilization elevated to a national strategic priority in China, with issues related to ecological protection and sustainability receiving increased academic and public attention. Against this backdrop, this article explores the ecological environment and human–land interactions in the Qinling-Daba Mountain Area and surrounding regions through the lens of ecological vulnerability [12].

The Qinling-Daba Mountain region exemplifies the intricate interactions between water resources, ecosystems, and socio-economic development. This region is a critical water resource for both the northern and southern slopes of the Qinling Mountains and exerts a profound influence on the ecological health of these mountainous areas, primarily through the socio-economic activities on both sides. Consequently, this region faces the dual responsibility of safeguarding ecological integrity while fostering economic development [13,14].

Recently, with the elevation of ecological civilization to a national strategy in China, issues related to ecological protection and sustainability have gained increasing academic attention and public recognition [15]. Therefore, this paper examines ecological sustainability in the study area through the lens of ecological vulnerability.

Ecological vulnerability serves as a crucial indicator for assessing ecosystem health in the context of human–environment interactions [16,17]. The Intergovernmental Panel on Climate Change (IPCC) defines ecological vulnerability as the susceptibility of the ecological environment to damage due to limited coping and adaptive capacity when exposed to internal or external disturbances [17]. Currently, the impact of climate change on the ecological vulnerability of mountainous regions is a prominent research topic [3]. Focusing on the social–ecological system from a coupled perspective and examining the interactions between densely populated areas and the Qinling Mountain region have emerged as key research directions for assessing the social–ecological vulnerability of this area. Moreover, studies on ecological vulnerability in this region primarily focus on evaluating ecological vulnerability and analyzing its driving mechanisms [5,18,19,20,21,22]. Accurately constructing indicators and assigning weights is crucial for assessing vulnerability. Several models have been employed in this context, including the pressure–state–response (PSR) model [23], the sensitivity–recovery–pressure (SPR) model [24], and the vulnerability–sensitivity–adaptability (VSD) model [25]. Notably, the VSD model constructs an evaluation index system for ecological vulnerability by systematically decomposing both the target layer and the indicator layer. This model can integrate evaluation indicators tailored to the characteristics of the system being evaluated, thereby revealing the interactions between natural and human elements and distinguishing it from other vulnerability assessment models. The VSD model has been widely applied and recognized in studies concerning human–environmental systems [20].

The diverse geography and rapid urbanization of the study area create complex couplings among social, ecological, and economic systems [11]. Further research is needed to examine ecological vulnerability in the context of climate change, particularly the impact of densely populated human activities on the region. Additionally, uncertainties persist regarding water resource processes, and the specific mechanisms and extent of impacts from agriculture and industry in the study area remain unclear [26]. Traditional ecological vulnerability models, which rely on empirical judgments to assess levels and thresholds, face significant challenges due to their complexity and uncertainty [27]. Bayesian networks are probabilistic graphical models that represent the conditional dependencies among variables using Bayes’ theorem. These models offer a significant advantage in handling uncertainty due to their intuitive graphical structure and powerful probabilistic reasoning capabilities. They are widely applied in disaster research and water resource management [28].

Consequently, this study aims to utilize the VSD framework evaluation model to conduct a systematic spatiotemporal assessment of ecological vulnerability in the study area from 2000 to 2020. Based on this assessment, Bayesian network models will be applied to predict and respond to future changes in ecological vulnerability in the Qinling-Daba Mountain region and its surrounding areas from multiple perspectives [29]. This research introduces the uncertain thinking of machine learning, providing new exploratory pathways for addressing the complexities of the ecological environment in this area, as well as the impacts of frequent human activities and climate change. This approach aids in accurately capturing the dynamic changes and potential trends of ecological vulnerability, offering a scientific basis for formulating effective ecological protection strategies and promoting sustainable regional development.

2. Materials and Methods

2.1. Overview of the Study Area

The northern and southern slopes of the Qinling Mountains in Shaanxi cover an area of approximately 1.08 × 10⁵ km² (Figure 1). Divided by the Qinling Mountains as a watershed, the northern slope falls within the Yellow River Basin and is characterized by a temperate, semi-humid, and semi-arid monsoon climate. This region represents the central area for population, economy, politics, and culture in Shaanxi Province. Conversely, the southern slope lies within the Yangtze River Basin and is characterized by a subtropical humid monsoon climate, encompassing the Han River Plain and the Daba Mountain region. The northern slope of the Qinling Mountains has a well-developed hydrological system, with a dense river network and good water quality, making it suitable for industrial, agricultural, and domestic use. The Wei River, the largest tributary of the Yellow River, originates from this area. Its abundant water resources are critical to the population growth and development of the Guanzhong urban area, where over half of the Guanzhong Basin’s population relies on the Wei River as their primary source of water. Moreover, the ecological health of the Guanzhong Basin is closely tied to the Wei River’s water supply, positioning the northern slope of the Qinling Mountains as a crucial ecological barrier and functional zone for the basin’s sustainability. The southern slope’s topography is predominantly hilly, with longer river courses and ample water volume. This area is the source of key Yangtze River tributaries, such as the Jialing River, Han River, and Dan River, and serves as an essential water conservation area for the South-to-North Water Transfer Project. Annually, it provides a stable and high-quality surface water supply to numerous cities across the country, benefiting a broad range of residents and playing a vital role in securing water resources for downstream regions [30].

In recent years, the demand for water on the northern and southern slopes of the Qinling Mountains has surged due to the rapid development of the socio-economic landscape. The northern slope, influenced by seasonal rainfall, experiences significant fluctuations in surface runoff and has an inadequate capacity to retain water resources. Additionally, inadequate resource management and increasing pollution have further negatively impacted water usage on the northern slope. These changes suggest that the rising water demand in the Guanzhong Basin will face substantial challenges in the future.

2.2. Data Sources and Pre-Processing

The research data utilized in this article primarily include (1) basic geographical information on the northern and southern slopes of the Qinling Mountains, including county-level administrative boundaries and elevation data (DEM) derived from the Landsat 8 satellite, with a 30 m resolution; (2) climatic data, including temperature and precipitation, based on daily observations from national meteorological stations and obtained through interpolation techniques; (3) Normalized Difference Vegetation Index (NDVI) data from NASA’s MOD13Q1 vegetation index product, with a spatial resolution of 250 m and a temporal resolution of 16 days. Annual-scale NDVI data are processed via the maximum composite method; (4) socio-economic data, including gridded population and GDP data for 2000, 2005, 2010, 2015, and 2020. Land use data are primarily derived from Landsat 8 remote-sensing images through interpretation methods, combined with the classification standards of land use types from the Chinese Academy of Sciences’ Resource Data Cloud platform. Additionally, statistical data are included on ecological and agricultural water use, industrial water use, emissions of wastewater and exhaust, and industrial activities. To ensure consistency between remote-sensing data products and observational data, resampling and geometric correction are conducted, with spatial coordinates standardized to the WGS_1984 system. Some socio-economic yearbook data are spatially interpolated using the inverse distance weighting (IDW) method for spatial visualization [31] (

Table 1. Data sources and descriptions.

Type	Data	Parameter
Topography Elements	DEM	m
Natural Elements	Temperature	°C
	Precipitation	mm
	NDVI	/
Social Elements	Land Use	/
	POP	/
	GDP	108
	Agricultural Water Consumption	t
	Fiscal Revenue	108
	Resident Storage	108
	Food Production	t
	Industrial Wastewater Emissions	t
	Sulfur Dioxide Emissions	t
	Industrial Water Consumption	t
	Ecosystem Water Consumption	t

Note: Topographic data were obtained from GSCloud (http://www.gscloud.cn); nature element data were sourced from the Resource and Environment Data Cloud Platform (https://www.resdc.cn/); and social element data were retrieved from the Shaanxi Statistical Yearbook.

).

2.3. Conceptual Model

As an important ecological functional area in China, the study area is rich in natural resources, which not only provide the essential material basis for local human activities but also play a crucial role in maintaining regional ecological balance and biodiversity. However, human activities have significantly impacted the ecological environment of the Qinling-Daba Mountain region, forming a complex coupled system in which natural and anthropogenic factors interact and constrain each other. For example, agricultural expansion may lead to a decrease in forest coverage, adversely affecting soil and water conservation and carbon storage capacity; industrial development may cause water quality degradation, threatening the drinking water safety of downstream areas; and urbanization processes may result in the destruction of natural landscapes. Figure 2 illustrates this intricate conceptual relationship.

2.4. Ecological Vulnerability Calculations

The selection of indicators is essential for a scientific and accurate assessment of ecological vulnerability. When constructing an evaluation index system, it is necessary to comprehensively consider both natural environmental and anthropogenic factors and ensure that the chosen indicators are practical and comparable. The VSD model divides ecological vulnerability into three main dimensions: exposure, sensitivity, and adaptability [32]. Exposure represents the degree to which the ecosystems in the study area are subject to external disturbances or pressures, particularly from human activities. In recent years, human activities and urbanization processes in the study area have notably affected exposure. The extent of land development is a key indicator for measuring exposure, as different land use types have varied impacts on ecosystems. Additionally, pollution from industrial and agricultural activities directly influences the regional ecological environment. Consequently, the selection of exposure indicators should emphasize population activities and economic development. Sensitivity denotes the inherent responsiveness of ecosystems to external pressures. When choosing indicators, factors such as water resources and vegetation cover should be considered. Precipitation and temperature data can indicate regional climate conditions, while elevation affects topography and species distribution. Vegetation cover reflects biodiversity levels and water conservation capacity, effectively revealing the resource endowment characteristics of the study area. Adaptability describes the ecosystem’s capacity to respond to disturbances, adapt to pressures, and recover from impacts. This study selects indicators related to economic factors to measure adaptability, including residents’ income, living standards, and investment in production materials. These indicators reflect the area’s capacity to utilize resources and the intent to enhance residents’ well-being (Figure 3).

Furthermore, this study discerns the relationship between the indicator data and the criteria layer. If the data in the indicator layer are positively correlated with the criteria layer, this signifies that the indicator data are favorable, represented by a “+”. Conversely, negative indicator data are represented by a “−”. This study considers the accessibility and timeliness of the data, selecting 15 layers to construct the regional ecological vulnerability assessment framework, as illustrated in

Table 2. Indicator system of ecological vulnerability assessment in the study area.

Target Layers	Guideline Layers	Metric Layers	Indicators
Vulnerability	Exposure	Population	+
		GDP	+
		Industrial Wastewater Emissions	+
		Sulfur Dioxide Emissions	+
		Agricultural Water Consumption	+
		Industrial Water Consumption	+
		Ecosystem Water Consumption	−
		Land Use	+
	Sensitivity	NDVI	+
		DEM	+
		Temperature	−
		Precipitation	−
	Adaptation	Fiscal Revenue	+
		Resident Storage	+
		Food Production	+

Note: Land use is classified according to the intensity of human activities for development and utilization. System with unutilized land is classified as 1; forest land as 2; watersheds and grasslands as 3; agricultural land as 4; and urban land as 5.

. The ecological vulnerability index, based on the determined indicator weights for the research area, is calculated using the following formula:

$$\mathrm{V}=\mathrm{E}+\mathrm{S}-\mathrm{A}$$

where V represents the ecological vulnerability index, reflecting the overall vulnerability of the ecosystem; E stands for exposure, S for sensitivity, and A for adaptability. Exposure and sensitivity are positively correlated with ecological vulnerability, while adaptability is negatively correlated, serving as a compensatory mechanism. First, E, S, and A are derived by multiplying the standardized values of all indicator layers under the criteria layer by their respective weights and then summing the results. Next, after obtaining E, S, and A, the ecological vulnerability index V is calculated according to the formula and normalized. The relationship between ecological vulnerability and its grades is shown in

Table 3. Ecological vulnerability levels.

Vulnerability Index	Evaluation Level	Features
0–0.2	Potential Vulnerability	Ecosystems are stable and resistant to disturbance
0.2–0.4	Slight Vulnerability	Ecosystems are relatively stable and resilient to disturbance, and some areas may be slightly dysfunctional
0.4–0.6	Moderate Vulnerability	Ecosystems are more unstable and resilient to disturbance in some areas
0.6–0.8	High Vulnerability	Ecosystem instability and poor resilience to disturbance
0.8–1	Extreme Vulnerability	Ecosystems are highly unstable, and the region is highly destabilized and in need of urgent improvement

[33].

Determining the weights of relevant indicators is the foundational work for the evaluation of ecological vulnerability. Given the extensive study area in this paper, a spatial resolution of 2 km × 2 km was chosen to effectively capture the region’s key ecological characteristics. This resolution balances data processing and storage demands while providing a robust sample for machine-learning methods, which is essential for effective model training and performance enhancement. Thus, selecting this resolution reflects a comprehensive consideration of data precision and model training accuracy. The entropy weight method, which uses the information entropy of the indicators, is employed to determine the weights of each indicator [34]. This paper uses the range method to standardize all indicators, converting them into dimensionless values. The weights of the indicators are shown in

Table 4. Indicator weights.

Target	Guideline Layers	Metric Layers	Year
			2000	2005	2010	2015	2020
Vulnerability (V)	Exposure (E)	GDP	0.31	0.18	0.19	0.25	0.27
		Population	0.12	0.12	0.03	0.13	0.14
		Agricultural Water Consumption	0.03	0.04	0.03	0.03	0.03
		Industrial Water Consumption	0.08	0.09	0.1	0.08	0.03
		Ecosystem Water Consumption	0.0005	0.0005	0.0004	0.0003	0.0004
		Industrial Wastewater Emissions	0.06	0.1	0.08	0.04	0.06
		Sulfur Dioxide Emissions	0.08	0.09	0.08	0.06	0.03
		Land Use	0.015	0.016	0.015	0.012	0.012
	Sensitivity (S)	NDVI	0.017	0.024	0.03	0.15	0.14
		DEM	0.03	0.03	0.03	0.02	0.02
		Temperature	0.02	0.03	0.02	0.02	0.02
		Precipitation	0.01	0.01	0.014	0.012	0.005
	Adaptation (A)	Fiscal Revenue	0.088	0.12	0.11	0.1	0.12
		Resident Storage	0.1	0.14	0.12	0.08	0.1
		Food Production	0.02	0.02	0.024	0.017	0.013

.

For positively correlated indicators, the standardization formula is as follows:

$${\mathrm{E}}_{\mathrm{ij}}={\displaystyle \frac{\left(\mathrm{x}-{\mathrm{x}}_{\min }\right)}{\left({\mathrm{x}}_{\max }-{\mathrm{x}}_{\min }\right)}}$$

For negatively correlated indicators, the formula for standardization processing is as follows:

$${\mathrm{E}}_{\mathrm{ij}}={\displaystyle \frac{\left({\mathrm{x}}_{\max }-\mathrm{x}\right)}{\left({\mathrm{x}}_{\max }-{\mathrm{x}}_{\min }\right)}}$$

where E_ij represents the standardized value of the j-th grid cell for the i-th indicator; x represents the original value of that indicator; x_max represents the maximum value of that indicator; and x_min represents the minimum value of that indicator.

2.5. Bayesian Network Model

Compared to other machine-learning models, Bayesian network models offer significant advantages and can be more effectively integrated with ecological vulnerability assessment models. They demonstrate a high degree of applicability in analyzing climate and vegetation changes within the study area, examining the connections between ecosystem services and residents’ well-being, and assessing the impact of various factors on the ecological environment. Bayesian network models provide robust decision support for sustainability research on the regional ecological environment and assist in developing informed environmental management and protection strategies.

The Bayesian network model, a directed acyclic probability graph, effectively demonstrates causal relationships between variables by analyzing changes in node probability distributions. This approach enables the exploration and prediction of factors driving ecological vulnerability in the study area [35,36,37].

This paper utilizes Netica 5.18 software to construct Bayesian network models, which are widely applied in disaster research and water resource management [38,39,40,41,42], and collects historical data on socio-economic, climatic, ecological, and water use aspects of the study area as case samples for parameter learning to determine the conditional probability tables and the probability distribution relationships for each node. The construction process of the Bayesian network model involves the following four key steps:

(1)

Establish the network topology structure based on the causal relationships among nodes [43].

(2)

Collect relevant data for sample parameter learning in the Bayesian network model.

(3)

Model accuracy is assessed using standard evaluation methods, including K-fold cross-validation [44], receiver operating characteristic (ROC) curves [45], and area under the curve (AUC) values [46]. This study employs the 10-fold cross-validation method to evaluate model accuracy.

(4)

Entropy reduction (VR) in Netica software is used to estimate the impact of input variables on ecological vulnerability, serving as a key parameter for quantifying sensitivity [47]. The standardization formula is as follows:

$$\mathrm{VR}=\mathrm{V}\left(\mathrm{Q}\right) - \mathrm{V}\left(\mathrm{Q}|\mathrm{F}\right)={\displaystyle \sum }_{\mathrm{q}}\mathrm{P}\left(\mathrm{q}\right)\left[{\mathrm{X}}_{\mathrm{q}} -{{\displaystyle \sum }}_{\mathrm{q}}\mathrm{P}\left(\mathrm{q}|\mathrm{f}\right)\right]{\left[{\mathrm{X}}_{\mathrm{q}} -{{\displaystyle \sum }}_{\mathrm{q}}\mathrm{P}\left(\mathrm{q}|\mathrm{f}\right){·\mathrm{X}}_{\mathrm{q}}\right]}^2$$

where V represents variance; H represents entropy; Q represents the target node; F represents other nodes; q and f represent the states of the target node and other nodes, respectively; and X_q is the actual value corresponding to state q.

(5)

The Bayesian network model developed in Netica is employed for posterior probability inference to determine the response states of various indicators [48].

Finally, using the state of the Bayesian network in 2020 as the baseline scenario, this paper assumes five different ecological vulnerability scenarios, each with probabilities set at 100%. By comparing the posterior and prior probabilities of the nodes, the direction of the impact of change factors on each node can be determined: if the posterior probability increases relative to the prior probability, the change factor is considered to have a promoting effect on the node; if the posterior probability decreases, it is regarded as an inhibiting effect. This method allows us to reveal the response of different impact indicators, thereby identifying key issues and areas requiring priority management in the region.

Scenario One: Ecological vulnerability is set at the “potential vulnerability” level, with a vulnerability index range of 0–0.2. Under this scenario, the Bayesian network recalculates the posterior probabilities and records the probability changes for each element. By comparing the differences between posterior and prior probabilities, we can identify which elements exhibit significant changes in response to potential vulnerability. This comparison helps assess the impact of potential vulnerability on each element and determine whether these elements have a promoting or inhibiting relationship with potential vulnerability.

Subsequent Scenarios: Similarly, ecological vulnerability is set to slight, moderate, very high, and extreme vulnerability levels, with recalculations of posterior probability changes conducted under each assumption to further assess the impacts on each element.

3. Results

3.1. Spatial and Temporal Variations in Ecological Vulnerability

From a spatial distribution perspective, ecological vulnerability in the study area showed a significant downward trend from 2000 to 2020, as visually demonstrated by the spatial distribution patterns in Figure 4. Specifically, ecological vulnerability decreases progressively from north to south, with the Guanzhong Basin—home to the cities of Xi’an and Xianyang—exhibiting the highest vulnerability indices, classified as very vulnerable and extremely vulnerable areas. These regions are marked by dense urban development, high population density, high urbanization levels, and concentrations of high-pollution industries. This combination has led to severe ecological degradation, increased ecological risk, and a simplified ecological structure and function, resulting in poor ecosystem stability and a reduced ability to resist external risks.

The northern part of Weinan City is the primary moderately vulnerable area, located on the outskirts of core cities. While this area experiences relatively high levels of human activity and certain ecological pressures from the development of nearby urban centers, it has suffered comparatively limited ecological degradation. In contrast, the Qinling Mountains and areas to the south are primarily characterized by potential and slight vulnerability owing to low human disturbance, high vegetation cover, and strengthened community succession, which bolsters resistance to external disturbances and preserves ecosystem integrity. Notably, the area of moderately, very, and extremely vulnerable regions has spatially decreased, showing that ecological vulnerability radiates outward from the basin’s central cities to the mountainous areas.

Table 5. Proportion of ecological vulnerability by grade.

Grade	Proportion of Area (%)
	2000	2005	2010	2015	2020
Potential Vulnerability	93.5	87.8	90.8	93.8	94.8
Slight Vulnerability	5	9.5	7.5	5.8	4.9
Moderate Vulnerability	1.1	1.1	0.8	0.3	0.17
High Vulnerability	0.2	0.4	0.2	0.11	0.01
Extreme Vulnerability	0.2	0.2	0.13	0.04	0.02

further reveals the temporal evolution of ecological vulnerability levels across the study area. The proportion of potential vulnerability areas decreased from 2000 to 2005, then began to rebound, while slight vulnerability areas showed the opposite trend. Moderately vulnerable areas declined continuously; very vulnerable areas increased from 2000 to 2005, then significantly decreased; and extremely vulnerable areas consistently decreased over the study period. Compared to 2000, the overall ecological vulnerability in 2020 significantly decreased, particularly in high and extreme vulnerability areas, such as the suburbs of Xi’an and northern Weinan. Moderate vulnerability also declined, while potential vulnerability markedly increased.

The results of spatiotemporal changes indicate that the ecological environment of the study area has improved, with enhanced ecosystem stability and a stronger capacity to withstand and manage ecological risks. These findings offer valuable insights for understanding spatial distribution variations and temporal trends in regional ecological vulnerability, supporting the development of more effective ecological protection and management strategies.

3.2. Bayesian Network Model Evaluation

This article employed a cross-validation method to assess the performance of the Bayesian network model (Figure 5). Specifically, we randomly divided the 25,482 data points in the study area into ten groups, using one group as the validation data each time, while the remaining nine groups were used to train the model. This process was repeated, so that each group of data had the opportunity to serve as the validation set in the assessment. First, we used the training dataset to construct the conditional probability tables for the Bayesian network. Then, with the Netica software, we performed probability reasoning on the already parameterized Bayesian network for the sample data in the validation set. The software calculates the probability values for the state of each vulnerability node and selects the state with the highest probability as the predicted outcome for that node. By comparing the predicted results with the actual values, we calculated the model accuracy for each validation. After 10 rounds of cross-validation, we obtained 10 models with varying accuracy rates, with an average accuracy of 82.8% [49,50]. This result indicates that the Bayesian network structural model has high accuracy and stability in the diagnosis and prediction of ecological vulnerability (

Table 6. Confusion matrix based on ecological vulnerability.

Ecological Vulnerability (Reality)	Ecological Vulnerability (Forecast)
	Potential Vulnerability	Slight Vulnerability	Moderate Vulnerability	High Vulnerability	Extreme Vulnerability	Row
Potential Vulnerability	18,947	457	0	0	0	25,482
Slight Vulnerability	1856	1673	0	0	0
Moderate Vulnerability	2	198	157	2	0
High Vulnerability	0	0	0	37	0
Extreme Vulnerability	0	0	0	0	21
Column	25,482
Accuracy	82.80%

).

3.3. Key Driving Indicators for Ecological Vulnerability Changes

From 2000 to 2020, the key indicators—including industrial water use, sulfur dioxide emissions, industrial wastewater discharge, and ecological water use—have consistently shown high VR values, indicating that these factors have played a decisive role in the ecological vulnerability of the study area. However, since 2010, the VR values for industrial water use and sulfur dioxide emissions have shown a downward trend, suggesting that their negative impact on ecological vulnerability has diminished. Meanwhile, the VR values for industrial wastewater discharge and ecological water use have remained relatively stable, indicating no significant change in their impact on ecological vulnerability. These trends imply that the study area has made some progress in managing water and air pollution, strengthened water resource management, and reduced the negative impact of human activities on the ecological environment.

Economic indicators, including grain output, fiscal revenue, and resident savings, have shown relatively low VR values between 2000 and 2020, indicating that these factors have had a limited impact on ecological vulnerability (Figure 6). The VR value for GDP was high before 2010 but significantly decreased afterward, indicating a clear trend of reduced impact on the ecological vulnerability of the study area. This suggests that the negative impact of economic growth on ecological vulnerability has been alleviated and that the quality of economic development has improved.

After 2015, the VR values for temperature, rainfall, elevation, and the Normalized Difference Vegetation Index (NDVI) in the study area have shown an upward trend, indicating that these indicators have become increasingly significant to ecological vulnerability. NDVI, an important indicator of vegetation cover, and elevation are closely associated with increased species diversity. This further suggests that the ecosystem’s sensitivity to external disturbances is rising, and the roles of vegetation cover and species diversity within the ecosystem are becoming more crucial. Climate change has emerged as an important factor affecting the ecological environment of the Qinling-Daba Mountain Area and cannot be ignored.

3.4. Scenario Analysis and Forecasting of Spatial and Temporal Distribution of Drivers

The Bayesian network model can assess the probability changes of the target variable under different ecological vulnerability scenarios (potential vulnerability, slight vulnerability, moderate vulnerability, high vulnerability, and extreme vulnerability) due to its capacity for performing inference in the reverse direction. As shown in Figure 7, the probability distribution of each driving factor changes under these scenarios. Additionally, a threshold of 1.5% change magnitude is used as a standard to identify the spatial distribution of key variables (Figure 8). This method not only enhances our understanding of the indicators affecting ecological vulnerability but also facilitates the effective formulation of ecological protection and restoration strategies.

Scenario One: Assuming this region is in a potential vulnerability state, the probability changes in NDVI, altitude, and temperature are particularly significant, indicating that these are the key drivers influencing potential vulnerability in the future. The probabilities of altitude and temperature in the 0.2–0.4 probability range increased by 3.8% and 3.3%, respectively, while those in the 0.4–0.6 range decreased by 3.5% and 3%, respectively, compared to 2020. At the same time, the probability of NDVI in the 0–0.2 range increased by 3.5%. These changes suggest that NDVI in the 0–0.2 range, and altitude and temperature in the 0.2–0.4 range, may promote the development of potential vulnerability, while they may inhibit its development at higher levels.

Scenario Two: Assuming this region is in a slight vulnerability state, the NDVI, altitude, and temperature show significant probability changes, suggesting that they are the key drivers of the future slightly vulnerable state. Compared to 2020, the probabilities of altitude and temperature in the 0.4–0.6 probability range increased by 17% and 14.6%, respectively, while in the 0.2–0.4 range, they decreased by 18.7% and 16.3%, respectively. Additionally, the probability of NDVI in the 0–0.2 range decreased by 17.6%. These changes reveal that altitude and temperature in the 0.4–0.6 range significantly promote slight vulnerability, while altitude and temperature in the 0.2–0.4 range, along with NDVI in the 0–0.2 range, have a significant inhibiting effect.

Scenario Three: Assuming this region is in a moderate vulnerability state, compared to 2020, industrial water use, ecological water use, industrial wastewater, agricultural water use, and SO₂ emissions have become the key indicators affecting the future moderately vulnerable state. The probabilities of industrial water use and industrial wastewater in the 0.2–0.4 probability range increased by 11.15% and 12.8%, respectively, while the probability of ecological water use in the 0.6–0.8 range increased by 10.22%. Meanwhile, the probability of agricultural water use in the 0–0.2 range decreased by 7.7%, and the probabilities of industrial water use, industrial wastewater, and SO₂ in the 0–0.2 range decreased by 13.4%, 16%, and 12.7%, respectively. These changes indicate that industrial water use and industrial wastewater within the 0.2–0.4 probability range, as well as ecological water use within the 0.6–0.8 range, significantly promote moderate vulnerability. In contrast, agricultural water use in the 0–0.2 range, and industrial water use, industrial wastewater, and SO₂ in the 0–0.2 range, have a significant inhibiting effect.

Scenario Four: Assuming this region is in a high vulnerability state, industrial water use, ecological water use, industrial wastewater, agricultural water use, and SO₂ are the key drivers of the future highly vulnerable state. Compared to 2020, the probabilities of industrial water use and industrial wastewater at moderate levels (0.2–0.4) increased by 16.15% and 18.5%, respectively. Meanwhile, the probability of ecological water use at 0.6–0.8 rose by 14.52% but declined by 18.2% at 0.8–1. In contrast, the probability of agricultural water use in the low range (0–0.2) decreased by 11.1%, and the probabilities of industrial water use, industrial wastewater, and SO₂ at very low levels (0–0.2) declined by 19.8%, 23.1%, and 18.4%, respectively. These changes indicate that industrial water use and industrial wastewater in the 0.2–0.4 range, along with ecological water use at 0.6–0.8, significantly promote high vulnerability. Conversely, ecological water use in the 0.8–1 range, as well as agricultural water use, industrial water use, industrial wastewater, and SO₂ in the 0–0.2 range, exhibit significant inhibitory effects.

Scenario Five: Assuming this region is in an extreme vulnerability state, industrial water use, ecological water use, industrial wastewater discharge, agricultural water use, and SO₂ emissions are identified as the key drivers exacerbating future ecological vulnerability. Compared to 2020, the probabilities of industrial water use and industrial wastewater at moderate levels (0.2–0.4) increased by 16.2% and 14%, respectively, while ecological water use at higher levels (0.6–0.8) increased by 10%. Conversely, the probability of agricultural water use at low levels (0–0.2) decreased by 10.8%, and the probabilities of industrial water use, industrial wastewater, and SO₂ at very low levels (0–0.2) decreased by 13.4%, 16.3%, and 13%, respectively, while ecological water use in the 0.8–1 range decreased by 12.2%. These data suggest that increases in industrial water use, industrial wastewater, and ecological water use at higher levels correlate positively with worsening ecological vulnerability. Conversely, reductions in agricultural water use, industrial water use, industrial wastewater, and SO₂ emissions at very low levels, along with high-level ecological water use, help mitigate ecological vulnerability.

It is notable that the impact of ecological water use on ecologically fragile areas varies significantly at different levels. When ecological water use is moderate to high (0.6–0.8), it considerably exacerbates ecological vulnerability in highly vulnerable and extremely vulnerable areas of the study region. However, at even higher levels of ecological water use (0.8–1), its impact shifts to a significant inhibitory effect. This phenomenon highlights a threshold effect: once ecological water use surpasses a specific threshold, its influence on ecological vulnerability may fundamentally change, shifting from exacerbating to inhibiting. This non-linear response may relate to varying effects on different ecosystem service functions. Additionally, the influence of the natural environment on high vulnerability and extreme vulnerability areas is relatively limited.

Given that most areas of the study region exhibited potential and slight vulnerability in 2020, this paper hypothesized that the entire region would be in a state of potential vulnerability and explored corresponding management strategies with the expectation of improving the ecological environment in the future. The results are presented in Figure 8.

As illustrated in the figure, the key indicators hindering ecological improvement include both natural factors and human activities. For example, indicators such as industrial water use, industrial wastewater, SO₂ emissions, and NDVI within the (0–0.2) grade range have a positive effect on the evolution of the central and eastern regions of the Guanzhong Basin into a potentially vulnerable state. Meanwhile, altitude, temperature, and NDVI within the (0.2–0.4) grade range positively influence the northern part of Xianyang in its transition to a potentially vulnerable state. In contrast, industrial wastewater within this range negatively affects the central and eastern parts of the northern slopes of Qinling, hindering them from becoming potentially vulnerable.

Within the (0.4–0.6) grade range, indicators such as altitude, temperature, and NDVI act as barriers to the Guanzhong Basin’s transition into a potential vulnerability state. Additionally, at the (0.8–1) grade range, precipitation serves as a negative factor, impeding the central and eastern parts of the Guanzhong Basin from reaching a state of potential vulnerability.

4. Discussion

This study, based on the VSD evaluation index system, assesses the ecological vulnerability index of the Qinling-Daba Mountain and adjacent areas from 2000 to 2020. Through the entropy reduction method of the Bayesian network model, the primary driving factors of ecological vulnerability in the region are analyzed and discussed, and different ecological vulnerability scenarios are simulated. The key findings are summarized as follows.

From a temporal perspective, between 2000 and 2010, the ecological vulnerability index of the study area exhibited a fluctuating downward trend. After 2010, this decline accelerated, particularly in the northern parts of Baoji City, northern Xianyang City, and eastern Weinan City, where the decrease in ecological vulnerability was especially pronounced. This trend suggests that the ecological environment has undergone continuous improvement over time. From a spatial perspective, potentially vulnerable areas constitute the largest proportion of the study region, whereas very vulnerable and extremely vulnerable areas account for a relatively smaller percentage. Specifically, the potentially vulnerable and mildly vulnerable areas are primarily located in the Qinling Mountains and regions to the south, where the ecological vulnerability index is low, and the ecosystem is relatively stable. In contrast, highly vulnerable areas, classified as very vulnerable or extremely vulnerable, are concentrated in central cities and their surrounding areas, particularly Xi’an and nearby regions, reflecting the pressure from urbanization and industrialization. Overall, the ecological vulnerability of the study area shows a marked decline across both temporal and spatial dimensions, indicating significant improvement in the balance between human activities and the natural environment, thus progressing toward harmonious coexistence [51,52]. Particularly in northern Baoji City, northern Xianyang City, and eastern Weinan City, the ecological civilization efforts have produced the anticipated positive results. The ongoing improvement in the region’s ecological environment is closely linked to several ecological restoration projects initiated after 2010, including large-scale restoration efforts, optimization of regional spatial management measures, and the advancement of ecological civilization initiatives. These interventions have effectively mitigated the negative effects of human activities on the environment, promoting the restoration and stabilization of ecosystems.

From the perspective of the driving factors, industrial water use, SO₂ emissions, industrial wastewater discharge, and ecological water use were identified as the key drivers of the region’s ecological vulnerability. This underscores the significant impact of industrial and agricultural activities on ecosystem vulnerability, with natural factors playing a comparatively minor role. However, the influence of these drivers has varied significantly over time. Since 2010, the impact of industrial water use and GDP on ecological vulnerability has gradually diminished, while sulfur dioxide and industrial wastewater emissions have significantly decreased since 2015, likely a result of government-led industrial upgrades. The technological advancements since 2010 not only enhanced resource efficiency but also introduced more sophisticated pollution control technologies. Through industrial restructuring and increased investments in green technologies and cleaner production, the government has substantially improved water resource utilization efficiency. Although total industrial water use has increased, the reduction in pollutant emissions has led to overall environmental improvement. Simultaneously, the government has enacted a series of sustainable agricultural policies, promoting organic agriculture, reducing chemical fertilizers and pesticides, and encouraging eco-friendly farming practices, which have effectively mitigated the adverse impacts of agriculture on ecosystems. These policies have driven sustainable agricultural development and significantly contributed to reducing ecological vulnerability. Notably, after 2015, ecological water use had a more pronounced effect in reducing ecological vulnerability, indicating the increasing effectiveness of these policies. Furthermore, the role of natural factors has evolved. Variables such as temperature, precipitation, elevation, and the Normalized Difference Vegetation Index (NDVI) have significantly influenced ecological vulnerability since 2010. This shift is likely linked to large-scale ecological restoration efforts, such as reforestation, wetland restoration, and biodiversity conservation programs. These measures have greatly improved the ecological conditions in the area. Regional climate warming has also promoted vegetation growth, further enhancing the stability and resilience of ecosystems. These comprehensive initiatives have played a critical role in mitigating ecological vulnerability and restoring ecological balance, laying a solid foundation for future sustainability in the region.

Compared to existing research on ecological vulnerability in the Qinling region, this study employs a Bayesian network model to simulate and predict various ecological vulnerability scenarios, demonstrating the practicality and feasibility of the model in complex geographical and climatic conditions. The scenario simulation results show significant differences in how natural and social factors affect ecological vulnerability: natural factors dominate in areas with low ecological vulnerability, whereas social factors are more prominent in densely populated urban areas. Under different scenarios, the simulation responses of different factors vary, indicating that the influence of different indicators on ecological vulnerability is closely related to ecosystem sensitivity and human activity intensity. Natural factors, such as the NDVI, elevation, temperature, and precipitation, positively impact potentially vulnerable areas at lower levels but have the opposite effect in mildly or moderately vulnerable areas, predominantly located south of the Qinling Mountains. Therefore, future studies should prioritize monitoring the changes in natural indicators in these regions. In contrast, social factors, especially those related to industrial and agricultural activities, have a more significant effect on highly vulnerable areas. Sulfur dioxide emissions, industrial water use, and industrial wastewater exert a substantial negative impact on the ecosystem, while ecological water use has a positive effect. Additionally, maintaining agricultural water use between 0.2 and 0.4 significantly reduces ecological vulnerability.

The scenario simulations further suggest that industrial water use and wastewater discharge may not immediately cause major ecological issues in areas with low ecological vulnerability. However, if left unchecked, increasing discharge volumes may gradually deteriorate water quality and ecosystem health. Similarly, changes in elevation and temperature may have a greater impact on ecosystems in low-vulnerability areas. A reduction in vegetation cover could decrease carbon sequestration, increase soil erosion, and affect biodiversity and ecosystem functions. The distribution of sensitive indicators in potential vulnerability scenarios shows that negative factors are concentrated in agricultural and urban land use areas north of the Qinling Mountains, while the positive effects of natural factors are prominent on the southern slopes of the mountains. Thus, the impact of land use on ecological vulnerability should not be overlooked.

The spatiotemporal evolution characteristics indicate that ecological vulnerability in the Qinling-Daba Mountain Area and its adjacent regions has exhibited a significant downward trend, aligning with findings from numerous studies [53,54,55]. Comparing this trend with that of the Loess Plateau—a mountainous region also deeply influenced by human–land interactions—we observe that since 2010, both the Qinling-Daba Mountain Area and the Loess Plateau have shown improving trends in ecological vulnerability. While both regions are affected by human activities and climate change, their primary driving factors differ [56,57]. In the Qinling-Daba Mountain Area, ecological vulnerability is closely linked to human activities, such as industrial water use, whereas in the Loess Plateau, vegetation coverage and precipitation serve as the main factors influencing ecological vulnerability. This finding underscores differences in the mechanisms shaping ecological vulnerability between the two regions. Furthermore, the decline in ecological vulnerability is more pronounced in the Qinling-Daba Mountain Area than in the Loess Plateau, suggesting that ecological civilization efforts have had more substantial effects in the Qinling-Daba Mountain Area. In summary, while both the Qinling-Daba Mountain Area and the Loess Plateau demonstrate similar trends in reduced ecological vulnerability, they differ significantly in the mechanisms driving these trends and the degree of improvement.

Based on the analysis above, this study proposes several recommendations. First, ecological protection measures, such as ecological migration programs, should be implemented in the Qinling region to mitigate the impact of human activities on ecosystems. Second, in areas significantly affected by industrial pollution, cost-effective pollution control measures should be adopted, such as promoting the development of green industries. Lastly, this study suggests combining nature-based restoration measures with environmentally driven governance strategies to further promote the health and sustainability of regional ecosystems.

Future environmental governance measures and policies may prioritize several key areas: urban development, climate change adaptation, and environmental monitoring. During urbanization, policymakers should carefully consider the environmental impact of human activities, especially those identified as key environmental concerns, such as sulfur dioxide (SO₂) emissions and industrial wastewater discharge. These activities challenge the protection of urban ecosystems and efforts to foster harmonious relationships between humans and nature.

In response to climate change, policies should emphasize monitoring and addressing the changes in temperature, precipitation patterns, and vegetation cover in the region to safeguard natural ecosystems and enhance ecosystem stability and resilience. Environmental monitoring is also essential for accurately assessing environmental quality trends and current status; it provides a scientific foundation for areas like land use, climate change, and vegetation management and plays a pivotal role in supporting sustainable environmental development and promoting human–nature coexistence.

The effectiveness of ecological water use varies significantly across different thresholds, and the non-linear changes triggered by these thresholds may have critical implications for water resource utilization efficiency and water management policies. This threshold effect could drive cities to make adaptive adjustments in water management to optimize resource allocation and enhance water use efficiency.

Bayesian network models offer significant guidance for the spatial planning and management of mountain ecosystems due to their capacity to handle complex ecological relationships and human–environment interactions. These models are particularly suitable for analyzing the intricate ecological dynamics within mountain systems. Their bidirectional reasoning capability—allowing for inference from causes to effects and vice versa—provides a powerful tool for diagnosing and predicting ecosystem dynamics. Specifically, Bayesian networks can elucidate interactions between ecosystem services, which is essential for developing effective strategies for ecological restoration and protection. This reasoning capacity also makes Bayesian networks particularly valuable for identifying the causes of ecosystem changes and forecasting future trends, thus offering a scientific basis for the sustainable development of mountain ecosystems. Consequently, Bayesian network models not only demonstrate a high level of theoretical applicability but also show great potential in practical research and applications in mountain ecology.

Despite their strengths in managing uncertainty and complex relationships, Bayesian network models face certain limitations in predicting ecological vulnerability. These limitations are primarily due to the high dependence of model predictions on input data quality and initial model configuration. Because Bayesian networks require substantial data to estimate parameters accurately, the sufficiency and precision of data are critical for model performance in practical applications. In this study, data precision may constrain the Bayesian network model’s effectiveness, affecting the reliability of predictions, particularly for long-term and large-scale ecological vulnerability forecasts. Furthermore, this study does not explore in depth how different driving factors influence ecological vulnerability thresholds across various simulation scenarios. To address these limitations and improve predictive accuracy and model applicability, it is essential to enhance data quality and precision and optimize initial model assumptions.

Future research should also examine the impacts of different driving factors on ecological vulnerability thresholds by refining the evaluation indicator ranges. By setting more specific indicator intervals and analyzing how each driving factor influences ecological vulnerability across these refined intervals, we can lay a more scientific foundation for devising targeted ecological protection and management strategies.

5. Conclusions

This paper innovatively combines the methods of the variance sensitive decision (VSD) assessment framework, entropy weighting, and Bayesian networks to conduct a systematic assessment and scenario simulation analysis of ecological vulnerability from 2000 to 2020 in the Shaanxi section of the Qinling-Daba Mountain Area and its adjacent regions. This approach demonstrates that the Bayesian network model is an efficient tool for studying the human–land relationship in the Qinling region. The core conclusions are as follows.

The majority of the study area exhibits potential ecological vulnerability, and from 2000 to 2020, there has been a significant reduction in areas classified as highly vulnerable or extremely vulnerable, indicating a clear trend of ecological improvement. Analyzing the changes in ecological vulnerability across regions and time periods enables the precise identification of areas needing targeted attention for the study area’s development. This analytical capacity supports the development of effective, targeted environmental protection and management measures, thereby promoting sustainable regional development.

Industrial water use, industrial wastewater discharge, sulfur dioxide emissions, and agricultural water use remain the most critical factors influencing ecological vulnerability, with a notable decrease in the impact of industrial water use and wastewater discharge after 2010. The influence of temperature, precipitation, and NDVI on ecological vulnerability has increased since 2015.

The key measures to improve the ecological environment in highly vulnerable areas include reducing water and wastewater discharges and emissions from agriculture and high-pollution industries, as well as increasing ecological water use. These actions can effectively support regional sustainability and contribute to a scientific understanding and assessment of ecological vulnerability. An in-depth analysis of the causes and manifestations of ecological vulnerability is essential, encompassing factors such as climate change, land use changes, and socio-economic development. These factors interact and directly impact ecosystem stability and sustainability, making them critical for accurately assessing ecosystem health and developing effective ecological protection strategies. A comprehensive consideration of these drivers provides a holistic understanding of ecological vulnerability, laying a scientific foundation for the long-term protection and management of ecosystems in the study area.

This study suggests that future research could utilize multi-source data and apply machine-learning techniques to identify the key drivers of ecological vulnerability in the study area. Analyzing the coupled spatiotemporal characteristics of the region from a dynamic perspective may provide deeper insights into the mechanisms driving regional ecological evolution.

The study reveals that the current policies have effectively improved water resource utilization efficiency. However, their impact on controlling key pollution sources, such as industrial production, remains insufficient and requires enhancement. Consequently, future policymaking should focus on more stringent regulation of emissions from industrial and other pollution sources to further advance environmental quality.

In the context of climate change, this study offers valuable guidance for regional ecological risk assessment and the development of early warning systems. These measures enable the prediction and identification of potential ecological risks and environmental issues, facilitating the adoption of proactive prevention and response strategies. Such strategies reduce the potential losses from ecological disasters and mitigate the adverse effects of human activities on ecosystems at a lower social cost.

Additionally, the study’s findings provide a scientific foundation for policymakers, supporting governments in formulating and adjusting policies on environmental protection, water resource management, and land use planning in response to trends in ecological vulnerability. This approach promotes harmonious socio-economic and ecological development, ensuring the successful implementation of sustainable development strategies.