Earthquake disasters rank as the most destructive natural disasters in the world (Albulescu, 2023). With the rapid development of urbanization, the growing exposure of populations and structures aggravates the seismic risk in urban areas (Zou et al., 2023). Such unpromising circumstances necessitate the optimization of earthquake disaster mitigation. The foundation of an optimal formulation lies in knowledge about the location, capacity, and spatial accessibility of critical emergency facilities (Albulescu, 2023). Emergency shelters are one of the most essential emergency facilities, providing temporary resettlement, medical services, and emergency command centers for residents. The spatial accessibility of emergency shelters depends on the complexity of evacuation in an emergency and reflects the effectiveness of mitigation strategies and emergency management. The SAA of emergency shelters can also assist in formulating strategies to optimize disaster mitigation for high seismic risk areas.

Once an earthquake occurs, taking shelter is the very first instinct of residents, which demands mobility and accessibility (Zou et al., 2023). Exploring the spatial distribution of emergency shelters is an effective approach for identifying their accessibility to refugees (Gall, 2004; Chang and Liao, 2015). A map of spatial accessibility distribution could visualize the ease of reaching emergency shelters and evaluate the rationality of their allocation.

The methods used for the SAA of urban facilities mainly encompass typical spatial analysis methods such as buffer analysis (Zhang, 2020) and network analysis (Hou and Jiang, 2014; Albulescu, 2023; Yan, 2023). Methods are constantly developing, with advanced methods combined with GIS technology. The gravity model, which considers the relationship between supply- and demand-sides, can be an essential companion to GIS spatial analysis technology to distinguish the radiation coverage of different-sized areas (Zhang, 2020). As a special form of the gravity model, the two-step floating catchment area (2SFCA) method can assist in the study of urban emergency facilities. For example, the Gaussian two-step mobile search method, which is an improved 2SFCA method, introduces a Gaussian function with an excellent fit for distance decay. This function can help obtain a more accurate accessibility indicator (Tong et al., 2021; Jiang et al., 2023). In addition, a 2SFCA method with a variable service radius and evacuation radius was proposed to deal with the differences in emergency shelter service capacities and population distribution (Su, Chen, and Cheng, 2022a). Moreover, typical map applications (e.g., Google Map and Baidu Map) also stimulate accessibility evaluation. For instance, the application programming interface (API) can provide automatic time cost calculations for real-time travel (Wang et al., 2019) and navigation data (Shen et al., 2020). As the accuracy of the accessibility index constantly improves, the quantitative assessment of spatial accessibility continues to evolve.

However, qualitative analysis involving particular scenarios has not been much emphasized, although multiple scenarios can affect the accessibility of emergency shelters. For example, demographic density, evacuation capacity across age groups, population spatial temporal distribution, and even topographic factors can lead to differences in evacuation time (Su, Chen, and Zhang, 2022b; Zou et al., 2023). Moreover, the accessibility of emergency shelters in disaster scenarios is of great concern (Zhang and Yun, 2019). The accessibility might be undermined at different levels due to severe damage. Earthquake damage can be distinguished into direct damage (e.g., shaking or failure of over-bridges and roads) and indirect damage (e.g., street blockages due to debris from collapsed buildings) (Anastassiadis and Argyroudis, 2007). Strong ground motion and fault rupturing can cause catastrophic damage to buildings and road surfaces. During the 2008 M8 Wenchuan earthquake, the buildings along the fault-ruptured area were all significantly destroyed because of sudden fault rupturing (Xu et al., 2009). Moreover, open space damage caused by shaking and collapsed buildings can indirectly impede evacuation and rescue in the immediate aftermath (Francini et al., 2020; Bernardini and Ferreira, 2022). Dramatic destruction due to fault rupturing and strong ground motion can significantly impede emergency evacuations. In such situations, accessibility to emergency shelters would be impaired. To better assist in emergency management, we must explore the reasonable SAA of emergency shelters for earthquake scenarios.

In this study, we developed an earthquake scenario-specific framework for SAA. This framework integrated the service capacity of urban facilities with the characteristics of earthquake scenarios. Except for the spatial distribution of emergency shelters, we were able to gain an insight into the fault rupturing and strong ground motion caused by earthquake disasters. First, we operated the spatial service area of emergency shelters through network analysis. Based on the service area distribution, we generated the impact of strong ground motion and fault rupturing on the level of accessibility. Finally, we generated a comprehensive distribution map of the level of accessibility to emergency shelters in earthquake scenarios.

We employed this framework in the urban area of Xichang City in western China. After analyzing the impact of strong ground motion and fault rupturing characteristics, we determined the changes in the accessibility level in several towns of Xichang City. We also identified the practical service area of emergency shelters in earthquake scenarios. Our results are reasonable as we considered the potential effects caused by earthquake disaster characteristics.

In the subsequent sections, we introduce the historical earthquake records and the extent of the study area in Section 2. Then, in Section 3, we illustrate the framework combined with the case study in Xichang City. We display the results in Section 4 and discuss them in Section 5. The last section draws conclusions for this study.

Xichang City, located in southwestern Sichuan Province, China, is the capital of the Liangshan Yi Autonomous Prefecture (Figure 1). Xichang City's population has increased from 700,000 to 966,000 in the past 10 years. Infrastructure construction there has expanded ∼ four times, with an urbanization rate of 68.48% (http://www.xichang.gov.cn/zfxxgk/zfxxgknr/tjsj_31356/202305/t20230526_2487868.html, last accessed in January 2024). The increased population and expanded inhabited structures can raise issues of natural disaster prevention, especially in high seismic risk zones. To cope with emergency evacuation issues in earthquake-prone areas, the construction of a reasonable SAA framework for emergency shelters is particularly important.

Xichang City is in an area with frequent earthquake occurrences. Due to the collision between the Indian and Eurasian plates, the eastern Tibetan plateau exhibits active crustal extrusion at a rate of approximately 10 mm/yr, resulting in frequent historical earthquakes. This pattern is particularly evident on large-scale boundary faults such as the Xianshuihe fault, the Anninghe fault, the Zemuhe fault and the Xiaojiang fault. Xichang City is located at the link of the Anninghe fault and the Zemuhe fault (Figure 2), both with high left-lateral strike–slip rates of ∼6.5 mm/yr (Ren et al., 1990; Xu et al., 2003; Shen et al., 2005; Cheng et al., 2012).

Historically, Xichang City has suffered significantly from devastating earthquakes—the 814 M7 earthquake, the 1489 M6.8 earthquake, the 1536 M7.5 earthquake, the 1732 M6.8 earthquake, and the 1850 M7.8 earthquake (Ren et al., 1993; Wen, 2000; Jia Cheng et al., 2021; Jia Cheng, 2023). These historical earthquakes and their damage (Liu, 1989; Zhang et al., 1998; Min et al., 2005) are illustrated in Supplementary Table S1.

In the 1489 M6.8 earthquake, Xichang City was near the assumed VIII or VIII+ intensity zone. The seismic intensity for cities situated about 100 km from Xichang (e.g., Mianning and Yuexi) also reached VIII or VIII+. Cities located approximately 600 km–700 km from Xichang (e.g., Suining, Yingshan, and Lingshui) are near the IV-intensity zone (Supplementary Figure S1). In the 1536 M7.5 earthquake, Xichang City was in the Ⅸ intensity zone. Cities situated approximately 400 km from Xichang (e.g., Dayi, Chengdu, Chongqing, and Ziyang) were near the assumed Ⅵ or Ⅵ+ intensity zone (Supplementary Figure S2A). In the 1732 M6.8 earthquake, Xichang City was in the Ⅸ intensity zone (Supplementary Figure S2B). In the 1850 M7.8 earthquake, Xichang City was in the region where seismic intensity is assumed to be Ⅹ (Supplementary Figure S2C).

Xichang City is still in a high-risk zone for earthquakes. Recent studies have proposed a sizable seismic gap in the Anninghe–Zemuhe–Daliangshan region, both from the lengthy intervals between historical earthquakes and the scarcity of M_L 2.5 earthquakes since 1980 (Yi et al., 2004; Wen et al., 2008; Yi, Wen, and Su, 2008). The expansion of the population and the number of high buildings make the evacuation of residents difficult in large earthquake events. Hence, we applied our SAA methodology to emergency shelters in Xichang City.

Prompt evacuation and rescue are crucial for preventing fatalities and severe injuries in residential buildings in an emergency (Chen et al., 2023). Increasingly large populations in urban areas also increase exposure to natural disasters. This situation should be emphasized in disaster mitigation research (Huang et al., 2019; Elliott, 2020; Iglesias et al., 2021; Albulescu, 2023). In this case, we employed SAA on emergency shelters in urban Xichang City, which is in a high seismic risk area. Using its population density data and remote sensing images, we extracted an area with an obviously high population density in Xichang City. Here, we analyzed raster data from Grid Population of the World Version 4 (GPW v4) in 2020 (https://sedac.ciesin. Columbia. edu/data/collection/gpw_v4, last accessed November 2023) and divided them into different categories according to the attribute values of each raster. Finally, we clipped out the areas with significant population density as the study area (Figure 3).

Here, we develop an earthquake scenario-specific SAA framework for cities with high seismic risks, such as Xichang City. Based on quantitative analysis with the network analysis method, we focus on the characteristics of earthquake scenarios (e.g., strong ground motion and fault rupturing). The procedure in this framework contains three main steps. First, we analyzed the accessibility of emergency shelters based on road network data using network analysis. Next, we assessed the impact of strong ground motion on accessibility. Finally, we calculated the impact of fault rupturing on accessibility and generated a comprehensive accessibility distribution map. We illustrate the workflow in Figure 4.

In this section, we present the result of each procedure (Figure 4).

For data collection, we selected 59 emergency shelter entrances in the urban area of Xichang City (Table 4). Table 4 shows that Xiaomiao Town, Xijiao Town, and Beicheng Block are the top three regarding emergency shelter quantity. Regarding density, emergency shelters are the most densely distributed in Beicheng, Changning, and Changan Blocks. Figure 7A illustrates the spatial distribution of selected emergency shelters and road networks. Our results show that Xichang City has more emergency shelters in the south but fewer in the north (Figure 7A). Emergency shelter distribution is consistent with the spatial distribution of population density. However, on both sides of the Dong River (the area linking Xicheng Block and Dongcheng Block), the number of emergency shelters is not proportional to the large population there. Lizhou and Xisheng Towns, with sparse populations and undeveloped road networks, have no proper emergency shelters according to the remote sensing imagery.

Figure 7B shows the road network-based accessibility distribution map. In this figure, we found that, in areas with dense emergency shelters and well-developed road networks, residents had faster access to emergency shelters (e.g., Beicheng, Xicheng, Dongcheng, and Changning Blocks, eastern Sihe and southern Anning Towns). In rural areas far from emergency shelters, residents have to spend more time accessing emergency shelters (e.g., Lizhou, Xingsheng, northern Anning, Xixiang, Chuanxing, and western Sihe Towns). For example, people living in Xingsheng Town need >20 min to access the nearest shelter, while people in Lizhou and Chuanxing Towns will spend more than 30 min escaping to selected emergency shelters. Figure 9A shows the accessibility level distribution based on road network analysis. In Lizhou, Xingsheng, western Anning, and Chuanxing Towns, the accessibility level is at Very-Low and Low levels, while for towns or blocks with a density of emergency shelters, accessibility presents Very-High and High levels.

After analyzing the impact of strong ground motion based on values of impact factors, we can see that the time cost escaping to emergency shelters increased in general (Supplementary Figure S3). In Figure 7B, we see five grades of classification based on the road network with clear accessibility distribution for different levels. For accessibility levels under the impact of ground motion, some of the time circle distributions are not clearly demonstrated (Figure 8). Figure 8C shows four grades of time-cost circle, while Figures 8A and 8B only display three distinct grades. This phenomenon might be caused by the increasing time cost in an earthquake scenario. Hence, we chose set C for the impact factors to analyze the influences from strong ground motion.

Based on the results in Figure 8C, we generated accessibility level distribution under the effect of strong ground motion. Our results show that the accessibility level in most areas changed after analyzing the influence from the predicted strong ground motion. Figure 9 shows the accessibility level distribution based on different sources. Figure 9A shows different accessibility levels only based on road networks, while Figures 9B and 9C show influence from predicted ground motion data calculated from the geodetic slip rates of the faults and those from geological fieldwork. Figure 9Bs and 9C show the accessibility of Low-level areas diminished to Very-Low (e.g., southern Lizhou, western Xingsheng, and central Chuanxing Towns). We also noted that the accessibility of Medium- and High-level areas in Figure 9A decreased to Low and Medium levels, as given in Figures 9B and 9C, respectively, such as the area in northeastern, northwestern, and southwestern Anning Town, central Xixiang Town, and the linked area of northeastern Xiaomiao Town, western Xijiao Town, and southwestern Sihe Town. For areas with a Very-High level in Figure 9A, the accessibility level is almost unchanged, such as the areas in Beicheng, Dongcheng, Xicheng, Changning and, Changan Blocks, and northern Gaoba, eastern Xiaomiao, and south-central Anning Towns. In Figures 9B and 9C, the accessibility levels of some specific towns or blocks, such as Anning and Xixiang Towns, cover four levels, including Very-High, Medium, Low, and Very-Low. There are three accessibility levels in Chuanxing Town: Very-High, Medium, and Very-Low. In the towns above, the accessibility level decreases with increasing distance from emergency shelters. While for the Beicheng Block, Dongcheng Block, Xicheng Block, and Changan Block, the accessibility level presents only the type of Very-High level. However, the accessibility level of the entirety of Lizhou Town (within the scope of our study) is only Very-Low.

Considering the fault rupturing effects in an earthquake, we also added the fault surface lines and generated the setback zone of regional mapped active faults. Two large-scale N–S trending faults spread across the urban areas of Xichang City, such as Anning, Xixiang, and Xiaomiao Towns (Figure 10). After reassigning the raster value of the setback zone, the accessibility level there is Very-Low. We found the apparent change in the accessibility level in towns or blocks crossed by setback zones, such as the Changan and Xicheng Blocks, at Very-High level decreased to Very-Low.

In this study, we developed an earthquake scenario-specific framework for SAA. We applied this framework to analyze the practical accessibility of emergency shelters in earthquakes for urban Xichang City. We employed active fault data and predicted strong ground motion data from geodetic and geological slip rates to explore the impact of earthquake damage.

Our results show that the areas with high accessibility to emergency shelters are in the middle of Xichang City, where emergency facilities are densely distributed, such as Beicheng, Dongcheng, Xicheng, Changan and Changning Blocks and Xijiao Town. Low accessibility levels are mainly distributed on the outskirts with few emergency shelters, such as in Lizhou, Xingsheng, central Chuanxing and, and western Anning Towns. Our results show that the spatial variation of accessibility level distribution decreases gradually with increasing distance from emergency shelters. It is worth noting that the accessibility level for Very-High areas almost did not change after adding the effect of potentially strong ground motion (see Figure 9), implying that people living in these areas can escape to emergency shelters in 5 min, even under strong ground motion from a sudden earthquake. We believed that the dense emergency shelters in urban areas can help residents evacuate in emergencies, even in earthquake scenarios.

Nevertheless, we also found that some areas with few emergency shelters still had a Very-High accessibility level, such as west Dongcheng Block and east Xicheng Block (Figure 11A). These areas have few emergency shelters but large populations and dense residential buildings. Our results show that their accessibility level is Very-High instead of Low. We considered this a result of their well-developed road networks. Thus, a developed road network can improve the accessibility level even in an area with few emergency shelters.

In contrast, an undeveloped road network can diminish accessibility, such as the linked area of northeastern Xiaomiao, western Xijiao, and southwestern Sihe Towns, which have plenty of emergency shelters. Their accessibility level is still Medium rather than Very High. In these areas, the hills (e.g., Panying, Wangjia, Macao, and Majingzi Hills) result in undeveloped road networks (Figure 11B), making evacuation more difficult. Hence, the role of road networks is also crucial to SAA.

We also demonstrated that the impact of potentially strong ground motion and fault rupturing on accessibility to emergency shelters should not be ignored. After analyzing potentially strong ground motion, the accessibility level in most areas declined due to the increasing time cost to access emergency shelters. In areas with mapped faults, the accessibility level decreased as far as Very-Low due to impedance from ruptured surfaces and damaged inhabited structures. These changes in the accessibility level to emergency shelters indicate that SAA should also consider the significant influence of extreme situations, such as strong earthquake scenarios. To cope with such an pessimistic emergency relief situation, the government should take action to refine emergency management. For areas with decreased levels of emergency shelter accessibility, the government should formulate an emergency evacuation plan and guidebook targeted at earthquake scenarios. For towns with multiple accessibility levels, the government should customize emergency evacuation work based on different accessibility levels (e.g., Anning, Xixiang, and Chuanxing Towns).

Our results indicate that people living on urban outskirts have difficulty escaping to emergency shelters. This does not directly mean that residents living there are more vulnerable to an earthquake. For these areas, we only analyzed the spatial accessibility of emergency shelters using the 59 selected emergency shelters, road network data, potential strong ground motion data, and active fault data, but we neglected the role of spaces for shelters, such as farmland and other open fields. In addition, it is easier for residents to exit structures via the lower height of rural residences and in lower populations than urban areas.

In summary, we developed an earthquake scenario-specific framework in the regional SAA for emergency shelters from the following aspects:

1. We developed an earthquake scenario-specific framework to generate a reasonable SAA for emergency shelters. Our framework is reasonable as it integrates the service capacity of urban emergency facilities with earthquake characteristics (strong ground motion and fault rupturing). We also added insights into earthquake scenario characteristics, which contribute to a reasonable simulation of emergency evacuation in earthquakes.

However, this framework still requires some improvements. 1)~We did not consider the relationship between population density and the capacity of emergency shelters. Among the 59 selected emergency shelters, the area varies from 994 m² to 2,789 m². Considering 2 m² per person to avoid overcrowding (Bernardini and Ferreira, 2022), the population capacity of emergency shelters ranges from 497 to 1,394, with a difference of nearly three times as significant.

2) We did not analyze the influence of the performance of buildings. The resistance of buildings can affect emergency evacuations. Building collapse will significantly impede emergency evacuation and even threaten the safety of residents (Guo et al., 2023; Magapu and Setia, 2023). The evacuation performance of buildings (e.g., distribution of separate stairs, maximum density, and vision time) is also related to evacuation efficiency (Chen et al., 2023).

3) In emergency evacuations, various responses to the surrounding environment can lead to multiple ways to avoid damage. For example, in a tunnel emergency evacuation, people’s behavior may be influenced by variables such as the best location to install the board, variable message signs (VMS) contents, and formats and how they interact (Ilkhani et al., 2023).

In this study, we proposed an earthquake scenario-specific framework to analyze spatial accessibility to emergency shelters. We applied this framework in urban Xichang City and achieved a comprehensive distribution of accessibility to emergency shelters by adding the impact of strong ground motion and fault rupturing.

Our results show that areas with dense emergency shelters and developed road networks present high accessibility levels (e.g., Beicheng, Dongcheng, and Xicheng Blocks). In contrast, areas far from emergency shelters with undeveloped road networks present low accessibility levels (e.g., Lizhou, Xisheng, Chuanxing, and western Xixiang Towns). As impacted by strong ground motion, the accessibility level decreases in several particular towns (Lizhou, Xingsheng, and Anning). For areas with mapped fault lines, residents and buildings would suffer most damage in earthquakes. The accessibility level to emergency shelters presents a Very-Low level.

Therefore, the SAA of emergency shelters is crucial for efficient evacuation, especially in complex earthquake scenarios. The blocks and towns with few emergency shelters and low accessibility levels must receive greatest attention. The seismic performance of buildings within the setback zone of active faults needs to be emphasized.