Petrochemical Risk Assessment in Coastal China and Implications for Land-Use Dynamics

Abstract

Land-use change and its interaction with petrochemical accident risk are critical for sustainable coastal development. This study established a multi-source data-integrated risk assessment framework, employing fuzzy C-means clustering to stratify petrochemical accident risk into six distinct levels. The analysis revealed the relationship between these risk levels and land-use type changes. Furthermore, the Takagi–Sugeno fuzzy dynamic model was applied to evaluate potential risks at representative coastal petrochemical enterprises. The findings were as follows: (1) Risk concentrates in small-to-medium private, newly established firms, primarily as explosion accidents. (2) The highest risk occurs in Bohai Bay, followed by Jiangsu, Zhejiang, and Guangdong; national policies have reduced affected zones from 352.61 km² (2019) to 43.67 km² (2022). (3) The total potential risk zone spans 2986.21 km², with high-risk cores in Hebei, Zhejiang, and Fujian (36.52%) and medium-risk in Shandong Peninsula (32.01%). (4) Risk primarily affects farmland and construction land; urban expansion has increased affected built-up areas from 16.36% (2012) to 47.02% (2022), shifting effects from ecological to combined socio-ecological consequences. These findings provide critical theoretical support and actionable management recommendations for integrating coastal land-use planning, urban expansion control, and coordinated petrochemical risk governance.

1. Introduction

Globally, the chemical industry generates a total annual revenue of approximately USD4 trillion, of which nearly 45% comes from China [1]. In recent decades, China’s economic growth has been largely driven by the chemical industry [2]. However, the rapid development of this industry has also brought about many problems, among which production safety is prominent [3]. Although the government and local authorities have developed relevant regulations and technical recommendations on this issue and taken many measures to improve the chemical safety environment in China, the growing risk of accidents cannot be ignored with the growth of chemical industry clusters and the high frequency of major accidents [4,5]. The likelihood and severity of accidents can be eliminated through proper risk assessment and management [6]. As defined by the Society for Risk Analysis (SRA) [7], the purpose of risk assessment is to study and address the risks associated with specific activities (e.g., those related to petrochemical company operations, investments, and natural phenomena) to prevent accidental occurrences and mitigate their consequences [8]. The assessment of potential risks in a petrochemical company can determine the severity of the current risks and whether they are acceptable. This enables managers to make rational decisions about risk control [9].

Petrochemical process safety research relies primarily on two pillars: analysis of historical accidents and experimentation [10,11]. Compared to other scientific and technological fields, experimental research on accident prediction and prevention in the petrochemical risk area is relatively limited. This makes historical accident analysis of irreplaceable importance within the industry [12], as reconstructing accident processes and assessing their probability and severity provides a key basis for developing accident prevention strategies.

Research on the impact of petrochemical risks has been conducted from multiple perspectives. Some studies have taken a statistical perspective, analyzing the frequency and spatial distribution of accidents. For example, Xiang et al. [13] systematically examined 478 major industrial accidents in China between 2000 and 2020, finding that they were largely concentrated in the eastern coastal provinces. Based on a review of 169 accidents in the French chemical industry, Dakkoune et al. [14] pointed out that the industry presents significant risks and urgently needs national regulations or recommendations. Chen and Reniers [2] summarized the statistical characteristics and direct causes of chemical accidents in China. Nivolianitou et al. [15] analyzed 85 major accidents in the European petrochemical industry and showed that the sector is one of the most severely affected by safety accidents. Similarly, some scholars have focused on specific time periods or types of accidents. For example, Duan et al. [16] and Zhang and Zheng [17] studied hazardous chemical accidents in China from 2000 to 2010, He et al. [18] examined the spatiotemporal distribution of such accidents, Wang et al. [19] conducted a statistical analysis of 76 serious hazardous chemical accidents during the hot season, and Wang et al. [20] identified patterns in accident occurrence among small- and medium-sized enterprises (SMEs) and emphasized the need to strengthen their accident prevention capabilities. Bai et al. [21] further suggested that loopholes in the implementation of process safety management elements in enterprises are a major cause of the continued occurrence of major accidents in China.

Another type of research takes an enterprise-level approach, aiming to identify the risks of petrochemical enterprises in terms of structure, site [22] and surrounding environmental health [23]. For example, Li [24] assessed corporate vulnerability to earthquakes, and Zhou [25] studied risk levels in terrorist attack scenarios. Other scholars have focused on specific risk factors within production processes [26,27,28,29], systematically analyzing production vulnerabilities, security countermeasures, and emergency responses [30], providing theoretical support and practical guidance for corporate risk prevention and control.

To avoid the impact of uncertainty on the accuracy of risk assessment due to incomplete information and inaccurate measurements of petrochemical accidents, scholars introduced probabilistic analyses into risk assessment, with Bayesian subjective probability methods being the most commonly used. Bayesian networks were used to address data uncertainty, and a priori risk was combined with data monitoring [31] to study safety risks in the chemical industry [32,33,34]. Furthermore, in instances where prior conditions or information are incomplete, researchers have increasingly identified the limitations of probabilistic analysis and have proposed the utilization of alternative methodologies, such as fuzzy probability or likelihood, for the assessment of uncertainty [35]. In this context, fuzzy set theory is also a common method for addressing uncertainty and ambiguity in risk assessment. Some scholars have employed the fuzzy bow-tie model to construct a risk management framework [36], while others have utilized Takagi–Sugeno (T-S) reasoning to enhance the bow-tie model for risk assessment in the chemical industry [37].

The fuzzy comprehension evaluation method enables the synthesis of the characteristics of the influencing factors, the construction of the affiliation function, and the determination of the affiliation degree of each indicator, thus facilitating the final ranking of importance [38]. The primary challenge in comprehensive evaluation at this stage is establishing a scientific index grading system and an objective affiliation matrix. The fuzzy clustering algorithm, which has reached a considerable degree of theoretical maturity and is extensively employed in unsupervised learning, allows for the affiliation level of elements within each cluster to be identified through unsupervised learning. The fuzzy characterization of each cluster enables the conversion between numerical values and fuzzy language [39]. The combination of Fuzzy C-Means (FCM) and T-S models allows for the efficient assessment of risk in the study area by leveraging the fuzziness inherent to FCM data clustering to identify patterns and groups of potential risks. Subsequently, fuzzy reasoning and decision-making on these patterns are conducted by T-S models. This combination is capable of handling complex and fuzzy risk information and accounting for nonlinear and multiple relationships between systems, thereby facilitating more accurate and comprehensive risk assessment results.

Although existing research has significantly advanced the theory and methods of petrochemical risk analysis, significant limitations remain. First, current research focuses primarily on the petrochemical industry itself, such as accident statistics analysis and risk assessments for individual companies, facilities, or production processes, while ignoring the comprehensive impact of accident risks on regional economies, the ecological environment, and sustainable social development. In the context of rapid urbanization [40], changes in land-use types directly affect land values and the distribution of petrochemical hazards, leading to spatial differentiation in population distribution, economic development levels, and disaster resilience. However, this cross-cutting impact has not yet been systematically assessed. Second, existing research generally relies on a single data source, primarily relying on internal company records or major accident archives. It lacks the integration and application of multi-source heterogeneous data, such as news and public opinion, remote sensing land use, and company information, making it difficult to support large-scale dynamic risk identification and assessment.

This study proposes a multi-source data fusion framework for petrochemical risk assessment. By integrating diverse datasets, such as historical accident records, enterprise operational information, and dynamic land-use patterns, the framework systematically delineates and classifies the spatial risk patterns and affect zones of the coastal petrochemical industry. Our primary contributions are threefold: First, we broaden data sources and enhance comprehensiveness and accuracy by extracting petrochemical accident data from news reports and integrating it with enterprise information. Second, leveraging multi-source data, we construct a risk assessment system centered on hazard intensity and accident likelihood. This system employs FCM clustering and T-S dynamic fuzzy modeling to grade risk levels, identify potential risk zones, and delineate petrochemical risk-affect boundaries. Third, by incorporating dynamic changes in coastal land-use types, we assess how the effects of petrochemical accidents correlate with urban expansion in coastal areas and evaluate their implications for coastal urban sustainability, thereby providing a scientific foundation for regional risk management.

2. Materials and Methods

2.1. Study Area and Data Sources

The petrochemical industry is a substantial component of China’s industrial system and plays a crucial role in the country’s economic and social development. The coastal areas, being market-oriented and in proximity to the raw material production area, offer the advantage of convenient transport and low freight costs, making them the ideal location for countries and regions that prioritize the import and export of raw materials and manufactured products [41]. The Petrochemical Industry Planning and Layout Scheme formulated by China’s National Development and Reform Commission prioritizes coastal locations for petrochemical industry development. Petrochemical accidents in coastal provinces account for 59.5% of the national total, with a total death toll of 60.8% of the national total. Petrochemical hazards have a significant negative impact on the safe development of coastal cities. Analyzing the current dangers and potential risks of coastal petrochemical accidents helps us understand the patterns of petrochemical risk and form recommendations for the safe development of coastal cities. Consequently, China’s coastal provinces were selected as the study area for analyzing petrochemical risks. The study area is shown in Figure 1.

This study selected typical coastal enterprises applying for licenses to produce or operate dangerous chemicals in the industries “2511 Crude oil processing and petroleum products” and “2614 Manufacture of organic chemical raw materials” in coastal cities (excluding petrol stations). This study focused on petrochemical accidents that occur during production and storage. As Wang’s [19] research, these phases are the most common for petrochemical accidents.

This study uses three categories of data: petrochemical accident, enterprise information data, and land-use data. The specific sources of these datasets are presented in

Table 1. Data source.

Data Type	Data Source	Data Description	Processing
Petrochemical accident data	China Chemical Safety Association	2012–2022/Test	Crawler technology is used on these sources of petrochemical accident data to obtain information about the incident itself (human error or equipment malfunction and condition of the site), the emergency response, consequences of the incident (injuries and property damage), and the location of the incident. A total of 960 chemical accidents have been documented.
	Chemical Accident Information Network
	Ministry of Emergency Management of the People’s Republic of China
Enterprise information data	National Enterprise Credit Information Publicity System	2012–2022/Test	The enterprise name can be used as a keyword to retrieve additional information from the national enterprise credit information publicity system, including the age, registration status, size, and nature of the enterprise.
Land-use data	China Land Cover Dataset (CLCD)	30 m/2012–2022/Raster	This dataset classifies land cover into nine distinct classes: cropland, forest, shrubland, grassland, water, snow/ice, bare land, impervious surfaces, and wetland [42]. Spatial projection transformation, study area extraction, cloud masking, and additional preprocessing are applied. CLCD is the first annual land cover dataset in China. The overall classification accuracy is above 80%, which can well meet the requirements of this study for event continuity and regional integrity.

Note: As this study focuses on characterizing land-use changes, the term built-up area (BUA) is used throughout to represent impervious surfaces.

.

2.2. Multi-Source Data-Driven Risk Assessment Frameworks for Coastal Petrochemical Clusters

Coastal petrochemical clusters face multifaceted risks originating from three interconnected sources: the spatial legacy of historical accidents, inherent operational and managerial safety hazards within production/storage/transport processes (internal enterprise risks), and dynamic land-use changes. These factors interact through complex spatiotemporal chain reactions and regional accumulation effects. To address this, we developed a multi-source data-driven risk assessment framework that systematically identifies risk levels, spatial distribution patterns, and dynamic interactions with land-use transformations.

First, multi-source data, including media reports (2012–2022), accident records from the Ministry of Emergency Management, and enterprise information, were integrated to construct a composite database. This database incorporates metrics for accident severity (fatalities/injuries, economic losses, environmental impacts) and processes hazard levels. Then, a dual-dimensional hazard-likelihood indicator system was established. FCM clustering was applied to classify historical accidents into refined risk grades. Building on this, a T-S dynamic fuzzy model was employed to assess potential hazards at coastal enterprises in 2022, generating spatial distributions of potential risk zones. Subsequently, land-use types (2012–2022) and urban expansion patterns were extracted to quantify the impact severity on dynamic land categories (e.g., BUA, ecological land) within petrochemical accident footprints. Figure 2 illustrates this multi-source data-driven framework for potential risk assessment in coastal petrochemical clusters.

2.2.1. Petrochemical Accident Risk Assessment System

The SRA glossary defines risk as a combination of the uncertainty of the occurrence of a negative event and the severity of the event [30,43]. The term “severity” is used to describe the intensity, size, scope, and other possible measures of magnitude in relation to things valued by human beings (life, the environment, money, etc.). In China, the existing petrochemical accident severity classification follows the 2007 “Reporting and Investigation of Production Safety Accidents and Handling Regulations” [31,44]. This is illustrated in

Table 2. Accident level classification criteria.

Severity	Classification	Number of Fatalities (N1)	Number of Seriously Injured (N2)	Direct Economic Loss (106 RMB) (N3)
IV	General accidents	N1 < 3	N2 < 10	N3 < 10
III	Larger accidents	3 ≤ N1 < 10	10 ≤ N2 < 50	10 ≤ N3 < 50
II	Major accidents	10 ≤ N1 < 30	50 ≤ N2 < 100	50 ≤ N3 < 100
I	Particularly major accidents	N1 ≥ 30	N2 ≥ 100	N3 ≥ 100

. However, given that production safety accidents occur across a range of industries [7,32,45], the classification standard for production safety accidents does not fully account for the distinctive features of petrochemical accidents. This includes the varying risks posed by different types of incidents, such as fires and explosions. Furthermore, the current classification system does not consider the effects at the enterprise level.

We referred to Aven and Renn [46], which employed the likelihood of an accident occurring in place of the probability of occurrence as a means of measuring the uncertainty of risk. Although probability is a commonly used tool for expressing uncertainty, it is more challenging to convert the available information on accidents into probability, and there are inherent inaccuracies. As observed by the SRA [7], if the probability cannot be readily determined or agreed upon, alternative methodologies may be employed. Accordingly, this study developed likelihood indicators to quantify the uncertainty of an accident occurring in a petrochemical enterprise.

The system for assessing the risk of petrochemical accidents is as follows:

Objective: Layer A involves creating a petrochemical risk index to assess accident risk levels.

Criterion: Layer B directly determines the scientific validity of the assessment results. By combining the two factors of accident severity and likelihood, it is possible to determine a petrochemical accident’s risk level more accurately. The following is a description of the conditional layer.

• The severity of accident risk reflects the negative impact of petrochemical accidents on the economy and society. It is directly related to economic losses, casualties, and accident types.
• The likelihood of an accident reflects the safety characteristics of the petrochemical enterprises. This refers to the likelihood of a petrochemical accident occurring in different enterprises. According to the research of Wang et al. [20], the likelihood of accidents in petrochemical enterprises is closely related to factors such as the equipment condition and management level of the enterprise. We use “enterprise age” to indirectly measure the degree of equipment aging, while “enterprise nature”, “enterprise scale” and “registration status” are mainly used to evaluate the management standards and safety culture level of the enterprise.

Indicator Layer C: For each criterion layer, we selected indicators that reflect the current carrying capacity and the potential for increasing the original capacity. We selected four to seven representative indicators for each criterion layer, as shown in

Table 3. Indicator system for assessing petrochemical risk levels.

Target Layer A	Criteria Layer B	No.	Indicator Layer C	Interpretation of Indicators
Petrochemical risk level assessment	Severity of accident	X1	Accident level	The classification is based on the criteria for classifying workplace safety accidents, as shown in Table 1.
		X2	Classification of accidents	Types of accidents are classified as explosions, fires, spills, asphyxiation, poisoning, and other accidents.
		X3	Amount of compensation for deceased persons	Under the provisions of the Work Safety Law of the People’s Republic of China, those who die in work safety accidents are compensated with a one-time death benefit, which is calculated at 20 times the per capita disposable income of urban residents nationwide for the previous year.
		X4	Death toll	Number of deaths due to accidents
		X5	Seriously injured	Number of people seriously injured from accidents
		X6	Minor injuries	Number of minor injuries due to accidents
		X7	Total number of casualties	Total number of casualties due to accidents
	Likelihood of accident	X8	Age of enterprise	Time between the establishment of the enterprise and the occurrence of the accident
		X9	Enterprise registration status	Includes three categories: surviving, revoked, and canceled
		X10	Nature of enterprise	Including state-owned enterprises, private enterprises, foreign enterprises, joint ventures, and Hong Kong, Macao, and Taiwan investment enterprises
		X11	Size of enterprise	Includes four categories: micro and small enterprises, small enterprises, medium-sized enterprises, and large enterprises

Note: Economic losses are more difficult to measure, and the amount of compensation for fatalities is used in place of this.

.

Based on multi-source data and after a comprehensive assessment of the likelihood and severity of accidents in enterprises, we classified them into six levels following the Guidelines for the Implementation of Risk Classification and Control of Chemical Enterprises [47] to assess petrochemical risks more accurately. Compared to the current four-level classification, the six-level classification incorporates the risk of no-impact accidents, without impact on the economy and society, and enterprise. Previously, the threshold of 30 fatalities was established between previous major and particularly major accidents. The classification of petrochemical accident risk was based on 10, 20, 30, 40, and 40 or more, respectively, with 10 as the progression unit. The classification of economic loss was similar. Determining the impact of petrochemical risks is therefore more refined, which allows for a more targeted approach to accident preparation and prevention. The six-level classification is therefore more comprehensive than the previous classification, as it considers not only the complexity and subsequent impact of the accident but also the impact on the economy and society, as well as business development. Specific grading content is shown in

Table 4. Level the impact of petrochemical accidents.

Name of Incident	Accident Level	Meaning	Significance	Area of Influence (n)/km
Extremely low-impact accidents	VI	No casualties, low economic losses	Little to no negative impact on business, the economy and society	0
Lower impact accidents	V	Death toll less than 4 and economic losses not exceeding 10 million	The degree to which negative impacts are low and do not harm the environment, population, or society, and the impacts on business are much greater than the impacts on the economy and society	1
General impact accidents	IV	Death toll less than 10 (greater than 4) and economic losses not exceeding 20 million	Minor impacts on the overall environment, population, or society, with a greater impact on business than on the economy and society	2
Higher impact accidents	III	Death toll less than 20 (greater than 10) and economic losses not exceeding 50 million	A greater negative impact on the environment, population, or society, with no higher or lower impact on business, economy, and society	4
Serious impact accidents	II	Death toll less than 40 (greater than 20) and economic losses not exceeding 100 million	Serious negative impacts on the environment, population, or society, with greater economic and social impacts than business impacts	6
Accidents with particularly serious effects	I	Death toll greater than 40 and economic losses exceeding 100 million	Very serious negative impact on the environment, population, or society, with an economic and social impact that is much greater than the impact on business	8

Note: The basis for the division of the scope of impact is extracted from the characteristics of the scope of impact of accidents that have occurred. The impact range of each accident is evaluated, thereby determining the typical impact radius under different accident levels.

.

2.2.2. Methods and Testing

1.

FCM clustering

The FCM algorithm incorporates the essence of fuzzy theory. Compared to the hard clustering of k-means, the FCM algorithm provides more flexible clustering results. This is a soft clustering method [48] that uses the degree of affiliation to express the relationship between each data point and to determine the clusters to which each data point belongs. The FCM algorithm is an objective function-based algorithm used to classify a dataset containing n data points: $\mathrm{X}=\left\{x_1, \right.x_2, \dots x_i, \left.x_n\right\}. X_I$ represents the ith feature vector and $X_{Ij}$ represents the jth attribute of $X_I$. Each sample contained j attributes. The FCM algorithm can classify this dataset into K classes, where K is a positive integer greater than one. The clustering centers of the K classes are $\left[v_1, v_2, \dots , v_n\right]$.

The objective function and the constraints of the FCM are defined as follows:

$$\mathrm{J}\left(U,V\right)={\sum }_{i=1}^n{\sum }_{j=1}^k{u}_{ij}^m{d}_{ij}^2$$

(1)

$${\sum }_{j=1}^k{u}_{ij}=1, u_{ij}\subset \left[\mathrm{0,1}\right]$$

(2)

The degree of membership between sample point $x_i$ and clustering center $v_j$ is represented by $u_{ij}$, where m is the fuzzy index (m > 1) and m = 2 is regarded as the most suitable for most applications, and $d_{ij}$ is the distance between the sample point $x_i$ and clustering center $v_j$. The Euclidean distance is generally used. The FCM algorithm obtains a fuzzy classification of a sample set by iteratively optimizing the objective function.

The Lagrange multiplier method is used to obtain the membership matrix U and the clustering center $v_j$ that minimize the objective function J, subject to the given constraints.

$$u_{ij}=1/{\sum }_{c=1}^k{\left(d_{ij}/d_{ik}\right)}^{2/m-1}$$

(3)

$$v_j={\sum }_{i=1}^n{u}_{ij}^m{x}_i/{\sum }_{i=1}^n{u}_{ij}^m$$

(4)

The optimal clustering center, designated as$v_j$, is identified, and the fuzzy affiliation matrix is generated to facilitate the automatic classification of samples based on the principle of maximum affiliation.

2.

T-S fuzzy model

The T-S fuzzy model is a common fuzzy dynamic model used in the fuzzy modeling of complex nonlinear systems. It was proposed by Takagi and Sugeno [49] and is based on a rigorous mathematical model that is essentially nonlinear. Each subsystem constituting the model is linear. The premise of the model is based on the existence of a local linear relationship between the system’s inputs and outputs. The system’s premise is to determine whether a local linear relationship exists between the inputs and outputs. This conclusion is expressed using polynomial linear equations that form a linear combination of rules. This results in a good linear descriptive property of the global output of a nonlinear system.

When describing the posterior of a multiple-input multiple-output nonlinear continuous dynamic system using a state space equation based on the T-S fuzzy model, the fuzzy rule can be expressed as follows:

Rule i:

$$\mathrm{If} {\mu }_1\left(t\right)\mathrm{is} {\mathrm{M}}_{i1}, \dots , \mathrm{and} {\mu }_g\mathrm{is} {\mathrm{M}}_{ig},$$

(5)

Then

$$x=A_{i}x\left(t\right)+B_{i}x\left(t\right), i=1, 2, \dots , r$$

where r is the number of fuzzy rules, i is an element in the set of fuzzy rule numbers, g is the number of IF partial variables, j is an element in the set of IF partial variables, ${\mathrm{M}}_{ij} (j=1, 2, \dots , g)$ is a fuzzy set, and ${\mu }_i\left(t\right)$ is the antecedent variable of the fuzzy rule. The global fuzzy model of system (1) can be obtained using single-point fuzzification, product inference, and weighted-average defuzzification.

$$x\left(t\right)={\sum }_{i=1}^r{h}_i\left(\mu \left(t\right)\right)\left(A_{i}x\left(t\right)+B_{i}x\left(t\right)\right) i=1, 2, \dots , r$$

(6)

where

$$\mu \left(t\right)=[{\mu }_1\left(t\right),{\mu }_2\left(t\right),\dots ,{\mu }_g\left(t\right)]$$

$$h_i\left(\mu \left(t\right)\right)={\displaystyle \frac{{\omega }_i\left(\mu \right(t\left)\right)}{{\sum }_{i=1}^N{\omega }_i\left(\mu \right(t\left)\right)}}$$

(7)

$${\omega }_i\left(\mu \left(t\right)\right)={\prod }_{j=1}^g{M}_{ij}\left(\mu \right(t\left)\right)$$

(8)

The normalized fuzzy affiliation function is

$$h_i\left(\mu \left(t\right)\right)\ge 0, and {\sum }_{i=1}^r{h}_i\left(\mu \right(t\left)\right)=1$$

3.

Modeling steps

In this study, the FCM model and the T-S fuzzy model were combined to classify 960 historical data risks into six levels and to predict the potential risk severity levels of 59 typical petrochemical companies in the study area in the following steps:

(i).

Data processing

Step 1: Pre-processing of 962 historical data, including cleaning, normalization, and removal of outliers.

(ii).

FCM processing incident sample classification

Step 2: Set the clustering objective to six and assign the input samples to six different clusters. Each cluster represents a pattern of risk or accident likelihood. Since the fuzzy C-mean can provide affiliation for each data point, it can reflect the fuzziness of an accident scenario belonging to different risk categories.

Step 3: Based on the clustering results, fuzzy affiliation functions are generated for the input variables. These affiliation functions will reflect the affiliation of each accident sample in different clusters. Define the affiliation function of each input factor by clustering center.

(iii).

Construction of T-S fuzzy model and case inference

Step 4: Use the affiliation function obtained from fuzzy C-mean clustering to construct the rule set of the T-S fuzzy model.

Step 5: Using the T-S fuzzy model and the previously generated fuzzy rules, combined with the input data, the model will predict or reason about accident severity. This inference process utilizes the fuzzy assignment in fuzzy C-mean clustering to ensure that the prediction considers the fuzziness in the data.

Step 6: When using the model to make a prediction, the petrochemical risk class and accident likelihood of the companies in the study area are entered, and the T-S model combines the previous rules to produce a new accident severity prediction. In this way, the risk of accidents can be assessed in real time.

4.

Model reliability

To verify the reliability and feasibility of the proposed model, this study focused its validation on fuzzy C-means (FCM) clustering. The rationale is that the membership functions and rule sets of the T–S fuzzy model constructed in this study are directly derived from the FCM cluster centers and membership degree assignments [50]. Therefore, the structural rationality of the clustering is fundamental to this methodological chain. The specific validation principles and process are as follows: We used specialized fuzzy clustering validity metrics, including the silhouette coefficient, Davies–Bouldin index, Xie–Beni index (XB Index, XB), fuzzy partition coefficient (FPC), and partition entropy (PE), to comprehensively assess clustering quality based on compactness, separation, and fuzziness [51], as shown in

Table 5. Model performance test.

Numbers of Clusters	5	6	7
FPC	0.85	0.88	0.86
FPE	0.05	0.04	0.04
XB	0.41	0.28	1.27
Silhouette Coefficient	0.67	0.69	0.70
Davies–Bouldin	0.46	0.46	0.46

. Robustness comparisons were also conducted for different numbers of clusters, as shown in the table. Among them, the silhouette coefficient measures the similarity between the sample and the points in the same cluster and the nearest neighbor cluster, and its value range is [–1, 1]. The higher the value, the better the cluster separation. The Davies–Bouldin index reflects the ratio of intra-class dispersion to inter-class distance. The lower the value, the better the clustering performance. The higher the FPC value, the clearer the classification, and the lower the FPE value, the smaller the clustering uncertainty. The XB index characterizes the clustering quality by calculating the ratio of intra-class compactness to inter-class separation. The lower the value, the better the clustering effect.

The results show that the clustering structure of the model under the six-category classification is the most reasonable and reliable. Therefore, this study ultimately chose to divide the FCM clustering results into six categories and use them to construct the membership function and rule set of the T–S fuzzy model.

3. Results of Analysis

3.1. Distribution Characteristics of Accidents at Various Risk Levels

Within China’s above-scale petrochemical enterprises, private firms account for over 90% of the sector, representing its dominant players. As most private enterprises are SMEs, analyzing characteristics across different risk grades provides critical foundations for subsequent potential risk assessments.

In China, hazardous chemical accidents are classified into five types: explosions, leakages, fires, poisonings, and others. As shown in Figure 3a, explosions (37.4%), fires (30.4%), and leakages (15.7%) are the most common accident types, whereas poisonings and other types occur less frequently. Among these, explosions are the predominant type in chemical enterprise accidents [2,5], closely linked to their inherent physical and chemical properties. Notably, hazardous chemical accidents occur predominantly in private enterprises (Figure 3b). Owing to their investment and operational models, these enterprises often prioritize efficiency and profit objectives [17]. This focus may lead to safety management deficiencies, including weak safety awareness, regulatory violations, ill-defined safety responsibilities, inadequate safety staffing/measures, and poor safety information management. Some private enterprises tend to prioritize productivity over safety. Deficient safety information systems and a lack of standardized procedures further hinder employees’ ability to implement timely and appropriate emergency responses. Simultaneously, SMEs are particularly vulnerable to hazardous chemical accidents (Figure 3c). For example, Chen et al. [2] report that approximately 80% of such accidents occur in SMEs. Given that about 99% of China’s hazardous chemical enterprises are SMEs, enhancing their safety management capabilities is critical [16].

Figure 3d shows that Level VI–V accidents are distributed over a wider period, indicating a lower likelihood of serious petrochemical accidents in veteran enterprises. However, Level II–I accidents are concentrated in newly established enterprises. Following the enterprise life cycle theory, an enterprise can be divided into three main stages: growth, maturity, and aging [52]. Unlike the “bathtub curve” [53], which correlates the age of equipment with its stage of failure, the firm life cycle theory can help explain the relationship between the frequency of petrochemical accidents and the age of an enterprise. There is no precise link between the age of an enterprise and its life cycle. In the initial stage of establishment, the company may exhibit greater flexibility and have new equipment. However, the level of control and expertise may be low, and operators may not be familiar with the equipment. Additionally, the safety management system may not yet be fully developed, leading to an increased risk of mistakes and accidents. As the company matures, it gains more control and stability, and its equipment functions well, resulting in a lower likelihood of petrochemical accidents. However, in the aging phase, companies may experience the same frequency of accidents as in the growth phase, possibly due to financial constraints or aging equipment.

3.2. Spatial Distribution of Petrochemical Risk Levels

Owing to China’s increasing reform and opening up, the country has become a global factory, with rising demand and productivity levels. The petrochemical industry is particularly developed in coastal areas due to their geographic location and transport conditions, making them areas of greater demand for chemical storage, production, and supply. This uneven development has resulted in notable disparities in the frequency and effects of petrochemical accidents.

Figure 4 illustrates the regional differences in petrochemical accidents in China [2,13]. Coastal provinces where petrochemicals serve as pillar industries exhibit higher accident frequencies. For instance, Shandong, Hebei, Zhejiang, and Jiangsu accounted for 54.6% of China’s hazardous chemical accidents (2012–2022), with 930 fatalities (56.57% of national deaths). Risk-level analysis reveals that Shanghai, Zhejiang, Fujian, and Guangdong predominantly experienced lower-grade incidents (Levels V–VI) with less severe effects. This is because the industrial development in these provinces is more advanced, and they have better measures and prevention strategies to deal with petrochemical accidents. Shandong, Hebei, and Jiangsu, which collectively account for the highest number of accidents in the country, also demonstrate distinct characteristics. Shandong and Hebei encompass a greater number of accident levels, spanning from VI to II. This reflects the complex risk landscape and management challenges within their petrochemical sectors. In contrast, Jiangsu has a higher prevalence of VI-level accidents. The only incident with a significant affect level is the I-level accident of the Xiangshui 3–21 Special Major Explosion Accident, which resulted in 78 deaths and profoundly adverse effects. This extreme case exposes critical gaps in preventing systemic risks and responding to worst-case scenarios. It underscores that even with successful low-grade risk control, continuous strengthening of prevention, monitoring, and emergency response capabilities for the highest-level (Level I) risks must remain paramount in safety management systems.

Potential risk assessments revealed a predominance of Level VI (35.6%), Level III (27.1%), and Level IV (23.7%) hazards among typical enterprises, while Level II and Level I risks were less frequent (3.4% each) (Figure 5). This distribution indicates predominantly medium-to-low composite risk levels across China’s coastal petrochemical sector, though high-level threats persist. Spatially, the Shandong Peninsula contained the highest concentration of potential risk zones, followed by the Liaodong Peninsula, with limited distribution in other coastal areas.

3.3. Spatiotemporal Evolution of Petrochemical Risk-Affect Zones

Higher risk levels of petrochemical accidents correspond to larger affect zones and more severe consequences. Figure 6a illustrates the relationship between risk level and affect zone, clearly demonstrating this positive correlation based on the specific spatial extent criteria defined for each risk level in Table 4.

Coastal petrochemical accident affect zones peaked in 2013, 2015, and 2019, subsequently exhibiting declines. As shown in Figure 6b, the affected zones decreased from 352.61 km² in 2019 to 43.67 km² in 2022. A sustained decrease has been observed, particularly since 2019. This trend closely aligns with the implementation of national petrochemical safety policies, most notably the release of the Guidelines on Comprehensively Strengthening Workplace Safety in the Chemical Industry in 2020, which significantly contributed to the decline. By contrast, the declines following the 2013 and 2015 peaks were driven primarily by major accidents occurring in those years; such events typically prompted intensified government safety oversight, leading to a temporary decrease in affect zones.

The proportion of zones affected by petrochemical accidents in coastal cities has exhibited a year-on-year decreasing trend. Starting from a baseline of 37.30% in 2020, this proportion has declined significantly. Analysis combining the previously discussed accident frequency and severity, which shows that coastal accidents mainly involve Level VI-V and Level I incidents, reveals that years witnessing an increase in the affected zone proportion consistently correspond to years when Level I accidents occurred along the coast. Following the implementation of stringent petrochemical safety management and enhanced safety requirements for enterprises in coastal cities, the accident-affected zone decreased markedly. This indicates that, for coastal cities, Level I petrochemical accidents pose the primary obstacle to sustainable socioeconomic development.

The total zone of potential risk affect spans 2986.21 km², posing a significant threat to urban safety and development. This necessitates special attention in planning and management. As indicated in Figure 5, potential high-risk zones are spatially highly concentrated. These zones are primarily located in Caofeidian Industrial Park (Hebei), Gulei (Fujian), Zhangzhou (Fujian), and Zhoushan (Zhejiang). These three provinces collectively account for 36.52% of the total potential risk zone. Shandong represents another highly concentrated region, comprising 32.01% of the medium-risk potential risk zone. Caofeidian and Gulei are designated national petrochemical industry bases, whereas Zhoushan’s development centers around the Port of Zhoushan. All three locations exhibit significant scale and clustering within the petrochemical sector, actively developing industrial symbiosis networks. This highly intensive development model directly results in the dense concentration of hazardous enterprises within limited geographical areas [54].

3.4. Association Between Petrochemical Risk and Land-Use Dynamics

Figure 7 illustrates the spatiotemporal evolution of land use within coastal petrochemical risk affect zones from 2012 to 2022. Figure 7a depicts temporal trends in the proportional coverage of dominant land types. Figure 7b,c contrast spatial distribution patterns of land categories within the risk zones, taking 2012 and 2022 as representative benchmark years. In 2012, agricultural land dominated the landscape at 81.50%, whereas BUA accounted for only 16.36%. Subsequently, the proportion of BUA increased steadily, reaching 54.79% by 2016—a period during which agricultural land declined to 37.25%. Post-2016, BUA stabilized at approximately 40%. The pattern of BUA growth closely corresponds to phases of urban expansion in coastal cities. Before 2015, these cities experienced rapid urban expansion, shifting to more stable growth afterward [55,56]. During the rapid growth phase, edge expansion and leapfrog development dominated urban growth [57,58]. This substantially enlarged urban boundaries, reducing distances between newly developed areas and petrochemical enterprises. This increased spatial proximity resulted in a higher proportion of BUA being affected during petrochemical accidents. After 2015, as urban expansion stabilized, the percentage of BUA affected by such accidents correspondingly ceased to show significant growth.

Over the decade, the region underwent a pronounced land-use transition: In 2012, agricultural land predominated while BUA constituted a minor share, spatially characterized by expansive contiguous farmland and fragmented built development. Throughout this period, agricultural coverage progressively decreased as BUA expanded substantially. Spatially, BUA exhibited rapid patch expansion and enhanced connectivity, progressively encroaching into former agricultural zones [59]. This resulted in highly fragmented and peripherally displaced farmland. Although land covers like woodland, shrubland, grassland, water bodies, and bare land collectively maintained minor proportions, they showed intermittent fluctuations during certain periods. These variations reflect transitional land conversions from ecological to developed uses during urban expansion. Between 2012 and 2022, BUA transitioned from block-like structures to grid-like distributions along roads and pipelines. This reorganization fragmented agricultural land multidirectionally while increasing spatial complexity in bare land and water body patterns. Such spatial restructuring reflects characteristic coastal urbanization and industrial expansion, simultaneously elevating population, assets, and environmental exposure within petrochemical risk zones. Consequently, hazardous material releases now follow expanded diffusion pathways with greater potential effects on farmlands, water bodies, and residential areas [60].

Significant north–south disparities exist in land-use types within potential risk affect zones. Figure 8 visually characterizes the spatial distribution of petrochemical enterprises and their associated risk zones across China’s representative coastal regions under varying land-use patterns. Enlarged insets (a–f) detail the spatial coupling between petrochemical risk zones and land cover types at key sites: Dalian, Caofeidian, Dongying, Qingdao, Zhoushan, and Zhangzhou. High-risk zones consistently concentrate in coastal industrial concentration zones, estuaries, and reclaimed coastal zones. These zones directly adjoin expansive BUA and farmland, with some interfaces extending to wetlands, water bodies, and woodlands. City-specific patterns emerge: Dalian and Qingdao exhibit near-complete spatial overlap between risk zones and coastal industrial districts. By contrast, Dongying and Zhangzhou feature risk zones predominantly embedded within agricultural and forested landscapes, whereas island-type regions like Zhoushan show risk zones encircling port shorelines. Local details reveal contiguous high-risk zones encircling numerous petrochemical enterprises, exhibiting strong spatial alignment with urban expansion axes and infrastructure networks. Overall, petrochemical risk distributions demonstrate high dependency on industrial siting patterns and urban growth trajectories, intricately interwoven with land-use configurations. These risk corridors predominantly follow coastlines and transportation arteries, concentrating pressures from human activity, ecological conservation, and industrial operations.

4. Discussion

There is growing concern in society regarding risk and risk assessment, which may be attributed to an increase in wealth [61]. The Sendai Framework for Disaster Risk Reduction argues that disaster risk management should be conducted preventatively. Although major accidents in the petrochemical industry are infrequent, they can result in serious injuries to workers, property damage, business interruptions, and environmental harm [62]. The early detection of signals or indications preceding an accident can thus assist in the mitigation of risk and the avoidance of economic and social losses. Since 2016, the Chinese government has prioritized lowering the occurrence of major accidents in safety work. China’s industrial structure has been upgraded with new trends in industrial park layouts, large-scale installations, and intelligent production. However, the country also faces the challenge of overlapping new and old risks and the difficulty of preventing and resolving major safety severity [63].

4.1. Drivers of Coastal Petrochemical Risk Dynamics

Analysis of changes in the spatial affect range of petrochemical risks in China reveals a significant dependence on national regulatory policies. Following Level I accidents, enhanced government safety oversight effectively contains hazard spread. Furthermore, the implementation of systematic safety policies, such as the 2020 Opinions on Comprehensively Strengthening Work Safety in the Hazardous Chemicals Sector, has resulted in a persistent decline in the affected zones. This pattern underscores a high reliance on state intervention for petrochemical risk control, concurrently highlighting insufficient autonomous risk management capabilities within enterprises. Notably, accident-prone enterprises are disproportionately concentrated within the privately owned sector [17], which typically exhibits weaker safety management capacity, further evidencing vulnerabilities in corporate-level risk control. Policymakers should prioritize targeted safety training for small- and medium-sized private petrochemical enterprises, make regulatory inspections more frequent and rigorous, and establish closed-loop mechanisms for rectifying identified hazards. These measures would systematically mitigate petrochemical risks arising from non-compliant operations.

4.2. Multi-Level Prevention Strategies for Coastal Potential Petrochemical Risks

Analysis of potential petrochemical risks in coastal cities indicates that provinces with large-scale petrochemical industries, such as Shandong and Liaoning, face a higher likelihood of moderate to high-level (III-IV) petrochemical risks in the future. This poses significant challenges for regional petrochemical enterprise development. Consequently, greater attention should be placed on the development status of petrochemical enterprises in these areas. Specifically, enhanced safety management support is crucial for SMEs within these regions to compensate for their inherent limitations in risk management capacity and prevent severe accidents [16]. Practical efforts should concentrate on strengthening personnel training, building accident early-warning capabilities, and improving safety awareness, particularly within privately owned enterprises. Simultaneously, strategic energy bases and port areas—such as Caofeidian, Zhoushan, and Zhangzhou—require heightened vigilance against high-hazard (I–II level) petrochemical risks that threaten urban safety and development. In these high-risk zones, policymakers should prioritize strengthening urban emergency rescue management systems for production safety. This includes accelerating the establishment of inter-agency emergency information sharing mechanisms and refining multi-departmental systems for coordinated early warning dissemination and incident response. These steps are critical to enhance disaster prevention, mitigation, response capabilities, and incident management proficiency. Furthermore, urban spatial planning must systematically integrate these safety considerations by rigorously avoiding the co-location of sensitive public or critical infrastructure within potential high-risk zones, thereby contributing to sustainable urban development.

4.3. BUA Expansion Drives Changes in Petrochemical Risk Zones

Petrochemical accidents occurring in 2012 significantly affected ecosystems, notably farmland and forests. Subsequent urban expansion [64] has led to a substantial increase in the area of built-up land affected by such incidents. This has resulted in a significant rise in exposed populations and assets, making the socioeconomic consequences of petrochemical accidents increasingly prominent. By 2022, the spatial impact of these accidents expanded beyond predominantly ecological systems to simultaneously affect both socioeconomic systems and the ecological environment.

Petrochemical accidents cause severe and frequently irreversible ecological damage that is difficult to remediate [65]. This includes impacts such as soil contamination from hazardous chemical accumulation, reduced crop yields, impaired forest growth, diminished biodiversity, and loss of ecosystem services. These accidents also trigger multi-dimensional socioeconomic consequences. These encompass toxic gas dispersion threatening public health [11], fire and explosion damage to infrastructure, and oil spills contaminating urban drainage systems [13] and water bodies [66]. Such impacts can further lead to widespread disruption of urban functions and social disorder. The spatial distribution of land-use types within potential risk zones provides critical guidance for urban emergency planning. In regions such as the Shandong and Liaodong Peninsulas, policymakers must prioritize mitigating socioeconomic impacts of petrochemical accidents through enhanced spatial governance. Specifically, urban expansion should avoid encroachment toward petrochemical industrial bases, and mandatory safety buffer zones must be strictly maintained between residential/commercial areas and hazardous enterprises to reduce risk exposure for populations and infrastructure. Conversely, in ecologically sensitive areas—including Zhoushan, Zhangzhou, and the Beibu Gulf—protection efforts should focus on safeguarding farmland, forests, and other vulnerable ecosystems from petrochemical threats through rigorous ecological risk controls.

4.4. A New Paradigm for Petrochemical Risk Assessment from a Land-Use Perspective and Its International Applicability

This study develops a petrochemical risk assessment framework based on dual indicators of hazard and likelihood, overcoming the limitations of traditional hazard-only approaches. By incorporating multi-source data fusion and an accident likelihood indicator, the framework enhances both the completeness and accuracy of risk identification.

When combined with land-use data, the results reveal that between 2012 and 2022, the built-up area within petrochemical risk zones expanded steadily. This pattern not only reflects the dynamic nature of risks under intensive land use, but also helps explain the observed trend of “declining accident frequency but increasing severity per incident” [2] in the petrochemical sector.

Although the case study focuses on China’s coastal provinces, the framework demonstrates strong adaptability, particularly for developing countries where safety management systems remain underdeveloped and petrochemical industry planning is often lacking. By coupling risk zoning with land-use analysis, the framework provides valuable insights for urban spatial planning and risk governance, offering promising potential for international application.

4.5. Limitations and Future Work

This study employs annual CLCD data to capture land-use changes when assessing petrochemical risks and the potential impacts on coastal urban risk zones. The data are mainly used to analyze the area and proportion of different land-use types within these zones. With an overall accuracy of around 80%, CLCD is considered suitable for reflecting land-use patterns and their dynamic evolution in the study area, a view supported by recent accuracy and consistency evaluations [67,68,69]. Nevertheless, uncertainties at the spatial scale still exist in land-use datasets [70,71]. To address this, future research will incorporate multi-source remote sensing and land-use products with varying levels of accuracy for cross-validation, thereby improving data robustness.

5. Conclusions

To mitigate the substantial impact of petrochemical accidents on the economy and society, we must assess the potential risks associated with petrochemical enterprises [2]. This prompts decision makers to develop appropriate regional strategies. The analysis of petrochemical risks encompasses an examination of both past accidents and the potential severity associated with future petrochemical enterprises. This analysis aimed to address the issue of petrochemical safety and provide a foundation for practical policies and behaviors to manage urban safety.

This study employed a fuzzy set-based approach to refine the classification of existing accident risks. This refined classification was then used to analyze the characteristics of relevant enterprises and the evolution of land-use types within the region. Subsequently, the approach facilitated the assessment of potential risks for enterprises in the study area and the visualization of high-risk zones. Analysis revealed a marked regional imbalance in both the occurrence and risk levels of petrochemical accidents across China’s coastal region from 2012 to 2022. Provinces such as Shandong, Hebei, and Jiangsu exhibited a higher frequency of accidents coupled with elevated risk levels, whereas areas such as Zhejiang and Guangdong experienced fewer incidents and lower associated risk levels. Analysis of the association between enterprise characteristics and risk levels reveals that high-risk accidents are predominantly concentrated in SMEs, privately owned firms, and newly established companies. Conversely, large enterprises, state-owned enterprises, and firms that have operated for over 20 years exhibit significantly fewer accidents; furthermore, incidents occurring within these established entities are predominantly classified as low-risk. Regarding land-use change, the primary land types impacted by petrochemical accidents have shifted from an initial predominance of farmland to a pattern in which farmland and construction land are now equally affected. Future research and risk management strategies should prioritize assessing the potential implications of petrochemical risk for both ecological security and socioeconomic security within these evolving landscapes. The spatial distribution of potential petrochemical zones closely aligns with China’s major chemical industry hubs. Notably, Shandong and Liaoning contain the highest concentration of potential risk zones, and Hebei is uniquely identified as the sole high-risk region. These regions represent major petrochemical hubs where rapid industrial expansion has generated significant economic benefits as well as heightened risks. Crucially, the pace of enhancing safety strategies has lagged behind the industry’s accelerated development in these areas.