DOI: 10.3303/CET2290029 Paper Received: 5 February 2022; Revised: 10 March 2022; Accepted: 27 April 2022 Please cite this article as: Men J., Chen G., Yang Y., 2022, A Macro-systematic Accident Propagation Analysis for Preventing Natural Hazard- induced Domino Chain in Chemical Industrial Parks, Chemical Engineering Transactions, 90, 169-174 DOI:10.3303/CET2290029 CHEMICAL ENGINEERING TRANSACTIONS VOL. 90, 2022 A publication of The Italian Association of Chemical Engineering Online at www.cetjournal.it Guest Editors: Aleš Bernatík, Bruno Fabiano Copyright © 2022, AIDIC Servizi S.r.l. ISBN 978-88-95608-88-4; ISSN 2283-9216 A Macro-systematic Accident Propagation Analysis for Preventing Natural Hazard-induced Domino Chain in Chemical Industrial Parks Jinkun Mena, b, Guohua Chena,b*, Yunfeng Yanga, b a Institute of Safety Science & Engineering, South China University of Technology, Guangzhou 510640, Guangdong, China b Guangdong Provincial Science and Technology Collaborative Innovation Center for Work Safety, Guangzhou 510640, Guangdong, China mmghchen@scut.edu.cn Chemical industrial park (CIP) is a typical accident-prone safety-critical system, which is usually congested with high-density hazardous installation units. Domino effects triggered by natural hazards are one of the emerging threats in CIPs, imposing tremendous challenges on society, environment, and economy. This work focuses on analysing the evolution mechanism of natural hazard-induced domino chain (NHDC) from a macro-systematic perspective. Based on the disaster chain theory and the thought of system science, a disaster chain evolution system (DCES) is developed to clarify the accident propagation characteristics. Inspired by the multi-source multi-level propagation pattern of NHDC, the Na-tech layer, the domino layer and the destruction layer are used to form the specific system structure of DCES. A Markov process-based accident propagation (MPAP) model is proposed to cope with the uncertain and complex accident scenarios associated with the evolution process of NHDC. Through formulating the system response process, the proposed MPAP model can reveal the general law of accident evolution. Finally, a general system dynamic response process of DCES is provided to reveal the propagation law of domino effects triggered by natural disaster from a macro-systematic perspective. 1. Introduction Natural hazards such as hurricanes, lightning, earthquakes and floods may rapidly lead to a series of loss of containment (LOC) events in chemical industrial parks (CIPs), causing fires, explosions, or toxic cloud emissions (Reniers et al., 2018). These technological accidents triggered by natural disasters are termed as Na-tech events (Showalter and Myers, 1994). Numerous previous studies (Ricci et al., 2021; Sengul et al., 2012; Young et al., 2004) have shown that there appears to be an increase in frequency and severity of Na- tech events. The typical examples of Na-tech events include: the Great East Japan Earthquake in 2011, caused serious fires and explosions in Sendai and Chiba (Huang et al., 2020); the Wenchuan earthquake in 2008, caused the release of over 100 tons of liquid ammonia in Shifang city (Cruz and Suarez-Paba, 2019); hurricanes “Katrina” and “Rita” in 2005, caused multiple damages to about 611 industrial installations in Gulf of Mexico (Ruckart et al., 2008). (Showalter and Myers, 1994) first coined the term “Na-tech” in 1994. Since then, the increasing catastrophic destruction associated with Na-tech events has quickly raised the awareness of industries, government and academia (Camila et al., 2019). The accident statistics (CHEN and ZOU, 2018) show that the most frequent technological scenarios in CIPs caused by natural disasters include fires and explosions, of which domino effects are easily triggered. However, most of research only focuses on installation failures caused by natural disasters and their secondary technological accidents, and rarely considers the propagation of subsequent domino accidents. The main characteristic of domino effects is the expansion and escalation of accident scenarios, linking a primary scenario with one or several higher level scenarios (Chen et al., 2018). The traditional quantitative domino risk assessment framework (Cozzani et al., 2005) only considers the first propagation level of domino effects. (Chen and Reniers, 2020) have pointed out however that the risk of high- 169 level domino propagation cannot be ignored. The escalation of domino accidents may result in multiple higher order scenarios, which can be seen as parallel effects. Accordingly, the escalation factors associated with multiple failure units may exacerbate the expansion of accident scenarios, and synergistic effects are defined (Reniers and Cozzani, 2013b). (Zhang et al., 2018) proposed an agent-based model to analyze the temporal dependencies of domino chains. (Chen et al., 2018) proposed a domino evolution graph model to capture the spatial-temporal evolution of domino accidents triggered by fire. In their follow-up study (Chen et al., 2021), a dynamic multi-agent approach was proposed to analyse the evolution of cascading technological accidents. (Huang et al., 2021) developed a dynamic analysis for domino chains under fire scenarios. This paper aims to reveal the propagation law of domino effects triggered by natural disaster from a macro- systematic perspective. Specifically, a disaster chain evolution system (DCES) is developed to clarify the accident propagation characteristics. The various features of DCES designs are given, including the system units, the system states, the system activation conditions, the system structure. A Markov process-based accident propagation (MPAP) model is proposed to formulate system response behaviors during the evolution process of NHDC. Through analyzing the evolution mechanism of NHDC, a conceptual loss prevention framework is established to guide the prevention and the mitigation of the domino effects triggered by natural hazards. 2. Disaster Chain Evolution System The general system response process of DCES is demonstrated in Figure 1. To analyze the propagation characteristics of domino effects triggered by natural hazards, a DCES is proposed in this section. The various features of DCES designs are stated in next sub-sections, including the system units in Section 2.1, the system activation conditions in Section 2.2, the system structure in Section 2.3, the system response process in Section 2.4. Figure 1: The general system dynamic response process 2.1 System Units The proposed DCES consists of hazard unit set (𝐻𝐻), environment unit set (𝐸𝐸) and vulnerable unit set (𝑉𝑉), i.e.: 𝐷𝐷𝐷𝐷𝐸𝐸𝐷𝐷 =< 𝐻𝐻, 𝑉𝑉, 𝐸𝐸 > (1) where 𝐻𝐻 refers to the hazard units that can cause adverse effects to vulnerable units. In CIPs, 𝐻𝐻 can be divided into natural hazards and technological hazards. Vulnerable units refer to the objectives affected and damaged by hazards, which mainly includes various HIUs. Environment unit is closely related to the derivation of hazards, which refer to the relationship among natural environment, human environment, and industrial environment, such as meteorological conditions, personnel distribution, management factors, and land-use layout. 2.2 System Activation Conditions The system activation conditions of DCES is equivalent to the occurrence conditions of NHDC. Based on the accident-causing theory (Chi and Han, 2013), the occurrence and development process of industrial accidents affected by natural disasters are analyzed, and the system activation conditions are stated as follows:  Natural hazards are regarded as the primary disasters causing failures of vulnerable units and generating secondary technological hazards;  Secondary technological hazards cause adverse effects to new vulnerable units;  The failure energy of hazard units exceeds the failure threshold of vulnerable units. 170 2.3 System Structure Usually, domino chains are propagated level by level (Reniers and Cozzani, 2013a). The propagation pattern of domino effects triggered by natural hazards is regarded as a multi-source multi-level parallel disaster chain. According to the propagation characteristics, a three-layer system structure is developed to identify the triggering relations among three system units, which mainly includes the Na-tech layer, the domino layer and the destruction layer. The following Figure 2 is adopted to demonstrate the structure of proposed system. In the Na-tech layer, hazard units are mainly composed of extreme natural phenomena caused by various natural disasters, which may trigger multiple failures of HIUs (Yang et al., 2020). In the domino layer, hazard units are converted into the technological hazards generated by the failures of HIUs. When the failure energy exceeds the failure threshold of the current vulnerable unit, domino accident may be triggered. In the destruction layer, the uncontrolled energy generated by various hazard units can spread to vulnerable units outside the system, causing serious accident consequences. Figure 2: Three-layer system structure of the disaster chain evolution system 3. Markov process-based accident propagation The core of the NHDC is the expansion and escalation of accident scenarios (Reniers and Cozzani, 2013a). The potential simultaneous damages caused by natural hazards lead to the complex and uncertain evolution process of NHDC. To facilitate the analysis of accident evolution mechanism, a reasonable assumption is established, that is, the accident scenario of the next accident expansion level is only related to the current accident scenario. Thus, the evolution process of NHDCs is regarded as a Markov process (Wang et al., 2020). In this work, the changes of system units during the system response process are used to describe the evolution of NHDC. The proposed MPAP model is stated as follows: 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 =< 𝑺𝑺, 𝑷𝑷 > (2) where 𝐷𝐷 is the state space of system units; 𝑀𝑀 is the transition probability space. The formulation of MPAP are stated in the following sub-sections. 3.1 State Space In CIPs, domino effects are mainly propagated among the HIUs (Zhang et al., 2019). According to the potential technological hazards associated with HIUs, five states of HIUs are shown in Table 1. Suppose that 𝑉𝑉 = {𝑣𝑣𝑖𝑖|𝑖𝑖 = 1,2, … |𝑉𝑉|} is a vulnerable set containing |𝑉𝑉| HIUs. The state of system units for kth-level accident scenario can be expressed by a state matrix 𝑺𝑺𝒌𝒌 = (s𝑘𝑘1, s𝑘𝑘2, … , s𝑘𝑘|𝑉𝑉|). s𝑖𝑖 = ⎩ ⎪ ⎨ ⎪ ⎧ 0, 𝑡𝑡ℎ𝑒𝑒 𝑠𝑠𝑡𝑡𝑠𝑠𝑡𝑡𝑒𝑒 𝑜𝑜𝑜𝑜 𝑣𝑣𝑖𝑖 𝑖𝑖𝑠𝑠 "𝑜𝑜𝑜𝑜𝑒𝑒𝑜𝑜𝑠𝑠𝑡𝑡𝑖𝑖𝑜𝑜𝑜𝑜𝑠𝑠𝑜𝑜" 1, 𝑡𝑡ℎ𝑒𝑒 𝑠𝑠𝑡𝑡𝑠𝑠𝑡𝑡𝑒𝑒 𝑜𝑜𝑜𝑜 𝑣𝑣𝑖𝑖 𝑖𝑖𝑠𝑠 "𝑅𝑅𝑒𝑒𝑜𝑜𝑒𝑒𝑠𝑠𝑠𝑠𝑒𝑒" 2, 𝑡𝑡ℎ𝑒𝑒 𝑠𝑠𝑡𝑡𝑠𝑠𝑡𝑡𝑒𝑒 𝑜𝑜𝑜𝑜 𝑣𝑣𝑖𝑖 𝑖𝑖𝑠𝑠 "𝐹𝐹𝑖𝑖𝑜𝑜𝑒𝑒" 3, 𝑡𝑡ℎ𝑒𝑒 𝑠𝑠𝑡𝑡𝑠𝑠𝑡𝑡𝑒𝑒 𝑜𝑜𝑜𝑜 𝑣𝑣𝑖𝑖 𝑖𝑖𝑠𝑠 "𝐸𝐸𝐸𝐸𝑜𝑜𝑜𝑜𝑜𝑜𝑠𝑠𝑖𝑖𝑜𝑜𝑜𝑜" 4, 𝑡𝑡ℎ𝑒𝑒 𝑠𝑠𝑡𝑡𝑠𝑠𝑡𝑡𝑒𝑒 𝑜𝑜𝑜𝑜 𝑣𝑣𝑖𝑖 𝑖𝑖𝑠𝑠 "𝐸𝐸𝐸𝐸𝑡𝑡𝑖𝑖𝑜𝑜𝐸𝐸𝐸𝐸𝑖𝑖𝑠𝑠ℎ𝑒𝑒𝑒𝑒" , 𝑖𝑖 = 1,2, . . , |𝑉𝑉| (3) Table 1: Five states of HIUs State Type State Description Operational The HIU is not failed. Release The HIU is physically damaged, resulting in the release of hazardous materials. Fire The HIU is on fire, causing heat radiation Explosion The LOC event of the HIU induces an explosion, causing heat radiation, causing shock wave overpressure and propellant fragments. Extinguished The HIU is failed but does not produce any technological hazards. 171 3.2 Transition Probability Space: Na-tech layer In the Na-tech layer, vulnerable units suffer from negative effects imposed by natural hazards. Suppose that 𝑉𝑉 = {𝑣𝑣𝑖𝑖|𝑖𝑖 = 1,2, … |𝑉𝑉|} is a vulnerable unit set containing |𝑉𝑉| HIUs, 𝒩𝒩 is a natural hazard unit 𝒩𝒩. The failure probability of HIU 𝑣𝑣𝑖𝑖 under the influence of natural hazard 𝒩𝒩 is given as follows: 𝑀𝑀𝐹𝐹1(𝑣𝑣𝑖𝑖) = 𝑜𝑜𝒩𝒩𝑀𝑀(𝑣𝑣𝑖𝑖|𝒩𝒩) (4) where 𝑜𝑜𝒩𝒩 is the frequency of the occurrence of 𝒩𝒩 , 𝑀𝑀(𝑣𝑣𝑖𝑖|𝒩𝒩) is the failure probability derived from the vulnerability assessment model. Generally, the vulnerability assessment model (Yang et al., 2020) is determined by comparing the relationship between the intensity of natural hazards and the resistance of HIUs, which can be expressed as follows: 𝑀𝑀(𝑣𝑣𝑖𝑖|𝒩𝒩) = 𝜙𝜙(𝐼𝐼(𝒩𝒩) > 𝑅𝑅(𝑣𝑣𝑖𝑖)) (5) where 𝜙𝜙(∙) is the mapping relation of the vulnerability model; 𝐼𝐼(𝒩𝒩) is the intensity of natural hazard unit 𝒩𝒩; 𝑅𝑅(𝑣𝑣𝑖𝑖) is the resistance of HIU 𝑣𝑣𝑖𝑖. The probabilities of 𝑣𝑣𝑖𝑖 being in five predefined states in Na-tech layer is given as follows: 𝑀𝑀𝑁𝑁𝑁𝑁�𝑺𝑺𝟏𝟏(𝒊𝒊)� = ⎩ ⎪ ⎨ ⎪ ⎧ 1 − 𝑀𝑀𝐹𝐹1(𝑣𝑣𝑖𝑖), 𝑺𝑺𝟏𝟏(𝒊𝒊) = 𝟎𝟎 𝑀𝑀𝐹𝐹1(𝑣𝑣𝑖𝑖)𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝑅𝑅), 𝑺𝑺𝟏𝟏(𝒊𝒊) = 𝟏𝟏 𝑀𝑀𝐹𝐹1(𝑣𝑣𝑖𝑖)𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐹𝐹), 𝑺𝑺𝟏𝟏(𝒊𝒊) = 𝟐𝟐 𝑀𝑀𝐹𝐹1(𝑣𝑣𝑖𝑖)𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐸𝐸), 𝑺𝑺𝟏𝟏(𝒊𝒊) = 𝟑𝟑 0, 𝑺𝑺𝟏𝟏(𝒊𝒊) = 𝟒𝟒 (6) where 𝑀𝑀𝑁𝑁𝑁𝑁�𝑺𝑺𝟏𝟏(𝑖𝑖)� gives the probabilities of 𝑣𝑣𝑖𝑖 being in five predefined states in Na-tech layer; 𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝑅𝑅) , 𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐹𝐹) and 𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐸𝐸) are the probabilities of three accident scenarios (release, fire and explosion) after installation failure. In practical engineering application, 𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝑅𝑅), 𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐹𝐹) and 𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐸𝐸) can be obtained by the event tree analysis (Vilchez et al., 2011). According to the system activation conditions, it is assumed that the initial state of the system is safe, i.e.: 𝑺𝑺𝟎𝟎 = 𝑶𝑶𝟏𝟏×|𝑽𝑽| (7) where all the entries in the matrix 𝑺𝑺𝟎𝟎 are 0. To sum up, the transition probability of the Na-tech layer can be formulated as follows: 𝑀𝑀(𝑺𝑺𝟎𝟎, 𝑺𝑺𝟏𝟏) = �𝑀𝑀𝑁𝑁𝑁𝑁�𝑺𝑺𝟏𝟏(𝑖𝑖)� |𝑉𝑉| 𝑖𝑖 (8) 3.3 Transition Probability Space: Domino layer The technological hazards that can trigger the domino effects are mainly thermal radiation and shock wave overpressure generated by fires and explosions. The domino extension probability can be obtained by the classical Probit model (Cozzani et al., 2005). In the Domino layer (𝑘𝑘 ≥ 2), for ∀ 𝑣𝑣𝑖𝑖 ∈ 𝑉𝑉𝑘𝑘, its failure probability 𝑀𝑀𝐷𝐷(𝑣𝑣𝑖𝑖) can be calculated as follows: 𝑀𝑀𝐷𝐷(𝑣𝑣𝑖𝑖) = 1 √2𝜋𝜋 � 𝑒𝑒− 𝑥𝑥2 2 𝑒𝑒𝐸𝐸 𝑌𝑌−5 −∞ (9) where 𝑌𝑌 is the probit variable, the probit variables of common installation are provided by (Cozzani et al., 2005). Similarly, the probabilities of 𝑣𝑣𝑖𝑖 being in five predefined states in next level accident scenario can be calculated as follows: 𝑀𝑀𝐷𝐷𝐷𝐷�𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖)� = ⎩ ⎪ ⎨ ⎪ ⎧1 − 𝑀𝑀𝐹𝐹 𝑘𝑘�𝑣𝑣𝑗𝑗�, 𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖) = 0 𝑀𝑀𝐹𝐹𝑘𝑘�𝑣𝑣𝑗𝑗�𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝑅𝑅), 𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖) = 1 𝑀𝑀𝐹𝐹𝑘𝑘�𝑣𝑣𝑗𝑗�𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐹𝐹), 𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖) = 2 𝑀𝑀𝐹𝐹𝑘𝑘�𝑣𝑣𝑗𝑗�𝑀𝑀𝑇𝑇(𝑣𝑣𝑖𝑖|𝐸𝐸), 𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖) = 3 0, 𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖) = 4 , 𝑘𝑘 ≥ 1 (13) where 𝑀𝑀𝐹𝐹𝑘𝑘�𝑣𝑣𝑗𝑗� is the failure probability of HIU 𝑣𝑣𝑗𝑗 in kth-level accident scenario. The transition probability of the Domino layer (𝑘𝑘 ≥ 1) can be formulated as follows: 𝑀𝑀(𝑺𝑺𝒌𝒌, 𝑺𝑺𝒌𝒌+𝟏𝟏) = � 𝑀𝑀𝐷𝐷𝐷𝐷�𝑺𝑺𝒌𝒌+𝟏𝟏(𝑖𝑖)� 𝑣𝑣𝑖𝑖∈𝑉𝑉𝑘𝑘 (14) It is worth mentioning that the shock wave overpressure caused by the explosion is a kind of instantaneous damage. For the unit in the explosion state, it will be directly converted to the extinguished state in the next level of accident scenario, and it will no longer participate in the subsequent accident evolution process. 172 4. Methodology illustration The following illustration shown in Figure 3 is adopted to demonstrate the transition probability space of the domino layer. Suppose that 𝑉𝑉 = {𝑣𝑣1, 𝑣𝑣2, 𝑣𝑣3, 𝑣𝑣4} is a vulnerable unit set containing four HIUs. All HIUs are pressure vessels. 𝐹𝐹𝑘𝑘 and 𝐸𝐸𝑘𝑘 are used to express the fire-related hazard units and the explosion-related hazard units corresponding to 𝑺𝑺𝒌𝒌. For the primary accident scenario, its system units are stated as follows: ⎩ ⎨ ⎧ 𝑺𝑺𝟏𝟏 = (𝟐𝟐 𝟎𝟎 𝟎𝟎 𝟎𝟎) 𝑉𝑉1 = {𝑣𝑣2, 𝑣𝑣3, 𝑣𝑣4} 𝐹𝐹1 = {𝑣𝑣1} 𝐸𝐸1 = ∅ (15) The fire domino effect propagation probability 𝑀𝑀𝑓𝑓𝑖𝑖 1 of 𝑣𝑣2 can be obtained as follows: ⎩ ⎪ ⎨ ⎪ ⎧ 𝑀𝑀𝑓𝑓𝑖𝑖 1 (𝑣𝑣2) = 1 √2𝜋𝜋 � 𝑒𝑒− 𝑥𝑥2 2 𝑒𝑒𝐸𝐸 𝑌𝑌(𝑣𝑣2)−5 −∞ 𝑌𝑌(𝑣𝑣2) = 16.82 − 1.847𝐷𝐷2 (𝑡𝑡𝑜𝑜𝑜𝑜) 𝐷𝐷2(𝑡𝑡𝑜𝑜𝑜𝑜) = −0.97 𝑜𝑜𝑜𝑜�𝜇𝜇𝑓𝑓 1(𝑣𝑣2)� − 8.835𝑉𝑉2 0.032 (16) where 𝜇𝜇𝑓𝑓 1(𝑣𝑣2) = 𝑜𝑜12; 𝑉𝑉2 is the volume of HIU 𝑣𝑣2. Since the primary accident scenario contains only one fire accident, the failure probability of HIU 𝑣𝑣2 in primary accident scenario is equivalent to 𝑀𝑀𝑓𝑓𝑖𝑖 1 (𝑣𝑣2) , 𝑀𝑀𝐹𝐹1(𝑣𝑣2) = 𝑀𝑀𝑓𝑓𝑖𝑖 1 (𝑣𝑣2). The failure probabilities of HIUs 𝑣𝑣3 and 𝑣𝑣4 in primary accident scenario can be obtained in the same way. Since the vulnerable unit set 𝑉𝑉4 = ∅ , the termination state condition is satisfied. Thus, 𝑺𝑺𝟒𝟒 is the termination state of the above NHDC. The transition probabilities are given as follows: 𝑀𝑀(𝑺𝑺𝟏𝟏, 𝑺𝑺𝟐𝟐) = �1 − 𝑀𝑀𝐹𝐹1(𝑣𝑣2)�𝑀𝑀𝐹𝐹1(𝑣𝑣3)𝑀𝑀𝑇𝑇(𝑣𝑣3|𝐸𝐸)�1 − 𝑀𝑀𝐹𝐹1(𝑣𝑣4)� (17) 𝑀𝑀(𝑺𝑺𝟐𝟐, 𝑺𝑺𝟑𝟑) = 𝑀𝑀𝐹𝐹2(𝑣𝑣2)𝑀𝑀𝑇𝑇(𝑣𝑣2|𝐹𝐹)�1 − 𝑀𝑀𝐹𝐹2(𝑣𝑣4)� (18) 𝑀𝑀(𝑺𝑺𝟑𝟑, 𝑺𝑺𝟒𝟒) = 𝑀𝑀𝐹𝐹3(𝑣𝑣4)𝑀𝑀𝑇𝑇(𝑣𝑣4|𝐹𝐹) (19) Figure 3: An illustration of state transition 5. Conclusions Natural hazards and their secondary fires and explosions may damage HIUs in chemical industrial parks, resulting in domino effects and enlarging the consequences of accidents. This work focuses on the multi- source multi-level domino accident chain induced by typical natural disasters such as lightning, floods, and earthquakes in chemical industrial parks. A disaster chain evolution system is developed to clarify the accident propagation characteristics. A Markov process-based accident propagation is proposed to formulate the system response behaviour during the evolution process of natural hazard-induced domino chain. The proposed MPAP model can quantify the possibility and uncertainty associated of the expansion of the accident scenario through the state transition probability, identify the most likely propagation chain and the most dangerous vulnerable unit. Natural hazards can produce LOC events in HIUs that store hazardous materials causing fires, explosions, or toxic cloud emissions. The technological hazards such as heat radiation, shock waves, and propellant fragments generated by fires and explosions can easily cause damage to the adjacent HIUs, triggering the domino chain. With the expansion and escalation of accident scenario, the hazard units and the vulnerable units are updated level by level. Acknowledgments This study was supported by the National Natural Science Foundation of China (22078109) and the Key-Area Research and Development Program of Guangdong Province (2019B111102001). 173 References Camila, S.-P.M., Mathis, P., Felipe, M., Maria, C.A., 2019. Systematic literature review and qualitative meta- analysis of Natech research in the past four decades. Safety Science 116, 58–77. Chen, C., Reniers, G., 2020. Chemical industry in China: The current status, safety problems, and pathways for future sustainable development. Safety Science 128, 104741. Chen, C., Reniers, G., Khakzad, N., 2021. A dynamic multi-agent approach for modeling the evolution of multi- hazard accident scenarios in chemical plants. Reliability Engineering and System Safety 207, 107349. Chen, C., Reniers, G., Zhang, L., 2018. An innovative methodology for quickly modeling the spatial-temporal evolution of domino accidents triggered by fire. Journal of Loss Prevention in the Process Industries 54, 312–324. CHEN, G., ZOU, M., 2018. Coupling relationship model of multi-hazard and pattern of chain-cutting disaster mitigation in Chemical Industry Park. Chemical Industry and Engineering Progress 37, 3271-3279. Chi, S., Han, S., 2013. Analyses of systems theory for construction accident prevention with specific reference to OSHA accident reports. International Journal of Project Management 31, 1027-1041. Cozzani, V., Gubinelli, G., Antonioni, G., Spadoni, G., Zanelli, S., 2005. The assessment of risk caused by domino effect in quantitative area risk analysis. Journal of Hazardous Materials 127, 14-30. Cruz, A.M., Suarez-Paba, M.C., 2019. Advances in Natech research: An overview. Progress in Disaster Science 1, 100013. Huang, K., Chen, G., Khan, F., Yang, Y., 2021. Dynamic analysis for fire-induced domino effects in chemical process industries. Process Safety and Environmental Protection 148, 686–697. Huang, K., Chen, G., Yang, Y., Chen, P., 2020. An innovative quantitative analysis methodology for Natech events triggered by earthquakes in chemical tank farms. Safety Science 128, 104744. Krausmann, E., Cruz, A.M., Salzano, E., 2017. Reducing the Risk of Natural-Hazard Impact on Hazardous Installations. Elsevier, Netherlands. Laurent, A., Pey, A., Gurtel, P., Fabiano, B., 2021. A critical perspective on the implementation of the EU Council Seveso Directives in France, Germany, Italy and Spain. Process Safety and Environmental Protection 148, 47-74. Reniers, G., Cozzani, V., 2013a. Domino Effects in the Process Industries Modelling, Prevention and Managing. Elsevier, Netherlands. Reniers, G., Cozzani, V., 2013b. Domino Effects in the Process Industries, Modeling, Prevention and Managing. Elsevier, Amsterdam, The Netherlands. Reniers, G., Khakzad, N., Cozzani, V., Khan, F., 2018. The impact of nature on chemical industrial facilities: Dealing with challenges for creating resilient chemical industrial parks. Journal of Loss Prevention in the Process Industries 56, 378-385. Ricci, F., Moreno, V.C., Cozzani, V., 2021. A comprehensive analysis of the occurrence of Natech events in the process industry. Process Safety and Environmental Protection 147, 703-713. Ruckart, P.Z., Orr, M.F., Lanier, K., Koehler, A., 2008. Hazardous substances releases associated with Hurricanes Katrina and Rita in industrial settings, Louisiana and Texas. Journal of Hazardous Materials 159, 53-57. Sengul, H., Santella, N., Steinberg, L.J., Cruz, A.M., 2012. Analysis of hazardous material releases due to natural hazards in the United States. Disasters 36, 723-743. Showalter, P.S., Myers, M.F., 1994. Natural Disasters in the United States as Release Agents of Oil, Chemicals, or Radiological Materials Between 1980‐1989: Analysis and Recommendations. Risk Analysis 14, 69-182. Vilchez, J.A., Espejo, V., Casal, J., 2011. Generic event trees and probabilities for the release of different types of hazardous materials. Journal of Loss Prevention in the Process Industries 24, 281-287. Wang, J., Wang, Y., Yan, H., Cao, J., Shen, H., 2020. Hybrid Event-Based Leader-Following Consensus of Nonlinear Multiagent Systems With Semi-Markov Jump Parameters. IEEE Systems Journal Early Access, 1-12. Yang, Y., Chen, G., Genserik, R., 2020. Vulnerability assessment of atmospheric storage tanks to floods based on logistic regression. Reliab. Eng. Syst. Saf. 196, 106721. Young, S., Balluz, L., Malilay, J., 2004. Natural and technologic hazardous material releases during and after natural disasters: a review. Sci. Total Environ. 322, 3-20. Zhang, L., Landucci, G., Reniers, G., Khakzad, N., Zhou, J., 2018. DAMS: A Model to Assess Domino Effects by Using Agent-Based Modeling and Simulation. Risk Analysis 38, 1585-1600. Zhang, M., Zheng, F., Chen, F., Pan, W., Mo, S., 2019. Propagation probability of domino effect based on analysis of accident chain in storage tank area. Journal of Loss Prevention in the Process Industries 62, 103962. 174 lp-2022-abstract-096.pdf A Macro-systematic Accident Propagation Analysis for Preventing Natural Hazard-induced Domino Chain in Chemical Industrial Parks