Front. Environ. Sci. Frontiers in Environmental Science Front. Environ. Sci. 2296-665X Frontiers Media S.A. 1761294 10.3389/fenvs.2026.1761294 Original Research Ecological degradation risk assessment and influence analysis of ecological conservation redlines in Xilingol League, Inner Mongolia, China Liu et al. 10.3389/fenvs.2026.1761294 Liu Jing 1 * † Funding acquisition Methodology Resources Visualization Writing - original draft Writing - review and editing Lin Dayi 1 † Formal Analysis Funding acquisition Investigation Writing - review and editing Wang Shuaihao 2 3 Software Writing - original draft Xu Xiaojuan 1 Conceptualization Formal Analysis Funding acquisition Methodology Software Writing - original draft Jiao Fusheng 1 Formal Analysis Writing - review and editing Qiu Jie 1 Resources Writing - review and editing Cai Mingyong 4 Funding acquisition Resources Supervision Writing - review and editing Shi Xuewei 4 Resources Writing - review and editing Zhang Kun 1 * † Resources Writing - review and editing Nanjing Institute of Environmental Sciences, Ministry of Ecology and Environment of the People’s Republic of China, Nanjing, China Trier University of Applied Sciences, Trier, Germany Yantai University Trier College of Sustainable Technology, Yantai University, Yantai, China Ministry of Ecology and Environmental Center for Satellite Application on Ecology and Environment, Beijing, China Correspondence: Jing Liu, liujing@nies.org; Kun Zhang, zhangkun@nies.org

These authors have contributed equally to this work

 27 02 2026 2026 14 1761294 09 12 2025 19 01 2026 29 01 2026 Copyright © 2026 Liu, Lin, Wang, Xu, Jiao, Qiu, Cai, Shi and Zhang. 2026 Liu, Lin, Wang, Xu, Jiao, Qiu, Cai, Shi and Zhang https://creativecommons.org/licenses/by/4.0/ This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Performing quantitative evaluations and diagnostic studies on human disruptions within ecological conservation redline areas is highly significant both theoretically and practically for enhancing redline management efficacy and safeguarding regional ecological security. This study examines the ecological conservation redline areas in Xilingol League of Inner Mongolia, using patches experiencing ecological degradation as a basis for assessing ecological risk and conducting diagnostic analysis. The study seeks to objectively identify the principal stressors, characterize risk areas, and suggest unique risk management techniques. The findings reveal that: (1) The risk of ecological degradation in the study area displays a spatial configuration characterized by a predominantly low overall level, interspersed with localized concentrations of high risk areas, with more than 80% of the area classified as having extremely low risk, but with a limited number of high-risk zones (4.61% of the study area), which are concentrated within the ecologically significant Hunshandak Sandland windbreak including a sand-fixation redline area; (2) the ecological degradation risk pattern is primarily influenced by the spatial heterogeneity of human-induced stress, which becomes more pronounced in regions with heightened ecological sensitivity; (3) human-induced stress serves as the principal catalyst for risk formation, with tourism development exerting an area-wide influence, while the impact of human patches is significant locally exhibiting a scale effect. This work elucidates the risk-driving process via nonlinear quantitative evaluation. The results can establish a scientific foundation for differentiated management, source-specific risk mitigation, and the sustainable development of ecological conservation redline areas in the Xilingol League.

 Bayesian network ecological conservation redline ecological degradation risk influence analysis patch The author(s) declared that financial support was received for this work and/or its publication. This research was supported by the Key Technologies Research and Development Program for Youth Scientists of China (No. 2024YFF1309100), the National Natural Science Foundation of China Youth Science Program (No. 42407635), the Youth Foundation in the Jiangsu Province of China (BK20240277; No. BK20220205), and the Youth Foundation in the Nanjing Institute of Environmental Sciences of China (No. GYZX230309). section-at-acceptance Ecosystem Restoration Introduction

Anthropogenic disturbance is widely recognized as a critical driver of ecosystem degradation and has attracted substantial attention from both researchers and practitioners in environmental conservation (Zhang et al., 2024; Zhang and Song, 2025). Various and diverse types human activity affect the structure and function of ecosystems, causing undesirable ecological consequences and may cause cascading responses that exacerbate ecological hazards (Kayumba et al., 2022). The intricate interconnections and transmission processes between sources of disturbance and ecosystem components impart unique nonlinear, dynamic, and unpredictable attributes to ecological degradation risk. Conducting reasonable evaluation and management of these ecological degradation risks is essential to maintaining ecosystem health and guaranteeing ecological security (Sun et al., 2020), particularly in fragile environments (Wang et al., 2025). In China, ecological conservation redline (ECR) areas serve as essential parameters for constructing a “Beautiful China,” highlighting the need of ecological conservation, while the scientific assessment and effective management of ecological concerns have become more critical (Liu et al., 2025). In this regard, regional ecological degradation risk assessment and causative diagnosis provide important scientifically-sound information that can be used to create strategies for managing ecosystems sustainability (Forbes and Galic, 2016; Qian et al., 2023).

Current research on the ecological degradation risk that can be attributed to human activity has predominantly employed various risk assessment methods and indicator systems, such as fragmentation risk assessment (Duan et al., 2025; Vargas-Jaimes et al., 2025) the InVEST model (Veisi Nabikandi et al., 2025; Wang and Liu, 2025), ArcGIS-based Ordered Weighted Averaging (Li et al., 2021; Zhang et al., 2021), ecological disturbance risk index methods (Qiu et al., 2023), and landscape ecological risk index approaches (Yan et al., 2023; Zeng et al., 2024). Although these types of frameworks provide a quantitative comparison and assessment of ecological risk levels across various locations, they mostly function as qualitative or semi-quantitative models (Liu et al., 2023; Lu et al., 2021). Consequently, these models demonstrate the difficulties involved in accurately revealing the interaction processes between anthropogenic disturbances and sensitive risk receptors, as well as in comprehensively defining the regional transmission channels and dispersion patterns of the risk of ecological degradation and damage (Li et al., 2025a; Liu et al., 2023; Zhang et al., 2021).

Probabilistic risk assessment (Chen et al., 2013; Suter et al., 2016; Yang et al., 2025) involves a methodological enhancement that involves quantifying the variability in risk and related uncertainty using probability distributions, therefore offering a more thorough and accurate characterization of risk than conventional qualitative or semi-quantitative methods. Bayesian Networks (BNs) (Kaikkonen et al., 2020), directed acyclic graphs integrating probability theory and graph theory (Pearl, 1986), provide a comprehensive framework for articulating casual relationships through directed acyclic graphs and conducting probabilistic risk assessment using conditional probability tables. This architecture facilitates the amalgamation of many data sources, including monitoring data, simulation results, and expert insights, combining both qualitative and quantitative information within a cohesive probabilistic framework. These characteristics render BNs especially appropriate for tackling intricate environmental issues and underscore their significant relevance in the assessment of ecological degradation risk, providing remarkable flexibility and adaptability (Harris et al., 2017; Orak, 2020). The amalgamation of BNs with spatiotemporal distribution models has been applied effectively in the evaluation of natural disasters (Lu et al., 2024; Nishino et al., 2024; Stritih et al., 2021; Zou et al., 2025), establishing innovative methodological frameworks for examining risks related to anthropogenic disturbance predicated on degradation occurring in ecological patches.

Xilingol League, noted for its vast grassland resources, serves as a crucial element of northern China’s ecological security barrier, preserving vital ecological functions across the landscape (Gao et al., 2023). The region has instituted ECRs covering 1.30 × 105 km2, which is 64.17% of the entire area of Xilingol League, therefore assuring the rigorous safeguarding of grassland ecosystems. Nonetheless, the ecological underpinning of this region remains tenuous, with grassland degradation problems still unresolved at a basic level. Moreover, shaped by previous resource-dependent economic growth models, many human activities continue inside the ECR regions, where zones with documented ecological degradation need ongoing attention to related ecological concerns. Thus, conducting quantitative evaluations and diagnostic analyses of human-induced disturbances within the ECRs of Xilingol League, along with formulating appropriate strategies related to managing ecological degradation risk, holds considerable theoretical and practical importance for improving the efficacy of ECR management and advancing regional ecological governance.

While existing studies have primarily evaluated the overall conservation outcomes of ECRs, systematic approaches to assessing the risk, diagnosing the causes, and formulating differentiated management strategies for localized degradation patches within ECRs remain underdeveloped. To address this research gap, this study conducts an integrated risk assessment and diagnostic analysis focusing on ecological degradation patches inside ECRs in the protected regions of Xilingol League. The specific contributions of this work are threefold: (1) to statistically identify and quantify the key stressors driving regional ecological degradation risk; (2) to spatially delineate zones of high, moderate, and low degradation risk; and (3) to propose a tiered governance framework with targeted risk mitigation measures aligned with different risk levels. The results offer a scientific basis for refined monitoring, risk-informed management, and dynamic optimization of ECRs, thereby supporting enhanced ecological governance and sustainable regional development.

 Methods and materials Study area and data sources

Xilingol League, located in the center of the Inner Mongolia Autonomous Region of China (42°32′‒46°41′N; 111°59′‒120°00′E), represents a quintessential grassland biological zone, with an area of roughly 203,000 km2 (Figure 1). This exemplifies a quintessential steppe habitat that forms a crucial ecological barrier in northern China. Xilingol consists of a typical steppe ecosystem that forms an important ecological barrier in northern China (Gao et al., 2023). The area has several grassland habitats, including temperate meadow steppe, typical steppe, and desert steppe; these grasslands are extensively scattered across the league (Wang et al., 2021). The moderate continental climate is characterized by frigid winters and sweltering summers. The average annual temperature varies from 1 °C to 5 °C, although yearly precipitation increases from west to east, ranging between 200 and 350 mm in the majority of the region. Precipitation mostly occurs throughout the summer months of June to August, constituting around 70% of the yearly total (Wang et al., 2019).

Geographical characteristics of the study area of Xilingol League, Inner Mongolia, China: (a) ecological conservation red lines; (b) slope; (c) digital elevation model (DEM); (d) ecosystem type.

Composite figure with five panels and a legend shows geographic and ecological data for Xilingol League, Inner Mongolia, China. The leftmost panel maps its location in red within China. The top right includes: (a) ecological conservation redlines, (b) elevation, (c) slope, and (d) ecosystem types such as forest, grassland, wetland, and others. Legends identify boundaries, conservation areas, elevation, slope, and ecosystem types.

In 2023, the designated ECR region in Xilingol League included roughly 1.30 × 105 km2, or 64.17% of the league’s entire area (Figure 1a). The ECR area includes several essential ecological functional zones including the Xilingol Grassland Biodiversity Conservation and Windbreak and Sand Fixation, the Hunshandak Sandland Windbreak and Sand Fixation, and the Greater Khingan Range Water Conservation and Biodiversity Maintenance ECRs. The primary objectives of these regions are the protection of biodiversity, establishment of windbreaks, and the stabilization of sand lands, along with water conservation. Collectively, they provide a fundamental geographical framework that underpins the ecological barrier in northern China.

The present study designated 2023 as the baseline year. A Digital Elevation Model (DEM) and corresponding slope data were sourced from the Chinese National Geospatial Data Cloud, at 30 m resolution (Figure 1c; http://www.gscloud.cn). Land use data for Xilingol League were sourced from China’s Third National Land Survey, featuring a spatial resolution of 30 m (Figure 1d). Information regarding the regional railway network, national and provincial highways, and expressways within the league was obtained from the Chinese National Geospatial Information Resources Directory Service System (https://www.webmap.cn). Data pertaining to demographics, administrative divisions, and socioeconomic factors were sourced from the Xilingol League Statistical Yearbook for 2023. Information regarding suspected ecological disturbance patches within the ECR zones was obtained from high-resolution satellite imagery (March‒April 2023) accessed through the Chinese National Ecological Protection Redline Supervision and Management Platform.

Geospatial processing, encompassing data vectorization, overlay analysis, classification, and attribute extraction, was performed using the ArcGIS spatial analysis software suite (https://developers.arcgis.com). A Bayesian network designed for modeling the risk of ecological degradation was constructed, and scenario-based case learning was conducted using Hugin software (http://www.hugin.com).

 Delineation of assessment units

The entire research region was divided into a 10 × 10 km grid system using ArcGIS tools to enable risk analysis and the spatial overlay processing, yielding a total of 1,845 risk assessment units. Each unit functioned as an fundamental autonomous spatial analysis entity, preserving the geographic attribute information of all risk factors—specifically, anthropogenic pressure (A), exposure potential (E), and ecological sensitivity (S)—along with the topological interactions among them. The identification of risk assessment units encompassed the necessary attribute information for developing conditional probability tables in the Bayesian Network, while also illustrating the successful integration of obtained geographic data with the research region (Liu et al., 2023).

 Model framework Bayesian network structure

A BN, as a semi-quantitative statistical model, consists of both qualitative and quantitative elements that facilitate the direct calculation of joint probability distributions. Equation 1 shows the joint conditional probability distribution of non-independent variables: P A i B = P A i × P B A i ∑ i = 1 n P A i × P B A i where B signifies an event, A i denotes all potential variables influencing event B , and P A i defines the prior probability of factor A i in relation to the occurrence of event B . The intrinsic probabilistic framework of BNs enables researchers to adeptly manage uncertainty and intricate nonlinear relationships among variables.

Bayesian networks are primarily used in the estimation of unknown node values based on the given values of certain observable nodes. This technique fundamentally entails probabilistic inference, whereby computations are executed depending on established circumstances and conditional probabilities inside the network. When specific node values, termed evidence (denoted as E), are given, the likelihood of an unknown node X assuming the value a was calculated as the conditional probability p X = a E , represented using Equation 2: p X = a | E = p E , X = a p E

This research, grounded in theoretical principles and taking into account the ecological condition of Xilingol League and the distribution of potential areas with ecological degradation within protection redlines, identified anthropogenic pressure (A), exposure potential (E), and ecological sensitivity (S) as critical risk factors, which were then used to develop a Bayesian network-based assessment model of the risk of ecological degradation (Figure 2).

Structure of the Bayesian network of the model framework.

Conceptual diagram displaying relationships among risk sources, accessibility, and risk receptors leading to ecological degradation risk. Boxes categorize factors, including anthropogenic pressure, exposure potential, and ecological sensitivity, each linked to relevant subfactors by labeled arrows.

The ecological degradation risk (R) is defined by zones of ecological degradation and critically regulated human activity within the designated boundaries; thus, the conditional probability distribution of R is learned via the BN based on the probabilistic dependencies among key nodes. Risk sources were categorized as patch hazards (P) and anthropogenic hazards (HA). Based on the actual conditions in Xilingol League, anthropogenic hazard (HA) factors were categorized into four types: industry (H1), mineral and energy (H2), transport (H3), and tourism (H4). Exposure potential (E) was categorized into minimum distance to the nearest road (E1), slope (E2), and elevation (E3). Factors for E1 comprised areas inside redlines including railroads (E01), national roads (E02), and provincial roads (E03). Risk receptors were delineated as ecological protection boundaries, categorized by functional ecological entities: cropland (S1), woodland (S2), garden land (S3), water bodies (S4), and grassland (S5).

 Stochastic variable parameterization

A BN, as a probabilistic graphical model, includes both the qualitative network structure illustrating dependencies among variables (Figure 2) and the quantitative parameters defining the distribution of node probabilities, specifically both prior and conditional probability tables. In model development, each node represents a discrete random variable with well-defined spatial states. This study used existing literature on the assessment of the risk of human disturbance and examined the delineation of ECRs alongside the distribution of suspected patches of ecological degradation in Xilingol League. It discretized all risk factors and quantified their gradations through expert judgment, subsequently calculating the corresponding prior probabilities. All risk factors were normalized to a [0,1] range using the min-max method and subsequently discretized into four distinct states: 0–0.25 (very low), 0.25–0.5 (low), 0.5–0.75 (medium), and 0.75–1 (high). The parameterization procedure for each variable is outlined below.

Anthropogenic pressure

Anthropogenic pressure denotes the degree of risk posed by human activity, reflecting the inherent capacity of human disruption to jeopardize protected areas of ecological importance. This research innovatively combined P with HA to collectively define A. In this study, P was derived from the area and quantity of patches within each assessment unit. The patches themselves were identified and classified through visual interpretation of remote sensing imagery. Meanwhile, HA was calculated based on the area proportion of the four human activity types (H1–H4). Spatial processing of these random variables was performed using ArcGIS, succeeded by data normalization and uniform discretization into four state intervals as defined above. As a result, A was categorized into the four states defined above.

 Exposure potential

Regional topographic features and transportation conditions primarily influence the extent of human activity, posing potential disturbances to ecosystems. Areas with lower elevation, gentle terrain, and convenient transportation access are more susceptible to human activity and interference. To quantitatively assess the disturbance of human activity on ECRs, this study selected three evaluation factors defined above as E1, E2, and E3. Here, E1, E2, and E3 were calculated based on a 10 × 10 km buffer around each assessment unit for discretization. Spatial processing for these variables was performed in ArcGIS, followed by data standardization and uniform discretization into four state intervals as defined above (see Figure 2).

 Ecological sensitivity

Ecological receptors generally denote the land types that may be exposed to stressors. Ecological units or land types inside the ECRs were categorized based on their functional distinctions as defined above as S1, S2, S3, S4, and S5. Their sensitivity was quantified by the degree of their exposure to ecological degradation risk receptors. The ecological sensitivity to risk factors reflects the susceptibility of ecological receptors to human disturbance or their inherent protective qualities, categorized into four levels as defined above. Ecological sensitivity is thus categorized into four state intervals as defined above.

 Risk of ecological degradation

The R was derived from the combined risk variables of A, E, and S. After data normalization, R was consistently discretized into the four state intervals as defined above.

 Parameter learning and validation

Learning by a BN was done to ascertain the prior probabilities of root nodes and the conditional probabilities of all subsequent nodes. The calculation outcomes from risk assessment units were used as training samples during model creation. The Expectation Maximization technique repeatedly updated parameters from random beginning values θ0 until convergence was reached, therefore determining the parameters for all nodes in a Bayesian network. This procedure produced conditional probabilities among the nodes in the network.

Subsequently, the k-fold cross-validation technique was used to assess model efficacy and aid in the selection of the Bayesian network architecture. The present study categorized 1,845 samples into 10 categories. Each group functioned as a validation dataset, while the other nine groups served as training data. The cross-validation procedure was executed ten times. Additionally, using Bayesian inference techniques, the posterior probability distributions of child nodes were generated automatically. The geographical likelihood of the risk of regional ecological degradation was quantitatively evaluated by layer-by-layer probabilistic reasoning using risk transmission nodes.

The model dependability was assessed using the Error Rate as the performance indicator to forecast and examine its efficacy. The precise formula for calculation is given in Equation 3: Error Rate = ∑ Error   rate   cases   ∑ Error   rate   cases + ∑ Correct   rate   cases   × 100 %

Diagnostic analysis

The diagnostic assessment of the risk of ecological degradation consisted of two elements: sensitivity and influence types of analyses. Sensitivity analysis measured the contribution and effect of risk variables on the assessment endpoint, while influence analysis delineated extreme risk scenarios by assessing the possible range of risk outcomes under maximum or minimum risk circumstances.

First, sensitivity analysis was performed focused on the assessment endpoint R, calculating the variation reduction (VR) for risk variables, including A, E, and S. The VR was calculated via the following cross-entropy formula, Equation 4: H p , q = − ∑ i p x i l o g q x i where p signifies the observed risk state distribution, q indicates the model-predicted or scenario-defined probability distribution, and p x i and q x i reflect the probability values of the true distribution and predicted distribution at state i , respectively. A higher VR value indicated a more substantial impact of the risk factor on the evaluation endpoint.

Subsequently, this study methodically set the aforementioned input parameters to their zero-state condition (i.e., minimum risk scenario) to compare the relative effects of P and HA on A and R, as well as to analyze how changes in the input parameters impact the distribution of the risk of ecological degradation. A comparison was made between the baseline distribution of risk obtained under the initial model settings and the associated distribution the risk of degradation.

 Results Basic training results

Figure 3 illustrates the baseline training results of the Bayesian network. Approximately 86.21% of the stochastic variables associated with risk factors were mostly classified as being in a very low condition. The parameters ranked as very low included P1; P2; P; H1; H2; H3; H4; HA; A; E01; E02; E03; E1; E3; S1; S2; S3; and S4 (Figure 3). The Xilingol League exemplifies a quintessential grassland environment, with the probability distribution of ecological vulnerability for S5 delineated as follows: very low (20.50%); low (14.06%); mid-range (17.41%); and high (48.00%). According to Bayesian probabilistic reasoning, the distribution of the risk of ecological degradation was as follows: mostly extremely low and low risk with less 1% of the area having either mid-range or high risk (Figure 3). These findings suggest that the overall risk of ecological degradation in the study region can be classified as extremely low to low.

Graphic representation of the basic training result for the assessment of the risk of ecological degradation of Xilingol League, Inner Mongolia, China.

Flowchart diagram showing factors influencing Ecological Degradation Risk, including Patch Hazard, Anthropogenic Pressure, Exposure Potential, and Ecological Sensitivity, with each factor subdivided into quantitative indicators color-coded by risk level from very low (green), low (yellow), medium (orange), to high (red).

The assessment of the error rate for the BN training is shown in Table 1. A total of 1,845 risk assessment units were classified into 10 categories, and the error rate for each category was computed and modified. Error rates for all groups remained below 3.50%, with a mean error rate of 1.25%. These findings demonstrate that the constructed Bayesian network has notable efficacy and reliability.

Error rate results for the Bayesian Network training.

Case number 1 2 3 4 5 6 7 8 9 10 Average error rate
Error rate % 3.24 2.70 1.67 0.54 1.08 1.08 0.32 0.24 0.54 1.08 1.25
Spatial distribution of the risk of ecological degradation

The spatial distributions of R, A, E, and S in Xilingol League are shown in Figure 4. Overall, the E within the ECRs of Xilingol League was relatively low. Among the 1,845 risk assessment units in the league, the number of very low-risk units was as high as 1,511 (81.90%), followed by 132 mid-range risk units (7.15%), 117 low-risk units (6.43%), and 85 high-risk units (4.61%). The top five areas with the highest R were mainly concentrated within the windbreak and sand-fixation ecological protection redline of the Hunshandak Sandland (Table 2). This area, with windbreak and sand fixation as its core ecological function, is an important zone for maintaining regional ecological stability.

Maps with city and county boundaries of Xilingol League, Inner Mongolia, China showing the regional distribution and ecological conservation redlines for water conservation, biodiversity conservation, and wind breaks and sand fixation related to: (a) risk of ecological degradation, (b) anthropogenic pressure, (c) exposure potential, and (d) ecological sensitivity.

Four-panel map graphic showing spatial distributions of (a) Ecological Degradation Risk, (b) Anthropogenic Pressure, (c) Exposure Potential, and (d) Ecological Sensitivity, using color-coded legends ranging from very low to high, over the same geographic area. Each panel highlights varying concentrations of risk or sensitivity, with clustering of high values differing per metric. North direction and scale bar are included.

Five principal assessment units for assessing the risk of ecological degradation inside the ecological conservation redlines (ECRs) of the Xilingol League, Inner Mongolia, China.

Assessment unit Risk values ECRs Function
1,338 0.8575 Hunshandak Windbreaks and sand fixation
277 0.85 Hunshandak Windbreaks and sand fixation
88 0.79 Hunshandak Windbreaks and sand fixation
1,212 0.7325 Greater Khingan range Water conservation
297 0.7225 Hunshandak Windbreaks and sand fixation

The R and A exhibited a significant level of spatial consistency in their distributions. In Xilingol League, of the 1,845 risk assessment units, 1,612 units (87.37%), 98 units (5.31%), 60 units (3.25%) and 75 units (4.07%) exhibited very low, low, mid-range, and high pressure, respectively. The distribution findings demonstrate that most regions exhibited relatively low-pressure levels, with a steady drop in corresponding areas as pressure levels rise.

The regional distribution of E demonstrated significant concentration. A predominant number of places exhibit poor accessibility levels, with 74 units classified as very low accessibility and 1,633 units as low accessibility, together representing 92.52% of the total. By comparison, 123 units with mid-range accessibility and only 15 units had high accessibility, both being quite low shares.

The regional distribution of S revealed considerable variation, substantially affected by grassland sensitivity (V5), with these areas being generally scattered. The bulk of regions exhibited low sensitivity levels, with eight units classified as very low sensitivity and 1,326 units categorized as low sensitivity, together representing 72.30% of the total. A total of 468 and 43 units exhibited mid-range and high sensitivity, respectively, being a rather small fraction of the total. These findings suggest that the ecological sensitivity of the research area mostly ranges from low to mid-range, with a restricted presence of regions classified with high sensitivity.

 Sensitivity analysis

The present study performed a sensitivity analysis on several risk variables and the resulting assessment of the risk of ecological degradation, with the objective of identifying the primary elements driving risk within the ECRs of Xilingol League. Based on the findings (Figure 5), the overall sensitivity of R was ranked from high to low as follows: Anthropogenic pressure (A = 0.16) exceeded ecological sensitivity (S = 0.07), which surpassed exposure potential (E = 0.06).

Sensitivity analysis of ecological degradation risk (a), anthropogenic pressure (b), exposure potential (c) and ecological sensitivity (d).

Four-panel figure with horizontal bar charts labeled a through d, each displaying variables versus VR values. Panel a shows Ecological Degradation Risk with category A having the highest VR of 0.16, followed by HA at 0.11. Panel b presents Anthropogenic Pressure, where HA and H4 lead with VR values of 0.23 and 0.17, respectively. Panel c visualizes Exposure Potential, showing E2 and E1 both at a VR value of 0.11. Panel d illustrates Ecological Sensitivity, with S5 having the highest VR at 0.02. All graphs use blue bars and axes labeled with VR values.

In the category of H, the three most sensitive elements were: anthropogenic pressure (HA = 0.23), tourist development (H4 = 0.17), and patch risk (P = 0.03). In terms of E, the parameters exhibiting considerable sensitivity include minimum distance (E1 = 0.11), slope (E2 = 0.11), and the presence of railroads inside the redline (E01 = 0.04). The primary elements affecting V include grassland (V5 = 0.02), water bodies (V4 = 9.96 × 10−3), and farmland (V1 = 1.74 × 10−3).

The findings of the present study indicate that anthropogenic pressure is the primary element contributing to the risk of ecological degradation in the region, with the intensity of human activity and tourism development being particularly relevant. Although the overall effects of ecological sensitivity and exposure accessibility are minimal, some factors, such as grassland type and transportation accessibility, significantly influence risk. These findings may provide a scientific basis for the rigorous implementation of the ecological protection redline and the formulation of risk minimization, mitigation, and management strategies in Xilingol League.

 Influence analysis

For the analysis of the impacts of the risks of ecological degradation, this study simulated the spatial distribution of H and R under the scenarios where tourism development H4 = 0 and human patch hazard P = 0 (Figure 6).

Maps with city and county boundaries of Xilingol League, Inner Mongolia, China showing the regional distribution and ecological conservation redlines for water conservation, biodiversity conservation, and wind breaks and sand fixation related to the risk of ecological degradation and anthropogenic pressure under two scenarios: (a) risk of ecological degradation in scenario P = 0; (b) risk of ecological degradation in scenario H4 = 0; (c) risk of anthropogenic pressure in scenario P = 0; (d) risk of anthropogenic pressure in scenario H4 = 0.

Four-panel scientific map compares ecological degradation risk (top row, panels a and b) and anthropogenic pressure (bottom row, panels c and d) in a region, using grid squares colored by severity levels from very low to high, with two different scenarios labeled P=0 and H4=0 for each measure, and includes distinct legends for each risk type and a north arrow.

Within the scenario where H4 is set to 0, the delineation of risk zones underwent substantial alterations: the number of high-, mid-range-, low- and very-low-risk zones decreased to five (or by 94.12%), 98 (or by 25.76%), and 63 (or by 46.15%), respectively, while the number of very low-risk zones increased to 1,679 (or by 11.12%). The findings demonstrate that H4 significantly influences R. A reduction of H4 to zero significantly lowered the total level of risk in the area.

In the case where the risk factor P was set to 0, the modifications in risk zoning were as follows: the number of high-risk zones decreased to 76 (or by 10.59%), mid-range risk zones increased to 235 (or by 78.03%), low-risk zones decreased to 11 (or by 90.60%), and very low-risk zones marginally increased to 1,523 (or by 0.79%). The findings suggest that P has a comparatively little influence on the risk R. When P is decreased to 0, this somewhat mitigated danger in high-risk areas but has a negligible effect on mid-range- and low-risk areas.

In the scenario where the stress factor H4 was set at 0, the stress distribution altered markedly: the number of high-, mid-range-, and low-stress zones decreased to four (or by 94.67%), four (or by 93.33%), and nine (or by 90.82%), respectively, while the number of very low-stress zones increased to 1,828 (or by 13.40%). The findings demonstrate that H4 significantly influences ecological stress. The reduction of H4 to 0 significantly reduced the total stress level of the area.

In the scenario where the stress factor P was set to 0, the alterations in stress zoning were as follows: the number of high-, mid-range-, and low-stress zones decreased to 72 (or by 4.00%), 53 (or by 11.67%), and 92 (or by 6.12%), respectively, while the number of very low-stress zones increased slightly to 1,628 (or by 1.00%). These findings suggest that P has a comparatively little influence on ecological stress. When P decreased to 0, this alleviated stress in high-stress areas to some degree, but its effect on other stress-level zones was comparatively small.

 Discussion Spatial risk analysis in Xilingol League

A diagnostic analysis of the risk of ecological degradation in the ECRs within Xilingol League, Inner Mongolia was conducted based on a spatial Bayesian Network. With the goal of quantitatively identifying key stress factors for the risk of regional ecological degradation, high, mid-range, low, and very low risk zones were identified, and accordingly regional risk prevention and mitigation strategies along with regulatory pathways were proposed. The findings suggest that the overall level of risk is quite modest, exhibiting considerable geographical variability. High-risk units were located within ECRs for windbreak and sand fixation in the Hunshandak Sandland. This pattern is mostly influenced by the geographical distribution of the anthropogenic pressure, which exhibited a significant spatial correlation within the distribution of risk, indicating that human activity is the principal determinant of regional ecological degradation risk. Despite the low overall exposure potential in the study region, significant geographical variation exists in ecological sensitivity, notably exhibiting localized high sensitivity in grassland-dominated ecosystems.

 Diagnostic analysis of the risk of ecological degradation

In this study, we explored the primary factors influencing the risk of ecological degradation and analyzed how they impacted that risk inside Xilingol League’s ECRs based on diagnostic findings. With a sensitivity coefficient (0.16) that is noticeably larger than that of exposure potential and ecological sensitivity, the findings demonstrate that anthropogenic pressure is the primary factor influencing the geographical variation in regional ecological degradation risk. The importance of human disturbance in the production of ecological degradation risk is further supported by the sensitivity of that risk to tourism and anthropogenic hazard within anthropogenic pressure. The number of high-risk sites dropped dramatically by 94.12% when H4 decreased to zero, according to impact analysis, and the pattern of anthropogenic pressure experienced a synchronized structural reversal. This suggests that the growth of the tourist industry is a major and real cause of ecological stress in addition to being a sensitive risk factor. It should be emphasized that this result is drawn specifically within the methodological and data framework of the present study, which focused on assessing risks from identifiable, spatially delineated human activity patches. The current analysis was constrained by data availability regarding other impactful anthropogenic pressures and did not encompass the full scope of natural climatic drivers (Gao et al., 2023; Li et al., 2025b), such as persistent drought or altered precipitation regimes, that govern vegetation changes and grassland degradation risk. Key pressures like grazing intensity and energy infrastructure expansion, which could further modulate the overall risk landscape, were not integrated but could be included in future refinements of the model.

The influence of patch hazard demonstrates a significant scale effect. Despite its low worldwide sensitivity (0.03), which results in a minimal impact on the overall risk profile, the 10.59% decrease of patch hazard in high-risk locations identified during the impact analysis demonstrated its considerable effect in specific locales. This discrepancy primarily arises from the spatial scale incongruity between the assessment units and the risk sources: this study employed a 10 × 10 km assessment grid from a regional viewpoint, while the actual destructive patch areas are smaller, varying from fractions to several square km. As a result, the pronounced yet limited impacts of the patch hazard factor are mitigated inside the macroscopic grid evaluation.

Moreover, while exposure potential and ecological sensitivity showed generally low overall sensitivities, grassland and woodland subtypes exhibited distinct spatial responsiveness. This highlights a critical interaction: under specific biophysical conditions, the intrinsic susceptibility of ecosystems, coupled with topographic accessibility, can significantly amplify the adverse impacts of anthropogenic disturbance. The diagnostic framework thus identifies spatially explicit intervention priorities, shifting the focus from generic risk assessment to targeted management. To operationalize these insights, we propose a tiered management approach. In high-risk clusters such as the Hunshandak Sandyland, immediate measures should include a moratorium on new tourism infrastructure and the implementation of seasonal grassland closures or regulated rotational grazing. Conversely, in areas where ecologically sensitive zones (e.g., S5 grassland types) intersect with intense anthropogenic pressure, management should emphasize adaptive governance-potentially through micro-zoning adjustments within the ECRs and enhanced high-frequency monitoring protocols to ensure compliance. These spatially differentiated strategies translate diagnostic risk patterns into a concrete basis for preventive ecosystem governance.

 Evaluation of the Bayesian model

Parameterization is a crucial component of Bayesian models during the assessment of the risk of ecological degradation. We created a spatial distribution pattern of that risk by combining existing information and pertinent research, and we examined the intrinsic processes by which different risk variables affect ecological degradation (Gao et al., 2023). In contrast to traditional evaluation techniques (Liu et al., 2023; 2025), the Bayesian model provided probabilistic risk information and aided in determining uncertainties associated with grassland degradation, particularly when comprehensive or adequate data is lacking. Additionally, the model is very adaptable, enabling ongoing updates and optimization in response to new data, which supports prompt decision-making in managerial interventions.

The probability links between nodes may be efficiently ascertained by integrating GIS data as prior knowledge. Geographical visualization assisted in identifying regions where risks related to human disturbance are most likely to occur, which is a crucial step in comprehending actual ecological processes. Notable changes in the distribution of ecological degradation risk were also noted during GIS-based spatial simulations. Furthermore, the spatial Bayesian network provided deeper insights into the linkages among risk indicators and their roles in complete risk assessment than conventional multi-criteria decision-making techniques such as landscape risk indices (Karimian et al., 2022; Yang et al., 2024). Consequently, there were fewer false positives, more precision in detecting high-risk areas, and better risk categorization accuracy. These benefits show that when it comes to assessing ecological degradation risk, the Bayesian approach is more dependable and portable.

 Limitations and prospects

It is essential to acknowledge the constraints and possible implementation difficulties related to this risk prediction model. Improvements may be required to address specific urban attributes, contextual situations, and localized risk issues, necessitating the active participation of local specialists. Future study may enhance the model’s reliability and practical usability through the following avenues.

Conducting dynamic risk assessment and scenario simulation: Future research should integrate regional development plans to produce scenarios that reflect differing intensities of human activity and effects of climate change. This can be achieved by prioritizing the acquisition and integration of multi-temporal spatial data spanning consecutive years. These efforts would provide dynamic forecasting of the progression of spatiotemporal risk and help researchers to accurately ascertain contribution levels of various risk factors, thus providing a quantitative basis for targeted risk source control.

Enhancing assessment scales: Although this study used bi-monthly remote sensing data, the resolution of the assessment units was still quite coarse. Future research could use higher-resolution data to conduct assessments at more granular sizes, such as kilometer-level grids or assessments at specific ecological units, to enhance the identification of localized and hidden threats.

It is proposed to include the provision, regulatory, and cultural functions of ecosystems as fundamental components in risk assessment by adopting an ecosystem services perspective. Assessing the impact of ecological degradation on the degradation of these services would transition risk management goals from preserving the “ecological baseline” to ensuring “human wellbeing”.

 Conclusion

This study constructed a Bayesian network-based model for assessing the risk of ecological degradation, using patches based on areas of ecological degradation, and deployed the model in Xilingol League for empirical validation. The study objectively assessed the spatial distribution patterns of the risk of ecological degradation and analyzed the inherent mechanisms by which different risk factors affect ecological degradation. The primary conclusions are as follows.

The danger of ecological degradation within the ECRs of Xilingol League has a pattern of “generally low risk levels with localized high-risk clusters;” meanwhile, zones with extremely low risk comprised 81.90% of the overall area. A restricted quantity of high-risk units (85 units, constituting 4.61%) exhibited notable spatial clustering inside the windbreak and sand-fixation ecological protection boundaries of the ecologically significant Hunshandak Sandland.

The probability of ecological degradation in Xilingol League is collectively influenced by anthropogenic pressure, exposure accessibility, and ecological sensitivity. The spatial variability of risk is predominantly influenced by human activity and is especially evident in limited regions with elevated ecological sensitivity.

Anthropogenic pressure is the primary factor influencing the danger of ecological degradation in Xilingol League. Tourism development has exhibited extensive worldwide impact on the global environment, whereas human patch risk has significant localized implications via scale-dependent mechanisms. This research employed nonlinear quantitative assessment to clarify the risk-driving mechanisms centered on anthropogenic pressure, thereby offering a crucial scientific basis for differentiated management strategies regarding ECRs and targeted governance of risk sources in Xilingol League.

 Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

 Author contributions

JL: Funding acquisition, Methodology, Resources, Visualization, Writing – original draft, Writing – review and editing. DL: Formal Analysis, Funding acquisition, Investigation, Writing – review and editing. SW: Software, Writing – original draft. XX: Conceptualization, Formal Analysis, Funding acquisition, Methodology, Software, Writing – original draft. FJ: Formal Analysis, Writing – review and editing. JQ: Resources, Writing – review and editing. MC: Funding acquisition, Resources, Supervision, Writing – review and editing. XS: Resources, Writing – review and editing. KZ: Resources, Writing – review and editing.

 Acknowledgements

The authors are grateful to the reviewers for their comments and suggestions which contributed to the further improvement of this paper.

 Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

 Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

 Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

 References Chen S. Chen B. Fath B. D. (2013). Ecological risk assessment on the system scale: a review of state-of-the-art models and future perspectives. Ecol. Model. 250, 25–33. 10.1016/j.ecolmodel.2012.10.015 Duan S. Yang Z. Han F. Liu T. (2025). Habitat identification and fragmentation risk assessment of key species in the Altai Mountains transboundary region. Environ. Sustain. Indic. 27, 100816. 10.1016/j.indic.2025.100816 Forbes V. E. Galic N. (2016). Next-generation ecological risk assessment: predicting risk from molecular initiation to ecosystem service delivery. Environ. Int. 91, 215–219. 10.1016/j.envint.2016.03.002 26985654 Gao X. Cao C. Xu M. Yang X. Li J. Shea Duerler R. (2023). Ecological risk assessment of grassland vegetation change based on Bayesian model in Xilin Gol League, China. Ecol. Indic. 157, 111199. 10.1016/j.ecolind.2023.111199 Harris M. J. Stinson J. Landis W. G. (2017). A Bayesian approach to integrated ecological and human health risk assessment for the South river, Virginia mercury‐contaminated site. Risk Anal. 37, 1341–1357. 10.1111/risa.12691 28121045 Kaikkonen L. Parviainen T. Rahikainen M. Uusitalo L. Lehikoinen A. (2020). Bayesian networks in environmental risk assessment: a review. Integr. Environ. Assess. Manag. 17, 62–78. 10.1002/ieam.4332 32841493 Karimian H. Zou W. Chen Y. Xia J. Wang Z. (2022). Landscape ecological risk assessment and driving factor analysis in Dongjiang river watershed. Chemosphere 307, 135835. 10.1016/j.chemosphere.2022.135835 35964726 Kayumba P. M. Chen Y. Mindje R. Mindje M. Uwamahoro S. (2022). Spatial ecological risk assessment and its landscape dynamics simulation in Bayanbulak wetland, Northwestern China. Land Use Policy 114, 105968. 10.1016/j.landusepol.2021.105968 Li J. Chen X. Kurban A. Van de Voorde T. De Maeyer P. Zhang C. (2021). Identification of conservation priorities in the major basins of central Asia: using an integrated GIS-based ordered weighted averaging approach. J. Environ. Manag. 298, 113442. 10.1016/j.jenvman.2021.113442 34371221 Li J. Liang W. Xu M. Tian H. Gao X. Yang Y. (2025a). Threshold extraction and early warning of key ecological factors for grassland degradation risk. Remote Sens. 17, 3098. 10.3390/rs17173098 Li J. Wang Y. Fu Y. Chen Y. (2025b). Ecological risk transmission network and its response characteristic in heavy metal-contaminated soil ecosystem under different scenarios. Environ. Pollut. 385, 127083. 10.1016/j.envpol.2025.127083 40935062 Liu J. Xu X. Zou C. Lin N. Zhang K. Shan N. (2023). A Bayesian network-GIS probabilistic model for addressing human disturbance risk to ecological conservation redline areas. J. Environ. Manag. 344, 118400. 10.1016/j.jenvman.2023.118400 37331314 Liu J. Luan C. Xu X. Zhu Y. Qiu J. Liu R. (2025). Assessment and prediction of ecological risk caused by anthropogenic activities during the implementation of China's ecological protection redline policy: a case study from the Yangtze river Delta. J. Clean. Prod. 533, 146962. 10.1016/j.jclepro.2025.146962 Lu J. Lu H. Wang W. Feng S. Lei K. (2021). Ecological risk assessment of heavy metal contamination of mining area soil based on land type changes: an information network environ analysis. Ecol. Model. 455, 109633. 10.1016/j.ecolmodel.2021.109633 Lu Y. Zhai G. Zhou S. (2024). An integrated Bayesian networks and geographic information system (BNs-GIS) approach for flood disaster risk assessment: a case study of Yinchuan, China. Ecol. Indic. 166, 112322. 10.1016/j.ecolind.2024.112322 Nishino T. Miyashita T. Mori N. (2024). Methodology for probabilistic tsunami-triggered oil spill fire hazard assessment based on Natech cascading disaster modeling. Reliab. Eng. and Syst. Saf. 242, 109789. 10.1016/j.ress.2023.109789 Orak N. H. (2020). A hybrid Bayesian network framework for risk assessment of arsenic exposure and adverse reproductive outcomes. Ecotoxicol. Environ. Saf. 192, 110270. 10.1016/j.ecoenv.2020.110270 32036100 Pearl J. (1986). Fusion, propagation, and structuring in belief networks. Artif. Intell. 29, 241–288. 10.1016/0004-3702(86)90072-x Qian L. Shi Y. Xu Q. Zhou X. Li X. Shao X. (2023). A prospective ecological risk assessment method based on exposure and ecological scenarios (ERA-EES) to determine soil ecological risks around metal mining areas. Sci. Total Environ. 901, 166371. 10.1016/j.scitotenv.2023.166371 37604368 Qiu M. Fu M. Zhang Z. Fu S. Yuan C. (2023). Assessing the ecological risk of croplands in loess drylands by combining environmental disturbance with ecosystem vulnerability. J. Environ. Manag. 347, 119231. 10.1016/j.jenvman.2023.119231 37804628 Stritih A. Bebi P. Rossi C. Grêt-Regamey A. (2021). Addressing disturbance risk to mountain forest ecosystem services. J. Environ. Manag. 296, 113188. 10.1016/j.jenvman.2021.113188 34225045 Sun M. Li X. Yang R. Zhang Y. Zhang L. Song Z. (2020). Comprehensive partitions and different strategies based on ecological security and economic development in Guizhou Province, China. J. Clean. Prod. 274, 122794. 10.1016/j.jclepro.2020.122794 Suter G. Barron M. Wayne M. (2016). Generic ecological assessment endpoints (GEAEs) for ecological risk assessment: second edition with generic ecosystem services endpoints added. Reston, VA: U.S. Environmental Protection Agency. Vargas-Jaimes J. Manjarrez J. Ramírez-Corona F. Woolrich-Piña G. A. González-Pedroza M. G. Estrada-Zúñiga M. E. (2025). Assessment of the vulnerability of spiny Lizards in central Mexico under anthropogenic pressure: integrating habitat suitability, landscape connectivity, fragmentation and protected area coverage. Nat. Conserv. 59, 351–386. 10.3897/natureconservation.59.159660 Veisi Nabikandi B. Hami A. Valizadeh Kamran K. Russo A. (2025). Evaluating urban ecosystem services and resilience using remote sensing and InVEST model: a case study of flood risk control and urban cooling in Tabriz. Sustain. Cities Soc. 130, 106654. 10.1016/j.scs.2025.106654 Wang Y. Liu H. (2025). Integrating landscape ecological risk and ecosystem services for ecological zoning and management in the Wuling Mountain Area. Ecol. Indic. 179, 114179. 10.1016/j.ecolind.2025.114179 Wang C. Wei M. Wu B. Wang S. Jiang K. (2019). Alpine grassland degradation reduced plant species diversity and stability of plant communities in the Northern Tibet Plateau. Acta Oecol. 98, 25–29. 10.1016/j.actao.2019.05.005 Wang H. Hu Y. Liang Y. (2021). Simulation and spatiotemporal evolution analysis of biocapacity in Xilingol based on CA-Markov land simulation. Environ. Sustain. Indic. 11, 100136. 10.1016/j.indic.2021.100136 Wang S. Xu Q. Yi J. (2025). Multi-scale identification of driving forces for landscape ecological risk: a case study of the basin containing three plateau lakes in Yunnan, China. Ecol. Indic. 170, 113037. 10.1016/j.ecolind.2024.113037 Yan F. Li N. Wang J. Wu H. (2023). Ecological footprint model of heavy metal pollution in water environment based on the potential ecological risk index. J. Environ. Manag. 344, 118708. 10.1016/j.jenvman.2023.118708 37541000 Yang N. Zhang T. Li J. Feng P. Yang N. (2024). Landscape ecological risk assessment and driving factors analysis based on optimal spatial scales in Luan river basin, China. Ecol. Indic. 169, 112821. 10.1016/j.ecolind.2024.112821 Yang W. Chen J. Ding T. Li Y. Gong W. (2025). Probabilistic assessment for drought risk: integrating drought hazard, ecological sensitivity, economic vulnerability, and their coupling coordination. J. Hydrology 658, 134451. 10.1016/j.jhydrol.2025.134451 Zeng J. Wu J. Chen W. (2024). Coupling analysis of land use change with landscape ecological risk in China: a multi-scenario simulation perspective. J. Clean. Prod. 435, 140518. 10.1016/j.jclepro.2023.140518 Zhang X. Song W. (2025). From static to predictive indicators: construction of China’s spatiotemporal risk zoning index for land use changes ecological risks. Ecol. Indic. 179, 114154. 10.1016/j.ecolind.2025.114154 Zhang Z. Hu B. Qiu H. (2021). Comprehensive assessment of ecological risk in southwest Guangxi-Beibu bay based on DPSIR model and OWA-GIS. Ecol. Indic. 132, 108334. 10.1016/j.ecolind.2021.108334 Zhang S. Jiang H. Yu H. Feng X. Fan M. (2024). Construction of landscape ecological network based on MCR risk assessment model: a case study of Liaoning Province, China. Ecol. Indic. 166, 112549. 10.1016/j.ecolind.2024.112549 Zou J. Zhou Z. Chu Y. Chen X. Zhang D. He Y. (2025). Dynamic risk assessment and sensitivity analysis of lightning-induced Natech tank fire based on probabilistic chain model and Bayesian network. J. Loss Prev. Process Industries 98, 105762. 10.1016/j.jlp.2025.105762

Edited by: Wei Jiang, Shandong University, China

Reviewed by: Lixiang Fu, Ocean University of China, China

Zemin Zhi, Fuyang Normal University, China