Analysis of Driving Factors for Vegetation Ecological Quality Based on Bayesian Network

: Vegetation is a crucial component of ecosystems, and understanding the drivers and spatial optimization patterns of its ecological quality is vital for environmental management in the middle reaches of the Yangtze River Urban Agglomeration. Traditional evaluations employing single indices may not fully capture the complexity of vegetation elements and require evaluation through various indicators. Therefore, this study introduced the Multi Criteria Vegetation Ecological Quality Index (VEQI), coupled with vegetation cover and vegetation ecological function indicators, to explore the driving factors of vegetation quality in the middle reaches of the Yangtze River and identify key areas where vegetation quality declines or improves. By constructing a Bayesian network for VEQI, we identified the driving variables that influence the index. Additionally, we delineated spatial optimization zones for VEQI. The results indicate that the VEQI exhibits a trend of transitioning from low values in urban centers to high values in suburban and rural areas. Over 20 years, the average VEQI of the study region ranged from 10.85% to 94.94%. Slope, DEM, and vegetation type were identified as significant drivers of VEQI, while precipitation, temperature, and nighttime light were considered secondary factors. Notably, areas in Hunan, Jiangxi, and Hubei provinces, especially the western part of Hunan, were pinpointed as spatial optimization regions. This research not only enhances the understanding of vegetation’s ecological quality in the urban agglomeration of the middle reaches of the Yangtze River but also provides scientific insights for the protection and management of vegetation.


Introduction
Vegetation plays a special role in the formation of the composition of species [1,2].However, over the past half-century, rapid economic development and intensified human activities have precipitated a marked decline in the vitality of vegetation ecosystems [3].In response to this pressing challenge, the monitoring and attribution analysis of vegetation ecological quality over extensive spatial scales and extended periods have become integral components of conservation efforts [4].
A foundational element in characterizing terrestrial vegetation landscapes is the quality of the vegetation ecology, with its changes having profound impacts on the energy fluxes and material cycles within both global and regional ecosystems [5,6].Accordingly, scholars have conducted extensive research in this area.For instance, Shiekh Marifatul [7] Forests 2024, 15, 1263 2 of 20 conducted an assessment of the vegetation ecosystem status in the Himalayas through field investigations, while Zhang [8] explored nitrogen deposition's effects on natural and planted forests through field surveys.Additionally, the advent and application of remote sensing technology have established a reliable method for studying vegetation characteristics, making it a popular choice for assessing vegetation ecological quality [9][10][11][12].Nevertheless, previous studies relying on remote sensing usually focused on single indicators such as the Normalized Difference Vegetation Index (NDVI) [13,14], Fraction of Vegetation Cover (FVC) [15], and Net Primary Productivity (NPP) [16].These indicators, individually, have proven effective in capturing specific aspects of vegetation health and productivity.However, given the complexity of vegetation ecosystems and the multitude of factors influencing their condition, a single indicator may oversimplify the complexities inherent in vegetation quality assessment, potentially leading to biased outcomes [17].
Therefore, there is a need to employ a combination of multiple indicators for a more holistic evaluation of vegetation ecological quality.Currently, comprehensive evaluation methods for environmental quality predominantly include the Environmental Index (EI) and the Remote Sensing Ecological Index (RSEI), which offer new avenues and perspectives for assessing ecological conditions.Nonetheless, the former faces challenges such as the difficulty of acquiring indicators and a limitation in the weighting process that can be subjective [18].The latter, despite its widespread use [19][20][21], is constrained by the spatial resolution of the land satellite data that it relies upon, making it primarily suited for evaluating vegetation ecological quality in smaller-scale areas.Its application in larger-scale regions encounters difficulties due to this limitation [22].Considering the complexity of vegetation ecosystems and the diversity of vegetation ecological elements, accurately assessing vegetation ecological quality requires the comprehensive consideration of multiple indicators.Integrating the ecological functions of vegetation with FVC in the study area can help reveal the spatial distribution pattern and changes in regional ecological quality more accurately.This approach aids in a comprehensive understanding of the crucial role of vegetation in maintaining ecosystem balance and the impact of human activities on vegetation ecological quality.
Human activities and natural factors are the primary drivers of variations in vegetation ecological quality, showcasing noticeable regional disparities [23][24][25].Scrutinizing the underlying determinants affecting vegetation ecological quality facilitates the pinpointing of environmental concerns and ecological vulnerabilities, offering a scientific foundation for environmental stewardship and spatial arrangement optimization [26].Traditional statistical methodologies, including partial least squares regression [27], correlation analysis [28], and multiple linear regression [29], have been employed to identify driving factors.Nonetheless, these methodologies are confined to investigating linear relationships between individual variables and the dependent variable.An abundance of research indicates that the influence of environmental variables on vegetation exhibits non-linear characteristics [30,31].Consequently, existing methodologies demand refinement to better comprehend the intricate non-linear relationships linking anthropogenic activities, natural variables, and vegetation quality.
Bayesian networks, as a powerful tool for large-scale data analysis, excel in representing non-linear relationships [32][33][34].They enable quantitative assessments of the relative contributions of both natural factors and variations in human activities to shifts in vegetation ecological quality, facilitating the identification of potential driving variables.Therefore, integrating a Bayesian network into ecological research has shown promising results in optimizing spatial patterns for environmental management.Unfortunately, despite the potential of a Bayesian network in vegetation quality research and spatial optimization, few studies have employed a Bayesian network for researching vegetation quality, let alone in optimizing its spatial patterns.Integrating spatial data into Bayesian network models could highlight areas susceptible to improvement or degradation, guiding strategic interventions, and enhancing our understanding of vegetation quality.
According to the above research gaps, this study adopts a multi-indicator approach from the perspective of vegetation, taking into account ecosystem service functions.Drawing inspiration from the improved Vegetation Ecological Quality Index (VEQI) developed by Cui et al. [35], we analyze the spatiotemporal variation characteristics of VEQI in the middle reaches of the Yangtze River Urban Agglomeration.Furthermore, we applied a Bayesian network to quantify the relative contributions of natural factors and human activities to VEQI, identify potential driving factors, and spatially delineate zones for optimizing VEQI within the region.This study aimed to (1) evaluate the long-term dynamic changes in vegetation ecological quality in the middle reaches of the Yangtze River from 2000 to 2020; (2) ascertain potential factors and their contributions that affect the vegetation ecological quality changes in the middle reaches of the Yangtze River; (3) identify spatial optimization areas for vegetation ecological quality.The research results can provide a theoretical basis for ecological governance and vegetation protection in the middle reaches of the Yangtze River, and contribute to the sustainable development of the region.

Study Area
The middle reaches of the Yangtze River Urban Agglomeration, depicted in Figure 1, are situated within the coordinates of 25 This area spans across Hubei, Hunan, and Jiangxi provinces in China, covering roughly 537,600 square kilometers.It integrates major metropolitan areas, including the Wuhan Metropolitan Area, the Chang-Zhu-Tan Metropolitan Circle, and the Poyang Lake Metropolitan Circle.Positioned in the subtropical monsoon climate zone, it is characterized by four distinct seasons and receives generous rainfall, with hot and humid summers, cold and dry winters, average annual temperatures between 14 and 18 • C, and annual precipitation levels ranging from 1000 to 1500 mm.The complex terrain is shaped by the Yangtze River and its tributaries, creating a varied riverine landscape [37].Vegetation is varied and generally well covered; terrestrial vegetation resources are abundant, with forest coverage rates in Hubei, Hunan, and Jiangxi provinces reaching 39.61%, 59.82%, and 64.00%, respectively, in 2015 [38].The vegetation types are diverse, mainly consisting of cultivated vegetation (50.26%), coniferous forests (21.53%), shrublands (12.55%), broadleaf forests (6.71%), and grasslands (6.26%) [39].The main vegetation resources include Metasequoia glyptostroboides, Ilex chinensis Sims, Cathaya argyrophylla, Davidia involucrata Baill, Ginkgo biloba L., Eucommia ulmoides, Camellia japonica, Acer kiangsiense, etc., boasting abundant forest resources that highlight the ecological significance of this area.In recent years, urban agglomeration has maintained a medium-to-high economic growth rate.From 2000 to 2020, the total gross domestic product (GDP) of the middle reaches of the Yangtze River Urban Agglomeration surged from CNY 0.85 trillion to CNY 10.61 trillion, increasing by approximately 12 times [40]; simultaneously, the population density increased from 454.65 person/km 2 to 490.6 person/km 2 [41].The middle reaches of the Yangtze River, influenced by the dual forces of climate change and human activities, exhibit a fragile ecological environment in some areas, with vegetation showing a high sensitivity to these changing conditions; the ecological quality of vegetation has consequently become a matter of significant public concern [42].Encouragingly, in 2022, the National Development and Reform Commission (NDRC) unveiled the "14th Five-Year Implementation Plan for the Development of the Mid-Yangtze River Urban Agglomerations", which underscores the principles of ecological priority and green development.Thus, investigating the vegetation ecological quality in the middle reaches of the Yangtze River Urban Agglomeration is of vital importance for regional ecological conservation and sustainable development.

Data Sources
The data sources for this study primarily consist of satellite imagery, meteorological data, human activity data, and land use data, all spanning the time period from 2000 to 2020.The terrain data comprised digital elevation model (DEM) data and slope information.The DEM dataset was downloaded from the Geospatial Data Cloud (https://www.gscloud.cn/sources/,accessed on 10 May 2024) and had a spatial resolution of 30 m.To ensure data consistency across datasets, the data were resampled to a 1 km resolution.Following this, DEM data for the study area were obtained through cropping, and subsequently, slope calculations were performed.
Climatic factor data primarily included temperature and precipitation, both sourced from the National Earth System Science Data Center (http://www.geodata.cn/,accessed on 10 May 2024), each with a spatial resolution of 1 km.Temperature stands as a cardinal determinant in vegetation growth and ecological processes.Diverse vegetation types exhibit varying degrees of thermal adaptability, with fluctuations in temperature directly impacting the overall ecological quality of plant communities.Precipitation critically influences vegetation growth; it serves as the primary source of moisture for plants, significantly affecting their physiological processes and, consequently, their ecological health.Vegetation type data were sourced from the Center for Resources and Environmental Data Science, Chinese Academy of Sciences (http://www.resdc.cn,accessed on 10 May 2024).

Data Sources
The data sources for this study primarily consist of satellite imagery, meteorological data, human activity data, and land use data, all spanning the time period from 2000 to 2020.The terrain data comprised digital elevation model (DEM) data and slope information.The DEM dataset was downloaded from the Geospatial Data Cloud (https://www.gscloud.cn/sources/, accessed on 10 May 2024) and had a spatial resolution of 30 m.To ensure data consistency across datasets, the data were resampled to a 1 km resolution.Following this, DEM data for the study area were obtained through cropping, and subsequently, slope calculations were performed.
Climatic factor data primarily included temperature and precipitation, both sourced from the National Earth System Science Data Center (http://www.geodata.cn/,accessed on 10 May 2024), each with a spatial resolution of 1 km.Temperature stands as a cardinal determinant in vegetation growth and ecological processes.Diverse vegetation types exhibit varying degrees of thermal adaptability, with fluctuations in temperature directly impacting the overall ecological quality of plant communities.Precipitation critically influences vegetation growth; it serves as the primary source of moisture for plants, significantly affecting their physiological processes and, consequently, their ecological health.Vegetation type data were sourced from the Center for Resources and Environmental Data Science, Chinese Academy of Sciences (http://www.resdc.cn,accessed on 10 May 2024).
The main vegetation types include coniferous forests, deciduous broad-leaved forests, evergreen broad-leaved forests, shrubs, grasslands, and cultivated vegetation.Normalized Difference Vegetation Index (NDVI) data were obtained from the MOD13Q1 product (https://code.earthengine.google.com/,accessed on 11 May 2024).Vegetation type is a critical driver because different vegetation types exhibit varying degrees of adaptability to environmental factors and possess distinct ecological functions.Taking vegetation type into account facilitates a deeper understanding of the factors and mechanisms influencing the ecological quality of different vegetation types.
Human activity data encompassed GDP, population (Pop), and night time light index (NTL) data.GDP and POP datasets were accessed from the Chinese Academy of Sciences' Center for Resources and Environmental Data Sciences (http://www.resdc.cn).The NTL data originated from the National Geophysical Data Center (NGDC) under the National Oceanic and Atmospheric Administration (NOAA) (https://eogdata.mines.edu/products/vnl/, accessed on 12 May 2024), with a spatial resolution of approximately 500 m (Table 1).GDP serves as a pivotal indicator of a nation or region's economic development level.Economic expansion often parallels industrialization, urban sprawl, and agricultural extension, all of which bear potential implications for vegetation ecological quality.The escalation in population numbers and density can alter land use patterns and intensify usage, potentially imposing detrimental effects on vegetation ecological health.Similarly, the NTL, which measures artificial luminosity in a given area, acts as an indirect proxy for human activity levels and may indirectly sway the ecological quality of vegetation.When conducting a comprehensive evaluation of vegetation ecological quality indicators (VEQI), it is essential to incorporate the functional aspects of the vegetation ecosystem.This study selects four key ecosystem functions-carbon sequestration (CS), water yield (WY), soil retention (SR), and habitat quality (HQ)-to underpin the development of an index for assessing vegetation ecological quality.CS services in the study area were assessed using the Carbon Storage and Sequestration module of the InVEST (Integrated Valuation of Ecosystem Services and Tradeoffs) model [43].Meanwhile, to evaluate the WY service provided by forests that regulate the water cycle through their trees and vegetation, this research employed the Water Yield module of the InVEST model [44,45].The function of SR was calculated using a revised Universal Soil Loss Equation [46,47].Biodiversity, indicative of HQ, was assessed by employing the Habitat Quality module within the InVEST framework.This module helps in quantifying biodiversity by evaluating the condition and intactness of habitats [48,49].
Table 2 presents detailed calculation formulas and data sources used in the study.Normalization was applied to the data of the four ecological functions, using their individual maximum and minimum values.These normalized values were then summed to derive the Total Ecosystem Service (TES) index [50], which succinctly represents the overarching level of ecosystem services provided by the middle reaches of the Yangtze River Urban Agglomeration [51].The formula is as follows: Y xj represents the annual water yield; P x denotes the annual average precipitation over grid cell x; AET xj signifies the actual annual average evapotranspiration from grid cell x under land use type j.
SC indicates average annual soil conservation; A P and A r are potential and actual soil erosion, respectively.R is the erosive factor of rainfall; K is the soil erodibility factor; LS is the slope length and slope factor; C is the vegetation coverage factor; P is the factor of soil and water conservation measures.
Q xj represents the habitat quality of pixel x on the j type of land use; H j represents the habitat suitability of the j type of land use; D xj represents the degree of habitat degradation of pixel x in the j land use type; K is the semi saturation constant.
In the formula, ES S represents the normalized result of the ecosystem service; ES is the initial value of the ecosystem service; ES max denotes the maximum value of the ecosystem service; and ES min signifies the minimum value.To mitigate the impact of outliers, the ES max and ES min are set as the 95% and 5% values, respectively.The Total Ecosystem Service Index TES is computed by summing up the normalized results of each service, where ES si is the normalized outcome of the i category of ecosystem service, with i = 4 indicating the total number of considered ecological functions.

Pixel Dichotomy Model
FVC effectively mirrors the state of vegetation growth and biomass.In this study, the pixel dichotomy model was adopted to calculate FVC [52], with the computation formula outlined as follows: In the formula, FVC represents the fractional vegetation cover of a particular pixel; NDVI is the NDVI value of that pixel; NDVI soil denotes the NDVI value of pure soil pixels within the image; and NDVI veg signifies the NDVI value of pure vegetation pixels within the image.

Construction of VEQI
The VEQI is based on TES and FVC.The formula is as follows: where VEQI denotes the Vegetation Ecological Quality Index, ranging from 0 to 100%; FVC represents the Fractional Vegetation Cover; TES represents the Total Ecosystem Service index, and f 1 and f 2 are the weighting coefficients employed in the calculation.

Bayesian Network Modeling 2.4.1. Construction of the BN
Bayesian networks (BNs), also known as belief networks, are probabilistic graphical models [53,54].They consist of two primary components: (1) a Directed Acyclic Graph (DAG), composed of nodes symbolizing a set of variables and arrows connecting these nodes from parent nodes to child nodes, which define dependencies and independencies among nodes, suggesting causal relationships; and (2) Conditional Probability Tables (CPT), which articulate the relationships and strengths between a node and its parent nodes.Each node within the network can represent either discrete or continuous variables, allowing BNs to model complex systems with mixed variable types effectively [55].
Initially, pertinent variables are selected to serve as nodes in constructing a two-layer BN, comprising a total of 9 nodes interconnected by 11 arrows.In this network, VEQI acts as the target node, representing the ultimate variable of interest, while eight other variables-DEM, SLO, and POP, among others-are designated as influencing factor nodes, each impacting the ecological quality of vegetation.Subsequently, natural breakpoint detection methods are employed to discretize and categorize these variables into four distinct levels or classes.This process segments the continuous data into intervals that naturally occur in the distribution, enhancing the interpretability and manageability of the model (Table 3).The raster values associated with each of these variables are then extracted and utilized to train the network [56].After constructing the BN, it is imperative to validate its performance.To do this, 20% of the processed dataset is randomly segregated to serve as a test set, distinct from the data used for training.The validation process employs Netica software (https://norsys.com,accessed on 10 May 2024), which furnishes three inbuilt metrics for assessing the model's efficacy: error rate, quadratic loss (Formula 5), and spherical payoff (Formula 6).The error rate and log loss are both measures of prediction error, with lower values nearing zero indicating superior predictive accuracy.A low error rate signifies that the model's predictions closely match the actual outcomes, while minimal log loss implies that the probabilities assigned by the model to the correct class are high, reflecting a good fit to the data.On the other hand, spherical gain, which operates on a scale from 0 to 1, gauges the overall precision of the model, with values closer to 1 signifying higher accuracy.A high spherical gain suggests that the BN is effectively capturing the underlying structure and dependencies within the data, leading to more reliable predictions [57].
In this study, the relative importance of each node in the BN with respect to the VEQI node was assessed through sensitivity analysis.This is primarily achieved through the reduction in belief variance and the reduction in entropy-based mutual information (MI); the formula is as follows: where a is the target variable, I is another known variable, s and I represent the states of a and I, respectively.V(a|I) represents the difference between the variance of the variable S and the variance of the variable S under the condition of known variable I. H denotes entropy; Q represents the target node; and F stands for other nodes.
Based on the evaluated Bayesian network, it becomes possible to analyze the relative likelihood of changes in the target variable's states under different management scenarios.Through the integration of prior knowledge and observational data, the conditional probability distributions of network nodes are obtained.When the probability distributions of certain nodes are adjusted, leveraging Bayes' theorem, the Bayesian network computes the posterior conditional probability distributions of other nodes.This capability enables both predictive and diagnostic analyses [58].Based on distinct developmental objectives, two scenarios have been designed.Scenario 1 is termed the Conservation Scenario, where the goal is set to maximize the ecological quality of vegetation, represented by setting the "good" state of the VEQI to 100%.Scenario 2 is referred to as the Development Scenario, characterized by maximizing economic development at the expense of ecological considerations.Here, the "ex_low" state of the VEQI is set to 100%.

Spatial Pattern Optimization
The integration of the BN model with ArcGIS 10.2 enables spatial visualization of the factors impacting ecological conditions, thereby facilitating the identification of areas where drivers of VEQI stability (or degradation) are concentrated.This process aids in pinpointing regions requiring ecological restoration and optimization strategies.The following section provides a comprehensive overview of the methodology employed: 1.
Utilizing variance reduction outcomes, factors that significantly influence changes in VEQI are selected as pivotal variables.Among these, the states with the highest probability are recognized as the critical states of the variables, which are the main drivers affecting VEQI dynamics.

2.
Capitalizing on the capabilities of GIS spatial analysis tools, a visual representation of the critical states of these key variables is created across the landscape [59,60].

Spatial and Temporal Distribution of VEQI
The mean values of ecosystem services and FVC for the Mid-Yangtze River Urban Agglomeration from 2000 to 2020 are depicted in Figure 2, these metrics reveal significant temporal and spatial variations in the region's ecological conditions over the two-decade period.Building upon these findings, we developed the VEQI, as illustrated in Figure 3.Over the course of 20 years, the average VEQI value in the study area ranged between 10.85% and 94.94%, with most regions having an average VEQI value greater than 60%, indicating good ecological quality (Figure 4a).From a provincial perspective, Jiangxi Province exhibits the highest vegetation ecological quality, with an average 20-year VEQI value of 75%.This indicates a high level of forest coverage, stable ecosystems, and superior ecological quality, which may be attributed to its rich natural resources, stringent environmental protection policies, and relatively lower degree of industrialization.Hunan Province follows closely behind Jiangxi, with an average 20-year VEQI value of 73%.This suggests good vegetation coverage, although there might be some regions with more fragile    Over the course of 20 years, the average VEQI value in the study area ranged between 10.85% and 94.94%, with most regions having an average VEQI value greater than 60%, indicating good ecological quality (Figure 4a).From a provincial perspective, Jiangxi Province exhibits the highest vegetation ecological quality, with an average 20-year VEQI value of 75%.This indicates a high level of forest coverage, stable ecosystems, and superior ecological quality, which may be attributed to its rich natural resources, stringent environmental protection policies, and relatively lower degree of industrialization.Hunan Province follows closely behind Jiangxi, with an average 20-year VEQI value of 73%.This suggests good vegetation coverage, although there might be some regions with more fragile The VEQI of the Mid-Yangtze River Urban Agglomeration shows an overall upward trend, presenting a "high-low-high" distribution pattern from periphery to center with distinct regional characteristics.Areas with dense populations, such as the provincial capitals, Nanchang, Wuhan, and Changsha, have seen their low VEQI zones gradually expand over two decades.Similarly, the low VEQI zones in suburban areas around cities like Yichang, Hengyang, and Jian have also expanded.The border region, where Hunan, Jiangxi, and Hubei provinces meet, has consistently been a high-value area for VEQI, indicating good ecological quality of vegetation.Areas with higher elevations, such as mountainous or hilly regions, tend to exhibit superior vegetation quality.This is primarily attributed to lesser human intervention in these areas, allowing for the natural growth and development of vegetation without significant disturbance.The rugged terrain often acts as a natural barrier to extensive human activities, preserving the integrity of the ecosystem.Contrastingly, plains or regions with gentler landscapes are more susceptible to intense socio-economic activities.Areas with more accessible and exploitable resources often experience more frequent and intense anthropogenic socioeconomic activities, such as agricultural cultivation, urban construction, and industrial development.These activities have the potential to cause the degradation and alteration of local vegetation and ecosystems.Agricultural practices, such as clearing land for farming, can lead to soil erosion, loss of native plant species, and disruption of the local food web.Urban expansion involves the conversion of natural landscapes into built environments, which can fragment habitats, reduce biodiversity, and alter hydrological cycles.Industrial activities, including mining and manufacturing, often result in pollution and habitat destruction, further impacting the health and resilience of ecosystems.Consequently, these areas often show lower VEQI values, reflecting diminished ecological quality.Furthermore, the proximity to large bodies of water, exemplified by the Poyang Lake and Dongting Lake regions, also plays a critical role in vegetation quality.Despite the potential benefits of water sources for vegetation, such as increased moisture and nutrient availability, the surrounding areas often experience lower VEQI scores.This could be due to several factors, including increased human activities like fishing, tourism, and urban sprawl near these lakes, which can negatively impact vegetation cover and ecological health.
Over the course of 20 years, the average VEQI value in the study area ranged between 10.85% and 94.94%, with most regions having an average VEQI value greater than 60%, indicating good ecological quality (Figure 4a).From a provincial perspective, Jiangxi Province exhibits the highest vegetation ecological quality, with an average 20-year VEQI value of 75%.This indicates a high level of forest coverage, stable ecosystems, and superior ecological quality, which may be attributed to its rich natural resources, stringent environmental protection policies, and relatively lower degree of industrialization.Hunan Province follows closely behind Jiangxi, with an average 20-year VEQI value of 73%.This suggests good vegetation coverage, although there might be some regions with more fragile ecosystems.This could be linked to the province's varied topography and climatic conditions, as well as its focus on ecological conservation.Hubei Province has an average 20-year VEQI value of 70%, slightly lower than Jiangxi and Hunan, yet still at a relatively high level.This demonstrates that overall vegetation coverage in the province is sound, but there may be specific areas contending with ecological issues.
Looking at individual cities (Figure 4b), the average VEQI value across all cities was 72%, with 16 cities scoring above this average.Most of these cities (nine in total) are located in Jiangxi Province, with Jingdezhen ranking the highest among them.This could be associated with Jiangxi's generally high ecological quality and the favorable vegetation coverage around Jingdezhen. Wuhan and Jingmen in Hubei Province are the only two cities with average VEQI values above the provincial average.This might be due to Wuhan's status as the provincial capital with well-developed urban greening and Jingmen possibly benefiting from certain ecological advantages.In Hunan Province, cities above and below the average are evenly distributed, indicating smaller differences in vegetation ecological quality among cities and a more balanced overall condition.Generally speaking, Jiangxi, situated in a subtropical monsoon climate zone with abundant rainfall, provides excellent conditions for vegetation growth and boasts a high forest coverage rate, thus resulting in superior vegetation ecological quality.Hunan Province features complex terrain with mountains and plains, leading to a variety of vegetation types and some regional disparities in ecological quality.Hubei Province, located on the middle and lower reaches of the Yangtze Plain, has vegetation dominated by farmland and plantation forests, with less natural vegetation cover, hence the vegetation ecological quality is slightly lower compared to Jiangxi and Hunan.
quality among cities and a more balanced overall condition.Generally speaking, Jiangxi, situated in a subtropical monsoon climate zone with abundant rainfall, provides excellent conditions for vegetation growth and boasts a high forest coverage rate, thus resulting in superior vegetation ecological quality.Hunan Province features complex terrain with mountains and plains, leading to a variety of vegetation types and some regional disparities in ecological quality.Hubei Province, located on the middle and lower reaches of the Yangtze Plain, has vegetation dominated by farmland and plantation forests, with less natural vegetation cover, hence the vegetation ecological quality is slightly lower compared to Jiangxi and Hunan.

Establishment and Validation of VEQI Based on the BN
Netica (https://norsys.com)was utilized to construct a two-layer BN model for the VEQI.This network included seven continuous variable nodes and one discrete variable node.After the nodes were established, 313,332 sample points obtained through 1 km grid sampling in the middle reaches of the Yangtze River region served as real-world input cases for learning the CPT of all nodes.Following parameter learning, the prior probability distributions for each state of the nodes were determined, as shown in Figure 5.In this region, the vegetation type that was most prevalent was cultivated vegetation, accounting for 46.6% of the total, while marshes were the least common, representing only 2.25% of the vegetation types.The probabilities of VEQI levels from low to high were 16.7%, 12.4%,

Establishment and Validation of VEQI Based on the BN
Netica (https://norsys.com)was utilized to construct a two-layer BN model for the VEQI.This network included seven continuous variable nodes and one discrete variable node.After the nodes were established, 313,332 sample points obtained through 1 km grid sampling in the middle reaches of the Yangtze River region served as real-world input cases for learning the CPT of all nodes.Following parameter learning, the prior probability distributions for each state of the nodes were determined, as shown in Figure 5.In this region, the vegetation type that was most prevalent was cultivated vegetation, accounting for 46.6% of the total, while marshes were the least common, representing only 2.25% of the vegetation types.The probabilities of VEQI levels from low to high were 16.7%, 12.4%, 32.5%, and 38.3%.This indicates that 16.7% of the area has lower vegetation ecological quality, while 38.3% of the area enjoys better vegetation ecological quality.The distribution of different vegetation types and the varying levels of VEQI in the region provide insights into the ecological health and management priorities for the area.Cultivated being the dominant type suggests intensive land use, possibly for agriculture.Given the VEQI level probabilities, it is clear that a significant portion of the region has relatively good ecological quality.Table 4 presents the outcomes of model validation.The error rate for the VEQI was determined to be 26.15%, with a quadratic algorithm loss of 0.4072 and a spherical gain of 0.7702.These metrics indicate that the BN model constructed in this study exhibits a high degree of accuracy.To further assess the model's predictive capabilities for VEQI, a confusion matrix for the "VEQI" node was calculated (Table 5).Focusing on the "good" state Table 4 presents the outcomes of model validation.The error rate for the VEQI was determined to be 26.15%, with a quadratic algorithm loss of 0.4072 and a spherical gain of 0.7702.These metrics indicate that the BN model constructed in this study exhibits a high degree of accuracy.To further assess the model's predictive capabilities for VEQI, a confusion matrix for the "VEQI" node was calculated (Table 5).Focusing on the "good" state as an example, out of 2330 samples where the actual state was "good," the model correctly predicted the "good" state for a significant number of these samples.The prediction accuracy rate for the "good" state was found to be 73.85%.This high accuracy rate underscores the reliability of the model's predictions for VEQI states.

Analysis of the Driving Factors Influencing VEQI
A sensitivity analysis was conducted based on Netica (https://norsys.com).The results demonstrate a consistent trend between mutual information and belief variance, indicating that the larger the mutual information, the greater the belief variance and the higher the sensitivity of the target variable (Figure 6).Moreover, when the number of intermediate variables between a node and the target increases, the influence on the target variable diminishes.intermediate variables between a node and the target increases, the influence on the target variable diminishes.The sensitivity analysis results indicated that among the natural factors, Slo, Veg, and DEM have a high sensitivity to VEQI.Regarding anthropogenic factors, NTL, showed the highest sensitivity to VEQI.This suggests that the gradient of the terrain (slope), the extent and condition of vegetation cover, and elevation play significant roles in determining the ecological quality of vegetation.These natural elements are crucial components that affect the overall health and vitality of plant communities.On the other hand, the intensity of nighttime lights, an indicator of human presence and activity, significantly influences VEQI.This finding implies that areas with higher levels of urbanization and human intervention tend to have a more pronounced effect on vegetation quality, potentially leading to the degradation or alteration of natural vegetation states.

Results of Scenario Analysis
The results of the sensitivity analysis indicated that Slo, Veg, and DEM were selected as observation factors.By setting up the two scenarios discussed in Section 2.4.3, changes in the probability of various states of these three critical variables were observed (Figure 7).
Compared to the current situation, under the conservation scenario ("good = 100%"), protecting vegetation on lower slopes and surfaces with less steep terrain proves to be most effective for enhancing vegetation ecological quality.This suggests that focusing conservation efforts in areas with gentler slopes can yield significant improvements in ecological health, as these areas are more conducive to the growth and sustenance of di- The sensitivity analysis results indicated that among the natural factors, Slo, Veg, and DEM have a high sensitivity to VEQI.Regarding anthropogenic factors, NTL, showed the highest sensitivity to VEQI.This suggests that the gradient of the terrain (slope), the extent and condition of vegetation cover, and elevation play significant roles in determining the Forests 2024, 15, 1263 13 of 20 ecological quality of vegetation.These natural elements are crucial components that affect the overall health and vitality of plant communities.On the other hand, the intensity of nighttime lights, an indicator of human presence and activity, significantly influences VEQI.This finding implies that areas with higher levels of urbanization and human intervention tend to have a more pronounced effect on vegetation quality, potentially leading to the degradation or alteration of natural vegetation states.

Results of Scenario Analysis
The results of the sensitivity analysis indicated that Slo, Veg, and DEM were selected as observation factors.By setting up the two scenarios discussed in Section 2.4.3, changes in the probability of various states of these three critical variables were observed (Figure 7).

The Optimal Development Area of VEQI
To optimize the spatial pattern of the VEQI, it is necessary to identify regions characterized by specific natural or socio-economic features that are more suitable for vegetation ecological conservation.This will allow for the delineation of priority areas for development and conservation efforts.In the context of this study, the primary method involves analyzing the states of key variables and their corresponding states of VEQI.By determining the states of these variables under which VEQI exhibits optimal conditions, one can pinpoint the most suitable areas for prioritized development or conservation.The optimization zones are identified by combining a subset of critical states with conditional probabilities.As shown in Table 6, when the VEQI exhibits probabilities corresponding to ex_low, low, medium, and good states, the subsets of critical states are at varying levels.Specifically, when the Slo is in the "medium" state, the Veg is coniferous forest, and the DEM is in the "low" state (Figure 8a), and the VEQI reaches its highest possible probability (0.95) in the "good" state.By spatially displaying the results, we identified the area most suitable for optimization.
As shown in Figure 8b, the suitable optimized region is located in the central region where vegetation coverage is high and human activity is minimal.These areas are concentrated at the junction of Hunan, Jiangxi, and Hubei provinces.The largest distribution is found within Hunan province, where the predominant land use types are forest and grassland, and the main vegetation types include coniferous forests and cultivated vegetation.For this region, proactive ecological protection strategies should continue to be implemented, promoting the enhancement and efficiency of forest lands.Specifically, stringent land use planning must be implemented to limit unnecessary human activities and infrastructure development, thereby reducing physical damage to coniferous forests and promoting sustainable forestry practices.Protected areas must be established to safeguard rare species and ecologically sensitive zones, while monitoring the health of wildlife and flora.Research must be conducted on the impacts of climate change on coniferous forests and adjust management strategies to enhance forest resilience.For cultivated vegetation, crop rotation systems and organic farming methods must be enforced to maintain soil fertility and biodiversity.Native plant species must be planted along borders to create ecological buffer zones.Diversified agricultural systems, such as agroforestry, which integrates agriculture and forestry to bolster ecological stability and economic returns, must Compared to the current situation, under the conservation scenario ("good = 100%"), protecting vegetation on lower slopes and surfaces with less steep terrain proves to be most effective for enhancing vegetation ecological quality.This suggests that focusing conservation efforts in areas with gentler slopes can yield significant improvements in ecological health, as these areas are more conducive to the growth and sustenance of diverse vegetation.Moreover, reducing the proportion of cultivated vegetation to 32.1% is found to be the most beneficial for vegetation ecological quality.Cultivated vegetation, often associated with agricultural practices, can sometimes lead to monocultures and reduced biodiversity, which can negatively impact the overall ecological quality.By limiting the extent of cultivated areas, there is an opportunity to promote a healthier mix of vegetation types that support a more robust and biodiverse ecosystem.
Compared to the current situation, under the development scenario ("ex_low = 100%"), the influence of slope and DEM on vegetation ecological quality appears to be less significant.Instead, the focus shifts to the impact of intensive cultivation of vegetation, particularly if the proportion of cultivated vegetation is dramatically increased to 56%.This expansion of cultivated vegetation comes at the expense of the growth space for other vegetation types, notably coniferous forests, leading to a degradation of the ecological quality of the vegetation to its lowest point.

The Optimal Development Area of VEQI
To optimize the spatial pattern of the VEQI, it is necessary to identify regions characterized by specific natural or socio-economic features that are more suitable for vegetation ecological conservation.This will allow for the delineation of priority areas for development and conservation efforts.In the context of this study, the primary method involves analyzing the states of key variables and their corresponding states of VEQI.By determining the states of these variables under which VEQI exhibits optimal conditions, one can pinpoint the most suitable areas for prioritized development or conservation.The optimization zones are identified by combining a subset of critical states with conditional probabilities.
As shown in Table 6, when the VEQI exhibits probabilities corresponding to ex_low, low, medium, and good states, the subsets of critical states are at varying levels.Specifically, when the Slo is in the "medium" state, the Veg is coniferous forest, and the DEM is in the "low" state (Figure 8a), and the VEQI reaches its highest possible probability (0.95) in the "good" state.By spatially displaying the results, we identified the area most suitable for optimization.

Characteristics of Changes in Regional VEQI
This research effectively integrates remote sensing data with ecosystem functions to provide a nuanced evaluation of vegetation ecological quality, factoring in both the perunit-area-and-time vegetation cover capacity of vegetation communities and their regional ecosystem functions.The index results provide a more comprehensive evaluation of the vegetation ecological quality in a specific region.In the midstream urban agglomeration of the Yangtze River, the VEQI demonstrates a typical zonal spatial distribution characteristic, decreasing from the border areas of Hunan, Jiangxi, and Hubei provinces toward the city centers and increasing toward the outskirts.This trend highlights the spatial variation in vegetation ecological quality across the region.This is consistent with the spatial distribution patterns of FVC [61] and NDVI [62] in the midstream urban agglomerations.This spatial consistency across multiple indicators reinforces the understanding of vegetation health and ecological quality in the region.Over time, the VEQI of the middle reaches of the Yangtze River Urban Agglomeration shows an upward trend, indicating that current ecological protection policies and conservation projects have effectively controlled soil erosion and significantly improved ecological quality [63].On a temporal scale, the ecological quality of vegetation within the Mid-Yangtze River Urban Agglomeration has demonstrated a general upward trajectory, as shown in Figure 4.More than half of the regions have an average ecological quality of over 60%, and only major central cities such as Wuhan, Nanchang, and Changsha have slightly lower ecological quality than other cities, indicating a significant improvement in the ecological quality of the middle reaches of the Yangtze River.As evidenced by Hu's study [64], there was a significant As shown in Figure 8b, the suitable optimized region is located in the central region where vegetation coverage is high and human activity is minimal.These areas are concentrated at the junction of Hunan, Jiangxi, and Hubei provinces.The largest distribution is found within Hunan province, where the predominant land use types are forest and grassland, and the main vegetation types include coniferous forests and cultivated vegetation.For this region, proactive ecological protection strategies should continue to be implemented, promoting the enhancement and efficiency of forest lands.Specifically, stringent land use planning must be implemented to limit unnecessary human activities and infrastructure development, thereby reducing physical damage to coniferous forests and promoting sustainable forestry practices.Protected areas must be established to safeguard rare species and ecologically sensitive zones, while monitoring the health of wildlife and flora.Research must be conducted on the impacts of climate change on coniferous forests and adjust management strategies to enhance forest resilience.For cultivated vegetation, crop rotation systems and organic farming methods must be enforced to maintain soil fertility and biodiversity.Native plant species must be planted along borders to create ecological buffer zones.Diversified agricultural systems, such as agroforestry, which integrates agriculture and forestry to bolster ecological stability and economic returns, must be developed.These measures will contribute to the long-term health and productivity of vegetation ecosystems, ensuring their ability to provide essential ecosystem services and support biodiversity.Additionally, monitoring and managing human activities in these areas to minimize negative impacts on the environment is crucial.

Characteristics of Changes in Regional VEQI
This research effectively integrates remote sensing data with ecosystem functions to provide a nuanced evaluation of vegetation ecological quality, factoring in both the perunit-area-and-time vegetation cover capacity of vegetation communities and their regional ecosystem functions.The index results provide a more comprehensive evaluation of the vegetation ecological quality in a specific region.In the midstream urban agglomeration of the Yangtze River, the VEQI demonstrates a typical zonal spatial distribution characteristic, decreasing from the border areas of Hunan, Jiangxi, and Hubei provinces toward the city centers and increasing toward the outskirts.This trend highlights the spatial variation in vegetation ecological quality across the region.This is consistent with the spatial distribution patterns of FVC [61] and NDVI [62] in the midstream urban agglomerations.This spatial consistency across multiple indicators reinforces the understanding of vegetation health and ecological quality in the region.Over time, the VEQI of the middle reaches of the Yangtze River Urban Agglomeration shows an upward trend, indicating that current ecological protection policies and conservation projects have effectively controlled soil erosion and significantly improved ecological quality [63].On a temporal scale, the ecological quality of vegetation within the Mid-Yangtze River Urban Agglomeration has demonstrated a general upward trajectory, as shown in Figure 4.More than half of the regions have an average ecological quality of over 60%, and only major central cities such as Wuhan, Nanchang, and Changsha have slightly lower ecological quality than other cities, indicating a significant improvement in the ecological quality of the middle reaches of the Yangtze River.As evidenced by Hu's study [64], there was a significant enhancement in vegetation coverage, with an estimated 66% increment recorded between 2005 and 2020.This marked growth proved that the current public welfare forest construction, the policy of returning farmland to forests and grasslands, and the comprehensive protection project for natural forest resources were effective in maintaining and improving ecological quality.

Influencing Factors of VEQI
Human activities and natural factors have long been a subject of interest for scholars in their study of vegetation.When exploring driving variables, scholars often use correlation analysis [65] and principal component analysis [66] to directly establish mathematical models between driving factors and vegetation quality.However, the impact of driving variables on vegetation ecological quality is not a simple linear relationship; instead, it is a non-linear relationship.In other studies, Bayesian networks have been widely employed to explore the non-linear relationships between driving variables and the study object.For instance, researchers such as He [67] used Bayesian network to investigate the non-linear relationship between rainfall intensity, slope, and their impact on herbaceous vegetation; Zeng [68] utilized a Bayesian network approach to identify critical variables, proposing a method to optimize the spatial pattern of water resource protection through a subset of key variables; Wang [69] combined InVEST-PLUSE with a Bayesian network, to investigate the spatial optimization patterns of four ecosystem services in the Chang-Ji-Tu region from 2005 to 2020.Bayesian networks possess robust analytical capabilities for exploring relationships among variables, making them particularly effective in uncovering complex interactions in ecological studies.In our study, we found that altitude, slope, and vegetation type have a significant impact on the VEQI.These factors are followed in importance by temperature, precipitation, and human activities, this is consistent with the research findings of Zhang [70] and Ma [71].Considering natural factors such as temperature and precipitation at different gradients and altitudes, along with human activities, will contribute significantly to the conservation of vegetation ecosystems and enhance the overall environmental quality.

Implications for Vegetation Ecological Conservation and Management
The conservation and management of vegetation ecosystems stand as pillars of environmental sustainability, playing a critical role in maintaining ecological balance and supporting life on Earth.Understanding and identifying critical areas for vegetation protection is indeed pivotal, particularly in light of the increasing pressures on natural ecosystems.In this study, we employed a scientific approach to pinpoint regions that are optimal for enhancing the VEQI.Our analysis reveals that these key areas predominantly cluster at the confluence of three provinces, highlighting their strategic importance for conservation efforts [72].This involves adopting an evidence-based approach to conservation, utilizing data and research to guide decision-making processes.Such strategies should focus on preserving biodiversity, restoring degraded ecosystems, and promoting sustainable land use practices that minimize human impact on natural habitats.In our study, two scenarios were set up to scrutinize the alterations in VEQI.A shared and powerful determinant affecting the VEQI across these scenarios was the vegetation type.Notably, the overplanting of cultivated vegetation could substantially alter the VEQI.Hence, it is imperative in management practices to define the boundaries for cultivated vegetation and regulate its extent.Moreover, given the cross-provincial nature of these areas, inter-provincial coordination becomes indispensable.Collaboration among neighboring provinces is necessary to ensure a unified approach to vegetation conservation.This includes sharing policies and regulations that affect land use and conservation efforts.By working together, provinces can leverage collective strengths and address common challenges more effectively, thereby enhancing the overall success of conservation initiatives.

Shortcomings and Prospects
Our study has several limitations.In selecting driver variables, we focused on temperature, precipitation, POP, and GDP, overlooking other critical factors highlighted in various studies, such as carbon dioxide levels, soil properties, wind speed, and their interrelationships.The categorization of input variables for the Bayesian network model impacts its accuracy, affecting subsequent analyses.As time progresses, regions suitable for vegetation ecological quality improvement may exhibit varying spatial changes.Given these shortcomings, future research should adopt a more comprehensive approach in selecting influencing factors, taking into account the synergistic effects of these variables.It will be crucial to determine the most appropriate classification criteria for model inputs, consider different future scenarios, identify optimal spatial optimization zones, and propose corresponding policy responses.

Conclusions
Based on multi-source data, this study integrates ecosystem functions into the assessment of vegetation ecological quality, constructing a framework within a Bayesian context.This research contributes to our understanding of vegetation ecological quality.Our findings are summarized as follows: (1) The VEQI in the midstream urban agglomeration of the Yangtze River has generally improved over time.Spatially, the VEQI shows a trend of increasing from low values in urban centers to higher values in suburban and rural areas.Over a period of 20 years, the average VEQI value in the study area ranged between 10.85% and 94.94%.Notably, the average VEQI value was above 60% in most regions, indicating a predominantly good ecological quality.(2) Among the natural factors, slope, DEM, and vegetation type are significant drivers impacting the VEQI.These are followed by Tem and Pre, which also play important roles in determining the ecological quality of vegetation.Human activities, represented by NTL intensity, emerge as another crucial factor affecting VEQI.(3) A multi-scenario analysis demonstrates that altering vegetation cover types has a significant impact on the VEQI, with a particular emphasis on the increase in cultivated vegetation crops.The areas at the intersection of Hunan, Jiangxi, and Hubei provinces, as well as western Hunan, are identified as spatial optimization regions and should become focal points for future ecological conservation efforts.
We aspire for our findings to provide policymakers and stakeholders with targeted insights and facilitate the formulation of efficacious strategies.Our aim is to deepen the collective understanding of ecological quality preservation and explore innovative models for environmental conservation.

Figure 1 .
Figure 1.Location of the study area: (a) represents the geographical location of China; (b) is the administrative division of the study area; and (c) represents the land use type in 2020.

Figure 1 .
Figure 1.Location of the study area: (a) represents the geographical location of China; (b) is the administrative division of the study area; and (c) represents the land use type in 2020.

Forests 2024 ,
15,  x FOR PEER REVIEW 10 of 21 often experience lower VEQI scores.This could be due to several factors, including increased human activities like fishing, tourism, and urban sprawl near these lakes, which can negatively impact vegetation cover and ecological health.

Figure 2 .
Figure 2. Spatial distribution of mean values of ecosystem services and FVC from 2000 to 2020.

Figure 3 .
Figure 3. Spatial and temporal distribution of different levels of VEQI from 2000 to 2020.

Figure 2 .
Figure 2. Spatial distribution of mean values of ecosystem services and FVC from 2000 to 2020.

Forests 2024 ,
15,  x FOR PEER REVIEW 10 of 21 often experience lower VEQI scores.This could be due to several factors, including increased human activities like fishing, tourism, and urban sprawl near these lakes, which can negatively impact vegetation cover and ecological health.

Figure 2 .
Figure 2. Spatial distribution of mean values of ecosystem services and FVC from 2000 to 2020.

Figure 3 .
Figure 3. Spatial and temporal distribution of different levels of VEQI from 2000 to 2020.

Figure 3 .
Figure 3. Spatial and temporal distribution of different levels of VEQI from 2000 to 2020.

Figure 4 .
Figure 4.The average VEQI over 20 years: (a) represents the mean value of VEQI in the study area over 20 years; (b) represents the average VEQI value for major cities over 20 years.

Figure 4 .
Figure 4.The average VEQI over 20 years: (a) represents the mean value of VEQI in the study area over 20 years; (b) represents the average VEQI value for major cities over 20 years.

Forests 2024 ,
15,  x FOR PEER REVIEW 12 of 2132.5%, and 38.3%.This indicates that 16.7% of the area has lower vegetation ecological quality, while 38.3% of the area enjoys better vegetation ecological quality.The distribution of different vegetation types and the varying levels of VEQI in the region provide insights into the ecological health and management priorities for the area.Cultivated being the dominant type suggests intensive land use, possibly for agriculture.Given the VEQI level probabilities, it is clear that a significant portion of the region has relatively good ecological quality.

Figure 5 .
Figure 5. Results of parameter learning in the BN model.

Figure 5 .
Figure 5. Results of parameter learning in the BN model.

Figure 7 .
Figure 7. of the two scenarios.

Table 1 .
Source of data.

Table 2 .
Assessment methods for ecosystem services.
total = C above + C below + C soil + C dead C total represents the total carbon stock; C abovel , C below , C soil , C dead represents above-ground, below-ground, soil and dead biogenic carbon stocks, respectively.WY Y xj = 1 −

Table 3 .
Description of BN nodes after discretization.

Table 4 .
Results of model accuracy evaluation.

Table 5 .
Accuracy validation based on the confusion matrix.